]>
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; |
d147543d | 349 | struct rbd_options *opts; |
602adf40 | 350 | |
0d7dbfce | 351 | char *header_name; |
971f839a | 352 | |
0903e875 AE |
353 | struct ceph_file_layout layout; |
354 | ||
59c2be1e | 355 | struct ceph_osd_event *watch_event; |
975241af | 356 | struct rbd_obj_request *watch_request; |
59c2be1e | 357 | |
86b00e0d AE |
358 | struct rbd_spec *parent_spec; |
359 | u64 parent_overlap; | |
a2acd00e | 360 | atomic_t parent_ref; |
2f82ee54 | 361 | struct rbd_device *parent; |
86b00e0d | 362 | |
7ad18afa CH |
363 | /* Block layer tags. */ |
364 | struct blk_mq_tag_set tag_set; | |
365 | ||
c666601a JD |
366 | /* protects updating the header */ |
367 | struct rw_semaphore header_rwsem; | |
f84344f3 AE |
368 | |
369 | struct rbd_mapping mapping; | |
602adf40 YS |
370 | |
371 | struct list_head node; | |
dfc5606d | 372 | |
dfc5606d YS |
373 | /* sysfs related */ |
374 | struct device dev; | |
b82d167b | 375 | unsigned long open_count; /* protected by lock */ |
dfc5606d YS |
376 | }; |
377 | ||
b82d167b AE |
378 | /* |
379 | * Flag bits for rbd_dev->flags. If atomicity is required, | |
380 | * rbd_dev->lock is used to protect access. | |
381 | * | |
382 | * Currently, only the "removing" flag (which is coupled with the | |
383 | * "open_count" field) requires atomic access. | |
384 | */ | |
6d292906 AE |
385 | enum rbd_dev_flags { |
386 | RBD_DEV_FLAG_EXISTS, /* mapped snapshot has not been deleted */ | |
b82d167b | 387 | RBD_DEV_FLAG_REMOVING, /* this mapping is being removed */ |
6d292906 AE |
388 | }; |
389 | ||
cfbf6377 | 390 | static DEFINE_MUTEX(client_mutex); /* Serialize client creation */ |
e124a82f | 391 | |
602adf40 | 392 | static LIST_HEAD(rbd_dev_list); /* devices */ |
e124a82f AE |
393 | static DEFINE_SPINLOCK(rbd_dev_list_lock); |
394 | ||
432b8587 AE |
395 | static LIST_HEAD(rbd_client_list); /* clients */ |
396 | static DEFINE_SPINLOCK(rbd_client_list_lock); | |
602adf40 | 397 | |
78c2a44a AE |
398 | /* Slab caches for frequently-allocated structures */ |
399 | ||
1c2a9dfe | 400 | static struct kmem_cache *rbd_img_request_cache; |
868311b1 | 401 | static struct kmem_cache *rbd_obj_request_cache; |
78c2a44a | 402 | static struct kmem_cache *rbd_segment_name_cache; |
1c2a9dfe | 403 | |
9b60e70b | 404 | static int rbd_major; |
f8a22fc2 ID |
405 | static DEFINE_IDA(rbd_dev_id_ida); |
406 | ||
f5ee37bd ID |
407 | static struct workqueue_struct *rbd_wq; |
408 | ||
9b60e70b ID |
409 | /* |
410 | * Default to false for now, as single-major requires >= 0.75 version of | |
411 | * userspace rbd utility. | |
412 | */ | |
413 | static bool single_major = false; | |
414 | module_param(single_major, bool, S_IRUGO); | |
415 | MODULE_PARM_DESC(single_major, "Use a single major number for all rbd devices (default: false)"); | |
416 | ||
3d7efd18 AE |
417 | static int rbd_img_request_submit(struct rbd_img_request *img_request); |
418 | ||
200a6a8b | 419 | static void rbd_dev_device_release(struct device *dev); |
dfc5606d | 420 | |
f0f8cef5 AE |
421 | static ssize_t rbd_add(struct bus_type *bus, const char *buf, |
422 | size_t count); | |
423 | static ssize_t rbd_remove(struct bus_type *bus, const char *buf, | |
424 | size_t count); | |
9b60e70b ID |
425 | static ssize_t rbd_add_single_major(struct bus_type *bus, const char *buf, |
426 | size_t count); | |
427 | static ssize_t rbd_remove_single_major(struct bus_type *bus, const char *buf, | |
428 | size_t count); | |
1f3ef788 | 429 | static int rbd_dev_image_probe(struct rbd_device *rbd_dev, bool mapping); |
a2acd00e | 430 | static void rbd_spec_put(struct rbd_spec *spec); |
f0f8cef5 | 431 | |
9b60e70b ID |
432 | static int rbd_dev_id_to_minor(int dev_id) |
433 | { | |
7e513d43 | 434 | return dev_id << RBD_SINGLE_MAJOR_PART_SHIFT; |
9b60e70b ID |
435 | } |
436 | ||
437 | static int minor_to_rbd_dev_id(int minor) | |
438 | { | |
7e513d43 | 439 | return minor >> RBD_SINGLE_MAJOR_PART_SHIFT; |
9b60e70b ID |
440 | } |
441 | ||
b15a21dd GKH |
442 | static BUS_ATTR(add, S_IWUSR, NULL, rbd_add); |
443 | static BUS_ATTR(remove, S_IWUSR, NULL, rbd_remove); | |
9b60e70b ID |
444 | static BUS_ATTR(add_single_major, S_IWUSR, NULL, rbd_add_single_major); |
445 | static BUS_ATTR(remove_single_major, S_IWUSR, NULL, rbd_remove_single_major); | |
b15a21dd GKH |
446 | |
447 | static struct attribute *rbd_bus_attrs[] = { | |
448 | &bus_attr_add.attr, | |
449 | &bus_attr_remove.attr, | |
9b60e70b ID |
450 | &bus_attr_add_single_major.attr, |
451 | &bus_attr_remove_single_major.attr, | |
b15a21dd | 452 | NULL, |
f0f8cef5 | 453 | }; |
92c76dc0 ID |
454 | |
455 | static umode_t rbd_bus_is_visible(struct kobject *kobj, | |
456 | struct attribute *attr, int index) | |
457 | { | |
9b60e70b ID |
458 | if (!single_major && |
459 | (attr == &bus_attr_add_single_major.attr || | |
460 | attr == &bus_attr_remove_single_major.attr)) | |
461 | return 0; | |
462 | ||
92c76dc0 ID |
463 | return attr->mode; |
464 | } | |
465 | ||
466 | static const struct attribute_group rbd_bus_group = { | |
467 | .attrs = rbd_bus_attrs, | |
468 | .is_visible = rbd_bus_is_visible, | |
469 | }; | |
470 | __ATTRIBUTE_GROUPS(rbd_bus); | |
f0f8cef5 AE |
471 | |
472 | static struct bus_type rbd_bus_type = { | |
473 | .name = "rbd", | |
b15a21dd | 474 | .bus_groups = rbd_bus_groups, |
f0f8cef5 AE |
475 | }; |
476 | ||
477 | static void rbd_root_dev_release(struct device *dev) | |
478 | { | |
479 | } | |
480 | ||
481 | static struct device rbd_root_dev = { | |
482 | .init_name = "rbd", | |
483 | .release = rbd_root_dev_release, | |
484 | }; | |
485 | ||
06ecc6cb AE |
486 | static __printf(2, 3) |
487 | void rbd_warn(struct rbd_device *rbd_dev, const char *fmt, ...) | |
488 | { | |
489 | struct va_format vaf; | |
490 | va_list args; | |
491 | ||
492 | va_start(args, fmt); | |
493 | vaf.fmt = fmt; | |
494 | vaf.va = &args; | |
495 | ||
496 | if (!rbd_dev) | |
497 | printk(KERN_WARNING "%s: %pV\n", RBD_DRV_NAME, &vaf); | |
498 | else if (rbd_dev->disk) | |
499 | printk(KERN_WARNING "%s: %s: %pV\n", | |
500 | RBD_DRV_NAME, rbd_dev->disk->disk_name, &vaf); | |
501 | else if (rbd_dev->spec && rbd_dev->spec->image_name) | |
502 | printk(KERN_WARNING "%s: image %s: %pV\n", | |
503 | RBD_DRV_NAME, rbd_dev->spec->image_name, &vaf); | |
504 | else if (rbd_dev->spec && rbd_dev->spec->image_id) | |
505 | printk(KERN_WARNING "%s: id %s: %pV\n", | |
506 | RBD_DRV_NAME, rbd_dev->spec->image_id, &vaf); | |
507 | else /* punt */ | |
508 | printk(KERN_WARNING "%s: rbd_dev %p: %pV\n", | |
509 | RBD_DRV_NAME, rbd_dev, &vaf); | |
510 | va_end(args); | |
511 | } | |
512 | ||
aafb230e AE |
513 | #ifdef RBD_DEBUG |
514 | #define rbd_assert(expr) \ | |
515 | if (unlikely(!(expr))) { \ | |
516 | printk(KERN_ERR "\nAssertion failure in %s() " \ | |
517 | "at line %d:\n\n" \ | |
518 | "\trbd_assert(%s);\n\n", \ | |
519 | __func__, __LINE__, #expr); \ | |
520 | BUG(); \ | |
521 | } | |
522 | #else /* !RBD_DEBUG */ | |
523 | # define rbd_assert(expr) ((void) 0) | |
524 | #endif /* !RBD_DEBUG */ | |
dfc5606d | 525 | |
b454e36d | 526 | static int rbd_img_obj_request_submit(struct rbd_obj_request *obj_request); |
05a46afd AE |
527 | static void rbd_img_parent_read(struct rbd_obj_request *obj_request); |
528 | static void rbd_dev_remove_parent(struct rbd_device *rbd_dev); | |
8b3e1a56 | 529 | |
cc4a38bd | 530 | static int rbd_dev_refresh(struct rbd_device *rbd_dev); |
2df3fac7 | 531 | static int rbd_dev_v2_header_onetime(struct rbd_device *rbd_dev); |
a720ae09 | 532 | static int rbd_dev_header_info(struct rbd_device *rbd_dev); |
e8f59b59 | 533 | static int rbd_dev_v2_parent_info(struct rbd_device *rbd_dev); |
54cac61f AE |
534 | static const char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev, |
535 | u64 snap_id); | |
2ad3d716 AE |
536 | static int _rbd_dev_v2_snap_size(struct rbd_device *rbd_dev, u64 snap_id, |
537 | u8 *order, u64 *snap_size); | |
538 | static int _rbd_dev_v2_snap_features(struct rbd_device *rbd_dev, u64 snap_id, | |
539 | u64 *snap_features); | |
540 | static u64 rbd_snap_id_by_name(struct rbd_device *rbd_dev, const char *name); | |
59c2be1e | 541 | |
602adf40 YS |
542 | static int rbd_open(struct block_device *bdev, fmode_t mode) |
543 | { | |
f0f8cef5 | 544 | struct rbd_device *rbd_dev = bdev->bd_disk->private_data; |
b82d167b | 545 | bool removing = false; |
602adf40 | 546 | |
f84344f3 | 547 | if ((mode & FMODE_WRITE) && rbd_dev->mapping.read_only) |
602adf40 YS |
548 | return -EROFS; |
549 | ||
a14ea269 | 550 | spin_lock_irq(&rbd_dev->lock); |
b82d167b AE |
551 | if (test_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags)) |
552 | removing = true; | |
553 | else | |
554 | rbd_dev->open_count++; | |
a14ea269 | 555 | spin_unlock_irq(&rbd_dev->lock); |
b82d167b AE |
556 | if (removing) |
557 | return -ENOENT; | |
558 | ||
c3e946ce | 559 | (void) get_device(&rbd_dev->dev); |
340c7a2b | 560 | |
602adf40 YS |
561 | return 0; |
562 | } | |
563 | ||
db2a144b | 564 | static void rbd_release(struct gendisk *disk, fmode_t mode) |
dfc5606d YS |
565 | { |
566 | struct rbd_device *rbd_dev = disk->private_data; | |
b82d167b AE |
567 | unsigned long open_count_before; |
568 | ||
a14ea269 | 569 | spin_lock_irq(&rbd_dev->lock); |
b82d167b | 570 | open_count_before = rbd_dev->open_count--; |
a14ea269 | 571 | spin_unlock_irq(&rbd_dev->lock); |
b82d167b | 572 | rbd_assert(open_count_before > 0); |
dfc5606d | 573 | |
c3e946ce | 574 | put_device(&rbd_dev->dev); |
dfc5606d YS |
575 | } |
576 | ||
131fd9f6 GZ |
577 | static int rbd_ioctl_set_ro(struct rbd_device *rbd_dev, unsigned long arg) |
578 | { | |
77f33c03 | 579 | int ret = 0; |
131fd9f6 GZ |
580 | int val; |
581 | bool ro; | |
77f33c03 | 582 | bool ro_changed = false; |
131fd9f6 | 583 | |
77f33c03 | 584 | /* get_user() may sleep, so call it before taking rbd_dev->lock */ |
131fd9f6 GZ |
585 | if (get_user(val, (int __user *)(arg))) |
586 | return -EFAULT; | |
587 | ||
588 | ro = val ? true : false; | |
589 | /* Snapshot doesn't allow to write*/ | |
590 | if (rbd_dev->spec->snap_id != CEPH_NOSNAP && !ro) | |
591 | return -EROFS; | |
592 | ||
77f33c03 JD |
593 | spin_lock_irq(&rbd_dev->lock); |
594 | /* prevent others open this device */ | |
595 | if (rbd_dev->open_count > 1) { | |
596 | ret = -EBUSY; | |
597 | goto out; | |
598 | } | |
599 | ||
131fd9f6 GZ |
600 | if (rbd_dev->mapping.read_only != ro) { |
601 | rbd_dev->mapping.read_only = ro; | |
77f33c03 | 602 | ro_changed = true; |
131fd9f6 GZ |
603 | } |
604 | ||
77f33c03 JD |
605 | out: |
606 | spin_unlock_irq(&rbd_dev->lock); | |
607 | /* set_disk_ro() may sleep, so call it after releasing rbd_dev->lock */ | |
608 | if (ret == 0 && ro_changed) | |
609 | set_disk_ro(rbd_dev->disk, ro ? 1 : 0); | |
610 | ||
611 | return ret; | |
131fd9f6 GZ |
612 | } |
613 | ||
614 | static int rbd_ioctl(struct block_device *bdev, fmode_t mode, | |
615 | unsigned int cmd, unsigned long arg) | |
616 | { | |
617 | struct rbd_device *rbd_dev = bdev->bd_disk->private_data; | |
618 | int ret = 0; | |
619 | ||
131fd9f6 GZ |
620 | switch (cmd) { |
621 | case BLKROSET: | |
622 | ret = rbd_ioctl_set_ro(rbd_dev, arg); | |
623 | break; | |
624 | default: | |
625 | ret = -ENOTTY; | |
626 | } | |
627 | ||
131fd9f6 GZ |
628 | return ret; |
629 | } | |
630 | ||
631 | #ifdef CONFIG_COMPAT | |
632 | static int rbd_compat_ioctl(struct block_device *bdev, fmode_t mode, | |
633 | unsigned int cmd, unsigned long arg) | |
634 | { | |
635 | return rbd_ioctl(bdev, mode, cmd, arg); | |
636 | } | |
637 | #endif /* CONFIG_COMPAT */ | |
638 | ||
602adf40 YS |
639 | static const struct block_device_operations rbd_bd_ops = { |
640 | .owner = THIS_MODULE, | |
641 | .open = rbd_open, | |
dfc5606d | 642 | .release = rbd_release, |
131fd9f6 GZ |
643 | .ioctl = rbd_ioctl, |
644 | #ifdef CONFIG_COMPAT | |
645 | .compat_ioctl = rbd_compat_ioctl, | |
646 | #endif | |
602adf40 YS |
647 | }; |
648 | ||
649 | /* | |
7262cfca | 650 | * Initialize an rbd client instance. Success or not, this function |
cfbf6377 | 651 | * consumes ceph_opts. Caller holds client_mutex. |
602adf40 | 652 | */ |
f8c38929 | 653 | static struct rbd_client *rbd_client_create(struct ceph_options *ceph_opts) |
602adf40 YS |
654 | { |
655 | struct rbd_client *rbdc; | |
656 | int ret = -ENOMEM; | |
657 | ||
37206ee5 | 658 | dout("%s:\n", __func__); |
602adf40 YS |
659 | rbdc = kmalloc(sizeof(struct rbd_client), GFP_KERNEL); |
660 | if (!rbdc) | |
661 | goto out_opt; | |
662 | ||
663 | kref_init(&rbdc->kref); | |
664 | INIT_LIST_HEAD(&rbdc->node); | |
665 | ||
43ae4701 | 666 | rbdc->client = ceph_create_client(ceph_opts, rbdc, 0, 0); |
602adf40 | 667 | if (IS_ERR(rbdc->client)) |
08f75463 | 668 | goto out_rbdc; |
43ae4701 | 669 | ceph_opts = NULL; /* Now rbdc->client is responsible for ceph_opts */ |
602adf40 YS |
670 | |
671 | ret = ceph_open_session(rbdc->client); | |
672 | if (ret < 0) | |
08f75463 | 673 | goto out_client; |
602adf40 | 674 | |
432b8587 | 675 | spin_lock(&rbd_client_list_lock); |
602adf40 | 676 | list_add_tail(&rbdc->node, &rbd_client_list); |
432b8587 | 677 | spin_unlock(&rbd_client_list_lock); |
602adf40 | 678 | |
37206ee5 | 679 | dout("%s: rbdc %p\n", __func__, rbdc); |
bc534d86 | 680 | |
602adf40 | 681 | return rbdc; |
08f75463 | 682 | out_client: |
602adf40 | 683 | ceph_destroy_client(rbdc->client); |
08f75463 | 684 | out_rbdc: |
602adf40 YS |
685 | kfree(rbdc); |
686 | out_opt: | |
43ae4701 AE |
687 | if (ceph_opts) |
688 | ceph_destroy_options(ceph_opts); | |
37206ee5 AE |
689 | dout("%s: error %d\n", __func__, ret); |
690 | ||
28f259b7 | 691 | return ERR_PTR(ret); |
602adf40 YS |
692 | } |
693 | ||
2f82ee54 AE |
694 | static struct rbd_client *__rbd_get_client(struct rbd_client *rbdc) |
695 | { | |
696 | kref_get(&rbdc->kref); | |
697 | ||
698 | return rbdc; | |
699 | } | |
700 | ||
602adf40 | 701 | /* |
1f7ba331 AE |
702 | * Find a ceph client with specific addr and configuration. If |
703 | * found, bump its reference count. | |
602adf40 | 704 | */ |
1f7ba331 | 705 | static struct rbd_client *rbd_client_find(struct ceph_options *ceph_opts) |
602adf40 YS |
706 | { |
707 | struct rbd_client *client_node; | |
1f7ba331 | 708 | bool found = false; |
602adf40 | 709 | |
43ae4701 | 710 | if (ceph_opts->flags & CEPH_OPT_NOSHARE) |
602adf40 YS |
711 | return NULL; |
712 | ||
1f7ba331 AE |
713 | spin_lock(&rbd_client_list_lock); |
714 | list_for_each_entry(client_node, &rbd_client_list, node) { | |
715 | if (!ceph_compare_options(ceph_opts, client_node->client)) { | |
2f82ee54 AE |
716 | __rbd_get_client(client_node); |
717 | ||
1f7ba331 AE |
718 | found = true; |
719 | break; | |
720 | } | |
721 | } | |
722 | spin_unlock(&rbd_client_list_lock); | |
723 | ||
724 | return found ? client_node : NULL; | |
602adf40 YS |
725 | } |
726 | ||
59c2be1e | 727 | /* |
210c104c | 728 | * (Per device) rbd map options |
59c2be1e YS |
729 | */ |
730 | enum { | |
b5584180 | 731 | Opt_queue_depth, |
59c2be1e YS |
732 | Opt_last_int, |
733 | /* int args above */ | |
734 | Opt_last_string, | |
735 | /* string args above */ | |
cc0538b6 AE |
736 | Opt_read_only, |
737 | Opt_read_write, | |
210c104c | 738 | Opt_err |
59c2be1e YS |
739 | }; |
740 | ||
43ae4701 | 741 | static match_table_t rbd_opts_tokens = { |
b5584180 | 742 | {Opt_queue_depth, "queue_depth=%d"}, |
59c2be1e YS |
743 | /* int args above */ |
744 | /* string args above */ | |
be466c1c | 745 | {Opt_read_only, "read_only"}, |
cc0538b6 AE |
746 | {Opt_read_only, "ro"}, /* Alternate spelling */ |
747 | {Opt_read_write, "read_write"}, | |
748 | {Opt_read_write, "rw"}, /* Alternate spelling */ | |
210c104c | 749 | {Opt_err, NULL} |
59c2be1e YS |
750 | }; |
751 | ||
98571b5a | 752 | struct rbd_options { |
b5584180 | 753 | int queue_depth; |
98571b5a AE |
754 | bool read_only; |
755 | }; | |
756 | ||
b5584180 | 757 | #define RBD_QUEUE_DEPTH_DEFAULT BLKDEV_MAX_RQ |
98571b5a AE |
758 | #define RBD_READ_ONLY_DEFAULT false |
759 | ||
59c2be1e YS |
760 | static int parse_rbd_opts_token(char *c, void *private) |
761 | { | |
43ae4701 | 762 | struct rbd_options *rbd_opts = private; |
59c2be1e YS |
763 | substring_t argstr[MAX_OPT_ARGS]; |
764 | int token, intval, ret; | |
765 | ||
43ae4701 | 766 | token = match_token(c, rbd_opts_tokens, argstr); |
59c2be1e YS |
767 | if (token < Opt_last_int) { |
768 | ret = match_int(&argstr[0], &intval); | |
769 | if (ret < 0) { | |
210c104c | 770 | pr_err("bad mount option arg (not int) at '%s'\n", c); |
59c2be1e YS |
771 | return ret; |
772 | } | |
773 | dout("got int token %d val %d\n", token, intval); | |
774 | } else if (token > Opt_last_int && token < Opt_last_string) { | |
210c104c | 775 | dout("got string token %d val %s\n", token, argstr[0].from); |
59c2be1e YS |
776 | } else { |
777 | dout("got token %d\n", token); | |
778 | } | |
779 | ||
780 | switch (token) { | |
b5584180 ID |
781 | case Opt_queue_depth: |
782 | if (intval < 1) { | |
783 | pr_err("queue_depth out of range\n"); | |
784 | return -EINVAL; | |
785 | } | |
786 | rbd_opts->queue_depth = intval; | |
787 | break; | |
cc0538b6 AE |
788 | case Opt_read_only: |
789 | rbd_opts->read_only = true; | |
790 | break; | |
791 | case Opt_read_write: | |
792 | rbd_opts->read_only = false; | |
793 | break; | |
59c2be1e | 794 | default: |
210c104c ID |
795 | /* libceph prints "bad option" msg */ |
796 | return -EINVAL; | |
59c2be1e | 797 | } |
210c104c | 798 | |
59c2be1e YS |
799 | return 0; |
800 | } | |
801 | ||
6d2940c8 GZ |
802 | static char* obj_op_name(enum obj_operation_type op_type) |
803 | { | |
804 | switch (op_type) { | |
805 | case OBJ_OP_READ: | |
806 | return "read"; | |
807 | case OBJ_OP_WRITE: | |
808 | return "write"; | |
90e98c52 GZ |
809 | case OBJ_OP_DISCARD: |
810 | return "discard"; | |
6d2940c8 GZ |
811 | default: |
812 | return "???"; | |
813 | } | |
814 | } | |
815 | ||
602adf40 YS |
816 | /* |
817 | * Get a ceph client with specific addr and configuration, if one does | |
7262cfca AE |
818 | * not exist create it. Either way, ceph_opts is consumed by this |
819 | * function. | |
602adf40 | 820 | */ |
9d3997fd | 821 | static struct rbd_client *rbd_get_client(struct ceph_options *ceph_opts) |
602adf40 | 822 | { |
f8c38929 | 823 | struct rbd_client *rbdc; |
59c2be1e | 824 | |
cfbf6377 | 825 | mutex_lock_nested(&client_mutex, SINGLE_DEPTH_NESTING); |
1f7ba331 | 826 | rbdc = rbd_client_find(ceph_opts); |
9d3997fd | 827 | if (rbdc) /* using an existing client */ |
43ae4701 | 828 | ceph_destroy_options(ceph_opts); |
9d3997fd | 829 | else |
f8c38929 | 830 | rbdc = rbd_client_create(ceph_opts); |
cfbf6377 | 831 | mutex_unlock(&client_mutex); |
602adf40 | 832 | |
9d3997fd | 833 | return rbdc; |
602adf40 YS |
834 | } |
835 | ||
836 | /* | |
837 | * Destroy ceph client | |
d23a4b3f | 838 | * |
432b8587 | 839 | * Caller must hold rbd_client_list_lock. |
602adf40 YS |
840 | */ |
841 | static void rbd_client_release(struct kref *kref) | |
842 | { | |
843 | struct rbd_client *rbdc = container_of(kref, struct rbd_client, kref); | |
844 | ||
37206ee5 | 845 | dout("%s: rbdc %p\n", __func__, rbdc); |
cd9d9f5d | 846 | spin_lock(&rbd_client_list_lock); |
602adf40 | 847 | list_del(&rbdc->node); |
cd9d9f5d | 848 | spin_unlock(&rbd_client_list_lock); |
602adf40 YS |
849 | |
850 | ceph_destroy_client(rbdc->client); | |
851 | kfree(rbdc); | |
852 | } | |
853 | ||
854 | /* | |
855 | * Drop reference to ceph client node. If it's not referenced anymore, release | |
856 | * it. | |
857 | */ | |
9d3997fd | 858 | static void rbd_put_client(struct rbd_client *rbdc) |
602adf40 | 859 | { |
c53d5893 AE |
860 | if (rbdc) |
861 | kref_put(&rbdc->kref, rbd_client_release); | |
602adf40 YS |
862 | } |
863 | ||
a30b71b9 AE |
864 | static bool rbd_image_format_valid(u32 image_format) |
865 | { | |
866 | return image_format == 1 || image_format == 2; | |
867 | } | |
868 | ||
8e94af8e AE |
869 | static bool rbd_dev_ondisk_valid(struct rbd_image_header_ondisk *ondisk) |
870 | { | |
103a150f AE |
871 | size_t size; |
872 | u32 snap_count; | |
873 | ||
874 | /* The header has to start with the magic rbd header text */ | |
875 | if (memcmp(&ondisk->text, RBD_HEADER_TEXT, sizeof (RBD_HEADER_TEXT))) | |
876 | return false; | |
877 | ||
db2388b6 AE |
878 | /* The bio layer requires at least sector-sized I/O */ |
879 | ||
880 | if (ondisk->options.order < SECTOR_SHIFT) | |
881 | return false; | |
882 | ||
883 | /* If we use u64 in a few spots we may be able to loosen this */ | |
884 | ||
885 | if (ondisk->options.order > 8 * sizeof (int) - 1) | |
886 | return false; | |
887 | ||
103a150f AE |
888 | /* |
889 | * The size of a snapshot header has to fit in a size_t, and | |
890 | * that limits the number of snapshots. | |
891 | */ | |
892 | snap_count = le32_to_cpu(ondisk->snap_count); | |
893 | size = SIZE_MAX - sizeof (struct ceph_snap_context); | |
894 | if (snap_count > size / sizeof (__le64)) | |
895 | return false; | |
896 | ||
897 | /* | |
898 | * Not only that, but the size of the entire the snapshot | |
899 | * header must also be representable in a size_t. | |
900 | */ | |
901 | size -= snap_count * sizeof (__le64); | |
902 | if ((u64) size < le64_to_cpu(ondisk->snap_names_len)) | |
903 | return false; | |
904 | ||
905 | return true; | |
8e94af8e AE |
906 | } |
907 | ||
602adf40 | 908 | /* |
bb23e37a AE |
909 | * Fill an rbd image header with information from the given format 1 |
910 | * on-disk header. | |
602adf40 | 911 | */ |
662518b1 | 912 | static int rbd_header_from_disk(struct rbd_device *rbd_dev, |
4156d998 | 913 | struct rbd_image_header_ondisk *ondisk) |
602adf40 | 914 | { |
662518b1 | 915 | struct rbd_image_header *header = &rbd_dev->header; |
bb23e37a AE |
916 | bool first_time = header->object_prefix == NULL; |
917 | struct ceph_snap_context *snapc; | |
918 | char *object_prefix = NULL; | |
919 | char *snap_names = NULL; | |
920 | u64 *snap_sizes = NULL; | |
ccece235 | 921 | u32 snap_count; |
d2bb24e5 | 922 | size_t size; |
bb23e37a | 923 | int ret = -ENOMEM; |
621901d6 | 924 | u32 i; |
602adf40 | 925 | |
bb23e37a | 926 | /* Allocate this now to avoid having to handle failure below */ |
6a52325f | 927 | |
bb23e37a AE |
928 | if (first_time) { |
929 | size_t len; | |
103a150f | 930 | |
bb23e37a AE |
931 | len = strnlen(ondisk->object_prefix, |
932 | sizeof (ondisk->object_prefix)); | |
933 | object_prefix = kmalloc(len + 1, GFP_KERNEL); | |
934 | if (!object_prefix) | |
935 | return -ENOMEM; | |
936 | memcpy(object_prefix, ondisk->object_prefix, len); | |
937 | object_prefix[len] = '\0'; | |
938 | } | |
00f1f36f | 939 | |
bb23e37a | 940 | /* Allocate the snapshot context and fill it in */ |
00f1f36f | 941 | |
bb23e37a AE |
942 | snap_count = le32_to_cpu(ondisk->snap_count); |
943 | snapc = ceph_create_snap_context(snap_count, GFP_KERNEL); | |
944 | if (!snapc) | |
945 | goto out_err; | |
946 | snapc->seq = le64_to_cpu(ondisk->snap_seq); | |
602adf40 | 947 | if (snap_count) { |
bb23e37a | 948 | struct rbd_image_snap_ondisk *snaps; |
f785cc1d AE |
949 | u64 snap_names_len = le64_to_cpu(ondisk->snap_names_len); |
950 | ||
bb23e37a | 951 | /* We'll keep a copy of the snapshot names... */ |
621901d6 | 952 | |
bb23e37a AE |
953 | if (snap_names_len > (u64)SIZE_MAX) |
954 | goto out_2big; | |
955 | snap_names = kmalloc(snap_names_len, GFP_KERNEL); | |
956 | if (!snap_names) | |
6a52325f AE |
957 | goto out_err; |
958 | ||
bb23e37a | 959 | /* ...as well as the array of their sizes. */ |
621901d6 | 960 | |
d2bb24e5 | 961 | size = snap_count * sizeof (*header->snap_sizes); |
bb23e37a AE |
962 | snap_sizes = kmalloc(size, GFP_KERNEL); |
963 | if (!snap_sizes) | |
6a52325f | 964 | goto out_err; |
bb23e37a | 965 | |
f785cc1d | 966 | /* |
bb23e37a AE |
967 | * Copy the names, and fill in each snapshot's id |
968 | * and size. | |
969 | * | |
99a41ebc | 970 | * Note that rbd_dev_v1_header_info() guarantees the |
bb23e37a | 971 | * ondisk buffer we're working with has |
f785cc1d AE |
972 | * snap_names_len bytes beyond the end of the |
973 | * snapshot id array, this memcpy() is safe. | |
974 | */ | |
bb23e37a AE |
975 | memcpy(snap_names, &ondisk->snaps[snap_count], snap_names_len); |
976 | snaps = ondisk->snaps; | |
977 | for (i = 0; i < snap_count; i++) { | |
978 | snapc->snaps[i] = le64_to_cpu(snaps[i].id); | |
979 | snap_sizes[i] = le64_to_cpu(snaps[i].image_size); | |
980 | } | |
602adf40 | 981 | } |
6a52325f | 982 | |
bb23e37a | 983 | /* We won't fail any more, fill in the header */ |
621901d6 | 984 | |
bb23e37a AE |
985 | if (first_time) { |
986 | header->object_prefix = object_prefix; | |
987 | header->obj_order = ondisk->options.order; | |
988 | header->crypt_type = ondisk->options.crypt_type; | |
989 | header->comp_type = ondisk->options.comp_type; | |
990 | /* The rest aren't used for format 1 images */ | |
991 | header->stripe_unit = 0; | |
992 | header->stripe_count = 0; | |
993 | header->features = 0; | |
602adf40 | 994 | } else { |
662518b1 AE |
995 | ceph_put_snap_context(header->snapc); |
996 | kfree(header->snap_names); | |
997 | kfree(header->snap_sizes); | |
602adf40 | 998 | } |
849b4260 | 999 | |
bb23e37a | 1000 | /* The remaining fields always get updated (when we refresh) */ |
621901d6 | 1001 | |
f84344f3 | 1002 | header->image_size = le64_to_cpu(ondisk->image_size); |
bb23e37a AE |
1003 | header->snapc = snapc; |
1004 | header->snap_names = snap_names; | |
1005 | header->snap_sizes = snap_sizes; | |
468521c1 | 1006 | |
602adf40 | 1007 | return 0; |
bb23e37a AE |
1008 | out_2big: |
1009 | ret = -EIO; | |
6a52325f | 1010 | out_err: |
bb23e37a AE |
1011 | kfree(snap_sizes); |
1012 | kfree(snap_names); | |
1013 | ceph_put_snap_context(snapc); | |
1014 | kfree(object_prefix); | |
ccece235 | 1015 | |
bb23e37a | 1016 | return ret; |
602adf40 YS |
1017 | } |
1018 | ||
9682fc6d AE |
1019 | static const char *_rbd_dev_v1_snap_name(struct rbd_device *rbd_dev, u32 which) |
1020 | { | |
1021 | const char *snap_name; | |
1022 | ||
1023 | rbd_assert(which < rbd_dev->header.snapc->num_snaps); | |
1024 | ||
1025 | /* Skip over names until we find the one we are looking for */ | |
1026 | ||
1027 | snap_name = rbd_dev->header.snap_names; | |
1028 | while (which--) | |
1029 | snap_name += strlen(snap_name) + 1; | |
1030 | ||
1031 | return kstrdup(snap_name, GFP_KERNEL); | |
1032 | } | |
1033 | ||
30d1cff8 AE |
1034 | /* |
1035 | * Snapshot id comparison function for use with qsort()/bsearch(). | |
1036 | * Note that result is for snapshots in *descending* order. | |
1037 | */ | |
1038 | static int snapid_compare_reverse(const void *s1, const void *s2) | |
1039 | { | |
1040 | u64 snap_id1 = *(u64 *)s1; | |
1041 | u64 snap_id2 = *(u64 *)s2; | |
1042 | ||
1043 | if (snap_id1 < snap_id2) | |
1044 | return 1; | |
1045 | return snap_id1 == snap_id2 ? 0 : -1; | |
1046 | } | |
1047 | ||
1048 | /* | |
1049 | * Search a snapshot context to see if the given snapshot id is | |
1050 | * present. | |
1051 | * | |
1052 | * Returns the position of the snapshot id in the array if it's found, | |
1053 | * or BAD_SNAP_INDEX otherwise. | |
1054 | * | |
1055 | * Note: The snapshot array is in kept sorted (by the osd) in | |
1056 | * reverse order, highest snapshot id first. | |
1057 | */ | |
9682fc6d AE |
1058 | static u32 rbd_dev_snap_index(struct rbd_device *rbd_dev, u64 snap_id) |
1059 | { | |
1060 | struct ceph_snap_context *snapc = rbd_dev->header.snapc; | |
30d1cff8 | 1061 | u64 *found; |
9682fc6d | 1062 | |
30d1cff8 AE |
1063 | found = bsearch(&snap_id, &snapc->snaps, snapc->num_snaps, |
1064 | sizeof (snap_id), snapid_compare_reverse); | |
9682fc6d | 1065 | |
30d1cff8 | 1066 | return found ? (u32)(found - &snapc->snaps[0]) : BAD_SNAP_INDEX; |
9682fc6d AE |
1067 | } |
1068 | ||
2ad3d716 AE |
1069 | static const char *rbd_dev_v1_snap_name(struct rbd_device *rbd_dev, |
1070 | u64 snap_id) | |
9e15b77d | 1071 | { |
54cac61f | 1072 | u32 which; |
da6a6b63 | 1073 | const char *snap_name; |
9e15b77d | 1074 | |
54cac61f AE |
1075 | which = rbd_dev_snap_index(rbd_dev, snap_id); |
1076 | if (which == BAD_SNAP_INDEX) | |
da6a6b63 | 1077 | return ERR_PTR(-ENOENT); |
54cac61f | 1078 | |
da6a6b63 JD |
1079 | snap_name = _rbd_dev_v1_snap_name(rbd_dev, which); |
1080 | return snap_name ? snap_name : ERR_PTR(-ENOMEM); | |
54cac61f AE |
1081 | } |
1082 | ||
1083 | static const char *rbd_snap_name(struct rbd_device *rbd_dev, u64 snap_id) | |
1084 | { | |
9e15b77d AE |
1085 | if (snap_id == CEPH_NOSNAP) |
1086 | return RBD_SNAP_HEAD_NAME; | |
1087 | ||
54cac61f AE |
1088 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); |
1089 | if (rbd_dev->image_format == 1) | |
1090 | return rbd_dev_v1_snap_name(rbd_dev, snap_id); | |
9e15b77d | 1091 | |
54cac61f | 1092 | return rbd_dev_v2_snap_name(rbd_dev, snap_id); |
9e15b77d AE |
1093 | } |
1094 | ||
2ad3d716 AE |
1095 | static int rbd_snap_size(struct rbd_device *rbd_dev, u64 snap_id, |
1096 | u64 *snap_size) | |
602adf40 | 1097 | { |
2ad3d716 AE |
1098 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); |
1099 | if (snap_id == CEPH_NOSNAP) { | |
1100 | *snap_size = rbd_dev->header.image_size; | |
1101 | } else if (rbd_dev->image_format == 1) { | |
1102 | u32 which; | |
602adf40 | 1103 | |
2ad3d716 AE |
1104 | which = rbd_dev_snap_index(rbd_dev, snap_id); |
1105 | if (which == BAD_SNAP_INDEX) | |
1106 | return -ENOENT; | |
e86924a8 | 1107 | |
2ad3d716 AE |
1108 | *snap_size = rbd_dev->header.snap_sizes[which]; |
1109 | } else { | |
1110 | u64 size = 0; | |
1111 | int ret; | |
1112 | ||
1113 | ret = _rbd_dev_v2_snap_size(rbd_dev, snap_id, NULL, &size); | |
1114 | if (ret) | |
1115 | return ret; | |
1116 | ||
1117 | *snap_size = size; | |
1118 | } | |
1119 | return 0; | |
602adf40 YS |
1120 | } |
1121 | ||
2ad3d716 AE |
1122 | static int rbd_snap_features(struct rbd_device *rbd_dev, u64 snap_id, |
1123 | u64 *snap_features) | |
602adf40 | 1124 | { |
2ad3d716 AE |
1125 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); |
1126 | if (snap_id == CEPH_NOSNAP) { | |
1127 | *snap_features = rbd_dev->header.features; | |
1128 | } else if (rbd_dev->image_format == 1) { | |
1129 | *snap_features = 0; /* No features for format 1 */ | |
602adf40 | 1130 | } else { |
2ad3d716 AE |
1131 | u64 features = 0; |
1132 | int ret; | |
8b0241f8 | 1133 | |
2ad3d716 AE |
1134 | ret = _rbd_dev_v2_snap_features(rbd_dev, snap_id, &features); |
1135 | if (ret) | |
1136 | return ret; | |
1137 | ||
1138 | *snap_features = features; | |
1139 | } | |
1140 | return 0; | |
1141 | } | |
1142 | ||
1143 | static int rbd_dev_mapping_set(struct rbd_device *rbd_dev) | |
1144 | { | |
8f4b7d98 | 1145 | u64 snap_id = rbd_dev->spec->snap_id; |
2ad3d716 AE |
1146 | u64 size = 0; |
1147 | u64 features = 0; | |
1148 | int ret; | |
1149 | ||
2ad3d716 AE |
1150 | ret = rbd_snap_size(rbd_dev, snap_id, &size); |
1151 | if (ret) | |
1152 | return ret; | |
1153 | ret = rbd_snap_features(rbd_dev, snap_id, &features); | |
1154 | if (ret) | |
1155 | return ret; | |
1156 | ||
1157 | rbd_dev->mapping.size = size; | |
1158 | rbd_dev->mapping.features = features; | |
1159 | ||
8b0241f8 | 1160 | return 0; |
602adf40 YS |
1161 | } |
1162 | ||
d1cf5788 AE |
1163 | static void rbd_dev_mapping_clear(struct rbd_device *rbd_dev) |
1164 | { | |
1165 | rbd_dev->mapping.size = 0; | |
1166 | rbd_dev->mapping.features = 0; | |
200a6a8b AE |
1167 | } |
1168 | ||
7d5079aa HS |
1169 | static void rbd_segment_name_free(const char *name) |
1170 | { | |
1171 | /* The explicit cast here is needed to drop the const qualifier */ | |
1172 | ||
1173 | kmem_cache_free(rbd_segment_name_cache, (void *)name); | |
1174 | } | |
1175 | ||
98571b5a | 1176 | static const char *rbd_segment_name(struct rbd_device *rbd_dev, u64 offset) |
602adf40 | 1177 | { |
65ccfe21 AE |
1178 | char *name; |
1179 | u64 segment; | |
1180 | int ret; | |
3a96d5cd | 1181 | char *name_format; |
602adf40 | 1182 | |
78c2a44a | 1183 | name = kmem_cache_alloc(rbd_segment_name_cache, GFP_NOIO); |
65ccfe21 AE |
1184 | if (!name) |
1185 | return NULL; | |
1186 | segment = offset >> rbd_dev->header.obj_order; | |
3a96d5cd JD |
1187 | name_format = "%s.%012llx"; |
1188 | if (rbd_dev->image_format == 2) | |
1189 | name_format = "%s.%016llx"; | |
2d0ebc5d | 1190 | ret = snprintf(name, CEPH_MAX_OID_NAME_LEN + 1, name_format, |
65ccfe21 | 1191 | rbd_dev->header.object_prefix, segment); |
2d0ebc5d | 1192 | if (ret < 0 || ret > CEPH_MAX_OID_NAME_LEN) { |
65ccfe21 AE |
1193 | pr_err("error formatting segment name for #%llu (%d)\n", |
1194 | segment, ret); | |
7d5079aa | 1195 | rbd_segment_name_free(name); |
65ccfe21 AE |
1196 | name = NULL; |
1197 | } | |
602adf40 | 1198 | |
65ccfe21 AE |
1199 | return name; |
1200 | } | |
602adf40 | 1201 | |
65ccfe21 AE |
1202 | static u64 rbd_segment_offset(struct rbd_device *rbd_dev, u64 offset) |
1203 | { | |
1204 | u64 segment_size = (u64) 1 << rbd_dev->header.obj_order; | |
602adf40 | 1205 | |
65ccfe21 AE |
1206 | return offset & (segment_size - 1); |
1207 | } | |
1208 | ||
1209 | static u64 rbd_segment_length(struct rbd_device *rbd_dev, | |
1210 | u64 offset, u64 length) | |
1211 | { | |
1212 | u64 segment_size = (u64) 1 << rbd_dev->header.obj_order; | |
1213 | ||
1214 | offset &= segment_size - 1; | |
1215 | ||
aafb230e | 1216 | rbd_assert(length <= U64_MAX - offset); |
65ccfe21 AE |
1217 | if (offset + length > segment_size) |
1218 | length = segment_size - offset; | |
1219 | ||
1220 | return length; | |
602adf40 YS |
1221 | } |
1222 | ||
029bcbd8 JD |
1223 | /* |
1224 | * returns the size of an object in the image | |
1225 | */ | |
1226 | static u64 rbd_obj_bytes(struct rbd_image_header *header) | |
1227 | { | |
1228 | return 1 << header->obj_order; | |
1229 | } | |
1230 | ||
602adf40 YS |
1231 | /* |
1232 | * bio helpers | |
1233 | */ | |
1234 | ||
1235 | static void bio_chain_put(struct bio *chain) | |
1236 | { | |
1237 | struct bio *tmp; | |
1238 | ||
1239 | while (chain) { | |
1240 | tmp = chain; | |
1241 | chain = chain->bi_next; | |
1242 | bio_put(tmp); | |
1243 | } | |
1244 | } | |
1245 | ||
1246 | /* | |
1247 | * zeros a bio chain, starting at specific offset | |
1248 | */ | |
1249 | static void zero_bio_chain(struct bio *chain, int start_ofs) | |
1250 | { | |
7988613b KO |
1251 | struct bio_vec bv; |
1252 | struct bvec_iter iter; | |
602adf40 YS |
1253 | unsigned long flags; |
1254 | void *buf; | |
602adf40 YS |
1255 | int pos = 0; |
1256 | ||
1257 | while (chain) { | |
7988613b KO |
1258 | bio_for_each_segment(bv, chain, iter) { |
1259 | if (pos + bv.bv_len > start_ofs) { | |
602adf40 | 1260 | int remainder = max(start_ofs - pos, 0); |
7988613b | 1261 | buf = bvec_kmap_irq(&bv, &flags); |
602adf40 | 1262 | memset(buf + remainder, 0, |
7988613b KO |
1263 | bv.bv_len - remainder); |
1264 | flush_dcache_page(bv.bv_page); | |
85b5aaa6 | 1265 | bvec_kunmap_irq(buf, &flags); |
602adf40 | 1266 | } |
7988613b | 1267 | pos += bv.bv_len; |
602adf40 YS |
1268 | } |
1269 | ||
1270 | chain = chain->bi_next; | |
1271 | } | |
1272 | } | |
1273 | ||
b9434c5b AE |
1274 | /* |
1275 | * similar to zero_bio_chain(), zeros data defined by a page array, | |
1276 | * starting at the given byte offset from the start of the array and | |
1277 | * continuing up to the given end offset. The pages array is | |
1278 | * assumed to be big enough to hold all bytes up to the end. | |
1279 | */ | |
1280 | static void zero_pages(struct page **pages, u64 offset, u64 end) | |
1281 | { | |
1282 | struct page **page = &pages[offset >> PAGE_SHIFT]; | |
1283 | ||
1284 | rbd_assert(end > offset); | |
1285 | rbd_assert(end - offset <= (u64)SIZE_MAX); | |
1286 | while (offset < end) { | |
1287 | size_t page_offset; | |
1288 | size_t length; | |
1289 | unsigned long flags; | |
1290 | void *kaddr; | |
1291 | ||
491205a8 GU |
1292 | page_offset = offset & ~PAGE_MASK; |
1293 | length = min_t(size_t, PAGE_SIZE - page_offset, end - offset); | |
b9434c5b AE |
1294 | local_irq_save(flags); |
1295 | kaddr = kmap_atomic(*page); | |
1296 | memset(kaddr + page_offset, 0, length); | |
e2156054 | 1297 | flush_dcache_page(*page); |
b9434c5b AE |
1298 | kunmap_atomic(kaddr); |
1299 | local_irq_restore(flags); | |
1300 | ||
1301 | offset += length; | |
1302 | page++; | |
1303 | } | |
1304 | } | |
1305 | ||
602adf40 | 1306 | /* |
f7760dad AE |
1307 | * Clone a portion of a bio, starting at the given byte offset |
1308 | * and continuing for the number of bytes indicated. | |
602adf40 | 1309 | */ |
f7760dad AE |
1310 | static struct bio *bio_clone_range(struct bio *bio_src, |
1311 | unsigned int offset, | |
1312 | unsigned int len, | |
1313 | gfp_t gfpmask) | |
602adf40 | 1314 | { |
f7760dad AE |
1315 | struct bio *bio; |
1316 | ||
5341a627 | 1317 | bio = bio_clone(bio_src, gfpmask); |
f7760dad AE |
1318 | if (!bio) |
1319 | return NULL; /* ENOMEM */ | |
602adf40 | 1320 | |
5341a627 | 1321 | bio_advance(bio, offset); |
4f024f37 | 1322 | bio->bi_iter.bi_size = len; |
f7760dad AE |
1323 | |
1324 | return bio; | |
1325 | } | |
1326 | ||
1327 | /* | |
1328 | * Clone a portion of a bio chain, starting at the given byte offset | |
1329 | * into the first bio in the source chain and continuing for the | |
1330 | * number of bytes indicated. The result is another bio chain of | |
1331 | * exactly the given length, or a null pointer on error. | |
1332 | * | |
1333 | * The bio_src and offset parameters are both in-out. On entry they | |
1334 | * refer to the first source bio and the offset into that bio where | |
1335 | * the start of data to be cloned is located. | |
1336 | * | |
1337 | * On return, bio_src is updated to refer to the bio in the source | |
1338 | * chain that contains first un-cloned byte, and *offset will | |
1339 | * contain the offset of that byte within that bio. | |
1340 | */ | |
1341 | static struct bio *bio_chain_clone_range(struct bio **bio_src, | |
1342 | unsigned int *offset, | |
1343 | unsigned int len, | |
1344 | gfp_t gfpmask) | |
1345 | { | |
1346 | struct bio *bi = *bio_src; | |
1347 | unsigned int off = *offset; | |
1348 | struct bio *chain = NULL; | |
1349 | struct bio **end; | |
1350 | ||
1351 | /* Build up a chain of clone bios up to the limit */ | |
1352 | ||
4f024f37 | 1353 | if (!bi || off >= bi->bi_iter.bi_size || !len) |
f7760dad | 1354 | return NULL; /* Nothing to clone */ |
602adf40 | 1355 | |
f7760dad AE |
1356 | end = &chain; |
1357 | while (len) { | |
1358 | unsigned int bi_size; | |
1359 | struct bio *bio; | |
1360 | ||
f5400b7a AE |
1361 | if (!bi) { |
1362 | rbd_warn(NULL, "bio_chain exhausted with %u left", len); | |
f7760dad | 1363 | goto out_err; /* EINVAL; ran out of bio's */ |
f5400b7a | 1364 | } |
4f024f37 | 1365 | bi_size = min_t(unsigned int, bi->bi_iter.bi_size - off, len); |
f7760dad AE |
1366 | bio = bio_clone_range(bi, off, bi_size, gfpmask); |
1367 | if (!bio) | |
1368 | goto out_err; /* ENOMEM */ | |
1369 | ||
1370 | *end = bio; | |
1371 | end = &bio->bi_next; | |
602adf40 | 1372 | |
f7760dad | 1373 | off += bi_size; |
4f024f37 | 1374 | if (off == bi->bi_iter.bi_size) { |
f7760dad AE |
1375 | bi = bi->bi_next; |
1376 | off = 0; | |
1377 | } | |
1378 | len -= bi_size; | |
1379 | } | |
1380 | *bio_src = bi; | |
1381 | *offset = off; | |
1382 | ||
1383 | return chain; | |
1384 | out_err: | |
1385 | bio_chain_put(chain); | |
602adf40 | 1386 | |
602adf40 YS |
1387 | return NULL; |
1388 | } | |
1389 | ||
926f9b3f AE |
1390 | /* |
1391 | * The default/initial value for all object request flags is 0. For | |
1392 | * each flag, once its value is set to 1 it is never reset to 0 | |
1393 | * again. | |
1394 | */ | |
57acbaa7 | 1395 | static void obj_request_img_data_set(struct rbd_obj_request *obj_request) |
926f9b3f | 1396 | { |
57acbaa7 | 1397 | if (test_and_set_bit(OBJ_REQ_IMG_DATA, &obj_request->flags)) { |
926f9b3f AE |
1398 | struct rbd_device *rbd_dev; |
1399 | ||
57acbaa7 | 1400 | rbd_dev = obj_request->img_request->rbd_dev; |
9584d508 | 1401 | rbd_warn(rbd_dev, "obj_request %p already marked img_data", |
926f9b3f AE |
1402 | obj_request); |
1403 | } | |
1404 | } | |
1405 | ||
57acbaa7 | 1406 | static bool obj_request_img_data_test(struct rbd_obj_request *obj_request) |
926f9b3f AE |
1407 | { |
1408 | smp_mb(); | |
57acbaa7 | 1409 | return test_bit(OBJ_REQ_IMG_DATA, &obj_request->flags) != 0; |
926f9b3f AE |
1410 | } |
1411 | ||
57acbaa7 | 1412 | static void obj_request_done_set(struct rbd_obj_request *obj_request) |
6365d33a | 1413 | { |
57acbaa7 AE |
1414 | if (test_and_set_bit(OBJ_REQ_DONE, &obj_request->flags)) { |
1415 | struct rbd_device *rbd_dev = NULL; | |
6365d33a | 1416 | |
57acbaa7 AE |
1417 | if (obj_request_img_data_test(obj_request)) |
1418 | rbd_dev = obj_request->img_request->rbd_dev; | |
9584d508 | 1419 | rbd_warn(rbd_dev, "obj_request %p already marked done", |
6365d33a AE |
1420 | obj_request); |
1421 | } | |
1422 | } | |
1423 | ||
57acbaa7 | 1424 | static bool obj_request_done_test(struct rbd_obj_request *obj_request) |
6365d33a AE |
1425 | { |
1426 | smp_mb(); | |
57acbaa7 | 1427 | return test_bit(OBJ_REQ_DONE, &obj_request->flags) != 0; |
6365d33a AE |
1428 | } |
1429 | ||
5679c59f AE |
1430 | /* |
1431 | * This sets the KNOWN flag after (possibly) setting the EXISTS | |
1432 | * flag. The latter is set based on the "exists" value provided. | |
1433 | * | |
1434 | * Note that for our purposes once an object exists it never goes | |
1435 | * away again. It's possible that the response from two existence | |
1436 | * checks are separated by the creation of the target object, and | |
1437 | * the first ("doesn't exist") response arrives *after* the second | |
1438 | * ("does exist"). In that case we ignore the second one. | |
1439 | */ | |
1440 | static void obj_request_existence_set(struct rbd_obj_request *obj_request, | |
1441 | bool exists) | |
1442 | { | |
1443 | if (exists) | |
1444 | set_bit(OBJ_REQ_EXISTS, &obj_request->flags); | |
1445 | set_bit(OBJ_REQ_KNOWN, &obj_request->flags); | |
1446 | smp_mb(); | |
1447 | } | |
1448 | ||
1449 | static bool obj_request_known_test(struct rbd_obj_request *obj_request) | |
1450 | { | |
1451 | smp_mb(); | |
1452 | return test_bit(OBJ_REQ_KNOWN, &obj_request->flags) != 0; | |
1453 | } | |
1454 | ||
1455 | static bool obj_request_exists_test(struct rbd_obj_request *obj_request) | |
1456 | { | |
1457 | smp_mb(); | |
1458 | return test_bit(OBJ_REQ_EXISTS, &obj_request->flags) != 0; | |
1459 | } | |
1460 | ||
9638556a ID |
1461 | static bool obj_request_overlaps_parent(struct rbd_obj_request *obj_request) |
1462 | { | |
1463 | struct rbd_device *rbd_dev = obj_request->img_request->rbd_dev; | |
1464 | ||
1465 | return obj_request->img_offset < | |
1466 | round_up(rbd_dev->parent_overlap, rbd_obj_bytes(&rbd_dev->header)); | |
1467 | } | |
1468 | ||
bf0d5f50 AE |
1469 | static void rbd_obj_request_get(struct rbd_obj_request *obj_request) |
1470 | { | |
37206ee5 AE |
1471 | dout("%s: obj %p (was %d)\n", __func__, obj_request, |
1472 | atomic_read(&obj_request->kref.refcount)); | |
bf0d5f50 AE |
1473 | kref_get(&obj_request->kref); |
1474 | } | |
1475 | ||
1476 | static void rbd_obj_request_destroy(struct kref *kref); | |
1477 | static void rbd_obj_request_put(struct rbd_obj_request *obj_request) | |
1478 | { | |
1479 | rbd_assert(obj_request != NULL); | |
37206ee5 AE |
1480 | dout("%s: obj %p (was %d)\n", __func__, obj_request, |
1481 | atomic_read(&obj_request->kref.refcount)); | |
bf0d5f50 AE |
1482 | kref_put(&obj_request->kref, rbd_obj_request_destroy); |
1483 | } | |
1484 | ||
0f2d5be7 AE |
1485 | static void rbd_img_request_get(struct rbd_img_request *img_request) |
1486 | { | |
1487 | dout("%s: img %p (was %d)\n", __func__, img_request, | |
1488 | atomic_read(&img_request->kref.refcount)); | |
1489 | kref_get(&img_request->kref); | |
1490 | } | |
1491 | ||
e93f3152 AE |
1492 | static bool img_request_child_test(struct rbd_img_request *img_request); |
1493 | static void rbd_parent_request_destroy(struct kref *kref); | |
bf0d5f50 AE |
1494 | static void rbd_img_request_destroy(struct kref *kref); |
1495 | static void rbd_img_request_put(struct rbd_img_request *img_request) | |
1496 | { | |
1497 | rbd_assert(img_request != NULL); | |
37206ee5 AE |
1498 | dout("%s: img %p (was %d)\n", __func__, img_request, |
1499 | atomic_read(&img_request->kref.refcount)); | |
e93f3152 AE |
1500 | if (img_request_child_test(img_request)) |
1501 | kref_put(&img_request->kref, rbd_parent_request_destroy); | |
1502 | else | |
1503 | kref_put(&img_request->kref, rbd_img_request_destroy); | |
bf0d5f50 AE |
1504 | } |
1505 | ||
1506 | static inline void rbd_img_obj_request_add(struct rbd_img_request *img_request, | |
1507 | struct rbd_obj_request *obj_request) | |
1508 | { | |
25dcf954 AE |
1509 | rbd_assert(obj_request->img_request == NULL); |
1510 | ||
b155e86c | 1511 | /* Image request now owns object's original reference */ |
bf0d5f50 | 1512 | obj_request->img_request = img_request; |
25dcf954 | 1513 | obj_request->which = img_request->obj_request_count; |
6365d33a AE |
1514 | rbd_assert(!obj_request_img_data_test(obj_request)); |
1515 | obj_request_img_data_set(obj_request); | |
bf0d5f50 | 1516 | rbd_assert(obj_request->which != BAD_WHICH); |
25dcf954 AE |
1517 | img_request->obj_request_count++; |
1518 | list_add_tail(&obj_request->links, &img_request->obj_requests); | |
37206ee5 AE |
1519 | dout("%s: img %p obj %p w=%u\n", __func__, img_request, obj_request, |
1520 | obj_request->which); | |
bf0d5f50 AE |
1521 | } |
1522 | ||
1523 | static inline void rbd_img_obj_request_del(struct rbd_img_request *img_request, | |
1524 | struct rbd_obj_request *obj_request) | |
1525 | { | |
1526 | rbd_assert(obj_request->which != BAD_WHICH); | |
25dcf954 | 1527 | |
37206ee5 AE |
1528 | dout("%s: img %p obj %p w=%u\n", __func__, img_request, obj_request, |
1529 | obj_request->which); | |
bf0d5f50 | 1530 | list_del(&obj_request->links); |
25dcf954 AE |
1531 | rbd_assert(img_request->obj_request_count > 0); |
1532 | img_request->obj_request_count--; | |
1533 | rbd_assert(obj_request->which == img_request->obj_request_count); | |
1534 | obj_request->which = BAD_WHICH; | |
6365d33a | 1535 | rbd_assert(obj_request_img_data_test(obj_request)); |
bf0d5f50 | 1536 | rbd_assert(obj_request->img_request == img_request); |
bf0d5f50 | 1537 | obj_request->img_request = NULL; |
25dcf954 | 1538 | obj_request->callback = NULL; |
bf0d5f50 AE |
1539 | rbd_obj_request_put(obj_request); |
1540 | } | |
1541 | ||
1542 | static bool obj_request_type_valid(enum obj_request_type type) | |
1543 | { | |
1544 | switch (type) { | |
9969ebc5 | 1545 | case OBJ_REQUEST_NODATA: |
bf0d5f50 | 1546 | case OBJ_REQUEST_BIO: |
788e2df3 | 1547 | case OBJ_REQUEST_PAGES: |
bf0d5f50 AE |
1548 | return true; |
1549 | default: | |
1550 | return false; | |
1551 | } | |
1552 | } | |
1553 | ||
bf0d5f50 AE |
1554 | static int rbd_obj_request_submit(struct ceph_osd_client *osdc, |
1555 | struct rbd_obj_request *obj_request) | |
1556 | { | |
71c20a06 | 1557 | dout("%s %p\n", __func__, obj_request); |
bf0d5f50 AE |
1558 | return ceph_osdc_start_request(osdc, obj_request->osd_req, false); |
1559 | } | |
1560 | ||
71c20a06 ID |
1561 | static void rbd_obj_request_end(struct rbd_obj_request *obj_request) |
1562 | { | |
1563 | dout("%s %p\n", __func__, obj_request); | |
1564 | ceph_osdc_cancel_request(obj_request->osd_req); | |
1565 | } | |
1566 | ||
1567 | /* | |
1568 | * Wait for an object request to complete. If interrupted, cancel the | |
1569 | * underlying osd request. | |
2894e1d7 ID |
1570 | * |
1571 | * @timeout: in jiffies, 0 means "wait forever" | |
71c20a06 | 1572 | */ |
2894e1d7 ID |
1573 | static int __rbd_obj_request_wait(struct rbd_obj_request *obj_request, |
1574 | unsigned long timeout) | |
71c20a06 | 1575 | { |
2894e1d7 | 1576 | long ret; |
71c20a06 ID |
1577 | |
1578 | dout("%s %p\n", __func__, obj_request); | |
2894e1d7 ID |
1579 | ret = wait_for_completion_interruptible_timeout( |
1580 | &obj_request->completion, | |
1581 | ceph_timeout_jiffies(timeout)); | |
1582 | if (ret <= 0) { | |
1583 | if (ret == 0) | |
1584 | ret = -ETIMEDOUT; | |
71c20a06 | 1585 | rbd_obj_request_end(obj_request); |
2894e1d7 ID |
1586 | } else { |
1587 | ret = 0; | |
71c20a06 ID |
1588 | } |
1589 | ||
2894e1d7 ID |
1590 | dout("%s %p ret %d\n", __func__, obj_request, (int)ret); |
1591 | return ret; | |
1592 | } | |
1593 | ||
1594 | static int rbd_obj_request_wait(struct rbd_obj_request *obj_request) | |
1595 | { | |
1596 | return __rbd_obj_request_wait(obj_request, 0); | |
1597 | } | |
1598 | ||
1599 | static int rbd_obj_request_wait_timeout(struct rbd_obj_request *obj_request, | |
1600 | unsigned long timeout) | |
1601 | { | |
1602 | return __rbd_obj_request_wait(obj_request, timeout); | |
71c20a06 ID |
1603 | } |
1604 | ||
bf0d5f50 AE |
1605 | static void rbd_img_request_complete(struct rbd_img_request *img_request) |
1606 | { | |
55f27e09 | 1607 | |
37206ee5 | 1608 | dout("%s: img %p\n", __func__, img_request); |
55f27e09 AE |
1609 | |
1610 | /* | |
1611 | * If no error occurred, compute the aggregate transfer | |
1612 | * count for the image request. We could instead use | |
1613 | * atomic64_cmpxchg() to update it as each object request | |
1614 | * completes; not clear which way is better off hand. | |
1615 | */ | |
1616 | if (!img_request->result) { | |
1617 | struct rbd_obj_request *obj_request; | |
1618 | u64 xferred = 0; | |
1619 | ||
1620 | for_each_obj_request(img_request, obj_request) | |
1621 | xferred += obj_request->xferred; | |
1622 | img_request->xferred = xferred; | |
1623 | } | |
1624 | ||
bf0d5f50 AE |
1625 | if (img_request->callback) |
1626 | img_request->callback(img_request); | |
1627 | else | |
1628 | rbd_img_request_put(img_request); | |
1629 | } | |
1630 | ||
0c425248 AE |
1631 | /* |
1632 | * The default/initial value for all image request flags is 0. Each | |
1633 | * is conditionally set to 1 at image request initialization time | |
1634 | * and currently never change thereafter. | |
1635 | */ | |
1636 | static void img_request_write_set(struct rbd_img_request *img_request) | |
1637 | { | |
1638 | set_bit(IMG_REQ_WRITE, &img_request->flags); | |
1639 | smp_mb(); | |
1640 | } | |
1641 | ||
1642 | static bool img_request_write_test(struct rbd_img_request *img_request) | |
1643 | { | |
1644 | smp_mb(); | |
1645 | return test_bit(IMG_REQ_WRITE, &img_request->flags) != 0; | |
1646 | } | |
1647 | ||
90e98c52 GZ |
1648 | /* |
1649 | * Set the discard flag when the img_request is an discard request | |
1650 | */ | |
1651 | static void img_request_discard_set(struct rbd_img_request *img_request) | |
1652 | { | |
1653 | set_bit(IMG_REQ_DISCARD, &img_request->flags); | |
1654 | smp_mb(); | |
1655 | } | |
1656 | ||
1657 | static bool img_request_discard_test(struct rbd_img_request *img_request) | |
1658 | { | |
1659 | smp_mb(); | |
1660 | return test_bit(IMG_REQ_DISCARD, &img_request->flags) != 0; | |
1661 | } | |
1662 | ||
9849e986 AE |
1663 | static void img_request_child_set(struct rbd_img_request *img_request) |
1664 | { | |
1665 | set_bit(IMG_REQ_CHILD, &img_request->flags); | |
1666 | smp_mb(); | |
1667 | } | |
1668 | ||
e93f3152 AE |
1669 | static void img_request_child_clear(struct rbd_img_request *img_request) |
1670 | { | |
1671 | clear_bit(IMG_REQ_CHILD, &img_request->flags); | |
1672 | smp_mb(); | |
1673 | } | |
1674 | ||
9849e986 AE |
1675 | static bool img_request_child_test(struct rbd_img_request *img_request) |
1676 | { | |
1677 | smp_mb(); | |
1678 | return test_bit(IMG_REQ_CHILD, &img_request->flags) != 0; | |
1679 | } | |
1680 | ||
d0b2e944 AE |
1681 | static void img_request_layered_set(struct rbd_img_request *img_request) |
1682 | { | |
1683 | set_bit(IMG_REQ_LAYERED, &img_request->flags); | |
1684 | smp_mb(); | |
1685 | } | |
1686 | ||
a2acd00e AE |
1687 | static void img_request_layered_clear(struct rbd_img_request *img_request) |
1688 | { | |
1689 | clear_bit(IMG_REQ_LAYERED, &img_request->flags); | |
1690 | smp_mb(); | |
1691 | } | |
1692 | ||
d0b2e944 AE |
1693 | static bool img_request_layered_test(struct rbd_img_request *img_request) |
1694 | { | |
1695 | smp_mb(); | |
1696 | return test_bit(IMG_REQ_LAYERED, &img_request->flags) != 0; | |
1697 | } | |
1698 | ||
3b434a2a JD |
1699 | static enum obj_operation_type |
1700 | rbd_img_request_op_type(struct rbd_img_request *img_request) | |
1701 | { | |
1702 | if (img_request_write_test(img_request)) | |
1703 | return OBJ_OP_WRITE; | |
1704 | else if (img_request_discard_test(img_request)) | |
1705 | return OBJ_OP_DISCARD; | |
1706 | else | |
1707 | return OBJ_OP_READ; | |
1708 | } | |
1709 | ||
6e2a4505 AE |
1710 | static void |
1711 | rbd_img_obj_request_read_callback(struct rbd_obj_request *obj_request) | |
1712 | { | |
b9434c5b AE |
1713 | u64 xferred = obj_request->xferred; |
1714 | u64 length = obj_request->length; | |
1715 | ||
6e2a4505 AE |
1716 | dout("%s: obj %p img %p result %d %llu/%llu\n", __func__, |
1717 | obj_request, obj_request->img_request, obj_request->result, | |
b9434c5b | 1718 | xferred, length); |
6e2a4505 | 1719 | /* |
17c1cc1d JD |
1720 | * ENOENT means a hole in the image. We zero-fill the entire |
1721 | * length of the request. A short read also implies zero-fill | |
1722 | * to the end of the request. An error requires the whole | |
1723 | * length of the request to be reported finished with an error | |
1724 | * to the block layer. In each case we update the xferred | |
1725 | * count to indicate the whole request was satisfied. | |
6e2a4505 | 1726 | */ |
b9434c5b | 1727 | rbd_assert(obj_request->type != OBJ_REQUEST_NODATA); |
6e2a4505 | 1728 | if (obj_request->result == -ENOENT) { |
b9434c5b AE |
1729 | if (obj_request->type == OBJ_REQUEST_BIO) |
1730 | zero_bio_chain(obj_request->bio_list, 0); | |
1731 | else | |
1732 | zero_pages(obj_request->pages, 0, length); | |
6e2a4505 | 1733 | obj_request->result = 0; |
b9434c5b AE |
1734 | } else if (xferred < length && !obj_request->result) { |
1735 | if (obj_request->type == OBJ_REQUEST_BIO) | |
1736 | zero_bio_chain(obj_request->bio_list, xferred); | |
1737 | else | |
1738 | zero_pages(obj_request->pages, xferred, length); | |
6e2a4505 | 1739 | } |
17c1cc1d | 1740 | obj_request->xferred = length; |
6e2a4505 AE |
1741 | obj_request_done_set(obj_request); |
1742 | } | |
1743 | ||
bf0d5f50 AE |
1744 | static void rbd_obj_request_complete(struct rbd_obj_request *obj_request) |
1745 | { | |
37206ee5 AE |
1746 | dout("%s: obj %p cb %p\n", __func__, obj_request, |
1747 | obj_request->callback); | |
bf0d5f50 AE |
1748 | if (obj_request->callback) |
1749 | obj_request->callback(obj_request); | |
788e2df3 AE |
1750 | else |
1751 | complete_all(&obj_request->completion); | |
bf0d5f50 AE |
1752 | } |
1753 | ||
c47f9371 | 1754 | static void rbd_osd_trivial_callback(struct rbd_obj_request *obj_request) |
39bf2c5d AE |
1755 | { |
1756 | dout("%s: obj %p\n", __func__, obj_request); | |
1757 | obj_request_done_set(obj_request); | |
1758 | } | |
1759 | ||
c47f9371 | 1760 | static void rbd_osd_read_callback(struct rbd_obj_request *obj_request) |
bf0d5f50 | 1761 | { |
57acbaa7 | 1762 | struct rbd_img_request *img_request = NULL; |
a9e8ba2c | 1763 | struct rbd_device *rbd_dev = NULL; |
57acbaa7 AE |
1764 | bool layered = false; |
1765 | ||
1766 | if (obj_request_img_data_test(obj_request)) { | |
1767 | img_request = obj_request->img_request; | |
1768 | layered = img_request && img_request_layered_test(img_request); | |
a9e8ba2c | 1769 | rbd_dev = img_request->rbd_dev; |
57acbaa7 | 1770 | } |
8b3e1a56 AE |
1771 | |
1772 | dout("%s: obj %p img %p result %d %llu/%llu\n", __func__, | |
1773 | obj_request, img_request, obj_request->result, | |
1774 | obj_request->xferred, obj_request->length); | |
a9e8ba2c AE |
1775 | if (layered && obj_request->result == -ENOENT && |
1776 | obj_request->img_offset < rbd_dev->parent_overlap) | |
8b3e1a56 AE |
1777 | rbd_img_parent_read(obj_request); |
1778 | else if (img_request) | |
6e2a4505 AE |
1779 | rbd_img_obj_request_read_callback(obj_request); |
1780 | else | |
1781 | obj_request_done_set(obj_request); | |
bf0d5f50 AE |
1782 | } |
1783 | ||
c47f9371 | 1784 | static void rbd_osd_write_callback(struct rbd_obj_request *obj_request) |
bf0d5f50 | 1785 | { |
1b83bef2 SW |
1786 | dout("%s: obj %p result %d %llu\n", __func__, obj_request, |
1787 | obj_request->result, obj_request->length); | |
1788 | /* | |
8b3e1a56 AE |
1789 | * There is no such thing as a successful short write. Set |
1790 | * it to our originally-requested length. | |
1b83bef2 SW |
1791 | */ |
1792 | obj_request->xferred = obj_request->length; | |
07741308 | 1793 | obj_request_done_set(obj_request); |
bf0d5f50 AE |
1794 | } |
1795 | ||
90e98c52 GZ |
1796 | static void rbd_osd_discard_callback(struct rbd_obj_request *obj_request) |
1797 | { | |
1798 | dout("%s: obj %p result %d %llu\n", __func__, obj_request, | |
1799 | obj_request->result, obj_request->length); | |
1800 | /* | |
1801 | * There is no such thing as a successful short discard. Set | |
1802 | * it to our originally-requested length. | |
1803 | */ | |
1804 | obj_request->xferred = obj_request->length; | |
d0265de7 JD |
1805 | /* discarding a non-existent object is not a problem */ |
1806 | if (obj_request->result == -ENOENT) | |
1807 | obj_request->result = 0; | |
90e98c52 GZ |
1808 | obj_request_done_set(obj_request); |
1809 | } | |
1810 | ||
fbfab539 AE |
1811 | /* |
1812 | * For a simple stat call there's nothing to do. We'll do more if | |
1813 | * this is part of a write sequence for a layered image. | |
1814 | */ | |
c47f9371 | 1815 | static void rbd_osd_stat_callback(struct rbd_obj_request *obj_request) |
fbfab539 | 1816 | { |
37206ee5 | 1817 | dout("%s: obj %p\n", __func__, obj_request); |
fbfab539 AE |
1818 | obj_request_done_set(obj_request); |
1819 | } | |
1820 | ||
bf0d5f50 AE |
1821 | static void rbd_osd_req_callback(struct ceph_osd_request *osd_req, |
1822 | struct ceph_msg *msg) | |
1823 | { | |
1824 | struct rbd_obj_request *obj_request = osd_req->r_priv; | |
bf0d5f50 AE |
1825 | u16 opcode; |
1826 | ||
37206ee5 | 1827 | dout("%s: osd_req %p msg %p\n", __func__, osd_req, msg); |
bf0d5f50 | 1828 | rbd_assert(osd_req == obj_request->osd_req); |
57acbaa7 AE |
1829 | if (obj_request_img_data_test(obj_request)) { |
1830 | rbd_assert(obj_request->img_request); | |
1831 | rbd_assert(obj_request->which != BAD_WHICH); | |
1832 | } else { | |
1833 | rbd_assert(obj_request->which == BAD_WHICH); | |
1834 | } | |
bf0d5f50 | 1835 | |
1b83bef2 SW |
1836 | if (osd_req->r_result < 0) |
1837 | obj_request->result = osd_req->r_result; | |
bf0d5f50 | 1838 | |
7cc69d42 | 1839 | rbd_assert(osd_req->r_num_ops <= CEPH_OSD_MAX_OP); |
bf0d5f50 | 1840 | |
c47f9371 AE |
1841 | /* |
1842 | * We support a 64-bit length, but ultimately it has to be | |
7ad18afa CH |
1843 | * passed to the block layer, which just supports a 32-bit |
1844 | * length field. | |
c47f9371 | 1845 | */ |
1b83bef2 | 1846 | obj_request->xferred = osd_req->r_reply_op_len[0]; |
8b3e1a56 | 1847 | rbd_assert(obj_request->xferred < (u64)UINT_MAX); |
0ccd5926 | 1848 | |
79528734 | 1849 | opcode = osd_req->r_ops[0].op; |
bf0d5f50 AE |
1850 | switch (opcode) { |
1851 | case CEPH_OSD_OP_READ: | |
c47f9371 | 1852 | rbd_osd_read_callback(obj_request); |
bf0d5f50 | 1853 | break; |
0ccd5926 ID |
1854 | case CEPH_OSD_OP_SETALLOCHINT: |
1855 | rbd_assert(osd_req->r_ops[1].op == CEPH_OSD_OP_WRITE); | |
1856 | /* fall through */ | |
bf0d5f50 | 1857 | case CEPH_OSD_OP_WRITE: |
c47f9371 | 1858 | rbd_osd_write_callback(obj_request); |
bf0d5f50 | 1859 | break; |
fbfab539 | 1860 | case CEPH_OSD_OP_STAT: |
c47f9371 | 1861 | rbd_osd_stat_callback(obj_request); |
fbfab539 | 1862 | break; |
90e98c52 GZ |
1863 | case CEPH_OSD_OP_DELETE: |
1864 | case CEPH_OSD_OP_TRUNCATE: | |
1865 | case CEPH_OSD_OP_ZERO: | |
1866 | rbd_osd_discard_callback(obj_request); | |
1867 | break; | |
36be9a76 | 1868 | case CEPH_OSD_OP_CALL: |
b8d70035 | 1869 | case CEPH_OSD_OP_NOTIFY_ACK: |
9969ebc5 | 1870 | case CEPH_OSD_OP_WATCH: |
c47f9371 | 1871 | rbd_osd_trivial_callback(obj_request); |
9969ebc5 | 1872 | break; |
bf0d5f50 | 1873 | default: |
9584d508 | 1874 | rbd_warn(NULL, "%s: unsupported op %hu", |
bf0d5f50 AE |
1875 | obj_request->object_name, (unsigned short) opcode); |
1876 | break; | |
1877 | } | |
1878 | ||
07741308 | 1879 | if (obj_request_done_test(obj_request)) |
bf0d5f50 AE |
1880 | rbd_obj_request_complete(obj_request); |
1881 | } | |
1882 | ||
9d4df01f | 1883 | static void rbd_osd_req_format_read(struct rbd_obj_request *obj_request) |
430c28c3 AE |
1884 | { |
1885 | struct rbd_img_request *img_request = obj_request->img_request; | |
8c042b0d | 1886 | struct ceph_osd_request *osd_req = obj_request->osd_req; |
9d4df01f | 1887 | u64 snap_id; |
430c28c3 | 1888 | |
8c042b0d | 1889 | rbd_assert(osd_req != NULL); |
430c28c3 | 1890 | |
9d4df01f | 1891 | snap_id = img_request ? img_request->snap_id : CEPH_NOSNAP; |
8c042b0d | 1892 | ceph_osdc_build_request(osd_req, obj_request->offset, |
9d4df01f AE |
1893 | NULL, snap_id, NULL); |
1894 | } | |
1895 | ||
1896 | static void rbd_osd_req_format_write(struct rbd_obj_request *obj_request) | |
1897 | { | |
1898 | struct rbd_img_request *img_request = obj_request->img_request; | |
1899 | struct ceph_osd_request *osd_req = obj_request->osd_req; | |
1900 | struct ceph_snap_context *snapc; | |
1901 | struct timespec mtime = CURRENT_TIME; | |
1902 | ||
1903 | rbd_assert(osd_req != NULL); | |
1904 | ||
1905 | snapc = img_request ? img_request->snapc : NULL; | |
1906 | ceph_osdc_build_request(osd_req, obj_request->offset, | |
1907 | snapc, CEPH_NOSNAP, &mtime); | |
430c28c3 AE |
1908 | } |
1909 | ||
0ccd5926 ID |
1910 | /* |
1911 | * Create an osd request. A read request has one osd op (read). | |
1912 | * A write request has either one (watch) or two (hint+write) osd ops. | |
1913 | * (All rbd data writes are prefixed with an allocation hint op, but | |
1914 | * technically osd watch is a write request, hence this distinction.) | |
1915 | */ | |
bf0d5f50 AE |
1916 | static struct ceph_osd_request *rbd_osd_req_create( |
1917 | struct rbd_device *rbd_dev, | |
6d2940c8 | 1918 | enum obj_operation_type op_type, |
deb236b3 | 1919 | unsigned int num_ops, |
430c28c3 | 1920 | struct rbd_obj_request *obj_request) |
bf0d5f50 | 1921 | { |
bf0d5f50 AE |
1922 | struct ceph_snap_context *snapc = NULL; |
1923 | struct ceph_osd_client *osdc; | |
1924 | struct ceph_osd_request *osd_req; | |
bf0d5f50 | 1925 | |
90e98c52 GZ |
1926 | if (obj_request_img_data_test(obj_request) && |
1927 | (op_type == OBJ_OP_DISCARD || op_type == OBJ_OP_WRITE)) { | |
6365d33a | 1928 | struct rbd_img_request *img_request = obj_request->img_request; |
90e98c52 GZ |
1929 | if (op_type == OBJ_OP_WRITE) { |
1930 | rbd_assert(img_request_write_test(img_request)); | |
1931 | } else { | |
1932 | rbd_assert(img_request_discard_test(img_request)); | |
1933 | } | |
6d2940c8 | 1934 | snapc = img_request->snapc; |
bf0d5f50 AE |
1935 | } |
1936 | ||
6d2940c8 | 1937 | rbd_assert(num_ops == 1 || ((op_type == OBJ_OP_WRITE) && num_ops == 2)); |
deb236b3 ID |
1938 | |
1939 | /* Allocate and initialize the request, for the num_ops ops */ | |
bf0d5f50 AE |
1940 | |
1941 | osdc = &rbd_dev->rbd_client->client->osdc; | |
deb236b3 ID |
1942 | osd_req = ceph_osdc_alloc_request(osdc, snapc, num_ops, false, |
1943 | GFP_ATOMIC); | |
bf0d5f50 AE |
1944 | if (!osd_req) |
1945 | return NULL; /* ENOMEM */ | |
bf0d5f50 | 1946 | |
90e98c52 | 1947 | if (op_type == OBJ_OP_WRITE || op_type == OBJ_OP_DISCARD) |
bf0d5f50 | 1948 | osd_req->r_flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK; |
430c28c3 | 1949 | else |
bf0d5f50 | 1950 | osd_req->r_flags = CEPH_OSD_FLAG_READ; |
bf0d5f50 AE |
1951 | |
1952 | osd_req->r_callback = rbd_osd_req_callback; | |
1953 | osd_req->r_priv = obj_request; | |
1954 | ||
3c972c95 ID |
1955 | osd_req->r_base_oloc.pool = ceph_file_layout_pg_pool(rbd_dev->layout); |
1956 | ceph_oid_set_name(&osd_req->r_base_oid, obj_request->object_name); | |
bf0d5f50 | 1957 | |
bf0d5f50 AE |
1958 | return osd_req; |
1959 | } | |
1960 | ||
0eefd470 | 1961 | /* |
d3246fb0 JD |
1962 | * Create a copyup osd request based on the information in the object |
1963 | * request supplied. A copyup request has two or three osd ops, a | |
1964 | * copyup method call, potentially a hint op, and a write or truncate | |
1965 | * or zero op. | |
0eefd470 AE |
1966 | */ |
1967 | static struct ceph_osd_request * | |
1968 | rbd_osd_req_create_copyup(struct rbd_obj_request *obj_request) | |
1969 | { | |
1970 | struct rbd_img_request *img_request; | |
1971 | struct ceph_snap_context *snapc; | |
1972 | struct rbd_device *rbd_dev; | |
1973 | struct ceph_osd_client *osdc; | |
1974 | struct ceph_osd_request *osd_req; | |
d3246fb0 | 1975 | int num_osd_ops = 3; |
0eefd470 AE |
1976 | |
1977 | rbd_assert(obj_request_img_data_test(obj_request)); | |
1978 | img_request = obj_request->img_request; | |
1979 | rbd_assert(img_request); | |
d3246fb0 JD |
1980 | rbd_assert(img_request_write_test(img_request) || |
1981 | img_request_discard_test(img_request)); | |
0eefd470 | 1982 | |
d3246fb0 JD |
1983 | if (img_request_discard_test(img_request)) |
1984 | num_osd_ops = 2; | |
1985 | ||
1986 | /* Allocate and initialize the request, for all the ops */ | |
0eefd470 AE |
1987 | |
1988 | snapc = img_request->snapc; | |
1989 | rbd_dev = img_request->rbd_dev; | |
1990 | osdc = &rbd_dev->rbd_client->client->osdc; | |
d3246fb0 JD |
1991 | osd_req = ceph_osdc_alloc_request(osdc, snapc, num_osd_ops, |
1992 | false, GFP_ATOMIC); | |
0eefd470 AE |
1993 | if (!osd_req) |
1994 | return NULL; /* ENOMEM */ | |
1995 | ||
1996 | osd_req->r_flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK; | |
1997 | osd_req->r_callback = rbd_osd_req_callback; | |
1998 | osd_req->r_priv = obj_request; | |
1999 | ||
3c972c95 ID |
2000 | osd_req->r_base_oloc.pool = ceph_file_layout_pg_pool(rbd_dev->layout); |
2001 | ceph_oid_set_name(&osd_req->r_base_oid, obj_request->object_name); | |
0eefd470 | 2002 | |
0eefd470 AE |
2003 | return osd_req; |
2004 | } | |
2005 | ||
2006 | ||
bf0d5f50 AE |
2007 | static void rbd_osd_req_destroy(struct ceph_osd_request *osd_req) |
2008 | { | |
2009 | ceph_osdc_put_request(osd_req); | |
2010 | } | |
2011 | ||
2012 | /* object_name is assumed to be a non-null pointer and NUL-terminated */ | |
2013 | ||
2014 | static struct rbd_obj_request *rbd_obj_request_create(const char *object_name, | |
2015 | u64 offset, u64 length, | |
2016 | enum obj_request_type type) | |
2017 | { | |
2018 | struct rbd_obj_request *obj_request; | |
2019 | size_t size; | |
2020 | char *name; | |
2021 | ||
2022 | rbd_assert(obj_request_type_valid(type)); | |
2023 | ||
2024 | size = strlen(object_name) + 1; | |
5a60e876 | 2025 | name = kmalloc(size, GFP_NOIO); |
f907ad55 | 2026 | if (!name) |
bf0d5f50 AE |
2027 | return NULL; |
2028 | ||
5a60e876 | 2029 | obj_request = kmem_cache_zalloc(rbd_obj_request_cache, GFP_NOIO); |
f907ad55 AE |
2030 | if (!obj_request) { |
2031 | kfree(name); | |
2032 | return NULL; | |
2033 | } | |
2034 | ||
bf0d5f50 AE |
2035 | obj_request->object_name = memcpy(name, object_name, size); |
2036 | obj_request->offset = offset; | |
2037 | obj_request->length = length; | |
926f9b3f | 2038 | obj_request->flags = 0; |
bf0d5f50 AE |
2039 | obj_request->which = BAD_WHICH; |
2040 | obj_request->type = type; | |
2041 | INIT_LIST_HEAD(&obj_request->links); | |
788e2df3 | 2042 | init_completion(&obj_request->completion); |
bf0d5f50 AE |
2043 | kref_init(&obj_request->kref); |
2044 | ||
37206ee5 AE |
2045 | dout("%s: \"%s\" %llu/%llu %d -> obj %p\n", __func__, object_name, |
2046 | offset, length, (int)type, obj_request); | |
2047 | ||
bf0d5f50 AE |
2048 | return obj_request; |
2049 | } | |
2050 | ||
2051 | static void rbd_obj_request_destroy(struct kref *kref) | |
2052 | { | |
2053 | struct rbd_obj_request *obj_request; | |
2054 | ||
2055 | obj_request = container_of(kref, struct rbd_obj_request, kref); | |
2056 | ||
37206ee5 AE |
2057 | dout("%s: obj %p\n", __func__, obj_request); |
2058 | ||
bf0d5f50 AE |
2059 | rbd_assert(obj_request->img_request == NULL); |
2060 | rbd_assert(obj_request->which == BAD_WHICH); | |
2061 | ||
2062 | if (obj_request->osd_req) | |
2063 | rbd_osd_req_destroy(obj_request->osd_req); | |
2064 | ||
2065 | rbd_assert(obj_request_type_valid(obj_request->type)); | |
2066 | switch (obj_request->type) { | |
9969ebc5 AE |
2067 | case OBJ_REQUEST_NODATA: |
2068 | break; /* Nothing to do */ | |
bf0d5f50 AE |
2069 | case OBJ_REQUEST_BIO: |
2070 | if (obj_request->bio_list) | |
2071 | bio_chain_put(obj_request->bio_list); | |
2072 | break; | |
788e2df3 AE |
2073 | case OBJ_REQUEST_PAGES: |
2074 | if (obj_request->pages) | |
2075 | ceph_release_page_vector(obj_request->pages, | |
2076 | obj_request->page_count); | |
2077 | break; | |
bf0d5f50 AE |
2078 | } |
2079 | ||
f907ad55 | 2080 | kfree(obj_request->object_name); |
868311b1 AE |
2081 | obj_request->object_name = NULL; |
2082 | kmem_cache_free(rbd_obj_request_cache, obj_request); | |
bf0d5f50 AE |
2083 | } |
2084 | ||
fb65d228 AE |
2085 | /* It's OK to call this for a device with no parent */ |
2086 | ||
2087 | static void rbd_spec_put(struct rbd_spec *spec); | |
2088 | static void rbd_dev_unparent(struct rbd_device *rbd_dev) | |
2089 | { | |
2090 | rbd_dev_remove_parent(rbd_dev); | |
2091 | rbd_spec_put(rbd_dev->parent_spec); | |
2092 | rbd_dev->parent_spec = NULL; | |
2093 | rbd_dev->parent_overlap = 0; | |
2094 | } | |
2095 | ||
a2acd00e AE |
2096 | /* |
2097 | * Parent image reference counting is used to determine when an | |
2098 | * image's parent fields can be safely torn down--after there are no | |
2099 | * more in-flight requests to the parent image. When the last | |
2100 | * reference is dropped, cleaning them up is safe. | |
2101 | */ | |
2102 | static void rbd_dev_parent_put(struct rbd_device *rbd_dev) | |
2103 | { | |
2104 | int counter; | |
2105 | ||
2106 | if (!rbd_dev->parent_spec) | |
2107 | return; | |
2108 | ||
2109 | counter = atomic_dec_return_safe(&rbd_dev->parent_ref); | |
2110 | if (counter > 0) | |
2111 | return; | |
2112 | ||
2113 | /* Last reference; clean up parent data structures */ | |
2114 | ||
2115 | if (!counter) | |
2116 | rbd_dev_unparent(rbd_dev); | |
2117 | else | |
9584d508 | 2118 | rbd_warn(rbd_dev, "parent reference underflow"); |
a2acd00e AE |
2119 | } |
2120 | ||
2121 | /* | |
2122 | * If an image has a non-zero parent overlap, get a reference to its | |
2123 | * parent. | |
2124 | * | |
2125 | * Returns true if the rbd device has a parent with a non-zero | |
2126 | * overlap and a reference for it was successfully taken, or | |
2127 | * false otherwise. | |
2128 | */ | |
2129 | static bool rbd_dev_parent_get(struct rbd_device *rbd_dev) | |
2130 | { | |
ae43e9d0 | 2131 | int counter = 0; |
a2acd00e AE |
2132 | |
2133 | if (!rbd_dev->parent_spec) | |
2134 | return false; | |
2135 | ||
ae43e9d0 ID |
2136 | down_read(&rbd_dev->header_rwsem); |
2137 | if (rbd_dev->parent_overlap) | |
2138 | counter = atomic_inc_return_safe(&rbd_dev->parent_ref); | |
2139 | up_read(&rbd_dev->header_rwsem); | |
a2acd00e AE |
2140 | |
2141 | if (counter < 0) | |
9584d508 | 2142 | rbd_warn(rbd_dev, "parent reference overflow"); |
a2acd00e | 2143 | |
ae43e9d0 | 2144 | return counter > 0; |
a2acd00e AE |
2145 | } |
2146 | ||
bf0d5f50 AE |
2147 | /* |
2148 | * Caller is responsible for filling in the list of object requests | |
2149 | * that comprises the image request, and the Linux request pointer | |
2150 | * (if there is one). | |
2151 | */ | |
cc344fa1 AE |
2152 | static struct rbd_img_request *rbd_img_request_create( |
2153 | struct rbd_device *rbd_dev, | |
bf0d5f50 | 2154 | u64 offset, u64 length, |
6d2940c8 | 2155 | enum obj_operation_type op_type, |
4e752f0a | 2156 | struct ceph_snap_context *snapc) |
bf0d5f50 AE |
2157 | { |
2158 | struct rbd_img_request *img_request; | |
bf0d5f50 | 2159 | |
7a716aac | 2160 | img_request = kmem_cache_alloc(rbd_img_request_cache, GFP_NOIO); |
bf0d5f50 AE |
2161 | if (!img_request) |
2162 | return NULL; | |
2163 | ||
bf0d5f50 AE |
2164 | img_request->rq = NULL; |
2165 | img_request->rbd_dev = rbd_dev; | |
2166 | img_request->offset = offset; | |
2167 | img_request->length = length; | |
0c425248 | 2168 | img_request->flags = 0; |
90e98c52 GZ |
2169 | if (op_type == OBJ_OP_DISCARD) { |
2170 | img_request_discard_set(img_request); | |
2171 | img_request->snapc = snapc; | |
2172 | } else if (op_type == OBJ_OP_WRITE) { | |
0c425248 | 2173 | img_request_write_set(img_request); |
4e752f0a | 2174 | img_request->snapc = snapc; |
0c425248 | 2175 | } else { |
bf0d5f50 | 2176 | img_request->snap_id = rbd_dev->spec->snap_id; |
0c425248 | 2177 | } |
a2acd00e | 2178 | if (rbd_dev_parent_get(rbd_dev)) |
d0b2e944 | 2179 | img_request_layered_set(img_request); |
bf0d5f50 AE |
2180 | spin_lock_init(&img_request->completion_lock); |
2181 | img_request->next_completion = 0; | |
2182 | img_request->callback = NULL; | |
a5a337d4 | 2183 | img_request->result = 0; |
bf0d5f50 AE |
2184 | img_request->obj_request_count = 0; |
2185 | INIT_LIST_HEAD(&img_request->obj_requests); | |
2186 | kref_init(&img_request->kref); | |
2187 | ||
37206ee5 | 2188 | dout("%s: rbd_dev %p %s %llu/%llu -> img %p\n", __func__, rbd_dev, |
6d2940c8 | 2189 | obj_op_name(op_type), offset, length, img_request); |
37206ee5 | 2190 | |
bf0d5f50 AE |
2191 | return img_request; |
2192 | } | |
2193 | ||
2194 | static void rbd_img_request_destroy(struct kref *kref) | |
2195 | { | |
2196 | struct rbd_img_request *img_request; | |
2197 | struct rbd_obj_request *obj_request; | |
2198 | struct rbd_obj_request *next_obj_request; | |
2199 | ||
2200 | img_request = container_of(kref, struct rbd_img_request, kref); | |
2201 | ||
37206ee5 AE |
2202 | dout("%s: img %p\n", __func__, img_request); |
2203 | ||
bf0d5f50 AE |
2204 | for_each_obj_request_safe(img_request, obj_request, next_obj_request) |
2205 | rbd_img_obj_request_del(img_request, obj_request); | |
25dcf954 | 2206 | rbd_assert(img_request->obj_request_count == 0); |
bf0d5f50 | 2207 | |
a2acd00e AE |
2208 | if (img_request_layered_test(img_request)) { |
2209 | img_request_layered_clear(img_request); | |
2210 | rbd_dev_parent_put(img_request->rbd_dev); | |
2211 | } | |
2212 | ||
bef95455 JD |
2213 | if (img_request_write_test(img_request) || |
2214 | img_request_discard_test(img_request)) | |
812164f8 | 2215 | ceph_put_snap_context(img_request->snapc); |
bf0d5f50 | 2216 | |
1c2a9dfe | 2217 | kmem_cache_free(rbd_img_request_cache, img_request); |
bf0d5f50 AE |
2218 | } |
2219 | ||
e93f3152 AE |
2220 | static struct rbd_img_request *rbd_parent_request_create( |
2221 | struct rbd_obj_request *obj_request, | |
2222 | u64 img_offset, u64 length) | |
2223 | { | |
2224 | struct rbd_img_request *parent_request; | |
2225 | struct rbd_device *rbd_dev; | |
2226 | ||
2227 | rbd_assert(obj_request->img_request); | |
2228 | rbd_dev = obj_request->img_request->rbd_dev; | |
2229 | ||
4e752f0a | 2230 | parent_request = rbd_img_request_create(rbd_dev->parent, img_offset, |
6d2940c8 | 2231 | length, OBJ_OP_READ, NULL); |
e93f3152 AE |
2232 | if (!parent_request) |
2233 | return NULL; | |
2234 | ||
2235 | img_request_child_set(parent_request); | |
2236 | rbd_obj_request_get(obj_request); | |
2237 | parent_request->obj_request = obj_request; | |
2238 | ||
2239 | return parent_request; | |
2240 | } | |
2241 | ||
2242 | static void rbd_parent_request_destroy(struct kref *kref) | |
2243 | { | |
2244 | struct rbd_img_request *parent_request; | |
2245 | struct rbd_obj_request *orig_request; | |
2246 | ||
2247 | parent_request = container_of(kref, struct rbd_img_request, kref); | |
2248 | orig_request = parent_request->obj_request; | |
2249 | ||
2250 | parent_request->obj_request = NULL; | |
2251 | rbd_obj_request_put(orig_request); | |
2252 | img_request_child_clear(parent_request); | |
2253 | ||
2254 | rbd_img_request_destroy(kref); | |
2255 | } | |
2256 | ||
1217857f AE |
2257 | static bool rbd_img_obj_end_request(struct rbd_obj_request *obj_request) |
2258 | { | |
6365d33a | 2259 | struct rbd_img_request *img_request; |
1217857f AE |
2260 | unsigned int xferred; |
2261 | int result; | |
8b3e1a56 | 2262 | bool more; |
1217857f | 2263 | |
6365d33a AE |
2264 | rbd_assert(obj_request_img_data_test(obj_request)); |
2265 | img_request = obj_request->img_request; | |
2266 | ||
1217857f AE |
2267 | rbd_assert(obj_request->xferred <= (u64)UINT_MAX); |
2268 | xferred = (unsigned int)obj_request->xferred; | |
2269 | result = obj_request->result; | |
2270 | if (result) { | |
2271 | struct rbd_device *rbd_dev = img_request->rbd_dev; | |
6d2940c8 GZ |
2272 | enum obj_operation_type op_type; |
2273 | ||
90e98c52 GZ |
2274 | if (img_request_discard_test(img_request)) |
2275 | op_type = OBJ_OP_DISCARD; | |
2276 | else if (img_request_write_test(img_request)) | |
2277 | op_type = OBJ_OP_WRITE; | |
2278 | else | |
2279 | op_type = OBJ_OP_READ; | |
1217857f | 2280 | |
9584d508 | 2281 | rbd_warn(rbd_dev, "%s %llx at %llx (%llx)", |
6d2940c8 GZ |
2282 | obj_op_name(op_type), obj_request->length, |
2283 | obj_request->img_offset, obj_request->offset); | |
9584d508 | 2284 | rbd_warn(rbd_dev, " result %d xferred %x", |
1217857f AE |
2285 | result, xferred); |
2286 | if (!img_request->result) | |
2287 | img_request->result = result; | |
082a75da ID |
2288 | /* |
2289 | * Need to end I/O on the entire obj_request worth of | |
2290 | * bytes in case of error. | |
2291 | */ | |
2292 | xferred = obj_request->length; | |
1217857f AE |
2293 | } |
2294 | ||
f1a4739f AE |
2295 | /* Image object requests don't own their page array */ |
2296 | ||
2297 | if (obj_request->type == OBJ_REQUEST_PAGES) { | |
2298 | obj_request->pages = NULL; | |
2299 | obj_request->page_count = 0; | |
2300 | } | |
2301 | ||
8b3e1a56 AE |
2302 | if (img_request_child_test(img_request)) { |
2303 | rbd_assert(img_request->obj_request != NULL); | |
2304 | more = obj_request->which < img_request->obj_request_count - 1; | |
2305 | } else { | |
2306 | rbd_assert(img_request->rq != NULL); | |
7ad18afa CH |
2307 | |
2308 | more = blk_update_request(img_request->rq, result, xferred); | |
2309 | if (!more) | |
2310 | __blk_mq_end_request(img_request->rq, result); | |
8b3e1a56 AE |
2311 | } |
2312 | ||
2313 | return more; | |
1217857f AE |
2314 | } |
2315 | ||
2169238d AE |
2316 | static void rbd_img_obj_callback(struct rbd_obj_request *obj_request) |
2317 | { | |
2318 | struct rbd_img_request *img_request; | |
2319 | u32 which = obj_request->which; | |
2320 | bool more = true; | |
2321 | ||
6365d33a | 2322 | rbd_assert(obj_request_img_data_test(obj_request)); |
2169238d AE |
2323 | img_request = obj_request->img_request; |
2324 | ||
2325 | dout("%s: img %p obj %p\n", __func__, img_request, obj_request); | |
2326 | rbd_assert(img_request != NULL); | |
2169238d AE |
2327 | rbd_assert(img_request->obj_request_count > 0); |
2328 | rbd_assert(which != BAD_WHICH); | |
2329 | rbd_assert(which < img_request->obj_request_count); | |
2169238d AE |
2330 | |
2331 | spin_lock_irq(&img_request->completion_lock); | |
2332 | if (which != img_request->next_completion) | |
2333 | goto out; | |
2334 | ||
2335 | for_each_obj_request_from(img_request, obj_request) { | |
2169238d AE |
2336 | rbd_assert(more); |
2337 | rbd_assert(which < img_request->obj_request_count); | |
2338 | ||
2339 | if (!obj_request_done_test(obj_request)) | |
2340 | break; | |
1217857f | 2341 | more = rbd_img_obj_end_request(obj_request); |
2169238d AE |
2342 | which++; |
2343 | } | |
2344 | ||
2345 | rbd_assert(more ^ (which == img_request->obj_request_count)); | |
2346 | img_request->next_completion = which; | |
2347 | out: | |
2348 | spin_unlock_irq(&img_request->completion_lock); | |
0f2d5be7 | 2349 | rbd_img_request_put(img_request); |
2169238d AE |
2350 | |
2351 | if (!more) | |
2352 | rbd_img_request_complete(img_request); | |
2353 | } | |
2354 | ||
3b434a2a JD |
2355 | /* |
2356 | * Add individual osd ops to the given ceph_osd_request and prepare | |
2357 | * them for submission. num_ops is the current number of | |
2358 | * osd operations already to the object request. | |
2359 | */ | |
2360 | static void rbd_img_obj_request_fill(struct rbd_obj_request *obj_request, | |
2361 | struct ceph_osd_request *osd_request, | |
2362 | enum obj_operation_type op_type, | |
2363 | unsigned int num_ops) | |
2364 | { | |
2365 | struct rbd_img_request *img_request = obj_request->img_request; | |
2366 | struct rbd_device *rbd_dev = img_request->rbd_dev; | |
2367 | u64 object_size = rbd_obj_bytes(&rbd_dev->header); | |
2368 | u64 offset = obj_request->offset; | |
2369 | u64 length = obj_request->length; | |
2370 | u64 img_end; | |
2371 | u16 opcode; | |
2372 | ||
2373 | if (op_type == OBJ_OP_DISCARD) { | |
d3246fb0 JD |
2374 | if (!offset && length == object_size && |
2375 | (!img_request_layered_test(img_request) || | |
2376 | !obj_request_overlaps_parent(obj_request))) { | |
3b434a2a JD |
2377 | opcode = CEPH_OSD_OP_DELETE; |
2378 | } else if ((offset + length == object_size)) { | |
2379 | opcode = CEPH_OSD_OP_TRUNCATE; | |
2380 | } else { | |
2381 | down_read(&rbd_dev->header_rwsem); | |
2382 | img_end = rbd_dev->header.image_size; | |
2383 | up_read(&rbd_dev->header_rwsem); | |
2384 | ||
2385 | if (obj_request->img_offset + length == img_end) | |
2386 | opcode = CEPH_OSD_OP_TRUNCATE; | |
2387 | else | |
2388 | opcode = CEPH_OSD_OP_ZERO; | |
2389 | } | |
2390 | } else if (op_type == OBJ_OP_WRITE) { | |
2391 | opcode = CEPH_OSD_OP_WRITE; | |
2392 | osd_req_op_alloc_hint_init(osd_request, num_ops, | |
2393 | object_size, object_size); | |
2394 | num_ops++; | |
2395 | } else { | |
2396 | opcode = CEPH_OSD_OP_READ; | |
2397 | } | |
2398 | ||
7e868b6e | 2399 | if (opcode == CEPH_OSD_OP_DELETE) |
144cba14 | 2400 | osd_req_op_init(osd_request, num_ops, opcode, 0); |
7e868b6e ID |
2401 | else |
2402 | osd_req_op_extent_init(osd_request, num_ops, opcode, | |
2403 | offset, length, 0, 0); | |
2404 | ||
3b434a2a JD |
2405 | if (obj_request->type == OBJ_REQUEST_BIO) |
2406 | osd_req_op_extent_osd_data_bio(osd_request, num_ops, | |
2407 | obj_request->bio_list, length); | |
2408 | else if (obj_request->type == OBJ_REQUEST_PAGES) | |
2409 | osd_req_op_extent_osd_data_pages(osd_request, num_ops, | |
2410 | obj_request->pages, length, | |
2411 | offset & ~PAGE_MASK, false, false); | |
2412 | ||
2413 | /* Discards are also writes */ | |
2414 | if (op_type == OBJ_OP_WRITE || op_type == OBJ_OP_DISCARD) | |
2415 | rbd_osd_req_format_write(obj_request); | |
2416 | else | |
2417 | rbd_osd_req_format_read(obj_request); | |
2418 | } | |
2419 | ||
f1a4739f AE |
2420 | /* |
2421 | * Split up an image request into one or more object requests, each | |
2422 | * to a different object. The "type" parameter indicates whether | |
2423 | * "data_desc" is the pointer to the head of a list of bio | |
2424 | * structures, or the base of a page array. In either case this | |
2425 | * function assumes data_desc describes memory sufficient to hold | |
2426 | * all data described by the image request. | |
2427 | */ | |
2428 | static int rbd_img_request_fill(struct rbd_img_request *img_request, | |
2429 | enum obj_request_type type, | |
2430 | void *data_desc) | |
bf0d5f50 AE |
2431 | { |
2432 | struct rbd_device *rbd_dev = img_request->rbd_dev; | |
2433 | struct rbd_obj_request *obj_request = NULL; | |
2434 | struct rbd_obj_request *next_obj_request; | |
a158073c | 2435 | struct bio *bio_list = NULL; |
f1a4739f | 2436 | unsigned int bio_offset = 0; |
a158073c | 2437 | struct page **pages = NULL; |
6d2940c8 | 2438 | enum obj_operation_type op_type; |
7da22d29 | 2439 | u64 img_offset; |
bf0d5f50 | 2440 | u64 resid; |
bf0d5f50 | 2441 | |
f1a4739f AE |
2442 | dout("%s: img %p type %d data_desc %p\n", __func__, img_request, |
2443 | (int)type, data_desc); | |
37206ee5 | 2444 | |
7da22d29 | 2445 | img_offset = img_request->offset; |
bf0d5f50 | 2446 | resid = img_request->length; |
4dda41d3 | 2447 | rbd_assert(resid > 0); |
3b434a2a | 2448 | op_type = rbd_img_request_op_type(img_request); |
f1a4739f AE |
2449 | |
2450 | if (type == OBJ_REQUEST_BIO) { | |
2451 | bio_list = data_desc; | |
4f024f37 KO |
2452 | rbd_assert(img_offset == |
2453 | bio_list->bi_iter.bi_sector << SECTOR_SHIFT); | |
90e98c52 | 2454 | } else if (type == OBJ_REQUEST_PAGES) { |
f1a4739f AE |
2455 | pages = data_desc; |
2456 | } | |
2457 | ||
bf0d5f50 | 2458 | while (resid) { |
2fa12320 | 2459 | struct ceph_osd_request *osd_req; |
bf0d5f50 | 2460 | const char *object_name; |
bf0d5f50 AE |
2461 | u64 offset; |
2462 | u64 length; | |
2463 | ||
7da22d29 | 2464 | object_name = rbd_segment_name(rbd_dev, img_offset); |
bf0d5f50 AE |
2465 | if (!object_name) |
2466 | goto out_unwind; | |
7da22d29 AE |
2467 | offset = rbd_segment_offset(rbd_dev, img_offset); |
2468 | length = rbd_segment_length(rbd_dev, img_offset, resid); | |
bf0d5f50 | 2469 | obj_request = rbd_obj_request_create(object_name, |
f1a4739f | 2470 | offset, length, type); |
78c2a44a AE |
2471 | /* object request has its own copy of the object name */ |
2472 | rbd_segment_name_free(object_name); | |
bf0d5f50 AE |
2473 | if (!obj_request) |
2474 | goto out_unwind; | |
62054da6 | 2475 | |
03507db6 JD |
2476 | /* |
2477 | * set obj_request->img_request before creating the | |
2478 | * osd_request so that it gets the right snapc | |
2479 | */ | |
2480 | rbd_img_obj_request_add(img_request, obj_request); | |
bf0d5f50 | 2481 | |
f1a4739f AE |
2482 | if (type == OBJ_REQUEST_BIO) { |
2483 | unsigned int clone_size; | |
2484 | ||
2485 | rbd_assert(length <= (u64)UINT_MAX); | |
2486 | clone_size = (unsigned int)length; | |
2487 | obj_request->bio_list = | |
2488 | bio_chain_clone_range(&bio_list, | |
2489 | &bio_offset, | |
2490 | clone_size, | |
2491 | GFP_ATOMIC); | |
2492 | if (!obj_request->bio_list) | |
62054da6 | 2493 | goto out_unwind; |
90e98c52 | 2494 | } else if (type == OBJ_REQUEST_PAGES) { |
f1a4739f AE |
2495 | unsigned int page_count; |
2496 | ||
2497 | obj_request->pages = pages; | |
2498 | page_count = (u32)calc_pages_for(offset, length); | |
2499 | obj_request->page_count = page_count; | |
2500 | if ((offset + length) & ~PAGE_MASK) | |
2501 | page_count--; /* more on last page */ | |
2502 | pages += page_count; | |
2503 | } | |
bf0d5f50 | 2504 | |
6d2940c8 GZ |
2505 | osd_req = rbd_osd_req_create(rbd_dev, op_type, |
2506 | (op_type == OBJ_OP_WRITE) ? 2 : 1, | |
2507 | obj_request); | |
2fa12320 | 2508 | if (!osd_req) |
62054da6 | 2509 | goto out_unwind; |
3b434a2a | 2510 | |
2fa12320 | 2511 | obj_request->osd_req = osd_req; |
2169238d | 2512 | obj_request->callback = rbd_img_obj_callback; |
3b434a2a | 2513 | obj_request->img_offset = img_offset; |
9d4df01f | 2514 | |
3b434a2a | 2515 | rbd_img_obj_request_fill(obj_request, osd_req, op_type, 0); |
430c28c3 | 2516 | |
3b434a2a | 2517 | rbd_img_request_get(img_request); |
bf0d5f50 | 2518 | |
7da22d29 | 2519 | img_offset += length; |
bf0d5f50 AE |
2520 | resid -= length; |
2521 | } | |
2522 | ||
2523 | return 0; | |
2524 | ||
bf0d5f50 AE |
2525 | out_unwind: |
2526 | for_each_obj_request_safe(img_request, obj_request, next_obj_request) | |
42dd037c | 2527 | rbd_img_obj_request_del(img_request, obj_request); |
bf0d5f50 AE |
2528 | |
2529 | return -ENOMEM; | |
2530 | } | |
2531 | ||
0eefd470 AE |
2532 | static void |
2533 | rbd_img_obj_copyup_callback(struct rbd_obj_request *obj_request) | |
2534 | { | |
2535 | struct rbd_img_request *img_request; | |
2536 | struct rbd_device *rbd_dev; | |
ebda6408 | 2537 | struct page **pages; |
0eefd470 AE |
2538 | u32 page_count; |
2539 | ||
d3246fb0 JD |
2540 | rbd_assert(obj_request->type == OBJ_REQUEST_BIO || |
2541 | obj_request->type == OBJ_REQUEST_NODATA); | |
0eefd470 AE |
2542 | rbd_assert(obj_request_img_data_test(obj_request)); |
2543 | img_request = obj_request->img_request; | |
2544 | rbd_assert(img_request); | |
2545 | ||
2546 | rbd_dev = img_request->rbd_dev; | |
2547 | rbd_assert(rbd_dev); | |
0eefd470 | 2548 | |
ebda6408 AE |
2549 | pages = obj_request->copyup_pages; |
2550 | rbd_assert(pages != NULL); | |
0eefd470 | 2551 | obj_request->copyup_pages = NULL; |
ebda6408 AE |
2552 | page_count = obj_request->copyup_page_count; |
2553 | rbd_assert(page_count); | |
2554 | obj_request->copyup_page_count = 0; | |
2555 | ceph_release_page_vector(pages, page_count); | |
0eefd470 AE |
2556 | |
2557 | /* | |
2558 | * We want the transfer count to reflect the size of the | |
2559 | * original write request. There is no such thing as a | |
2560 | * successful short write, so if the request was successful | |
2561 | * we can just set it to the originally-requested length. | |
2562 | */ | |
2563 | if (!obj_request->result) | |
2564 | obj_request->xferred = obj_request->length; | |
2565 | ||
2566 | /* Finish up with the normal image object callback */ | |
2567 | ||
2568 | rbd_img_obj_callback(obj_request); | |
2569 | } | |
2570 | ||
3d7efd18 AE |
2571 | static void |
2572 | rbd_img_obj_parent_read_full_callback(struct rbd_img_request *img_request) | |
2573 | { | |
2574 | struct rbd_obj_request *orig_request; | |
0eefd470 AE |
2575 | struct ceph_osd_request *osd_req; |
2576 | struct ceph_osd_client *osdc; | |
2577 | struct rbd_device *rbd_dev; | |
3d7efd18 | 2578 | struct page **pages; |
d3246fb0 | 2579 | enum obj_operation_type op_type; |
ebda6408 | 2580 | u32 page_count; |
bbea1c1a | 2581 | int img_result; |
ebda6408 | 2582 | u64 parent_length; |
3d7efd18 AE |
2583 | |
2584 | rbd_assert(img_request_child_test(img_request)); | |
2585 | ||
2586 | /* First get what we need from the image request */ | |
2587 | ||
2588 | pages = img_request->copyup_pages; | |
2589 | rbd_assert(pages != NULL); | |
2590 | img_request->copyup_pages = NULL; | |
ebda6408 AE |
2591 | page_count = img_request->copyup_page_count; |
2592 | rbd_assert(page_count); | |
2593 | img_request->copyup_page_count = 0; | |
3d7efd18 AE |
2594 | |
2595 | orig_request = img_request->obj_request; | |
2596 | rbd_assert(orig_request != NULL); | |
b91f09f1 | 2597 | rbd_assert(obj_request_type_valid(orig_request->type)); |
bbea1c1a | 2598 | img_result = img_request->result; |
ebda6408 AE |
2599 | parent_length = img_request->length; |
2600 | rbd_assert(parent_length == img_request->xferred); | |
91c6febb | 2601 | rbd_img_request_put(img_request); |
3d7efd18 | 2602 | |
91c6febb AE |
2603 | rbd_assert(orig_request->img_request); |
2604 | rbd_dev = orig_request->img_request->rbd_dev; | |
0eefd470 | 2605 | rbd_assert(rbd_dev); |
0eefd470 | 2606 | |
bbea1c1a AE |
2607 | /* |
2608 | * If the overlap has become 0 (most likely because the | |
2609 | * image has been flattened) we need to free the pages | |
2610 | * and re-submit the original write request. | |
2611 | */ | |
2612 | if (!rbd_dev->parent_overlap) { | |
2613 | struct ceph_osd_client *osdc; | |
3d7efd18 | 2614 | |
bbea1c1a AE |
2615 | ceph_release_page_vector(pages, page_count); |
2616 | osdc = &rbd_dev->rbd_client->client->osdc; | |
2617 | img_result = rbd_obj_request_submit(osdc, orig_request); | |
2618 | if (!img_result) | |
2619 | return; | |
2620 | } | |
0eefd470 | 2621 | |
bbea1c1a | 2622 | if (img_result) |
0eefd470 | 2623 | goto out_err; |
0eefd470 | 2624 | |
8785b1d4 AE |
2625 | /* |
2626 | * The original osd request is of no use to use any more. | |
0ccd5926 | 2627 | * We need a new one that can hold the three ops in a copyup |
8785b1d4 AE |
2628 | * request. Allocate the new copyup osd request for the |
2629 | * original request, and release the old one. | |
2630 | */ | |
bbea1c1a | 2631 | img_result = -ENOMEM; |
0eefd470 AE |
2632 | osd_req = rbd_osd_req_create_copyup(orig_request); |
2633 | if (!osd_req) | |
2634 | goto out_err; | |
8785b1d4 | 2635 | rbd_osd_req_destroy(orig_request->osd_req); |
0eefd470 AE |
2636 | orig_request->osd_req = osd_req; |
2637 | orig_request->copyup_pages = pages; | |
ebda6408 | 2638 | orig_request->copyup_page_count = page_count; |
3d7efd18 | 2639 | |
0eefd470 | 2640 | /* Initialize the copyup op */ |
3d7efd18 | 2641 | |
0eefd470 | 2642 | osd_req_op_cls_init(osd_req, 0, CEPH_OSD_OP_CALL, "rbd", "copyup"); |
ebda6408 | 2643 | osd_req_op_cls_request_data_pages(osd_req, 0, pages, parent_length, 0, |
0eefd470 | 2644 | false, false); |
3d7efd18 | 2645 | |
d3246fb0 | 2646 | /* Add the other op(s) */ |
0eefd470 | 2647 | |
d3246fb0 JD |
2648 | op_type = rbd_img_request_op_type(orig_request->img_request); |
2649 | rbd_img_obj_request_fill(orig_request, osd_req, op_type, 1); | |
0eefd470 AE |
2650 | |
2651 | /* All set, send it off. */ | |
2652 | ||
2653 | orig_request->callback = rbd_img_obj_copyup_callback; | |
2654 | osdc = &rbd_dev->rbd_client->client->osdc; | |
bbea1c1a AE |
2655 | img_result = rbd_obj_request_submit(osdc, orig_request); |
2656 | if (!img_result) | |
0eefd470 AE |
2657 | return; |
2658 | out_err: | |
2659 | /* Record the error code and complete the request */ | |
2660 | ||
bbea1c1a | 2661 | orig_request->result = img_result; |
0eefd470 AE |
2662 | orig_request->xferred = 0; |
2663 | obj_request_done_set(orig_request); | |
2664 | rbd_obj_request_complete(orig_request); | |
3d7efd18 AE |
2665 | } |
2666 | ||
2667 | /* | |
2668 | * Read from the parent image the range of data that covers the | |
2669 | * entire target of the given object request. This is used for | |
2670 | * satisfying a layered image write request when the target of an | |
2671 | * object request from the image request does not exist. | |
2672 | * | |
2673 | * A page array big enough to hold the returned data is allocated | |
2674 | * and supplied to rbd_img_request_fill() as the "data descriptor." | |
2675 | * When the read completes, this page array will be transferred to | |
2676 | * the original object request for the copyup operation. | |
2677 | * | |
2678 | * If an error occurs, record it as the result of the original | |
2679 | * object request and mark it done so it gets completed. | |
2680 | */ | |
2681 | static int rbd_img_obj_parent_read_full(struct rbd_obj_request *obj_request) | |
2682 | { | |
2683 | struct rbd_img_request *img_request = NULL; | |
2684 | struct rbd_img_request *parent_request = NULL; | |
2685 | struct rbd_device *rbd_dev; | |
2686 | u64 img_offset; | |
2687 | u64 length; | |
2688 | struct page **pages = NULL; | |
2689 | u32 page_count; | |
2690 | int result; | |
2691 | ||
2692 | rbd_assert(obj_request_img_data_test(obj_request)); | |
b91f09f1 | 2693 | rbd_assert(obj_request_type_valid(obj_request->type)); |
3d7efd18 AE |
2694 | |
2695 | img_request = obj_request->img_request; | |
2696 | rbd_assert(img_request != NULL); | |
2697 | rbd_dev = img_request->rbd_dev; | |
2698 | rbd_assert(rbd_dev->parent != NULL); | |
2699 | ||
2700 | /* | |
2701 | * Determine the byte range covered by the object in the | |
2702 | * child image to which the original request was to be sent. | |
2703 | */ | |
2704 | img_offset = obj_request->img_offset - obj_request->offset; | |
2705 | length = (u64)1 << rbd_dev->header.obj_order; | |
2706 | ||
a9e8ba2c AE |
2707 | /* |
2708 | * There is no defined parent data beyond the parent | |
2709 | * overlap, so limit what we read at that boundary if | |
2710 | * necessary. | |
2711 | */ | |
2712 | if (img_offset + length > rbd_dev->parent_overlap) { | |
2713 | rbd_assert(img_offset < rbd_dev->parent_overlap); | |
2714 | length = rbd_dev->parent_overlap - img_offset; | |
2715 | } | |
2716 | ||
3d7efd18 AE |
2717 | /* |
2718 | * Allocate a page array big enough to receive the data read | |
2719 | * from the parent. | |
2720 | */ | |
2721 | page_count = (u32)calc_pages_for(0, length); | |
2722 | pages = ceph_alloc_page_vector(page_count, GFP_KERNEL); | |
2723 | if (IS_ERR(pages)) { | |
2724 | result = PTR_ERR(pages); | |
2725 | pages = NULL; | |
2726 | goto out_err; | |
2727 | } | |
2728 | ||
2729 | result = -ENOMEM; | |
e93f3152 AE |
2730 | parent_request = rbd_parent_request_create(obj_request, |
2731 | img_offset, length); | |
3d7efd18 AE |
2732 | if (!parent_request) |
2733 | goto out_err; | |
3d7efd18 AE |
2734 | |
2735 | result = rbd_img_request_fill(parent_request, OBJ_REQUEST_PAGES, pages); | |
2736 | if (result) | |
2737 | goto out_err; | |
2738 | parent_request->copyup_pages = pages; | |
ebda6408 | 2739 | parent_request->copyup_page_count = page_count; |
3d7efd18 AE |
2740 | |
2741 | parent_request->callback = rbd_img_obj_parent_read_full_callback; | |
2742 | result = rbd_img_request_submit(parent_request); | |
2743 | if (!result) | |
2744 | return 0; | |
2745 | ||
2746 | parent_request->copyup_pages = NULL; | |
ebda6408 | 2747 | parent_request->copyup_page_count = 0; |
3d7efd18 AE |
2748 | parent_request->obj_request = NULL; |
2749 | rbd_obj_request_put(obj_request); | |
2750 | out_err: | |
2751 | if (pages) | |
2752 | ceph_release_page_vector(pages, page_count); | |
2753 | if (parent_request) | |
2754 | rbd_img_request_put(parent_request); | |
2755 | obj_request->result = result; | |
2756 | obj_request->xferred = 0; | |
2757 | obj_request_done_set(obj_request); | |
2758 | ||
2759 | return result; | |
2760 | } | |
2761 | ||
c5b5ef6c AE |
2762 | static void rbd_img_obj_exists_callback(struct rbd_obj_request *obj_request) |
2763 | { | |
c5b5ef6c | 2764 | struct rbd_obj_request *orig_request; |
638f5abe | 2765 | struct rbd_device *rbd_dev; |
c5b5ef6c AE |
2766 | int result; |
2767 | ||
2768 | rbd_assert(!obj_request_img_data_test(obj_request)); | |
2769 | ||
2770 | /* | |
2771 | * All we need from the object request is the original | |
2772 | * request and the result of the STAT op. Grab those, then | |
2773 | * we're done with the request. | |
2774 | */ | |
2775 | orig_request = obj_request->obj_request; | |
2776 | obj_request->obj_request = NULL; | |
912c317d | 2777 | rbd_obj_request_put(orig_request); |
c5b5ef6c AE |
2778 | rbd_assert(orig_request); |
2779 | rbd_assert(orig_request->img_request); | |
2780 | ||
2781 | result = obj_request->result; | |
2782 | obj_request->result = 0; | |
2783 | ||
2784 | dout("%s: obj %p for obj %p result %d %llu/%llu\n", __func__, | |
2785 | obj_request, orig_request, result, | |
2786 | obj_request->xferred, obj_request->length); | |
2787 | rbd_obj_request_put(obj_request); | |
2788 | ||
638f5abe AE |
2789 | /* |
2790 | * If the overlap has become 0 (most likely because the | |
2791 | * image has been flattened) we need to free the pages | |
2792 | * and re-submit the original write request. | |
2793 | */ | |
2794 | rbd_dev = orig_request->img_request->rbd_dev; | |
2795 | if (!rbd_dev->parent_overlap) { | |
2796 | struct ceph_osd_client *osdc; | |
2797 | ||
638f5abe AE |
2798 | osdc = &rbd_dev->rbd_client->client->osdc; |
2799 | result = rbd_obj_request_submit(osdc, orig_request); | |
2800 | if (!result) | |
2801 | return; | |
2802 | } | |
c5b5ef6c AE |
2803 | |
2804 | /* | |
2805 | * Our only purpose here is to determine whether the object | |
2806 | * exists, and we don't want to treat the non-existence as | |
2807 | * an error. If something else comes back, transfer the | |
2808 | * error to the original request and complete it now. | |
2809 | */ | |
2810 | if (!result) { | |
2811 | obj_request_existence_set(orig_request, true); | |
2812 | } else if (result == -ENOENT) { | |
2813 | obj_request_existence_set(orig_request, false); | |
2814 | } else if (result) { | |
2815 | orig_request->result = result; | |
3d7efd18 | 2816 | goto out; |
c5b5ef6c AE |
2817 | } |
2818 | ||
2819 | /* | |
2820 | * Resubmit the original request now that we have recorded | |
2821 | * whether the target object exists. | |
2822 | */ | |
b454e36d | 2823 | orig_request->result = rbd_img_obj_request_submit(orig_request); |
3d7efd18 | 2824 | out: |
c5b5ef6c AE |
2825 | if (orig_request->result) |
2826 | rbd_obj_request_complete(orig_request); | |
c5b5ef6c AE |
2827 | } |
2828 | ||
2829 | static int rbd_img_obj_exists_submit(struct rbd_obj_request *obj_request) | |
2830 | { | |
2831 | struct rbd_obj_request *stat_request; | |
2832 | struct rbd_device *rbd_dev; | |
2833 | struct ceph_osd_client *osdc; | |
2834 | struct page **pages = NULL; | |
2835 | u32 page_count; | |
2836 | size_t size; | |
2837 | int ret; | |
2838 | ||
2839 | /* | |
2840 | * The response data for a STAT call consists of: | |
2841 | * le64 length; | |
2842 | * struct { | |
2843 | * le32 tv_sec; | |
2844 | * le32 tv_nsec; | |
2845 | * } mtime; | |
2846 | */ | |
2847 | size = sizeof (__le64) + sizeof (__le32) + sizeof (__le32); | |
2848 | page_count = (u32)calc_pages_for(0, size); | |
2849 | pages = ceph_alloc_page_vector(page_count, GFP_KERNEL); | |
2850 | if (IS_ERR(pages)) | |
2851 | return PTR_ERR(pages); | |
2852 | ||
2853 | ret = -ENOMEM; | |
2854 | stat_request = rbd_obj_request_create(obj_request->object_name, 0, 0, | |
2855 | OBJ_REQUEST_PAGES); | |
2856 | if (!stat_request) | |
2857 | goto out; | |
2858 | ||
2859 | rbd_obj_request_get(obj_request); | |
2860 | stat_request->obj_request = obj_request; | |
2861 | stat_request->pages = pages; | |
2862 | stat_request->page_count = page_count; | |
2863 | ||
2864 | rbd_assert(obj_request->img_request); | |
2865 | rbd_dev = obj_request->img_request->rbd_dev; | |
6d2940c8 | 2866 | stat_request->osd_req = rbd_osd_req_create(rbd_dev, OBJ_OP_READ, 1, |
deb236b3 | 2867 | stat_request); |
c5b5ef6c AE |
2868 | if (!stat_request->osd_req) |
2869 | goto out; | |
2870 | stat_request->callback = rbd_img_obj_exists_callback; | |
2871 | ||
144cba14 | 2872 | osd_req_op_init(stat_request->osd_req, 0, CEPH_OSD_OP_STAT, 0); |
c5b5ef6c AE |
2873 | osd_req_op_raw_data_in_pages(stat_request->osd_req, 0, pages, size, 0, |
2874 | false, false); | |
9d4df01f | 2875 | rbd_osd_req_format_read(stat_request); |
c5b5ef6c AE |
2876 | |
2877 | osdc = &rbd_dev->rbd_client->client->osdc; | |
2878 | ret = rbd_obj_request_submit(osdc, stat_request); | |
2879 | out: | |
2880 | if (ret) | |
2881 | rbd_obj_request_put(obj_request); | |
2882 | ||
2883 | return ret; | |
2884 | } | |
2885 | ||
70d045f6 | 2886 | static bool img_obj_request_simple(struct rbd_obj_request *obj_request) |
b454e36d AE |
2887 | { |
2888 | struct rbd_img_request *img_request; | |
a9e8ba2c | 2889 | struct rbd_device *rbd_dev; |
b454e36d AE |
2890 | |
2891 | rbd_assert(obj_request_img_data_test(obj_request)); | |
2892 | ||
2893 | img_request = obj_request->img_request; | |
2894 | rbd_assert(img_request); | |
a9e8ba2c | 2895 | rbd_dev = img_request->rbd_dev; |
b454e36d | 2896 | |
70d045f6 | 2897 | /* Reads */ |
1c220881 JD |
2898 | if (!img_request_write_test(img_request) && |
2899 | !img_request_discard_test(img_request)) | |
70d045f6 ID |
2900 | return true; |
2901 | ||
2902 | /* Non-layered writes */ | |
2903 | if (!img_request_layered_test(img_request)) | |
2904 | return true; | |
2905 | ||
b454e36d | 2906 | /* |
70d045f6 ID |
2907 | * Layered writes outside of the parent overlap range don't |
2908 | * share any data with the parent. | |
b454e36d | 2909 | */ |
70d045f6 ID |
2910 | if (!obj_request_overlaps_parent(obj_request)) |
2911 | return true; | |
b454e36d | 2912 | |
c622d226 GZ |
2913 | /* |
2914 | * Entire-object layered writes - we will overwrite whatever | |
2915 | * parent data there is anyway. | |
2916 | */ | |
2917 | if (!obj_request->offset && | |
2918 | obj_request->length == rbd_obj_bytes(&rbd_dev->header)) | |
2919 | return true; | |
2920 | ||
70d045f6 ID |
2921 | /* |
2922 | * If the object is known to already exist, its parent data has | |
2923 | * already been copied. | |
2924 | */ | |
2925 | if (obj_request_known_test(obj_request) && | |
2926 | obj_request_exists_test(obj_request)) | |
2927 | return true; | |
2928 | ||
2929 | return false; | |
2930 | } | |
2931 | ||
2932 | static int rbd_img_obj_request_submit(struct rbd_obj_request *obj_request) | |
2933 | { | |
2934 | if (img_obj_request_simple(obj_request)) { | |
b454e36d AE |
2935 | struct rbd_device *rbd_dev; |
2936 | struct ceph_osd_client *osdc; | |
2937 | ||
2938 | rbd_dev = obj_request->img_request->rbd_dev; | |
2939 | osdc = &rbd_dev->rbd_client->client->osdc; | |
2940 | ||
2941 | return rbd_obj_request_submit(osdc, obj_request); | |
2942 | } | |
2943 | ||
2944 | /* | |
3d7efd18 AE |
2945 | * It's a layered write. The target object might exist but |
2946 | * we may not know that yet. If we know it doesn't exist, | |
2947 | * start by reading the data for the full target object from | |
2948 | * the parent so we can use it for a copyup to the target. | |
b454e36d | 2949 | */ |
70d045f6 | 2950 | if (obj_request_known_test(obj_request)) |
3d7efd18 AE |
2951 | return rbd_img_obj_parent_read_full(obj_request); |
2952 | ||
2953 | /* We don't know whether the target exists. Go find out. */ | |
b454e36d AE |
2954 | |
2955 | return rbd_img_obj_exists_submit(obj_request); | |
2956 | } | |
2957 | ||
bf0d5f50 AE |
2958 | static int rbd_img_request_submit(struct rbd_img_request *img_request) |
2959 | { | |
bf0d5f50 | 2960 | struct rbd_obj_request *obj_request; |
46faeed4 | 2961 | struct rbd_obj_request *next_obj_request; |
bf0d5f50 | 2962 | |
37206ee5 | 2963 | dout("%s: img %p\n", __func__, img_request); |
46faeed4 | 2964 | for_each_obj_request_safe(img_request, obj_request, next_obj_request) { |
bf0d5f50 AE |
2965 | int ret; |
2966 | ||
b454e36d | 2967 | ret = rbd_img_obj_request_submit(obj_request); |
bf0d5f50 AE |
2968 | if (ret) |
2969 | return ret; | |
bf0d5f50 AE |
2970 | } |
2971 | ||
2972 | return 0; | |
2973 | } | |
8b3e1a56 AE |
2974 | |
2975 | static void rbd_img_parent_read_callback(struct rbd_img_request *img_request) | |
2976 | { | |
2977 | struct rbd_obj_request *obj_request; | |
a9e8ba2c AE |
2978 | struct rbd_device *rbd_dev; |
2979 | u64 obj_end; | |
02c74fba AE |
2980 | u64 img_xferred; |
2981 | int img_result; | |
8b3e1a56 AE |
2982 | |
2983 | rbd_assert(img_request_child_test(img_request)); | |
2984 | ||
02c74fba AE |
2985 | /* First get what we need from the image request and release it */ |
2986 | ||
8b3e1a56 | 2987 | obj_request = img_request->obj_request; |
02c74fba AE |
2988 | img_xferred = img_request->xferred; |
2989 | img_result = img_request->result; | |
2990 | rbd_img_request_put(img_request); | |
2991 | ||
2992 | /* | |
2993 | * If the overlap has become 0 (most likely because the | |
2994 | * image has been flattened) we need to re-submit the | |
2995 | * original request. | |
2996 | */ | |
a9e8ba2c AE |
2997 | rbd_assert(obj_request); |
2998 | rbd_assert(obj_request->img_request); | |
02c74fba AE |
2999 | rbd_dev = obj_request->img_request->rbd_dev; |
3000 | if (!rbd_dev->parent_overlap) { | |
3001 | struct ceph_osd_client *osdc; | |
3002 | ||
3003 | osdc = &rbd_dev->rbd_client->client->osdc; | |
3004 | img_result = rbd_obj_request_submit(osdc, obj_request); | |
3005 | if (!img_result) | |
3006 | return; | |
3007 | } | |
a9e8ba2c | 3008 | |
02c74fba | 3009 | obj_request->result = img_result; |
a9e8ba2c AE |
3010 | if (obj_request->result) |
3011 | goto out; | |
3012 | ||
3013 | /* | |
3014 | * We need to zero anything beyond the parent overlap | |
3015 | * boundary. Since rbd_img_obj_request_read_callback() | |
3016 | * will zero anything beyond the end of a short read, an | |
3017 | * easy way to do this is to pretend the data from the | |
3018 | * parent came up short--ending at the overlap boundary. | |
3019 | */ | |
3020 | rbd_assert(obj_request->img_offset < U64_MAX - obj_request->length); | |
3021 | obj_end = obj_request->img_offset + obj_request->length; | |
a9e8ba2c AE |
3022 | if (obj_end > rbd_dev->parent_overlap) { |
3023 | u64 xferred = 0; | |
3024 | ||
3025 | if (obj_request->img_offset < rbd_dev->parent_overlap) | |
3026 | xferred = rbd_dev->parent_overlap - | |
3027 | obj_request->img_offset; | |
8b3e1a56 | 3028 | |
02c74fba | 3029 | obj_request->xferred = min(img_xferred, xferred); |
a9e8ba2c | 3030 | } else { |
02c74fba | 3031 | obj_request->xferred = img_xferred; |
a9e8ba2c AE |
3032 | } |
3033 | out: | |
8b3e1a56 AE |
3034 | rbd_img_obj_request_read_callback(obj_request); |
3035 | rbd_obj_request_complete(obj_request); | |
3036 | } | |
3037 | ||
3038 | static void rbd_img_parent_read(struct rbd_obj_request *obj_request) | |
3039 | { | |
8b3e1a56 AE |
3040 | struct rbd_img_request *img_request; |
3041 | int result; | |
3042 | ||
3043 | rbd_assert(obj_request_img_data_test(obj_request)); | |
3044 | rbd_assert(obj_request->img_request != NULL); | |
3045 | rbd_assert(obj_request->result == (s32) -ENOENT); | |
5b2ab72d | 3046 | rbd_assert(obj_request_type_valid(obj_request->type)); |
8b3e1a56 | 3047 | |
8b3e1a56 | 3048 | /* rbd_read_finish(obj_request, obj_request->length); */ |
e93f3152 | 3049 | img_request = rbd_parent_request_create(obj_request, |
8b3e1a56 | 3050 | obj_request->img_offset, |
e93f3152 | 3051 | obj_request->length); |
8b3e1a56 AE |
3052 | result = -ENOMEM; |
3053 | if (!img_request) | |
3054 | goto out_err; | |
3055 | ||
5b2ab72d AE |
3056 | if (obj_request->type == OBJ_REQUEST_BIO) |
3057 | result = rbd_img_request_fill(img_request, OBJ_REQUEST_BIO, | |
3058 | obj_request->bio_list); | |
3059 | else | |
3060 | result = rbd_img_request_fill(img_request, OBJ_REQUEST_PAGES, | |
3061 | obj_request->pages); | |
8b3e1a56 AE |
3062 | if (result) |
3063 | goto out_err; | |
3064 | ||
3065 | img_request->callback = rbd_img_parent_read_callback; | |
3066 | result = rbd_img_request_submit(img_request); | |
3067 | if (result) | |
3068 | goto out_err; | |
3069 | ||
3070 | return; | |
3071 | out_err: | |
3072 | if (img_request) | |
3073 | rbd_img_request_put(img_request); | |
3074 | obj_request->result = result; | |
3075 | obj_request->xferred = 0; | |
3076 | obj_request_done_set(obj_request); | |
3077 | } | |
bf0d5f50 | 3078 | |
20e0af67 | 3079 | static int rbd_obj_notify_ack_sync(struct rbd_device *rbd_dev, u64 notify_id) |
b8d70035 AE |
3080 | { |
3081 | struct rbd_obj_request *obj_request; | |
2169238d | 3082 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
b8d70035 AE |
3083 | int ret; |
3084 | ||
3085 | obj_request = rbd_obj_request_create(rbd_dev->header_name, 0, 0, | |
3086 | OBJ_REQUEST_NODATA); | |
3087 | if (!obj_request) | |
3088 | return -ENOMEM; | |
3089 | ||
3090 | ret = -ENOMEM; | |
6d2940c8 | 3091 | obj_request->osd_req = rbd_osd_req_create(rbd_dev, OBJ_OP_READ, 1, |
deb236b3 | 3092 | obj_request); |
b8d70035 AE |
3093 | if (!obj_request->osd_req) |
3094 | goto out; | |
3095 | ||
c99d2d4a | 3096 | osd_req_op_watch_init(obj_request->osd_req, 0, CEPH_OSD_OP_NOTIFY_ACK, |
cc4a38bd | 3097 | notify_id, 0, 0); |
9d4df01f | 3098 | rbd_osd_req_format_read(obj_request); |
430c28c3 | 3099 | |
b8d70035 | 3100 | ret = rbd_obj_request_submit(osdc, obj_request); |
cf81b60e | 3101 | if (ret) |
20e0af67 JD |
3102 | goto out; |
3103 | ret = rbd_obj_request_wait(obj_request); | |
3104 | out: | |
3105 | rbd_obj_request_put(obj_request); | |
b8d70035 AE |
3106 | |
3107 | return ret; | |
3108 | } | |
3109 | ||
3110 | static void rbd_watch_cb(u64 ver, u64 notify_id, u8 opcode, void *data) | |
3111 | { | |
3112 | struct rbd_device *rbd_dev = (struct rbd_device *)data; | |
e627db08 | 3113 | int ret; |
b8d70035 AE |
3114 | |
3115 | if (!rbd_dev) | |
3116 | return; | |
3117 | ||
37206ee5 | 3118 | dout("%s: \"%s\" notify_id %llu opcode %u\n", __func__, |
cc4a38bd AE |
3119 | rbd_dev->header_name, (unsigned long long)notify_id, |
3120 | (unsigned int)opcode); | |
52bb1f9b ID |
3121 | |
3122 | /* | |
3123 | * Until adequate refresh error handling is in place, there is | |
3124 | * not much we can do here, except warn. | |
3125 | * | |
3126 | * See http://tracker.ceph.com/issues/5040 | |
3127 | */ | |
e627db08 AE |
3128 | ret = rbd_dev_refresh(rbd_dev); |
3129 | if (ret) | |
9584d508 | 3130 | rbd_warn(rbd_dev, "refresh failed: %d", ret); |
b8d70035 | 3131 | |
52bb1f9b ID |
3132 | ret = rbd_obj_notify_ack_sync(rbd_dev, notify_id); |
3133 | if (ret) | |
9584d508 | 3134 | rbd_warn(rbd_dev, "notify_ack ret %d", ret); |
b8d70035 AE |
3135 | } |
3136 | ||
bb040aa0 ID |
3137 | /* |
3138 | * Send a (un)watch request and wait for the ack. Return a request | |
3139 | * with a ref held on success or error. | |
3140 | */ | |
3141 | static struct rbd_obj_request *rbd_obj_watch_request_helper( | |
3142 | struct rbd_device *rbd_dev, | |
3143 | bool watch) | |
3144 | { | |
3145 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; | |
2894e1d7 | 3146 | struct ceph_options *opts = osdc->client->options; |
bb040aa0 ID |
3147 | struct rbd_obj_request *obj_request; |
3148 | int ret; | |
3149 | ||
3150 | obj_request = rbd_obj_request_create(rbd_dev->header_name, 0, 0, | |
3151 | OBJ_REQUEST_NODATA); | |
3152 | if (!obj_request) | |
3153 | return ERR_PTR(-ENOMEM); | |
3154 | ||
6d2940c8 | 3155 | obj_request->osd_req = rbd_osd_req_create(rbd_dev, OBJ_OP_WRITE, 1, |
bb040aa0 ID |
3156 | obj_request); |
3157 | if (!obj_request->osd_req) { | |
3158 | ret = -ENOMEM; | |
3159 | goto out; | |
3160 | } | |
3161 | ||
3162 | osd_req_op_watch_init(obj_request->osd_req, 0, CEPH_OSD_OP_WATCH, | |
3163 | rbd_dev->watch_event->cookie, 0, watch); | |
3164 | rbd_osd_req_format_write(obj_request); | |
3165 | ||
3166 | if (watch) | |
3167 | ceph_osdc_set_request_linger(osdc, obj_request->osd_req); | |
3168 | ||
3169 | ret = rbd_obj_request_submit(osdc, obj_request); | |
3170 | if (ret) | |
3171 | goto out; | |
3172 | ||
2894e1d7 | 3173 | ret = rbd_obj_request_wait_timeout(obj_request, opts->mount_timeout); |
bb040aa0 ID |
3174 | if (ret) |
3175 | goto out; | |
3176 | ||
3177 | ret = obj_request->result; | |
3178 | if (ret) { | |
3179 | if (watch) | |
3180 | rbd_obj_request_end(obj_request); | |
3181 | goto out; | |
3182 | } | |
3183 | ||
3184 | return obj_request; | |
3185 | ||
3186 | out: | |
3187 | rbd_obj_request_put(obj_request); | |
3188 | return ERR_PTR(ret); | |
3189 | } | |
3190 | ||
9969ebc5 | 3191 | /* |
b30a01f2 | 3192 | * Initiate a watch request, synchronously. |
9969ebc5 | 3193 | */ |
b30a01f2 | 3194 | static int rbd_dev_header_watch_sync(struct rbd_device *rbd_dev) |
9969ebc5 AE |
3195 | { |
3196 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; | |
3197 | struct rbd_obj_request *obj_request; | |
9969ebc5 AE |
3198 | int ret; |
3199 | ||
b30a01f2 ID |
3200 | rbd_assert(!rbd_dev->watch_event); |
3201 | rbd_assert(!rbd_dev->watch_request); | |
9969ebc5 | 3202 | |
b30a01f2 ID |
3203 | ret = ceph_osdc_create_event(osdc, rbd_watch_cb, rbd_dev, |
3204 | &rbd_dev->watch_event); | |
3205 | if (ret < 0) | |
3206 | return ret; | |
3207 | ||
76756a51 ID |
3208 | obj_request = rbd_obj_watch_request_helper(rbd_dev, true); |
3209 | if (IS_ERR(obj_request)) { | |
3210 | ceph_osdc_cancel_event(rbd_dev->watch_event); | |
3211 | rbd_dev->watch_event = NULL; | |
3212 | return PTR_ERR(obj_request); | |
b30a01f2 | 3213 | } |
9969ebc5 | 3214 | |
8eb87565 AE |
3215 | /* |
3216 | * A watch request is set to linger, so the underlying osd | |
3217 | * request won't go away until we unregister it. We retain | |
3218 | * a pointer to the object request during that time (in | |
76756a51 ID |
3219 | * rbd_dev->watch_request), so we'll keep a reference to it. |
3220 | * We'll drop that reference after we've unregistered it in | |
3221 | * rbd_dev_header_unwatch_sync(). | |
8eb87565 | 3222 | */ |
b30a01f2 | 3223 | rbd_dev->watch_request = obj_request; |
8eb87565 | 3224 | |
b30a01f2 | 3225 | return 0; |
b30a01f2 ID |
3226 | } |
3227 | ||
3228 | /* | |
3229 | * Tear down a watch request, synchronously. | |
3230 | */ | |
76756a51 | 3231 | static void rbd_dev_header_unwatch_sync(struct rbd_device *rbd_dev) |
b30a01f2 | 3232 | { |
b30a01f2 | 3233 | struct rbd_obj_request *obj_request; |
b30a01f2 ID |
3234 | |
3235 | rbd_assert(rbd_dev->watch_event); | |
3236 | rbd_assert(rbd_dev->watch_request); | |
3237 | ||
76756a51 | 3238 | rbd_obj_request_end(rbd_dev->watch_request); |
8eb87565 AE |
3239 | rbd_obj_request_put(rbd_dev->watch_request); |
3240 | rbd_dev->watch_request = NULL; | |
b30a01f2 | 3241 | |
76756a51 ID |
3242 | obj_request = rbd_obj_watch_request_helper(rbd_dev, false); |
3243 | if (!IS_ERR(obj_request)) | |
3244 | rbd_obj_request_put(obj_request); | |
3245 | else | |
3246 | rbd_warn(rbd_dev, "unable to tear down watch request (%ld)", | |
3247 | PTR_ERR(obj_request)); | |
3248 | ||
9969ebc5 AE |
3249 | ceph_osdc_cancel_event(rbd_dev->watch_event); |
3250 | rbd_dev->watch_event = NULL; | |
fca27065 ID |
3251 | } |
3252 | ||
36be9a76 | 3253 | /* |
f40eb349 AE |
3254 | * Synchronous osd object method call. Returns the number of bytes |
3255 | * returned in the outbound buffer, or a negative error code. | |
36be9a76 AE |
3256 | */ |
3257 | static int rbd_obj_method_sync(struct rbd_device *rbd_dev, | |
3258 | const char *object_name, | |
3259 | const char *class_name, | |
3260 | const char *method_name, | |
4157976b | 3261 | const void *outbound, |
36be9a76 | 3262 | size_t outbound_size, |
4157976b | 3263 | void *inbound, |
e2a58ee5 | 3264 | size_t inbound_size) |
36be9a76 | 3265 | { |
2169238d | 3266 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
36be9a76 | 3267 | struct rbd_obj_request *obj_request; |
36be9a76 AE |
3268 | struct page **pages; |
3269 | u32 page_count; | |
3270 | int ret; | |
3271 | ||
3272 | /* | |
6010a451 AE |
3273 | * Method calls are ultimately read operations. The result |
3274 | * should placed into the inbound buffer provided. They | |
3275 | * also supply outbound data--parameters for the object | |
3276 | * method. Currently if this is present it will be a | |
3277 | * snapshot id. | |
36be9a76 | 3278 | */ |
57385b51 | 3279 | page_count = (u32)calc_pages_for(0, inbound_size); |
36be9a76 AE |
3280 | pages = ceph_alloc_page_vector(page_count, GFP_KERNEL); |
3281 | if (IS_ERR(pages)) | |
3282 | return PTR_ERR(pages); | |
3283 | ||
3284 | ret = -ENOMEM; | |
6010a451 | 3285 | obj_request = rbd_obj_request_create(object_name, 0, inbound_size, |
36be9a76 AE |
3286 | OBJ_REQUEST_PAGES); |
3287 | if (!obj_request) | |
3288 | goto out; | |
3289 | ||
3290 | obj_request->pages = pages; | |
3291 | obj_request->page_count = page_count; | |
3292 | ||
6d2940c8 | 3293 | obj_request->osd_req = rbd_osd_req_create(rbd_dev, OBJ_OP_READ, 1, |
deb236b3 | 3294 | obj_request); |
36be9a76 AE |
3295 | if (!obj_request->osd_req) |
3296 | goto out; | |
3297 | ||
c99d2d4a | 3298 | osd_req_op_cls_init(obj_request->osd_req, 0, CEPH_OSD_OP_CALL, |
04017e29 AE |
3299 | class_name, method_name); |
3300 | if (outbound_size) { | |
3301 | struct ceph_pagelist *pagelist; | |
3302 | ||
3303 | pagelist = kmalloc(sizeof (*pagelist), GFP_NOFS); | |
3304 | if (!pagelist) | |
3305 | goto out; | |
3306 | ||
3307 | ceph_pagelist_init(pagelist); | |
3308 | ceph_pagelist_append(pagelist, outbound, outbound_size); | |
3309 | osd_req_op_cls_request_data_pagelist(obj_request->osd_req, 0, | |
3310 | pagelist); | |
3311 | } | |
a4ce40a9 AE |
3312 | osd_req_op_cls_response_data_pages(obj_request->osd_req, 0, |
3313 | obj_request->pages, inbound_size, | |
44cd188d | 3314 | 0, false, false); |
9d4df01f | 3315 | rbd_osd_req_format_read(obj_request); |
430c28c3 | 3316 | |
36be9a76 AE |
3317 | ret = rbd_obj_request_submit(osdc, obj_request); |
3318 | if (ret) | |
3319 | goto out; | |
3320 | ret = rbd_obj_request_wait(obj_request); | |
3321 | if (ret) | |
3322 | goto out; | |
3323 | ||
3324 | ret = obj_request->result; | |
3325 | if (ret < 0) | |
3326 | goto out; | |
57385b51 AE |
3327 | |
3328 | rbd_assert(obj_request->xferred < (u64)INT_MAX); | |
3329 | ret = (int)obj_request->xferred; | |
903bb32e | 3330 | ceph_copy_from_page_vector(pages, inbound, 0, obj_request->xferred); |
36be9a76 AE |
3331 | out: |
3332 | if (obj_request) | |
3333 | rbd_obj_request_put(obj_request); | |
3334 | else | |
3335 | ceph_release_page_vector(pages, page_count); | |
3336 | ||
3337 | return ret; | |
3338 | } | |
3339 | ||
7ad18afa | 3340 | static void rbd_queue_workfn(struct work_struct *work) |
bf0d5f50 | 3341 | { |
7ad18afa CH |
3342 | struct request *rq = blk_mq_rq_from_pdu(work); |
3343 | struct rbd_device *rbd_dev = rq->q->queuedata; | |
bc1ecc65 | 3344 | struct rbd_img_request *img_request; |
4e752f0a | 3345 | struct ceph_snap_context *snapc = NULL; |
bc1ecc65 ID |
3346 | u64 offset = (u64)blk_rq_pos(rq) << SECTOR_SHIFT; |
3347 | u64 length = blk_rq_bytes(rq); | |
6d2940c8 | 3348 | enum obj_operation_type op_type; |
4e752f0a | 3349 | u64 mapping_size; |
bf0d5f50 AE |
3350 | int result; |
3351 | ||
7ad18afa CH |
3352 | if (rq->cmd_type != REQ_TYPE_FS) { |
3353 | dout("%s: non-fs request type %d\n", __func__, | |
3354 | (int) rq->cmd_type); | |
3355 | result = -EIO; | |
3356 | goto err; | |
3357 | } | |
3358 | ||
90e98c52 GZ |
3359 | if (rq->cmd_flags & REQ_DISCARD) |
3360 | op_type = OBJ_OP_DISCARD; | |
3361 | else if (rq->cmd_flags & REQ_WRITE) | |
6d2940c8 GZ |
3362 | op_type = OBJ_OP_WRITE; |
3363 | else | |
3364 | op_type = OBJ_OP_READ; | |
3365 | ||
bc1ecc65 | 3366 | /* Ignore/skip any zero-length requests */ |
bf0d5f50 | 3367 | |
bc1ecc65 ID |
3368 | if (!length) { |
3369 | dout("%s: zero-length request\n", __func__); | |
3370 | result = 0; | |
3371 | goto err_rq; | |
3372 | } | |
bf0d5f50 | 3373 | |
6d2940c8 | 3374 | /* Only reads are allowed to a read-only device */ |
bc1ecc65 | 3375 | |
6d2940c8 | 3376 | if (op_type != OBJ_OP_READ) { |
bc1ecc65 ID |
3377 | if (rbd_dev->mapping.read_only) { |
3378 | result = -EROFS; | |
3379 | goto err_rq; | |
4dda41d3 | 3380 | } |
bc1ecc65 ID |
3381 | rbd_assert(rbd_dev->spec->snap_id == CEPH_NOSNAP); |
3382 | } | |
4dda41d3 | 3383 | |
bc1ecc65 ID |
3384 | /* |
3385 | * Quit early if the mapped snapshot no longer exists. It's | |
3386 | * still possible the snapshot will have disappeared by the | |
3387 | * time our request arrives at the osd, but there's no sense in | |
3388 | * sending it if we already know. | |
3389 | */ | |
3390 | if (!test_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags)) { | |
3391 | dout("request for non-existent snapshot"); | |
3392 | rbd_assert(rbd_dev->spec->snap_id != CEPH_NOSNAP); | |
3393 | result = -ENXIO; | |
3394 | goto err_rq; | |
3395 | } | |
4dda41d3 | 3396 | |
bc1ecc65 ID |
3397 | if (offset && length > U64_MAX - offset + 1) { |
3398 | rbd_warn(rbd_dev, "bad request range (%llu~%llu)", offset, | |
3399 | length); | |
3400 | result = -EINVAL; | |
3401 | goto err_rq; /* Shouldn't happen */ | |
3402 | } | |
4dda41d3 | 3403 | |
7ad18afa CH |
3404 | blk_mq_start_request(rq); |
3405 | ||
4e752f0a JD |
3406 | down_read(&rbd_dev->header_rwsem); |
3407 | mapping_size = rbd_dev->mapping.size; | |
6d2940c8 | 3408 | if (op_type != OBJ_OP_READ) { |
4e752f0a JD |
3409 | snapc = rbd_dev->header.snapc; |
3410 | ceph_get_snap_context(snapc); | |
3411 | } | |
3412 | up_read(&rbd_dev->header_rwsem); | |
3413 | ||
3414 | if (offset + length > mapping_size) { | |
bc1ecc65 | 3415 | rbd_warn(rbd_dev, "beyond EOD (%llu~%llu > %llu)", offset, |
4e752f0a | 3416 | length, mapping_size); |
bc1ecc65 ID |
3417 | result = -EIO; |
3418 | goto err_rq; | |
3419 | } | |
bf0d5f50 | 3420 | |
6d2940c8 | 3421 | img_request = rbd_img_request_create(rbd_dev, offset, length, op_type, |
4e752f0a | 3422 | snapc); |
bc1ecc65 ID |
3423 | if (!img_request) { |
3424 | result = -ENOMEM; | |
3425 | goto err_rq; | |
3426 | } | |
3427 | img_request->rq = rq; | |
bf0d5f50 | 3428 | |
90e98c52 GZ |
3429 | if (op_type == OBJ_OP_DISCARD) |
3430 | result = rbd_img_request_fill(img_request, OBJ_REQUEST_NODATA, | |
3431 | NULL); | |
3432 | else | |
3433 | result = rbd_img_request_fill(img_request, OBJ_REQUEST_BIO, | |
3434 | rq->bio); | |
bc1ecc65 ID |
3435 | if (result) |
3436 | goto err_img_request; | |
bf0d5f50 | 3437 | |
bc1ecc65 ID |
3438 | result = rbd_img_request_submit(img_request); |
3439 | if (result) | |
3440 | goto err_img_request; | |
bf0d5f50 | 3441 | |
bc1ecc65 | 3442 | return; |
bf0d5f50 | 3443 | |
bc1ecc65 ID |
3444 | err_img_request: |
3445 | rbd_img_request_put(img_request); | |
3446 | err_rq: | |
3447 | if (result) | |
3448 | rbd_warn(rbd_dev, "%s %llx at %llx result %d", | |
6d2940c8 | 3449 | obj_op_name(op_type), length, offset, result); |
e96a650a | 3450 | ceph_put_snap_context(snapc); |
7ad18afa CH |
3451 | err: |
3452 | blk_mq_end_request(rq, result); | |
bc1ecc65 | 3453 | } |
bf0d5f50 | 3454 | |
7ad18afa CH |
3455 | static int rbd_queue_rq(struct blk_mq_hw_ctx *hctx, |
3456 | const struct blk_mq_queue_data *bd) | |
bc1ecc65 | 3457 | { |
7ad18afa CH |
3458 | struct request *rq = bd->rq; |
3459 | struct work_struct *work = blk_mq_rq_to_pdu(rq); | |
bf0d5f50 | 3460 | |
7ad18afa CH |
3461 | queue_work(rbd_wq, work); |
3462 | return BLK_MQ_RQ_QUEUE_OK; | |
bf0d5f50 AE |
3463 | } |
3464 | ||
602adf40 YS |
3465 | /* |
3466 | * a queue callback. Makes sure that we don't create a bio that spans across | |
3467 | * multiple osd objects. One exception would be with a single page bios, | |
f7760dad | 3468 | * which we handle later at bio_chain_clone_range() |
602adf40 YS |
3469 | */ |
3470 | static int rbd_merge_bvec(struct request_queue *q, struct bvec_merge_data *bmd, | |
3471 | struct bio_vec *bvec) | |
3472 | { | |
3473 | struct rbd_device *rbd_dev = q->queuedata; | |
e5cfeed2 AE |
3474 | sector_t sector_offset; |
3475 | sector_t sectors_per_obj; | |
3476 | sector_t obj_sector_offset; | |
3477 | int ret; | |
3478 | ||
3479 | /* | |
3480 | * Find how far into its rbd object the partition-relative | |
3481 | * bio start sector is to offset relative to the enclosing | |
3482 | * device. | |
3483 | */ | |
3484 | sector_offset = get_start_sect(bmd->bi_bdev) + bmd->bi_sector; | |
3485 | sectors_per_obj = 1 << (rbd_dev->header.obj_order - SECTOR_SHIFT); | |
3486 | obj_sector_offset = sector_offset & (sectors_per_obj - 1); | |
3487 | ||
3488 | /* | |
3489 | * Compute the number of bytes from that offset to the end | |
3490 | * of the object. Account for what's already used by the bio. | |
3491 | */ | |
3492 | ret = (int) (sectors_per_obj - obj_sector_offset) << SECTOR_SHIFT; | |
3493 | if (ret > bmd->bi_size) | |
3494 | ret -= bmd->bi_size; | |
3495 | else | |
3496 | ret = 0; | |
3497 | ||
3498 | /* | |
3499 | * Don't send back more than was asked for. And if the bio | |
3500 | * was empty, let the whole thing through because: "Note | |
3501 | * that a block device *must* allow a single page to be | |
3502 | * added to an empty bio." | |
3503 | */ | |
3504 | rbd_assert(bvec->bv_len <= PAGE_SIZE); | |
3505 | if (ret > (int) bvec->bv_len || !bmd->bi_size) | |
3506 | ret = (int) bvec->bv_len; | |
3507 | ||
3508 | return ret; | |
602adf40 YS |
3509 | } |
3510 | ||
3511 | static void rbd_free_disk(struct rbd_device *rbd_dev) | |
3512 | { | |
3513 | struct gendisk *disk = rbd_dev->disk; | |
3514 | ||
3515 | if (!disk) | |
3516 | return; | |
3517 | ||
a0cab924 AE |
3518 | rbd_dev->disk = NULL; |
3519 | if (disk->flags & GENHD_FL_UP) { | |
602adf40 | 3520 | del_gendisk(disk); |
a0cab924 AE |
3521 | if (disk->queue) |
3522 | blk_cleanup_queue(disk->queue); | |
7ad18afa | 3523 | blk_mq_free_tag_set(&rbd_dev->tag_set); |
a0cab924 | 3524 | } |
602adf40 YS |
3525 | put_disk(disk); |
3526 | } | |
3527 | ||
788e2df3 AE |
3528 | static int rbd_obj_read_sync(struct rbd_device *rbd_dev, |
3529 | const char *object_name, | |
7097f8df | 3530 | u64 offset, u64 length, void *buf) |
788e2df3 AE |
3531 | |
3532 | { | |
2169238d | 3533 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
788e2df3 | 3534 | struct rbd_obj_request *obj_request; |
788e2df3 AE |
3535 | struct page **pages = NULL; |
3536 | u32 page_count; | |
1ceae7ef | 3537 | size_t size; |
788e2df3 AE |
3538 | int ret; |
3539 | ||
3540 | page_count = (u32) calc_pages_for(offset, length); | |
3541 | pages = ceph_alloc_page_vector(page_count, GFP_KERNEL); | |
3542 | if (IS_ERR(pages)) | |
a8d42056 | 3543 | return PTR_ERR(pages); |
788e2df3 AE |
3544 | |
3545 | ret = -ENOMEM; | |
3546 | obj_request = rbd_obj_request_create(object_name, offset, length, | |
36be9a76 | 3547 | OBJ_REQUEST_PAGES); |
788e2df3 AE |
3548 | if (!obj_request) |
3549 | goto out; | |
3550 | ||
3551 | obj_request->pages = pages; | |
3552 | obj_request->page_count = page_count; | |
3553 | ||
6d2940c8 | 3554 | obj_request->osd_req = rbd_osd_req_create(rbd_dev, OBJ_OP_READ, 1, |
deb236b3 | 3555 | obj_request); |
788e2df3 AE |
3556 | if (!obj_request->osd_req) |
3557 | goto out; | |
3558 | ||
c99d2d4a AE |
3559 | osd_req_op_extent_init(obj_request->osd_req, 0, CEPH_OSD_OP_READ, |
3560 | offset, length, 0, 0); | |
406e2c9f | 3561 | osd_req_op_extent_osd_data_pages(obj_request->osd_req, 0, |
a4ce40a9 | 3562 | obj_request->pages, |
44cd188d AE |
3563 | obj_request->length, |
3564 | obj_request->offset & ~PAGE_MASK, | |
3565 | false, false); | |
9d4df01f | 3566 | rbd_osd_req_format_read(obj_request); |
430c28c3 | 3567 | |
788e2df3 AE |
3568 | ret = rbd_obj_request_submit(osdc, obj_request); |
3569 | if (ret) | |
3570 | goto out; | |
3571 | ret = rbd_obj_request_wait(obj_request); | |
3572 | if (ret) | |
3573 | goto out; | |
3574 | ||
3575 | ret = obj_request->result; | |
3576 | if (ret < 0) | |
3577 | goto out; | |
1ceae7ef AE |
3578 | |
3579 | rbd_assert(obj_request->xferred <= (u64) SIZE_MAX); | |
3580 | size = (size_t) obj_request->xferred; | |
903bb32e | 3581 | ceph_copy_from_page_vector(pages, buf, 0, size); |
7097f8df AE |
3582 | rbd_assert(size <= (size_t)INT_MAX); |
3583 | ret = (int)size; | |
788e2df3 AE |
3584 | out: |
3585 | if (obj_request) | |
3586 | rbd_obj_request_put(obj_request); | |
3587 | else | |
3588 | ceph_release_page_vector(pages, page_count); | |
3589 | ||
3590 | return ret; | |
3591 | } | |
3592 | ||
602adf40 | 3593 | /* |
662518b1 AE |
3594 | * Read the complete header for the given rbd device. On successful |
3595 | * return, the rbd_dev->header field will contain up-to-date | |
3596 | * information about the image. | |
602adf40 | 3597 | */ |
99a41ebc | 3598 | static int rbd_dev_v1_header_info(struct rbd_device *rbd_dev) |
602adf40 | 3599 | { |
4156d998 | 3600 | struct rbd_image_header_ondisk *ondisk = NULL; |
50f7c4c9 | 3601 | u32 snap_count = 0; |
4156d998 AE |
3602 | u64 names_size = 0; |
3603 | u32 want_count; | |
3604 | int ret; | |
602adf40 | 3605 | |
00f1f36f | 3606 | /* |
4156d998 AE |
3607 | * The complete header will include an array of its 64-bit |
3608 | * snapshot ids, followed by the names of those snapshots as | |
3609 | * a contiguous block of NUL-terminated strings. Note that | |
3610 | * the number of snapshots could change by the time we read | |
3611 | * it in, in which case we re-read it. | |
00f1f36f | 3612 | */ |
4156d998 AE |
3613 | do { |
3614 | size_t size; | |
3615 | ||
3616 | kfree(ondisk); | |
3617 | ||
3618 | size = sizeof (*ondisk); | |
3619 | size += snap_count * sizeof (struct rbd_image_snap_ondisk); | |
3620 | size += names_size; | |
3621 | ondisk = kmalloc(size, GFP_KERNEL); | |
3622 | if (!ondisk) | |
662518b1 | 3623 | return -ENOMEM; |
4156d998 | 3624 | |
788e2df3 | 3625 | ret = rbd_obj_read_sync(rbd_dev, rbd_dev->header_name, |
7097f8df | 3626 | 0, size, ondisk); |
4156d998 | 3627 | if (ret < 0) |
662518b1 | 3628 | goto out; |
c0cd10db | 3629 | if ((size_t)ret < size) { |
4156d998 | 3630 | ret = -ENXIO; |
06ecc6cb AE |
3631 | rbd_warn(rbd_dev, "short header read (want %zd got %d)", |
3632 | size, ret); | |
662518b1 | 3633 | goto out; |
4156d998 AE |
3634 | } |
3635 | if (!rbd_dev_ondisk_valid(ondisk)) { | |
3636 | ret = -ENXIO; | |
06ecc6cb | 3637 | rbd_warn(rbd_dev, "invalid header"); |
662518b1 | 3638 | goto out; |
81e759fb | 3639 | } |
602adf40 | 3640 | |
4156d998 AE |
3641 | names_size = le64_to_cpu(ondisk->snap_names_len); |
3642 | want_count = snap_count; | |
3643 | snap_count = le32_to_cpu(ondisk->snap_count); | |
3644 | } while (snap_count != want_count); | |
00f1f36f | 3645 | |
662518b1 AE |
3646 | ret = rbd_header_from_disk(rbd_dev, ondisk); |
3647 | out: | |
4156d998 AE |
3648 | kfree(ondisk); |
3649 | ||
3650 | return ret; | |
602adf40 YS |
3651 | } |
3652 | ||
15228ede AE |
3653 | /* |
3654 | * Clear the rbd device's EXISTS flag if the snapshot it's mapped to | |
3655 | * has disappeared from the (just updated) snapshot context. | |
3656 | */ | |
3657 | static void rbd_exists_validate(struct rbd_device *rbd_dev) | |
3658 | { | |
3659 | u64 snap_id; | |
3660 | ||
3661 | if (!test_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags)) | |
3662 | return; | |
3663 | ||
3664 | snap_id = rbd_dev->spec->snap_id; | |
3665 | if (snap_id == CEPH_NOSNAP) | |
3666 | return; | |
3667 | ||
3668 | if (rbd_dev_snap_index(rbd_dev, snap_id) == BAD_SNAP_INDEX) | |
3669 | clear_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags); | |
3670 | } | |
3671 | ||
9875201e JD |
3672 | static void rbd_dev_update_size(struct rbd_device *rbd_dev) |
3673 | { | |
3674 | sector_t size; | |
3675 | bool removing; | |
3676 | ||
3677 | /* | |
3678 | * Don't hold the lock while doing disk operations, | |
3679 | * or lock ordering will conflict with the bdev mutex via: | |
3680 | * rbd_add() -> blkdev_get() -> rbd_open() | |
3681 | */ | |
3682 | spin_lock_irq(&rbd_dev->lock); | |
3683 | removing = test_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags); | |
3684 | spin_unlock_irq(&rbd_dev->lock); | |
3685 | /* | |
3686 | * If the device is being removed, rbd_dev->disk has | |
3687 | * been destroyed, so don't try to update its size | |
3688 | */ | |
3689 | if (!removing) { | |
3690 | size = (sector_t)rbd_dev->mapping.size / SECTOR_SIZE; | |
3691 | dout("setting size to %llu sectors", (unsigned long long)size); | |
3692 | set_capacity(rbd_dev->disk, size); | |
3693 | revalidate_disk(rbd_dev->disk); | |
3694 | } | |
3695 | } | |
3696 | ||
cc4a38bd | 3697 | static int rbd_dev_refresh(struct rbd_device *rbd_dev) |
1fe5e993 | 3698 | { |
e627db08 | 3699 | u64 mapping_size; |
1fe5e993 AE |
3700 | int ret; |
3701 | ||
cfbf6377 | 3702 | down_write(&rbd_dev->header_rwsem); |
3b5cf2a2 | 3703 | mapping_size = rbd_dev->mapping.size; |
a720ae09 ID |
3704 | |
3705 | ret = rbd_dev_header_info(rbd_dev); | |
52bb1f9b | 3706 | if (ret) |
73e39e4d | 3707 | goto out; |
15228ede | 3708 | |
e8f59b59 ID |
3709 | /* |
3710 | * If there is a parent, see if it has disappeared due to the | |
3711 | * mapped image getting flattened. | |
3712 | */ | |
3713 | if (rbd_dev->parent) { | |
3714 | ret = rbd_dev_v2_parent_info(rbd_dev); | |
3715 | if (ret) | |
73e39e4d | 3716 | goto out; |
e8f59b59 ID |
3717 | } |
3718 | ||
5ff1108c | 3719 | if (rbd_dev->spec->snap_id == CEPH_NOSNAP) { |
73e39e4d | 3720 | rbd_dev->mapping.size = rbd_dev->header.image_size; |
5ff1108c ID |
3721 | } else { |
3722 | /* validate mapped snapshot's EXISTS flag */ | |
3723 | rbd_exists_validate(rbd_dev); | |
3724 | } | |
15228ede | 3725 | |
73e39e4d | 3726 | out: |
cfbf6377 | 3727 | up_write(&rbd_dev->header_rwsem); |
73e39e4d | 3728 | if (!ret && mapping_size != rbd_dev->mapping.size) |
9875201e | 3729 | rbd_dev_update_size(rbd_dev); |
1fe5e993 | 3730 | |
73e39e4d | 3731 | return ret; |
1fe5e993 AE |
3732 | } |
3733 | ||
7ad18afa CH |
3734 | static int rbd_init_request(void *data, struct request *rq, |
3735 | unsigned int hctx_idx, unsigned int request_idx, | |
3736 | unsigned int numa_node) | |
3737 | { | |
3738 | struct work_struct *work = blk_mq_rq_to_pdu(rq); | |
3739 | ||
3740 | INIT_WORK(work, rbd_queue_workfn); | |
3741 | return 0; | |
3742 | } | |
3743 | ||
3744 | static struct blk_mq_ops rbd_mq_ops = { | |
3745 | .queue_rq = rbd_queue_rq, | |
3746 | .map_queue = blk_mq_map_queue, | |
3747 | .init_request = rbd_init_request, | |
3748 | }; | |
3749 | ||
602adf40 YS |
3750 | static int rbd_init_disk(struct rbd_device *rbd_dev) |
3751 | { | |
3752 | struct gendisk *disk; | |
3753 | struct request_queue *q; | |
593a9e7b | 3754 | u64 segment_size; |
7ad18afa | 3755 | int err; |
602adf40 | 3756 | |
602adf40 | 3757 | /* create gendisk info */ |
7e513d43 ID |
3758 | disk = alloc_disk(single_major ? |
3759 | (1 << RBD_SINGLE_MAJOR_PART_SHIFT) : | |
3760 | RBD_MINORS_PER_MAJOR); | |
602adf40 | 3761 | if (!disk) |
1fcdb8aa | 3762 | return -ENOMEM; |
602adf40 | 3763 | |
f0f8cef5 | 3764 | snprintf(disk->disk_name, sizeof(disk->disk_name), RBD_DRV_NAME "%d", |
de71a297 | 3765 | rbd_dev->dev_id); |
602adf40 | 3766 | disk->major = rbd_dev->major; |
dd82fff1 | 3767 | disk->first_minor = rbd_dev->minor; |
7e513d43 ID |
3768 | if (single_major) |
3769 | disk->flags |= GENHD_FL_EXT_DEVT; | |
602adf40 YS |
3770 | disk->fops = &rbd_bd_ops; |
3771 | disk->private_data = rbd_dev; | |
3772 | ||
7ad18afa CH |
3773 | memset(&rbd_dev->tag_set, 0, sizeof(rbd_dev->tag_set)); |
3774 | rbd_dev->tag_set.ops = &rbd_mq_ops; | |
b5584180 | 3775 | rbd_dev->tag_set.queue_depth = rbd_dev->opts->queue_depth; |
7ad18afa | 3776 | rbd_dev->tag_set.numa_node = NUMA_NO_NODE; |
b5584180 | 3777 | rbd_dev->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE; |
7ad18afa CH |
3778 | rbd_dev->tag_set.nr_hw_queues = 1; |
3779 | rbd_dev->tag_set.cmd_size = sizeof(struct work_struct); | |
3780 | ||
3781 | err = blk_mq_alloc_tag_set(&rbd_dev->tag_set); | |
3782 | if (err) | |
602adf40 | 3783 | goto out_disk; |
029bcbd8 | 3784 | |
7ad18afa CH |
3785 | q = blk_mq_init_queue(&rbd_dev->tag_set); |
3786 | if (IS_ERR(q)) { | |
3787 | err = PTR_ERR(q); | |
3788 | goto out_tag_set; | |
3789 | } | |
3790 | ||
d8a2c89c ID |
3791 | queue_flag_set_unlocked(QUEUE_FLAG_NONROT, q); |
3792 | /* QUEUE_FLAG_ADD_RANDOM is off by default for blk-mq */ | |
593a9e7b | 3793 | |
029bcbd8 | 3794 | /* set io sizes to object size */ |
593a9e7b AE |
3795 | segment_size = rbd_obj_bytes(&rbd_dev->header); |
3796 | blk_queue_max_hw_sectors(q, segment_size / SECTOR_SIZE); | |
d3834fef | 3797 | blk_queue_max_segments(q, segment_size / SECTOR_SIZE); |
593a9e7b AE |
3798 | blk_queue_max_segment_size(q, segment_size); |
3799 | blk_queue_io_min(q, segment_size); | |
3800 | blk_queue_io_opt(q, segment_size); | |
029bcbd8 | 3801 | |
90e98c52 GZ |
3802 | /* enable the discard support */ |
3803 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q); | |
3804 | q->limits.discard_granularity = segment_size; | |
3805 | q->limits.discard_alignment = segment_size; | |
b76f8239 JD |
3806 | q->limits.max_discard_sectors = segment_size / SECTOR_SIZE; |
3807 | q->limits.discard_zeroes_data = 1; | |
90e98c52 | 3808 | |
602adf40 YS |
3809 | blk_queue_merge_bvec(q, rbd_merge_bvec); |
3810 | disk->queue = q; | |
3811 | ||
3812 | q->queuedata = rbd_dev; | |
3813 | ||
3814 | rbd_dev->disk = disk; | |
602adf40 | 3815 | |
602adf40 | 3816 | return 0; |
7ad18afa CH |
3817 | out_tag_set: |
3818 | blk_mq_free_tag_set(&rbd_dev->tag_set); | |
602adf40 YS |
3819 | out_disk: |
3820 | put_disk(disk); | |
7ad18afa | 3821 | return err; |
602adf40 YS |
3822 | } |
3823 | ||
dfc5606d YS |
3824 | /* |
3825 | sysfs | |
3826 | */ | |
3827 | ||
593a9e7b AE |
3828 | static struct rbd_device *dev_to_rbd_dev(struct device *dev) |
3829 | { | |
3830 | return container_of(dev, struct rbd_device, dev); | |
3831 | } | |
3832 | ||
dfc5606d YS |
3833 | static ssize_t rbd_size_show(struct device *dev, |
3834 | struct device_attribute *attr, char *buf) | |
3835 | { | |
593a9e7b | 3836 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
a51aa0c0 | 3837 | |
fc71d833 AE |
3838 | return sprintf(buf, "%llu\n", |
3839 | (unsigned long long)rbd_dev->mapping.size); | |
dfc5606d YS |
3840 | } |
3841 | ||
34b13184 AE |
3842 | /* |
3843 | * Note this shows the features for whatever's mapped, which is not | |
3844 | * necessarily the base image. | |
3845 | */ | |
3846 | static ssize_t rbd_features_show(struct device *dev, | |
3847 | struct device_attribute *attr, char *buf) | |
3848 | { | |
3849 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
3850 | ||
3851 | return sprintf(buf, "0x%016llx\n", | |
fc71d833 | 3852 | (unsigned long long)rbd_dev->mapping.features); |
34b13184 AE |
3853 | } |
3854 | ||
dfc5606d YS |
3855 | static ssize_t rbd_major_show(struct device *dev, |
3856 | struct device_attribute *attr, char *buf) | |
3857 | { | |
593a9e7b | 3858 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
602adf40 | 3859 | |
fc71d833 AE |
3860 | if (rbd_dev->major) |
3861 | return sprintf(buf, "%d\n", rbd_dev->major); | |
3862 | ||
3863 | return sprintf(buf, "(none)\n"); | |
dd82fff1 ID |
3864 | } |
3865 | ||
3866 | static ssize_t rbd_minor_show(struct device *dev, | |
3867 | struct device_attribute *attr, char *buf) | |
3868 | { | |
3869 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
fc71d833 | 3870 | |
dd82fff1 | 3871 | return sprintf(buf, "%d\n", rbd_dev->minor); |
dfc5606d YS |
3872 | } |
3873 | ||
3874 | static ssize_t rbd_client_id_show(struct device *dev, | |
3875 | struct device_attribute *attr, char *buf) | |
602adf40 | 3876 | { |
593a9e7b | 3877 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 3878 | |
1dbb4399 AE |
3879 | return sprintf(buf, "client%lld\n", |
3880 | ceph_client_id(rbd_dev->rbd_client->client)); | |
602adf40 YS |
3881 | } |
3882 | ||
dfc5606d YS |
3883 | static ssize_t rbd_pool_show(struct device *dev, |
3884 | struct device_attribute *attr, char *buf) | |
602adf40 | 3885 | { |
593a9e7b | 3886 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 3887 | |
0d7dbfce | 3888 | return sprintf(buf, "%s\n", rbd_dev->spec->pool_name); |
dfc5606d YS |
3889 | } |
3890 | ||
9bb2f334 AE |
3891 | static ssize_t rbd_pool_id_show(struct device *dev, |
3892 | struct device_attribute *attr, char *buf) | |
3893 | { | |
3894 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
3895 | ||
0d7dbfce | 3896 | return sprintf(buf, "%llu\n", |
fc71d833 | 3897 | (unsigned long long) rbd_dev->spec->pool_id); |
9bb2f334 AE |
3898 | } |
3899 | ||
dfc5606d YS |
3900 | static ssize_t rbd_name_show(struct device *dev, |
3901 | struct device_attribute *attr, char *buf) | |
3902 | { | |
593a9e7b | 3903 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 3904 | |
a92ffdf8 AE |
3905 | if (rbd_dev->spec->image_name) |
3906 | return sprintf(buf, "%s\n", rbd_dev->spec->image_name); | |
3907 | ||
3908 | return sprintf(buf, "(unknown)\n"); | |
dfc5606d YS |
3909 | } |
3910 | ||
589d30e0 AE |
3911 | static ssize_t rbd_image_id_show(struct device *dev, |
3912 | struct device_attribute *attr, char *buf) | |
3913 | { | |
3914 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
3915 | ||
0d7dbfce | 3916 | return sprintf(buf, "%s\n", rbd_dev->spec->image_id); |
589d30e0 AE |
3917 | } |
3918 | ||
34b13184 AE |
3919 | /* |
3920 | * Shows the name of the currently-mapped snapshot (or | |
3921 | * RBD_SNAP_HEAD_NAME for the base image). | |
3922 | */ | |
dfc5606d YS |
3923 | static ssize_t rbd_snap_show(struct device *dev, |
3924 | struct device_attribute *attr, | |
3925 | char *buf) | |
3926 | { | |
593a9e7b | 3927 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 3928 | |
0d7dbfce | 3929 | return sprintf(buf, "%s\n", rbd_dev->spec->snap_name); |
dfc5606d YS |
3930 | } |
3931 | ||
86b00e0d | 3932 | /* |
ff96128f ID |
3933 | * For a v2 image, shows the chain of parent images, separated by empty |
3934 | * lines. For v1 images or if there is no parent, shows "(no parent | |
3935 | * image)". | |
86b00e0d AE |
3936 | */ |
3937 | static ssize_t rbd_parent_show(struct device *dev, | |
ff96128f ID |
3938 | struct device_attribute *attr, |
3939 | char *buf) | |
86b00e0d AE |
3940 | { |
3941 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
ff96128f | 3942 | ssize_t count = 0; |
86b00e0d | 3943 | |
ff96128f | 3944 | if (!rbd_dev->parent) |
86b00e0d AE |
3945 | return sprintf(buf, "(no parent image)\n"); |
3946 | ||
ff96128f ID |
3947 | for ( ; rbd_dev->parent; rbd_dev = rbd_dev->parent) { |
3948 | struct rbd_spec *spec = rbd_dev->parent_spec; | |
3949 | ||
3950 | count += sprintf(&buf[count], "%s" | |
3951 | "pool_id %llu\npool_name %s\n" | |
3952 | "image_id %s\nimage_name %s\n" | |
3953 | "snap_id %llu\nsnap_name %s\n" | |
3954 | "overlap %llu\n", | |
3955 | !count ? "" : "\n", /* first? */ | |
3956 | spec->pool_id, spec->pool_name, | |
3957 | spec->image_id, spec->image_name ?: "(unknown)", | |
3958 | spec->snap_id, spec->snap_name, | |
3959 | rbd_dev->parent_overlap); | |
3960 | } | |
3961 | ||
3962 | return count; | |
86b00e0d AE |
3963 | } |
3964 | ||
dfc5606d YS |
3965 | static ssize_t rbd_image_refresh(struct device *dev, |
3966 | struct device_attribute *attr, | |
3967 | const char *buf, | |
3968 | size_t size) | |
3969 | { | |
593a9e7b | 3970 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
b813623a | 3971 | int ret; |
602adf40 | 3972 | |
cc4a38bd | 3973 | ret = rbd_dev_refresh(rbd_dev); |
e627db08 | 3974 | if (ret) |
52bb1f9b | 3975 | return ret; |
b813623a | 3976 | |
52bb1f9b | 3977 | return size; |
dfc5606d | 3978 | } |
602adf40 | 3979 | |
dfc5606d | 3980 | static DEVICE_ATTR(size, S_IRUGO, rbd_size_show, NULL); |
34b13184 | 3981 | static DEVICE_ATTR(features, S_IRUGO, rbd_features_show, NULL); |
dfc5606d | 3982 | static DEVICE_ATTR(major, S_IRUGO, rbd_major_show, NULL); |
dd82fff1 | 3983 | static DEVICE_ATTR(minor, S_IRUGO, rbd_minor_show, NULL); |
dfc5606d YS |
3984 | static DEVICE_ATTR(client_id, S_IRUGO, rbd_client_id_show, NULL); |
3985 | static DEVICE_ATTR(pool, S_IRUGO, rbd_pool_show, NULL); | |
9bb2f334 | 3986 | static DEVICE_ATTR(pool_id, S_IRUGO, rbd_pool_id_show, NULL); |
dfc5606d | 3987 | static DEVICE_ATTR(name, S_IRUGO, rbd_name_show, NULL); |
589d30e0 | 3988 | static DEVICE_ATTR(image_id, S_IRUGO, rbd_image_id_show, NULL); |
dfc5606d YS |
3989 | static DEVICE_ATTR(refresh, S_IWUSR, NULL, rbd_image_refresh); |
3990 | static DEVICE_ATTR(current_snap, S_IRUGO, rbd_snap_show, NULL); | |
86b00e0d | 3991 | static DEVICE_ATTR(parent, S_IRUGO, rbd_parent_show, NULL); |
dfc5606d YS |
3992 | |
3993 | static struct attribute *rbd_attrs[] = { | |
3994 | &dev_attr_size.attr, | |
34b13184 | 3995 | &dev_attr_features.attr, |
dfc5606d | 3996 | &dev_attr_major.attr, |
dd82fff1 | 3997 | &dev_attr_minor.attr, |
dfc5606d YS |
3998 | &dev_attr_client_id.attr, |
3999 | &dev_attr_pool.attr, | |
9bb2f334 | 4000 | &dev_attr_pool_id.attr, |
dfc5606d | 4001 | &dev_attr_name.attr, |
589d30e0 | 4002 | &dev_attr_image_id.attr, |
dfc5606d | 4003 | &dev_attr_current_snap.attr, |
86b00e0d | 4004 | &dev_attr_parent.attr, |
dfc5606d | 4005 | &dev_attr_refresh.attr, |
dfc5606d YS |
4006 | NULL |
4007 | }; | |
4008 | ||
4009 | static struct attribute_group rbd_attr_group = { | |
4010 | .attrs = rbd_attrs, | |
4011 | }; | |
4012 | ||
4013 | static const struct attribute_group *rbd_attr_groups[] = { | |
4014 | &rbd_attr_group, | |
4015 | NULL | |
4016 | }; | |
4017 | ||
4018 | static void rbd_sysfs_dev_release(struct device *dev) | |
4019 | { | |
4020 | } | |
4021 | ||
4022 | static struct device_type rbd_device_type = { | |
4023 | .name = "rbd", | |
4024 | .groups = rbd_attr_groups, | |
4025 | .release = rbd_sysfs_dev_release, | |
4026 | }; | |
4027 | ||
8b8fb99c AE |
4028 | static struct rbd_spec *rbd_spec_get(struct rbd_spec *spec) |
4029 | { | |
4030 | kref_get(&spec->kref); | |
4031 | ||
4032 | return spec; | |
4033 | } | |
4034 | ||
4035 | static void rbd_spec_free(struct kref *kref); | |
4036 | static void rbd_spec_put(struct rbd_spec *spec) | |
4037 | { | |
4038 | if (spec) | |
4039 | kref_put(&spec->kref, rbd_spec_free); | |
4040 | } | |
4041 | ||
4042 | static struct rbd_spec *rbd_spec_alloc(void) | |
4043 | { | |
4044 | struct rbd_spec *spec; | |
4045 | ||
4046 | spec = kzalloc(sizeof (*spec), GFP_KERNEL); | |
4047 | if (!spec) | |
4048 | return NULL; | |
04077599 ID |
4049 | |
4050 | spec->pool_id = CEPH_NOPOOL; | |
4051 | spec->snap_id = CEPH_NOSNAP; | |
8b8fb99c AE |
4052 | kref_init(&spec->kref); |
4053 | ||
8b8fb99c AE |
4054 | return spec; |
4055 | } | |
4056 | ||
4057 | static void rbd_spec_free(struct kref *kref) | |
4058 | { | |
4059 | struct rbd_spec *spec = container_of(kref, struct rbd_spec, kref); | |
4060 | ||
4061 | kfree(spec->pool_name); | |
4062 | kfree(spec->image_id); | |
4063 | kfree(spec->image_name); | |
4064 | kfree(spec->snap_name); | |
4065 | kfree(spec); | |
4066 | } | |
4067 | ||
cc344fa1 | 4068 | static struct rbd_device *rbd_dev_create(struct rbd_client *rbdc, |
d147543d ID |
4069 | struct rbd_spec *spec, |
4070 | struct rbd_options *opts) | |
c53d5893 AE |
4071 | { |
4072 | struct rbd_device *rbd_dev; | |
4073 | ||
4074 | rbd_dev = kzalloc(sizeof (*rbd_dev), GFP_KERNEL); | |
4075 | if (!rbd_dev) | |
4076 | return NULL; | |
4077 | ||
4078 | spin_lock_init(&rbd_dev->lock); | |
6d292906 | 4079 | rbd_dev->flags = 0; |
a2acd00e | 4080 | atomic_set(&rbd_dev->parent_ref, 0); |
c53d5893 | 4081 | INIT_LIST_HEAD(&rbd_dev->node); |
c53d5893 AE |
4082 | init_rwsem(&rbd_dev->header_rwsem); |
4083 | ||
c53d5893 | 4084 | rbd_dev->rbd_client = rbdc; |
d147543d ID |
4085 | rbd_dev->spec = spec; |
4086 | rbd_dev->opts = opts; | |
c53d5893 | 4087 | |
0903e875 AE |
4088 | /* Initialize the layout used for all rbd requests */ |
4089 | ||
4090 | rbd_dev->layout.fl_stripe_unit = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER); | |
4091 | rbd_dev->layout.fl_stripe_count = cpu_to_le32(1); | |
4092 | rbd_dev->layout.fl_object_size = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER); | |
4093 | rbd_dev->layout.fl_pg_pool = cpu_to_le32((u32) spec->pool_id); | |
4094 | ||
c53d5893 AE |
4095 | return rbd_dev; |
4096 | } | |
4097 | ||
4098 | static void rbd_dev_destroy(struct rbd_device *rbd_dev) | |
4099 | { | |
c53d5893 AE |
4100 | rbd_put_client(rbd_dev->rbd_client); |
4101 | rbd_spec_put(rbd_dev->spec); | |
d147543d | 4102 | kfree(rbd_dev->opts); |
c53d5893 AE |
4103 | kfree(rbd_dev); |
4104 | } | |
4105 | ||
9d475de5 AE |
4106 | /* |
4107 | * Get the size and object order for an image snapshot, or if | |
4108 | * snap_id is CEPH_NOSNAP, gets this information for the base | |
4109 | * image. | |
4110 | */ | |
4111 | static int _rbd_dev_v2_snap_size(struct rbd_device *rbd_dev, u64 snap_id, | |
4112 | u8 *order, u64 *snap_size) | |
4113 | { | |
4114 | __le64 snapid = cpu_to_le64(snap_id); | |
4115 | int ret; | |
4116 | struct { | |
4117 | u8 order; | |
4118 | __le64 size; | |
4119 | } __attribute__ ((packed)) size_buf = { 0 }; | |
4120 | ||
36be9a76 | 4121 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
9d475de5 | 4122 | "rbd", "get_size", |
4157976b | 4123 | &snapid, sizeof (snapid), |
e2a58ee5 | 4124 | &size_buf, sizeof (size_buf)); |
36be9a76 | 4125 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
9d475de5 AE |
4126 | if (ret < 0) |
4127 | return ret; | |
57385b51 AE |
4128 | if (ret < sizeof (size_buf)) |
4129 | return -ERANGE; | |
9d475de5 | 4130 | |
c3545579 | 4131 | if (order) { |
c86f86e9 | 4132 | *order = size_buf.order; |
c3545579 JD |
4133 | dout(" order %u", (unsigned int)*order); |
4134 | } | |
9d475de5 AE |
4135 | *snap_size = le64_to_cpu(size_buf.size); |
4136 | ||
c3545579 JD |
4137 | dout(" snap_id 0x%016llx snap_size = %llu\n", |
4138 | (unsigned long long)snap_id, | |
57385b51 | 4139 | (unsigned long long)*snap_size); |
9d475de5 AE |
4140 | |
4141 | return 0; | |
4142 | } | |
4143 | ||
4144 | static int rbd_dev_v2_image_size(struct rbd_device *rbd_dev) | |
4145 | { | |
4146 | return _rbd_dev_v2_snap_size(rbd_dev, CEPH_NOSNAP, | |
4147 | &rbd_dev->header.obj_order, | |
4148 | &rbd_dev->header.image_size); | |
4149 | } | |
4150 | ||
1e130199 AE |
4151 | static int rbd_dev_v2_object_prefix(struct rbd_device *rbd_dev) |
4152 | { | |
4153 | void *reply_buf; | |
4154 | int ret; | |
4155 | void *p; | |
4156 | ||
4157 | reply_buf = kzalloc(RBD_OBJ_PREFIX_LEN_MAX, GFP_KERNEL); | |
4158 | if (!reply_buf) | |
4159 | return -ENOMEM; | |
4160 | ||
36be9a76 | 4161 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
4157976b | 4162 | "rbd", "get_object_prefix", NULL, 0, |
e2a58ee5 | 4163 | reply_buf, RBD_OBJ_PREFIX_LEN_MAX); |
36be9a76 | 4164 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
1e130199 AE |
4165 | if (ret < 0) |
4166 | goto out; | |
4167 | ||
4168 | p = reply_buf; | |
4169 | rbd_dev->header.object_prefix = ceph_extract_encoded_string(&p, | |
57385b51 AE |
4170 | p + ret, NULL, GFP_NOIO); |
4171 | ret = 0; | |
1e130199 AE |
4172 | |
4173 | if (IS_ERR(rbd_dev->header.object_prefix)) { | |
4174 | ret = PTR_ERR(rbd_dev->header.object_prefix); | |
4175 | rbd_dev->header.object_prefix = NULL; | |
4176 | } else { | |
4177 | dout(" object_prefix = %s\n", rbd_dev->header.object_prefix); | |
4178 | } | |
1e130199 AE |
4179 | out: |
4180 | kfree(reply_buf); | |
4181 | ||
4182 | return ret; | |
4183 | } | |
4184 | ||
b1b5402a AE |
4185 | static int _rbd_dev_v2_snap_features(struct rbd_device *rbd_dev, u64 snap_id, |
4186 | u64 *snap_features) | |
4187 | { | |
4188 | __le64 snapid = cpu_to_le64(snap_id); | |
4189 | struct { | |
4190 | __le64 features; | |
4191 | __le64 incompat; | |
4157976b | 4192 | } __attribute__ ((packed)) features_buf = { 0 }; |
d889140c | 4193 | u64 incompat; |
b1b5402a AE |
4194 | int ret; |
4195 | ||
36be9a76 | 4196 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
b1b5402a | 4197 | "rbd", "get_features", |
4157976b | 4198 | &snapid, sizeof (snapid), |
e2a58ee5 | 4199 | &features_buf, sizeof (features_buf)); |
36be9a76 | 4200 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
b1b5402a AE |
4201 | if (ret < 0) |
4202 | return ret; | |
57385b51 AE |
4203 | if (ret < sizeof (features_buf)) |
4204 | return -ERANGE; | |
d889140c AE |
4205 | |
4206 | incompat = le64_to_cpu(features_buf.incompat); | |
5cbf6f12 | 4207 | if (incompat & ~RBD_FEATURES_SUPPORTED) |
b8f5c6ed | 4208 | return -ENXIO; |
d889140c | 4209 | |
b1b5402a AE |
4210 | *snap_features = le64_to_cpu(features_buf.features); |
4211 | ||
4212 | dout(" snap_id 0x%016llx features = 0x%016llx incompat = 0x%016llx\n", | |
57385b51 AE |
4213 | (unsigned long long)snap_id, |
4214 | (unsigned long long)*snap_features, | |
4215 | (unsigned long long)le64_to_cpu(features_buf.incompat)); | |
b1b5402a AE |
4216 | |
4217 | return 0; | |
4218 | } | |
4219 | ||
4220 | static int rbd_dev_v2_features(struct rbd_device *rbd_dev) | |
4221 | { | |
4222 | return _rbd_dev_v2_snap_features(rbd_dev, CEPH_NOSNAP, | |
4223 | &rbd_dev->header.features); | |
4224 | } | |
4225 | ||
86b00e0d AE |
4226 | static int rbd_dev_v2_parent_info(struct rbd_device *rbd_dev) |
4227 | { | |
4228 | struct rbd_spec *parent_spec; | |
4229 | size_t size; | |
4230 | void *reply_buf = NULL; | |
4231 | __le64 snapid; | |
4232 | void *p; | |
4233 | void *end; | |
642a2537 | 4234 | u64 pool_id; |
86b00e0d | 4235 | char *image_id; |
3b5cf2a2 | 4236 | u64 snap_id; |
86b00e0d | 4237 | u64 overlap; |
86b00e0d AE |
4238 | int ret; |
4239 | ||
4240 | parent_spec = rbd_spec_alloc(); | |
4241 | if (!parent_spec) | |
4242 | return -ENOMEM; | |
4243 | ||
4244 | size = sizeof (__le64) + /* pool_id */ | |
4245 | sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX + /* image_id */ | |
4246 | sizeof (__le64) + /* snap_id */ | |
4247 | sizeof (__le64); /* overlap */ | |
4248 | reply_buf = kmalloc(size, GFP_KERNEL); | |
4249 | if (!reply_buf) { | |
4250 | ret = -ENOMEM; | |
4251 | goto out_err; | |
4252 | } | |
4253 | ||
4d9b67cd | 4254 | snapid = cpu_to_le64(rbd_dev->spec->snap_id); |
36be9a76 | 4255 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
86b00e0d | 4256 | "rbd", "get_parent", |
4157976b | 4257 | &snapid, sizeof (snapid), |
e2a58ee5 | 4258 | reply_buf, size); |
36be9a76 | 4259 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
86b00e0d AE |
4260 | if (ret < 0) |
4261 | goto out_err; | |
4262 | ||
86b00e0d | 4263 | p = reply_buf; |
57385b51 AE |
4264 | end = reply_buf + ret; |
4265 | ret = -ERANGE; | |
642a2537 | 4266 | ceph_decode_64_safe(&p, end, pool_id, out_err); |
392a9dad AE |
4267 | if (pool_id == CEPH_NOPOOL) { |
4268 | /* | |
4269 | * Either the parent never existed, or we have | |
4270 | * record of it but the image got flattened so it no | |
4271 | * longer has a parent. When the parent of a | |
4272 | * layered image disappears we immediately set the | |
4273 | * overlap to 0. The effect of this is that all new | |
4274 | * requests will be treated as if the image had no | |
4275 | * parent. | |
4276 | */ | |
4277 | if (rbd_dev->parent_overlap) { | |
4278 | rbd_dev->parent_overlap = 0; | |
392a9dad AE |
4279 | rbd_dev_parent_put(rbd_dev); |
4280 | pr_info("%s: clone image has been flattened\n", | |
4281 | rbd_dev->disk->disk_name); | |
4282 | } | |
4283 | ||
86b00e0d | 4284 | goto out; /* No parent? No problem. */ |
392a9dad | 4285 | } |
86b00e0d | 4286 | |
0903e875 AE |
4287 | /* The ceph file layout needs to fit pool id in 32 bits */ |
4288 | ||
4289 | ret = -EIO; | |
642a2537 | 4290 | if (pool_id > (u64)U32_MAX) { |
9584d508 | 4291 | rbd_warn(NULL, "parent pool id too large (%llu > %u)", |
642a2537 | 4292 | (unsigned long long)pool_id, U32_MAX); |
57385b51 | 4293 | goto out_err; |
c0cd10db | 4294 | } |
0903e875 | 4295 | |
979ed480 | 4296 | image_id = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL); |
86b00e0d AE |
4297 | if (IS_ERR(image_id)) { |
4298 | ret = PTR_ERR(image_id); | |
4299 | goto out_err; | |
4300 | } | |
3b5cf2a2 | 4301 | ceph_decode_64_safe(&p, end, snap_id, out_err); |
86b00e0d AE |
4302 | ceph_decode_64_safe(&p, end, overlap, out_err); |
4303 | ||
3b5cf2a2 AE |
4304 | /* |
4305 | * The parent won't change (except when the clone is | |
4306 | * flattened, already handled that). So we only need to | |
4307 | * record the parent spec we have not already done so. | |
4308 | */ | |
4309 | if (!rbd_dev->parent_spec) { | |
4310 | parent_spec->pool_id = pool_id; | |
4311 | parent_spec->image_id = image_id; | |
4312 | parent_spec->snap_id = snap_id; | |
70cf49cf AE |
4313 | rbd_dev->parent_spec = parent_spec; |
4314 | parent_spec = NULL; /* rbd_dev now owns this */ | |
fbba11b3 ID |
4315 | } else { |
4316 | kfree(image_id); | |
3b5cf2a2 AE |
4317 | } |
4318 | ||
4319 | /* | |
cf32bd9c ID |
4320 | * We always update the parent overlap. If it's zero we issue |
4321 | * a warning, as we will proceed as if there was no parent. | |
3b5cf2a2 | 4322 | */ |
3b5cf2a2 | 4323 | if (!overlap) { |
3b5cf2a2 | 4324 | if (parent_spec) { |
cf32bd9c ID |
4325 | /* refresh, careful to warn just once */ |
4326 | if (rbd_dev->parent_overlap) | |
4327 | rbd_warn(rbd_dev, | |
4328 | "clone now standalone (overlap became 0)"); | |
3b5cf2a2 | 4329 | } else { |
cf32bd9c ID |
4330 | /* initial probe */ |
4331 | rbd_warn(rbd_dev, "clone is standalone (overlap 0)"); | |
3b5cf2a2 | 4332 | } |
70cf49cf | 4333 | } |
cf32bd9c ID |
4334 | rbd_dev->parent_overlap = overlap; |
4335 | ||
86b00e0d AE |
4336 | out: |
4337 | ret = 0; | |
4338 | out_err: | |
4339 | kfree(reply_buf); | |
4340 | rbd_spec_put(parent_spec); | |
4341 | ||
4342 | return ret; | |
4343 | } | |
4344 | ||
cc070d59 AE |
4345 | static int rbd_dev_v2_striping_info(struct rbd_device *rbd_dev) |
4346 | { | |
4347 | struct { | |
4348 | __le64 stripe_unit; | |
4349 | __le64 stripe_count; | |
4350 | } __attribute__ ((packed)) striping_info_buf = { 0 }; | |
4351 | size_t size = sizeof (striping_info_buf); | |
4352 | void *p; | |
4353 | u64 obj_size; | |
4354 | u64 stripe_unit; | |
4355 | u64 stripe_count; | |
4356 | int ret; | |
4357 | ||
4358 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, | |
4359 | "rbd", "get_stripe_unit_count", NULL, 0, | |
e2a58ee5 | 4360 | (char *)&striping_info_buf, size); |
cc070d59 AE |
4361 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
4362 | if (ret < 0) | |
4363 | return ret; | |
4364 | if (ret < size) | |
4365 | return -ERANGE; | |
4366 | ||
4367 | /* | |
4368 | * We don't actually support the "fancy striping" feature | |
4369 | * (STRIPINGV2) yet, but if the striping sizes are the | |
4370 | * defaults the behavior is the same as before. So find | |
4371 | * out, and only fail if the image has non-default values. | |
4372 | */ | |
4373 | ret = -EINVAL; | |
4374 | obj_size = (u64)1 << rbd_dev->header.obj_order; | |
4375 | p = &striping_info_buf; | |
4376 | stripe_unit = ceph_decode_64(&p); | |
4377 | if (stripe_unit != obj_size) { | |
4378 | rbd_warn(rbd_dev, "unsupported stripe unit " | |
4379 | "(got %llu want %llu)", | |
4380 | stripe_unit, obj_size); | |
4381 | return -EINVAL; | |
4382 | } | |
4383 | stripe_count = ceph_decode_64(&p); | |
4384 | if (stripe_count != 1) { | |
4385 | rbd_warn(rbd_dev, "unsupported stripe count " | |
4386 | "(got %llu want 1)", stripe_count); | |
4387 | return -EINVAL; | |
4388 | } | |
500d0c0f AE |
4389 | rbd_dev->header.stripe_unit = stripe_unit; |
4390 | rbd_dev->header.stripe_count = stripe_count; | |
cc070d59 AE |
4391 | |
4392 | return 0; | |
4393 | } | |
4394 | ||
9e15b77d AE |
4395 | static char *rbd_dev_image_name(struct rbd_device *rbd_dev) |
4396 | { | |
4397 | size_t image_id_size; | |
4398 | char *image_id; | |
4399 | void *p; | |
4400 | void *end; | |
4401 | size_t size; | |
4402 | void *reply_buf = NULL; | |
4403 | size_t len = 0; | |
4404 | char *image_name = NULL; | |
4405 | int ret; | |
4406 | ||
4407 | rbd_assert(!rbd_dev->spec->image_name); | |
4408 | ||
69e7a02f AE |
4409 | len = strlen(rbd_dev->spec->image_id); |
4410 | image_id_size = sizeof (__le32) + len; | |
9e15b77d AE |
4411 | image_id = kmalloc(image_id_size, GFP_KERNEL); |
4412 | if (!image_id) | |
4413 | return NULL; | |
4414 | ||
4415 | p = image_id; | |
4157976b | 4416 | end = image_id + image_id_size; |
57385b51 | 4417 | ceph_encode_string(&p, end, rbd_dev->spec->image_id, (u32)len); |
9e15b77d AE |
4418 | |
4419 | size = sizeof (__le32) + RBD_IMAGE_NAME_LEN_MAX; | |
4420 | reply_buf = kmalloc(size, GFP_KERNEL); | |
4421 | if (!reply_buf) | |
4422 | goto out; | |
4423 | ||
36be9a76 | 4424 | ret = rbd_obj_method_sync(rbd_dev, RBD_DIRECTORY, |
9e15b77d AE |
4425 | "rbd", "dir_get_name", |
4426 | image_id, image_id_size, | |
e2a58ee5 | 4427 | reply_buf, size); |
9e15b77d AE |
4428 | if (ret < 0) |
4429 | goto out; | |
4430 | p = reply_buf; | |
f40eb349 AE |
4431 | end = reply_buf + ret; |
4432 | ||
9e15b77d AE |
4433 | image_name = ceph_extract_encoded_string(&p, end, &len, GFP_KERNEL); |
4434 | if (IS_ERR(image_name)) | |
4435 | image_name = NULL; | |
4436 | else | |
4437 | dout("%s: name is %s len is %zd\n", __func__, image_name, len); | |
4438 | out: | |
4439 | kfree(reply_buf); | |
4440 | kfree(image_id); | |
4441 | ||
4442 | return image_name; | |
4443 | } | |
4444 | ||
2ad3d716 AE |
4445 | static u64 rbd_v1_snap_id_by_name(struct rbd_device *rbd_dev, const char *name) |
4446 | { | |
4447 | struct ceph_snap_context *snapc = rbd_dev->header.snapc; | |
4448 | const char *snap_name; | |
4449 | u32 which = 0; | |
4450 | ||
4451 | /* Skip over names until we find the one we are looking for */ | |
4452 | ||
4453 | snap_name = rbd_dev->header.snap_names; | |
4454 | while (which < snapc->num_snaps) { | |
4455 | if (!strcmp(name, snap_name)) | |
4456 | return snapc->snaps[which]; | |
4457 | snap_name += strlen(snap_name) + 1; | |
4458 | which++; | |
4459 | } | |
4460 | return CEPH_NOSNAP; | |
4461 | } | |
4462 | ||
4463 | static u64 rbd_v2_snap_id_by_name(struct rbd_device *rbd_dev, const char *name) | |
4464 | { | |
4465 | struct ceph_snap_context *snapc = rbd_dev->header.snapc; | |
4466 | u32 which; | |
4467 | bool found = false; | |
4468 | u64 snap_id; | |
4469 | ||
4470 | for (which = 0; !found && which < snapc->num_snaps; which++) { | |
4471 | const char *snap_name; | |
4472 | ||
4473 | snap_id = snapc->snaps[which]; | |
4474 | snap_name = rbd_dev_v2_snap_name(rbd_dev, snap_id); | |
efadc98a JD |
4475 | if (IS_ERR(snap_name)) { |
4476 | /* ignore no-longer existing snapshots */ | |
4477 | if (PTR_ERR(snap_name) == -ENOENT) | |
4478 | continue; | |
4479 | else | |
4480 | break; | |
4481 | } | |
2ad3d716 AE |
4482 | found = !strcmp(name, snap_name); |
4483 | kfree(snap_name); | |
4484 | } | |
4485 | return found ? snap_id : CEPH_NOSNAP; | |
4486 | } | |
4487 | ||
4488 | /* | |
4489 | * Assumes name is never RBD_SNAP_HEAD_NAME; returns CEPH_NOSNAP if | |
4490 | * no snapshot by that name is found, or if an error occurs. | |
4491 | */ | |
4492 | static u64 rbd_snap_id_by_name(struct rbd_device *rbd_dev, const char *name) | |
4493 | { | |
4494 | if (rbd_dev->image_format == 1) | |
4495 | return rbd_v1_snap_id_by_name(rbd_dev, name); | |
4496 | ||
4497 | return rbd_v2_snap_id_by_name(rbd_dev, name); | |
4498 | } | |
4499 | ||
9e15b77d | 4500 | /* |
04077599 ID |
4501 | * An image being mapped will have everything but the snap id. |
4502 | */ | |
4503 | static int rbd_spec_fill_snap_id(struct rbd_device *rbd_dev) | |
4504 | { | |
4505 | struct rbd_spec *spec = rbd_dev->spec; | |
4506 | ||
4507 | rbd_assert(spec->pool_id != CEPH_NOPOOL && spec->pool_name); | |
4508 | rbd_assert(spec->image_id && spec->image_name); | |
4509 | rbd_assert(spec->snap_name); | |
4510 | ||
4511 | if (strcmp(spec->snap_name, RBD_SNAP_HEAD_NAME)) { | |
4512 | u64 snap_id; | |
4513 | ||
4514 | snap_id = rbd_snap_id_by_name(rbd_dev, spec->snap_name); | |
4515 | if (snap_id == CEPH_NOSNAP) | |
4516 | return -ENOENT; | |
4517 | ||
4518 | spec->snap_id = snap_id; | |
4519 | } else { | |
4520 | spec->snap_id = CEPH_NOSNAP; | |
4521 | } | |
4522 | ||
4523 | return 0; | |
4524 | } | |
4525 | ||
4526 | /* | |
4527 | * A parent image will have all ids but none of the names. | |
e1d4213f | 4528 | * |
04077599 ID |
4529 | * All names in an rbd spec are dynamically allocated. It's OK if we |
4530 | * can't figure out the name for an image id. | |
9e15b77d | 4531 | */ |
04077599 | 4532 | static int rbd_spec_fill_names(struct rbd_device *rbd_dev) |
9e15b77d | 4533 | { |
2e9f7f1c AE |
4534 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
4535 | struct rbd_spec *spec = rbd_dev->spec; | |
4536 | const char *pool_name; | |
4537 | const char *image_name; | |
4538 | const char *snap_name; | |
9e15b77d AE |
4539 | int ret; |
4540 | ||
04077599 ID |
4541 | rbd_assert(spec->pool_id != CEPH_NOPOOL); |
4542 | rbd_assert(spec->image_id); | |
4543 | rbd_assert(spec->snap_id != CEPH_NOSNAP); | |
9e15b77d | 4544 | |
2e9f7f1c | 4545 | /* Get the pool name; we have to make our own copy of this */ |
9e15b77d | 4546 | |
2e9f7f1c AE |
4547 | pool_name = ceph_pg_pool_name_by_id(osdc->osdmap, spec->pool_id); |
4548 | if (!pool_name) { | |
4549 | rbd_warn(rbd_dev, "no pool with id %llu", spec->pool_id); | |
935dc89f AE |
4550 | return -EIO; |
4551 | } | |
2e9f7f1c AE |
4552 | pool_name = kstrdup(pool_name, GFP_KERNEL); |
4553 | if (!pool_name) | |
9e15b77d AE |
4554 | return -ENOMEM; |
4555 | ||
4556 | /* Fetch the image name; tolerate failure here */ | |
4557 | ||
2e9f7f1c AE |
4558 | image_name = rbd_dev_image_name(rbd_dev); |
4559 | if (!image_name) | |
06ecc6cb | 4560 | rbd_warn(rbd_dev, "unable to get image name"); |
9e15b77d | 4561 | |
04077599 | 4562 | /* Fetch the snapshot name */ |
9e15b77d | 4563 | |
2e9f7f1c | 4564 | snap_name = rbd_snap_name(rbd_dev, spec->snap_id); |
da6a6b63 JD |
4565 | if (IS_ERR(snap_name)) { |
4566 | ret = PTR_ERR(snap_name); | |
9e15b77d | 4567 | goto out_err; |
2e9f7f1c AE |
4568 | } |
4569 | ||
4570 | spec->pool_name = pool_name; | |
4571 | spec->image_name = image_name; | |
4572 | spec->snap_name = snap_name; | |
9e15b77d AE |
4573 | |
4574 | return 0; | |
04077599 | 4575 | |
9e15b77d | 4576 | out_err: |
2e9f7f1c AE |
4577 | kfree(image_name); |
4578 | kfree(pool_name); | |
9e15b77d AE |
4579 | return ret; |
4580 | } | |
4581 | ||
cc4a38bd | 4582 | static int rbd_dev_v2_snap_context(struct rbd_device *rbd_dev) |
35d489f9 AE |
4583 | { |
4584 | size_t size; | |
4585 | int ret; | |
4586 | void *reply_buf; | |
4587 | void *p; | |
4588 | void *end; | |
4589 | u64 seq; | |
4590 | u32 snap_count; | |
4591 | struct ceph_snap_context *snapc; | |
4592 | u32 i; | |
4593 | ||
4594 | /* | |
4595 | * We'll need room for the seq value (maximum snapshot id), | |
4596 | * snapshot count, and array of that many snapshot ids. | |
4597 | * For now we have a fixed upper limit on the number we're | |
4598 | * prepared to receive. | |
4599 | */ | |
4600 | size = sizeof (__le64) + sizeof (__le32) + | |
4601 | RBD_MAX_SNAP_COUNT * sizeof (__le64); | |
4602 | reply_buf = kzalloc(size, GFP_KERNEL); | |
4603 | if (!reply_buf) | |
4604 | return -ENOMEM; | |
4605 | ||
36be9a76 | 4606 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
4157976b | 4607 | "rbd", "get_snapcontext", NULL, 0, |
e2a58ee5 | 4608 | reply_buf, size); |
36be9a76 | 4609 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
35d489f9 AE |
4610 | if (ret < 0) |
4611 | goto out; | |
4612 | ||
35d489f9 | 4613 | p = reply_buf; |
57385b51 AE |
4614 | end = reply_buf + ret; |
4615 | ret = -ERANGE; | |
35d489f9 AE |
4616 | ceph_decode_64_safe(&p, end, seq, out); |
4617 | ceph_decode_32_safe(&p, end, snap_count, out); | |
4618 | ||
4619 | /* | |
4620 | * Make sure the reported number of snapshot ids wouldn't go | |
4621 | * beyond the end of our buffer. But before checking that, | |
4622 | * make sure the computed size of the snapshot context we | |
4623 | * allocate is representable in a size_t. | |
4624 | */ | |
4625 | if (snap_count > (SIZE_MAX - sizeof (struct ceph_snap_context)) | |
4626 | / sizeof (u64)) { | |
4627 | ret = -EINVAL; | |
4628 | goto out; | |
4629 | } | |
4630 | if (!ceph_has_room(&p, end, snap_count * sizeof (__le64))) | |
4631 | goto out; | |
468521c1 | 4632 | ret = 0; |
35d489f9 | 4633 | |
812164f8 | 4634 | snapc = ceph_create_snap_context(snap_count, GFP_KERNEL); |
35d489f9 AE |
4635 | if (!snapc) { |
4636 | ret = -ENOMEM; | |
4637 | goto out; | |
4638 | } | |
35d489f9 | 4639 | snapc->seq = seq; |
35d489f9 AE |
4640 | for (i = 0; i < snap_count; i++) |
4641 | snapc->snaps[i] = ceph_decode_64(&p); | |
4642 | ||
49ece554 | 4643 | ceph_put_snap_context(rbd_dev->header.snapc); |
35d489f9 AE |
4644 | rbd_dev->header.snapc = snapc; |
4645 | ||
4646 | dout(" snap context seq = %llu, snap_count = %u\n", | |
57385b51 | 4647 | (unsigned long long)seq, (unsigned int)snap_count); |
35d489f9 AE |
4648 | out: |
4649 | kfree(reply_buf); | |
4650 | ||
57385b51 | 4651 | return ret; |
35d489f9 AE |
4652 | } |
4653 | ||
54cac61f AE |
4654 | static const char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev, |
4655 | u64 snap_id) | |
b8b1e2db AE |
4656 | { |
4657 | size_t size; | |
4658 | void *reply_buf; | |
54cac61f | 4659 | __le64 snapid; |
b8b1e2db AE |
4660 | int ret; |
4661 | void *p; | |
4662 | void *end; | |
b8b1e2db AE |
4663 | char *snap_name; |
4664 | ||
4665 | size = sizeof (__le32) + RBD_MAX_SNAP_NAME_LEN; | |
4666 | reply_buf = kmalloc(size, GFP_KERNEL); | |
4667 | if (!reply_buf) | |
4668 | return ERR_PTR(-ENOMEM); | |
4669 | ||
54cac61f | 4670 | snapid = cpu_to_le64(snap_id); |
36be9a76 | 4671 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
b8b1e2db | 4672 | "rbd", "get_snapshot_name", |
54cac61f | 4673 | &snapid, sizeof (snapid), |
e2a58ee5 | 4674 | reply_buf, size); |
36be9a76 | 4675 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
f40eb349 AE |
4676 | if (ret < 0) { |
4677 | snap_name = ERR_PTR(ret); | |
b8b1e2db | 4678 | goto out; |
f40eb349 | 4679 | } |
b8b1e2db AE |
4680 | |
4681 | p = reply_buf; | |
f40eb349 | 4682 | end = reply_buf + ret; |
e5c35534 | 4683 | snap_name = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL); |
f40eb349 | 4684 | if (IS_ERR(snap_name)) |
b8b1e2db | 4685 | goto out; |
b8b1e2db | 4686 | |
f40eb349 | 4687 | dout(" snap_id 0x%016llx snap_name = %s\n", |
54cac61f | 4688 | (unsigned long long)snap_id, snap_name); |
b8b1e2db AE |
4689 | out: |
4690 | kfree(reply_buf); | |
4691 | ||
f40eb349 | 4692 | return snap_name; |
b8b1e2db AE |
4693 | } |
4694 | ||
2df3fac7 | 4695 | static int rbd_dev_v2_header_info(struct rbd_device *rbd_dev) |
117973fb | 4696 | { |
2df3fac7 | 4697 | bool first_time = rbd_dev->header.object_prefix == NULL; |
117973fb | 4698 | int ret; |
117973fb | 4699 | |
1617e40c JD |
4700 | ret = rbd_dev_v2_image_size(rbd_dev); |
4701 | if (ret) | |
cfbf6377 | 4702 | return ret; |
1617e40c | 4703 | |
2df3fac7 AE |
4704 | if (first_time) { |
4705 | ret = rbd_dev_v2_header_onetime(rbd_dev); | |
4706 | if (ret) | |
cfbf6377 | 4707 | return ret; |
2df3fac7 AE |
4708 | } |
4709 | ||
cc4a38bd | 4710 | ret = rbd_dev_v2_snap_context(rbd_dev); |
117973fb | 4711 | dout("rbd_dev_v2_snap_context returned %d\n", ret); |
117973fb AE |
4712 | |
4713 | return ret; | |
4714 | } | |
4715 | ||
a720ae09 ID |
4716 | static int rbd_dev_header_info(struct rbd_device *rbd_dev) |
4717 | { | |
4718 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); | |
4719 | ||
4720 | if (rbd_dev->image_format == 1) | |
4721 | return rbd_dev_v1_header_info(rbd_dev); | |
4722 | ||
4723 | return rbd_dev_v2_header_info(rbd_dev); | |
4724 | } | |
4725 | ||
dfc5606d YS |
4726 | static int rbd_bus_add_dev(struct rbd_device *rbd_dev) |
4727 | { | |
dfc5606d | 4728 | struct device *dev; |
cd789ab9 | 4729 | int ret; |
dfc5606d | 4730 | |
cd789ab9 | 4731 | dev = &rbd_dev->dev; |
dfc5606d YS |
4732 | dev->bus = &rbd_bus_type; |
4733 | dev->type = &rbd_device_type; | |
4734 | dev->parent = &rbd_root_dev; | |
200a6a8b | 4735 | dev->release = rbd_dev_device_release; |
de71a297 | 4736 | dev_set_name(dev, "%d", rbd_dev->dev_id); |
dfc5606d | 4737 | ret = device_register(dev); |
dfc5606d | 4738 | |
dfc5606d | 4739 | return ret; |
602adf40 YS |
4740 | } |
4741 | ||
dfc5606d YS |
4742 | static void rbd_bus_del_dev(struct rbd_device *rbd_dev) |
4743 | { | |
4744 | device_unregister(&rbd_dev->dev); | |
4745 | } | |
4746 | ||
1ddbe94e | 4747 | /* |
499afd5b | 4748 | * Get a unique rbd identifier for the given new rbd_dev, and add |
f8a22fc2 | 4749 | * the rbd_dev to the global list. |
1ddbe94e | 4750 | */ |
f8a22fc2 | 4751 | static int rbd_dev_id_get(struct rbd_device *rbd_dev) |
b7f23c36 | 4752 | { |
f8a22fc2 ID |
4753 | int new_dev_id; |
4754 | ||
9b60e70b ID |
4755 | new_dev_id = ida_simple_get(&rbd_dev_id_ida, |
4756 | 0, minor_to_rbd_dev_id(1 << MINORBITS), | |
4757 | GFP_KERNEL); | |
f8a22fc2 ID |
4758 | if (new_dev_id < 0) |
4759 | return new_dev_id; | |
4760 | ||
4761 | rbd_dev->dev_id = new_dev_id; | |
499afd5b AE |
4762 | |
4763 | spin_lock(&rbd_dev_list_lock); | |
4764 | list_add_tail(&rbd_dev->node, &rbd_dev_list); | |
4765 | spin_unlock(&rbd_dev_list_lock); | |
f8a22fc2 | 4766 | |
70eebd20 | 4767 | dout("rbd_dev %p given dev id %d\n", rbd_dev, rbd_dev->dev_id); |
f8a22fc2 ID |
4768 | |
4769 | return 0; | |
1ddbe94e | 4770 | } |
b7f23c36 | 4771 | |
1ddbe94e | 4772 | /* |
499afd5b AE |
4773 | * Remove an rbd_dev from the global list, and record that its |
4774 | * identifier is no longer in use. | |
1ddbe94e | 4775 | */ |
e2839308 | 4776 | static void rbd_dev_id_put(struct rbd_device *rbd_dev) |
1ddbe94e | 4777 | { |
499afd5b AE |
4778 | spin_lock(&rbd_dev_list_lock); |
4779 | list_del_init(&rbd_dev->node); | |
4780 | spin_unlock(&rbd_dev_list_lock); | |
b7f23c36 | 4781 | |
f8a22fc2 ID |
4782 | ida_simple_remove(&rbd_dev_id_ida, rbd_dev->dev_id); |
4783 | ||
4784 | dout("rbd_dev %p released dev id %d\n", rbd_dev, rbd_dev->dev_id); | |
b7f23c36 AE |
4785 | } |
4786 | ||
e28fff26 AE |
4787 | /* |
4788 | * Skips over white space at *buf, and updates *buf to point to the | |
4789 | * first found non-space character (if any). Returns the length of | |
593a9e7b AE |
4790 | * the token (string of non-white space characters) found. Note |
4791 | * that *buf must be terminated with '\0'. | |
e28fff26 AE |
4792 | */ |
4793 | static inline size_t next_token(const char **buf) | |
4794 | { | |
4795 | /* | |
4796 | * These are the characters that produce nonzero for | |
4797 | * isspace() in the "C" and "POSIX" locales. | |
4798 | */ | |
4799 | const char *spaces = " \f\n\r\t\v"; | |
4800 | ||
4801 | *buf += strspn(*buf, spaces); /* Find start of token */ | |
4802 | ||
4803 | return strcspn(*buf, spaces); /* Return token length */ | |
4804 | } | |
4805 | ||
ea3352f4 AE |
4806 | /* |
4807 | * Finds the next token in *buf, dynamically allocates a buffer big | |
4808 | * enough to hold a copy of it, and copies the token into the new | |
4809 | * buffer. The copy is guaranteed to be terminated with '\0'. Note | |
4810 | * that a duplicate buffer is created even for a zero-length token. | |
4811 | * | |
4812 | * Returns a pointer to the newly-allocated duplicate, or a null | |
4813 | * pointer if memory for the duplicate was not available. If | |
4814 | * the lenp argument is a non-null pointer, the length of the token | |
4815 | * (not including the '\0') is returned in *lenp. | |
4816 | * | |
4817 | * If successful, the *buf pointer will be updated to point beyond | |
4818 | * the end of the found token. | |
4819 | * | |
4820 | * Note: uses GFP_KERNEL for allocation. | |
4821 | */ | |
4822 | static inline char *dup_token(const char **buf, size_t *lenp) | |
4823 | { | |
4824 | char *dup; | |
4825 | size_t len; | |
4826 | ||
4827 | len = next_token(buf); | |
4caf35f9 | 4828 | dup = kmemdup(*buf, len + 1, GFP_KERNEL); |
ea3352f4 AE |
4829 | if (!dup) |
4830 | return NULL; | |
ea3352f4 AE |
4831 | *(dup + len) = '\0'; |
4832 | *buf += len; | |
4833 | ||
4834 | if (lenp) | |
4835 | *lenp = len; | |
4836 | ||
4837 | return dup; | |
4838 | } | |
4839 | ||
a725f65e | 4840 | /* |
859c31df AE |
4841 | * Parse the options provided for an "rbd add" (i.e., rbd image |
4842 | * mapping) request. These arrive via a write to /sys/bus/rbd/add, | |
4843 | * and the data written is passed here via a NUL-terminated buffer. | |
4844 | * Returns 0 if successful or an error code otherwise. | |
d22f76e7 | 4845 | * |
859c31df AE |
4846 | * The information extracted from these options is recorded in |
4847 | * the other parameters which return dynamically-allocated | |
4848 | * structures: | |
4849 | * ceph_opts | |
4850 | * The address of a pointer that will refer to a ceph options | |
4851 | * structure. Caller must release the returned pointer using | |
4852 | * ceph_destroy_options() when it is no longer needed. | |
4853 | * rbd_opts | |
4854 | * Address of an rbd options pointer. Fully initialized by | |
4855 | * this function; caller must release with kfree(). | |
4856 | * spec | |
4857 | * Address of an rbd image specification pointer. Fully | |
4858 | * initialized by this function based on parsed options. | |
4859 | * Caller must release with rbd_spec_put(). | |
4860 | * | |
4861 | * The options passed take this form: | |
4862 | * <mon_addrs> <options> <pool_name> <image_name> [<snap_id>] | |
4863 | * where: | |
4864 | * <mon_addrs> | |
4865 | * A comma-separated list of one or more monitor addresses. | |
4866 | * A monitor address is an ip address, optionally followed | |
4867 | * by a port number (separated by a colon). | |
4868 | * I.e.: ip1[:port1][,ip2[:port2]...] | |
4869 | * <options> | |
4870 | * A comma-separated list of ceph and/or rbd options. | |
4871 | * <pool_name> | |
4872 | * The name of the rados pool containing the rbd image. | |
4873 | * <image_name> | |
4874 | * The name of the image in that pool to map. | |
4875 | * <snap_id> | |
4876 | * An optional snapshot id. If provided, the mapping will | |
4877 | * present data from the image at the time that snapshot was | |
4878 | * created. The image head is used if no snapshot id is | |
4879 | * provided. Snapshot mappings are always read-only. | |
a725f65e | 4880 | */ |
859c31df | 4881 | static int rbd_add_parse_args(const char *buf, |
dc79b113 | 4882 | struct ceph_options **ceph_opts, |
859c31df AE |
4883 | struct rbd_options **opts, |
4884 | struct rbd_spec **rbd_spec) | |
e28fff26 | 4885 | { |
d22f76e7 | 4886 | size_t len; |
859c31df | 4887 | char *options; |
0ddebc0c | 4888 | const char *mon_addrs; |
ecb4dc22 | 4889 | char *snap_name; |
0ddebc0c | 4890 | size_t mon_addrs_size; |
859c31df | 4891 | struct rbd_spec *spec = NULL; |
4e9afeba | 4892 | struct rbd_options *rbd_opts = NULL; |
859c31df | 4893 | struct ceph_options *copts; |
dc79b113 | 4894 | int ret; |
e28fff26 AE |
4895 | |
4896 | /* The first four tokens are required */ | |
4897 | ||
7ef3214a | 4898 | len = next_token(&buf); |
4fb5d671 AE |
4899 | if (!len) { |
4900 | rbd_warn(NULL, "no monitor address(es) provided"); | |
4901 | return -EINVAL; | |
4902 | } | |
0ddebc0c | 4903 | mon_addrs = buf; |
f28e565a | 4904 | mon_addrs_size = len + 1; |
7ef3214a | 4905 | buf += len; |
a725f65e | 4906 | |
dc79b113 | 4907 | ret = -EINVAL; |
f28e565a AE |
4908 | options = dup_token(&buf, NULL); |
4909 | if (!options) | |
dc79b113 | 4910 | return -ENOMEM; |
4fb5d671 AE |
4911 | if (!*options) { |
4912 | rbd_warn(NULL, "no options provided"); | |
4913 | goto out_err; | |
4914 | } | |
e28fff26 | 4915 | |
859c31df AE |
4916 | spec = rbd_spec_alloc(); |
4917 | if (!spec) | |
f28e565a | 4918 | goto out_mem; |
859c31df AE |
4919 | |
4920 | spec->pool_name = dup_token(&buf, NULL); | |
4921 | if (!spec->pool_name) | |
4922 | goto out_mem; | |
4fb5d671 AE |
4923 | if (!*spec->pool_name) { |
4924 | rbd_warn(NULL, "no pool name provided"); | |
4925 | goto out_err; | |
4926 | } | |
e28fff26 | 4927 | |
69e7a02f | 4928 | spec->image_name = dup_token(&buf, NULL); |
859c31df | 4929 | if (!spec->image_name) |
f28e565a | 4930 | goto out_mem; |
4fb5d671 AE |
4931 | if (!*spec->image_name) { |
4932 | rbd_warn(NULL, "no image name provided"); | |
4933 | goto out_err; | |
4934 | } | |
d4b125e9 | 4935 | |
f28e565a AE |
4936 | /* |
4937 | * Snapshot name is optional; default is to use "-" | |
4938 | * (indicating the head/no snapshot). | |
4939 | */ | |
3feeb894 | 4940 | len = next_token(&buf); |
820a5f3e | 4941 | if (!len) { |
3feeb894 AE |
4942 | buf = RBD_SNAP_HEAD_NAME; /* No snapshot supplied */ |
4943 | len = sizeof (RBD_SNAP_HEAD_NAME) - 1; | |
f28e565a | 4944 | } else if (len > RBD_MAX_SNAP_NAME_LEN) { |
dc79b113 | 4945 | ret = -ENAMETOOLONG; |
f28e565a | 4946 | goto out_err; |
849b4260 | 4947 | } |
ecb4dc22 AE |
4948 | snap_name = kmemdup(buf, len + 1, GFP_KERNEL); |
4949 | if (!snap_name) | |
f28e565a | 4950 | goto out_mem; |
ecb4dc22 AE |
4951 | *(snap_name + len) = '\0'; |
4952 | spec->snap_name = snap_name; | |
e5c35534 | 4953 | |
0ddebc0c | 4954 | /* Initialize all rbd options to the defaults */ |
e28fff26 | 4955 | |
4e9afeba AE |
4956 | rbd_opts = kzalloc(sizeof (*rbd_opts), GFP_KERNEL); |
4957 | if (!rbd_opts) | |
4958 | goto out_mem; | |
4959 | ||
4960 | rbd_opts->read_only = RBD_READ_ONLY_DEFAULT; | |
b5584180 | 4961 | rbd_opts->queue_depth = RBD_QUEUE_DEPTH_DEFAULT; |
d22f76e7 | 4962 | |
859c31df | 4963 | copts = ceph_parse_options(options, mon_addrs, |
0ddebc0c | 4964 | mon_addrs + mon_addrs_size - 1, |
4e9afeba | 4965 | parse_rbd_opts_token, rbd_opts); |
859c31df AE |
4966 | if (IS_ERR(copts)) { |
4967 | ret = PTR_ERR(copts); | |
dc79b113 AE |
4968 | goto out_err; |
4969 | } | |
859c31df AE |
4970 | kfree(options); |
4971 | ||
4972 | *ceph_opts = copts; | |
4e9afeba | 4973 | *opts = rbd_opts; |
859c31df | 4974 | *rbd_spec = spec; |
0ddebc0c | 4975 | |
dc79b113 | 4976 | return 0; |
f28e565a | 4977 | out_mem: |
dc79b113 | 4978 | ret = -ENOMEM; |
d22f76e7 | 4979 | out_err: |
859c31df AE |
4980 | kfree(rbd_opts); |
4981 | rbd_spec_put(spec); | |
f28e565a | 4982 | kfree(options); |
d22f76e7 | 4983 | |
dc79b113 | 4984 | return ret; |
a725f65e AE |
4985 | } |
4986 | ||
30ba1f02 ID |
4987 | /* |
4988 | * Return pool id (>= 0) or a negative error code. | |
4989 | */ | |
4990 | static int rbd_add_get_pool_id(struct rbd_client *rbdc, const char *pool_name) | |
4991 | { | |
a319bf56 | 4992 | struct ceph_options *opts = rbdc->client->options; |
30ba1f02 | 4993 | u64 newest_epoch; |
30ba1f02 ID |
4994 | int tries = 0; |
4995 | int ret; | |
4996 | ||
4997 | again: | |
4998 | ret = ceph_pg_poolid_by_name(rbdc->client->osdc.osdmap, pool_name); | |
4999 | if (ret == -ENOENT && tries++ < 1) { | |
5000 | ret = ceph_monc_do_get_version(&rbdc->client->monc, "osdmap", | |
5001 | &newest_epoch); | |
5002 | if (ret < 0) | |
5003 | return ret; | |
5004 | ||
5005 | if (rbdc->client->osdc.osdmap->epoch < newest_epoch) { | |
5006 | ceph_monc_request_next_osdmap(&rbdc->client->monc); | |
5007 | (void) ceph_monc_wait_osdmap(&rbdc->client->monc, | |
a319bf56 ID |
5008 | newest_epoch, |
5009 | opts->mount_timeout); | |
30ba1f02 ID |
5010 | goto again; |
5011 | } else { | |
5012 | /* the osdmap we have is new enough */ | |
5013 | return -ENOENT; | |
5014 | } | |
5015 | } | |
5016 | ||
5017 | return ret; | |
5018 | } | |
5019 | ||
589d30e0 AE |
5020 | /* |
5021 | * An rbd format 2 image has a unique identifier, distinct from the | |
5022 | * name given to it by the user. Internally, that identifier is | |
5023 | * what's used to specify the names of objects related to the image. | |
5024 | * | |
5025 | * A special "rbd id" object is used to map an rbd image name to its | |
5026 | * id. If that object doesn't exist, then there is no v2 rbd image | |
5027 | * with the supplied name. | |
5028 | * | |
5029 | * This function will record the given rbd_dev's image_id field if | |
5030 | * it can be determined, and in that case will return 0. If any | |
5031 | * errors occur a negative errno will be returned and the rbd_dev's | |
5032 | * image_id field will be unchanged (and should be NULL). | |
5033 | */ | |
5034 | static int rbd_dev_image_id(struct rbd_device *rbd_dev) | |
5035 | { | |
5036 | int ret; | |
5037 | size_t size; | |
5038 | char *object_name; | |
5039 | void *response; | |
c0fba368 | 5040 | char *image_id; |
2f82ee54 | 5041 | |
2c0d0a10 AE |
5042 | /* |
5043 | * When probing a parent image, the image id is already | |
5044 | * known (and the image name likely is not). There's no | |
c0fba368 AE |
5045 | * need to fetch the image id again in this case. We |
5046 | * do still need to set the image format though. | |
2c0d0a10 | 5047 | */ |
c0fba368 AE |
5048 | if (rbd_dev->spec->image_id) { |
5049 | rbd_dev->image_format = *rbd_dev->spec->image_id ? 2 : 1; | |
5050 | ||
2c0d0a10 | 5051 | return 0; |
c0fba368 | 5052 | } |
2c0d0a10 | 5053 | |
589d30e0 AE |
5054 | /* |
5055 | * First, see if the format 2 image id file exists, and if | |
5056 | * so, get the image's persistent id from it. | |
5057 | */ | |
69e7a02f | 5058 | size = sizeof (RBD_ID_PREFIX) + strlen(rbd_dev->spec->image_name); |
589d30e0 AE |
5059 | object_name = kmalloc(size, GFP_NOIO); |
5060 | if (!object_name) | |
5061 | return -ENOMEM; | |
0d7dbfce | 5062 | sprintf(object_name, "%s%s", RBD_ID_PREFIX, rbd_dev->spec->image_name); |
589d30e0 AE |
5063 | dout("rbd id object name is %s\n", object_name); |
5064 | ||
5065 | /* Response will be an encoded string, which includes a length */ | |
5066 | ||
5067 | size = sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX; | |
5068 | response = kzalloc(size, GFP_NOIO); | |
5069 | if (!response) { | |
5070 | ret = -ENOMEM; | |
5071 | goto out; | |
5072 | } | |
5073 | ||
c0fba368 AE |
5074 | /* If it doesn't exist we'll assume it's a format 1 image */ |
5075 | ||
36be9a76 | 5076 | ret = rbd_obj_method_sync(rbd_dev, object_name, |
4157976b | 5077 | "rbd", "get_id", NULL, 0, |
e2a58ee5 | 5078 | response, RBD_IMAGE_ID_LEN_MAX); |
36be9a76 | 5079 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
c0fba368 AE |
5080 | if (ret == -ENOENT) { |
5081 | image_id = kstrdup("", GFP_KERNEL); | |
5082 | ret = image_id ? 0 : -ENOMEM; | |
5083 | if (!ret) | |
5084 | rbd_dev->image_format = 1; | |
7dd440c9 | 5085 | } else if (ret >= 0) { |
c0fba368 AE |
5086 | void *p = response; |
5087 | ||
5088 | image_id = ceph_extract_encoded_string(&p, p + ret, | |
979ed480 | 5089 | NULL, GFP_NOIO); |
461f758a | 5090 | ret = PTR_ERR_OR_ZERO(image_id); |
c0fba368 AE |
5091 | if (!ret) |
5092 | rbd_dev->image_format = 2; | |
c0fba368 AE |
5093 | } |
5094 | ||
5095 | if (!ret) { | |
5096 | rbd_dev->spec->image_id = image_id; | |
5097 | dout("image_id is %s\n", image_id); | |
589d30e0 AE |
5098 | } |
5099 | out: | |
5100 | kfree(response); | |
5101 | kfree(object_name); | |
5102 | ||
5103 | return ret; | |
5104 | } | |
5105 | ||
3abef3b3 AE |
5106 | /* |
5107 | * Undo whatever state changes are made by v1 or v2 header info | |
5108 | * call. | |
5109 | */ | |
6fd48b3b AE |
5110 | static void rbd_dev_unprobe(struct rbd_device *rbd_dev) |
5111 | { | |
5112 | struct rbd_image_header *header; | |
5113 | ||
e69b8d41 | 5114 | rbd_dev_parent_put(rbd_dev); |
6fd48b3b AE |
5115 | |
5116 | /* Free dynamic fields from the header, then zero it out */ | |
5117 | ||
5118 | header = &rbd_dev->header; | |
812164f8 | 5119 | ceph_put_snap_context(header->snapc); |
6fd48b3b AE |
5120 | kfree(header->snap_sizes); |
5121 | kfree(header->snap_names); | |
5122 | kfree(header->object_prefix); | |
5123 | memset(header, 0, sizeof (*header)); | |
5124 | } | |
5125 | ||
2df3fac7 | 5126 | static int rbd_dev_v2_header_onetime(struct rbd_device *rbd_dev) |
a30b71b9 AE |
5127 | { |
5128 | int ret; | |
a30b71b9 | 5129 | |
1e130199 | 5130 | ret = rbd_dev_v2_object_prefix(rbd_dev); |
57385b51 | 5131 | if (ret) |
b1b5402a AE |
5132 | goto out_err; |
5133 | ||
2df3fac7 AE |
5134 | /* |
5135 | * Get the and check features for the image. Currently the | |
5136 | * features are assumed to never change. | |
5137 | */ | |
b1b5402a | 5138 | ret = rbd_dev_v2_features(rbd_dev); |
57385b51 | 5139 | if (ret) |
9d475de5 | 5140 | goto out_err; |
35d489f9 | 5141 | |
cc070d59 AE |
5142 | /* If the image supports fancy striping, get its parameters */ |
5143 | ||
5144 | if (rbd_dev->header.features & RBD_FEATURE_STRIPINGV2) { | |
5145 | ret = rbd_dev_v2_striping_info(rbd_dev); | |
5146 | if (ret < 0) | |
5147 | goto out_err; | |
5148 | } | |
2df3fac7 | 5149 | /* No support for crypto and compression type format 2 images */ |
a30b71b9 | 5150 | |
35152979 | 5151 | return 0; |
9d475de5 | 5152 | out_err: |
642a2537 | 5153 | rbd_dev->header.features = 0; |
1e130199 AE |
5154 | kfree(rbd_dev->header.object_prefix); |
5155 | rbd_dev->header.object_prefix = NULL; | |
9d475de5 AE |
5156 | |
5157 | return ret; | |
a30b71b9 AE |
5158 | } |
5159 | ||
124afba2 | 5160 | static int rbd_dev_probe_parent(struct rbd_device *rbd_dev) |
83a06263 | 5161 | { |
2f82ee54 | 5162 | struct rbd_device *parent = NULL; |
124afba2 AE |
5163 | struct rbd_spec *parent_spec; |
5164 | struct rbd_client *rbdc; | |
5165 | int ret; | |
5166 | ||
5167 | if (!rbd_dev->parent_spec) | |
5168 | return 0; | |
5169 | /* | |
5170 | * We need to pass a reference to the client and the parent | |
5171 | * spec when creating the parent rbd_dev. Images related by | |
5172 | * parent/child relationships always share both. | |
5173 | */ | |
5174 | parent_spec = rbd_spec_get(rbd_dev->parent_spec); | |
5175 | rbdc = __rbd_get_client(rbd_dev->rbd_client); | |
5176 | ||
5177 | ret = -ENOMEM; | |
d147543d | 5178 | parent = rbd_dev_create(rbdc, parent_spec, NULL); |
124afba2 AE |
5179 | if (!parent) |
5180 | goto out_err; | |
5181 | ||
1f3ef788 | 5182 | ret = rbd_dev_image_probe(parent, false); |
124afba2 AE |
5183 | if (ret < 0) |
5184 | goto out_err; | |
5185 | rbd_dev->parent = parent; | |
a2acd00e | 5186 | atomic_set(&rbd_dev->parent_ref, 1); |
124afba2 AE |
5187 | |
5188 | return 0; | |
5189 | out_err: | |
5190 | if (parent) { | |
fb65d228 | 5191 | rbd_dev_unparent(rbd_dev); |
124afba2 AE |
5192 | kfree(rbd_dev->header_name); |
5193 | rbd_dev_destroy(parent); | |
5194 | } else { | |
5195 | rbd_put_client(rbdc); | |
5196 | rbd_spec_put(parent_spec); | |
5197 | } | |
5198 | ||
5199 | return ret; | |
5200 | } | |
5201 | ||
200a6a8b | 5202 | static int rbd_dev_device_setup(struct rbd_device *rbd_dev) |
124afba2 | 5203 | { |
83a06263 | 5204 | int ret; |
d1cf5788 | 5205 | |
f8a22fc2 ID |
5206 | /* Get an id and fill in device name. */ |
5207 | ||
5208 | ret = rbd_dev_id_get(rbd_dev); | |
5209 | if (ret) | |
5210 | return ret; | |
83a06263 | 5211 | |
83a06263 AE |
5212 | BUILD_BUG_ON(DEV_NAME_LEN |
5213 | < sizeof (RBD_DRV_NAME) + MAX_INT_FORMAT_WIDTH); | |
5214 | sprintf(rbd_dev->name, "%s%d", RBD_DRV_NAME, rbd_dev->dev_id); | |
5215 | ||
9b60e70b | 5216 | /* Record our major and minor device numbers. */ |
83a06263 | 5217 | |
9b60e70b ID |
5218 | if (!single_major) { |
5219 | ret = register_blkdev(0, rbd_dev->name); | |
5220 | if (ret < 0) | |
5221 | goto err_out_id; | |
5222 | ||
5223 | rbd_dev->major = ret; | |
5224 | rbd_dev->minor = 0; | |
5225 | } else { | |
5226 | rbd_dev->major = rbd_major; | |
5227 | rbd_dev->minor = rbd_dev_id_to_minor(rbd_dev->dev_id); | |
5228 | } | |
83a06263 AE |
5229 | |
5230 | /* Set up the blkdev mapping. */ | |
5231 | ||
5232 | ret = rbd_init_disk(rbd_dev); | |
5233 | if (ret) | |
5234 | goto err_out_blkdev; | |
5235 | ||
f35a4dee | 5236 | ret = rbd_dev_mapping_set(rbd_dev); |
83a06263 AE |
5237 | if (ret) |
5238 | goto err_out_disk; | |
bc1ecc65 | 5239 | |
f35a4dee | 5240 | set_capacity(rbd_dev->disk, rbd_dev->mapping.size / SECTOR_SIZE); |
22001f61 | 5241 | set_disk_ro(rbd_dev->disk, rbd_dev->mapping.read_only); |
f35a4dee AE |
5242 | |
5243 | ret = rbd_bus_add_dev(rbd_dev); | |
5244 | if (ret) | |
f5ee37bd | 5245 | goto err_out_mapping; |
83a06263 | 5246 | |
83a06263 AE |
5247 | /* Everything's ready. Announce the disk to the world. */ |
5248 | ||
129b79d4 | 5249 | set_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags); |
83a06263 AE |
5250 | add_disk(rbd_dev->disk); |
5251 | ||
5252 | pr_info("%s: added with size 0x%llx\n", rbd_dev->disk->disk_name, | |
5253 | (unsigned long long) rbd_dev->mapping.size); | |
5254 | ||
5255 | return ret; | |
2f82ee54 | 5256 | |
f35a4dee AE |
5257 | err_out_mapping: |
5258 | rbd_dev_mapping_clear(rbd_dev); | |
83a06263 AE |
5259 | err_out_disk: |
5260 | rbd_free_disk(rbd_dev); | |
5261 | err_out_blkdev: | |
9b60e70b ID |
5262 | if (!single_major) |
5263 | unregister_blkdev(rbd_dev->major, rbd_dev->name); | |
83a06263 AE |
5264 | err_out_id: |
5265 | rbd_dev_id_put(rbd_dev); | |
d1cf5788 | 5266 | rbd_dev_mapping_clear(rbd_dev); |
83a06263 AE |
5267 | |
5268 | return ret; | |
5269 | } | |
5270 | ||
332bb12d AE |
5271 | static int rbd_dev_header_name(struct rbd_device *rbd_dev) |
5272 | { | |
5273 | struct rbd_spec *spec = rbd_dev->spec; | |
5274 | size_t size; | |
5275 | ||
5276 | /* Record the header object name for this rbd image. */ | |
5277 | ||
5278 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); | |
5279 | ||
5280 | if (rbd_dev->image_format == 1) | |
5281 | size = strlen(spec->image_name) + sizeof (RBD_SUFFIX); | |
5282 | else | |
5283 | size = sizeof (RBD_HEADER_PREFIX) + strlen(spec->image_id); | |
5284 | ||
5285 | rbd_dev->header_name = kmalloc(size, GFP_KERNEL); | |
5286 | if (!rbd_dev->header_name) | |
5287 | return -ENOMEM; | |
5288 | ||
5289 | if (rbd_dev->image_format == 1) | |
5290 | sprintf(rbd_dev->header_name, "%s%s", | |
5291 | spec->image_name, RBD_SUFFIX); | |
5292 | else | |
5293 | sprintf(rbd_dev->header_name, "%s%s", | |
5294 | RBD_HEADER_PREFIX, spec->image_id); | |
5295 | return 0; | |
5296 | } | |
5297 | ||
200a6a8b AE |
5298 | static void rbd_dev_image_release(struct rbd_device *rbd_dev) |
5299 | { | |
6fd48b3b | 5300 | rbd_dev_unprobe(rbd_dev); |
200a6a8b | 5301 | kfree(rbd_dev->header_name); |
6fd48b3b AE |
5302 | rbd_dev->header_name = NULL; |
5303 | rbd_dev->image_format = 0; | |
5304 | kfree(rbd_dev->spec->image_id); | |
5305 | rbd_dev->spec->image_id = NULL; | |
5306 | ||
200a6a8b AE |
5307 | rbd_dev_destroy(rbd_dev); |
5308 | } | |
5309 | ||
a30b71b9 AE |
5310 | /* |
5311 | * Probe for the existence of the header object for the given rbd | |
1f3ef788 AE |
5312 | * device. If this image is the one being mapped (i.e., not a |
5313 | * parent), initiate a watch on its header object before using that | |
5314 | * object to get detailed information about the rbd image. | |
a30b71b9 | 5315 | */ |
1f3ef788 | 5316 | static int rbd_dev_image_probe(struct rbd_device *rbd_dev, bool mapping) |
a30b71b9 AE |
5317 | { |
5318 | int ret; | |
5319 | ||
5320 | /* | |
3abef3b3 AE |
5321 | * Get the id from the image id object. Unless there's an |
5322 | * error, rbd_dev->spec->image_id will be filled in with | |
5323 | * a dynamically-allocated string, and rbd_dev->image_format | |
5324 | * will be set to either 1 or 2. | |
a30b71b9 AE |
5325 | */ |
5326 | ret = rbd_dev_image_id(rbd_dev); | |
5327 | if (ret) | |
c0fba368 | 5328 | return ret; |
c0fba368 | 5329 | |
332bb12d AE |
5330 | ret = rbd_dev_header_name(rbd_dev); |
5331 | if (ret) | |
5332 | goto err_out_format; | |
5333 | ||
1f3ef788 | 5334 | if (mapping) { |
fca27065 | 5335 | ret = rbd_dev_header_watch_sync(rbd_dev); |
1fe48023 ID |
5336 | if (ret) { |
5337 | if (ret == -ENOENT) | |
5338 | pr_info("image %s/%s does not exist\n", | |
5339 | rbd_dev->spec->pool_name, | |
5340 | rbd_dev->spec->image_name); | |
1f3ef788 | 5341 | goto out_header_name; |
1fe48023 | 5342 | } |
1f3ef788 | 5343 | } |
b644de2b | 5344 | |
a720ae09 | 5345 | ret = rbd_dev_header_info(rbd_dev); |
5655c4d9 | 5346 | if (ret) |
b644de2b | 5347 | goto err_out_watch; |
83a06263 | 5348 | |
04077599 ID |
5349 | /* |
5350 | * If this image is the one being mapped, we have pool name and | |
5351 | * id, image name and id, and snap name - need to fill snap id. | |
5352 | * Otherwise this is a parent image, identified by pool, image | |
5353 | * and snap ids - need to fill in names for those ids. | |
5354 | */ | |
5355 | if (mapping) | |
5356 | ret = rbd_spec_fill_snap_id(rbd_dev); | |
5357 | else | |
5358 | ret = rbd_spec_fill_names(rbd_dev); | |
1fe48023 ID |
5359 | if (ret) { |
5360 | if (ret == -ENOENT) | |
5361 | pr_info("snap %s/%s@%s does not exist\n", | |
5362 | rbd_dev->spec->pool_name, | |
5363 | rbd_dev->spec->image_name, | |
5364 | rbd_dev->spec->snap_name); | |
33dca39f | 5365 | goto err_out_probe; |
1fe48023 | 5366 | } |
9bb81c9b | 5367 | |
e8f59b59 ID |
5368 | if (rbd_dev->header.features & RBD_FEATURE_LAYERING) { |
5369 | ret = rbd_dev_v2_parent_info(rbd_dev); | |
5370 | if (ret) | |
5371 | goto err_out_probe; | |
5372 | ||
5373 | /* | |
5374 | * Need to warn users if this image is the one being | |
5375 | * mapped and has a parent. | |
5376 | */ | |
5377 | if (mapping && rbd_dev->parent_spec) | |
5378 | rbd_warn(rbd_dev, | |
5379 | "WARNING: kernel layering is EXPERIMENTAL!"); | |
5380 | } | |
5381 | ||
9bb81c9b | 5382 | ret = rbd_dev_probe_parent(rbd_dev); |
30d60ba2 AE |
5383 | if (ret) |
5384 | goto err_out_probe; | |
5385 | ||
5386 | dout("discovered format %u image, header name is %s\n", | |
5387 | rbd_dev->image_format, rbd_dev->header_name); | |
30d60ba2 | 5388 | return 0; |
e8f59b59 | 5389 | |
6fd48b3b AE |
5390 | err_out_probe: |
5391 | rbd_dev_unprobe(rbd_dev); | |
b644de2b | 5392 | err_out_watch: |
fca27065 ID |
5393 | if (mapping) |
5394 | rbd_dev_header_unwatch_sync(rbd_dev); | |
332bb12d AE |
5395 | out_header_name: |
5396 | kfree(rbd_dev->header_name); | |
5397 | rbd_dev->header_name = NULL; | |
5398 | err_out_format: | |
5399 | rbd_dev->image_format = 0; | |
5655c4d9 AE |
5400 | kfree(rbd_dev->spec->image_id); |
5401 | rbd_dev->spec->image_id = NULL; | |
a30b71b9 AE |
5402 | return ret; |
5403 | } | |
5404 | ||
9b60e70b ID |
5405 | static ssize_t do_rbd_add(struct bus_type *bus, |
5406 | const char *buf, | |
5407 | size_t count) | |
602adf40 | 5408 | { |
cb8627c7 | 5409 | struct rbd_device *rbd_dev = NULL; |
dc79b113 | 5410 | struct ceph_options *ceph_opts = NULL; |
4e9afeba | 5411 | struct rbd_options *rbd_opts = NULL; |
859c31df | 5412 | struct rbd_spec *spec = NULL; |
9d3997fd | 5413 | struct rbd_client *rbdc; |
51344a38 | 5414 | bool read_only; |
27cc2594 | 5415 | int rc = -ENOMEM; |
602adf40 YS |
5416 | |
5417 | if (!try_module_get(THIS_MODULE)) | |
5418 | return -ENODEV; | |
5419 | ||
602adf40 | 5420 | /* parse add command */ |
859c31df | 5421 | rc = rbd_add_parse_args(buf, &ceph_opts, &rbd_opts, &spec); |
dc79b113 | 5422 | if (rc < 0) |
bd4ba655 | 5423 | goto err_out_module; |
78cea76e | 5424 | |
9d3997fd AE |
5425 | rbdc = rbd_get_client(ceph_opts); |
5426 | if (IS_ERR(rbdc)) { | |
5427 | rc = PTR_ERR(rbdc); | |
0ddebc0c | 5428 | goto err_out_args; |
9d3997fd | 5429 | } |
602adf40 | 5430 | |
602adf40 | 5431 | /* pick the pool */ |
30ba1f02 | 5432 | rc = rbd_add_get_pool_id(rbdc, spec->pool_name); |
1fe48023 ID |
5433 | if (rc < 0) { |
5434 | if (rc == -ENOENT) | |
5435 | pr_info("pool %s does not exist\n", spec->pool_name); | |
602adf40 | 5436 | goto err_out_client; |
1fe48023 | 5437 | } |
c0cd10db | 5438 | spec->pool_id = (u64)rc; |
859c31df | 5439 | |
0903e875 AE |
5440 | /* The ceph file layout needs to fit pool id in 32 bits */ |
5441 | ||
c0cd10db | 5442 | if (spec->pool_id > (u64)U32_MAX) { |
9584d508 | 5443 | rbd_warn(NULL, "pool id too large (%llu > %u)", |
c0cd10db | 5444 | (unsigned long long)spec->pool_id, U32_MAX); |
0903e875 AE |
5445 | rc = -EIO; |
5446 | goto err_out_client; | |
5447 | } | |
5448 | ||
d147543d | 5449 | rbd_dev = rbd_dev_create(rbdc, spec, rbd_opts); |
bd4ba655 AE |
5450 | if (!rbd_dev) |
5451 | goto err_out_client; | |
c53d5893 AE |
5452 | rbdc = NULL; /* rbd_dev now owns this */ |
5453 | spec = NULL; /* rbd_dev now owns this */ | |
d147543d | 5454 | rbd_opts = NULL; /* rbd_dev now owns this */ |
602adf40 | 5455 | |
1f3ef788 | 5456 | rc = rbd_dev_image_probe(rbd_dev, true); |
a30b71b9 | 5457 | if (rc < 0) |
c53d5893 | 5458 | goto err_out_rbd_dev; |
05fd6f6f | 5459 | |
7ce4eef7 AE |
5460 | /* If we are mapping a snapshot it must be marked read-only */ |
5461 | ||
d147543d | 5462 | read_only = rbd_dev->opts->read_only; |
7ce4eef7 AE |
5463 | if (rbd_dev->spec->snap_id != CEPH_NOSNAP) |
5464 | read_only = true; | |
5465 | rbd_dev->mapping.read_only = read_only; | |
5466 | ||
b536f69a | 5467 | rc = rbd_dev_device_setup(rbd_dev); |
3abef3b3 | 5468 | if (rc) { |
e37180c0 ID |
5469 | /* |
5470 | * rbd_dev_header_unwatch_sync() can't be moved into | |
5471 | * rbd_dev_image_release() without refactoring, see | |
5472 | * commit 1f3ef78861ac. | |
5473 | */ | |
5474 | rbd_dev_header_unwatch_sync(rbd_dev); | |
3abef3b3 AE |
5475 | rbd_dev_image_release(rbd_dev); |
5476 | goto err_out_module; | |
5477 | } | |
5478 | ||
5479 | return count; | |
b536f69a | 5480 | |
c53d5893 AE |
5481 | err_out_rbd_dev: |
5482 | rbd_dev_destroy(rbd_dev); | |
bd4ba655 | 5483 | err_out_client: |
9d3997fd | 5484 | rbd_put_client(rbdc); |
0ddebc0c | 5485 | err_out_args: |
859c31df | 5486 | rbd_spec_put(spec); |
d147543d | 5487 | kfree(rbd_opts); |
bd4ba655 AE |
5488 | err_out_module: |
5489 | module_put(THIS_MODULE); | |
27cc2594 | 5490 | |
602adf40 | 5491 | dout("Error adding device %s\n", buf); |
27cc2594 | 5492 | |
c0cd10db | 5493 | return (ssize_t)rc; |
602adf40 YS |
5494 | } |
5495 | ||
9b60e70b ID |
5496 | static ssize_t rbd_add(struct bus_type *bus, |
5497 | const char *buf, | |
5498 | size_t count) | |
5499 | { | |
5500 | if (single_major) | |
5501 | return -EINVAL; | |
5502 | ||
5503 | return do_rbd_add(bus, buf, count); | |
5504 | } | |
5505 | ||
5506 | static ssize_t rbd_add_single_major(struct bus_type *bus, | |
5507 | const char *buf, | |
5508 | size_t count) | |
5509 | { | |
5510 | return do_rbd_add(bus, buf, count); | |
5511 | } | |
5512 | ||
200a6a8b | 5513 | static void rbd_dev_device_release(struct device *dev) |
602adf40 | 5514 | { |
593a9e7b | 5515 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
602adf40 | 5516 | |
602adf40 | 5517 | rbd_free_disk(rbd_dev); |
200a6a8b | 5518 | clear_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags); |
6d80b130 | 5519 | rbd_dev_mapping_clear(rbd_dev); |
9b60e70b ID |
5520 | if (!single_major) |
5521 | unregister_blkdev(rbd_dev->major, rbd_dev->name); | |
e2839308 | 5522 | rbd_dev_id_put(rbd_dev); |
d1cf5788 | 5523 | rbd_dev_mapping_clear(rbd_dev); |
602adf40 YS |
5524 | } |
5525 | ||
05a46afd AE |
5526 | static void rbd_dev_remove_parent(struct rbd_device *rbd_dev) |
5527 | { | |
ad945fc1 | 5528 | while (rbd_dev->parent) { |
05a46afd AE |
5529 | struct rbd_device *first = rbd_dev; |
5530 | struct rbd_device *second = first->parent; | |
5531 | struct rbd_device *third; | |
5532 | ||
5533 | /* | |
5534 | * Follow to the parent with no grandparent and | |
5535 | * remove it. | |
5536 | */ | |
5537 | while (second && (third = second->parent)) { | |
5538 | first = second; | |
5539 | second = third; | |
5540 | } | |
ad945fc1 | 5541 | rbd_assert(second); |
8ad42cd0 | 5542 | rbd_dev_image_release(second); |
ad945fc1 AE |
5543 | first->parent = NULL; |
5544 | first->parent_overlap = 0; | |
5545 | ||
5546 | rbd_assert(first->parent_spec); | |
05a46afd AE |
5547 | rbd_spec_put(first->parent_spec); |
5548 | first->parent_spec = NULL; | |
05a46afd AE |
5549 | } |
5550 | } | |
5551 | ||
9b60e70b ID |
5552 | static ssize_t do_rbd_remove(struct bus_type *bus, |
5553 | const char *buf, | |
5554 | size_t count) | |
602adf40 YS |
5555 | { |
5556 | struct rbd_device *rbd_dev = NULL; | |
751cc0e3 AE |
5557 | struct list_head *tmp; |
5558 | int dev_id; | |
602adf40 | 5559 | unsigned long ul; |
82a442d2 | 5560 | bool already = false; |
0d8189e1 | 5561 | int ret; |
602adf40 | 5562 | |
bb8e0e84 | 5563 | ret = kstrtoul(buf, 10, &ul); |
0d8189e1 AE |
5564 | if (ret) |
5565 | return ret; | |
602adf40 YS |
5566 | |
5567 | /* convert to int; abort if we lost anything in the conversion */ | |
751cc0e3 AE |
5568 | dev_id = (int)ul; |
5569 | if (dev_id != ul) | |
602adf40 YS |
5570 | return -EINVAL; |
5571 | ||
751cc0e3 AE |
5572 | ret = -ENOENT; |
5573 | spin_lock(&rbd_dev_list_lock); | |
5574 | list_for_each(tmp, &rbd_dev_list) { | |
5575 | rbd_dev = list_entry(tmp, struct rbd_device, node); | |
5576 | if (rbd_dev->dev_id == dev_id) { | |
5577 | ret = 0; | |
5578 | break; | |
5579 | } | |
42382b70 | 5580 | } |
751cc0e3 AE |
5581 | if (!ret) { |
5582 | spin_lock_irq(&rbd_dev->lock); | |
5583 | if (rbd_dev->open_count) | |
5584 | ret = -EBUSY; | |
5585 | else | |
82a442d2 AE |
5586 | already = test_and_set_bit(RBD_DEV_FLAG_REMOVING, |
5587 | &rbd_dev->flags); | |
751cc0e3 AE |
5588 | spin_unlock_irq(&rbd_dev->lock); |
5589 | } | |
5590 | spin_unlock(&rbd_dev_list_lock); | |
82a442d2 | 5591 | if (ret < 0 || already) |
1ba0f1e7 | 5592 | return ret; |
751cc0e3 | 5593 | |
fca27065 | 5594 | rbd_dev_header_unwatch_sync(rbd_dev); |
9abc5990 JD |
5595 | /* |
5596 | * flush remaining watch callbacks - these must be complete | |
5597 | * before the osd_client is shutdown | |
5598 | */ | |
5599 | dout("%s: flushing notifies", __func__); | |
5600 | ceph_osdc_flush_notifies(&rbd_dev->rbd_client->client->osdc); | |
fca27065 | 5601 | |
9875201e JD |
5602 | /* |
5603 | * Don't free anything from rbd_dev->disk until after all | |
5604 | * notifies are completely processed. Otherwise | |
5605 | * rbd_bus_del_dev() will race with rbd_watch_cb(), resulting | |
5606 | * in a potential use after free of rbd_dev->disk or rbd_dev. | |
5607 | */ | |
5608 | rbd_bus_del_dev(rbd_dev); | |
8ad42cd0 | 5609 | rbd_dev_image_release(rbd_dev); |
79ab7558 | 5610 | module_put(THIS_MODULE); |
aafb230e | 5611 | |
1ba0f1e7 | 5612 | return count; |
602adf40 YS |
5613 | } |
5614 | ||
9b60e70b ID |
5615 | static ssize_t rbd_remove(struct bus_type *bus, |
5616 | const char *buf, | |
5617 | size_t count) | |
5618 | { | |
5619 | if (single_major) | |
5620 | return -EINVAL; | |
5621 | ||
5622 | return do_rbd_remove(bus, buf, count); | |
5623 | } | |
5624 | ||
5625 | static ssize_t rbd_remove_single_major(struct bus_type *bus, | |
5626 | const char *buf, | |
5627 | size_t count) | |
5628 | { | |
5629 | return do_rbd_remove(bus, buf, count); | |
5630 | } | |
5631 | ||
602adf40 YS |
5632 | /* |
5633 | * create control files in sysfs | |
dfc5606d | 5634 | * /sys/bus/rbd/... |
602adf40 YS |
5635 | */ |
5636 | static int rbd_sysfs_init(void) | |
5637 | { | |
dfc5606d | 5638 | int ret; |
602adf40 | 5639 | |
fed4c143 | 5640 | ret = device_register(&rbd_root_dev); |
21079786 | 5641 | if (ret < 0) |
dfc5606d | 5642 | return ret; |
602adf40 | 5643 | |
fed4c143 AE |
5644 | ret = bus_register(&rbd_bus_type); |
5645 | if (ret < 0) | |
5646 | device_unregister(&rbd_root_dev); | |
602adf40 | 5647 | |
602adf40 YS |
5648 | return ret; |
5649 | } | |
5650 | ||
5651 | static void rbd_sysfs_cleanup(void) | |
5652 | { | |
dfc5606d | 5653 | bus_unregister(&rbd_bus_type); |
fed4c143 | 5654 | device_unregister(&rbd_root_dev); |
602adf40 YS |
5655 | } |
5656 | ||
1c2a9dfe AE |
5657 | static int rbd_slab_init(void) |
5658 | { | |
5659 | rbd_assert(!rbd_img_request_cache); | |
5660 | rbd_img_request_cache = kmem_cache_create("rbd_img_request", | |
5661 | sizeof (struct rbd_img_request), | |
5662 | __alignof__(struct rbd_img_request), | |
5663 | 0, NULL); | |
868311b1 AE |
5664 | if (!rbd_img_request_cache) |
5665 | return -ENOMEM; | |
5666 | ||
5667 | rbd_assert(!rbd_obj_request_cache); | |
5668 | rbd_obj_request_cache = kmem_cache_create("rbd_obj_request", | |
5669 | sizeof (struct rbd_obj_request), | |
5670 | __alignof__(struct rbd_obj_request), | |
5671 | 0, NULL); | |
78c2a44a AE |
5672 | if (!rbd_obj_request_cache) |
5673 | goto out_err; | |
5674 | ||
5675 | rbd_assert(!rbd_segment_name_cache); | |
5676 | rbd_segment_name_cache = kmem_cache_create("rbd_segment_name", | |
2d0ebc5d | 5677 | CEPH_MAX_OID_NAME_LEN + 1, 1, 0, NULL); |
78c2a44a | 5678 | if (rbd_segment_name_cache) |
1c2a9dfe | 5679 | return 0; |
78c2a44a AE |
5680 | out_err: |
5681 | if (rbd_obj_request_cache) { | |
5682 | kmem_cache_destroy(rbd_obj_request_cache); | |
5683 | rbd_obj_request_cache = NULL; | |
5684 | } | |
1c2a9dfe | 5685 | |
868311b1 AE |
5686 | kmem_cache_destroy(rbd_img_request_cache); |
5687 | rbd_img_request_cache = NULL; | |
5688 | ||
1c2a9dfe AE |
5689 | return -ENOMEM; |
5690 | } | |
5691 | ||
5692 | static void rbd_slab_exit(void) | |
5693 | { | |
78c2a44a AE |
5694 | rbd_assert(rbd_segment_name_cache); |
5695 | kmem_cache_destroy(rbd_segment_name_cache); | |
5696 | rbd_segment_name_cache = NULL; | |
5697 | ||
868311b1 AE |
5698 | rbd_assert(rbd_obj_request_cache); |
5699 | kmem_cache_destroy(rbd_obj_request_cache); | |
5700 | rbd_obj_request_cache = NULL; | |
5701 | ||
1c2a9dfe AE |
5702 | rbd_assert(rbd_img_request_cache); |
5703 | kmem_cache_destroy(rbd_img_request_cache); | |
5704 | rbd_img_request_cache = NULL; | |
5705 | } | |
5706 | ||
cc344fa1 | 5707 | static int __init rbd_init(void) |
602adf40 YS |
5708 | { |
5709 | int rc; | |
5710 | ||
1e32d34c AE |
5711 | if (!libceph_compatible(NULL)) { |
5712 | rbd_warn(NULL, "libceph incompatibility (quitting)"); | |
1e32d34c AE |
5713 | return -EINVAL; |
5714 | } | |
e1b4d96d | 5715 | |
1c2a9dfe | 5716 | rc = rbd_slab_init(); |
602adf40 YS |
5717 | if (rc) |
5718 | return rc; | |
e1b4d96d | 5719 | |
f5ee37bd ID |
5720 | /* |
5721 | * The number of active work items is limited by the number of | |
f77303bd | 5722 | * rbd devices * queue depth, so leave @max_active at default. |
f5ee37bd ID |
5723 | */ |
5724 | rbd_wq = alloc_workqueue(RBD_DRV_NAME, WQ_MEM_RECLAIM, 0); | |
5725 | if (!rbd_wq) { | |
5726 | rc = -ENOMEM; | |
5727 | goto err_out_slab; | |
5728 | } | |
5729 | ||
9b60e70b ID |
5730 | if (single_major) { |
5731 | rbd_major = register_blkdev(0, RBD_DRV_NAME); | |
5732 | if (rbd_major < 0) { | |
5733 | rc = rbd_major; | |
f5ee37bd | 5734 | goto err_out_wq; |
9b60e70b ID |
5735 | } |
5736 | } | |
5737 | ||
1c2a9dfe AE |
5738 | rc = rbd_sysfs_init(); |
5739 | if (rc) | |
9b60e70b ID |
5740 | goto err_out_blkdev; |
5741 | ||
5742 | if (single_major) | |
5743 | pr_info("loaded (major %d)\n", rbd_major); | |
5744 | else | |
5745 | pr_info("loaded\n"); | |
1c2a9dfe | 5746 | |
e1b4d96d ID |
5747 | return 0; |
5748 | ||
9b60e70b ID |
5749 | err_out_blkdev: |
5750 | if (single_major) | |
5751 | unregister_blkdev(rbd_major, RBD_DRV_NAME); | |
f5ee37bd ID |
5752 | err_out_wq: |
5753 | destroy_workqueue(rbd_wq); | |
e1b4d96d ID |
5754 | err_out_slab: |
5755 | rbd_slab_exit(); | |
1c2a9dfe | 5756 | return rc; |
602adf40 YS |
5757 | } |
5758 | ||
cc344fa1 | 5759 | static void __exit rbd_exit(void) |
602adf40 | 5760 | { |
ffe312cf | 5761 | ida_destroy(&rbd_dev_id_ida); |
602adf40 | 5762 | rbd_sysfs_cleanup(); |
9b60e70b ID |
5763 | if (single_major) |
5764 | unregister_blkdev(rbd_major, RBD_DRV_NAME); | |
f5ee37bd | 5765 | destroy_workqueue(rbd_wq); |
1c2a9dfe | 5766 | rbd_slab_exit(); |
602adf40 YS |
5767 | } |
5768 | ||
5769 | module_init(rbd_init); | |
5770 | module_exit(rbd_exit); | |
5771 | ||
d552c619 | 5772 | MODULE_AUTHOR("Alex Elder <elder@inktank.com>"); |
602adf40 YS |
5773 | MODULE_AUTHOR("Sage Weil <sage@newdream.net>"); |
5774 | MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>"); | |
602adf40 YS |
5775 | /* following authorship retained from original osdblk.c */ |
5776 | MODULE_AUTHOR("Jeff Garzik <jeff@garzik.org>"); | |
5777 | ||
90da258b | 5778 | MODULE_DESCRIPTION("RADOS Block Device (RBD) driver"); |
602adf40 | 5779 | MODULE_LICENSE("GPL"); |