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 | |
d4b125e9 AE |
99 | #define RBD_SNAP_DEV_NAME_PREFIX "snap_" |
100 | #define RBD_MAX_SNAP_NAME_LEN \ | |
101 | (NAME_MAX - (sizeof (RBD_SNAP_DEV_NAME_PREFIX) - 1)) | |
102 | ||
35d489f9 | 103 | #define RBD_MAX_SNAP_COUNT 510 /* allows max snapc to fit in 4KB */ |
602adf40 YS |
104 | |
105 | #define RBD_SNAP_HEAD_NAME "-" | |
106 | ||
9682fc6d AE |
107 | #define BAD_SNAP_INDEX U32_MAX /* invalid index into snap array */ |
108 | ||
9e15b77d AE |
109 | /* This allows a single page to hold an image name sent by OSD */ |
110 | #define RBD_IMAGE_NAME_LEN_MAX (PAGE_SIZE - sizeof (__le32) - 1) | |
1e130199 | 111 | #define RBD_IMAGE_ID_LEN_MAX 64 |
9e15b77d | 112 | |
1e130199 | 113 | #define RBD_OBJ_PREFIX_LEN_MAX 64 |
589d30e0 | 114 | |
d889140c AE |
115 | /* Feature bits */ |
116 | ||
5cbf6f12 AE |
117 | #define RBD_FEATURE_LAYERING (1<<0) |
118 | #define RBD_FEATURE_STRIPINGV2 (1<<1) | |
119 | #define RBD_FEATURES_ALL \ | |
120 | (RBD_FEATURE_LAYERING | RBD_FEATURE_STRIPINGV2) | |
d889140c AE |
121 | |
122 | /* Features supported by this (client software) implementation. */ | |
123 | ||
770eba6e | 124 | #define RBD_FEATURES_SUPPORTED (RBD_FEATURES_ALL) |
d889140c | 125 | |
81a89793 AE |
126 | /* |
127 | * An RBD device name will be "rbd#", where the "rbd" comes from | |
128 | * RBD_DRV_NAME above, and # is a unique integer identifier. | |
129 | * MAX_INT_FORMAT_WIDTH is used in ensuring DEV_NAME_LEN is big | |
130 | * enough to hold all possible device names. | |
131 | */ | |
602adf40 | 132 | #define DEV_NAME_LEN 32 |
81a89793 | 133 | #define MAX_INT_FORMAT_WIDTH ((5 * sizeof (int)) / 2 + 1) |
602adf40 YS |
134 | |
135 | /* | |
136 | * block device image metadata (in-memory version) | |
137 | */ | |
138 | struct rbd_image_header { | |
f35a4dee | 139 | /* These six fields never change for a given rbd image */ |
849b4260 | 140 | char *object_prefix; |
602adf40 YS |
141 | __u8 obj_order; |
142 | __u8 crypt_type; | |
143 | __u8 comp_type; | |
f35a4dee AE |
144 | u64 stripe_unit; |
145 | u64 stripe_count; | |
146 | u64 features; /* Might be changeable someday? */ | |
602adf40 | 147 | |
f84344f3 AE |
148 | /* The remaining fields need to be updated occasionally */ |
149 | u64 image_size; | |
150 | struct ceph_snap_context *snapc; | |
f35a4dee AE |
151 | char *snap_names; /* format 1 only */ |
152 | u64 *snap_sizes; /* format 1 only */ | |
59c2be1e YS |
153 | }; |
154 | ||
0d7dbfce AE |
155 | /* |
156 | * An rbd image specification. | |
157 | * | |
158 | * The tuple (pool_id, image_id, snap_id) is sufficient to uniquely | |
c66c6e0c AE |
159 | * identify an image. Each rbd_dev structure includes a pointer to |
160 | * an rbd_spec structure that encapsulates this identity. | |
161 | * | |
162 | * Each of the id's in an rbd_spec has an associated name. For a | |
163 | * user-mapped image, the names are supplied and the id's associated | |
164 | * with them are looked up. For a layered image, a parent image is | |
165 | * defined by the tuple, and the names are looked up. | |
166 | * | |
167 | * An rbd_dev structure contains a parent_spec pointer which is | |
168 | * non-null if the image it represents is a child in a layered | |
169 | * image. This pointer will refer to the rbd_spec structure used | |
170 | * by the parent rbd_dev for its own identity (i.e., the structure | |
171 | * is shared between the parent and child). | |
172 | * | |
173 | * Since these structures are populated once, during the discovery | |
174 | * phase of image construction, they are effectively immutable so | |
175 | * we make no effort to synchronize access to them. | |
176 | * | |
177 | * Note that code herein does not assume the image name is known (it | |
178 | * could be a null pointer). | |
0d7dbfce AE |
179 | */ |
180 | struct rbd_spec { | |
181 | u64 pool_id; | |
ecb4dc22 | 182 | const char *pool_name; |
0d7dbfce | 183 | |
ecb4dc22 AE |
184 | const char *image_id; |
185 | const char *image_name; | |
0d7dbfce AE |
186 | |
187 | u64 snap_id; | |
ecb4dc22 | 188 | const char *snap_name; |
0d7dbfce AE |
189 | |
190 | struct kref kref; | |
191 | }; | |
192 | ||
602adf40 | 193 | /* |
f0f8cef5 | 194 | * an instance of the client. multiple devices may share an rbd client. |
602adf40 YS |
195 | */ |
196 | struct rbd_client { | |
197 | struct ceph_client *client; | |
198 | struct kref kref; | |
199 | struct list_head node; | |
200 | }; | |
201 | ||
bf0d5f50 AE |
202 | struct rbd_img_request; |
203 | typedef void (*rbd_img_callback_t)(struct rbd_img_request *); | |
204 | ||
205 | #define BAD_WHICH U32_MAX /* Good which or bad which, which? */ | |
206 | ||
207 | struct rbd_obj_request; | |
208 | typedef void (*rbd_obj_callback_t)(struct rbd_obj_request *); | |
209 | ||
9969ebc5 AE |
210 | enum obj_request_type { |
211 | OBJ_REQUEST_NODATA, OBJ_REQUEST_BIO, OBJ_REQUEST_PAGES | |
212 | }; | |
bf0d5f50 | 213 | |
6d2940c8 GZ |
214 | enum obj_operation_type { |
215 | OBJ_OP_WRITE, | |
216 | OBJ_OP_READ, | |
90e98c52 | 217 | OBJ_OP_DISCARD, |
6d2940c8 GZ |
218 | }; |
219 | ||
926f9b3f AE |
220 | enum obj_req_flags { |
221 | OBJ_REQ_DONE, /* completion flag: not done = 0, done = 1 */ | |
6365d33a | 222 | OBJ_REQ_IMG_DATA, /* object usage: standalone = 0, image = 1 */ |
5679c59f AE |
223 | OBJ_REQ_KNOWN, /* EXISTS flag valid: no = 0, yes = 1 */ |
224 | OBJ_REQ_EXISTS, /* target exists: no = 0, yes = 1 */ | |
926f9b3f AE |
225 | }; |
226 | ||
bf0d5f50 AE |
227 | struct rbd_obj_request { |
228 | const char *object_name; | |
229 | u64 offset; /* object start byte */ | |
230 | u64 length; /* bytes from offset */ | |
926f9b3f | 231 | unsigned long flags; |
bf0d5f50 | 232 | |
c5b5ef6c AE |
233 | /* |
234 | * An object request associated with an image will have its | |
235 | * img_data flag set; a standalone object request will not. | |
236 | * | |
237 | * A standalone object request will have which == BAD_WHICH | |
238 | * and a null obj_request pointer. | |
239 | * | |
240 | * An object request initiated in support of a layered image | |
241 | * object (to check for its existence before a write) will | |
242 | * have which == BAD_WHICH and a non-null obj_request pointer. | |
243 | * | |
244 | * Finally, an object request for rbd image data will have | |
245 | * which != BAD_WHICH, and will have a non-null img_request | |
246 | * pointer. The value of which will be in the range | |
247 | * 0..(img_request->obj_request_count-1). | |
248 | */ | |
249 | union { | |
250 | struct rbd_obj_request *obj_request; /* STAT op */ | |
251 | struct { | |
252 | struct rbd_img_request *img_request; | |
253 | u64 img_offset; | |
254 | /* links for img_request->obj_requests list */ | |
255 | struct list_head links; | |
256 | }; | |
257 | }; | |
bf0d5f50 AE |
258 | u32 which; /* posn image request list */ |
259 | ||
260 | enum obj_request_type type; | |
788e2df3 AE |
261 | union { |
262 | struct bio *bio_list; | |
263 | struct { | |
264 | struct page **pages; | |
265 | u32 page_count; | |
266 | }; | |
267 | }; | |
0eefd470 | 268 | struct page **copyup_pages; |
ebda6408 | 269 | u32 copyup_page_count; |
bf0d5f50 AE |
270 | |
271 | struct ceph_osd_request *osd_req; | |
272 | ||
273 | u64 xferred; /* bytes transferred */ | |
1b83bef2 | 274 | int result; |
bf0d5f50 AE |
275 | |
276 | rbd_obj_callback_t callback; | |
788e2df3 | 277 | struct completion completion; |
bf0d5f50 AE |
278 | |
279 | struct kref kref; | |
280 | }; | |
281 | ||
0c425248 | 282 | enum img_req_flags { |
9849e986 AE |
283 | IMG_REQ_WRITE, /* I/O direction: read = 0, write = 1 */ |
284 | IMG_REQ_CHILD, /* initiator: block = 0, child image = 1 */ | |
d0b2e944 | 285 | IMG_REQ_LAYERED, /* ENOENT handling: normal = 0, layered = 1 */ |
90e98c52 | 286 | IMG_REQ_DISCARD, /* discard: normal = 0, discard request = 1 */ |
0c425248 AE |
287 | }; |
288 | ||
bf0d5f50 | 289 | struct rbd_img_request { |
bf0d5f50 AE |
290 | struct rbd_device *rbd_dev; |
291 | u64 offset; /* starting image byte offset */ | |
292 | u64 length; /* byte count from offset */ | |
0c425248 | 293 | unsigned long flags; |
bf0d5f50 | 294 | union { |
9849e986 | 295 | u64 snap_id; /* for reads */ |
bf0d5f50 | 296 | struct ceph_snap_context *snapc; /* for writes */ |
9849e986 AE |
297 | }; |
298 | union { | |
299 | struct request *rq; /* block request */ | |
300 | struct rbd_obj_request *obj_request; /* obj req initiator */ | |
bf0d5f50 | 301 | }; |
3d7efd18 | 302 | struct page **copyup_pages; |
ebda6408 | 303 | u32 copyup_page_count; |
bf0d5f50 AE |
304 | spinlock_t completion_lock;/* protects next_completion */ |
305 | u32 next_completion; | |
306 | rbd_img_callback_t callback; | |
55f27e09 | 307 | u64 xferred;/* aggregate bytes transferred */ |
a5a337d4 | 308 | int result; /* first nonzero obj_request result */ |
bf0d5f50 AE |
309 | |
310 | u32 obj_request_count; | |
311 | struct list_head obj_requests; /* rbd_obj_request structs */ | |
312 | ||
313 | struct kref kref; | |
314 | }; | |
315 | ||
316 | #define for_each_obj_request(ireq, oreq) \ | |
ef06f4d3 | 317 | list_for_each_entry(oreq, &(ireq)->obj_requests, links) |
bf0d5f50 | 318 | #define for_each_obj_request_from(ireq, oreq) \ |
ef06f4d3 | 319 | list_for_each_entry_from(oreq, &(ireq)->obj_requests, links) |
bf0d5f50 | 320 | #define for_each_obj_request_safe(ireq, oreq, n) \ |
ef06f4d3 | 321 | list_for_each_entry_safe_reverse(oreq, n, &(ireq)->obj_requests, links) |
bf0d5f50 | 322 | |
f84344f3 | 323 | struct rbd_mapping { |
99c1f08f | 324 | u64 size; |
34b13184 | 325 | u64 features; |
f84344f3 AE |
326 | bool read_only; |
327 | }; | |
328 | ||
602adf40 YS |
329 | /* |
330 | * a single device | |
331 | */ | |
332 | struct rbd_device { | |
de71a297 | 333 | int dev_id; /* blkdev unique id */ |
602adf40 YS |
334 | |
335 | int major; /* blkdev assigned major */ | |
dd82fff1 | 336 | int minor; |
602adf40 | 337 | struct gendisk *disk; /* blkdev's gendisk and rq */ |
602adf40 | 338 | |
a30b71b9 | 339 | u32 image_format; /* Either 1 or 2 */ |
602adf40 YS |
340 | struct rbd_client *rbd_client; |
341 | ||
342 | char name[DEV_NAME_LEN]; /* blkdev name, e.g. rbd3 */ | |
343 | ||
b82d167b | 344 | spinlock_t lock; /* queue, flags, open_count */ |
602adf40 YS |
345 | |
346 | struct rbd_image_header header; | |
b82d167b | 347 | unsigned long flags; /* possibly lock protected */ |
0d7dbfce | 348 | struct rbd_spec *spec; |
602adf40 | 349 | |
0d7dbfce | 350 | char *header_name; |
971f839a | 351 | |
0903e875 AE |
352 | struct ceph_file_layout layout; |
353 | ||
59c2be1e | 354 | struct ceph_osd_event *watch_event; |
975241af | 355 | struct rbd_obj_request *watch_request; |
59c2be1e | 356 | |
86b00e0d AE |
357 | struct rbd_spec *parent_spec; |
358 | u64 parent_overlap; | |
a2acd00e | 359 | atomic_t parent_ref; |
2f82ee54 | 360 | struct rbd_device *parent; |
86b00e0d | 361 | |
7ad18afa CH |
362 | /* Block layer tags. */ |
363 | struct blk_mq_tag_set tag_set; | |
364 | ||
c666601a JD |
365 | /* protects updating the header */ |
366 | struct rw_semaphore header_rwsem; | |
f84344f3 AE |
367 | |
368 | struct rbd_mapping mapping; | |
602adf40 YS |
369 | |
370 | struct list_head node; | |
dfc5606d | 371 | |
dfc5606d YS |
372 | /* sysfs related */ |
373 | struct device dev; | |
b82d167b | 374 | unsigned long open_count; /* protected by lock */ |
dfc5606d YS |
375 | }; |
376 | ||
b82d167b AE |
377 | /* |
378 | * Flag bits for rbd_dev->flags. If atomicity is required, | |
379 | * rbd_dev->lock is used to protect access. | |
380 | * | |
381 | * Currently, only the "removing" flag (which is coupled with the | |
382 | * "open_count" field) requires atomic access. | |
383 | */ | |
6d292906 AE |
384 | enum rbd_dev_flags { |
385 | RBD_DEV_FLAG_EXISTS, /* mapped snapshot has not been deleted */ | |
b82d167b | 386 | RBD_DEV_FLAG_REMOVING, /* this mapping is being removed */ |
6d292906 AE |
387 | }; |
388 | ||
cfbf6377 | 389 | static DEFINE_MUTEX(client_mutex); /* Serialize client creation */ |
e124a82f | 390 | |
602adf40 | 391 | static LIST_HEAD(rbd_dev_list); /* devices */ |
e124a82f AE |
392 | static DEFINE_SPINLOCK(rbd_dev_list_lock); |
393 | ||
432b8587 AE |
394 | static LIST_HEAD(rbd_client_list); /* clients */ |
395 | static DEFINE_SPINLOCK(rbd_client_list_lock); | |
602adf40 | 396 | |
78c2a44a AE |
397 | /* Slab caches for frequently-allocated structures */ |
398 | ||
1c2a9dfe | 399 | static struct kmem_cache *rbd_img_request_cache; |
868311b1 | 400 | static struct kmem_cache *rbd_obj_request_cache; |
78c2a44a | 401 | static struct kmem_cache *rbd_segment_name_cache; |
1c2a9dfe | 402 | |
9b60e70b | 403 | static int rbd_major; |
f8a22fc2 ID |
404 | static DEFINE_IDA(rbd_dev_id_ida); |
405 | ||
f5ee37bd ID |
406 | static struct workqueue_struct *rbd_wq; |
407 | ||
9b60e70b ID |
408 | /* |
409 | * Default to false for now, as single-major requires >= 0.75 version of | |
410 | * userspace rbd utility. | |
411 | */ | |
412 | static bool single_major = false; | |
413 | module_param(single_major, bool, S_IRUGO); | |
414 | MODULE_PARM_DESC(single_major, "Use a single major number for all rbd devices (default: false)"); | |
415 | ||
3d7efd18 AE |
416 | static int rbd_img_request_submit(struct rbd_img_request *img_request); |
417 | ||
200a6a8b | 418 | static void rbd_dev_device_release(struct device *dev); |
dfc5606d | 419 | |
f0f8cef5 AE |
420 | static ssize_t rbd_add(struct bus_type *bus, const char *buf, |
421 | size_t count); | |
422 | static ssize_t rbd_remove(struct bus_type *bus, const char *buf, | |
423 | size_t count); | |
9b60e70b ID |
424 | static ssize_t rbd_add_single_major(struct bus_type *bus, const char *buf, |
425 | size_t count); | |
426 | static ssize_t rbd_remove_single_major(struct bus_type *bus, const char *buf, | |
427 | size_t count); | |
1f3ef788 | 428 | static int rbd_dev_image_probe(struct rbd_device *rbd_dev, bool mapping); |
a2acd00e | 429 | static void rbd_spec_put(struct rbd_spec *spec); |
f0f8cef5 | 430 | |
9b60e70b ID |
431 | static int rbd_dev_id_to_minor(int dev_id) |
432 | { | |
7e513d43 | 433 | return dev_id << RBD_SINGLE_MAJOR_PART_SHIFT; |
9b60e70b ID |
434 | } |
435 | ||
436 | static int minor_to_rbd_dev_id(int minor) | |
437 | { | |
7e513d43 | 438 | return minor >> RBD_SINGLE_MAJOR_PART_SHIFT; |
9b60e70b ID |
439 | } |
440 | ||
b15a21dd GKH |
441 | static BUS_ATTR(add, S_IWUSR, NULL, rbd_add); |
442 | static BUS_ATTR(remove, S_IWUSR, NULL, rbd_remove); | |
9b60e70b ID |
443 | static BUS_ATTR(add_single_major, S_IWUSR, NULL, rbd_add_single_major); |
444 | static BUS_ATTR(remove_single_major, S_IWUSR, NULL, rbd_remove_single_major); | |
b15a21dd GKH |
445 | |
446 | static struct attribute *rbd_bus_attrs[] = { | |
447 | &bus_attr_add.attr, | |
448 | &bus_attr_remove.attr, | |
9b60e70b ID |
449 | &bus_attr_add_single_major.attr, |
450 | &bus_attr_remove_single_major.attr, | |
b15a21dd | 451 | NULL, |
f0f8cef5 | 452 | }; |
92c76dc0 ID |
453 | |
454 | static umode_t rbd_bus_is_visible(struct kobject *kobj, | |
455 | struct attribute *attr, int index) | |
456 | { | |
9b60e70b ID |
457 | if (!single_major && |
458 | (attr == &bus_attr_add_single_major.attr || | |
459 | attr == &bus_attr_remove_single_major.attr)) | |
460 | return 0; | |
461 | ||
92c76dc0 ID |
462 | return attr->mode; |
463 | } | |
464 | ||
465 | static const struct attribute_group rbd_bus_group = { | |
466 | .attrs = rbd_bus_attrs, | |
467 | .is_visible = rbd_bus_is_visible, | |
468 | }; | |
469 | __ATTRIBUTE_GROUPS(rbd_bus); | |
f0f8cef5 AE |
470 | |
471 | static struct bus_type rbd_bus_type = { | |
472 | .name = "rbd", | |
b15a21dd | 473 | .bus_groups = rbd_bus_groups, |
f0f8cef5 AE |
474 | }; |
475 | ||
476 | static void rbd_root_dev_release(struct device *dev) | |
477 | { | |
478 | } | |
479 | ||
480 | static struct device rbd_root_dev = { | |
481 | .init_name = "rbd", | |
482 | .release = rbd_root_dev_release, | |
483 | }; | |
484 | ||
06ecc6cb AE |
485 | static __printf(2, 3) |
486 | void rbd_warn(struct rbd_device *rbd_dev, const char *fmt, ...) | |
487 | { | |
488 | struct va_format vaf; | |
489 | va_list args; | |
490 | ||
491 | va_start(args, fmt); | |
492 | vaf.fmt = fmt; | |
493 | vaf.va = &args; | |
494 | ||
495 | if (!rbd_dev) | |
496 | printk(KERN_WARNING "%s: %pV\n", RBD_DRV_NAME, &vaf); | |
497 | else if (rbd_dev->disk) | |
498 | printk(KERN_WARNING "%s: %s: %pV\n", | |
499 | RBD_DRV_NAME, rbd_dev->disk->disk_name, &vaf); | |
500 | else if (rbd_dev->spec && rbd_dev->spec->image_name) | |
501 | printk(KERN_WARNING "%s: image %s: %pV\n", | |
502 | RBD_DRV_NAME, rbd_dev->spec->image_name, &vaf); | |
503 | else if (rbd_dev->spec && rbd_dev->spec->image_id) | |
504 | printk(KERN_WARNING "%s: id %s: %pV\n", | |
505 | RBD_DRV_NAME, rbd_dev->spec->image_id, &vaf); | |
506 | else /* punt */ | |
507 | printk(KERN_WARNING "%s: rbd_dev %p: %pV\n", | |
508 | RBD_DRV_NAME, rbd_dev, &vaf); | |
509 | va_end(args); | |
510 | } | |
511 | ||
aafb230e AE |
512 | #ifdef RBD_DEBUG |
513 | #define rbd_assert(expr) \ | |
514 | if (unlikely(!(expr))) { \ | |
515 | printk(KERN_ERR "\nAssertion failure in %s() " \ | |
516 | "at line %d:\n\n" \ | |
517 | "\trbd_assert(%s);\n\n", \ | |
518 | __func__, __LINE__, #expr); \ | |
519 | BUG(); \ | |
520 | } | |
521 | #else /* !RBD_DEBUG */ | |
522 | # define rbd_assert(expr) ((void) 0) | |
523 | #endif /* !RBD_DEBUG */ | |
dfc5606d | 524 | |
b454e36d | 525 | static int rbd_img_obj_request_submit(struct rbd_obj_request *obj_request); |
05a46afd AE |
526 | static void rbd_img_parent_read(struct rbd_obj_request *obj_request); |
527 | static void rbd_dev_remove_parent(struct rbd_device *rbd_dev); | |
8b3e1a56 | 528 | |
cc4a38bd | 529 | static int rbd_dev_refresh(struct rbd_device *rbd_dev); |
2df3fac7 | 530 | static int rbd_dev_v2_header_onetime(struct rbd_device *rbd_dev); |
a720ae09 | 531 | static int rbd_dev_header_info(struct rbd_device *rbd_dev); |
e8f59b59 | 532 | static int rbd_dev_v2_parent_info(struct rbd_device *rbd_dev); |
54cac61f AE |
533 | static const char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev, |
534 | u64 snap_id); | |
2ad3d716 AE |
535 | static int _rbd_dev_v2_snap_size(struct rbd_device *rbd_dev, u64 snap_id, |
536 | u8 *order, u64 *snap_size); | |
537 | static int _rbd_dev_v2_snap_features(struct rbd_device *rbd_dev, u64 snap_id, | |
538 | u64 *snap_features); | |
539 | static u64 rbd_snap_id_by_name(struct rbd_device *rbd_dev, const char *name); | |
59c2be1e | 540 | |
602adf40 YS |
541 | static int rbd_open(struct block_device *bdev, fmode_t mode) |
542 | { | |
f0f8cef5 | 543 | struct rbd_device *rbd_dev = bdev->bd_disk->private_data; |
b82d167b | 544 | bool removing = false; |
602adf40 | 545 | |
f84344f3 | 546 | if ((mode & FMODE_WRITE) && rbd_dev->mapping.read_only) |
602adf40 YS |
547 | return -EROFS; |
548 | ||
a14ea269 | 549 | spin_lock_irq(&rbd_dev->lock); |
b82d167b AE |
550 | if (test_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags)) |
551 | removing = true; | |
552 | else | |
553 | rbd_dev->open_count++; | |
a14ea269 | 554 | spin_unlock_irq(&rbd_dev->lock); |
b82d167b AE |
555 | if (removing) |
556 | return -ENOENT; | |
557 | ||
c3e946ce | 558 | (void) get_device(&rbd_dev->dev); |
340c7a2b | 559 | |
602adf40 YS |
560 | return 0; |
561 | } | |
562 | ||
db2a144b | 563 | static void rbd_release(struct gendisk *disk, fmode_t mode) |
dfc5606d YS |
564 | { |
565 | struct rbd_device *rbd_dev = disk->private_data; | |
b82d167b AE |
566 | unsigned long open_count_before; |
567 | ||
a14ea269 | 568 | spin_lock_irq(&rbd_dev->lock); |
b82d167b | 569 | open_count_before = rbd_dev->open_count--; |
a14ea269 | 570 | spin_unlock_irq(&rbd_dev->lock); |
b82d167b | 571 | rbd_assert(open_count_before > 0); |
dfc5606d | 572 | |
c3e946ce | 573 | put_device(&rbd_dev->dev); |
dfc5606d YS |
574 | } |
575 | ||
131fd9f6 GZ |
576 | static int rbd_ioctl_set_ro(struct rbd_device *rbd_dev, unsigned long arg) |
577 | { | |
77f33c03 | 578 | int ret = 0; |
131fd9f6 GZ |
579 | int val; |
580 | bool ro; | |
77f33c03 | 581 | bool ro_changed = false; |
131fd9f6 | 582 | |
77f33c03 | 583 | /* get_user() may sleep, so call it before taking rbd_dev->lock */ |
131fd9f6 GZ |
584 | if (get_user(val, (int __user *)(arg))) |
585 | return -EFAULT; | |
586 | ||
587 | ro = val ? true : false; | |
588 | /* Snapshot doesn't allow to write*/ | |
589 | if (rbd_dev->spec->snap_id != CEPH_NOSNAP && !ro) | |
590 | return -EROFS; | |
591 | ||
77f33c03 JD |
592 | spin_lock_irq(&rbd_dev->lock); |
593 | /* prevent others open this device */ | |
594 | if (rbd_dev->open_count > 1) { | |
595 | ret = -EBUSY; | |
596 | goto out; | |
597 | } | |
598 | ||
131fd9f6 GZ |
599 | if (rbd_dev->mapping.read_only != ro) { |
600 | rbd_dev->mapping.read_only = ro; | |
77f33c03 | 601 | ro_changed = true; |
131fd9f6 GZ |
602 | } |
603 | ||
77f33c03 JD |
604 | out: |
605 | spin_unlock_irq(&rbd_dev->lock); | |
606 | /* set_disk_ro() may sleep, so call it after releasing rbd_dev->lock */ | |
607 | if (ret == 0 && ro_changed) | |
608 | set_disk_ro(rbd_dev->disk, ro ? 1 : 0); | |
609 | ||
610 | return ret; | |
131fd9f6 GZ |
611 | } |
612 | ||
613 | static int rbd_ioctl(struct block_device *bdev, fmode_t mode, | |
614 | unsigned int cmd, unsigned long arg) | |
615 | { | |
616 | struct rbd_device *rbd_dev = bdev->bd_disk->private_data; | |
617 | int ret = 0; | |
618 | ||
131fd9f6 GZ |
619 | switch (cmd) { |
620 | case BLKROSET: | |
621 | ret = rbd_ioctl_set_ro(rbd_dev, arg); | |
622 | break; | |
623 | default: | |
624 | ret = -ENOTTY; | |
625 | } | |
626 | ||
131fd9f6 GZ |
627 | return ret; |
628 | } | |
629 | ||
630 | #ifdef CONFIG_COMPAT | |
631 | static int rbd_compat_ioctl(struct block_device *bdev, fmode_t mode, | |
632 | unsigned int cmd, unsigned long arg) | |
633 | { | |
634 | return rbd_ioctl(bdev, mode, cmd, arg); | |
635 | } | |
636 | #endif /* CONFIG_COMPAT */ | |
637 | ||
602adf40 YS |
638 | static const struct block_device_operations rbd_bd_ops = { |
639 | .owner = THIS_MODULE, | |
640 | .open = rbd_open, | |
dfc5606d | 641 | .release = rbd_release, |
131fd9f6 GZ |
642 | .ioctl = rbd_ioctl, |
643 | #ifdef CONFIG_COMPAT | |
644 | .compat_ioctl = rbd_compat_ioctl, | |
645 | #endif | |
602adf40 YS |
646 | }; |
647 | ||
648 | /* | |
7262cfca | 649 | * Initialize an rbd client instance. Success or not, this function |
cfbf6377 | 650 | * consumes ceph_opts. Caller holds client_mutex. |
602adf40 | 651 | */ |
f8c38929 | 652 | static struct rbd_client *rbd_client_create(struct ceph_options *ceph_opts) |
602adf40 YS |
653 | { |
654 | struct rbd_client *rbdc; | |
655 | int ret = -ENOMEM; | |
656 | ||
37206ee5 | 657 | dout("%s:\n", __func__); |
602adf40 YS |
658 | rbdc = kmalloc(sizeof(struct rbd_client), GFP_KERNEL); |
659 | if (!rbdc) | |
660 | goto out_opt; | |
661 | ||
662 | kref_init(&rbdc->kref); | |
663 | INIT_LIST_HEAD(&rbdc->node); | |
664 | ||
43ae4701 | 665 | rbdc->client = ceph_create_client(ceph_opts, rbdc, 0, 0); |
602adf40 | 666 | if (IS_ERR(rbdc->client)) |
08f75463 | 667 | goto out_rbdc; |
43ae4701 | 668 | ceph_opts = NULL; /* Now rbdc->client is responsible for ceph_opts */ |
602adf40 YS |
669 | |
670 | ret = ceph_open_session(rbdc->client); | |
671 | if (ret < 0) | |
08f75463 | 672 | goto out_client; |
602adf40 | 673 | |
432b8587 | 674 | spin_lock(&rbd_client_list_lock); |
602adf40 | 675 | list_add_tail(&rbdc->node, &rbd_client_list); |
432b8587 | 676 | spin_unlock(&rbd_client_list_lock); |
602adf40 | 677 | |
37206ee5 | 678 | dout("%s: rbdc %p\n", __func__, rbdc); |
bc534d86 | 679 | |
602adf40 | 680 | return rbdc; |
08f75463 | 681 | out_client: |
602adf40 | 682 | ceph_destroy_client(rbdc->client); |
08f75463 | 683 | out_rbdc: |
602adf40 YS |
684 | kfree(rbdc); |
685 | out_opt: | |
43ae4701 AE |
686 | if (ceph_opts) |
687 | ceph_destroy_options(ceph_opts); | |
37206ee5 AE |
688 | dout("%s: error %d\n", __func__, ret); |
689 | ||
28f259b7 | 690 | return ERR_PTR(ret); |
602adf40 YS |
691 | } |
692 | ||
2f82ee54 AE |
693 | static struct rbd_client *__rbd_get_client(struct rbd_client *rbdc) |
694 | { | |
695 | kref_get(&rbdc->kref); | |
696 | ||
697 | return rbdc; | |
698 | } | |
699 | ||
602adf40 | 700 | /* |
1f7ba331 AE |
701 | * Find a ceph client with specific addr and configuration. If |
702 | * found, bump its reference count. | |
602adf40 | 703 | */ |
1f7ba331 | 704 | static struct rbd_client *rbd_client_find(struct ceph_options *ceph_opts) |
602adf40 YS |
705 | { |
706 | struct rbd_client *client_node; | |
1f7ba331 | 707 | bool found = false; |
602adf40 | 708 | |
43ae4701 | 709 | if (ceph_opts->flags & CEPH_OPT_NOSHARE) |
602adf40 YS |
710 | return NULL; |
711 | ||
1f7ba331 AE |
712 | spin_lock(&rbd_client_list_lock); |
713 | list_for_each_entry(client_node, &rbd_client_list, node) { | |
714 | if (!ceph_compare_options(ceph_opts, client_node->client)) { | |
2f82ee54 AE |
715 | __rbd_get_client(client_node); |
716 | ||
1f7ba331 AE |
717 | found = true; |
718 | break; | |
719 | } | |
720 | } | |
721 | spin_unlock(&rbd_client_list_lock); | |
722 | ||
723 | return found ? client_node : NULL; | |
602adf40 YS |
724 | } |
725 | ||
59c2be1e | 726 | /* |
210c104c | 727 | * (Per device) rbd map options |
59c2be1e YS |
728 | */ |
729 | enum { | |
59c2be1e YS |
730 | Opt_last_int, |
731 | /* int args above */ | |
732 | Opt_last_string, | |
733 | /* string args above */ | |
cc0538b6 AE |
734 | Opt_read_only, |
735 | Opt_read_write, | |
210c104c | 736 | Opt_err |
59c2be1e YS |
737 | }; |
738 | ||
43ae4701 | 739 | static match_table_t rbd_opts_tokens = { |
59c2be1e YS |
740 | /* int args above */ |
741 | /* string args above */ | |
be466c1c | 742 | {Opt_read_only, "read_only"}, |
cc0538b6 AE |
743 | {Opt_read_only, "ro"}, /* Alternate spelling */ |
744 | {Opt_read_write, "read_write"}, | |
745 | {Opt_read_write, "rw"}, /* Alternate spelling */ | |
210c104c | 746 | {Opt_err, NULL} |
59c2be1e YS |
747 | }; |
748 | ||
98571b5a AE |
749 | struct rbd_options { |
750 | bool read_only; | |
751 | }; | |
752 | ||
753 | #define RBD_READ_ONLY_DEFAULT false | |
754 | ||
59c2be1e YS |
755 | static int parse_rbd_opts_token(char *c, void *private) |
756 | { | |
43ae4701 | 757 | struct rbd_options *rbd_opts = private; |
59c2be1e YS |
758 | substring_t argstr[MAX_OPT_ARGS]; |
759 | int token, intval, ret; | |
760 | ||
43ae4701 | 761 | token = match_token(c, rbd_opts_tokens, argstr); |
59c2be1e YS |
762 | if (token < Opt_last_int) { |
763 | ret = match_int(&argstr[0], &intval); | |
764 | if (ret < 0) { | |
210c104c | 765 | pr_err("bad mount option arg (not int) at '%s'\n", c); |
59c2be1e YS |
766 | return ret; |
767 | } | |
768 | dout("got int token %d val %d\n", token, intval); | |
769 | } else if (token > Opt_last_int && token < Opt_last_string) { | |
210c104c | 770 | dout("got string token %d val %s\n", token, argstr[0].from); |
59c2be1e YS |
771 | } else { |
772 | dout("got token %d\n", token); | |
773 | } | |
774 | ||
775 | switch (token) { | |
cc0538b6 AE |
776 | case Opt_read_only: |
777 | rbd_opts->read_only = true; | |
778 | break; | |
779 | case Opt_read_write: | |
780 | rbd_opts->read_only = false; | |
781 | break; | |
59c2be1e | 782 | default: |
210c104c ID |
783 | /* libceph prints "bad option" msg */ |
784 | return -EINVAL; | |
59c2be1e | 785 | } |
210c104c | 786 | |
59c2be1e YS |
787 | return 0; |
788 | } | |
789 | ||
6d2940c8 GZ |
790 | static char* obj_op_name(enum obj_operation_type op_type) |
791 | { | |
792 | switch (op_type) { | |
793 | case OBJ_OP_READ: | |
794 | return "read"; | |
795 | case OBJ_OP_WRITE: | |
796 | return "write"; | |
90e98c52 GZ |
797 | case OBJ_OP_DISCARD: |
798 | return "discard"; | |
6d2940c8 GZ |
799 | default: |
800 | return "???"; | |
801 | } | |
802 | } | |
803 | ||
602adf40 YS |
804 | /* |
805 | * Get a ceph client with specific addr and configuration, if one does | |
7262cfca AE |
806 | * not exist create it. Either way, ceph_opts is consumed by this |
807 | * function. | |
602adf40 | 808 | */ |
9d3997fd | 809 | static struct rbd_client *rbd_get_client(struct ceph_options *ceph_opts) |
602adf40 | 810 | { |
f8c38929 | 811 | struct rbd_client *rbdc; |
59c2be1e | 812 | |
cfbf6377 | 813 | mutex_lock_nested(&client_mutex, SINGLE_DEPTH_NESTING); |
1f7ba331 | 814 | rbdc = rbd_client_find(ceph_opts); |
9d3997fd | 815 | if (rbdc) /* using an existing client */ |
43ae4701 | 816 | ceph_destroy_options(ceph_opts); |
9d3997fd | 817 | else |
f8c38929 | 818 | rbdc = rbd_client_create(ceph_opts); |
cfbf6377 | 819 | mutex_unlock(&client_mutex); |
602adf40 | 820 | |
9d3997fd | 821 | return rbdc; |
602adf40 YS |
822 | } |
823 | ||
824 | /* | |
825 | * Destroy ceph client | |
d23a4b3f | 826 | * |
432b8587 | 827 | * Caller must hold rbd_client_list_lock. |
602adf40 YS |
828 | */ |
829 | static void rbd_client_release(struct kref *kref) | |
830 | { | |
831 | struct rbd_client *rbdc = container_of(kref, struct rbd_client, kref); | |
832 | ||
37206ee5 | 833 | dout("%s: rbdc %p\n", __func__, rbdc); |
cd9d9f5d | 834 | spin_lock(&rbd_client_list_lock); |
602adf40 | 835 | list_del(&rbdc->node); |
cd9d9f5d | 836 | spin_unlock(&rbd_client_list_lock); |
602adf40 YS |
837 | |
838 | ceph_destroy_client(rbdc->client); | |
839 | kfree(rbdc); | |
840 | } | |
841 | ||
842 | /* | |
843 | * Drop reference to ceph client node. If it's not referenced anymore, release | |
844 | * it. | |
845 | */ | |
9d3997fd | 846 | static void rbd_put_client(struct rbd_client *rbdc) |
602adf40 | 847 | { |
c53d5893 AE |
848 | if (rbdc) |
849 | kref_put(&rbdc->kref, rbd_client_release); | |
602adf40 YS |
850 | } |
851 | ||
a30b71b9 AE |
852 | static bool rbd_image_format_valid(u32 image_format) |
853 | { | |
854 | return image_format == 1 || image_format == 2; | |
855 | } | |
856 | ||
8e94af8e AE |
857 | static bool rbd_dev_ondisk_valid(struct rbd_image_header_ondisk *ondisk) |
858 | { | |
103a150f AE |
859 | size_t size; |
860 | u32 snap_count; | |
861 | ||
862 | /* The header has to start with the magic rbd header text */ | |
863 | if (memcmp(&ondisk->text, RBD_HEADER_TEXT, sizeof (RBD_HEADER_TEXT))) | |
864 | return false; | |
865 | ||
db2388b6 AE |
866 | /* The bio layer requires at least sector-sized I/O */ |
867 | ||
868 | if (ondisk->options.order < SECTOR_SHIFT) | |
869 | return false; | |
870 | ||
871 | /* If we use u64 in a few spots we may be able to loosen this */ | |
872 | ||
873 | if (ondisk->options.order > 8 * sizeof (int) - 1) | |
874 | return false; | |
875 | ||
103a150f AE |
876 | /* |
877 | * The size of a snapshot header has to fit in a size_t, and | |
878 | * that limits the number of snapshots. | |
879 | */ | |
880 | snap_count = le32_to_cpu(ondisk->snap_count); | |
881 | size = SIZE_MAX - sizeof (struct ceph_snap_context); | |
882 | if (snap_count > size / sizeof (__le64)) | |
883 | return false; | |
884 | ||
885 | /* | |
886 | * Not only that, but the size of the entire the snapshot | |
887 | * header must also be representable in a size_t. | |
888 | */ | |
889 | size -= snap_count * sizeof (__le64); | |
890 | if ((u64) size < le64_to_cpu(ondisk->snap_names_len)) | |
891 | return false; | |
892 | ||
893 | return true; | |
8e94af8e AE |
894 | } |
895 | ||
602adf40 | 896 | /* |
bb23e37a AE |
897 | * Fill an rbd image header with information from the given format 1 |
898 | * on-disk header. | |
602adf40 | 899 | */ |
662518b1 | 900 | static int rbd_header_from_disk(struct rbd_device *rbd_dev, |
4156d998 | 901 | struct rbd_image_header_ondisk *ondisk) |
602adf40 | 902 | { |
662518b1 | 903 | struct rbd_image_header *header = &rbd_dev->header; |
bb23e37a AE |
904 | bool first_time = header->object_prefix == NULL; |
905 | struct ceph_snap_context *snapc; | |
906 | char *object_prefix = NULL; | |
907 | char *snap_names = NULL; | |
908 | u64 *snap_sizes = NULL; | |
ccece235 | 909 | u32 snap_count; |
d2bb24e5 | 910 | size_t size; |
bb23e37a | 911 | int ret = -ENOMEM; |
621901d6 | 912 | u32 i; |
602adf40 | 913 | |
bb23e37a | 914 | /* Allocate this now to avoid having to handle failure below */ |
6a52325f | 915 | |
bb23e37a AE |
916 | if (first_time) { |
917 | size_t len; | |
103a150f | 918 | |
bb23e37a AE |
919 | len = strnlen(ondisk->object_prefix, |
920 | sizeof (ondisk->object_prefix)); | |
921 | object_prefix = kmalloc(len + 1, GFP_KERNEL); | |
922 | if (!object_prefix) | |
923 | return -ENOMEM; | |
924 | memcpy(object_prefix, ondisk->object_prefix, len); | |
925 | object_prefix[len] = '\0'; | |
926 | } | |
00f1f36f | 927 | |
bb23e37a | 928 | /* Allocate the snapshot context and fill it in */ |
00f1f36f | 929 | |
bb23e37a AE |
930 | snap_count = le32_to_cpu(ondisk->snap_count); |
931 | snapc = ceph_create_snap_context(snap_count, GFP_KERNEL); | |
932 | if (!snapc) | |
933 | goto out_err; | |
934 | snapc->seq = le64_to_cpu(ondisk->snap_seq); | |
602adf40 | 935 | if (snap_count) { |
bb23e37a | 936 | struct rbd_image_snap_ondisk *snaps; |
f785cc1d AE |
937 | u64 snap_names_len = le64_to_cpu(ondisk->snap_names_len); |
938 | ||
bb23e37a | 939 | /* We'll keep a copy of the snapshot names... */ |
621901d6 | 940 | |
bb23e37a AE |
941 | if (snap_names_len > (u64)SIZE_MAX) |
942 | goto out_2big; | |
943 | snap_names = kmalloc(snap_names_len, GFP_KERNEL); | |
944 | if (!snap_names) | |
6a52325f AE |
945 | goto out_err; |
946 | ||
bb23e37a | 947 | /* ...as well as the array of their sizes. */ |
621901d6 | 948 | |
d2bb24e5 | 949 | size = snap_count * sizeof (*header->snap_sizes); |
bb23e37a AE |
950 | snap_sizes = kmalloc(size, GFP_KERNEL); |
951 | if (!snap_sizes) | |
6a52325f | 952 | goto out_err; |
bb23e37a | 953 | |
f785cc1d | 954 | /* |
bb23e37a AE |
955 | * Copy the names, and fill in each snapshot's id |
956 | * and size. | |
957 | * | |
99a41ebc | 958 | * Note that rbd_dev_v1_header_info() guarantees the |
bb23e37a | 959 | * ondisk buffer we're working with has |
f785cc1d AE |
960 | * snap_names_len bytes beyond the end of the |
961 | * snapshot id array, this memcpy() is safe. | |
962 | */ | |
bb23e37a AE |
963 | memcpy(snap_names, &ondisk->snaps[snap_count], snap_names_len); |
964 | snaps = ondisk->snaps; | |
965 | for (i = 0; i < snap_count; i++) { | |
966 | snapc->snaps[i] = le64_to_cpu(snaps[i].id); | |
967 | snap_sizes[i] = le64_to_cpu(snaps[i].image_size); | |
968 | } | |
602adf40 | 969 | } |
6a52325f | 970 | |
bb23e37a | 971 | /* We won't fail any more, fill in the header */ |
621901d6 | 972 | |
bb23e37a AE |
973 | if (first_time) { |
974 | header->object_prefix = object_prefix; | |
975 | header->obj_order = ondisk->options.order; | |
976 | header->crypt_type = ondisk->options.crypt_type; | |
977 | header->comp_type = ondisk->options.comp_type; | |
978 | /* The rest aren't used for format 1 images */ | |
979 | header->stripe_unit = 0; | |
980 | header->stripe_count = 0; | |
981 | header->features = 0; | |
602adf40 | 982 | } else { |
662518b1 AE |
983 | ceph_put_snap_context(header->snapc); |
984 | kfree(header->snap_names); | |
985 | kfree(header->snap_sizes); | |
602adf40 | 986 | } |
849b4260 | 987 | |
bb23e37a | 988 | /* The remaining fields always get updated (when we refresh) */ |
621901d6 | 989 | |
f84344f3 | 990 | header->image_size = le64_to_cpu(ondisk->image_size); |
bb23e37a AE |
991 | header->snapc = snapc; |
992 | header->snap_names = snap_names; | |
993 | header->snap_sizes = snap_sizes; | |
468521c1 | 994 | |
602adf40 | 995 | return 0; |
bb23e37a AE |
996 | out_2big: |
997 | ret = -EIO; | |
6a52325f | 998 | out_err: |
bb23e37a AE |
999 | kfree(snap_sizes); |
1000 | kfree(snap_names); | |
1001 | ceph_put_snap_context(snapc); | |
1002 | kfree(object_prefix); | |
ccece235 | 1003 | |
bb23e37a | 1004 | return ret; |
602adf40 YS |
1005 | } |
1006 | ||
9682fc6d AE |
1007 | static const char *_rbd_dev_v1_snap_name(struct rbd_device *rbd_dev, u32 which) |
1008 | { | |
1009 | const char *snap_name; | |
1010 | ||
1011 | rbd_assert(which < rbd_dev->header.snapc->num_snaps); | |
1012 | ||
1013 | /* Skip over names until we find the one we are looking for */ | |
1014 | ||
1015 | snap_name = rbd_dev->header.snap_names; | |
1016 | while (which--) | |
1017 | snap_name += strlen(snap_name) + 1; | |
1018 | ||
1019 | return kstrdup(snap_name, GFP_KERNEL); | |
1020 | } | |
1021 | ||
30d1cff8 AE |
1022 | /* |
1023 | * Snapshot id comparison function for use with qsort()/bsearch(). | |
1024 | * Note that result is for snapshots in *descending* order. | |
1025 | */ | |
1026 | static int snapid_compare_reverse(const void *s1, const void *s2) | |
1027 | { | |
1028 | u64 snap_id1 = *(u64 *)s1; | |
1029 | u64 snap_id2 = *(u64 *)s2; | |
1030 | ||
1031 | if (snap_id1 < snap_id2) | |
1032 | return 1; | |
1033 | return snap_id1 == snap_id2 ? 0 : -1; | |
1034 | } | |
1035 | ||
1036 | /* | |
1037 | * Search a snapshot context to see if the given snapshot id is | |
1038 | * present. | |
1039 | * | |
1040 | * Returns the position of the snapshot id in the array if it's found, | |
1041 | * or BAD_SNAP_INDEX otherwise. | |
1042 | * | |
1043 | * Note: The snapshot array is in kept sorted (by the osd) in | |
1044 | * reverse order, highest snapshot id first. | |
1045 | */ | |
9682fc6d AE |
1046 | static u32 rbd_dev_snap_index(struct rbd_device *rbd_dev, u64 snap_id) |
1047 | { | |
1048 | struct ceph_snap_context *snapc = rbd_dev->header.snapc; | |
30d1cff8 | 1049 | u64 *found; |
9682fc6d | 1050 | |
30d1cff8 AE |
1051 | found = bsearch(&snap_id, &snapc->snaps, snapc->num_snaps, |
1052 | sizeof (snap_id), snapid_compare_reverse); | |
9682fc6d | 1053 | |
30d1cff8 | 1054 | return found ? (u32)(found - &snapc->snaps[0]) : BAD_SNAP_INDEX; |
9682fc6d AE |
1055 | } |
1056 | ||
2ad3d716 AE |
1057 | static const char *rbd_dev_v1_snap_name(struct rbd_device *rbd_dev, |
1058 | u64 snap_id) | |
9e15b77d | 1059 | { |
54cac61f | 1060 | u32 which; |
da6a6b63 | 1061 | const char *snap_name; |
9e15b77d | 1062 | |
54cac61f AE |
1063 | which = rbd_dev_snap_index(rbd_dev, snap_id); |
1064 | if (which == BAD_SNAP_INDEX) | |
da6a6b63 | 1065 | return ERR_PTR(-ENOENT); |
54cac61f | 1066 | |
da6a6b63 JD |
1067 | snap_name = _rbd_dev_v1_snap_name(rbd_dev, which); |
1068 | return snap_name ? snap_name : ERR_PTR(-ENOMEM); | |
54cac61f AE |
1069 | } |
1070 | ||
1071 | static const char *rbd_snap_name(struct rbd_device *rbd_dev, u64 snap_id) | |
1072 | { | |
9e15b77d AE |
1073 | if (snap_id == CEPH_NOSNAP) |
1074 | return RBD_SNAP_HEAD_NAME; | |
1075 | ||
54cac61f AE |
1076 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); |
1077 | if (rbd_dev->image_format == 1) | |
1078 | return rbd_dev_v1_snap_name(rbd_dev, snap_id); | |
9e15b77d | 1079 | |
54cac61f | 1080 | return rbd_dev_v2_snap_name(rbd_dev, snap_id); |
9e15b77d AE |
1081 | } |
1082 | ||
2ad3d716 AE |
1083 | static int rbd_snap_size(struct rbd_device *rbd_dev, u64 snap_id, |
1084 | u64 *snap_size) | |
602adf40 | 1085 | { |
2ad3d716 AE |
1086 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); |
1087 | if (snap_id == CEPH_NOSNAP) { | |
1088 | *snap_size = rbd_dev->header.image_size; | |
1089 | } else if (rbd_dev->image_format == 1) { | |
1090 | u32 which; | |
602adf40 | 1091 | |
2ad3d716 AE |
1092 | which = rbd_dev_snap_index(rbd_dev, snap_id); |
1093 | if (which == BAD_SNAP_INDEX) | |
1094 | return -ENOENT; | |
e86924a8 | 1095 | |
2ad3d716 AE |
1096 | *snap_size = rbd_dev->header.snap_sizes[which]; |
1097 | } else { | |
1098 | u64 size = 0; | |
1099 | int ret; | |
1100 | ||
1101 | ret = _rbd_dev_v2_snap_size(rbd_dev, snap_id, NULL, &size); | |
1102 | if (ret) | |
1103 | return ret; | |
1104 | ||
1105 | *snap_size = size; | |
1106 | } | |
1107 | return 0; | |
602adf40 YS |
1108 | } |
1109 | ||
2ad3d716 AE |
1110 | static int rbd_snap_features(struct rbd_device *rbd_dev, u64 snap_id, |
1111 | u64 *snap_features) | |
602adf40 | 1112 | { |
2ad3d716 AE |
1113 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); |
1114 | if (snap_id == CEPH_NOSNAP) { | |
1115 | *snap_features = rbd_dev->header.features; | |
1116 | } else if (rbd_dev->image_format == 1) { | |
1117 | *snap_features = 0; /* No features for format 1 */ | |
602adf40 | 1118 | } else { |
2ad3d716 AE |
1119 | u64 features = 0; |
1120 | int ret; | |
8b0241f8 | 1121 | |
2ad3d716 AE |
1122 | ret = _rbd_dev_v2_snap_features(rbd_dev, snap_id, &features); |
1123 | if (ret) | |
1124 | return ret; | |
1125 | ||
1126 | *snap_features = features; | |
1127 | } | |
1128 | return 0; | |
1129 | } | |
1130 | ||
1131 | static int rbd_dev_mapping_set(struct rbd_device *rbd_dev) | |
1132 | { | |
8f4b7d98 | 1133 | u64 snap_id = rbd_dev->spec->snap_id; |
2ad3d716 AE |
1134 | u64 size = 0; |
1135 | u64 features = 0; | |
1136 | int ret; | |
1137 | ||
2ad3d716 AE |
1138 | ret = rbd_snap_size(rbd_dev, snap_id, &size); |
1139 | if (ret) | |
1140 | return ret; | |
1141 | ret = rbd_snap_features(rbd_dev, snap_id, &features); | |
1142 | if (ret) | |
1143 | return ret; | |
1144 | ||
1145 | rbd_dev->mapping.size = size; | |
1146 | rbd_dev->mapping.features = features; | |
1147 | ||
8b0241f8 | 1148 | return 0; |
602adf40 YS |
1149 | } |
1150 | ||
d1cf5788 AE |
1151 | static void rbd_dev_mapping_clear(struct rbd_device *rbd_dev) |
1152 | { | |
1153 | rbd_dev->mapping.size = 0; | |
1154 | rbd_dev->mapping.features = 0; | |
200a6a8b AE |
1155 | } |
1156 | ||
7d5079aa HS |
1157 | static void rbd_segment_name_free(const char *name) |
1158 | { | |
1159 | /* The explicit cast here is needed to drop the const qualifier */ | |
1160 | ||
1161 | kmem_cache_free(rbd_segment_name_cache, (void *)name); | |
1162 | } | |
1163 | ||
98571b5a | 1164 | static const char *rbd_segment_name(struct rbd_device *rbd_dev, u64 offset) |
602adf40 | 1165 | { |
65ccfe21 AE |
1166 | char *name; |
1167 | u64 segment; | |
1168 | int ret; | |
3a96d5cd | 1169 | char *name_format; |
602adf40 | 1170 | |
78c2a44a | 1171 | name = kmem_cache_alloc(rbd_segment_name_cache, GFP_NOIO); |
65ccfe21 AE |
1172 | if (!name) |
1173 | return NULL; | |
1174 | segment = offset >> rbd_dev->header.obj_order; | |
3a96d5cd JD |
1175 | name_format = "%s.%012llx"; |
1176 | if (rbd_dev->image_format == 2) | |
1177 | name_format = "%s.%016llx"; | |
2d0ebc5d | 1178 | ret = snprintf(name, CEPH_MAX_OID_NAME_LEN + 1, name_format, |
65ccfe21 | 1179 | rbd_dev->header.object_prefix, segment); |
2d0ebc5d | 1180 | if (ret < 0 || ret > CEPH_MAX_OID_NAME_LEN) { |
65ccfe21 AE |
1181 | pr_err("error formatting segment name for #%llu (%d)\n", |
1182 | segment, ret); | |
7d5079aa | 1183 | rbd_segment_name_free(name); |
65ccfe21 AE |
1184 | name = NULL; |
1185 | } | |
602adf40 | 1186 | |
65ccfe21 AE |
1187 | return name; |
1188 | } | |
602adf40 | 1189 | |
65ccfe21 AE |
1190 | static u64 rbd_segment_offset(struct rbd_device *rbd_dev, u64 offset) |
1191 | { | |
1192 | u64 segment_size = (u64) 1 << rbd_dev->header.obj_order; | |
602adf40 | 1193 | |
65ccfe21 AE |
1194 | return offset & (segment_size - 1); |
1195 | } | |
1196 | ||
1197 | static u64 rbd_segment_length(struct rbd_device *rbd_dev, | |
1198 | u64 offset, u64 length) | |
1199 | { | |
1200 | u64 segment_size = (u64) 1 << rbd_dev->header.obj_order; | |
1201 | ||
1202 | offset &= segment_size - 1; | |
1203 | ||
aafb230e | 1204 | rbd_assert(length <= U64_MAX - offset); |
65ccfe21 AE |
1205 | if (offset + length > segment_size) |
1206 | length = segment_size - offset; | |
1207 | ||
1208 | return length; | |
602adf40 YS |
1209 | } |
1210 | ||
029bcbd8 JD |
1211 | /* |
1212 | * returns the size of an object in the image | |
1213 | */ | |
1214 | static u64 rbd_obj_bytes(struct rbd_image_header *header) | |
1215 | { | |
1216 | return 1 << header->obj_order; | |
1217 | } | |
1218 | ||
602adf40 YS |
1219 | /* |
1220 | * bio helpers | |
1221 | */ | |
1222 | ||
1223 | static void bio_chain_put(struct bio *chain) | |
1224 | { | |
1225 | struct bio *tmp; | |
1226 | ||
1227 | while (chain) { | |
1228 | tmp = chain; | |
1229 | chain = chain->bi_next; | |
1230 | bio_put(tmp); | |
1231 | } | |
1232 | } | |
1233 | ||
1234 | /* | |
1235 | * zeros a bio chain, starting at specific offset | |
1236 | */ | |
1237 | static void zero_bio_chain(struct bio *chain, int start_ofs) | |
1238 | { | |
7988613b KO |
1239 | struct bio_vec bv; |
1240 | struct bvec_iter iter; | |
602adf40 YS |
1241 | unsigned long flags; |
1242 | void *buf; | |
602adf40 YS |
1243 | int pos = 0; |
1244 | ||
1245 | while (chain) { | |
7988613b KO |
1246 | bio_for_each_segment(bv, chain, iter) { |
1247 | if (pos + bv.bv_len > start_ofs) { | |
602adf40 | 1248 | int remainder = max(start_ofs - pos, 0); |
7988613b | 1249 | buf = bvec_kmap_irq(&bv, &flags); |
602adf40 | 1250 | memset(buf + remainder, 0, |
7988613b KO |
1251 | bv.bv_len - remainder); |
1252 | flush_dcache_page(bv.bv_page); | |
85b5aaa6 | 1253 | bvec_kunmap_irq(buf, &flags); |
602adf40 | 1254 | } |
7988613b | 1255 | pos += bv.bv_len; |
602adf40 YS |
1256 | } |
1257 | ||
1258 | chain = chain->bi_next; | |
1259 | } | |
1260 | } | |
1261 | ||
b9434c5b AE |
1262 | /* |
1263 | * similar to zero_bio_chain(), zeros data defined by a page array, | |
1264 | * starting at the given byte offset from the start of the array and | |
1265 | * continuing up to the given end offset. The pages array is | |
1266 | * assumed to be big enough to hold all bytes up to the end. | |
1267 | */ | |
1268 | static void zero_pages(struct page **pages, u64 offset, u64 end) | |
1269 | { | |
1270 | struct page **page = &pages[offset >> PAGE_SHIFT]; | |
1271 | ||
1272 | rbd_assert(end > offset); | |
1273 | rbd_assert(end - offset <= (u64)SIZE_MAX); | |
1274 | while (offset < end) { | |
1275 | size_t page_offset; | |
1276 | size_t length; | |
1277 | unsigned long flags; | |
1278 | void *kaddr; | |
1279 | ||
491205a8 GU |
1280 | page_offset = offset & ~PAGE_MASK; |
1281 | length = min_t(size_t, PAGE_SIZE - page_offset, end - offset); | |
b9434c5b AE |
1282 | local_irq_save(flags); |
1283 | kaddr = kmap_atomic(*page); | |
1284 | memset(kaddr + page_offset, 0, length); | |
e2156054 | 1285 | flush_dcache_page(*page); |
b9434c5b AE |
1286 | kunmap_atomic(kaddr); |
1287 | local_irq_restore(flags); | |
1288 | ||
1289 | offset += length; | |
1290 | page++; | |
1291 | } | |
1292 | } | |
1293 | ||
602adf40 | 1294 | /* |
f7760dad AE |
1295 | * Clone a portion of a bio, starting at the given byte offset |
1296 | * and continuing for the number of bytes indicated. | |
602adf40 | 1297 | */ |
f7760dad AE |
1298 | static struct bio *bio_clone_range(struct bio *bio_src, |
1299 | unsigned int offset, | |
1300 | unsigned int len, | |
1301 | gfp_t gfpmask) | |
602adf40 | 1302 | { |
f7760dad AE |
1303 | struct bio *bio; |
1304 | ||
5341a627 | 1305 | bio = bio_clone(bio_src, gfpmask); |
f7760dad AE |
1306 | if (!bio) |
1307 | return NULL; /* ENOMEM */ | |
602adf40 | 1308 | |
5341a627 | 1309 | bio_advance(bio, offset); |
4f024f37 | 1310 | bio->bi_iter.bi_size = len; |
f7760dad AE |
1311 | |
1312 | return bio; | |
1313 | } | |
1314 | ||
1315 | /* | |
1316 | * Clone a portion of a bio chain, starting at the given byte offset | |
1317 | * into the first bio in the source chain and continuing for the | |
1318 | * number of bytes indicated. The result is another bio chain of | |
1319 | * exactly the given length, or a null pointer on error. | |
1320 | * | |
1321 | * The bio_src and offset parameters are both in-out. On entry they | |
1322 | * refer to the first source bio and the offset into that bio where | |
1323 | * the start of data to be cloned is located. | |
1324 | * | |
1325 | * On return, bio_src is updated to refer to the bio in the source | |
1326 | * chain that contains first un-cloned byte, and *offset will | |
1327 | * contain the offset of that byte within that bio. | |
1328 | */ | |
1329 | static struct bio *bio_chain_clone_range(struct bio **bio_src, | |
1330 | unsigned int *offset, | |
1331 | unsigned int len, | |
1332 | gfp_t gfpmask) | |
1333 | { | |
1334 | struct bio *bi = *bio_src; | |
1335 | unsigned int off = *offset; | |
1336 | struct bio *chain = NULL; | |
1337 | struct bio **end; | |
1338 | ||
1339 | /* Build up a chain of clone bios up to the limit */ | |
1340 | ||
4f024f37 | 1341 | if (!bi || off >= bi->bi_iter.bi_size || !len) |
f7760dad | 1342 | return NULL; /* Nothing to clone */ |
602adf40 | 1343 | |
f7760dad AE |
1344 | end = &chain; |
1345 | while (len) { | |
1346 | unsigned int bi_size; | |
1347 | struct bio *bio; | |
1348 | ||
f5400b7a AE |
1349 | if (!bi) { |
1350 | rbd_warn(NULL, "bio_chain exhausted with %u left", len); | |
f7760dad | 1351 | goto out_err; /* EINVAL; ran out of bio's */ |
f5400b7a | 1352 | } |
4f024f37 | 1353 | bi_size = min_t(unsigned int, bi->bi_iter.bi_size - off, len); |
f7760dad AE |
1354 | bio = bio_clone_range(bi, off, bi_size, gfpmask); |
1355 | if (!bio) | |
1356 | goto out_err; /* ENOMEM */ | |
1357 | ||
1358 | *end = bio; | |
1359 | end = &bio->bi_next; | |
602adf40 | 1360 | |
f7760dad | 1361 | off += bi_size; |
4f024f37 | 1362 | if (off == bi->bi_iter.bi_size) { |
f7760dad AE |
1363 | bi = bi->bi_next; |
1364 | off = 0; | |
1365 | } | |
1366 | len -= bi_size; | |
1367 | } | |
1368 | *bio_src = bi; | |
1369 | *offset = off; | |
1370 | ||
1371 | return chain; | |
1372 | out_err: | |
1373 | bio_chain_put(chain); | |
602adf40 | 1374 | |
602adf40 YS |
1375 | return NULL; |
1376 | } | |
1377 | ||
926f9b3f AE |
1378 | /* |
1379 | * The default/initial value for all object request flags is 0. For | |
1380 | * each flag, once its value is set to 1 it is never reset to 0 | |
1381 | * again. | |
1382 | */ | |
57acbaa7 | 1383 | static void obj_request_img_data_set(struct rbd_obj_request *obj_request) |
926f9b3f | 1384 | { |
57acbaa7 | 1385 | if (test_and_set_bit(OBJ_REQ_IMG_DATA, &obj_request->flags)) { |
926f9b3f AE |
1386 | struct rbd_device *rbd_dev; |
1387 | ||
57acbaa7 | 1388 | rbd_dev = obj_request->img_request->rbd_dev; |
9584d508 | 1389 | rbd_warn(rbd_dev, "obj_request %p already marked img_data", |
926f9b3f AE |
1390 | obj_request); |
1391 | } | |
1392 | } | |
1393 | ||
57acbaa7 | 1394 | static bool obj_request_img_data_test(struct rbd_obj_request *obj_request) |
926f9b3f AE |
1395 | { |
1396 | smp_mb(); | |
57acbaa7 | 1397 | return test_bit(OBJ_REQ_IMG_DATA, &obj_request->flags) != 0; |
926f9b3f AE |
1398 | } |
1399 | ||
57acbaa7 | 1400 | static void obj_request_done_set(struct rbd_obj_request *obj_request) |
6365d33a | 1401 | { |
57acbaa7 AE |
1402 | if (test_and_set_bit(OBJ_REQ_DONE, &obj_request->flags)) { |
1403 | struct rbd_device *rbd_dev = NULL; | |
6365d33a | 1404 | |
57acbaa7 AE |
1405 | if (obj_request_img_data_test(obj_request)) |
1406 | rbd_dev = obj_request->img_request->rbd_dev; | |
9584d508 | 1407 | rbd_warn(rbd_dev, "obj_request %p already marked done", |
6365d33a AE |
1408 | obj_request); |
1409 | } | |
1410 | } | |
1411 | ||
57acbaa7 | 1412 | static bool obj_request_done_test(struct rbd_obj_request *obj_request) |
6365d33a AE |
1413 | { |
1414 | smp_mb(); | |
57acbaa7 | 1415 | return test_bit(OBJ_REQ_DONE, &obj_request->flags) != 0; |
6365d33a AE |
1416 | } |
1417 | ||
5679c59f AE |
1418 | /* |
1419 | * This sets the KNOWN flag after (possibly) setting the EXISTS | |
1420 | * flag. The latter is set based on the "exists" value provided. | |
1421 | * | |
1422 | * Note that for our purposes once an object exists it never goes | |
1423 | * away again. It's possible that the response from two existence | |
1424 | * checks are separated by the creation of the target object, and | |
1425 | * the first ("doesn't exist") response arrives *after* the second | |
1426 | * ("does exist"). In that case we ignore the second one. | |
1427 | */ | |
1428 | static void obj_request_existence_set(struct rbd_obj_request *obj_request, | |
1429 | bool exists) | |
1430 | { | |
1431 | if (exists) | |
1432 | set_bit(OBJ_REQ_EXISTS, &obj_request->flags); | |
1433 | set_bit(OBJ_REQ_KNOWN, &obj_request->flags); | |
1434 | smp_mb(); | |
1435 | } | |
1436 | ||
1437 | static bool obj_request_known_test(struct rbd_obj_request *obj_request) | |
1438 | { | |
1439 | smp_mb(); | |
1440 | return test_bit(OBJ_REQ_KNOWN, &obj_request->flags) != 0; | |
1441 | } | |
1442 | ||
1443 | static bool obj_request_exists_test(struct rbd_obj_request *obj_request) | |
1444 | { | |
1445 | smp_mb(); | |
1446 | return test_bit(OBJ_REQ_EXISTS, &obj_request->flags) != 0; | |
1447 | } | |
1448 | ||
9638556a ID |
1449 | static bool obj_request_overlaps_parent(struct rbd_obj_request *obj_request) |
1450 | { | |
1451 | struct rbd_device *rbd_dev = obj_request->img_request->rbd_dev; | |
1452 | ||
1453 | return obj_request->img_offset < | |
1454 | round_up(rbd_dev->parent_overlap, rbd_obj_bytes(&rbd_dev->header)); | |
1455 | } | |
1456 | ||
bf0d5f50 AE |
1457 | static void rbd_obj_request_get(struct rbd_obj_request *obj_request) |
1458 | { | |
37206ee5 AE |
1459 | dout("%s: obj %p (was %d)\n", __func__, obj_request, |
1460 | atomic_read(&obj_request->kref.refcount)); | |
bf0d5f50 AE |
1461 | kref_get(&obj_request->kref); |
1462 | } | |
1463 | ||
1464 | static void rbd_obj_request_destroy(struct kref *kref); | |
1465 | static void rbd_obj_request_put(struct rbd_obj_request *obj_request) | |
1466 | { | |
1467 | rbd_assert(obj_request != NULL); | |
37206ee5 AE |
1468 | dout("%s: obj %p (was %d)\n", __func__, obj_request, |
1469 | atomic_read(&obj_request->kref.refcount)); | |
bf0d5f50 AE |
1470 | kref_put(&obj_request->kref, rbd_obj_request_destroy); |
1471 | } | |
1472 | ||
0f2d5be7 AE |
1473 | static void rbd_img_request_get(struct rbd_img_request *img_request) |
1474 | { | |
1475 | dout("%s: img %p (was %d)\n", __func__, img_request, | |
1476 | atomic_read(&img_request->kref.refcount)); | |
1477 | kref_get(&img_request->kref); | |
1478 | } | |
1479 | ||
e93f3152 AE |
1480 | static bool img_request_child_test(struct rbd_img_request *img_request); |
1481 | static void rbd_parent_request_destroy(struct kref *kref); | |
bf0d5f50 AE |
1482 | static void rbd_img_request_destroy(struct kref *kref); |
1483 | static void rbd_img_request_put(struct rbd_img_request *img_request) | |
1484 | { | |
1485 | rbd_assert(img_request != NULL); | |
37206ee5 AE |
1486 | dout("%s: img %p (was %d)\n", __func__, img_request, |
1487 | atomic_read(&img_request->kref.refcount)); | |
e93f3152 AE |
1488 | if (img_request_child_test(img_request)) |
1489 | kref_put(&img_request->kref, rbd_parent_request_destroy); | |
1490 | else | |
1491 | kref_put(&img_request->kref, rbd_img_request_destroy); | |
bf0d5f50 AE |
1492 | } |
1493 | ||
1494 | static inline void rbd_img_obj_request_add(struct rbd_img_request *img_request, | |
1495 | struct rbd_obj_request *obj_request) | |
1496 | { | |
25dcf954 AE |
1497 | rbd_assert(obj_request->img_request == NULL); |
1498 | ||
b155e86c | 1499 | /* Image request now owns object's original reference */ |
bf0d5f50 | 1500 | obj_request->img_request = img_request; |
25dcf954 | 1501 | obj_request->which = img_request->obj_request_count; |
6365d33a AE |
1502 | rbd_assert(!obj_request_img_data_test(obj_request)); |
1503 | obj_request_img_data_set(obj_request); | |
bf0d5f50 | 1504 | rbd_assert(obj_request->which != BAD_WHICH); |
25dcf954 AE |
1505 | img_request->obj_request_count++; |
1506 | list_add_tail(&obj_request->links, &img_request->obj_requests); | |
37206ee5 AE |
1507 | dout("%s: img %p obj %p w=%u\n", __func__, img_request, obj_request, |
1508 | obj_request->which); | |
bf0d5f50 AE |
1509 | } |
1510 | ||
1511 | static inline void rbd_img_obj_request_del(struct rbd_img_request *img_request, | |
1512 | struct rbd_obj_request *obj_request) | |
1513 | { | |
1514 | rbd_assert(obj_request->which != BAD_WHICH); | |
25dcf954 | 1515 | |
37206ee5 AE |
1516 | dout("%s: img %p obj %p w=%u\n", __func__, img_request, obj_request, |
1517 | obj_request->which); | |
bf0d5f50 | 1518 | list_del(&obj_request->links); |
25dcf954 AE |
1519 | rbd_assert(img_request->obj_request_count > 0); |
1520 | img_request->obj_request_count--; | |
1521 | rbd_assert(obj_request->which == img_request->obj_request_count); | |
1522 | obj_request->which = BAD_WHICH; | |
6365d33a | 1523 | rbd_assert(obj_request_img_data_test(obj_request)); |
bf0d5f50 | 1524 | rbd_assert(obj_request->img_request == img_request); |
bf0d5f50 | 1525 | obj_request->img_request = NULL; |
25dcf954 | 1526 | obj_request->callback = NULL; |
bf0d5f50 AE |
1527 | rbd_obj_request_put(obj_request); |
1528 | } | |
1529 | ||
1530 | static bool obj_request_type_valid(enum obj_request_type type) | |
1531 | { | |
1532 | switch (type) { | |
9969ebc5 | 1533 | case OBJ_REQUEST_NODATA: |
bf0d5f50 | 1534 | case OBJ_REQUEST_BIO: |
788e2df3 | 1535 | case OBJ_REQUEST_PAGES: |
bf0d5f50 AE |
1536 | return true; |
1537 | default: | |
1538 | return false; | |
1539 | } | |
1540 | } | |
1541 | ||
bf0d5f50 AE |
1542 | static int rbd_obj_request_submit(struct ceph_osd_client *osdc, |
1543 | struct rbd_obj_request *obj_request) | |
1544 | { | |
71c20a06 | 1545 | dout("%s %p\n", __func__, obj_request); |
bf0d5f50 AE |
1546 | return ceph_osdc_start_request(osdc, obj_request->osd_req, false); |
1547 | } | |
1548 | ||
71c20a06 ID |
1549 | static void rbd_obj_request_end(struct rbd_obj_request *obj_request) |
1550 | { | |
1551 | dout("%s %p\n", __func__, obj_request); | |
1552 | ceph_osdc_cancel_request(obj_request->osd_req); | |
1553 | } | |
1554 | ||
1555 | /* | |
1556 | * Wait for an object request to complete. If interrupted, cancel the | |
1557 | * underlying osd request. | |
2894e1d7 ID |
1558 | * |
1559 | * @timeout: in jiffies, 0 means "wait forever" | |
71c20a06 | 1560 | */ |
2894e1d7 ID |
1561 | static int __rbd_obj_request_wait(struct rbd_obj_request *obj_request, |
1562 | unsigned long timeout) | |
71c20a06 | 1563 | { |
2894e1d7 | 1564 | long ret; |
71c20a06 ID |
1565 | |
1566 | dout("%s %p\n", __func__, obj_request); | |
2894e1d7 ID |
1567 | ret = wait_for_completion_interruptible_timeout( |
1568 | &obj_request->completion, | |
1569 | ceph_timeout_jiffies(timeout)); | |
1570 | if (ret <= 0) { | |
1571 | if (ret == 0) | |
1572 | ret = -ETIMEDOUT; | |
71c20a06 | 1573 | rbd_obj_request_end(obj_request); |
2894e1d7 ID |
1574 | } else { |
1575 | ret = 0; | |
71c20a06 ID |
1576 | } |
1577 | ||
2894e1d7 ID |
1578 | dout("%s %p ret %d\n", __func__, obj_request, (int)ret); |
1579 | return ret; | |
1580 | } | |
1581 | ||
1582 | static int rbd_obj_request_wait(struct rbd_obj_request *obj_request) | |
1583 | { | |
1584 | return __rbd_obj_request_wait(obj_request, 0); | |
1585 | } | |
1586 | ||
1587 | static int rbd_obj_request_wait_timeout(struct rbd_obj_request *obj_request, | |
1588 | unsigned long timeout) | |
1589 | { | |
1590 | return __rbd_obj_request_wait(obj_request, timeout); | |
71c20a06 ID |
1591 | } |
1592 | ||
bf0d5f50 AE |
1593 | static void rbd_img_request_complete(struct rbd_img_request *img_request) |
1594 | { | |
55f27e09 | 1595 | |
37206ee5 | 1596 | dout("%s: img %p\n", __func__, img_request); |
55f27e09 AE |
1597 | |
1598 | /* | |
1599 | * If no error occurred, compute the aggregate transfer | |
1600 | * count for the image request. We could instead use | |
1601 | * atomic64_cmpxchg() to update it as each object request | |
1602 | * completes; not clear which way is better off hand. | |
1603 | */ | |
1604 | if (!img_request->result) { | |
1605 | struct rbd_obj_request *obj_request; | |
1606 | u64 xferred = 0; | |
1607 | ||
1608 | for_each_obj_request(img_request, obj_request) | |
1609 | xferred += obj_request->xferred; | |
1610 | img_request->xferred = xferred; | |
1611 | } | |
1612 | ||
bf0d5f50 AE |
1613 | if (img_request->callback) |
1614 | img_request->callback(img_request); | |
1615 | else | |
1616 | rbd_img_request_put(img_request); | |
1617 | } | |
1618 | ||
0c425248 AE |
1619 | /* |
1620 | * The default/initial value for all image request flags is 0. Each | |
1621 | * is conditionally set to 1 at image request initialization time | |
1622 | * and currently never change thereafter. | |
1623 | */ | |
1624 | static void img_request_write_set(struct rbd_img_request *img_request) | |
1625 | { | |
1626 | set_bit(IMG_REQ_WRITE, &img_request->flags); | |
1627 | smp_mb(); | |
1628 | } | |
1629 | ||
1630 | static bool img_request_write_test(struct rbd_img_request *img_request) | |
1631 | { | |
1632 | smp_mb(); | |
1633 | return test_bit(IMG_REQ_WRITE, &img_request->flags) != 0; | |
1634 | } | |
1635 | ||
90e98c52 GZ |
1636 | /* |
1637 | * Set the discard flag when the img_request is an discard request | |
1638 | */ | |
1639 | static void img_request_discard_set(struct rbd_img_request *img_request) | |
1640 | { | |
1641 | set_bit(IMG_REQ_DISCARD, &img_request->flags); | |
1642 | smp_mb(); | |
1643 | } | |
1644 | ||
1645 | static bool img_request_discard_test(struct rbd_img_request *img_request) | |
1646 | { | |
1647 | smp_mb(); | |
1648 | return test_bit(IMG_REQ_DISCARD, &img_request->flags) != 0; | |
1649 | } | |
1650 | ||
9849e986 AE |
1651 | static void img_request_child_set(struct rbd_img_request *img_request) |
1652 | { | |
1653 | set_bit(IMG_REQ_CHILD, &img_request->flags); | |
1654 | smp_mb(); | |
1655 | } | |
1656 | ||
e93f3152 AE |
1657 | static void img_request_child_clear(struct rbd_img_request *img_request) |
1658 | { | |
1659 | clear_bit(IMG_REQ_CHILD, &img_request->flags); | |
1660 | smp_mb(); | |
1661 | } | |
1662 | ||
9849e986 AE |
1663 | static bool img_request_child_test(struct rbd_img_request *img_request) |
1664 | { | |
1665 | smp_mb(); | |
1666 | return test_bit(IMG_REQ_CHILD, &img_request->flags) != 0; | |
1667 | } | |
1668 | ||
d0b2e944 AE |
1669 | static void img_request_layered_set(struct rbd_img_request *img_request) |
1670 | { | |
1671 | set_bit(IMG_REQ_LAYERED, &img_request->flags); | |
1672 | smp_mb(); | |
1673 | } | |
1674 | ||
a2acd00e AE |
1675 | static void img_request_layered_clear(struct rbd_img_request *img_request) |
1676 | { | |
1677 | clear_bit(IMG_REQ_LAYERED, &img_request->flags); | |
1678 | smp_mb(); | |
1679 | } | |
1680 | ||
d0b2e944 AE |
1681 | static bool img_request_layered_test(struct rbd_img_request *img_request) |
1682 | { | |
1683 | smp_mb(); | |
1684 | return test_bit(IMG_REQ_LAYERED, &img_request->flags) != 0; | |
1685 | } | |
1686 | ||
3b434a2a JD |
1687 | static enum obj_operation_type |
1688 | rbd_img_request_op_type(struct rbd_img_request *img_request) | |
1689 | { | |
1690 | if (img_request_write_test(img_request)) | |
1691 | return OBJ_OP_WRITE; | |
1692 | else if (img_request_discard_test(img_request)) | |
1693 | return OBJ_OP_DISCARD; | |
1694 | else | |
1695 | return OBJ_OP_READ; | |
1696 | } | |
1697 | ||
6e2a4505 AE |
1698 | static void |
1699 | rbd_img_obj_request_read_callback(struct rbd_obj_request *obj_request) | |
1700 | { | |
b9434c5b AE |
1701 | u64 xferred = obj_request->xferred; |
1702 | u64 length = obj_request->length; | |
1703 | ||
6e2a4505 AE |
1704 | dout("%s: obj %p img %p result %d %llu/%llu\n", __func__, |
1705 | obj_request, obj_request->img_request, obj_request->result, | |
b9434c5b | 1706 | xferred, length); |
6e2a4505 | 1707 | /* |
17c1cc1d JD |
1708 | * ENOENT means a hole in the image. We zero-fill the entire |
1709 | * length of the request. A short read also implies zero-fill | |
1710 | * to the end of the request. An error requires the whole | |
1711 | * length of the request to be reported finished with an error | |
1712 | * to the block layer. In each case we update the xferred | |
1713 | * count to indicate the whole request was satisfied. | |
6e2a4505 | 1714 | */ |
b9434c5b | 1715 | rbd_assert(obj_request->type != OBJ_REQUEST_NODATA); |
6e2a4505 | 1716 | if (obj_request->result == -ENOENT) { |
b9434c5b AE |
1717 | if (obj_request->type == OBJ_REQUEST_BIO) |
1718 | zero_bio_chain(obj_request->bio_list, 0); | |
1719 | else | |
1720 | zero_pages(obj_request->pages, 0, length); | |
6e2a4505 | 1721 | obj_request->result = 0; |
b9434c5b AE |
1722 | } else if (xferred < length && !obj_request->result) { |
1723 | if (obj_request->type == OBJ_REQUEST_BIO) | |
1724 | zero_bio_chain(obj_request->bio_list, xferred); | |
1725 | else | |
1726 | zero_pages(obj_request->pages, xferred, length); | |
6e2a4505 | 1727 | } |
17c1cc1d | 1728 | obj_request->xferred = length; |
6e2a4505 AE |
1729 | obj_request_done_set(obj_request); |
1730 | } | |
1731 | ||
bf0d5f50 AE |
1732 | static void rbd_obj_request_complete(struct rbd_obj_request *obj_request) |
1733 | { | |
37206ee5 AE |
1734 | dout("%s: obj %p cb %p\n", __func__, obj_request, |
1735 | obj_request->callback); | |
bf0d5f50 AE |
1736 | if (obj_request->callback) |
1737 | obj_request->callback(obj_request); | |
788e2df3 AE |
1738 | else |
1739 | complete_all(&obj_request->completion); | |
bf0d5f50 AE |
1740 | } |
1741 | ||
c47f9371 | 1742 | static void rbd_osd_trivial_callback(struct rbd_obj_request *obj_request) |
39bf2c5d AE |
1743 | { |
1744 | dout("%s: obj %p\n", __func__, obj_request); | |
1745 | obj_request_done_set(obj_request); | |
1746 | } | |
1747 | ||
c47f9371 | 1748 | static void rbd_osd_read_callback(struct rbd_obj_request *obj_request) |
bf0d5f50 | 1749 | { |
57acbaa7 | 1750 | struct rbd_img_request *img_request = NULL; |
a9e8ba2c | 1751 | struct rbd_device *rbd_dev = NULL; |
57acbaa7 AE |
1752 | bool layered = false; |
1753 | ||
1754 | if (obj_request_img_data_test(obj_request)) { | |
1755 | img_request = obj_request->img_request; | |
1756 | layered = img_request && img_request_layered_test(img_request); | |
a9e8ba2c | 1757 | rbd_dev = img_request->rbd_dev; |
57acbaa7 | 1758 | } |
8b3e1a56 AE |
1759 | |
1760 | dout("%s: obj %p img %p result %d %llu/%llu\n", __func__, | |
1761 | obj_request, img_request, obj_request->result, | |
1762 | obj_request->xferred, obj_request->length); | |
a9e8ba2c AE |
1763 | if (layered && obj_request->result == -ENOENT && |
1764 | obj_request->img_offset < rbd_dev->parent_overlap) | |
8b3e1a56 AE |
1765 | rbd_img_parent_read(obj_request); |
1766 | else if (img_request) | |
6e2a4505 AE |
1767 | rbd_img_obj_request_read_callback(obj_request); |
1768 | else | |
1769 | obj_request_done_set(obj_request); | |
bf0d5f50 AE |
1770 | } |
1771 | ||
c47f9371 | 1772 | static void rbd_osd_write_callback(struct rbd_obj_request *obj_request) |
bf0d5f50 | 1773 | { |
1b83bef2 SW |
1774 | dout("%s: obj %p result %d %llu\n", __func__, obj_request, |
1775 | obj_request->result, obj_request->length); | |
1776 | /* | |
8b3e1a56 AE |
1777 | * There is no such thing as a successful short write. Set |
1778 | * it to our originally-requested length. | |
1b83bef2 SW |
1779 | */ |
1780 | obj_request->xferred = obj_request->length; | |
07741308 | 1781 | obj_request_done_set(obj_request); |
bf0d5f50 AE |
1782 | } |
1783 | ||
90e98c52 GZ |
1784 | static void rbd_osd_discard_callback(struct rbd_obj_request *obj_request) |
1785 | { | |
1786 | dout("%s: obj %p result %d %llu\n", __func__, obj_request, | |
1787 | obj_request->result, obj_request->length); | |
1788 | /* | |
1789 | * There is no such thing as a successful short discard. Set | |
1790 | * it to our originally-requested length. | |
1791 | */ | |
1792 | obj_request->xferred = obj_request->length; | |
d0265de7 JD |
1793 | /* discarding a non-existent object is not a problem */ |
1794 | if (obj_request->result == -ENOENT) | |
1795 | obj_request->result = 0; | |
90e98c52 GZ |
1796 | obj_request_done_set(obj_request); |
1797 | } | |
1798 | ||
fbfab539 AE |
1799 | /* |
1800 | * For a simple stat call there's nothing to do. We'll do more if | |
1801 | * this is part of a write sequence for a layered image. | |
1802 | */ | |
c47f9371 | 1803 | static void rbd_osd_stat_callback(struct rbd_obj_request *obj_request) |
fbfab539 | 1804 | { |
37206ee5 | 1805 | dout("%s: obj %p\n", __func__, obj_request); |
fbfab539 AE |
1806 | obj_request_done_set(obj_request); |
1807 | } | |
1808 | ||
bf0d5f50 AE |
1809 | static void rbd_osd_req_callback(struct ceph_osd_request *osd_req, |
1810 | struct ceph_msg *msg) | |
1811 | { | |
1812 | struct rbd_obj_request *obj_request = osd_req->r_priv; | |
bf0d5f50 AE |
1813 | u16 opcode; |
1814 | ||
37206ee5 | 1815 | dout("%s: osd_req %p msg %p\n", __func__, osd_req, msg); |
bf0d5f50 | 1816 | rbd_assert(osd_req == obj_request->osd_req); |
57acbaa7 AE |
1817 | if (obj_request_img_data_test(obj_request)) { |
1818 | rbd_assert(obj_request->img_request); | |
1819 | rbd_assert(obj_request->which != BAD_WHICH); | |
1820 | } else { | |
1821 | rbd_assert(obj_request->which == BAD_WHICH); | |
1822 | } | |
bf0d5f50 | 1823 | |
1b83bef2 SW |
1824 | if (osd_req->r_result < 0) |
1825 | obj_request->result = osd_req->r_result; | |
bf0d5f50 | 1826 | |
7cc69d42 | 1827 | rbd_assert(osd_req->r_num_ops <= CEPH_OSD_MAX_OP); |
bf0d5f50 | 1828 | |
c47f9371 AE |
1829 | /* |
1830 | * We support a 64-bit length, but ultimately it has to be | |
7ad18afa CH |
1831 | * passed to the block layer, which just supports a 32-bit |
1832 | * length field. | |
c47f9371 | 1833 | */ |
1b83bef2 | 1834 | obj_request->xferred = osd_req->r_reply_op_len[0]; |
8b3e1a56 | 1835 | rbd_assert(obj_request->xferred < (u64)UINT_MAX); |
0ccd5926 | 1836 | |
79528734 | 1837 | opcode = osd_req->r_ops[0].op; |
bf0d5f50 AE |
1838 | switch (opcode) { |
1839 | case CEPH_OSD_OP_READ: | |
c47f9371 | 1840 | rbd_osd_read_callback(obj_request); |
bf0d5f50 | 1841 | break; |
0ccd5926 ID |
1842 | case CEPH_OSD_OP_SETALLOCHINT: |
1843 | rbd_assert(osd_req->r_ops[1].op == CEPH_OSD_OP_WRITE); | |
1844 | /* fall through */ | |
bf0d5f50 | 1845 | case CEPH_OSD_OP_WRITE: |
c47f9371 | 1846 | rbd_osd_write_callback(obj_request); |
bf0d5f50 | 1847 | break; |
fbfab539 | 1848 | case CEPH_OSD_OP_STAT: |
c47f9371 | 1849 | rbd_osd_stat_callback(obj_request); |
fbfab539 | 1850 | break; |
90e98c52 GZ |
1851 | case CEPH_OSD_OP_DELETE: |
1852 | case CEPH_OSD_OP_TRUNCATE: | |
1853 | case CEPH_OSD_OP_ZERO: | |
1854 | rbd_osd_discard_callback(obj_request); | |
1855 | break; | |
36be9a76 | 1856 | case CEPH_OSD_OP_CALL: |
b8d70035 | 1857 | case CEPH_OSD_OP_NOTIFY_ACK: |
9969ebc5 | 1858 | case CEPH_OSD_OP_WATCH: |
c47f9371 | 1859 | rbd_osd_trivial_callback(obj_request); |
9969ebc5 | 1860 | break; |
bf0d5f50 | 1861 | default: |
9584d508 | 1862 | rbd_warn(NULL, "%s: unsupported op %hu", |
bf0d5f50 AE |
1863 | obj_request->object_name, (unsigned short) opcode); |
1864 | break; | |
1865 | } | |
1866 | ||
07741308 | 1867 | if (obj_request_done_test(obj_request)) |
bf0d5f50 AE |
1868 | rbd_obj_request_complete(obj_request); |
1869 | } | |
1870 | ||
9d4df01f | 1871 | static void rbd_osd_req_format_read(struct rbd_obj_request *obj_request) |
430c28c3 AE |
1872 | { |
1873 | struct rbd_img_request *img_request = obj_request->img_request; | |
8c042b0d | 1874 | struct ceph_osd_request *osd_req = obj_request->osd_req; |
9d4df01f | 1875 | u64 snap_id; |
430c28c3 | 1876 | |
8c042b0d | 1877 | rbd_assert(osd_req != NULL); |
430c28c3 | 1878 | |
9d4df01f | 1879 | snap_id = img_request ? img_request->snap_id : CEPH_NOSNAP; |
8c042b0d | 1880 | ceph_osdc_build_request(osd_req, obj_request->offset, |
9d4df01f AE |
1881 | NULL, snap_id, NULL); |
1882 | } | |
1883 | ||
1884 | static void rbd_osd_req_format_write(struct rbd_obj_request *obj_request) | |
1885 | { | |
1886 | struct rbd_img_request *img_request = obj_request->img_request; | |
1887 | struct ceph_osd_request *osd_req = obj_request->osd_req; | |
1888 | struct ceph_snap_context *snapc; | |
1889 | struct timespec mtime = CURRENT_TIME; | |
1890 | ||
1891 | rbd_assert(osd_req != NULL); | |
1892 | ||
1893 | snapc = img_request ? img_request->snapc : NULL; | |
1894 | ceph_osdc_build_request(osd_req, obj_request->offset, | |
1895 | snapc, CEPH_NOSNAP, &mtime); | |
430c28c3 AE |
1896 | } |
1897 | ||
0ccd5926 ID |
1898 | /* |
1899 | * Create an osd request. A read request has one osd op (read). | |
1900 | * A write request has either one (watch) or two (hint+write) osd ops. | |
1901 | * (All rbd data writes are prefixed with an allocation hint op, but | |
1902 | * technically osd watch is a write request, hence this distinction.) | |
1903 | */ | |
bf0d5f50 AE |
1904 | static struct ceph_osd_request *rbd_osd_req_create( |
1905 | struct rbd_device *rbd_dev, | |
6d2940c8 | 1906 | enum obj_operation_type op_type, |
deb236b3 | 1907 | unsigned int num_ops, |
430c28c3 | 1908 | struct rbd_obj_request *obj_request) |
bf0d5f50 | 1909 | { |
bf0d5f50 AE |
1910 | struct ceph_snap_context *snapc = NULL; |
1911 | struct ceph_osd_client *osdc; | |
1912 | struct ceph_osd_request *osd_req; | |
bf0d5f50 | 1913 | |
90e98c52 GZ |
1914 | if (obj_request_img_data_test(obj_request) && |
1915 | (op_type == OBJ_OP_DISCARD || op_type == OBJ_OP_WRITE)) { | |
6365d33a | 1916 | struct rbd_img_request *img_request = obj_request->img_request; |
90e98c52 GZ |
1917 | if (op_type == OBJ_OP_WRITE) { |
1918 | rbd_assert(img_request_write_test(img_request)); | |
1919 | } else { | |
1920 | rbd_assert(img_request_discard_test(img_request)); | |
1921 | } | |
6d2940c8 | 1922 | snapc = img_request->snapc; |
bf0d5f50 AE |
1923 | } |
1924 | ||
6d2940c8 | 1925 | rbd_assert(num_ops == 1 || ((op_type == OBJ_OP_WRITE) && num_ops == 2)); |
deb236b3 ID |
1926 | |
1927 | /* Allocate and initialize the request, for the num_ops ops */ | |
bf0d5f50 AE |
1928 | |
1929 | osdc = &rbd_dev->rbd_client->client->osdc; | |
deb236b3 ID |
1930 | osd_req = ceph_osdc_alloc_request(osdc, snapc, num_ops, false, |
1931 | GFP_ATOMIC); | |
bf0d5f50 AE |
1932 | if (!osd_req) |
1933 | return NULL; /* ENOMEM */ | |
bf0d5f50 | 1934 | |
90e98c52 | 1935 | if (op_type == OBJ_OP_WRITE || op_type == OBJ_OP_DISCARD) |
bf0d5f50 | 1936 | osd_req->r_flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK; |
430c28c3 | 1937 | else |
bf0d5f50 | 1938 | osd_req->r_flags = CEPH_OSD_FLAG_READ; |
bf0d5f50 AE |
1939 | |
1940 | osd_req->r_callback = rbd_osd_req_callback; | |
1941 | osd_req->r_priv = obj_request; | |
1942 | ||
3c972c95 ID |
1943 | osd_req->r_base_oloc.pool = ceph_file_layout_pg_pool(rbd_dev->layout); |
1944 | ceph_oid_set_name(&osd_req->r_base_oid, obj_request->object_name); | |
bf0d5f50 | 1945 | |
bf0d5f50 AE |
1946 | return osd_req; |
1947 | } | |
1948 | ||
0eefd470 | 1949 | /* |
d3246fb0 JD |
1950 | * Create a copyup osd request based on the information in the object |
1951 | * request supplied. A copyup request has two or three osd ops, a | |
1952 | * copyup method call, potentially a hint op, and a write or truncate | |
1953 | * or zero op. | |
0eefd470 AE |
1954 | */ |
1955 | static struct ceph_osd_request * | |
1956 | rbd_osd_req_create_copyup(struct rbd_obj_request *obj_request) | |
1957 | { | |
1958 | struct rbd_img_request *img_request; | |
1959 | struct ceph_snap_context *snapc; | |
1960 | struct rbd_device *rbd_dev; | |
1961 | struct ceph_osd_client *osdc; | |
1962 | struct ceph_osd_request *osd_req; | |
d3246fb0 | 1963 | int num_osd_ops = 3; |
0eefd470 AE |
1964 | |
1965 | rbd_assert(obj_request_img_data_test(obj_request)); | |
1966 | img_request = obj_request->img_request; | |
1967 | rbd_assert(img_request); | |
d3246fb0 JD |
1968 | rbd_assert(img_request_write_test(img_request) || |
1969 | img_request_discard_test(img_request)); | |
0eefd470 | 1970 | |
d3246fb0 JD |
1971 | if (img_request_discard_test(img_request)) |
1972 | num_osd_ops = 2; | |
1973 | ||
1974 | /* Allocate and initialize the request, for all the ops */ | |
0eefd470 AE |
1975 | |
1976 | snapc = img_request->snapc; | |
1977 | rbd_dev = img_request->rbd_dev; | |
1978 | osdc = &rbd_dev->rbd_client->client->osdc; | |
d3246fb0 JD |
1979 | osd_req = ceph_osdc_alloc_request(osdc, snapc, num_osd_ops, |
1980 | false, GFP_ATOMIC); | |
0eefd470 AE |
1981 | if (!osd_req) |
1982 | return NULL; /* ENOMEM */ | |
1983 | ||
1984 | osd_req->r_flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK; | |
1985 | osd_req->r_callback = rbd_osd_req_callback; | |
1986 | osd_req->r_priv = obj_request; | |
1987 | ||
3c972c95 ID |
1988 | osd_req->r_base_oloc.pool = ceph_file_layout_pg_pool(rbd_dev->layout); |
1989 | ceph_oid_set_name(&osd_req->r_base_oid, obj_request->object_name); | |
0eefd470 | 1990 | |
0eefd470 AE |
1991 | return osd_req; |
1992 | } | |
1993 | ||
1994 | ||
bf0d5f50 AE |
1995 | static void rbd_osd_req_destroy(struct ceph_osd_request *osd_req) |
1996 | { | |
1997 | ceph_osdc_put_request(osd_req); | |
1998 | } | |
1999 | ||
2000 | /* object_name is assumed to be a non-null pointer and NUL-terminated */ | |
2001 | ||
2002 | static struct rbd_obj_request *rbd_obj_request_create(const char *object_name, | |
2003 | u64 offset, u64 length, | |
2004 | enum obj_request_type type) | |
2005 | { | |
2006 | struct rbd_obj_request *obj_request; | |
2007 | size_t size; | |
2008 | char *name; | |
2009 | ||
2010 | rbd_assert(obj_request_type_valid(type)); | |
2011 | ||
2012 | size = strlen(object_name) + 1; | |
f907ad55 AE |
2013 | name = kmalloc(size, GFP_KERNEL); |
2014 | if (!name) | |
bf0d5f50 AE |
2015 | return NULL; |
2016 | ||
868311b1 | 2017 | obj_request = kmem_cache_zalloc(rbd_obj_request_cache, GFP_KERNEL); |
f907ad55 AE |
2018 | if (!obj_request) { |
2019 | kfree(name); | |
2020 | return NULL; | |
2021 | } | |
2022 | ||
bf0d5f50 AE |
2023 | obj_request->object_name = memcpy(name, object_name, size); |
2024 | obj_request->offset = offset; | |
2025 | obj_request->length = length; | |
926f9b3f | 2026 | obj_request->flags = 0; |
bf0d5f50 AE |
2027 | obj_request->which = BAD_WHICH; |
2028 | obj_request->type = type; | |
2029 | INIT_LIST_HEAD(&obj_request->links); | |
788e2df3 | 2030 | init_completion(&obj_request->completion); |
bf0d5f50 AE |
2031 | kref_init(&obj_request->kref); |
2032 | ||
37206ee5 AE |
2033 | dout("%s: \"%s\" %llu/%llu %d -> obj %p\n", __func__, object_name, |
2034 | offset, length, (int)type, obj_request); | |
2035 | ||
bf0d5f50 AE |
2036 | return obj_request; |
2037 | } | |
2038 | ||
2039 | static void rbd_obj_request_destroy(struct kref *kref) | |
2040 | { | |
2041 | struct rbd_obj_request *obj_request; | |
2042 | ||
2043 | obj_request = container_of(kref, struct rbd_obj_request, kref); | |
2044 | ||
37206ee5 AE |
2045 | dout("%s: obj %p\n", __func__, obj_request); |
2046 | ||
bf0d5f50 AE |
2047 | rbd_assert(obj_request->img_request == NULL); |
2048 | rbd_assert(obj_request->which == BAD_WHICH); | |
2049 | ||
2050 | if (obj_request->osd_req) | |
2051 | rbd_osd_req_destroy(obj_request->osd_req); | |
2052 | ||
2053 | rbd_assert(obj_request_type_valid(obj_request->type)); | |
2054 | switch (obj_request->type) { | |
9969ebc5 AE |
2055 | case OBJ_REQUEST_NODATA: |
2056 | break; /* Nothing to do */ | |
bf0d5f50 AE |
2057 | case OBJ_REQUEST_BIO: |
2058 | if (obj_request->bio_list) | |
2059 | bio_chain_put(obj_request->bio_list); | |
2060 | break; | |
788e2df3 AE |
2061 | case OBJ_REQUEST_PAGES: |
2062 | if (obj_request->pages) | |
2063 | ceph_release_page_vector(obj_request->pages, | |
2064 | obj_request->page_count); | |
2065 | break; | |
bf0d5f50 AE |
2066 | } |
2067 | ||
f907ad55 | 2068 | kfree(obj_request->object_name); |
868311b1 AE |
2069 | obj_request->object_name = NULL; |
2070 | kmem_cache_free(rbd_obj_request_cache, obj_request); | |
bf0d5f50 AE |
2071 | } |
2072 | ||
fb65d228 AE |
2073 | /* It's OK to call this for a device with no parent */ |
2074 | ||
2075 | static void rbd_spec_put(struct rbd_spec *spec); | |
2076 | static void rbd_dev_unparent(struct rbd_device *rbd_dev) | |
2077 | { | |
2078 | rbd_dev_remove_parent(rbd_dev); | |
2079 | rbd_spec_put(rbd_dev->parent_spec); | |
2080 | rbd_dev->parent_spec = NULL; | |
2081 | rbd_dev->parent_overlap = 0; | |
2082 | } | |
2083 | ||
a2acd00e AE |
2084 | /* |
2085 | * Parent image reference counting is used to determine when an | |
2086 | * image's parent fields can be safely torn down--after there are no | |
2087 | * more in-flight requests to the parent image. When the last | |
2088 | * reference is dropped, cleaning them up is safe. | |
2089 | */ | |
2090 | static void rbd_dev_parent_put(struct rbd_device *rbd_dev) | |
2091 | { | |
2092 | int counter; | |
2093 | ||
2094 | if (!rbd_dev->parent_spec) | |
2095 | return; | |
2096 | ||
2097 | counter = atomic_dec_return_safe(&rbd_dev->parent_ref); | |
2098 | if (counter > 0) | |
2099 | return; | |
2100 | ||
2101 | /* Last reference; clean up parent data structures */ | |
2102 | ||
2103 | if (!counter) | |
2104 | rbd_dev_unparent(rbd_dev); | |
2105 | else | |
9584d508 | 2106 | rbd_warn(rbd_dev, "parent reference underflow"); |
a2acd00e AE |
2107 | } |
2108 | ||
2109 | /* | |
2110 | * If an image has a non-zero parent overlap, get a reference to its | |
2111 | * parent. | |
2112 | * | |
2113 | * Returns true if the rbd device has a parent with a non-zero | |
2114 | * overlap and a reference for it was successfully taken, or | |
2115 | * false otherwise. | |
2116 | */ | |
2117 | static bool rbd_dev_parent_get(struct rbd_device *rbd_dev) | |
2118 | { | |
ae43e9d0 | 2119 | int counter = 0; |
a2acd00e AE |
2120 | |
2121 | if (!rbd_dev->parent_spec) | |
2122 | return false; | |
2123 | ||
ae43e9d0 ID |
2124 | down_read(&rbd_dev->header_rwsem); |
2125 | if (rbd_dev->parent_overlap) | |
2126 | counter = atomic_inc_return_safe(&rbd_dev->parent_ref); | |
2127 | up_read(&rbd_dev->header_rwsem); | |
a2acd00e AE |
2128 | |
2129 | if (counter < 0) | |
9584d508 | 2130 | rbd_warn(rbd_dev, "parent reference overflow"); |
a2acd00e | 2131 | |
ae43e9d0 | 2132 | return counter > 0; |
a2acd00e AE |
2133 | } |
2134 | ||
bf0d5f50 AE |
2135 | /* |
2136 | * Caller is responsible for filling in the list of object requests | |
2137 | * that comprises the image request, and the Linux request pointer | |
2138 | * (if there is one). | |
2139 | */ | |
cc344fa1 AE |
2140 | static struct rbd_img_request *rbd_img_request_create( |
2141 | struct rbd_device *rbd_dev, | |
bf0d5f50 | 2142 | u64 offset, u64 length, |
6d2940c8 | 2143 | enum obj_operation_type op_type, |
4e752f0a | 2144 | struct ceph_snap_context *snapc) |
bf0d5f50 AE |
2145 | { |
2146 | struct rbd_img_request *img_request; | |
bf0d5f50 | 2147 | |
7a716aac | 2148 | img_request = kmem_cache_alloc(rbd_img_request_cache, GFP_NOIO); |
bf0d5f50 AE |
2149 | if (!img_request) |
2150 | return NULL; | |
2151 | ||
bf0d5f50 AE |
2152 | img_request->rq = NULL; |
2153 | img_request->rbd_dev = rbd_dev; | |
2154 | img_request->offset = offset; | |
2155 | img_request->length = length; | |
0c425248 | 2156 | img_request->flags = 0; |
90e98c52 GZ |
2157 | if (op_type == OBJ_OP_DISCARD) { |
2158 | img_request_discard_set(img_request); | |
2159 | img_request->snapc = snapc; | |
2160 | } else if (op_type == OBJ_OP_WRITE) { | |
0c425248 | 2161 | img_request_write_set(img_request); |
4e752f0a | 2162 | img_request->snapc = snapc; |
0c425248 | 2163 | } else { |
bf0d5f50 | 2164 | img_request->snap_id = rbd_dev->spec->snap_id; |
0c425248 | 2165 | } |
a2acd00e | 2166 | if (rbd_dev_parent_get(rbd_dev)) |
d0b2e944 | 2167 | img_request_layered_set(img_request); |
bf0d5f50 AE |
2168 | spin_lock_init(&img_request->completion_lock); |
2169 | img_request->next_completion = 0; | |
2170 | img_request->callback = NULL; | |
a5a337d4 | 2171 | img_request->result = 0; |
bf0d5f50 AE |
2172 | img_request->obj_request_count = 0; |
2173 | INIT_LIST_HEAD(&img_request->obj_requests); | |
2174 | kref_init(&img_request->kref); | |
2175 | ||
37206ee5 | 2176 | dout("%s: rbd_dev %p %s %llu/%llu -> img %p\n", __func__, rbd_dev, |
6d2940c8 | 2177 | obj_op_name(op_type), offset, length, img_request); |
37206ee5 | 2178 | |
bf0d5f50 AE |
2179 | return img_request; |
2180 | } | |
2181 | ||
2182 | static void rbd_img_request_destroy(struct kref *kref) | |
2183 | { | |
2184 | struct rbd_img_request *img_request; | |
2185 | struct rbd_obj_request *obj_request; | |
2186 | struct rbd_obj_request *next_obj_request; | |
2187 | ||
2188 | img_request = container_of(kref, struct rbd_img_request, kref); | |
2189 | ||
37206ee5 AE |
2190 | dout("%s: img %p\n", __func__, img_request); |
2191 | ||
bf0d5f50 AE |
2192 | for_each_obj_request_safe(img_request, obj_request, next_obj_request) |
2193 | rbd_img_obj_request_del(img_request, obj_request); | |
25dcf954 | 2194 | rbd_assert(img_request->obj_request_count == 0); |
bf0d5f50 | 2195 | |
a2acd00e AE |
2196 | if (img_request_layered_test(img_request)) { |
2197 | img_request_layered_clear(img_request); | |
2198 | rbd_dev_parent_put(img_request->rbd_dev); | |
2199 | } | |
2200 | ||
bef95455 JD |
2201 | if (img_request_write_test(img_request) || |
2202 | img_request_discard_test(img_request)) | |
812164f8 | 2203 | ceph_put_snap_context(img_request->snapc); |
bf0d5f50 | 2204 | |
1c2a9dfe | 2205 | kmem_cache_free(rbd_img_request_cache, img_request); |
bf0d5f50 AE |
2206 | } |
2207 | ||
e93f3152 AE |
2208 | static struct rbd_img_request *rbd_parent_request_create( |
2209 | struct rbd_obj_request *obj_request, | |
2210 | u64 img_offset, u64 length) | |
2211 | { | |
2212 | struct rbd_img_request *parent_request; | |
2213 | struct rbd_device *rbd_dev; | |
2214 | ||
2215 | rbd_assert(obj_request->img_request); | |
2216 | rbd_dev = obj_request->img_request->rbd_dev; | |
2217 | ||
4e752f0a | 2218 | parent_request = rbd_img_request_create(rbd_dev->parent, img_offset, |
6d2940c8 | 2219 | length, OBJ_OP_READ, NULL); |
e93f3152 AE |
2220 | if (!parent_request) |
2221 | return NULL; | |
2222 | ||
2223 | img_request_child_set(parent_request); | |
2224 | rbd_obj_request_get(obj_request); | |
2225 | parent_request->obj_request = obj_request; | |
2226 | ||
2227 | return parent_request; | |
2228 | } | |
2229 | ||
2230 | static void rbd_parent_request_destroy(struct kref *kref) | |
2231 | { | |
2232 | struct rbd_img_request *parent_request; | |
2233 | struct rbd_obj_request *orig_request; | |
2234 | ||
2235 | parent_request = container_of(kref, struct rbd_img_request, kref); | |
2236 | orig_request = parent_request->obj_request; | |
2237 | ||
2238 | parent_request->obj_request = NULL; | |
2239 | rbd_obj_request_put(orig_request); | |
2240 | img_request_child_clear(parent_request); | |
2241 | ||
2242 | rbd_img_request_destroy(kref); | |
2243 | } | |
2244 | ||
1217857f AE |
2245 | static bool rbd_img_obj_end_request(struct rbd_obj_request *obj_request) |
2246 | { | |
6365d33a | 2247 | struct rbd_img_request *img_request; |
1217857f AE |
2248 | unsigned int xferred; |
2249 | int result; | |
8b3e1a56 | 2250 | bool more; |
1217857f | 2251 | |
6365d33a AE |
2252 | rbd_assert(obj_request_img_data_test(obj_request)); |
2253 | img_request = obj_request->img_request; | |
2254 | ||
1217857f AE |
2255 | rbd_assert(obj_request->xferred <= (u64)UINT_MAX); |
2256 | xferred = (unsigned int)obj_request->xferred; | |
2257 | result = obj_request->result; | |
2258 | if (result) { | |
2259 | struct rbd_device *rbd_dev = img_request->rbd_dev; | |
6d2940c8 GZ |
2260 | enum obj_operation_type op_type; |
2261 | ||
90e98c52 GZ |
2262 | if (img_request_discard_test(img_request)) |
2263 | op_type = OBJ_OP_DISCARD; | |
2264 | else if (img_request_write_test(img_request)) | |
2265 | op_type = OBJ_OP_WRITE; | |
2266 | else | |
2267 | op_type = OBJ_OP_READ; | |
1217857f | 2268 | |
9584d508 | 2269 | rbd_warn(rbd_dev, "%s %llx at %llx (%llx)", |
6d2940c8 GZ |
2270 | obj_op_name(op_type), obj_request->length, |
2271 | obj_request->img_offset, obj_request->offset); | |
9584d508 | 2272 | rbd_warn(rbd_dev, " result %d xferred %x", |
1217857f AE |
2273 | result, xferred); |
2274 | if (!img_request->result) | |
2275 | img_request->result = result; | |
082a75da ID |
2276 | /* |
2277 | * Need to end I/O on the entire obj_request worth of | |
2278 | * bytes in case of error. | |
2279 | */ | |
2280 | xferred = obj_request->length; | |
1217857f AE |
2281 | } |
2282 | ||
f1a4739f AE |
2283 | /* Image object requests don't own their page array */ |
2284 | ||
2285 | if (obj_request->type == OBJ_REQUEST_PAGES) { | |
2286 | obj_request->pages = NULL; | |
2287 | obj_request->page_count = 0; | |
2288 | } | |
2289 | ||
8b3e1a56 AE |
2290 | if (img_request_child_test(img_request)) { |
2291 | rbd_assert(img_request->obj_request != NULL); | |
2292 | more = obj_request->which < img_request->obj_request_count - 1; | |
2293 | } else { | |
2294 | rbd_assert(img_request->rq != NULL); | |
7ad18afa CH |
2295 | |
2296 | more = blk_update_request(img_request->rq, result, xferred); | |
2297 | if (!more) | |
2298 | __blk_mq_end_request(img_request->rq, result); | |
8b3e1a56 AE |
2299 | } |
2300 | ||
2301 | return more; | |
1217857f AE |
2302 | } |
2303 | ||
2169238d AE |
2304 | static void rbd_img_obj_callback(struct rbd_obj_request *obj_request) |
2305 | { | |
2306 | struct rbd_img_request *img_request; | |
2307 | u32 which = obj_request->which; | |
2308 | bool more = true; | |
2309 | ||
6365d33a | 2310 | rbd_assert(obj_request_img_data_test(obj_request)); |
2169238d AE |
2311 | img_request = obj_request->img_request; |
2312 | ||
2313 | dout("%s: img %p obj %p\n", __func__, img_request, obj_request); | |
2314 | rbd_assert(img_request != NULL); | |
2169238d AE |
2315 | rbd_assert(img_request->obj_request_count > 0); |
2316 | rbd_assert(which != BAD_WHICH); | |
2317 | rbd_assert(which < img_request->obj_request_count); | |
2169238d AE |
2318 | |
2319 | spin_lock_irq(&img_request->completion_lock); | |
2320 | if (which != img_request->next_completion) | |
2321 | goto out; | |
2322 | ||
2323 | for_each_obj_request_from(img_request, obj_request) { | |
2169238d AE |
2324 | rbd_assert(more); |
2325 | rbd_assert(which < img_request->obj_request_count); | |
2326 | ||
2327 | if (!obj_request_done_test(obj_request)) | |
2328 | break; | |
1217857f | 2329 | more = rbd_img_obj_end_request(obj_request); |
2169238d AE |
2330 | which++; |
2331 | } | |
2332 | ||
2333 | rbd_assert(more ^ (which == img_request->obj_request_count)); | |
2334 | img_request->next_completion = which; | |
2335 | out: | |
2336 | spin_unlock_irq(&img_request->completion_lock); | |
0f2d5be7 | 2337 | rbd_img_request_put(img_request); |
2169238d AE |
2338 | |
2339 | if (!more) | |
2340 | rbd_img_request_complete(img_request); | |
2341 | } | |
2342 | ||
3b434a2a JD |
2343 | /* |
2344 | * Add individual osd ops to the given ceph_osd_request and prepare | |
2345 | * them for submission. num_ops is the current number of | |
2346 | * osd operations already to the object request. | |
2347 | */ | |
2348 | static void rbd_img_obj_request_fill(struct rbd_obj_request *obj_request, | |
2349 | struct ceph_osd_request *osd_request, | |
2350 | enum obj_operation_type op_type, | |
2351 | unsigned int num_ops) | |
2352 | { | |
2353 | struct rbd_img_request *img_request = obj_request->img_request; | |
2354 | struct rbd_device *rbd_dev = img_request->rbd_dev; | |
2355 | u64 object_size = rbd_obj_bytes(&rbd_dev->header); | |
2356 | u64 offset = obj_request->offset; | |
2357 | u64 length = obj_request->length; | |
2358 | u64 img_end; | |
2359 | u16 opcode; | |
2360 | ||
2361 | if (op_type == OBJ_OP_DISCARD) { | |
d3246fb0 JD |
2362 | if (!offset && length == object_size && |
2363 | (!img_request_layered_test(img_request) || | |
2364 | !obj_request_overlaps_parent(obj_request))) { | |
3b434a2a JD |
2365 | opcode = CEPH_OSD_OP_DELETE; |
2366 | } else if ((offset + length == object_size)) { | |
2367 | opcode = CEPH_OSD_OP_TRUNCATE; | |
2368 | } else { | |
2369 | down_read(&rbd_dev->header_rwsem); | |
2370 | img_end = rbd_dev->header.image_size; | |
2371 | up_read(&rbd_dev->header_rwsem); | |
2372 | ||
2373 | if (obj_request->img_offset + length == img_end) | |
2374 | opcode = CEPH_OSD_OP_TRUNCATE; | |
2375 | else | |
2376 | opcode = CEPH_OSD_OP_ZERO; | |
2377 | } | |
2378 | } else if (op_type == OBJ_OP_WRITE) { | |
2379 | opcode = CEPH_OSD_OP_WRITE; | |
2380 | osd_req_op_alloc_hint_init(osd_request, num_ops, | |
2381 | object_size, object_size); | |
2382 | num_ops++; | |
2383 | } else { | |
2384 | opcode = CEPH_OSD_OP_READ; | |
2385 | } | |
2386 | ||
7e868b6e | 2387 | if (opcode == CEPH_OSD_OP_DELETE) |
144cba14 | 2388 | osd_req_op_init(osd_request, num_ops, opcode, 0); |
7e868b6e ID |
2389 | else |
2390 | osd_req_op_extent_init(osd_request, num_ops, opcode, | |
2391 | offset, length, 0, 0); | |
2392 | ||
3b434a2a JD |
2393 | if (obj_request->type == OBJ_REQUEST_BIO) |
2394 | osd_req_op_extent_osd_data_bio(osd_request, num_ops, | |
2395 | obj_request->bio_list, length); | |
2396 | else if (obj_request->type == OBJ_REQUEST_PAGES) | |
2397 | osd_req_op_extent_osd_data_pages(osd_request, num_ops, | |
2398 | obj_request->pages, length, | |
2399 | offset & ~PAGE_MASK, false, false); | |
2400 | ||
2401 | /* Discards are also writes */ | |
2402 | if (op_type == OBJ_OP_WRITE || op_type == OBJ_OP_DISCARD) | |
2403 | rbd_osd_req_format_write(obj_request); | |
2404 | else | |
2405 | rbd_osd_req_format_read(obj_request); | |
2406 | } | |
2407 | ||
f1a4739f AE |
2408 | /* |
2409 | * Split up an image request into one or more object requests, each | |
2410 | * to a different object. The "type" parameter indicates whether | |
2411 | * "data_desc" is the pointer to the head of a list of bio | |
2412 | * structures, or the base of a page array. In either case this | |
2413 | * function assumes data_desc describes memory sufficient to hold | |
2414 | * all data described by the image request. | |
2415 | */ | |
2416 | static int rbd_img_request_fill(struct rbd_img_request *img_request, | |
2417 | enum obj_request_type type, | |
2418 | void *data_desc) | |
bf0d5f50 AE |
2419 | { |
2420 | struct rbd_device *rbd_dev = img_request->rbd_dev; | |
2421 | struct rbd_obj_request *obj_request = NULL; | |
2422 | struct rbd_obj_request *next_obj_request; | |
a158073c | 2423 | struct bio *bio_list = NULL; |
f1a4739f | 2424 | unsigned int bio_offset = 0; |
a158073c | 2425 | struct page **pages = NULL; |
6d2940c8 | 2426 | enum obj_operation_type op_type; |
7da22d29 | 2427 | u64 img_offset; |
bf0d5f50 | 2428 | u64 resid; |
bf0d5f50 | 2429 | |
f1a4739f AE |
2430 | dout("%s: img %p type %d data_desc %p\n", __func__, img_request, |
2431 | (int)type, data_desc); | |
37206ee5 | 2432 | |
7da22d29 | 2433 | img_offset = img_request->offset; |
bf0d5f50 | 2434 | resid = img_request->length; |
4dda41d3 | 2435 | rbd_assert(resid > 0); |
3b434a2a | 2436 | op_type = rbd_img_request_op_type(img_request); |
f1a4739f AE |
2437 | |
2438 | if (type == OBJ_REQUEST_BIO) { | |
2439 | bio_list = data_desc; | |
4f024f37 KO |
2440 | rbd_assert(img_offset == |
2441 | bio_list->bi_iter.bi_sector << SECTOR_SHIFT); | |
90e98c52 | 2442 | } else if (type == OBJ_REQUEST_PAGES) { |
f1a4739f AE |
2443 | pages = data_desc; |
2444 | } | |
2445 | ||
bf0d5f50 | 2446 | while (resid) { |
2fa12320 | 2447 | struct ceph_osd_request *osd_req; |
bf0d5f50 | 2448 | const char *object_name; |
bf0d5f50 AE |
2449 | u64 offset; |
2450 | u64 length; | |
2451 | ||
7da22d29 | 2452 | object_name = rbd_segment_name(rbd_dev, img_offset); |
bf0d5f50 AE |
2453 | if (!object_name) |
2454 | goto out_unwind; | |
7da22d29 AE |
2455 | offset = rbd_segment_offset(rbd_dev, img_offset); |
2456 | length = rbd_segment_length(rbd_dev, img_offset, resid); | |
bf0d5f50 | 2457 | obj_request = rbd_obj_request_create(object_name, |
f1a4739f | 2458 | offset, length, type); |
78c2a44a AE |
2459 | /* object request has its own copy of the object name */ |
2460 | rbd_segment_name_free(object_name); | |
bf0d5f50 AE |
2461 | if (!obj_request) |
2462 | goto out_unwind; | |
62054da6 | 2463 | |
03507db6 JD |
2464 | /* |
2465 | * set obj_request->img_request before creating the | |
2466 | * osd_request so that it gets the right snapc | |
2467 | */ | |
2468 | rbd_img_obj_request_add(img_request, obj_request); | |
bf0d5f50 | 2469 | |
f1a4739f AE |
2470 | if (type == OBJ_REQUEST_BIO) { |
2471 | unsigned int clone_size; | |
2472 | ||
2473 | rbd_assert(length <= (u64)UINT_MAX); | |
2474 | clone_size = (unsigned int)length; | |
2475 | obj_request->bio_list = | |
2476 | bio_chain_clone_range(&bio_list, | |
2477 | &bio_offset, | |
2478 | clone_size, | |
2479 | GFP_ATOMIC); | |
2480 | if (!obj_request->bio_list) | |
62054da6 | 2481 | goto out_unwind; |
90e98c52 | 2482 | } else if (type == OBJ_REQUEST_PAGES) { |
f1a4739f AE |
2483 | unsigned int page_count; |
2484 | ||
2485 | obj_request->pages = pages; | |
2486 | page_count = (u32)calc_pages_for(offset, length); | |
2487 | obj_request->page_count = page_count; | |
2488 | if ((offset + length) & ~PAGE_MASK) | |
2489 | page_count--; /* more on last page */ | |
2490 | pages += page_count; | |
2491 | } | |
bf0d5f50 | 2492 | |
6d2940c8 GZ |
2493 | osd_req = rbd_osd_req_create(rbd_dev, op_type, |
2494 | (op_type == OBJ_OP_WRITE) ? 2 : 1, | |
2495 | obj_request); | |
2fa12320 | 2496 | if (!osd_req) |
62054da6 | 2497 | goto out_unwind; |
3b434a2a | 2498 | |
2fa12320 | 2499 | obj_request->osd_req = osd_req; |
2169238d | 2500 | obj_request->callback = rbd_img_obj_callback; |
3b434a2a | 2501 | obj_request->img_offset = img_offset; |
9d4df01f | 2502 | |
3b434a2a | 2503 | rbd_img_obj_request_fill(obj_request, osd_req, op_type, 0); |
430c28c3 | 2504 | |
3b434a2a | 2505 | rbd_img_request_get(img_request); |
bf0d5f50 | 2506 | |
7da22d29 | 2507 | img_offset += length; |
bf0d5f50 AE |
2508 | resid -= length; |
2509 | } | |
2510 | ||
2511 | return 0; | |
2512 | ||
bf0d5f50 AE |
2513 | out_unwind: |
2514 | for_each_obj_request_safe(img_request, obj_request, next_obj_request) | |
42dd037c | 2515 | rbd_img_obj_request_del(img_request, obj_request); |
bf0d5f50 AE |
2516 | |
2517 | return -ENOMEM; | |
2518 | } | |
2519 | ||
0eefd470 AE |
2520 | static void |
2521 | rbd_img_obj_copyup_callback(struct rbd_obj_request *obj_request) | |
2522 | { | |
2523 | struct rbd_img_request *img_request; | |
2524 | struct rbd_device *rbd_dev; | |
ebda6408 | 2525 | struct page **pages; |
0eefd470 AE |
2526 | u32 page_count; |
2527 | ||
d3246fb0 JD |
2528 | rbd_assert(obj_request->type == OBJ_REQUEST_BIO || |
2529 | obj_request->type == OBJ_REQUEST_NODATA); | |
0eefd470 AE |
2530 | rbd_assert(obj_request_img_data_test(obj_request)); |
2531 | img_request = obj_request->img_request; | |
2532 | rbd_assert(img_request); | |
2533 | ||
2534 | rbd_dev = img_request->rbd_dev; | |
2535 | rbd_assert(rbd_dev); | |
0eefd470 | 2536 | |
ebda6408 AE |
2537 | pages = obj_request->copyup_pages; |
2538 | rbd_assert(pages != NULL); | |
0eefd470 | 2539 | obj_request->copyup_pages = NULL; |
ebda6408 AE |
2540 | page_count = obj_request->copyup_page_count; |
2541 | rbd_assert(page_count); | |
2542 | obj_request->copyup_page_count = 0; | |
2543 | ceph_release_page_vector(pages, page_count); | |
0eefd470 AE |
2544 | |
2545 | /* | |
2546 | * We want the transfer count to reflect the size of the | |
2547 | * original write request. There is no such thing as a | |
2548 | * successful short write, so if the request was successful | |
2549 | * we can just set it to the originally-requested length. | |
2550 | */ | |
2551 | if (!obj_request->result) | |
2552 | obj_request->xferred = obj_request->length; | |
2553 | ||
2554 | /* Finish up with the normal image object callback */ | |
2555 | ||
2556 | rbd_img_obj_callback(obj_request); | |
2557 | } | |
2558 | ||
3d7efd18 AE |
2559 | static void |
2560 | rbd_img_obj_parent_read_full_callback(struct rbd_img_request *img_request) | |
2561 | { | |
2562 | struct rbd_obj_request *orig_request; | |
0eefd470 AE |
2563 | struct ceph_osd_request *osd_req; |
2564 | struct ceph_osd_client *osdc; | |
2565 | struct rbd_device *rbd_dev; | |
3d7efd18 | 2566 | struct page **pages; |
d3246fb0 | 2567 | enum obj_operation_type op_type; |
ebda6408 | 2568 | u32 page_count; |
bbea1c1a | 2569 | int img_result; |
ebda6408 | 2570 | u64 parent_length; |
3d7efd18 AE |
2571 | |
2572 | rbd_assert(img_request_child_test(img_request)); | |
2573 | ||
2574 | /* First get what we need from the image request */ | |
2575 | ||
2576 | pages = img_request->copyup_pages; | |
2577 | rbd_assert(pages != NULL); | |
2578 | img_request->copyup_pages = NULL; | |
ebda6408 AE |
2579 | page_count = img_request->copyup_page_count; |
2580 | rbd_assert(page_count); | |
2581 | img_request->copyup_page_count = 0; | |
3d7efd18 AE |
2582 | |
2583 | orig_request = img_request->obj_request; | |
2584 | rbd_assert(orig_request != NULL); | |
b91f09f1 | 2585 | rbd_assert(obj_request_type_valid(orig_request->type)); |
bbea1c1a | 2586 | img_result = img_request->result; |
ebda6408 AE |
2587 | parent_length = img_request->length; |
2588 | rbd_assert(parent_length == img_request->xferred); | |
91c6febb | 2589 | rbd_img_request_put(img_request); |
3d7efd18 | 2590 | |
91c6febb AE |
2591 | rbd_assert(orig_request->img_request); |
2592 | rbd_dev = orig_request->img_request->rbd_dev; | |
0eefd470 | 2593 | rbd_assert(rbd_dev); |
0eefd470 | 2594 | |
bbea1c1a AE |
2595 | /* |
2596 | * If the overlap has become 0 (most likely because the | |
2597 | * image has been flattened) we need to free the pages | |
2598 | * and re-submit the original write request. | |
2599 | */ | |
2600 | if (!rbd_dev->parent_overlap) { | |
2601 | struct ceph_osd_client *osdc; | |
3d7efd18 | 2602 | |
bbea1c1a AE |
2603 | ceph_release_page_vector(pages, page_count); |
2604 | osdc = &rbd_dev->rbd_client->client->osdc; | |
2605 | img_result = rbd_obj_request_submit(osdc, orig_request); | |
2606 | if (!img_result) | |
2607 | return; | |
2608 | } | |
0eefd470 | 2609 | |
bbea1c1a | 2610 | if (img_result) |
0eefd470 | 2611 | goto out_err; |
0eefd470 | 2612 | |
8785b1d4 AE |
2613 | /* |
2614 | * The original osd request is of no use to use any more. | |
0ccd5926 | 2615 | * We need a new one that can hold the three ops in a copyup |
8785b1d4 AE |
2616 | * request. Allocate the new copyup osd request for the |
2617 | * original request, and release the old one. | |
2618 | */ | |
bbea1c1a | 2619 | img_result = -ENOMEM; |
0eefd470 AE |
2620 | osd_req = rbd_osd_req_create_copyup(orig_request); |
2621 | if (!osd_req) | |
2622 | goto out_err; | |
8785b1d4 | 2623 | rbd_osd_req_destroy(orig_request->osd_req); |
0eefd470 AE |
2624 | orig_request->osd_req = osd_req; |
2625 | orig_request->copyup_pages = pages; | |
ebda6408 | 2626 | orig_request->copyup_page_count = page_count; |
3d7efd18 | 2627 | |
0eefd470 | 2628 | /* Initialize the copyup op */ |
3d7efd18 | 2629 | |
0eefd470 | 2630 | osd_req_op_cls_init(osd_req, 0, CEPH_OSD_OP_CALL, "rbd", "copyup"); |
ebda6408 | 2631 | osd_req_op_cls_request_data_pages(osd_req, 0, pages, parent_length, 0, |
0eefd470 | 2632 | false, false); |
3d7efd18 | 2633 | |
d3246fb0 | 2634 | /* Add the other op(s) */ |
0eefd470 | 2635 | |
d3246fb0 JD |
2636 | op_type = rbd_img_request_op_type(orig_request->img_request); |
2637 | rbd_img_obj_request_fill(orig_request, osd_req, op_type, 1); | |
0eefd470 AE |
2638 | |
2639 | /* All set, send it off. */ | |
2640 | ||
2641 | orig_request->callback = rbd_img_obj_copyup_callback; | |
2642 | osdc = &rbd_dev->rbd_client->client->osdc; | |
bbea1c1a AE |
2643 | img_result = rbd_obj_request_submit(osdc, orig_request); |
2644 | if (!img_result) | |
0eefd470 AE |
2645 | return; |
2646 | out_err: | |
2647 | /* Record the error code and complete the request */ | |
2648 | ||
bbea1c1a | 2649 | orig_request->result = img_result; |
0eefd470 AE |
2650 | orig_request->xferred = 0; |
2651 | obj_request_done_set(orig_request); | |
2652 | rbd_obj_request_complete(orig_request); | |
3d7efd18 AE |
2653 | } |
2654 | ||
2655 | /* | |
2656 | * Read from the parent image the range of data that covers the | |
2657 | * entire target of the given object request. This is used for | |
2658 | * satisfying a layered image write request when the target of an | |
2659 | * object request from the image request does not exist. | |
2660 | * | |
2661 | * A page array big enough to hold the returned data is allocated | |
2662 | * and supplied to rbd_img_request_fill() as the "data descriptor." | |
2663 | * When the read completes, this page array will be transferred to | |
2664 | * the original object request for the copyup operation. | |
2665 | * | |
2666 | * If an error occurs, record it as the result of the original | |
2667 | * object request and mark it done so it gets completed. | |
2668 | */ | |
2669 | static int rbd_img_obj_parent_read_full(struct rbd_obj_request *obj_request) | |
2670 | { | |
2671 | struct rbd_img_request *img_request = NULL; | |
2672 | struct rbd_img_request *parent_request = NULL; | |
2673 | struct rbd_device *rbd_dev; | |
2674 | u64 img_offset; | |
2675 | u64 length; | |
2676 | struct page **pages = NULL; | |
2677 | u32 page_count; | |
2678 | int result; | |
2679 | ||
2680 | rbd_assert(obj_request_img_data_test(obj_request)); | |
b91f09f1 | 2681 | rbd_assert(obj_request_type_valid(obj_request->type)); |
3d7efd18 AE |
2682 | |
2683 | img_request = obj_request->img_request; | |
2684 | rbd_assert(img_request != NULL); | |
2685 | rbd_dev = img_request->rbd_dev; | |
2686 | rbd_assert(rbd_dev->parent != NULL); | |
2687 | ||
2688 | /* | |
2689 | * Determine the byte range covered by the object in the | |
2690 | * child image to which the original request was to be sent. | |
2691 | */ | |
2692 | img_offset = obj_request->img_offset - obj_request->offset; | |
2693 | length = (u64)1 << rbd_dev->header.obj_order; | |
2694 | ||
a9e8ba2c AE |
2695 | /* |
2696 | * There is no defined parent data beyond the parent | |
2697 | * overlap, so limit what we read at that boundary if | |
2698 | * necessary. | |
2699 | */ | |
2700 | if (img_offset + length > rbd_dev->parent_overlap) { | |
2701 | rbd_assert(img_offset < rbd_dev->parent_overlap); | |
2702 | length = rbd_dev->parent_overlap - img_offset; | |
2703 | } | |
2704 | ||
3d7efd18 AE |
2705 | /* |
2706 | * Allocate a page array big enough to receive the data read | |
2707 | * from the parent. | |
2708 | */ | |
2709 | page_count = (u32)calc_pages_for(0, length); | |
2710 | pages = ceph_alloc_page_vector(page_count, GFP_KERNEL); | |
2711 | if (IS_ERR(pages)) { | |
2712 | result = PTR_ERR(pages); | |
2713 | pages = NULL; | |
2714 | goto out_err; | |
2715 | } | |
2716 | ||
2717 | result = -ENOMEM; | |
e93f3152 AE |
2718 | parent_request = rbd_parent_request_create(obj_request, |
2719 | img_offset, length); | |
3d7efd18 AE |
2720 | if (!parent_request) |
2721 | goto out_err; | |
3d7efd18 AE |
2722 | |
2723 | result = rbd_img_request_fill(parent_request, OBJ_REQUEST_PAGES, pages); | |
2724 | if (result) | |
2725 | goto out_err; | |
2726 | parent_request->copyup_pages = pages; | |
ebda6408 | 2727 | parent_request->copyup_page_count = page_count; |
3d7efd18 AE |
2728 | |
2729 | parent_request->callback = rbd_img_obj_parent_read_full_callback; | |
2730 | result = rbd_img_request_submit(parent_request); | |
2731 | if (!result) | |
2732 | return 0; | |
2733 | ||
2734 | parent_request->copyup_pages = NULL; | |
ebda6408 | 2735 | parent_request->copyup_page_count = 0; |
3d7efd18 AE |
2736 | parent_request->obj_request = NULL; |
2737 | rbd_obj_request_put(obj_request); | |
2738 | out_err: | |
2739 | if (pages) | |
2740 | ceph_release_page_vector(pages, page_count); | |
2741 | if (parent_request) | |
2742 | rbd_img_request_put(parent_request); | |
2743 | obj_request->result = result; | |
2744 | obj_request->xferred = 0; | |
2745 | obj_request_done_set(obj_request); | |
2746 | ||
2747 | return result; | |
2748 | } | |
2749 | ||
c5b5ef6c AE |
2750 | static void rbd_img_obj_exists_callback(struct rbd_obj_request *obj_request) |
2751 | { | |
c5b5ef6c | 2752 | struct rbd_obj_request *orig_request; |
638f5abe | 2753 | struct rbd_device *rbd_dev; |
c5b5ef6c AE |
2754 | int result; |
2755 | ||
2756 | rbd_assert(!obj_request_img_data_test(obj_request)); | |
2757 | ||
2758 | /* | |
2759 | * All we need from the object request is the original | |
2760 | * request and the result of the STAT op. Grab those, then | |
2761 | * we're done with the request. | |
2762 | */ | |
2763 | orig_request = obj_request->obj_request; | |
2764 | obj_request->obj_request = NULL; | |
912c317d | 2765 | rbd_obj_request_put(orig_request); |
c5b5ef6c AE |
2766 | rbd_assert(orig_request); |
2767 | rbd_assert(orig_request->img_request); | |
2768 | ||
2769 | result = obj_request->result; | |
2770 | obj_request->result = 0; | |
2771 | ||
2772 | dout("%s: obj %p for obj %p result %d %llu/%llu\n", __func__, | |
2773 | obj_request, orig_request, result, | |
2774 | obj_request->xferred, obj_request->length); | |
2775 | rbd_obj_request_put(obj_request); | |
2776 | ||
638f5abe AE |
2777 | /* |
2778 | * If the overlap has become 0 (most likely because the | |
2779 | * image has been flattened) we need to free the pages | |
2780 | * and re-submit the original write request. | |
2781 | */ | |
2782 | rbd_dev = orig_request->img_request->rbd_dev; | |
2783 | if (!rbd_dev->parent_overlap) { | |
2784 | struct ceph_osd_client *osdc; | |
2785 | ||
638f5abe AE |
2786 | osdc = &rbd_dev->rbd_client->client->osdc; |
2787 | result = rbd_obj_request_submit(osdc, orig_request); | |
2788 | if (!result) | |
2789 | return; | |
2790 | } | |
c5b5ef6c AE |
2791 | |
2792 | /* | |
2793 | * Our only purpose here is to determine whether the object | |
2794 | * exists, and we don't want to treat the non-existence as | |
2795 | * an error. If something else comes back, transfer the | |
2796 | * error to the original request and complete it now. | |
2797 | */ | |
2798 | if (!result) { | |
2799 | obj_request_existence_set(orig_request, true); | |
2800 | } else if (result == -ENOENT) { | |
2801 | obj_request_existence_set(orig_request, false); | |
2802 | } else if (result) { | |
2803 | orig_request->result = result; | |
3d7efd18 | 2804 | goto out; |
c5b5ef6c AE |
2805 | } |
2806 | ||
2807 | /* | |
2808 | * Resubmit the original request now that we have recorded | |
2809 | * whether the target object exists. | |
2810 | */ | |
b454e36d | 2811 | orig_request->result = rbd_img_obj_request_submit(orig_request); |
3d7efd18 | 2812 | out: |
c5b5ef6c AE |
2813 | if (orig_request->result) |
2814 | rbd_obj_request_complete(orig_request); | |
c5b5ef6c AE |
2815 | } |
2816 | ||
2817 | static int rbd_img_obj_exists_submit(struct rbd_obj_request *obj_request) | |
2818 | { | |
2819 | struct rbd_obj_request *stat_request; | |
2820 | struct rbd_device *rbd_dev; | |
2821 | struct ceph_osd_client *osdc; | |
2822 | struct page **pages = NULL; | |
2823 | u32 page_count; | |
2824 | size_t size; | |
2825 | int ret; | |
2826 | ||
2827 | /* | |
2828 | * The response data for a STAT call consists of: | |
2829 | * le64 length; | |
2830 | * struct { | |
2831 | * le32 tv_sec; | |
2832 | * le32 tv_nsec; | |
2833 | * } mtime; | |
2834 | */ | |
2835 | size = sizeof (__le64) + sizeof (__le32) + sizeof (__le32); | |
2836 | page_count = (u32)calc_pages_for(0, size); | |
2837 | pages = ceph_alloc_page_vector(page_count, GFP_KERNEL); | |
2838 | if (IS_ERR(pages)) | |
2839 | return PTR_ERR(pages); | |
2840 | ||
2841 | ret = -ENOMEM; | |
2842 | stat_request = rbd_obj_request_create(obj_request->object_name, 0, 0, | |
2843 | OBJ_REQUEST_PAGES); | |
2844 | if (!stat_request) | |
2845 | goto out; | |
2846 | ||
2847 | rbd_obj_request_get(obj_request); | |
2848 | stat_request->obj_request = obj_request; | |
2849 | stat_request->pages = pages; | |
2850 | stat_request->page_count = page_count; | |
2851 | ||
2852 | rbd_assert(obj_request->img_request); | |
2853 | rbd_dev = obj_request->img_request->rbd_dev; | |
6d2940c8 | 2854 | stat_request->osd_req = rbd_osd_req_create(rbd_dev, OBJ_OP_READ, 1, |
deb236b3 | 2855 | stat_request); |
c5b5ef6c AE |
2856 | if (!stat_request->osd_req) |
2857 | goto out; | |
2858 | stat_request->callback = rbd_img_obj_exists_callback; | |
2859 | ||
144cba14 | 2860 | osd_req_op_init(stat_request->osd_req, 0, CEPH_OSD_OP_STAT, 0); |
c5b5ef6c AE |
2861 | osd_req_op_raw_data_in_pages(stat_request->osd_req, 0, pages, size, 0, |
2862 | false, false); | |
9d4df01f | 2863 | rbd_osd_req_format_read(stat_request); |
c5b5ef6c AE |
2864 | |
2865 | osdc = &rbd_dev->rbd_client->client->osdc; | |
2866 | ret = rbd_obj_request_submit(osdc, stat_request); | |
2867 | out: | |
2868 | if (ret) | |
2869 | rbd_obj_request_put(obj_request); | |
2870 | ||
2871 | return ret; | |
2872 | } | |
2873 | ||
70d045f6 | 2874 | static bool img_obj_request_simple(struct rbd_obj_request *obj_request) |
b454e36d AE |
2875 | { |
2876 | struct rbd_img_request *img_request; | |
a9e8ba2c | 2877 | struct rbd_device *rbd_dev; |
b454e36d AE |
2878 | |
2879 | rbd_assert(obj_request_img_data_test(obj_request)); | |
2880 | ||
2881 | img_request = obj_request->img_request; | |
2882 | rbd_assert(img_request); | |
a9e8ba2c | 2883 | rbd_dev = img_request->rbd_dev; |
b454e36d | 2884 | |
70d045f6 | 2885 | /* Reads */ |
1c220881 JD |
2886 | if (!img_request_write_test(img_request) && |
2887 | !img_request_discard_test(img_request)) | |
70d045f6 ID |
2888 | return true; |
2889 | ||
2890 | /* Non-layered writes */ | |
2891 | if (!img_request_layered_test(img_request)) | |
2892 | return true; | |
2893 | ||
b454e36d | 2894 | /* |
70d045f6 ID |
2895 | * Layered writes outside of the parent overlap range don't |
2896 | * share any data with the parent. | |
b454e36d | 2897 | */ |
70d045f6 ID |
2898 | if (!obj_request_overlaps_parent(obj_request)) |
2899 | return true; | |
b454e36d | 2900 | |
c622d226 GZ |
2901 | /* |
2902 | * Entire-object layered writes - we will overwrite whatever | |
2903 | * parent data there is anyway. | |
2904 | */ | |
2905 | if (!obj_request->offset && | |
2906 | obj_request->length == rbd_obj_bytes(&rbd_dev->header)) | |
2907 | return true; | |
2908 | ||
70d045f6 ID |
2909 | /* |
2910 | * If the object is known to already exist, its parent data has | |
2911 | * already been copied. | |
2912 | */ | |
2913 | if (obj_request_known_test(obj_request) && | |
2914 | obj_request_exists_test(obj_request)) | |
2915 | return true; | |
2916 | ||
2917 | return false; | |
2918 | } | |
2919 | ||
2920 | static int rbd_img_obj_request_submit(struct rbd_obj_request *obj_request) | |
2921 | { | |
2922 | if (img_obj_request_simple(obj_request)) { | |
b454e36d AE |
2923 | struct rbd_device *rbd_dev; |
2924 | struct ceph_osd_client *osdc; | |
2925 | ||
2926 | rbd_dev = obj_request->img_request->rbd_dev; | |
2927 | osdc = &rbd_dev->rbd_client->client->osdc; | |
2928 | ||
2929 | return rbd_obj_request_submit(osdc, obj_request); | |
2930 | } | |
2931 | ||
2932 | /* | |
3d7efd18 AE |
2933 | * It's a layered write. The target object might exist but |
2934 | * we may not know that yet. If we know it doesn't exist, | |
2935 | * start by reading the data for the full target object from | |
2936 | * the parent so we can use it for a copyup to the target. | |
b454e36d | 2937 | */ |
70d045f6 | 2938 | if (obj_request_known_test(obj_request)) |
3d7efd18 AE |
2939 | return rbd_img_obj_parent_read_full(obj_request); |
2940 | ||
2941 | /* We don't know whether the target exists. Go find out. */ | |
b454e36d AE |
2942 | |
2943 | return rbd_img_obj_exists_submit(obj_request); | |
2944 | } | |
2945 | ||
bf0d5f50 AE |
2946 | static int rbd_img_request_submit(struct rbd_img_request *img_request) |
2947 | { | |
bf0d5f50 | 2948 | struct rbd_obj_request *obj_request; |
46faeed4 | 2949 | struct rbd_obj_request *next_obj_request; |
bf0d5f50 | 2950 | |
37206ee5 | 2951 | dout("%s: img %p\n", __func__, img_request); |
46faeed4 | 2952 | for_each_obj_request_safe(img_request, obj_request, next_obj_request) { |
bf0d5f50 AE |
2953 | int ret; |
2954 | ||
b454e36d | 2955 | ret = rbd_img_obj_request_submit(obj_request); |
bf0d5f50 AE |
2956 | if (ret) |
2957 | return ret; | |
bf0d5f50 AE |
2958 | } |
2959 | ||
2960 | return 0; | |
2961 | } | |
8b3e1a56 AE |
2962 | |
2963 | static void rbd_img_parent_read_callback(struct rbd_img_request *img_request) | |
2964 | { | |
2965 | struct rbd_obj_request *obj_request; | |
a9e8ba2c AE |
2966 | struct rbd_device *rbd_dev; |
2967 | u64 obj_end; | |
02c74fba AE |
2968 | u64 img_xferred; |
2969 | int img_result; | |
8b3e1a56 AE |
2970 | |
2971 | rbd_assert(img_request_child_test(img_request)); | |
2972 | ||
02c74fba AE |
2973 | /* First get what we need from the image request and release it */ |
2974 | ||
8b3e1a56 | 2975 | obj_request = img_request->obj_request; |
02c74fba AE |
2976 | img_xferred = img_request->xferred; |
2977 | img_result = img_request->result; | |
2978 | rbd_img_request_put(img_request); | |
2979 | ||
2980 | /* | |
2981 | * If the overlap has become 0 (most likely because the | |
2982 | * image has been flattened) we need to re-submit the | |
2983 | * original request. | |
2984 | */ | |
a9e8ba2c AE |
2985 | rbd_assert(obj_request); |
2986 | rbd_assert(obj_request->img_request); | |
02c74fba AE |
2987 | rbd_dev = obj_request->img_request->rbd_dev; |
2988 | if (!rbd_dev->parent_overlap) { | |
2989 | struct ceph_osd_client *osdc; | |
2990 | ||
2991 | osdc = &rbd_dev->rbd_client->client->osdc; | |
2992 | img_result = rbd_obj_request_submit(osdc, obj_request); | |
2993 | if (!img_result) | |
2994 | return; | |
2995 | } | |
a9e8ba2c | 2996 | |
02c74fba | 2997 | obj_request->result = img_result; |
a9e8ba2c AE |
2998 | if (obj_request->result) |
2999 | goto out; | |
3000 | ||
3001 | /* | |
3002 | * We need to zero anything beyond the parent overlap | |
3003 | * boundary. Since rbd_img_obj_request_read_callback() | |
3004 | * will zero anything beyond the end of a short read, an | |
3005 | * easy way to do this is to pretend the data from the | |
3006 | * parent came up short--ending at the overlap boundary. | |
3007 | */ | |
3008 | rbd_assert(obj_request->img_offset < U64_MAX - obj_request->length); | |
3009 | obj_end = obj_request->img_offset + obj_request->length; | |
a9e8ba2c AE |
3010 | if (obj_end > rbd_dev->parent_overlap) { |
3011 | u64 xferred = 0; | |
3012 | ||
3013 | if (obj_request->img_offset < rbd_dev->parent_overlap) | |
3014 | xferred = rbd_dev->parent_overlap - | |
3015 | obj_request->img_offset; | |
8b3e1a56 | 3016 | |
02c74fba | 3017 | obj_request->xferred = min(img_xferred, xferred); |
a9e8ba2c | 3018 | } else { |
02c74fba | 3019 | obj_request->xferred = img_xferred; |
a9e8ba2c AE |
3020 | } |
3021 | out: | |
8b3e1a56 AE |
3022 | rbd_img_obj_request_read_callback(obj_request); |
3023 | rbd_obj_request_complete(obj_request); | |
3024 | } | |
3025 | ||
3026 | static void rbd_img_parent_read(struct rbd_obj_request *obj_request) | |
3027 | { | |
8b3e1a56 AE |
3028 | struct rbd_img_request *img_request; |
3029 | int result; | |
3030 | ||
3031 | rbd_assert(obj_request_img_data_test(obj_request)); | |
3032 | rbd_assert(obj_request->img_request != NULL); | |
3033 | rbd_assert(obj_request->result == (s32) -ENOENT); | |
5b2ab72d | 3034 | rbd_assert(obj_request_type_valid(obj_request->type)); |
8b3e1a56 | 3035 | |
8b3e1a56 | 3036 | /* rbd_read_finish(obj_request, obj_request->length); */ |
e93f3152 | 3037 | img_request = rbd_parent_request_create(obj_request, |
8b3e1a56 | 3038 | obj_request->img_offset, |
e93f3152 | 3039 | obj_request->length); |
8b3e1a56 AE |
3040 | result = -ENOMEM; |
3041 | if (!img_request) | |
3042 | goto out_err; | |
3043 | ||
5b2ab72d AE |
3044 | if (obj_request->type == OBJ_REQUEST_BIO) |
3045 | result = rbd_img_request_fill(img_request, OBJ_REQUEST_BIO, | |
3046 | obj_request->bio_list); | |
3047 | else | |
3048 | result = rbd_img_request_fill(img_request, OBJ_REQUEST_PAGES, | |
3049 | obj_request->pages); | |
8b3e1a56 AE |
3050 | if (result) |
3051 | goto out_err; | |
3052 | ||
3053 | img_request->callback = rbd_img_parent_read_callback; | |
3054 | result = rbd_img_request_submit(img_request); | |
3055 | if (result) | |
3056 | goto out_err; | |
3057 | ||
3058 | return; | |
3059 | out_err: | |
3060 | if (img_request) | |
3061 | rbd_img_request_put(img_request); | |
3062 | obj_request->result = result; | |
3063 | obj_request->xferred = 0; | |
3064 | obj_request_done_set(obj_request); | |
3065 | } | |
bf0d5f50 | 3066 | |
20e0af67 | 3067 | static int rbd_obj_notify_ack_sync(struct rbd_device *rbd_dev, u64 notify_id) |
b8d70035 AE |
3068 | { |
3069 | struct rbd_obj_request *obj_request; | |
2169238d | 3070 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
b8d70035 AE |
3071 | int ret; |
3072 | ||
3073 | obj_request = rbd_obj_request_create(rbd_dev->header_name, 0, 0, | |
3074 | OBJ_REQUEST_NODATA); | |
3075 | if (!obj_request) | |
3076 | return -ENOMEM; | |
3077 | ||
3078 | ret = -ENOMEM; | |
6d2940c8 | 3079 | obj_request->osd_req = rbd_osd_req_create(rbd_dev, OBJ_OP_READ, 1, |
deb236b3 | 3080 | obj_request); |
b8d70035 AE |
3081 | if (!obj_request->osd_req) |
3082 | goto out; | |
3083 | ||
c99d2d4a | 3084 | osd_req_op_watch_init(obj_request->osd_req, 0, CEPH_OSD_OP_NOTIFY_ACK, |
cc4a38bd | 3085 | notify_id, 0, 0); |
9d4df01f | 3086 | rbd_osd_req_format_read(obj_request); |
430c28c3 | 3087 | |
b8d70035 | 3088 | ret = rbd_obj_request_submit(osdc, obj_request); |
cf81b60e | 3089 | if (ret) |
20e0af67 JD |
3090 | goto out; |
3091 | ret = rbd_obj_request_wait(obj_request); | |
3092 | out: | |
3093 | rbd_obj_request_put(obj_request); | |
b8d70035 AE |
3094 | |
3095 | return ret; | |
3096 | } | |
3097 | ||
3098 | static void rbd_watch_cb(u64 ver, u64 notify_id, u8 opcode, void *data) | |
3099 | { | |
3100 | struct rbd_device *rbd_dev = (struct rbd_device *)data; | |
e627db08 | 3101 | int ret; |
b8d70035 AE |
3102 | |
3103 | if (!rbd_dev) | |
3104 | return; | |
3105 | ||
37206ee5 | 3106 | dout("%s: \"%s\" notify_id %llu opcode %u\n", __func__, |
cc4a38bd AE |
3107 | rbd_dev->header_name, (unsigned long long)notify_id, |
3108 | (unsigned int)opcode); | |
52bb1f9b ID |
3109 | |
3110 | /* | |
3111 | * Until adequate refresh error handling is in place, there is | |
3112 | * not much we can do here, except warn. | |
3113 | * | |
3114 | * See http://tracker.ceph.com/issues/5040 | |
3115 | */ | |
e627db08 AE |
3116 | ret = rbd_dev_refresh(rbd_dev); |
3117 | if (ret) | |
9584d508 | 3118 | rbd_warn(rbd_dev, "refresh failed: %d", ret); |
b8d70035 | 3119 | |
52bb1f9b ID |
3120 | ret = rbd_obj_notify_ack_sync(rbd_dev, notify_id); |
3121 | if (ret) | |
9584d508 | 3122 | rbd_warn(rbd_dev, "notify_ack ret %d", ret); |
b8d70035 AE |
3123 | } |
3124 | ||
bb040aa0 ID |
3125 | /* |
3126 | * Send a (un)watch request and wait for the ack. Return a request | |
3127 | * with a ref held on success or error. | |
3128 | */ | |
3129 | static struct rbd_obj_request *rbd_obj_watch_request_helper( | |
3130 | struct rbd_device *rbd_dev, | |
3131 | bool watch) | |
3132 | { | |
3133 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; | |
2894e1d7 | 3134 | struct ceph_options *opts = osdc->client->options; |
bb040aa0 ID |
3135 | struct rbd_obj_request *obj_request; |
3136 | int ret; | |
3137 | ||
3138 | obj_request = rbd_obj_request_create(rbd_dev->header_name, 0, 0, | |
3139 | OBJ_REQUEST_NODATA); | |
3140 | if (!obj_request) | |
3141 | return ERR_PTR(-ENOMEM); | |
3142 | ||
6d2940c8 | 3143 | obj_request->osd_req = rbd_osd_req_create(rbd_dev, OBJ_OP_WRITE, 1, |
bb040aa0 ID |
3144 | obj_request); |
3145 | if (!obj_request->osd_req) { | |
3146 | ret = -ENOMEM; | |
3147 | goto out; | |
3148 | } | |
3149 | ||
3150 | osd_req_op_watch_init(obj_request->osd_req, 0, CEPH_OSD_OP_WATCH, | |
3151 | rbd_dev->watch_event->cookie, 0, watch); | |
3152 | rbd_osd_req_format_write(obj_request); | |
3153 | ||
3154 | if (watch) | |
3155 | ceph_osdc_set_request_linger(osdc, obj_request->osd_req); | |
3156 | ||
3157 | ret = rbd_obj_request_submit(osdc, obj_request); | |
3158 | if (ret) | |
3159 | goto out; | |
3160 | ||
2894e1d7 | 3161 | ret = rbd_obj_request_wait_timeout(obj_request, opts->mount_timeout); |
bb040aa0 ID |
3162 | if (ret) |
3163 | goto out; | |
3164 | ||
3165 | ret = obj_request->result; | |
3166 | if (ret) { | |
3167 | if (watch) | |
3168 | rbd_obj_request_end(obj_request); | |
3169 | goto out; | |
3170 | } | |
3171 | ||
3172 | return obj_request; | |
3173 | ||
3174 | out: | |
3175 | rbd_obj_request_put(obj_request); | |
3176 | return ERR_PTR(ret); | |
3177 | } | |
3178 | ||
9969ebc5 | 3179 | /* |
b30a01f2 | 3180 | * Initiate a watch request, synchronously. |
9969ebc5 | 3181 | */ |
b30a01f2 | 3182 | static int rbd_dev_header_watch_sync(struct rbd_device *rbd_dev) |
9969ebc5 AE |
3183 | { |
3184 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; | |
3185 | struct rbd_obj_request *obj_request; | |
9969ebc5 AE |
3186 | int ret; |
3187 | ||
b30a01f2 ID |
3188 | rbd_assert(!rbd_dev->watch_event); |
3189 | rbd_assert(!rbd_dev->watch_request); | |
9969ebc5 | 3190 | |
b30a01f2 ID |
3191 | ret = ceph_osdc_create_event(osdc, rbd_watch_cb, rbd_dev, |
3192 | &rbd_dev->watch_event); | |
3193 | if (ret < 0) | |
3194 | return ret; | |
3195 | ||
76756a51 ID |
3196 | obj_request = rbd_obj_watch_request_helper(rbd_dev, true); |
3197 | if (IS_ERR(obj_request)) { | |
3198 | ceph_osdc_cancel_event(rbd_dev->watch_event); | |
3199 | rbd_dev->watch_event = NULL; | |
3200 | return PTR_ERR(obj_request); | |
b30a01f2 | 3201 | } |
9969ebc5 | 3202 | |
8eb87565 AE |
3203 | /* |
3204 | * A watch request is set to linger, so the underlying osd | |
3205 | * request won't go away until we unregister it. We retain | |
3206 | * a pointer to the object request during that time (in | |
76756a51 ID |
3207 | * rbd_dev->watch_request), so we'll keep a reference to it. |
3208 | * We'll drop that reference after we've unregistered it in | |
3209 | * rbd_dev_header_unwatch_sync(). | |
8eb87565 | 3210 | */ |
b30a01f2 | 3211 | rbd_dev->watch_request = obj_request; |
8eb87565 | 3212 | |
b30a01f2 | 3213 | return 0; |
b30a01f2 ID |
3214 | } |
3215 | ||
3216 | /* | |
3217 | * Tear down a watch request, synchronously. | |
3218 | */ | |
76756a51 | 3219 | static void rbd_dev_header_unwatch_sync(struct rbd_device *rbd_dev) |
b30a01f2 | 3220 | { |
b30a01f2 | 3221 | struct rbd_obj_request *obj_request; |
b30a01f2 ID |
3222 | |
3223 | rbd_assert(rbd_dev->watch_event); | |
3224 | rbd_assert(rbd_dev->watch_request); | |
3225 | ||
76756a51 | 3226 | rbd_obj_request_end(rbd_dev->watch_request); |
8eb87565 AE |
3227 | rbd_obj_request_put(rbd_dev->watch_request); |
3228 | rbd_dev->watch_request = NULL; | |
b30a01f2 | 3229 | |
76756a51 ID |
3230 | obj_request = rbd_obj_watch_request_helper(rbd_dev, false); |
3231 | if (!IS_ERR(obj_request)) | |
3232 | rbd_obj_request_put(obj_request); | |
3233 | else | |
3234 | rbd_warn(rbd_dev, "unable to tear down watch request (%ld)", | |
3235 | PTR_ERR(obj_request)); | |
3236 | ||
9969ebc5 AE |
3237 | ceph_osdc_cancel_event(rbd_dev->watch_event); |
3238 | rbd_dev->watch_event = NULL; | |
fca27065 ID |
3239 | } |
3240 | ||
36be9a76 | 3241 | /* |
f40eb349 AE |
3242 | * Synchronous osd object method call. Returns the number of bytes |
3243 | * returned in the outbound buffer, or a negative error code. | |
36be9a76 AE |
3244 | */ |
3245 | static int rbd_obj_method_sync(struct rbd_device *rbd_dev, | |
3246 | const char *object_name, | |
3247 | const char *class_name, | |
3248 | const char *method_name, | |
4157976b | 3249 | const void *outbound, |
36be9a76 | 3250 | size_t outbound_size, |
4157976b | 3251 | void *inbound, |
e2a58ee5 | 3252 | size_t inbound_size) |
36be9a76 | 3253 | { |
2169238d | 3254 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
36be9a76 | 3255 | struct rbd_obj_request *obj_request; |
36be9a76 AE |
3256 | struct page **pages; |
3257 | u32 page_count; | |
3258 | int ret; | |
3259 | ||
3260 | /* | |
6010a451 AE |
3261 | * Method calls are ultimately read operations. The result |
3262 | * should placed into the inbound buffer provided. They | |
3263 | * also supply outbound data--parameters for the object | |
3264 | * method. Currently if this is present it will be a | |
3265 | * snapshot id. | |
36be9a76 | 3266 | */ |
57385b51 | 3267 | page_count = (u32)calc_pages_for(0, inbound_size); |
36be9a76 AE |
3268 | pages = ceph_alloc_page_vector(page_count, GFP_KERNEL); |
3269 | if (IS_ERR(pages)) | |
3270 | return PTR_ERR(pages); | |
3271 | ||
3272 | ret = -ENOMEM; | |
6010a451 | 3273 | obj_request = rbd_obj_request_create(object_name, 0, inbound_size, |
36be9a76 AE |
3274 | OBJ_REQUEST_PAGES); |
3275 | if (!obj_request) | |
3276 | goto out; | |
3277 | ||
3278 | obj_request->pages = pages; | |
3279 | obj_request->page_count = page_count; | |
3280 | ||
6d2940c8 | 3281 | obj_request->osd_req = rbd_osd_req_create(rbd_dev, OBJ_OP_READ, 1, |
deb236b3 | 3282 | obj_request); |
36be9a76 AE |
3283 | if (!obj_request->osd_req) |
3284 | goto out; | |
3285 | ||
c99d2d4a | 3286 | osd_req_op_cls_init(obj_request->osd_req, 0, CEPH_OSD_OP_CALL, |
04017e29 AE |
3287 | class_name, method_name); |
3288 | if (outbound_size) { | |
3289 | struct ceph_pagelist *pagelist; | |
3290 | ||
3291 | pagelist = kmalloc(sizeof (*pagelist), GFP_NOFS); | |
3292 | if (!pagelist) | |
3293 | goto out; | |
3294 | ||
3295 | ceph_pagelist_init(pagelist); | |
3296 | ceph_pagelist_append(pagelist, outbound, outbound_size); | |
3297 | osd_req_op_cls_request_data_pagelist(obj_request->osd_req, 0, | |
3298 | pagelist); | |
3299 | } | |
a4ce40a9 AE |
3300 | osd_req_op_cls_response_data_pages(obj_request->osd_req, 0, |
3301 | obj_request->pages, inbound_size, | |
44cd188d | 3302 | 0, false, false); |
9d4df01f | 3303 | rbd_osd_req_format_read(obj_request); |
430c28c3 | 3304 | |
36be9a76 AE |
3305 | ret = rbd_obj_request_submit(osdc, obj_request); |
3306 | if (ret) | |
3307 | goto out; | |
3308 | ret = rbd_obj_request_wait(obj_request); | |
3309 | if (ret) | |
3310 | goto out; | |
3311 | ||
3312 | ret = obj_request->result; | |
3313 | if (ret < 0) | |
3314 | goto out; | |
57385b51 AE |
3315 | |
3316 | rbd_assert(obj_request->xferred < (u64)INT_MAX); | |
3317 | ret = (int)obj_request->xferred; | |
903bb32e | 3318 | ceph_copy_from_page_vector(pages, inbound, 0, obj_request->xferred); |
36be9a76 AE |
3319 | out: |
3320 | if (obj_request) | |
3321 | rbd_obj_request_put(obj_request); | |
3322 | else | |
3323 | ceph_release_page_vector(pages, page_count); | |
3324 | ||
3325 | return ret; | |
3326 | } | |
3327 | ||
7ad18afa | 3328 | static void rbd_queue_workfn(struct work_struct *work) |
bf0d5f50 | 3329 | { |
7ad18afa CH |
3330 | struct request *rq = blk_mq_rq_from_pdu(work); |
3331 | struct rbd_device *rbd_dev = rq->q->queuedata; | |
bc1ecc65 | 3332 | struct rbd_img_request *img_request; |
4e752f0a | 3333 | struct ceph_snap_context *snapc = NULL; |
bc1ecc65 ID |
3334 | u64 offset = (u64)blk_rq_pos(rq) << SECTOR_SHIFT; |
3335 | u64 length = blk_rq_bytes(rq); | |
6d2940c8 | 3336 | enum obj_operation_type op_type; |
4e752f0a | 3337 | u64 mapping_size; |
bf0d5f50 AE |
3338 | int result; |
3339 | ||
7ad18afa CH |
3340 | if (rq->cmd_type != REQ_TYPE_FS) { |
3341 | dout("%s: non-fs request type %d\n", __func__, | |
3342 | (int) rq->cmd_type); | |
3343 | result = -EIO; | |
3344 | goto err; | |
3345 | } | |
3346 | ||
90e98c52 GZ |
3347 | if (rq->cmd_flags & REQ_DISCARD) |
3348 | op_type = OBJ_OP_DISCARD; | |
3349 | else if (rq->cmd_flags & REQ_WRITE) | |
6d2940c8 GZ |
3350 | op_type = OBJ_OP_WRITE; |
3351 | else | |
3352 | op_type = OBJ_OP_READ; | |
3353 | ||
bc1ecc65 | 3354 | /* Ignore/skip any zero-length requests */ |
bf0d5f50 | 3355 | |
bc1ecc65 ID |
3356 | if (!length) { |
3357 | dout("%s: zero-length request\n", __func__); | |
3358 | result = 0; | |
3359 | goto err_rq; | |
3360 | } | |
bf0d5f50 | 3361 | |
6d2940c8 | 3362 | /* Only reads are allowed to a read-only device */ |
bc1ecc65 | 3363 | |
6d2940c8 | 3364 | if (op_type != OBJ_OP_READ) { |
bc1ecc65 ID |
3365 | if (rbd_dev->mapping.read_only) { |
3366 | result = -EROFS; | |
3367 | goto err_rq; | |
4dda41d3 | 3368 | } |
bc1ecc65 ID |
3369 | rbd_assert(rbd_dev->spec->snap_id == CEPH_NOSNAP); |
3370 | } | |
4dda41d3 | 3371 | |
bc1ecc65 ID |
3372 | /* |
3373 | * Quit early if the mapped snapshot no longer exists. It's | |
3374 | * still possible the snapshot will have disappeared by the | |
3375 | * time our request arrives at the osd, but there's no sense in | |
3376 | * sending it if we already know. | |
3377 | */ | |
3378 | if (!test_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags)) { | |
3379 | dout("request for non-existent snapshot"); | |
3380 | rbd_assert(rbd_dev->spec->snap_id != CEPH_NOSNAP); | |
3381 | result = -ENXIO; | |
3382 | goto err_rq; | |
3383 | } | |
4dda41d3 | 3384 | |
bc1ecc65 ID |
3385 | if (offset && length > U64_MAX - offset + 1) { |
3386 | rbd_warn(rbd_dev, "bad request range (%llu~%llu)", offset, | |
3387 | length); | |
3388 | result = -EINVAL; | |
3389 | goto err_rq; /* Shouldn't happen */ | |
3390 | } | |
4dda41d3 | 3391 | |
7ad18afa CH |
3392 | blk_mq_start_request(rq); |
3393 | ||
4e752f0a JD |
3394 | down_read(&rbd_dev->header_rwsem); |
3395 | mapping_size = rbd_dev->mapping.size; | |
6d2940c8 | 3396 | if (op_type != OBJ_OP_READ) { |
4e752f0a JD |
3397 | snapc = rbd_dev->header.snapc; |
3398 | ceph_get_snap_context(snapc); | |
3399 | } | |
3400 | up_read(&rbd_dev->header_rwsem); | |
3401 | ||
3402 | if (offset + length > mapping_size) { | |
bc1ecc65 | 3403 | rbd_warn(rbd_dev, "beyond EOD (%llu~%llu > %llu)", offset, |
4e752f0a | 3404 | length, mapping_size); |
bc1ecc65 ID |
3405 | result = -EIO; |
3406 | goto err_rq; | |
3407 | } | |
bf0d5f50 | 3408 | |
6d2940c8 | 3409 | img_request = rbd_img_request_create(rbd_dev, offset, length, op_type, |
4e752f0a | 3410 | snapc); |
bc1ecc65 ID |
3411 | if (!img_request) { |
3412 | result = -ENOMEM; | |
3413 | goto err_rq; | |
3414 | } | |
3415 | img_request->rq = rq; | |
bf0d5f50 | 3416 | |
90e98c52 GZ |
3417 | if (op_type == OBJ_OP_DISCARD) |
3418 | result = rbd_img_request_fill(img_request, OBJ_REQUEST_NODATA, | |
3419 | NULL); | |
3420 | else | |
3421 | result = rbd_img_request_fill(img_request, OBJ_REQUEST_BIO, | |
3422 | rq->bio); | |
bc1ecc65 ID |
3423 | if (result) |
3424 | goto err_img_request; | |
bf0d5f50 | 3425 | |
bc1ecc65 ID |
3426 | result = rbd_img_request_submit(img_request); |
3427 | if (result) | |
3428 | goto err_img_request; | |
bf0d5f50 | 3429 | |
bc1ecc65 | 3430 | return; |
bf0d5f50 | 3431 | |
bc1ecc65 ID |
3432 | err_img_request: |
3433 | rbd_img_request_put(img_request); | |
3434 | err_rq: | |
3435 | if (result) | |
3436 | rbd_warn(rbd_dev, "%s %llx at %llx result %d", | |
6d2940c8 | 3437 | obj_op_name(op_type), length, offset, result); |
e96a650a | 3438 | ceph_put_snap_context(snapc); |
7ad18afa CH |
3439 | err: |
3440 | blk_mq_end_request(rq, result); | |
bc1ecc65 | 3441 | } |
bf0d5f50 | 3442 | |
7ad18afa CH |
3443 | static int rbd_queue_rq(struct blk_mq_hw_ctx *hctx, |
3444 | const struct blk_mq_queue_data *bd) | |
bc1ecc65 | 3445 | { |
7ad18afa CH |
3446 | struct request *rq = bd->rq; |
3447 | struct work_struct *work = blk_mq_rq_to_pdu(rq); | |
bf0d5f50 | 3448 | |
7ad18afa CH |
3449 | queue_work(rbd_wq, work); |
3450 | return BLK_MQ_RQ_QUEUE_OK; | |
bf0d5f50 AE |
3451 | } |
3452 | ||
602adf40 YS |
3453 | /* |
3454 | * a queue callback. Makes sure that we don't create a bio that spans across | |
3455 | * multiple osd objects. One exception would be with a single page bios, | |
f7760dad | 3456 | * which we handle later at bio_chain_clone_range() |
602adf40 YS |
3457 | */ |
3458 | static int rbd_merge_bvec(struct request_queue *q, struct bvec_merge_data *bmd, | |
3459 | struct bio_vec *bvec) | |
3460 | { | |
3461 | struct rbd_device *rbd_dev = q->queuedata; | |
e5cfeed2 AE |
3462 | sector_t sector_offset; |
3463 | sector_t sectors_per_obj; | |
3464 | sector_t obj_sector_offset; | |
3465 | int ret; | |
3466 | ||
3467 | /* | |
3468 | * Find how far into its rbd object the partition-relative | |
3469 | * bio start sector is to offset relative to the enclosing | |
3470 | * device. | |
3471 | */ | |
3472 | sector_offset = get_start_sect(bmd->bi_bdev) + bmd->bi_sector; | |
3473 | sectors_per_obj = 1 << (rbd_dev->header.obj_order - SECTOR_SHIFT); | |
3474 | obj_sector_offset = sector_offset & (sectors_per_obj - 1); | |
3475 | ||
3476 | /* | |
3477 | * Compute the number of bytes from that offset to the end | |
3478 | * of the object. Account for what's already used by the bio. | |
3479 | */ | |
3480 | ret = (int) (sectors_per_obj - obj_sector_offset) << SECTOR_SHIFT; | |
3481 | if (ret > bmd->bi_size) | |
3482 | ret -= bmd->bi_size; | |
3483 | else | |
3484 | ret = 0; | |
3485 | ||
3486 | /* | |
3487 | * Don't send back more than was asked for. And if the bio | |
3488 | * was empty, let the whole thing through because: "Note | |
3489 | * that a block device *must* allow a single page to be | |
3490 | * added to an empty bio." | |
3491 | */ | |
3492 | rbd_assert(bvec->bv_len <= PAGE_SIZE); | |
3493 | if (ret > (int) bvec->bv_len || !bmd->bi_size) | |
3494 | ret = (int) bvec->bv_len; | |
3495 | ||
3496 | return ret; | |
602adf40 YS |
3497 | } |
3498 | ||
3499 | static void rbd_free_disk(struct rbd_device *rbd_dev) | |
3500 | { | |
3501 | struct gendisk *disk = rbd_dev->disk; | |
3502 | ||
3503 | if (!disk) | |
3504 | return; | |
3505 | ||
a0cab924 AE |
3506 | rbd_dev->disk = NULL; |
3507 | if (disk->flags & GENHD_FL_UP) { | |
602adf40 | 3508 | del_gendisk(disk); |
a0cab924 AE |
3509 | if (disk->queue) |
3510 | blk_cleanup_queue(disk->queue); | |
7ad18afa | 3511 | blk_mq_free_tag_set(&rbd_dev->tag_set); |
a0cab924 | 3512 | } |
602adf40 YS |
3513 | put_disk(disk); |
3514 | } | |
3515 | ||
788e2df3 AE |
3516 | static int rbd_obj_read_sync(struct rbd_device *rbd_dev, |
3517 | const char *object_name, | |
7097f8df | 3518 | u64 offset, u64 length, void *buf) |
788e2df3 AE |
3519 | |
3520 | { | |
2169238d | 3521 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
788e2df3 | 3522 | struct rbd_obj_request *obj_request; |
788e2df3 AE |
3523 | struct page **pages = NULL; |
3524 | u32 page_count; | |
1ceae7ef | 3525 | size_t size; |
788e2df3 AE |
3526 | int ret; |
3527 | ||
3528 | page_count = (u32) calc_pages_for(offset, length); | |
3529 | pages = ceph_alloc_page_vector(page_count, GFP_KERNEL); | |
3530 | if (IS_ERR(pages)) | |
a8d42056 | 3531 | return PTR_ERR(pages); |
788e2df3 AE |
3532 | |
3533 | ret = -ENOMEM; | |
3534 | obj_request = rbd_obj_request_create(object_name, offset, length, | |
36be9a76 | 3535 | OBJ_REQUEST_PAGES); |
788e2df3 AE |
3536 | if (!obj_request) |
3537 | goto out; | |
3538 | ||
3539 | obj_request->pages = pages; | |
3540 | obj_request->page_count = page_count; | |
3541 | ||
6d2940c8 | 3542 | obj_request->osd_req = rbd_osd_req_create(rbd_dev, OBJ_OP_READ, 1, |
deb236b3 | 3543 | obj_request); |
788e2df3 AE |
3544 | if (!obj_request->osd_req) |
3545 | goto out; | |
3546 | ||
c99d2d4a AE |
3547 | osd_req_op_extent_init(obj_request->osd_req, 0, CEPH_OSD_OP_READ, |
3548 | offset, length, 0, 0); | |
406e2c9f | 3549 | osd_req_op_extent_osd_data_pages(obj_request->osd_req, 0, |
a4ce40a9 | 3550 | obj_request->pages, |
44cd188d AE |
3551 | obj_request->length, |
3552 | obj_request->offset & ~PAGE_MASK, | |
3553 | false, false); | |
9d4df01f | 3554 | rbd_osd_req_format_read(obj_request); |
430c28c3 | 3555 | |
788e2df3 AE |
3556 | ret = rbd_obj_request_submit(osdc, obj_request); |
3557 | if (ret) | |
3558 | goto out; | |
3559 | ret = rbd_obj_request_wait(obj_request); | |
3560 | if (ret) | |
3561 | goto out; | |
3562 | ||
3563 | ret = obj_request->result; | |
3564 | if (ret < 0) | |
3565 | goto out; | |
1ceae7ef AE |
3566 | |
3567 | rbd_assert(obj_request->xferred <= (u64) SIZE_MAX); | |
3568 | size = (size_t) obj_request->xferred; | |
903bb32e | 3569 | ceph_copy_from_page_vector(pages, buf, 0, size); |
7097f8df AE |
3570 | rbd_assert(size <= (size_t)INT_MAX); |
3571 | ret = (int)size; | |
788e2df3 AE |
3572 | out: |
3573 | if (obj_request) | |
3574 | rbd_obj_request_put(obj_request); | |
3575 | else | |
3576 | ceph_release_page_vector(pages, page_count); | |
3577 | ||
3578 | return ret; | |
3579 | } | |
3580 | ||
602adf40 | 3581 | /* |
662518b1 AE |
3582 | * Read the complete header for the given rbd device. On successful |
3583 | * return, the rbd_dev->header field will contain up-to-date | |
3584 | * information about the image. | |
602adf40 | 3585 | */ |
99a41ebc | 3586 | static int rbd_dev_v1_header_info(struct rbd_device *rbd_dev) |
602adf40 | 3587 | { |
4156d998 | 3588 | struct rbd_image_header_ondisk *ondisk = NULL; |
50f7c4c9 | 3589 | u32 snap_count = 0; |
4156d998 AE |
3590 | u64 names_size = 0; |
3591 | u32 want_count; | |
3592 | int ret; | |
602adf40 | 3593 | |
00f1f36f | 3594 | /* |
4156d998 AE |
3595 | * The complete header will include an array of its 64-bit |
3596 | * snapshot ids, followed by the names of those snapshots as | |
3597 | * a contiguous block of NUL-terminated strings. Note that | |
3598 | * the number of snapshots could change by the time we read | |
3599 | * it in, in which case we re-read it. | |
00f1f36f | 3600 | */ |
4156d998 AE |
3601 | do { |
3602 | size_t size; | |
3603 | ||
3604 | kfree(ondisk); | |
3605 | ||
3606 | size = sizeof (*ondisk); | |
3607 | size += snap_count * sizeof (struct rbd_image_snap_ondisk); | |
3608 | size += names_size; | |
3609 | ondisk = kmalloc(size, GFP_KERNEL); | |
3610 | if (!ondisk) | |
662518b1 | 3611 | return -ENOMEM; |
4156d998 | 3612 | |
788e2df3 | 3613 | ret = rbd_obj_read_sync(rbd_dev, rbd_dev->header_name, |
7097f8df | 3614 | 0, size, ondisk); |
4156d998 | 3615 | if (ret < 0) |
662518b1 | 3616 | goto out; |
c0cd10db | 3617 | if ((size_t)ret < size) { |
4156d998 | 3618 | ret = -ENXIO; |
06ecc6cb AE |
3619 | rbd_warn(rbd_dev, "short header read (want %zd got %d)", |
3620 | size, ret); | |
662518b1 | 3621 | goto out; |
4156d998 AE |
3622 | } |
3623 | if (!rbd_dev_ondisk_valid(ondisk)) { | |
3624 | ret = -ENXIO; | |
06ecc6cb | 3625 | rbd_warn(rbd_dev, "invalid header"); |
662518b1 | 3626 | goto out; |
81e759fb | 3627 | } |
602adf40 | 3628 | |
4156d998 AE |
3629 | names_size = le64_to_cpu(ondisk->snap_names_len); |
3630 | want_count = snap_count; | |
3631 | snap_count = le32_to_cpu(ondisk->snap_count); | |
3632 | } while (snap_count != want_count); | |
00f1f36f | 3633 | |
662518b1 AE |
3634 | ret = rbd_header_from_disk(rbd_dev, ondisk); |
3635 | out: | |
4156d998 AE |
3636 | kfree(ondisk); |
3637 | ||
3638 | return ret; | |
602adf40 YS |
3639 | } |
3640 | ||
15228ede AE |
3641 | /* |
3642 | * Clear the rbd device's EXISTS flag if the snapshot it's mapped to | |
3643 | * has disappeared from the (just updated) snapshot context. | |
3644 | */ | |
3645 | static void rbd_exists_validate(struct rbd_device *rbd_dev) | |
3646 | { | |
3647 | u64 snap_id; | |
3648 | ||
3649 | if (!test_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags)) | |
3650 | return; | |
3651 | ||
3652 | snap_id = rbd_dev->spec->snap_id; | |
3653 | if (snap_id == CEPH_NOSNAP) | |
3654 | return; | |
3655 | ||
3656 | if (rbd_dev_snap_index(rbd_dev, snap_id) == BAD_SNAP_INDEX) | |
3657 | clear_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags); | |
3658 | } | |
3659 | ||
9875201e JD |
3660 | static void rbd_dev_update_size(struct rbd_device *rbd_dev) |
3661 | { | |
3662 | sector_t size; | |
3663 | bool removing; | |
3664 | ||
3665 | /* | |
3666 | * Don't hold the lock while doing disk operations, | |
3667 | * or lock ordering will conflict with the bdev mutex via: | |
3668 | * rbd_add() -> blkdev_get() -> rbd_open() | |
3669 | */ | |
3670 | spin_lock_irq(&rbd_dev->lock); | |
3671 | removing = test_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags); | |
3672 | spin_unlock_irq(&rbd_dev->lock); | |
3673 | /* | |
3674 | * If the device is being removed, rbd_dev->disk has | |
3675 | * been destroyed, so don't try to update its size | |
3676 | */ | |
3677 | if (!removing) { | |
3678 | size = (sector_t)rbd_dev->mapping.size / SECTOR_SIZE; | |
3679 | dout("setting size to %llu sectors", (unsigned long long)size); | |
3680 | set_capacity(rbd_dev->disk, size); | |
3681 | revalidate_disk(rbd_dev->disk); | |
3682 | } | |
3683 | } | |
3684 | ||
cc4a38bd | 3685 | static int rbd_dev_refresh(struct rbd_device *rbd_dev) |
1fe5e993 | 3686 | { |
e627db08 | 3687 | u64 mapping_size; |
1fe5e993 AE |
3688 | int ret; |
3689 | ||
cfbf6377 | 3690 | down_write(&rbd_dev->header_rwsem); |
3b5cf2a2 | 3691 | mapping_size = rbd_dev->mapping.size; |
a720ae09 ID |
3692 | |
3693 | ret = rbd_dev_header_info(rbd_dev); | |
52bb1f9b | 3694 | if (ret) |
73e39e4d | 3695 | goto out; |
15228ede | 3696 | |
e8f59b59 ID |
3697 | /* |
3698 | * If there is a parent, see if it has disappeared due to the | |
3699 | * mapped image getting flattened. | |
3700 | */ | |
3701 | if (rbd_dev->parent) { | |
3702 | ret = rbd_dev_v2_parent_info(rbd_dev); | |
3703 | if (ret) | |
73e39e4d | 3704 | goto out; |
e8f59b59 ID |
3705 | } |
3706 | ||
5ff1108c | 3707 | if (rbd_dev->spec->snap_id == CEPH_NOSNAP) { |
73e39e4d | 3708 | rbd_dev->mapping.size = rbd_dev->header.image_size; |
5ff1108c ID |
3709 | } else { |
3710 | /* validate mapped snapshot's EXISTS flag */ | |
3711 | rbd_exists_validate(rbd_dev); | |
3712 | } | |
15228ede | 3713 | |
73e39e4d | 3714 | out: |
cfbf6377 | 3715 | up_write(&rbd_dev->header_rwsem); |
73e39e4d | 3716 | if (!ret && mapping_size != rbd_dev->mapping.size) |
9875201e | 3717 | rbd_dev_update_size(rbd_dev); |
1fe5e993 | 3718 | |
73e39e4d | 3719 | return ret; |
1fe5e993 AE |
3720 | } |
3721 | ||
7ad18afa CH |
3722 | static int rbd_init_request(void *data, struct request *rq, |
3723 | unsigned int hctx_idx, unsigned int request_idx, | |
3724 | unsigned int numa_node) | |
3725 | { | |
3726 | struct work_struct *work = blk_mq_rq_to_pdu(rq); | |
3727 | ||
3728 | INIT_WORK(work, rbd_queue_workfn); | |
3729 | return 0; | |
3730 | } | |
3731 | ||
3732 | static struct blk_mq_ops rbd_mq_ops = { | |
3733 | .queue_rq = rbd_queue_rq, | |
3734 | .map_queue = blk_mq_map_queue, | |
3735 | .init_request = rbd_init_request, | |
3736 | }; | |
3737 | ||
602adf40 YS |
3738 | static int rbd_init_disk(struct rbd_device *rbd_dev) |
3739 | { | |
3740 | struct gendisk *disk; | |
3741 | struct request_queue *q; | |
593a9e7b | 3742 | u64 segment_size; |
7ad18afa | 3743 | int err; |
602adf40 | 3744 | |
602adf40 | 3745 | /* create gendisk info */ |
7e513d43 ID |
3746 | disk = alloc_disk(single_major ? |
3747 | (1 << RBD_SINGLE_MAJOR_PART_SHIFT) : | |
3748 | RBD_MINORS_PER_MAJOR); | |
602adf40 | 3749 | if (!disk) |
1fcdb8aa | 3750 | return -ENOMEM; |
602adf40 | 3751 | |
f0f8cef5 | 3752 | snprintf(disk->disk_name, sizeof(disk->disk_name), RBD_DRV_NAME "%d", |
de71a297 | 3753 | rbd_dev->dev_id); |
602adf40 | 3754 | disk->major = rbd_dev->major; |
dd82fff1 | 3755 | disk->first_minor = rbd_dev->minor; |
7e513d43 ID |
3756 | if (single_major) |
3757 | disk->flags |= GENHD_FL_EXT_DEVT; | |
602adf40 YS |
3758 | disk->fops = &rbd_bd_ops; |
3759 | disk->private_data = rbd_dev; | |
3760 | ||
7ad18afa CH |
3761 | memset(&rbd_dev->tag_set, 0, sizeof(rbd_dev->tag_set)); |
3762 | rbd_dev->tag_set.ops = &rbd_mq_ops; | |
3763 | rbd_dev->tag_set.queue_depth = BLKDEV_MAX_RQ; | |
3764 | rbd_dev->tag_set.numa_node = NUMA_NO_NODE; | |
3765 | rbd_dev->tag_set.flags = | |
3766 | BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE; | |
3767 | rbd_dev->tag_set.nr_hw_queues = 1; | |
3768 | rbd_dev->tag_set.cmd_size = sizeof(struct work_struct); | |
3769 | ||
3770 | err = blk_mq_alloc_tag_set(&rbd_dev->tag_set); | |
3771 | if (err) | |
602adf40 | 3772 | goto out_disk; |
029bcbd8 | 3773 | |
7ad18afa CH |
3774 | q = blk_mq_init_queue(&rbd_dev->tag_set); |
3775 | if (IS_ERR(q)) { | |
3776 | err = PTR_ERR(q); | |
3777 | goto out_tag_set; | |
3778 | } | |
3779 | ||
d8a2c89c ID |
3780 | queue_flag_set_unlocked(QUEUE_FLAG_NONROT, q); |
3781 | /* QUEUE_FLAG_ADD_RANDOM is off by default for blk-mq */ | |
593a9e7b | 3782 | |
029bcbd8 | 3783 | /* set io sizes to object size */ |
593a9e7b AE |
3784 | segment_size = rbd_obj_bytes(&rbd_dev->header); |
3785 | blk_queue_max_hw_sectors(q, segment_size / SECTOR_SIZE); | |
d3834fef | 3786 | blk_queue_max_segments(q, segment_size / SECTOR_SIZE); |
593a9e7b AE |
3787 | blk_queue_max_segment_size(q, segment_size); |
3788 | blk_queue_io_min(q, segment_size); | |
3789 | blk_queue_io_opt(q, segment_size); | |
029bcbd8 | 3790 | |
90e98c52 GZ |
3791 | /* enable the discard support */ |
3792 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q); | |
3793 | q->limits.discard_granularity = segment_size; | |
3794 | q->limits.discard_alignment = segment_size; | |
b76f8239 JD |
3795 | q->limits.max_discard_sectors = segment_size / SECTOR_SIZE; |
3796 | q->limits.discard_zeroes_data = 1; | |
90e98c52 | 3797 | |
602adf40 YS |
3798 | blk_queue_merge_bvec(q, rbd_merge_bvec); |
3799 | disk->queue = q; | |
3800 | ||
3801 | q->queuedata = rbd_dev; | |
3802 | ||
3803 | rbd_dev->disk = disk; | |
602adf40 | 3804 | |
602adf40 | 3805 | return 0; |
7ad18afa CH |
3806 | out_tag_set: |
3807 | blk_mq_free_tag_set(&rbd_dev->tag_set); | |
602adf40 YS |
3808 | out_disk: |
3809 | put_disk(disk); | |
7ad18afa | 3810 | return err; |
602adf40 YS |
3811 | } |
3812 | ||
dfc5606d YS |
3813 | /* |
3814 | sysfs | |
3815 | */ | |
3816 | ||
593a9e7b AE |
3817 | static struct rbd_device *dev_to_rbd_dev(struct device *dev) |
3818 | { | |
3819 | return container_of(dev, struct rbd_device, dev); | |
3820 | } | |
3821 | ||
dfc5606d YS |
3822 | static ssize_t rbd_size_show(struct device *dev, |
3823 | struct device_attribute *attr, char *buf) | |
3824 | { | |
593a9e7b | 3825 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
a51aa0c0 | 3826 | |
fc71d833 AE |
3827 | return sprintf(buf, "%llu\n", |
3828 | (unsigned long long)rbd_dev->mapping.size); | |
dfc5606d YS |
3829 | } |
3830 | ||
34b13184 AE |
3831 | /* |
3832 | * Note this shows the features for whatever's mapped, which is not | |
3833 | * necessarily the base image. | |
3834 | */ | |
3835 | static ssize_t rbd_features_show(struct device *dev, | |
3836 | struct device_attribute *attr, char *buf) | |
3837 | { | |
3838 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
3839 | ||
3840 | return sprintf(buf, "0x%016llx\n", | |
fc71d833 | 3841 | (unsigned long long)rbd_dev->mapping.features); |
34b13184 AE |
3842 | } |
3843 | ||
dfc5606d YS |
3844 | static ssize_t rbd_major_show(struct device *dev, |
3845 | struct device_attribute *attr, char *buf) | |
3846 | { | |
593a9e7b | 3847 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
602adf40 | 3848 | |
fc71d833 AE |
3849 | if (rbd_dev->major) |
3850 | return sprintf(buf, "%d\n", rbd_dev->major); | |
3851 | ||
3852 | return sprintf(buf, "(none)\n"); | |
dd82fff1 ID |
3853 | } |
3854 | ||
3855 | static ssize_t rbd_minor_show(struct device *dev, | |
3856 | struct device_attribute *attr, char *buf) | |
3857 | { | |
3858 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
fc71d833 | 3859 | |
dd82fff1 | 3860 | return sprintf(buf, "%d\n", rbd_dev->minor); |
dfc5606d YS |
3861 | } |
3862 | ||
3863 | static ssize_t rbd_client_id_show(struct device *dev, | |
3864 | struct device_attribute *attr, char *buf) | |
602adf40 | 3865 | { |
593a9e7b | 3866 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 3867 | |
1dbb4399 AE |
3868 | return sprintf(buf, "client%lld\n", |
3869 | ceph_client_id(rbd_dev->rbd_client->client)); | |
602adf40 YS |
3870 | } |
3871 | ||
dfc5606d YS |
3872 | static ssize_t rbd_pool_show(struct device *dev, |
3873 | struct device_attribute *attr, char *buf) | |
602adf40 | 3874 | { |
593a9e7b | 3875 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 3876 | |
0d7dbfce | 3877 | return sprintf(buf, "%s\n", rbd_dev->spec->pool_name); |
dfc5606d YS |
3878 | } |
3879 | ||
9bb2f334 AE |
3880 | static ssize_t rbd_pool_id_show(struct device *dev, |
3881 | struct device_attribute *attr, char *buf) | |
3882 | { | |
3883 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
3884 | ||
0d7dbfce | 3885 | return sprintf(buf, "%llu\n", |
fc71d833 | 3886 | (unsigned long long) rbd_dev->spec->pool_id); |
9bb2f334 AE |
3887 | } |
3888 | ||
dfc5606d YS |
3889 | static ssize_t rbd_name_show(struct device *dev, |
3890 | struct device_attribute *attr, char *buf) | |
3891 | { | |
593a9e7b | 3892 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 3893 | |
a92ffdf8 AE |
3894 | if (rbd_dev->spec->image_name) |
3895 | return sprintf(buf, "%s\n", rbd_dev->spec->image_name); | |
3896 | ||
3897 | return sprintf(buf, "(unknown)\n"); | |
dfc5606d YS |
3898 | } |
3899 | ||
589d30e0 AE |
3900 | static ssize_t rbd_image_id_show(struct device *dev, |
3901 | struct device_attribute *attr, char *buf) | |
3902 | { | |
3903 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
3904 | ||
0d7dbfce | 3905 | return sprintf(buf, "%s\n", rbd_dev->spec->image_id); |
589d30e0 AE |
3906 | } |
3907 | ||
34b13184 AE |
3908 | /* |
3909 | * Shows the name of the currently-mapped snapshot (or | |
3910 | * RBD_SNAP_HEAD_NAME for the base image). | |
3911 | */ | |
dfc5606d YS |
3912 | static ssize_t rbd_snap_show(struct device *dev, |
3913 | struct device_attribute *attr, | |
3914 | char *buf) | |
3915 | { | |
593a9e7b | 3916 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 3917 | |
0d7dbfce | 3918 | return sprintf(buf, "%s\n", rbd_dev->spec->snap_name); |
dfc5606d YS |
3919 | } |
3920 | ||
86b00e0d | 3921 | /* |
ff96128f ID |
3922 | * For a v2 image, shows the chain of parent images, separated by empty |
3923 | * lines. For v1 images or if there is no parent, shows "(no parent | |
3924 | * image)". | |
86b00e0d AE |
3925 | */ |
3926 | static ssize_t rbd_parent_show(struct device *dev, | |
ff96128f ID |
3927 | struct device_attribute *attr, |
3928 | char *buf) | |
86b00e0d AE |
3929 | { |
3930 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
ff96128f | 3931 | ssize_t count = 0; |
86b00e0d | 3932 | |
ff96128f | 3933 | if (!rbd_dev->parent) |
86b00e0d AE |
3934 | return sprintf(buf, "(no parent image)\n"); |
3935 | ||
ff96128f ID |
3936 | for ( ; rbd_dev->parent; rbd_dev = rbd_dev->parent) { |
3937 | struct rbd_spec *spec = rbd_dev->parent_spec; | |
3938 | ||
3939 | count += sprintf(&buf[count], "%s" | |
3940 | "pool_id %llu\npool_name %s\n" | |
3941 | "image_id %s\nimage_name %s\n" | |
3942 | "snap_id %llu\nsnap_name %s\n" | |
3943 | "overlap %llu\n", | |
3944 | !count ? "" : "\n", /* first? */ | |
3945 | spec->pool_id, spec->pool_name, | |
3946 | spec->image_id, spec->image_name ?: "(unknown)", | |
3947 | spec->snap_id, spec->snap_name, | |
3948 | rbd_dev->parent_overlap); | |
3949 | } | |
3950 | ||
3951 | return count; | |
86b00e0d AE |
3952 | } |
3953 | ||
dfc5606d YS |
3954 | static ssize_t rbd_image_refresh(struct device *dev, |
3955 | struct device_attribute *attr, | |
3956 | const char *buf, | |
3957 | size_t size) | |
3958 | { | |
593a9e7b | 3959 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
b813623a | 3960 | int ret; |
602adf40 | 3961 | |
cc4a38bd | 3962 | ret = rbd_dev_refresh(rbd_dev); |
e627db08 | 3963 | if (ret) |
52bb1f9b | 3964 | return ret; |
b813623a | 3965 | |
52bb1f9b | 3966 | return size; |
dfc5606d | 3967 | } |
602adf40 | 3968 | |
dfc5606d | 3969 | static DEVICE_ATTR(size, S_IRUGO, rbd_size_show, NULL); |
34b13184 | 3970 | static DEVICE_ATTR(features, S_IRUGO, rbd_features_show, NULL); |
dfc5606d | 3971 | static DEVICE_ATTR(major, S_IRUGO, rbd_major_show, NULL); |
dd82fff1 | 3972 | static DEVICE_ATTR(minor, S_IRUGO, rbd_minor_show, NULL); |
dfc5606d YS |
3973 | static DEVICE_ATTR(client_id, S_IRUGO, rbd_client_id_show, NULL); |
3974 | static DEVICE_ATTR(pool, S_IRUGO, rbd_pool_show, NULL); | |
9bb2f334 | 3975 | static DEVICE_ATTR(pool_id, S_IRUGO, rbd_pool_id_show, NULL); |
dfc5606d | 3976 | static DEVICE_ATTR(name, S_IRUGO, rbd_name_show, NULL); |
589d30e0 | 3977 | static DEVICE_ATTR(image_id, S_IRUGO, rbd_image_id_show, NULL); |
dfc5606d YS |
3978 | static DEVICE_ATTR(refresh, S_IWUSR, NULL, rbd_image_refresh); |
3979 | static DEVICE_ATTR(current_snap, S_IRUGO, rbd_snap_show, NULL); | |
86b00e0d | 3980 | static DEVICE_ATTR(parent, S_IRUGO, rbd_parent_show, NULL); |
dfc5606d YS |
3981 | |
3982 | static struct attribute *rbd_attrs[] = { | |
3983 | &dev_attr_size.attr, | |
34b13184 | 3984 | &dev_attr_features.attr, |
dfc5606d | 3985 | &dev_attr_major.attr, |
dd82fff1 | 3986 | &dev_attr_minor.attr, |
dfc5606d YS |
3987 | &dev_attr_client_id.attr, |
3988 | &dev_attr_pool.attr, | |
9bb2f334 | 3989 | &dev_attr_pool_id.attr, |
dfc5606d | 3990 | &dev_attr_name.attr, |
589d30e0 | 3991 | &dev_attr_image_id.attr, |
dfc5606d | 3992 | &dev_attr_current_snap.attr, |
86b00e0d | 3993 | &dev_attr_parent.attr, |
dfc5606d | 3994 | &dev_attr_refresh.attr, |
dfc5606d YS |
3995 | NULL |
3996 | }; | |
3997 | ||
3998 | static struct attribute_group rbd_attr_group = { | |
3999 | .attrs = rbd_attrs, | |
4000 | }; | |
4001 | ||
4002 | static const struct attribute_group *rbd_attr_groups[] = { | |
4003 | &rbd_attr_group, | |
4004 | NULL | |
4005 | }; | |
4006 | ||
4007 | static void rbd_sysfs_dev_release(struct device *dev) | |
4008 | { | |
4009 | } | |
4010 | ||
4011 | static struct device_type rbd_device_type = { | |
4012 | .name = "rbd", | |
4013 | .groups = rbd_attr_groups, | |
4014 | .release = rbd_sysfs_dev_release, | |
4015 | }; | |
4016 | ||
8b8fb99c AE |
4017 | static struct rbd_spec *rbd_spec_get(struct rbd_spec *spec) |
4018 | { | |
4019 | kref_get(&spec->kref); | |
4020 | ||
4021 | return spec; | |
4022 | } | |
4023 | ||
4024 | static void rbd_spec_free(struct kref *kref); | |
4025 | static void rbd_spec_put(struct rbd_spec *spec) | |
4026 | { | |
4027 | if (spec) | |
4028 | kref_put(&spec->kref, rbd_spec_free); | |
4029 | } | |
4030 | ||
4031 | static struct rbd_spec *rbd_spec_alloc(void) | |
4032 | { | |
4033 | struct rbd_spec *spec; | |
4034 | ||
4035 | spec = kzalloc(sizeof (*spec), GFP_KERNEL); | |
4036 | if (!spec) | |
4037 | return NULL; | |
04077599 ID |
4038 | |
4039 | spec->pool_id = CEPH_NOPOOL; | |
4040 | spec->snap_id = CEPH_NOSNAP; | |
8b8fb99c AE |
4041 | kref_init(&spec->kref); |
4042 | ||
8b8fb99c AE |
4043 | return spec; |
4044 | } | |
4045 | ||
4046 | static void rbd_spec_free(struct kref *kref) | |
4047 | { | |
4048 | struct rbd_spec *spec = container_of(kref, struct rbd_spec, kref); | |
4049 | ||
4050 | kfree(spec->pool_name); | |
4051 | kfree(spec->image_id); | |
4052 | kfree(spec->image_name); | |
4053 | kfree(spec->snap_name); | |
4054 | kfree(spec); | |
4055 | } | |
4056 | ||
cc344fa1 | 4057 | static struct rbd_device *rbd_dev_create(struct rbd_client *rbdc, |
c53d5893 AE |
4058 | struct rbd_spec *spec) |
4059 | { | |
4060 | struct rbd_device *rbd_dev; | |
4061 | ||
4062 | rbd_dev = kzalloc(sizeof (*rbd_dev), GFP_KERNEL); | |
4063 | if (!rbd_dev) | |
4064 | return NULL; | |
4065 | ||
4066 | spin_lock_init(&rbd_dev->lock); | |
6d292906 | 4067 | rbd_dev->flags = 0; |
a2acd00e | 4068 | atomic_set(&rbd_dev->parent_ref, 0); |
c53d5893 | 4069 | INIT_LIST_HEAD(&rbd_dev->node); |
c53d5893 AE |
4070 | init_rwsem(&rbd_dev->header_rwsem); |
4071 | ||
4072 | rbd_dev->spec = spec; | |
4073 | rbd_dev->rbd_client = rbdc; | |
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 | ||
c53d5893 AE |
4082 | return rbd_dev; |
4083 | } | |
4084 | ||
4085 | static void rbd_dev_destroy(struct rbd_device *rbd_dev) | |
4086 | { | |
c53d5893 AE |
4087 | rbd_put_client(rbd_dev->rbd_client); |
4088 | rbd_spec_put(rbd_dev->spec); | |
4089 | kfree(rbd_dev); | |
4090 | } | |
4091 | ||
9d475de5 AE |
4092 | /* |
4093 | * Get the size and object order for an image snapshot, or if | |
4094 | * snap_id is CEPH_NOSNAP, gets this information for the base | |
4095 | * image. | |
4096 | */ | |
4097 | static int _rbd_dev_v2_snap_size(struct rbd_device *rbd_dev, u64 snap_id, | |
4098 | u8 *order, u64 *snap_size) | |
4099 | { | |
4100 | __le64 snapid = cpu_to_le64(snap_id); | |
4101 | int ret; | |
4102 | struct { | |
4103 | u8 order; | |
4104 | __le64 size; | |
4105 | } __attribute__ ((packed)) size_buf = { 0 }; | |
4106 | ||
36be9a76 | 4107 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
9d475de5 | 4108 | "rbd", "get_size", |
4157976b | 4109 | &snapid, sizeof (snapid), |
e2a58ee5 | 4110 | &size_buf, sizeof (size_buf)); |
36be9a76 | 4111 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
9d475de5 AE |
4112 | if (ret < 0) |
4113 | return ret; | |
57385b51 AE |
4114 | if (ret < sizeof (size_buf)) |
4115 | return -ERANGE; | |
9d475de5 | 4116 | |
c3545579 | 4117 | if (order) { |
c86f86e9 | 4118 | *order = size_buf.order; |
c3545579 JD |
4119 | dout(" order %u", (unsigned int)*order); |
4120 | } | |
9d475de5 AE |
4121 | *snap_size = le64_to_cpu(size_buf.size); |
4122 | ||
c3545579 JD |
4123 | dout(" snap_id 0x%016llx snap_size = %llu\n", |
4124 | (unsigned long long)snap_id, | |
57385b51 | 4125 | (unsigned long long)*snap_size); |
9d475de5 AE |
4126 | |
4127 | return 0; | |
4128 | } | |
4129 | ||
4130 | static int rbd_dev_v2_image_size(struct rbd_device *rbd_dev) | |
4131 | { | |
4132 | return _rbd_dev_v2_snap_size(rbd_dev, CEPH_NOSNAP, | |
4133 | &rbd_dev->header.obj_order, | |
4134 | &rbd_dev->header.image_size); | |
4135 | } | |
4136 | ||
1e130199 AE |
4137 | static int rbd_dev_v2_object_prefix(struct rbd_device *rbd_dev) |
4138 | { | |
4139 | void *reply_buf; | |
4140 | int ret; | |
4141 | void *p; | |
4142 | ||
4143 | reply_buf = kzalloc(RBD_OBJ_PREFIX_LEN_MAX, GFP_KERNEL); | |
4144 | if (!reply_buf) | |
4145 | return -ENOMEM; | |
4146 | ||
36be9a76 | 4147 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
4157976b | 4148 | "rbd", "get_object_prefix", NULL, 0, |
e2a58ee5 | 4149 | reply_buf, RBD_OBJ_PREFIX_LEN_MAX); |
36be9a76 | 4150 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
1e130199 AE |
4151 | if (ret < 0) |
4152 | goto out; | |
4153 | ||
4154 | p = reply_buf; | |
4155 | rbd_dev->header.object_prefix = ceph_extract_encoded_string(&p, | |
57385b51 AE |
4156 | p + ret, NULL, GFP_NOIO); |
4157 | ret = 0; | |
1e130199 AE |
4158 | |
4159 | if (IS_ERR(rbd_dev->header.object_prefix)) { | |
4160 | ret = PTR_ERR(rbd_dev->header.object_prefix); | |
4161 | rbd_dev->header.object_prefix = NULL; | |
4162 | } else { | |
4163 | dout(" object_prefix = %s\n", rbd_dev->header.object_prefix); | |
4164 | } | |
1e130199 AE |
4165 | out: |
4166 | kfree(reply_buf); | |
4167 | ||
4168 | return ret; | |
4169 | } | |
4170 | ||
b1b5402a AE |
4171 | static int _rbd_dev_v2_snap_features(struct rbd_device *rbd_dev, u64 snap_id, |
4172 | u64 *snap_features) | |
4173 | { | |
4174 | __le64 snapid = cpu_to_le64(snap_id); | |
4175 | struct { | |
4176 | __le64 features; | |
4177 | __le64 incompat; | |
4157976b | 4178 | } __attribute__ ((packed)) features_buf = { 0 }; |
d889140c | 4179 | u64 incompat; |
b1b5402a AE |
4180 | int ret; |
4181 | ||
36be9a76 | 4182 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
b1b5402a | 4183 | "rbd", "get_features", |
4157976b | 4184 | &snapid, sizeof (snapid), |
e2a58ee5 | 4185 | &features_buf, sizeof (features_buf)); |
36be9a76 | 4186 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
b1b5402a AE |
4187 | if (ret < 0) |
4188 | return ret; | |
57385b51 AE |
4189 | if (ret < sizeof (features_buf)) |
4190 | return -ERANGE; | |
d889140c AE |
4191 | |
4192 | incompat = le64_to_cpu(features_buf.incompat); | |
5cbf6f12 | 4193 | if (incompat & ~RBD_FEATURES_SUPPORTED) |
b8f5c6ed | 4194 | return -ENXIO; |
d889140c | 4195 | |
b1b5402a AE |
4196 | *snap_features = le64_to_cpu(features_buf.features); |
4197 | ||
4198 | dout(" snap_id 0x%016llx features = 0x%016llx incompat = 0x%016llx\n", | |
57385b51 AE |
4199 | (unsigned long long)snap_id, |
4200 | (unsigned long long)*snap_features, | |
4201 | (unsigned long long)le64_to_cpu(features_buf.incompat)); | |
b1b5402a AE |
4202 | |
4203 | return 0; | |
4204 | } | |
4205 | ||
4206 | static int rbd_dev_v2_features(struct rbd_device *rbd_dev) | |
4207 | { | |
4208 | return _rbd_dev_v2_snap_features(rbd_dev, CEPH_NOSNAP, | |
4209 | &rbd_dev->header.features); | |
4210 | } | |
4211 | ||
86b00e0d AE |
4212 | static int rbd_dev_v2_parent_info(struct rbd_device *rbd_dev) |
4213 | { | |
4214 | struct rbd_spec *parent_spec; | |
4215 | size_t size; | |
4216 | void *reply_buf = NULL; | |
4217 | __le64 snapid; | |
4218 | void *p; | |
4219 | void *end; | |
642a2537 | 4220 | u64 pool_id; |
86b00e0d | 4221 | char *image_id; |
3b5cf2a2 | 4222 | u64 snap_id; |
86b00e0d | 4223 | u64 overlap; |
86b00e0d AE |
4224 | int ret; |
4225 | ||
4226 | parent_spec = rbd_spec_alloc(); | |
4227 | if (!parent_spec) | |
4228 | return -ENOMEM; | |
4229 | ||
4230 | size = sizeof (__le64) + /* pool_id */ | |
4231 | sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX + /* image_id */ | |
4232 | sizeof (__le64) + /* snap_id */ | |
4233 | sizeof (__le64); /* overlap */ | |
4234 | reply_buf = kmalloc(size, GFP_KERNEL); | |
4235 | if (!reply_buf) { | |
4236 | ret = -ENOMEM; | |
4237 | goto out_err; | |
4238 | } | |
4239 | ||
4d9b67cd | 4240 | snapid = cpu_to_le64(rbd_dev->spec->snap_id); |
36be9a76 | 4241 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
86b00e0d | 4242 | "rbd", "get_parent", |
4157976b | 4243 | &snapid, sizeof (snapid), |
e2a58ee5 | 4244 | reply_buf, size); |
36be9a76 | 4245 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
86b00e0d AE |
4246 | if (ret < 0) |
4247 | goto out_err; | |
4248 | ||
86b00e0d | 4249 | p = reply_buf; |
57385b51 AE |
4250 | end = reply_buf + ret; |
4251 | ret = -ERANGE; | |
642a2537 | 4252 | ceph_decode_64_safe(&p, end, pool_id, out_err); |
392a9dad AE |
4253 | if (pool_id == CEPH_NOPOOL) { |
4254 | /* | |
4255 | * Either the parent never existed, or we have | |
4256 | * record of it but the image got flattened so it no | |
4257 | * longer has a parent. When the parent of a | |
4258 | * layered image disappears we immediately set the | |
4259 | * overlap to 0. The effect of this is that all new | |
4260 | * requests will be treated as if the image had no | |
4261 | * parent. | |
4262 | */ | |
4263 | if (rbd_dev->parent_overlap) { | |
4264 | rbd_dev->parent_overlap = 0; | |
392a9dad AE |
4265 | rbd_dev_parent_put(rbd_dev); |
4266 | pr_info("%s: clone image has been flattened\n", | |
4267 | rbd_dev->disk->disk_name); | |
4268 | } | |
4269 | ||
86b00e0d | 4270 | goto out; /* No parent? No problem. */ |
392a9dad | 4271 | } |
86b00e0d | 4272 | |
0903e875 AE |
4273 | /* The ceph file layout needs to fit pool id in 32 bits */ |
4274 | ||
4275 | ret = -EIO; | |
642a2537 | 4276 | if (pool_id > (u64)U32_MAX) { |
9584d508 | 4277 | rbd_warn(NULL, "parent pool id too large (%llu > %u)", |
642a2537 | 4278 | (unsigned long long)pool_id, U32_MAX); |
57385b51 | 4279 | goto out_err; |
c0cd10db | 4280 | } |
0903e875 | 4281 | |
979ed480 | 4282 | image_id = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL); |
86b00e0d AE |
4283 | if (IS_ERR(image_id)) { |
4284 | ret = PTR_ERR(image_id); | |
4285 | goto out_err; | |
4286 | } | |
3b5cf2a2 | 4287 | ceph_decode_64_safe(&p, end, snap_id, out_err); |
86b00e0d AE |
4288 | ceph_decode_64_safe(&p, end, overlap, out_err); |
4289 | ||
3b5cf2a2 AE |
4290 | /* |
4291 | * The parent won't change (except when the clone is | |
4292 | * flattened, already handled that). So we only need to | |
4293 | * record the parent spec we have not already done so. | |
4294 | */ | |
4295 | if (!rbd_dev->parent_spec) { | |
4296 | parent_spec->pool_id = pool_id; | |
4297 | parent_spec->image_id = image_id; | |
4298 | parent_spec->snap_id = snap_id; | |
70cf49cf AE |
4299 | rbd_dev->parent_spec = parent_spec; |
4300 | parent_spec = NULL; /* rbd_dev now owns this */ | |
fbba11b3 ID |
4301 | } else { |
4302 | kfree(image_id); | |
3b5cf2a2 AE |
4303 | } |
4304 | ||
4305 | /* | |
cf32bd9c ID |
4306 | * We always update the parent overlap. If it's zero we issue |
4307 | * a warning, as we will proceed as if there was no parent. | |
3b5cf2a2 | 4308 | */ |
3b5cf2a2 | 4309 | if (!overlap) { |
3b5cf2a2 | 4310 | if (parent_spec) { |
cf32bd9c ID |
4311 | /* refresh, careful to warn just once */ |
4312 | if (rbd_dev->parent_overlap) | |
4313 | rbd_warn(rbd_dev, | |
4314 | "clone now standalone (overlap became 0)"); | |
3b5cf2a2 | 4315 | } else { |
cf32bd9c ID |
4316 | /* initial probe */ |
4317 | rbd_warn(rbd_dev, "clone is standalone (overlap 0)"); | |
3b5cf2a2 | 4318 | } |
70cf49cf | 4319 | } |
cf32bd9c ID |
4320 | rbd_dev->parent_overlap = overlap; |
4321 | ||
86b00e0d AE |
4322 | out: |
4323 | ret = 0; | |
4324 | out_err: | |
4325 | kfree(reply_buf); | |
4326 | rbd_spec_put(parent_spec); | |
4327 | ||
4328 | return ret; | |
4329 | } | |
4330 | ||
cc070d59 AE |
4331 | static int rbd_dev_v2_striping_info(struct rbd_device *rbd_dev) |
4332 | { | |
4333 | struct { | |
4334 | __le64 stripe_unit; | |
4335 | __le64 stripe_count; | |
4336 | } __attribute__ ((packed)) striping_info_buf = { 0 }; | |
4337 | size_t size = sizeof (striping_info_buf); | |
4338 | void *p; | |
4339 | u64 obj_size; | |
4340 | u64 stripe_unit; | |
4341 | u64 stripe_count; | |
4342 | int ret; | |
4343 | ||
4344 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, | |
4345 | "rbd", "get_stripe_unit_count", NULL, 0, | |
e2a58ee5 | 4346 | (char *)&striping_info_buf, size); |
cc070d59 AE |
4347 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
4348 | if (ret < 0) | |
4349 | return ret; | |
4350 | if (ret < size) | |
4351 | return -ERANGE; | |
4352 | ||
4353 | /* | |
4354 | * We don't actually support the "fancy striping" feature | |
4355 | * (STRIPINGV2) yet, but if the striping sizes are the | |
4356 | * defaults the behavior is the same as before. So find | |
4357 | * out, and only fail if the image has non-default values. | |
4358 | */ | |
4359 | ret = -EINVAL; | |
4360 | obj_size = (u64)1 << rbd_dev->header.obj_order; | |
4361 | p = &striping_info_buf; | |
4362 | stripe_unit = ceph_decode_64(&p); | |
4363 | if (stripe_unit != obj_size) { | |
4364 | rbd_warn(rbd_dev, "unsupported stripe unit " | |
4365 | "(got %llu want %llu)", | |
4366 | stripe_unit, obj_size); | |
4367 | return -EINVAL; | |
4368 | } | |
4369 | stripe_count = ceph_decode_64(&p); | |
4370 | if (stripe_count != 1) { | |
4371 | rbd_warn(rbd_dev, "unsupported stripe count " | |
4372 | "(got %llu want 1)", stripe_count); | |
4373 | return -EINVAL; | |
4374 | } | |
500d0c0f AE |
4375 | rbd_dev->header.stripe_unit = stripe_unit; |
4376 | rbd_dev->header.stripe_count = stripe_count; | |
cc070d59 AE |
4377 | |
4378 | return 0; | |
4379 | } | |
4380 | ||
9e15b77d AE |
4381 | static char *rbd_dev_image_name(struct rbd_device *rbd_dev) |
4382 | { | |
4383 | size_t image_id_size; | |
4384 | char *image_id; | |
4385 | void *p; | |
4386 | void *end; | |
4387 | size_t size; | |
4388 | void *reply_buf = NULL; | |
4389 | size_t len = 0; | |
4390 | char *image_name = NULL; | |
4391 | int ret; | |
4392 | ||
4393 | rbd_assert(!rbd_dev->spec->image_name); | |
4394 | ||
69e7a02f AE |
4395 | len = strlen(rbd_dev->spec->image_id); |
4396 | image_id_size = sizeof (__le32) + len; | |
9e15b77d AE |
4397 | image_id = kmalloc(image_id_size, GFP_KERNEL); |
4398 | if (!image_id) | |
4399 | return NULL; | |
4400 | ||
4401 | p = image_id; | |
4157976b | 4402 | end = image_id + image_id_size; |
57385b51 | 4403 | ceph_encode_string(&p, end, rbd_dev->spec->image_id, (u32)len); |
9e15b77d AE |
4404 | |
4405 | size = sizeof (__le32) + RBD_IMAGE_NAME_LEN_MAX; | |
4406 | reply_buf = kmalloc(size, GFP_KERNEL); | |
4407 | if (!reply_buf) | |
4408 | goto out; | |
4409 | ||
36be9a76 | 4410 | ret = rbd_obj_method_sync(rbd_dev, RBD_DIRECTORY, |
9e15b77d AE |
4411 | "rbd", "dir_get_name", |
4412 | image_id, image_id_size, | |
e2a58ee5 | 4413 | reply_buf, size); |
9e15b77d AE |
4414 | if (ret < 0) |
4415 | goto out; | |
4416 | p = reply_buf; | |
f40eb349 AE |
4417 | end = reply_buf + ret; |
4418 | ||
9e15b77d AE |
4419 | image_name = ceph_extract_encoded_string(&p, end, &len, GFP_KERNEL); |
4420 | if (IS_ERR(image_name)) | |
4421 | image_name = NULL; | |
4422 | else | |
4423 | dout("%s: name is %s len is %zd\n", __func__, image_name, len); | |
4424 | out: | |
4425 | kfree(reply_buf); | |
4426 | kfree(image_id); | |
4427 | ||
4428 | return image_name; | |
4429 | } | |
4430 | ||
2ad3d716 AE |
4431 | static u64 rbd_v1_snap_id_by_name(struct rbd_device *rbd_dev, const char *name) |
4432 | { | |
4433 | struct ceph_snap_context *snapc = rbd_dev->header.snapc; | |
4434 | const char *snap_name; | |
4435 | u32 which = 0; | |
4436 | ||
4437 | /* Skip over names until we find the one we are looking for */ | |
4438 | ||
4439 | snap_name = rbd_dev->header.snap_names; | |
4440 | while (which < snapc->num_snaps) { | |
4441 | if (!strcmp(name, snap_name)) | |
4442 | return snapc->snaps[which]; | |
4443 | snap_name += strlen(snap_name) + 1; | |
4444 | which++; | |
4445 | } | |
4446 | return CEPH_NOSNAP; | |
4447 | } | |
4448 | ||
4449 | static u64 rbd_v2_snap_id_by_name(struct rbd_device *rbd_dev, const char *name) | |
4450 | { | |
4451 | struct ceph_snap_context *snapc = rbd_dev->header.snapc; | |
4452 | u32 which; | |
4453 | bool found = false; | |
4454 | u64 snap_id; | |
4455 | ||
4456 | for (which = 0; !found && which < snapc->num_snaps; which++) { | |
4457 | const char *snap_name; | |
4458 | ||
4459 | snap_id = snapc->snaps[which]; | |
4460 | snap_name = rbd_dev_v2_snap_name(rbd_dev, snap_id); | |
efadc98a JD |
4461 | if (IS_ERR(snap_name)) { |
4462 | /* ignore no-longer existing snapshots */ | |
4463 | if (PTR_ERR(snap_name) == -ENOENT) | |
4464 | continue; | |
4465 | else | |
4466 | break; | |
4467 | } | |
2ad3d716 AE |
4468 | found = !strcmp(name, snap_name); |
4469 | kfree(snap_name); | |
4470 | } | |
4471 | return found ? snap_id : CEPH_NOSNAP; | |
4472 | } | |
4473 | ||
4474 | /* | |
4475 | * Assumes name is never RBD_SNAP_HEAD_NAME; returns CEPH_NOSNAP if | |
4476 | * no snapshot by that name is found, or if an error occurs. | |
4477 | */ | |
4478 | static u64 rbd_snap_id_by_name(struct rbd_device *rbd_dev, const char *name) | |
4479 | { | |
4480 | if (rbd_dev->image_format == 1) | |
4481 | return rbd_v1_snap_id_by_name(rbd_dev, name); | |
4482 | ||
4483 | return rbd_v2_snap_id_by_name(rbd_dev, name); | |
4484 | } | |
4485 | ||
9e15b77d | 4486 | /* |
04077599 ID |
4487 | * An image being mapped will have everything but the snap id. |
4488 | */ | |
4489 | static int rbd_spec_fill_snap_id(struct rbd_device *rbd_dev) | |
4490 | { | |
4491 | struct rbd_spec *spec = rbd_dev->spec; | |
4492 | ||
4493 | rbd_assert(spec->pool_id != CEPH_NOPOOL && spec->pool_name); | |
4494 | rbd_assert(spec->image_id && spec->image_name); | |
4495 | rbd_assert(spec->snap_name); | |
4496 | ||
4497 | if (strcmp(spec->snap_name, RBD_SNAP_HEAD_NAME)) { | |
4498 | u64 snap_id; | |
4499 | ||
4500 | snap_id = rbd_snap_id_by_name(rbd_dev, spec->snap_name); | |
4501 | if (snap_id == CEPH_NOSNAP) | |
4502 | return -ENOENT; | |
4503 | ||
4504 | spec->snap_id = snap_id; | |
4505 | } else { | |
4506 | spec->snap_id = CEPH_NOSNAP; | |
4507 | } | |
4508 | ||
4509 | return 0; | |
4510 | } | |
4511 | ||
4512 | /* | |
4513 | * A parent image will have all ids but none of the names. | |
e1d4213f | 4514 | * |
04077599 ID |
4515 | * All names in an rbd spec are dynamically allocated. It's OK if we |
4516 | * can't figure out the name for an image id. | |
9e15b77d | 4517 | */ |
04077599 | 4518 | static int rbd_spec_fill_names(struct rbd_device *rbd_dev) |
9e15b77d | 4519 | { |
2e9f7f1c AE |
4520 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
4521 | struct rbd_spec *spec = rbd_dev->spec; | |
4522 | const char *pool_name; | |
4523 | const char *image_name; | |
4524 | const char *snap_name; | |
9e15b77d AE |
4525 | int ret; |
4526 | ||
04077599 ID |
4527 | rbd_assert(spec->pool_id != CEPH_NOPOOL); |
4528 | rbd_assert(spec->image_id); | |
4529 | rbd_assert(spec->snap_id != CEPH_NOSNAP); | |
9e15b77d | 4530 | |
2e9f7f1c | 4531 | /* Get the pool name; we have to make our own copy of this */ |
9e15b77d | 4532 | |
2e9f7f1c AE |
4533 | pool_name = ceph_pg_pool_name_by_id(osdc->osdmap, spec->pool_id); |
4534 | if (!pool_name) { | |
4535 | rbd_warn(rbd_dev, "no pool with id %llu", spec->pool_id); | |
935dc89f AE |
4536 | return -EIO; |
4537 | } | |
2e9f7f1c AE |
4538 | pool_name = kstrdup(pool_name, GFP_KERNEL); |
4539 | if (!pool_name) | |
9e15b77d AE |
4540 | return -ENOMEM; |
4541 | ||
4542 | /* Fetch the image name; tolerate failure here */ | |
4543 | ||
2e9f7f1c AE |
4544 | image_name = rbd_dev_image_name(rbd_dev); |
4545 | if (!image_name) | |
06ecc6cb | 4546 | rbd_warn(rbd_dev, "unable to get image name"); |
9e15b77d | 4547 | |
04077599 | 4548 | /* Fetch the snapshot name */ |
9e15b77d | 4549 | |
2e9f7f1c | 4550 | snap_name = rbd_snap_name(rbd_dev, spec->snap_id); |
da6a6b63 JD |
4551 | if (IS_ERR(snap_name)) { |
4552 | ret = PTR_ERR(snap_name); | |
9e15b77d | 4553 | goto out_err; |
2e9f7f1c AE |
4554 | } |
4555 | ||
4556 | spec->pool_name = pool_name; | |
4557 | spec->image_name = image_name; | |
4558 | spec->snap_name = snap_name; | |
9e15b77d AE |
4559 | |
4560 | return 0; | |
04077599 | 4561 | |
9e15b77d | 4562 | out_err: |
2e9f7f1c AE |
4563 | kfree(image_name); |
4564 | kfree(pool_name); | |
9e15b77d AE |
4565 | return ret; |
4566 | } | |
4567 | ||
cc4a38bd | 4568 | static int rbd_dev_v2_snap_context(struct rbd_device *rbd_dev) |
35d489f9 AE |
4569 | { |
4570 | size_t size; | |
4571 | int ret; | |
4572 | void *reply_buf; | |
4573 | void *p; | |
4574 | void *end; | |
4575 | u64 seq; | |
4576 | u32 snap_count; | |
4577 | struct ceph_snap_context *snapc; | |
4578 | u32 i; | |
4579 | ||
4580 | /* | |
4581 | * We'll need room for the seq value (maximum snapshot id), | |
4582 | * snapshot count, and array of that many snapshot ids. | |
4583 | * For now we have a fixed upper limit on the number we're | |
4584 | * prepared to receive. | |
4585 | */ | |
4586 | size = sizeof (__le64) + sizeof (__le32) + | |
4587 | RBD_MAX_SNAP_COUNT * sizeof (__le64); | |
4588 | reply_buf = kzalloc(size, GFP_KERNEL); | |
4589 | if (!reply_buf) | |
4590 | return -ENOMEM; | |
4591 | ||
36be9a76 | 4592 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
4157976b | 4593 | "rbd", "get_snapcontext", NULL, 0, |
e2a58ee5 | 4594 | reply_buf, size); |
36be9a76 | 4595 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
35d489f9 AE |
4596 | if (ret < 0) |
4597 | goto out; | |
4598 | ||
35d489f9 | 4599 | p = reply_buf; |
57385b51 AE |
4600 | end = reply_buf + ret; |
4601 | ret = -ERANGE; | |
35d489f9 AE |
4602 | ceph_decode_64_safe(&p, end, seq, out); |
4603 | ceph_decode_32_safe(&p, end, snap_count, out); | |
4604 | ||
4605 | /* | |
4606 | * Make sure the reported number of snapshot ids wouldn't go | |
4607 | * beyond the end of our buffer. But before checking that, | |
4608 | * make sure the computed size of the snapshot context we | |
4609 | * allocate is representable in a size_t. | |
4610 | */ | |
4611 | if (snap_count > (SIZE_MAX - sizeof (struct ceph_snap_context)) | |
4612 | / sizeof (u64)) { | |
4613 | ret = -EINVAL; | |
4614 | goto out; | |
4615 | } | |
4616 | if (!ceph_has_room(&p, end, snap_count * sizeof (__le64))) | |
4617 | goto out; | |
468521c1 | 4618 | ret = 0; |
35d489f9 | 4619 | |
812164f8 | 4620 | snapc = ceph_create_snap_context(snap_count, GFP_KERNEL); |
35d489f9 AE |
4621 | if (!snapc) { |
4622 | ret = -ENOMEM; | |
4623 | goto out; | |
4624 | } | |
35d489f9 | 4625 | snapc->seq = seq; |
35d489f9 AE |
4626 | for (i = 0; i < snap_count; i++) |
4627 | snapc->snaps[i] = ceph_decode_64(&p); | |
4628 | ||
49ece554 | 4629 | ceph_put_snap_context(rbd_dev->header.snapc); |
35d489f9 AE |
4630 | rbd_dev->header.snapc = snapc; |
4631 | ||
4632 | dout(" snap context seq = %llu, snap_count = %u\n", | |
57385b51 | 4633 | (unsigned long long)seq, (unsigned int)snap_count); |
35d489f9 AE |
4634 | out: |
4635 | kfree(reply_buf); | |
4636 | ||
57385b51 | 4637 | return ret; |
35d489f9 AE |
4638 | } |
4639 | ||
54cac61f AE |
4640 | static const char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev, |
4641 | u64 snap_id) | |
b8b1e2db AE |
4642 | { |
4643 | size_t size; | |
4644 | void *reply_buf; | |
54cac61f | 4645 | __le64 snapid; |
b8b1e2db AE |
4646 | int ret; |
4647 | void *p; | |
4648 | void *end; | |
b8b1e2db AE |
4649 | char *snap_name; |
4650 | ||
4651 | size = sizeof (__le32) + RBD_MAX_SNAP_NAME_LEN; | |
4652 | reply_buf = kmalloc(size, GFP_KERNEL); | |
4653 | if (!reply_buf) | |
4654 | return ERR_PTR(-ENOMEM); | |
4655 | ||
54cac61f | 4656 | snapid = cpu_to_le64(snap_id); |
36be9a76 | 4657 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
b8b1e2db | 4658 | "rbd", "get_snapshot_name", |
54cac61f | 4659 | &snapid, sizeof (snapid), |
e2a58ee5 | 4660 | reply_buf, size); |
36be9a76 | 4661 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
f40eb349 AE |
4662 | if (ret < 0) { |
4663 | snap_name = ERR_PTR(ret); | |
b8b1e2db | 4664 | goto out; |
f40eb349 | 4665 | } |
b8b1e2db AE |
4666 | |
4667 | p = reply_buf; | |
f40eb349 | 4668 | end = reply_buf + ret; |
e5c35534 | 4669 | snap_name = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL); |
f40eb349 | 4670 | if (IS_ERR(snap_name)) |
b8b1e2db | 4671 | goto out; |
b8b1e2db | 4672 | |
f40eb349 | 4673 | dout(" snap_id 0x%016llx snap_name = %s\n", |
54cac61f | 4674 | (unsigned long long)snap_id, snap_name); |
b8b1e2db AE |
4675 | out: |
4676 | kfree(reply_buf); | |
4677 | ||
f40eb349 | 4678 | return snap_name; |
b8b1e2db AE |
4679 | } |
4680 | ||
2df3fac7 | 4681 | static int rbd_dev_v2_header_info(struct rbd_device *rbd_dev) |
117973fb | 4682 | { |
2df3fac7 | 4683 | bool first_time = rbd_dev->header.object_prefix == NULL; |
117973fb | 4684 | int ret; |
117973fb | 4685 | |
1617e40c JD |
4686 | ret = rbd_dev_v2_image_size(rbd_dev); |
4687 | if (ret) | |
cfbf6377 | 4688 | return ret; |
1617e40c | 4689 | |
2df3fac7 AE |
4690 | if (first_time) { |
4691 | ret = rbd_dev_v2_header_onetime(rbd_dev); | |
4692 | if (ret) | |
cfbf6377 | 4693 | return ret; |
2df3fac7 AE |
4694 | } |
4695 | ||
cc4a38bd | 4696 | ret = rbd_dev_v2_snap_context(rbd_dev); |
117973fb | 4697 | dout("rbd_dev_v2_snap_context returned %d\n", ret); |
117973fb AE |
4698 | |
4699 | return ret; | |
4700 | } | |
4701 | ||
a720ae09 ID |
4702 | static int rbd_dev_header_info(struct rbd_device *rbd_dev) |
4703 | { | |
4704 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); | |
4705 | ||
4706 | if (rbd_dev->image_format == 1) | |
4707 | return rbd_dev_v1_header_info(rbd_dev); | |
4708 | ||
4709 | return rbd_dev_v2_header_info(rbd_dev); | |
4710 | } | |
4711 | ||
dfc5606d YS |
4712 | static int rbd_bus_add_dev(struct rbd_device *rbd_dev) |
4713 | { | |
dfc5606d | 4714 | struct device *dev; |
cd789ab9 | 4715 | int ret; |
dfc5606d | 4716 | |
cd789ab9 | 4717 | dev = &rbd_dev->dev; |
dfc5606d YS |
4718 | dev->bus = &rbd_bus_type; |
4719 | dev->type = &rbd_device_type; | |
4720 | dev->parent = &rbd_root_dev; | |
200a6a8b | 4721 | dev->release = rbd_dev_device_release; |
de71a297 | 4722 | dev_set_name(dev, "%d", rbd_dev->dev_id); |
dfc5606d | 4723 | ret = device_register(dev); |
dfc5606d | 4724 | |
dfc5606d | 4725 | return ret; |
602adf40 YS |
4726 | } |
4727 | ||
dfc5606d YS |
4728 | static void rbd_bus_del_dev(struct rbd_device *rbd_dev) |
4729 | { | |
4730 | device_unregister(&rbd_dev->dev); | |
4731 | } | |
4732 | ||
1ddbe94e | 4733 | /* |
499afd5b | 4734 | * Get a unique rbd identifier for the given new rbd_dev, and add |
f8a22fc2 | 4735 | * the rbd_dev to the global list. |
1ddbe94e | 4736 | */ |
f8a22fc2 | 4737 | static int rbd_dev_id_get(struct rbd_device *rbd_dev) |
b7f23c36 | 4738 | { |
f8a22fc2 ID |
4739 | int new_dev_id; |
4740 | ||
9b60e70b ID |
4741 | new_dev_id = ida_simple_get(&rbd_dev_id_ida, |
4742 | 0, minor_to_rbd_dev_id(1 << MINORBITS), | |
4743 | GFP_KERNEL); | |
f8a22fc2 ID |
4744 | if (new_dev_id < 0) |
4745 | return new_dev_id; | |
4746 | ||
4747 | rbd_dev->dev_id = new_dev_id; | |
499afd5b AE |
4748 | |
4749 | spin_lock(&rbd_dev_list_lock); | |
4750 | list_add_tail(&rbd_dev->node, &rbd_dev_list); | |
4751 | spin_unlock(&rbd_dev_list_lock); | |
f8a22fc2 | 4752 | |
70eebd20 | 4753 | dout("rbd_dev %p given dev id %d\n", rbd_dev, rbd_dev->dev_id); |
f8a22fc2 ID |
4754 | |
4755 | return 0; | |
1ddbe94e | 4756 | } |
b7f23c36 | 4757 | |
1ddbe94e | 4758 | /* |
499afd5b AE |
4759 | * Remove an rbd_dev from the global list, and record that its |
4760 | * identifier is no longer in use. | |
1ddbe94e | 4761 | */ |
e2839308 | 4762 | static void rbd_dev_id_put(struct rbd_device *rbd_dev) |
1ddbe94e | 4763 | { |
499afd5b AE |
4764 | spin_lock(&rbd_dev_list_lock); |
4765 | list_del_init(&rbd_dev->node); | |
4766 | spin_unlock(&rbd_dev_list_lock); | |
b7f23c36 | 4767 | |
f8a22fc2 ID |
4768 | ida_simple_remove(&rbd_dev_id_ida, rbd_dev->dev_id); |
4769 | ||
4770 | dout("rbd_dev %p released dev id %d\n", rbd_dev, rbd_dev->dev_id); | |
b7f23c36 AE |
4771 | } |
4772 | ||
e28fff26 AE |
4773 | /* |
4774 | * Skips over white space at *buf, and updates *buf to point to the | |
4775 | * first found non-space character (if any). Returns the length of | |
593a9e7b AE |
4776 | * the token (string of non-white space characters) found. Note |
4777 | * that *buf must be terminated with '\0'. | |
e28fff26 AE |
4778 | */ |
4779 | static inline size_t next_token(const char **buf) | |
4780 | { | |
4781 | /* | |
4782 | * These are the characters that produce nonzero for | |
4783 | * isspace() in the "C" and "POSIX" locales. | |
4784 | */ | |
4785 | const char *spaces = " \f\n\r\t\v"; | |
4786 | ||
4787 | *buf += strspn(*buf, spaces); /* Find start of token */ | |
4788 | ||
4789 | return strcspn(*buf, spaces); /* Return token length */ | |
4790 | } | |
4791 | ||
ea3352f4 AE |
4792 | /* |
4793 | * Finds the next token in *buf, dynamically allocates a buffer big | |
4794 | * enough to hold a copy of it, and copies the token into the new | |
4795 | * buffer. The copy is guaranteed to be terminated with '\0'. Note | |
4796 | * that a duplicate buffer is created even for a zero-length token. | |
4797 | * | |
4798 | * Returns a pointer to the newly-allocated duplicate, or a null | |
4799 | * pointer if memory for the duplicate was not available. If | |
4800 | * the lenp argument is a non-null pointer, the length of the token | |
4801 | * (not including the '\0') is returned in *lenp. | |
4802 | * | |
4803 | * If successful, the *buf pointer will be updated to point beyond | |
4804 | * the end of the found token. | |
4805 | * | |
4806 | * Note: uses GFP_KERNEL for allocation. | |
4807 | */ | |
4808 | static inline char *dup_token(const char **buf, size_t *lenp) | |
4809 | { | |
4810 | char *dup; | |
4811 | size_t len; | |
4812 | ||
4813 | len = next_token(buf); | |
4caf35f9 | 4814 | dup = kmemdup(*buf, len + 1, GFP_KERNEL); |
ea3352f4 AE |
4815 | if (!dup) |
4816 | return NULL; | |
ea3352f4 AE |
4817 | *(dup + len) = '\0'; |
4818 | *buf += len; | |
4819 | ||
4820 | if (lenp) | |
4821 | *lenp = len; | |
4822 | ||
4823 | return dup; | |
4824 | } | |
4825 | ||
a725f65e | 4826 | /* |
859c31df AE |
4827 | * Parse the options provided for an "rbd add" (i.e., rbd image |
4828 | * mapping) request. These arrive via a write to /sys/bus/rbd/add, | |
4829 | * and the data written is passed here via a NUL-terminated buffer. | |
4830 | * Returns 0 if successful or an error code otherwise. | |
d22f76e7 | 4831 | * |
859c31df AE |
4832 | * The information extracted from these options is recorded in |
4833 | * the other parameters which return dynamically-allocated | |
4834 | * structures: | |
4835 | * ceph_opts | |
4836 | * The address of a pointer that will refer to a ceph options | |
4837 | * structure. Caller must release the returned pointer using | |
4838 | * ceph_destroy_options() when it is no longer needed. | |
4839 | * rbd_opts | |
4840 | * Address of an rbd options pointer. Fully initialized by | |
4841 | * this function; caller must release with kfree(). | |
4842 | * spec | |
4843 | * Address of an rbd image specification pointer. Fully | |
4844 | * initialized by this function based on parsed options. | |
4845 | * Caller must release with rbd_spec_put(). | |
4846 | * | |
4847 | * The options passed take this form: | |
4848 | * <mon_addrs> <options> <pool_name> <image_name> [<snap_id>] | |
4849 | * where: | |
4850 | * <mon_addrs> | |
4851 | * A comma-separated list of one or more monitor addresses. | |
4852 | * A monitor address is an ip address, optionally followed | |
4853 | * by a port number (separated by a colon). | |
4854 | * I.e.: ip1[:port1][,ip2[:port2]...] | |
4855 | * <options> | |
4856 | * A comma-separated list of ceph and/or rbd options. | |
4857 | * <pool_name> | |
4858 | * The name of the rados pool containing the rbd image. | |
4859 | * <image_name> | |
4860 | * The name of the image in that pool to map. | |
4861 | * <snap_id> | |
4862 | * An optional snapshot id. If provided, the mapping will | |
4863 | * present data from the image at the time that snapshot was | |
4864 | * created. The image head is used if no snapshot id is | |
4865 | * provided. Snapshot mappings are always read-only. | |
a725f65e | 4866 | */ |
859c31df | 4867 | static int rbd_add_parse_args(const char *buf, |
dc79b113 | 4868 | struct ceph_options **ceph_opts, |
859c31df AE |
4869 | struct rbd_options **opts, |
4870 | struct rbd_spec **rbd_spec) | |
e28fff26 | 4871 | { |
d22f76e7 | 4872 | size_t len; |
859c31df | 4873 | char *options; |
0ddebc0c | 4874 | const char *mon_addrs; |
ecb4dc22 | 4875 | char *snap_name; |
0ddebc0c | 4876 | size_t mon_addrs_size; |
859c31df | 4877 | struct rbd_spec *spec = NULL; |
4e9afeba | 4878 | struct rbd_options *rbd_opts = NULL; |
859c31df | 4879 | struct ceph_options *copts; |
dc79b113 | 4880 | int ret; |
e28fff26 AE |
4881 | |
4882 | /* The first four tokens are required */ | |
4883 | ||
7ef3214a | 4884 | len = next_token(&buf); |
4fb5d671 AE |
4885 | if (!len) { |
4886 | rbd_warn(NULL, "no monitor address(es) provided"); | |
4887 | return -EINVAL; | |
4888 | } | |
0ddebc0c | 4889 | mon_addrs = buf; |
f28e565a | 4890 | mon_addrs_size = len + 1; |
7ef3214a | 4891 | buf += len; |
a725f65e | 4892 | |
dc79b113 | 4893 | ret = -EINVAL; |
f28e565a AE |
4894 | options = dup_token(&buf, NULL); |
4895 | if (!options) | |
dc79b113 | 4896 | return -ENOMEM; |
4fb5d671 AE |
4897 | if (!*options) { |
4898 | rbd_warn(NULL, "no options provided"); | |
4899 | goto out_err; | |
4900 | } | |
e28fff26 | 4901 | |
859c31df AE |
4902 | spec = rbd_spec_alloc(); |
4903 | if (!spec) | |
f28e565a | 4904 | goto out_mem; |
859c31df AE |
4905 | |
4906 | spec->pool_name = dup_token(&buf, NULL); | |
4907 | if (!spec->pool_name) | |
4908 | goto out_mem; | |
4fb5d671 AE |
4909 | if (!*spec->pool_name) { |
4910 | rbd_warn(NULL, "no pool name provided"); | |
4911 | goto out_err; | |
4912 | } | |
e28fff26 | 4913 | |
69e7a02f | 4914 | spec->image_name = dup_token(&buf, NULL); |
859c31df | 4915 | if (!spec->image_name) |
f28e565a | 4916 | goto out_mem; |
4fb5d671 AE |
4917 | if (!*spec->image_name) { |
4918 | rbd_warn(NULL, "no image name provided"); | |
4919 | goto out_err; | |
4920 | } | |
d4b125e9 | 4921 | |
f28e565a AE |
4922 | /* |
4923 | * Snapshot name is optional; default is to use "-" | |
4924 | * (indicating the head/no snapshot). | |
4925 | */ | |
3feeb894 | 4926 | len = next_token(&buf); |
820a5f3e | 4927 | if (!len) { |
3feeb894 AE |
4928 | buf = RBD_SNAP_HEAD_NAME; /* No snapshot supplied */ |
4929 | len = sizeof (RBD_SNAP_HEAD_NAME) - 1; | |
f28e565a | 4930 | } else if (len > RBD_MAX_SNAP_NAME_LEN) { |
dc79b113 | 4931 | ret = -ENAMETOOLONG; |
f28e565a | 4932 | goto out_err; |
849b4260 | 4933 | } |
ecb4dc22 AE |
4934 | snap_name = kmemdup(buf, len + 1, GFP_KERNEL); |
4935 | if (!snap_name) | |
f28e565a | 4936 | goto out_mem; |
ecb4dc22 AE |
4937 | *(snap_name + len) = '\0'; |
4938 | spec->snap_name = snap_name; | |
e5c35534 | 4939 | |
0ddebc0c | 4940 | /* Initialize all rbd options to the defaults */ |
e28fff26 | 4941 | |
4e9afeba AE |
4942 | rbd_opts = kzalloc(sizeof (*rbd_opts), GFP_KERNEL); |
4943 | if (!rbd_opts) | |
4944 | goto out_mem; | |
4945 | ||
4946 | rbd_opts->read_only = RBD_READ_ONLY_DEFAULT; | |
d22f76e7 | 4947 | |
859c31df | 4948 | copts = ceph_parse_options(options, mon_addrs, |
0ddebc0c | 4949 | mon_addrs + mon_addrs_size - 1, |
4e9afeba | 4950 | parse_rbd_opts_token, rbd_opts); |
859c31df AE |
4951 | if (IS_ERR(copts)) { |
4952 | ret = PTR_ERR(copts); | |
dc79b113 AE |
4953 | goto out_err; |
4954 | } | |
859c31df AE |
4955 | kfree(options); |
4956 | ||
4957 | *ceph_opts = copts; | |
4e9afeba | 4958 | *opts = rbd_opts; |
859c31df | 4959 | *rbd_spec = spec; |
0ddebc0c | 4960 | |
dc79b113 | 4961 | return 0; |
f28e565a | 4962 | out_mem: |
dc79b113 | 4963 | ret = -ENOMEM; |
d22f76e7 | 4964 | out_err: |
859c31df AE |
4965 | kfree(rbd_opts); |
4966 | rbd_spec_put(spec); | |
f28e565a | 4967 | kfree(options); |
d22f76e7 | 4968 | |
dc79b113 | 4969 | return ret; |
a725f65e AE |
4970 | } |
4971 | ||
30ba1f02 ID |
4972 | /* |
4973 | * Return pool id (>= 0) or a negative error code. | |
4974 | */ | |
4975 | static int rbd_add_get_pool_id(struct rbd_client *rbdc, const char *pool_name) | |
4976 | { | |
a319bf56 | 4977 | struct ceph_options *opts = rbdc->client->options; |
30ba1f02 | 4978 | u64 newest_epoch; |
30ba1f02 ID |
4979 | int tries = 0; |
4980 | int ret; | |
4981 | ||
4982 | again: | |
4983 | ret = ceph_pg_poolid_by_name(rbdc->client->osdc.osdmap, pool_name); | |
4984 | if (ret == -ENOENT && tries++ < 1) { | |
4985 | ret = ceph_monc_do_get_version(&rbdc->client->monc, "osdmap", | |
4986 | &newest_epoch); | |
4987 | if (ret < 0) | |
4988 | return ret; | |
4989 | ||
4990 | if (rbdc->client->osdc.osdmap->epoch < newest_epoch) { | |
4991 | ceph_monc_request_next_osdmap(&rbdc->client->monc); | |
4992 | (void) ceph_monc_wait_osdmap(&rbdc->client->monc, | |
a319bf56 ID |
4993 | newest_epoch, |
4994 | opts->mount_timeout); | |
30ba1f02 ID |
4995 | goto again; |
4996 | } else { | |
4997 | /* the osdmap we have is new enough */ | |
4998 | return -ENOENT; | |
4999 | } | |
5000 | } | |
5001 | ||
5002 | return ret; | |
5003 | } | |
5004 | ||
589d30e0 AE |
5005 | /* |
5006 | * An rbd format 2 image has a unique identifier, distinct from the | |
5007 | * name given to it by the user. Internally, that identifier is | |
5008 | * what's used to specify the names of objects related to the image. | |
5009 | * | |
5010 | * A special "rbd id" object is used to map an rbd image name to its | |
5011 | * id. If that object doesn't exist, then there is no v2 rbd image | |
5012 | * with the supplied name. | |
5013 | * | |
5014 | * This function will record the given rbd_dev's image_id field if | |
5015 | * it can be determined, and in that case will return 0. If any | |
5016 | * errors occur a negative errno will be returned and the rbd_dev's | |
5017 | * image_id field will be unchanged (and should be NULL). | |
5018 | */ | |
5019 | static int rbd_dev_image_id(struct rbd_device *rbd_dev) | |
5020 | { | |
5021 | int ret; | |
5022 | size_t size; | |
5023 | char *object_name; | |
5024 | void *response; | |
c0fba368 | 5025 | char *image_id; |
2f82ee54 | 5026 | |
2c0d0a10 AE |
5027 | /* |
5028 | * When probing a parent image, the image id is already | |
5029 | * known (and the image name likely is not). There's no | |
c0fba368 AE |
5030 | * need to fetch the image id again in this case. We |
5031 | * do still need to set the image format though. | |
2c0d0a10 | 5032 | */ |
c0fba368 AE |
5033 | if (rbd_dev->spec->image_id) { |
5034 | rbd_dev->image_format = *rbd_dev->spec->image_id ? 2 : 1; | |
5035 | ||
2c0d0a10 | 5036 | return 0; |
c0fba368 | 5037 | } |
2c0d0a10 | 5038 | |
589d30e0 AE |
5039 | /* |
5040 | * First, see if the format 2 image id file exists, and if | |
5041 | * so, get the image's persistent id from it. | |
5042 | */ | |
69e7a02f | 5043 | size = sizeof (RBD_ID_PREFIX) + strlen(rbd_dev->spec->image_name); |
589d30e0 AE |
5044 | object_name = kmalloc(size, GFP_NOIO); |
5045 | if (!object_name) | |
5046 | return -ENOMEM; | |
0d7dbfce | 5047 | sprintf(object_name, "%s%s", RBD_ID_PREFIX, rbd_dev->spec->image_name); |
589d30e0 AE |
5048 | dout("rbd id object name is %s\n", object_name); |
5049 | ||
5050 | /* Response will be an encoded string, which includes a length */ | |
5051 | ||
5052 | size = sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX; | |
5053 | response = kzalloc(size, GFP_NOIO); | |
5054 | if (!response) { | |
5055 | ret = -ENOMEM; | |
5056 | goto out; | |
5057 | } | |
5058 | ||
c0fba368 AE |
5059 | /* If it doesn't exist we'll assume it's a format 1 image */ |
5060 | ||
36be9a76 | 5061 | ret = rbd_obj_method_sync(rbd_dev, object_name, |
4157976b | 5062 | "rbd", "get_id", NULL, 0, |
e2a58ee5 | 5063 | response, RBD_IMAGE_ID_LEN_MAX); |
36be9a76 | 5064 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
c0fba368 AE |
5065 | if (ret == -ENOENT) { |
5066 | image_id = kstrdup("", GFP_KERNEL); | |
5067 | ret = image_id ? 0 : -ENOMEM; | |
5068 | if (!ret) | |
5069 | rbd_dev->image_format = 1; | |
7dd440c9 | 5070 | } else if (ret >= 0) { |
c0fba368 AE |
5071 | void *p = response; |
5072 | ||
5073 | image_id = ceph_extract_encoded_string(&p, p + ret, | |
979ed480 | 5074 | NULL, GFP_NOIO); |
461f758a | 5075 | ret = PTR_ERR_OR_ZERO(image_id); |
c0fba368 AE |
5076 | if (!ret) |
5077 | rbd_dev->image_format = 2; | |
c0fba368 AE |
5078 | } |
5079 | ||
5080 | if (!ret) { | |
5081 | rbd_dev->spec->image_id = image_id; | |
5082 | dout("image_id is %s\n", image_id); | |
589d30e0 AE |
5083 | } |
5084 | out: | |
5085 | kfree(response); | |
5086 | kfree(object_name); | |
5087 | ||
5088 | return ret; | |
5089 | } | |
5090 | ||
3abef3b3 AE |
5091 | /* |
5092 | * Undo whatever state changes are made by v1 or v2 header info | |
5093 | * call. | |
5094 | */ | |
6fd48b3b AE |
5095 | static void rbd_dev_unprobe(struct rbd_device *rbd_dev) |
5096 | { | |
5097 | struct rbd_image_header *header; | |
5098 | ||
e69b8d41 | 5099 | rbd_dev_parent_put(rbd_dev); |
6fd48b3b AE |
5100 | |
5101 | /* Free dynamic fields from the header, then zero it out */ | |
5102 | ||
5103 | header = &rbd_dev->header; | |
812164f8 | 5104 | ceph_put_snap_context(header->snapc); |
6fd48b3b AE |
5105 | kfree(header->snap_sizes); |
5106 | kfree(header->snap_names); | |
5107 | kfree(header->object_prefix); | |
5108 | memset(header, 0, sizeof (*header)); | |
5109 | } | |
5110 | ||
2df3fac7 | 5111 | static int rbd_dev_v2_header_onetime(struct rbd_device *rbd_dev) |
a30b71b9 AE |
5112 | { |
5113 | int ret; | |
a30b71b9 | 5114 | |
1e130199 | 5115 | ret = rbd_dev_v2_object_prefix(rbd_dev); |
57385b51 | 5116 | if (ret) |
b1b5402a AE |
5117 | goto out_err; |
5118 | ||
2df3fac7 AE |
5119 | /* |
5120 | * Get the and check features for the image. Currently the | |
5121 | * features are assumed to never change. | |
5122 | */ | |
b1b5402a | 5123 | ret = rbd_dev_v2_features(rbd_dev); |
57385b51 | 5124 | if (ret) |
9d475de5 | 5125 | goto out_err; |
35d489f9 | 5126 | |
cc070d59 AE |
5127 | /* If the image supports fancy striping, get its parameters */ |
5128 | ||
5129 | if (rbd_dev->header.features & RBD_FEATURE_STRIPINGV2) { | |
5130 | ret = rbd_dev_v2_striping_info(rbd_dev); | |
5131 | if (ret < 0) | |
5132 | goto out_err; | |
5133 | } | |
2df3fac7 | 5134 | /* No support for crypto and compression type format 2 images */ |
a30b71b9 | 5135 | |
35152979 | 5136 | return 0; |
9d475de5 | 5137 | out_err: |
642a2537 | 5138 | rbd_dev->header.features = 0; |
1e130199 AE |
5139 | kfree(rbd_dev->header.object_prefix); |
5140 | rbd_dev->header.object_prefix = NULL; | |
9d475de5 AE |
5141 | |
5142 | return ret; | |
a30b71b9 AE |
5143 | } |
5144 | ||
124afba2 | 5145 | static int rbd_dev_probe_parent(struct rbd_device *rbd_dev) |
83a06263 | 5146 | { |
2f82ee54 | 5147 | struct rbd_device *parent = NULL; |
124afba2 AE |
5148 | struct rbd_spec *parent_spec; |
5149 | struct rbd_client *rbdc; | |
5150 | int ret; | |
5151 | ||
5152 | if (!rbd_dev->parent_spec) | |
5153 | return 0; | |
5154 | /* | |
5155 | * We need to pass a reference to the client and the parent | |
5156 | * spec when creating the parent rbd_dev. Images related by | |
5157 | * parent/child relationships always share both. | |
5158 | */ | |
5159 | parent_spec = rbd_spec_get(rbd_dev->parent_spec); | |
5160 | rbdc = __rbd_get_client(rbd_dev->rbd_client); | |
5161 | ||
5162 | ret = -ENOMEM; | |
5163 | parent = rbd_dev_create(rbdc, parent_spec); | |
5164 | if (!parent) | |
5165 | goto out_err; | |
5166 | ||
1f3ef788 | 5167 | ret = rbd_dev_image_probe(parent, false); |
124afba2 AE |
5168 | if (ret < 0) |
5169 | goto out_err; | |
5170 | rbd_dev->parent = parent; | |
a2acd00e | 5171 | atomic_set(&rbd_dev->parent_ref, 1); |
124afba2 AE |
5172 | |
5173 | return 0; | |
5174 | out_err: | |
5175 | if (parent) { | |
fb65d228 | 5176 | rbd_dev_unparent(rbd_dev); |
124afba2 AE |
5177 | kfree(rbd_dev->header_name); |
5178 | rbd_dev_destroy(parent); | |
5179 | } else { | |
5180 | rbd_put_client(rbdc); | |
5181 | rbd_spec_put(parent_spec); | |
5182 | } | |
5183 | ||
5184 | return ret; | |
5185 | } | |
5186 | ||
200a6a8b | 5187 | static int rbd_dev_device_setup(struct rbd_device *rbd_dev) |
124afba2 | 5188 | { |
83a06263 | 5189 | int ret; |
d1cf5788 | 5190 | |
f8a22fc2 ID |
5191 | /* Get an id and fill in device name. */ |
5192 | ||
5193 | ret = rbd_dev_id_get(rbd_dev); | |
5194 | if (ret) | |
5195 | return ret; | |
83a06263 | 5196 | |
83a06263 AE |
5197 | BUILD_BUG_ON(DEV_NAME_LEN |
5198 | < sizeof (RBD_DRV_NAME) + MAX_INT_FORMAT_WIDTH); | |
5199 | sprintf(rbd_dev->name, "%s%d", RBD_DRV_NAME, rbd_dev->dev_id); | |
5200 | ||
9b60e70b | 5201 | /* Record our major and minor device numbers. */ |
83a06263 | 5202 | |
9b60e70b ID |
5203 | if (!single_major) { |
5204 | ret = register_blkdev(0, rbd_dev->name); | |
5205 | if (ret < 0) | |
5206 | goto err_out_id; | |
5207 | ||
5208 | rbd_dev->major = ret; | |
5209 | rbd_dev->minor = 0; | |
5210 | } else { | |
5211 | rbd_dev->major = rbd_major; | |
5212 | rbd_dev->minor = rbd_dev_id_to_minor(rbd_dev->dev_id); | |
5213 | } | |
83a06263 AE |
5214 | |
5215 | /* Set up the blkdev mapping. */ | |
5216 | ||
5217 | ret = rbd_init_disk(rbd_dev); | |
5218 | if (ret) | |
5219 | goto err_out_blkdev; | |
5220 | ||
f35a4dee | 5221 | ret = rbd_dev_mapping_set(rbd_dev); |
83a06263 AE |
5222 | if (ret) |
5223 | goto err_out_disk; | |
bc1ecc65 | 5224 | |
f35a4dee | 5225 | set_capacity(rbd_dev->disk, rbd_dev->mapping.size / SECTOR_SIZE); |
22001f61 | 5226 | set_disk_ro(rbd_dev->disk, rbd_dev->mapping.read_only); |
f35a4dee AE |
5227 | |
5228 | ret = rbd_bus_add_dev(rbd_dev); | |
5229 | if (ret) | |
f5ee37bd | 5230 | goto err_out_mapping; |
83a06263 | 5231 | |
83a06263 AE |
5232 | /* Everything's ready. Announce the disk to the world. */ |
5233 | ||
129b79d4 | 5234 | set_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags); |
83a06263 AE |
5235 | add_disk(rbd_dev->disk); |
5236 | ||
5237 | pr_info("%s: added with size 0x%llx\n", rbd_dev->disk->disk_name, | |
5238 | (unsigned long long) rbd_dev->mapping.size); | |
5239 | ||
5240 | return ret; | |
2f82ee54 | 5241 | |
f35a4dee AE |
5242 | err_out_mapping: |
5243 | rbd_dev_mapping_clear(rbd_dev); | |
83a06263 AE |
5244 | err_out_disk: |
5245 | rbd_free_disk(rbd_dev); | |
5246 | err_out_blkdev: | |
9b60e70b ID |
5247 | if (!single_major) |
5248 | unregister_blkdev(rbd_dev->major, rbd_dev->name); | |
83a06263 AE |
5249 | err_out_id: |
5250 | rbd_dev_id_put(rbd_dev); | |
d1cf5788 | 5251 | rbd_dev_mapping_clear(rbd_dev); |
83a06263 AE |
5252 | |
5253 | return ret; | |
5254 | } | |
5255 | ||
332bb12d AE |
5256 | static int rbd_dev_header_name(struct rbd_device *rbd_dev) |
5257 | { | |
5258 | struct rbd_spec *spec = rbd_dev->spec; | |
5259 | size_t size; | |
5260 | ||
5261 | /* Record the header object name for this rbd image. */ | |
5262 | ||
5263 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); | |
5264 | ||
5265 | if (rbd_dev->image_format == 1) | |
5266 | size = strlen(spec->image_name) + sizeof (RBD_SUFFIX); | |
5267 | else | |
5268 | size = sizeof (RBD_HEADER_PREFIX) + strlen(spec->image_id); | |
5269 | ||
5270 | rbd_dev->header_name = kmalloc(size, GFP_KERNEL); | |
5271 | if (!rbd_dev->header_name) | |
5272 | return -ENOMEM; | |
5273 | ||
5274 | if (rbd_dev->image_format == 1) | |
5275 | sprintf(rbd_dev->header_name, "%s%s", | |
5276 | spec->image_name, RBD_SUFFIX); | |
5277 | else | |
5278 | sprintf(rbd_dev->header_name, "%s%s", | |
5279 | RBD_HEADER_PREFIX, spec->image_id); | |
5280 | return 0; | |
5281 | } | |
5282 | ||
200a6a8b AE |
5283 | static void rbd_dev_image_release(struct rbd_device *rbd_dev) |
5284 | { | |
6fd48b3b | 5285 | rbd_dev_unprobe(rbd_dev); |
200a6a8b | 5286 | kfree(rbd_dev->header_name); |
6fd48b3b AE |
5287 | rbd_dev->header_name = NULL; |
5288 | rbd_dev->image_format = 0; | |
5289 | kfree(rbd_dev->spec->image_id); | |
5290 | rbd_dev->spec->image_id = NULL; | |
5291 | ||
200a6a8b AE |
5292 | rbd_dev_destroy(rbd_dev); |
5293 | } | |
5294 | ||
a30b71b9 AE |
5295 | /* |
5296 | * Probe for the existence of the header object for the given rbd | |
1f3ef788 AE |
5297 | * device. If this image is the one being mapped (i.e., not a |
5298 | * parent), initiate a watch on its header object before using that | |
5299 | * object to get detailed information about the rbd image. | |
a30b71b9 | 5300 | */ |
1f3ef788 | 5301 | static int rbd_dev_image_probe(struct rbd_device *rbd_dev, bool mapping) |
a30b71b9 AE |
5302 | { |
5303 | int ret; | |
5304 | ||
5305 | /* | |
3abef3b3 AE |
5306 | * Get the id from the image id object. Unless there's an |
5307 | * error, rbd_dev->spec->image_id will be filled in with | |
5308 | * a dynamically-allocated string, and rbd_dev->image_format | |
5309 | * will be set to either 1 or 2. | |
a30b71b9 AE |
5310 | */ |
5311 | ret = rbd_dev_image_id(rbd_dev); | |
5312 | if (ret) | |
c0fba368 | 5313 | return ret; |
c0fba368 | 5314 | |
332bb12d AE |
5315 | ret = rbd_dev_header_name(rbd_dev); |
5316 | if (ret) | |
5317 | goto err_out_format; | |
5318 | ||
1f3ef788 | 5319 | if (mapping) { |
fca27065 | 5320 | ret = rbd_dev_header_watch_sync(rbd_dev); |
1fe48023 ID |
5321 | if (ret) { |
5322 | if (ret == -ENOENT) | |
5323 | pr_info("image %s/%s does not exist\n", | |
5324 | rbd_dev->spec->pool_name, | |
5325 | rbd_dev->spec->image_name); | |
1f3ef788 | 5326 | goto out_header_name; |
1fe48023 | 5327 | } |
1f3ef788 | 5328 | } |
b644de2b | 5329 | |
a720ae09 | 5330 | ret = rbd_dev_header_info(rbd_dev); |
5655c4d9 | 5331 | if (ret) |
b644de2b | 5332 | goto err_out_watch; |
83a06263 | 5333 | |
04077599 ID |
5334 | /* |
5335 | * If this image is the one being mapped, we have pool name and | |
5336 | * id, image name and id, and snap name - need to fill snap id. | |
5337 | * Otherwise this is a parent image, identified by pool, image | |
5338 | * and snap ids - need to fill in names for those ids. | |
5339 | */ | |
5340 | if (mapping) | |
5341 | ret = rbd_spec_fill_snap_id(rbd_dev); | |
5342 | else | |
5343 | ret = rbd_spec_fill_names(rbd_dev); | |
1fe48023 ID |
5344 | if (ret) { |
5345 | if (ret == -ENOENT) | |
5346 | pr_info("snap %s/%s@%s does not exist\n", | |
5347 | rbd_dev->spec->pool_name, | |
5348 | rbd_dev->spec->image_name, | |
5349 | rbd_dev->spec->snap_name); | |
33dca39f | 5350 | goto err_out_probe; |
1fe48023 | 5351 | } |
9bb81c9b | 5352 | |
e8f59b59 ID |
5353 | if (rbd_dev->header.features & RBD_FEATURE_LAYERING) { |
5354 | ret = rbd_dev_v2_parent_info(rbd_dev); | |
5355 | if (ret) | |
5356 | goto err_out_probe; | |
5357 | ||
5358 | /* | |
5359 | * Need to warn users if this image is the one being | |
5360 | * mapped and has a parent. | |
5361 | */ | |
5362 | if (mapping && rbd_dev->parent_spec) | |
5363 | rbd_warn(rbd_dev, | |
5364 | "WARNING: kernel layering is EXPERIMENTAL!"); | |
5365 | } | |
5366 | ||
9bb81c9b | 5367 | ret = rbd_dev_probe_parent(rbd_dev); |
30d60ba2 AE |
5368 | if (ret) |
5369 | goto err_out_probe; | |
5370 | ||
5371 | dout("discovered format %u image, header name is %s\n", | |
5372 | rbd_dev->image_format, rbd_dev->header_name); | |
30d60ba2 | 5373 | return 0; |
e8f59b59 | 5374 | |
6fd48b3b AE |
5375 | err_out_probe: |
5376 | rbd_dev_unprobe(rbd_dev); | |
b644de2b | 5377 | err_out_watch: |
fca27065 ID |
5378 | if (mapping) |
5379 | rbd_dev_header_unwatch_sync(rbd_dev); | |
332bb12d AE |
5380 | out_header_name: |
5381 | kfree(rbd_dev->header_name); | |
5382 | rbd_dev->header_name = NULL; | |
5383 | err_out_format: | |
5384 | rbd_dev->image_format = 0; | |
5655c4d9 AE |
5385 | kfree(rbd_dev->spec->image_id); |
5386 | rbd_dev->spec->image_id = NULL; | |
a30b71b9 AE |
5387 | return ret; |
5388 | } | |
5389 | ||
9b60e70b ID |
5390 | static ssize_t do_rbd_add(struct bus_type *bus, |
5391 | const char *buf, | |
5392 | size_t count) | |
602adf40 | 5393 | { |
cb8627c7 | 5394 | struct rbd_device *rbd_dev = NULL; |
dc79b113 | 5395 | struct ceph_options *ceph_opts = NULL; |
4e9afeba | 5396 | struct rbd_options *rbd_opts = NULL; |
859c31df | 5397 | struct rbd_spec *spec = NULL; |
9d3997fd | 5398 | struct rbd_client *rbdc; |
51344a38 | 5399 | bool read_only; |
27cc2594 | 5400 | int rc = -ENOMEM; |
602adf40 YS |
5401 | |
5402 | if (!try_module_get(THIS_MODULE)) | |
5403 | return -ENODEV; | |
5404 | ||
602adf40 | 5405 | /* parse add command */ |
859c31df | 5406 | rc = rbd_add_parse_args(buf, &ceph_opts, &rbd_opts, &spec); |
dc79b113 | 5407 | if (rc < 0) |
bd4ba655 | 5408 | goto err_out_module; |
51344a38 AE |
5409 | read_only = rbd_opts->read_only; |
5410 | kfree(rbd_opts); | |
5411 | rbd_opts = NULL; /* done with this */ | |
78cea76e | 5412 | |
9d3997fd AE |
5413 | rbdc = rbd_get_client(ceph_opts); |
5414 | if (IS_ERR(rbdc)) { | |
5415 | rc = PTR_ERR(rbdc); | |
0ddebc0c | 5416 | goto err_out_args; |
9d3997fd | 5417 | } |
602adf40 | 5418 | |
602adf40 | 5419 | /* pick the pool */ |
30ba1f02 | 5420 | rc = rbd_add_get_pool_id(rbdc, spec->pool_name); |
1fe48023 ID |
5421 | if (rc < 0) { |
5422 | if (rc == -ENOENT) | |
5423 | pr_info("pool %s does not exist\n", spec->pool_name); | |
602adf40 | 5424 | goto err_out_client; |
1fe48023 | 5425 | } |
c0cd10db | 5426 | spec->pool_id = (u64)rc; |
859c31df | 5427 | |
0903e875 AE |
5428 | /* The ceph file layout needs to fit pool id in 32 bits */ |
5429 | ||
c0cd10db | 5430 | if (spec->pool_id > (u64)U32_MAX) { |
9584d508 | 5431 | rbd_warn(NULL, "pool id too large (%llu > %u)", |
c0cd10db | 5432 | (unsigned long long)spec->pool_id, U32_MAX); |
0903e875 AE |
5433 | rc = -EIO; |
5434 | goto err_out_client; | |
5435 | } | |
5436 | ||
c53d5893 | 5437 | rbd_dev = rbd_dev_create(rbdc, spec); |
bd4ba655 AE |
5438 | if (!rbd_dev) |
5439 | goto err_out_client; | |
c53d5893 AE |
5440 | rbdc = NULL; /* rbd_dev now owns this */ |
5441 | spec = NULL; /* rbd_dev now owns this */ | |
602adf40 | 5442 | |
1f3ef788 | 5443 | rc = rbd_dev_image_probe(rbd_dev, true); |
a30b71b9 | 5444 | if (rc < 0) |
c53d5893 | 5445 | goto err_out_rbd_dev; |
05fd6f6f | 5446 | |
7ce4eef7 AE |
5447 | /* If we are mapping a snapshot it must be marked read-only */ |
5448 | ||
5449 | if (rbd_dev->spec->snap_id != CEPH_NOSNAP) | |
5450 | read_only = true; | |
5451 | rbd_dev->mapping.read_only = read_only; | |
5452 | ||
b536f69a | 5453 | rc = rbd_dev_device_setup(rbd_dev); |
3abef3b3 | 5454 | if (rc) { |
e37180c0 ID |
5455 | /* |
5456 | * rbd_dev_header_unwatch_sync() can't be moved into | |
5457 | * rbd_dev_image_release() without refactoring, see | |
5458 | * commit 1f3ef78861ac. | |
5459 | */ | |
5460 | rbd_dev_header_unwatch_sync(rbd_dev); | |
3abef3b3 AE |
5461 | rbd_dev_image_release(rbd_dev); |
5462 | goto err_out_module; | |
5463 | } | |
5464 | ||
5465 | return count; | |
b536f69a | 5466 | |
c53d5893 AE |
5467 | err_out_rbd_dev: |
5468 | rbd_dev_destroy(rbd_dev); | |
bd4ba655 | 5469 | err_out_client: |
9d3997fd | 5470 | rbd_put_client(rbdc); |
0ddebc0c | 5471 | err_out_args: |
859c31df | 5472 | rbd_spec_put(spec); |
bd4ba655 AE |
5473 | err_out_module: |
5474 | module_put(THIS_MODULE); | |
27cc2594 | 5475 | |
602adf40 | 5476 | dout("Error adding device %s\n", buf); |
27cc2594 | 5477 | |
c0cd10db | 5478 | return (ssize_t)rc; |
602adf40 YS |
5479 | } |
5480 | ||
9b60e70b ID |
5481 | static ssize_t rbd_add(struct bus_type *bus, |
5482 | const char *buf, | |
5483 | size_t count) | |
5484 | { | |
5485 | if (single_major) | |
5486 | return -EINVAL; | |
5487 | ||
5488 | return do_rbd_add(bus, buf, count); | |
5489 | } | |
5490 | ||
5491 | static ssize_t rbd_add_single_major(struct bus_type *bus, | |
5492 | const char *buf, | |
5493 | size_t count) | |
5494 | { | |
5495 | return do_rbd_add(bus, buf, count); | |
5496 | } | |
5497 | ||
200a6a8b | 5498 | static void rbd_dev_device_release(struct device *dev) |
602adf40 | 5499 | { |
593a9e7b | 5500 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
602adf40 | 5501 | |
602adf40 | 5502 | rbd_free_disk(rbd_dev); |
200a6a8b | 5503 | clear_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags); |
6d80b130 | 5504 | rbd_dev_mapping_clear(rbd_dev); |
9b60e70b ID |
5505 | if (!single_major) |
5506 | unregister_blkdev(rbd_dev->major, rbd_dev->name); | |
e2839308 | 5507 | rbd_dev_id_put(rbd_dev); |
d1cf5788 | 5508 | rbd_dev_mapping_clear(rbd_dev); |
602adf40 YS |
5509 | } |
5510 | ||
05a46afd AE |
5511 | static void rbd_dev_remove_parent(struct rbd_device *rbd_dev) |
5512 | { | |
ad945fc1 | 5513 | while (rbd_dev->parent) { |
05a46afd AE |
5514 | struct rbd_device *first = rbd_dev; |
5515 | struct rbd_device *second = first->parent; | |
5516 | struct rbd_device *third; | |
5517 | ||
5518 | /* | |
5519 | * Follow to the parent with no grandparent and | |
5520 | * remove it. | |
5521 | */ | |
5522 | while (second && (third = second->parent)) { | |
5523 | first = second; | |
5524 | second = third; | |
5525 | } | |
ad945fc1 | 5526 | rbd_assert(second); |
8ad42cd0 | 5527 | rbd_dev_image_release(second); |
ad945fc1 AE |
5528 | first->parent = NULL; |
5529 | first->parent_overlap = 0; | |
5530 | ||
5531 | rbd_assert(first->parent_spec); | |
05a46afd AE |
5532 | rbd_spec_put(first->parent_spec); |
5533 | first->parent_spec = NULL; | |
05a46afd AE |
5534 | } |
5535 | } | |
5536 | ||
9b60e70b ID |
5537 | static ssize_t do_rbd_remove(struct bus_type *bus, |
5538 | const char *buf, | |
5539 | size_t count) | |
602adf40 YS |
5540 | { |
5541 | struct rbd_device *rbd_dev = NULL; | |
751cc0e3 AE |
5542 | struct list_head *tmp; |
5543 | int dev_id; | |
602adf40 | 5544 | unsigned long ul; |
82a442d2 | 5545 | bool already = false; |
0d8189e1 | 5546 | int ret; |
602adf40 | 5547 | |
bb8e0e84 | 5548 | ret = kstrtoul(buf, 10, &ul); |
0d8189e1 AE |
5549 | if (ret) |
5550 | return ret; | |
602adf40 YS |
5551 | |
5552 | /* convert to int; abort if we lost anything in the conversion */ | |
751cc0e3 AE |
5553 | dev_id = (int)ul; |
5554 | if (dev_id != ul) | |
602adf40 YS |
5555 | return -EINVAL; |
5556 | ||
751cc0e3 AE |
5557 | ret = -ENOENT; |
5558 | spin_lock(&rbd_dev_list_lock); | |
5559 | list_for_each(tmp, &rbd_dev_list) { | |
5560 | rbd_dev = list_entry(tmp, struct rbd_device, node); | |
5561 | if (rbd_dev->dev_id == dev_id) { | |
5562 | ret = 0; | |
5563 | break; | |
5564 | } | |
42382b70 | 5565 | } |
751cc0e3 AE |
5566 | if (!ret) { |
5567 | spin_lock_irq(&rbd_dev->lock); | |
5568 | if (rbd_dev->open_count) | |
5569 | ret = -EBUSY; | |
5570 | else | |
82a442d2 AE |
5571 | already = test_and_set_bit(RBD_DEV_FLAG_REMOVING, |
5572 | &rbd_dev->flags); | |
751cc0e3 AE |
5573 | spin_unlock_irq(&rbd_dev->lock); |
5574 | } | |
5575 | spin_unlock(&rbd_dev_list_lock); | |
82a442d2 | 5576 | if (ret < 0 || already) |
1ba0f1e7 | 5577 | return ret; |
751cc0e3 | 5578 | |
fca27065 | 5579 | rbd_dev_header_unwatch_sync(rbd_dev); |
9abc5990 JD |
5580 | /* |
5581 | * flush remaining watch callbacks - these must be complete | |
5582 | * before the osd_client is shutdown | |
5583 | */ | |
5584 | dout("%s: flushing notifies", __func__); | |
5585 | ceph_osdc_flush_notifies(&rbd_dev->rbd_client->client->osdc); | |
fca27065 | 5586 | |
9875201e JD |
5587 | /* |
5588 | * Don't free anything from rbd_dev->disk until after all | |
5589 | * notifies are completely processed. Otherwise | |
5590 | * rbd_bus_del_dev() will race with rbd_watch_cb(), resulting | |
5591 | * in a potential use after free of rbd_dev->disk or rbd_dev. | |
5592 | */ | |
5593 | rbd_bus_del_dev(rbd_dev); | |
8ad42cd0 | 5594 | rbd_dev_image_release(rbd_dev); |
79ab7558 | 5595 | module_put(THIS_MODULE); |
aafb230e | 5596 | |
1ba0f1e7 | 5597 | return count; |
602adf40 YS |
5598 | } |
5599 | ||
9b60e70b ID |
5600 | static ssize_t rbd_remove(struct bus_type *bus, |
5601 | const char *buf, | |
5602 | size_t count) | |
5603 | { | |
5604 | if (single_major) | |
5605 | return -EINVAL; | |
5606 | ||
5607 | return do_rbd_remove(bus, buf, count); | |
5608 | } | |
5609 | ||
5610 | static ssize_t rbd_remove_single_major(struct bus_type *bus, | |
5611 | const char *buf, | |
5612 | size_t count) | |
5613 | { | |
5614 | return do_rbd_remove(bus, buf, count); | |
5615 | } | |
5616 | ||
602adf40 YS |
5617 | /* |
5618 | * create control files in sysfs | |
dfc5606d | 5619 | * /sys/bus/rbd/... |
602adf40 YS |
5620 | */ |
5621 | static int rbd_sysfs_init(void) | |
5622 | { | |
dfc5606d | 5623 | int ret; |
602adf40 | 5624 | |
fed4c143 | 5625 | ret = device_register(&rbd_root_dev); |
21079786 | 5626 | if (ret < 0) |
dfc5606d | 5627 | return ret; |
602adf40 | 5628 | |
fed4c143 AE |
5629 | ret = bus_register(&rbd_bus_type); |
5630 | if (ret < 0) | |
5631 | device_unregister(&rbd_root_dev); | |
602adf40 | 5632 | |
602adf40 YS |
5633 | return ret; |
5634 | } | |
5635 | ||
5636 | static void rbd_sysfs_cleanup(void) | |
5637 | { | |
dfc5606d | 5638 | bus_unregister(&rbd_bus_type); |
fed4c143 | 5639 | device_unregister(&rbd_root_dev); |
602adf40 YS |
5640 | } |
5641 | ||
1c2a9dfe AE |
5642 | static int rbd_slab_init(void) |
5643 | { | |
5644 | rbd_assert(!rbd_img_request_cache); | |
5645 | rbd_img_request_cache = kmem_cache_create("rbd_img_request", | |
5646 | sizeof (struct rbd_img_request), | |
5647 | __alignof__(struct rbd_img_request), | |
5648 | 0, NULL); | |
868311b1 AE |
5649 | if (!rbd_img_request_cache) |
5650 | return -ENOMEM; | |
5651 | ||
5652 | rbd_assert(!rbd_obj_request_cache); | |
5653 | rbd_obj_request_cache = kmem_cache_create("rbd_obj_request", | |
5654 | sizeof (struct rbd_obj_request), | |
5655 | __alignof__(struct rbd_obj_request), | |
5656 | 0, NULL); | |
78c2a44a AE |
5657 | if (!rbd_obj_request_cache) |
5658 | goto out_err; | |
5659 | ||
5660 | rbd_assert(!rbd_segment_name_cache); | |
5661 | rbd_segment_name_cache = kmem_cache_create("rbd_segment_name", | |
2d0ebc5d | 5662 | CEPH_MAX_OID_NAME_LEN + 1, 1, 0, NULL); |
78c2a44a | 5663 | if (rbd_segment_name_cache) |
1c2a9dfe | 5664 | return 0; |
78c2a44a AE |
5665 | out_err: |
5666 | if (rbd_obj_request_cache) { | |
5667 | kmem_cache_destroy(rbd_obj_request_cache); | |
5668 | rbd_obj_request_cache = NULL; | |
5669 | } | |
1c2a9dfe | 5670 | |
868311b1 AE |
5671 | kmem_cache_destroy(rbd_img_request_cache); |
5672 | rbd_img_request_cache = NULL; | |
5673 | ||
1c2a9dfe AE |
5674 | return -ENOMEM; |
5675 | } | |
5676 | ||
5677 | static void rbd_slab_exit(void) | |
5678 | { | |
78c2a44a AE |
5679 | rbd_assert(rbd_segment_name_cache); |
5680 | kmem_cache_destroy(rbd_segment_name_cache); | |
5681 | rbd_segment_name_cache = NULL; | |
5682 | ||
868311b1 AE |
5683 | rbd_assert(rbd_obj_request_cache); |
5684 | kmem_cache_destroy(rbd_obj_request_cache); | |
5685 | rbd_obj_request_cache = NULL; | |
5686 | ||
1c2a9dfe AE |
5687 | rbd_assert(rbd_img_request_cache); |
5688 | kmem_cache_destroy(rbd_img_request_cache); | |
5689 | rbd_img_request_cache = NULL; | |
5690 | } | |
5691 | ||
cc344fa1 | 5692 | static int __init rbd_init(void) |
602adf40 YS |
5693 | { |
5694 | int rc; | |
5695 | ||
1e32d34c AE |
5696 | if (!libceph_compatible(NULL)) { |
5697 | rbd_warn(NULL, "libceph incompatibility (quitting)"); | |
1e32d34c AE |
5698 | return -EINVAL; |
5699 | } | |
e1b4d96d | 5700 | |
1c2a9dfe | 5701 | rc = rbd_slab_init(); |
602adf40 YS |
5702 | if (rc) |
5703 | return rc; | |
e1b4d96d | 5704 | |
f5ee37bd ID |
5705 | /* |
5706 | * The number of active work items is limited by the number of | |
f77303bd | 5707 | * rbd devices * queue depth, so leave @max_active at default. |
f5ee37bd ID |
5708 | */ |
5709 | rbd_wq = alloc_workqueue(RBD_DRV_NAME, WQ_MEM_RECLAIM, 0); | |
5710 | if (!rbd_wq) { | |
5711 | rc = -ENOMEM; | |
5712 | goto err_out_slab; | |
5713 | } | |
5714 | ||
9b60e70b ID |
5715 | if (single_major) { |
5716 | rbd_major = register_blkdev(0, RBD_DRV_NAME); | |
5717 | if (rbd_major < 0) { | |
5718 | rc = rbd_major; | |
f5ee37bd | 5719 | goto err_out_wq; |
9b60e70b ID |
5720 | } |
5721 | } | |
5722 | ||
1c2a9dfe AE |
5723 | rc = rbd_sysfs_init(); |
5724 | if (rc) | |
9b60e70b ID |
5725 | goto err_out_blkdev; |
5726 | ||
5727 | if (single_major) | |
5728 | pr_info("loaded (major %d)\n", rbd_major); | |
5729 | else | |
5730 | pr_info("loaded\n"); | |
1c2a9dfe | 5731 | |
e1b4d96d ID |
5732 | return 0; |
5733 | ||
9b60e70b ID |
5734 | err_out_blkdev: |
5735 | if (single_major) | |
5736 | unregister_blkdev(rbd_major, RBD_DRV_NAME); | |
f5ee37bd ID |
5737 | err_out_wq: |
5738 | destroy_workqueue(rbd_wq); | |
e1b4d96d ID |
5739 | err_out_slab: |
5740 | rbd_slab_exit(); | |
1c2a9dfe | 5741 | return rc; |
602adf40 YS |
5742 | } |
5743 | ||
cc344fa1 | 5744 | static void __exit rbd_exit(void) |
602adf40 | 5745 | { |
ffe312cf | 5746 | ida_destroy(&rbd_dev_id_ida); |
602adf40 | 5747 | rbd_sysfs_cleanup(); |
9b60e70b ID |
5748 | if (single_major) |
5749 | unregister_blkdev(rbd_major, RBD_DRV_NAME); | |
f5ee37bd | 5750 | destroy_workqueue(rbd_wq); |
1c2a9dfe | 5751 | rbd_slab_exit(); |
602adf40 YS |
5752 | } |
5753 | ||
5754 | module_init(rbd_init); | |
5755 | module_exit(rbd_exit); | |
5756 | ||
d552c619 | 5757 | MODULE_AUTHOR("Alex Elder <elder@inktank.com>"); |
602adf40 YS |
5758 | MODULE_AUTHOR("Sage Weil <sage@newdream.net>"); |
5759 | MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>"); | |
602adf40 YS |
5760 | /* following authorship retained from original osdblk.c */ |
5761 | MODULE_AUTHOR("Jeff Garzik <jeff@garzik.org>"); | |
5762 | ||
90da258b | 5763 | MODULE_DESCRIPTION("RADOS Block Device (RBD) driver"); |
602adf40 | 5764 | MODULE_LICENSE("GPL"); |