Commit | Line | Data |
---|---|---|
602adf40 YS |
1 | /* |
2 | rbd.c -- Export ceph rados objects as a Linux block device | |
3 | ||
4 | ||
5 | based on drivers/block/osdblk.c: | |
6 | ||
7 | Copyright 2009 Red Hat, Inc. | |
8 | ||
9 | This program is free software; you can redistribute it and/or modify | |
10 | it under the terms of the GNU General Public License as published by | |
11 | the Free Software Foundation. | |
12 | ||
13 | This program is distributed in the hope that it will be useful, | |
14 | but WITHOUT ANY WARRANTY; without even the implied warranty of | |
15 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
16 | GNU General Public License for more details. | |
17 | ||
18 | You should have received a copy of the GNU General Public License | |
19 | along with this program; see the file COPYING. If not, write to | |
20 | the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA. | |
21 | ||
22 | ||
23 | ||
dfc5606d | 24 | For usage instructions, please refer to: |
602adf40 | 25 | |
dfc5606d | 26 | Documentation/ABI/testing/sysfs-bus-rbd |
602adf40 YS |
27 | |
28 | */ | |
29 | ||
30 | #include <linux/ceph/libceph.h> | |
31 | #include <linux/ceph/osd_client.h> | |
32 | #include <linux/ceph/mon_client.h> | |
33 | #include <linux/ceph/decode.h> | |
59c2be1e | 34 | #include <linux/parser.h> |
602adf40 YS |
35 | |
36 | #include <linux/kernel.h> | |
37 | #include <linux/device.h> | |
38 | #include <linux/module.h> | |
39 | #include <linux/fs.h> | |
40 | #include <linux/blkdev.h> | |
41 | ||
42 | #include "rbd_types.h" | |
43 | ||
aafb230e AE |
44 | #define RBD_DEBUG /* Activate rbd_assert() calls */ |
45 | ||
593a9e7b AE |
46 | /* |
47 | * The basic unit of block I/O is a sector. It is interpreted in a | |
48 | * number of contexts in Linux (blk, bio, genhd), but the default is | |
49 | * universally 512 bytes. These symbols are just slightly more | |
50 | * meaningful than the bare numbers they represent. | |
51 | */ | |
52 | #define SECTOR_SHIFT 9 | |
53 | #define SECTOR_SIZE (1ULL << SECTOR_SHIFT) | |
54 | ||
2647ba38 | 55 | /* It might be useful to have these defined elsewhere */ |
df111be6 | 56 | |
2647ba38 AE |
57 | #define U8_MAX ((u8) (~0U)) |
58 | #define U16_MAX ((u16) (~0U)) | |
59 | #define U32_MAX ((u32) (~0U)) | |
60 | #define U64_MAX ((u64) (~0ULL)) | |
df111be6 | 61 | |
f0f8cef5 AE |
62 | #define RBD_DRV_NAME "rbd" |
63 | #define RBD_DRV_NAME_LONG "rbd (rados block device)" | |
602adf40 YS |
64 | |
65 | #define RBD_MINORS_PER_MAJOR 256 /* max minors per blkdev */ | |
66 | ||
d4b125e9 AE |
67 | #define RBD_SNAP_DEV_NAME_PREFIX "snap_" |
68 | #define RBD_MAX_SNAP_NAME_LEN \ | |
69 | (NAME_MAX - (sizeof (RBD_SNAP_DEV_NAME_PREFIX) - 1)) | |
70 | ||
35d489f9 | 71 | #define RBD_MAX_SNAP_COUNT 510 /* allows max snapc to fit in 4KB */ |
602adf40 YS |
72 | |
73 | #define RBD_SNAP_HEAD_NAME "-" | |
74 | ||
9e15b77d AE |
75 | /* This allows a single page to hold an image name sent by OSD */ |
76 | #define RBD_IMAGE_NAME_LEN_MAX (PAGE_SIZE - sizeof (__le32) - 1) | |
1e130199 | 77 | #define RBD_IMAGE_ID_LEN_MAX 64 |
9e15b77d | 78 | |
1e130199 | 79 | #define RBD_OBJ_PREFIX_LEN_MAX 64 |
589d30e0 | 80 | |
d889140c AE |
81 | /* Feature bits */ |
82 | ||
83 | #define RBD_FEATURE_LAYERING 1 | |
84 | ||
85 | /* Features supported by this (client software) implementation. */ | |
86 | ||
87 | #define RBD_FEATURES_ALL (0) | |
88 | ||
81a89793 AE |
89 | /* |
90 | * An RBD device name will be "rbd#", where the "rbd" comes from | |
91 | * RBD_DRV_NAME above, and # is a unique integer identifier. | |
92 | * MAX_INT_FORMAT_WIDTH is used in ensuring DEV_NAME_LEN is big | |
93 | * enough to hold all possible device names. | |
94 | */ | |
602adf40 | 95 | #define DEV_NAME_LEN 32 |
81a89793 | 96 | #define MAX_INT_FORMAT_WIDTH ((5 * sizeof (int)) / 2 + 1) |
602adf40 YS |
97 | |
98 | /* | |
99 | * block device image metadata (in-memory version) | |
100 | */ | |
101 | struct rbd_image_header { | |
f84344f3 | 102 | /* These four fields never change for a given rbd image */ |
849b4260 | 103 | char *object_prefix; |
34b13184 | 104 | u64 features; |
602adf40 YS |
105 | __u8 obj_order; |
106 | __u8 crypt_type; | |
107 | __u8 comp_type; | |
602adf40 | 108 | |
f84344f3 AE |
109 | /* The remaining fields need to be updated occasionally */ |
110 | u64 image_size; | |
111 | struct ceph_snap_context *snapc; | |
602adf40 YS |
112 | char *snap_names; |
113 | u64 *snap_sizes; | |
59c2be1e YS |
114 | |
115 | u64 obj_version; | |
116 | }; | |
117 | ||
0d7dbfce AE |
118 | /* |
119 | * An rbd image specification. | |
120 | * | |
121 | * The tuple (pool_id, image_id, snap_id) is sufficient to uniquely | |
c66c6e0c AE |
122 | * identify an image. Each rbd_dev structure includes a pointer to |
123 | * an rbd_spec structure that encapsulates this identity. | |
124 | * | |
125 | * Each of the id's in an rbd_spec has an associated name. For a | |
126 | * user-mapped image, the names are supplied and the id's associated | |
127 | * with them are looked up. For a layered image, a parent image is | |
128 | * defined by the tuple, and the names are looked up. | |
129 | * | |
130 | * An rbd_dev structure contains a parent_spec pointer which is | |
131 | * non-null if the image it represents is a child in a layered | |
132 | * image. This pointer will refer to the rbd_spec structure used | |
133 | * by the parent rbd_dev for its own identity (i.e., the structure | |
134 | * is shared between the parent and child). | |
135 | * | |
136 | * Since these structures are populated once, during the discovery | |
137 | * phase of image construction, they are effectively immutable so | |
138 | * we make no effort to synchronize access to them. | |
139 | * | |
140 | * Note that code herein does not assume the image name is known (it | |
141 | * could be a null pointer). | |
0d7dbfce AE |
142 | */ |
143 | struct rbd_spec { | |
144 | u64 pool_id; | |
145 | char *pool_name; | |
146 | ||
147 | char *image_id; | |
0d7dbfce | 148 | char *image_name; |
0d7dbfce AE |
149 | |
150 | u64 snap_id; | |
151 | char *snap_name; | |
152 | ||
153 | struct kref kref; | |
154 | }; | |
155 | ||
602adf40 | 156 | /* |
f0f8cef5 | 157 | * an instance of the client. multiple devices may share an rbd client. |
602adf40 YS |
158 | */ |
159 | struct rbd_client { | |
160 | struct ceph_client *client; | |
161 | struct kref kref; | |
162 | struct list_head node; | |
163 | }; | |
164 | ||
bf0d5f50 AE |
165 | struct rbd_img_request; |
166 | typedef void (*rbd_img_callback_t)(struct rbd_img_request *); | |
167 | ||
168 | #define BAD_WHICH U32_MAX /* Good which or bad which, which? */ | |
169 | ||
170 | struct rbd_obj_request; | |
171 | typedef void (*rbd_obj_callback_t)(struct rbd_obj_request *); | |
172 | ||
9969ebc5 AE |
173 | enum obj_request_type { |
174 | OBJ_REQUEST_NODATA, OBJ_REQUEST_BIO, OBJ_REQUEST_PAGES | |
175 | }; | |
bf0d5f50 AE |
176 | |
177 | struct rbd_obj_request { | |
178 | const char *object_name; | |
179 | u64 offset; /* object start byte */ | |
180 | u64 length; /* bytes from offset */ | |
181 | ||
182 | struct rbd_img_request *img_request; | |
183 | struct list_head links; /* img_request->obj_requests */ | |
184 | u32 which; /* posn image request list */ | |
185 | ||
186 | enum obj_request_type type; | |
788e2df3 AE |
187 | union { |
188 | struct bio *bio_list; | |
189 | struct { | |
190 | struct page **pages; | |
191 | u32 page_count; | |
192 | }; | |
193 | }; | |
bf0d5f50 AE |
194 | |
195 | struct ceph_osd_request *osd_req; | |
196 | ||
197 | u64 xferred; /* bytes transferred */ | |
198 | u64 version; | |
199 | s32 result; | |
200 | atomic_t done; | |
201 | ||
202 | rbd_obj_callback_t callback; | |
788e2df3 | 203 | struct completion completion; |
bf0d5f50 AE |
204 | |
205 | struct kref kref; | |
206 | }; | |
207 | ||
208 | struct rbd_img_request { | |
209 | struct request *rq; | |
210 | struct rbd_device *rbd_dev; | |
211 | u64 offset; /* starting image byte offset */ | |
212 | u64 length; /* byte count from offset */ | |
213 | bool write_request; /* false for read */ | |
214 | union { | |
215 | struct ceph_snap_context *snapc; /* for writes */ | |
216 | u64 snap_id; /* for reads */ | |
217 | }; | |
218 | spinlock_t completion_lock;/* protects next_completion */ | |
219 | u32 next_completion; | |
220 | rbd_img_callback_t callback; | |
221 | ||
222 | u32 obj_request_count; | |
223 | struct list_head obj_requests; /* rbd_obj_request structs */ | |
224 | ||
225 | struct kref kref; | |
226 | }; | |
227 | ||
228 | #define for_each_obj_request(ireq, oreq) \ | |
ef06f4d3 | 229 | list_for_each_entry(oreq, &(ireq)->obj_requests, links) |
bf0d5f50 | 230 | #define for_each_obj_request_from(ireq, oreq) \ |
ef06f4d3 | 231 | list_for_each_entry_from(oreq, &(ireq)->obj_requests, links) |
bf0d5f50 | 232 | #define for_each_obj_request_safe(ireq, oreq, n) \ |
ef06f4d3 | 233 | list_for_each_entry_safe_reverse(oreq, n, &(ireq)->obj_requests, links) |
bf0d5f50 | 234 | |
dfc5606d YS |
235 | struct rbd_snap { |
236 | struct device dev; | |
237 | const char *name; | |
3591538f | 238 | u64 size; |
dfc5606d YS |
239 | struct list_head node; |
240 | u64 id; | |
34b13184 | 241 | u64 features; |
dfc5606d YS |
242 | }; |
243 | ||
f84344f3 | 244 | struct rbd_mapping { |
99c1f08f | 245 | u64 size; |
34b13184 | 246 | u64 features; |
f84344f3 AE |
247 | bool read_only; |
248 | }; | |
249 | ||
602adf40 YS |
250 | /* |
251 | * a single device | |
252 | */ | |
253 | struct rbd_device { | |
de71a297 | 254 | int dev_id; /* blkdev unique id */ |
602adf40 YS |
255 | |
256 | int major; /* blkdev assigned major */ | |
257 | struct gendisk *disk; /* blkdev's gendisk and rq */ | |
602adf40 | 258 | |
a30b71b9 | 259 | u32 image_format; /* Either 1 or 2 */ |
602adf40 YS |
260 | struct rbd_client *rbd_client; |
261 | ||
262 | char name[DEV_NAME_LEN]; /* blkdev name, e.g. rbd3 */ | |
263 | ||
b82d167b | 264 | spinlock_t lock; /* queue, flags, open_count */ |
602adf40 YS |
265 | |
266 | struct rbd_image_header header; | |
b82d167b | 267 | unsigned long flags; /* possibly lock protected */ |
0d7dbfce | 268 | struct rbd_spec *spec; |
602adf40 | 269 | |
0d7dbfce | 270 | char *header_name; |
971f839a | 271 | |
0903e875 AE |
272 | struct ceph_file_layout layout; |
273 | ||
59c2be1e | 274 | struct ceph_osd_event *watch_event; |
975241af | 275 | struct rbd_obj_request *watch_request; |
59c2be1e | 276 | |
86b00e0d AE |
277 | struct rbd_spec *parent_spec; |
278 | u64 parent_overlap; | |
279 | ||
c666601a JD |
280 | /* protects updating the header */ |
281 | struct rw_semaphore header_rwsem; | |
f84344f3 AE |
282 | |
283 | struct rbd_mapping mapping; | |
602adf40 YS |
284 | |
285 | struct list_head node; | |
dfc5606d YS |
286 | |
287 | /* list of snapshots */ | |
288 | struct list_head snaps; | |
289 | ||
290 | /* sysfs related */ | |
291 | struct device dev; | |
b82d167b | 292 | unsigned long open_count; /* protected by lock */ |
dfc5606d YS |
293 | }; |
294 | ||
b82d167b AE |
295 | /* |
296 | * Flag bits for rbd_dev->flags. If atomicity is required, | |
297 | * rbd_dev->lock is used to protect access. | |
298 | * | |
299 | * Currently, only the "removing" flag (which is coupled with the | |
300 | * "open_count" field) requires atomic access. | |
301 | */ | |
6d292906 AE |
302 | enum rbd_dev_flags { |
303 | RBD_DEV_FLAG_EXISTS, /* mapped snapshot has not been deleted */ | |
b82d167b | 304 | RBD_DEV_FLAG_REMOVING, /* this mapping is being removed */ |
6d292906 AE |
305 | }; |
306 | ||
602adf40 | 307 | static DEFINE_MUTEX(ctl_mutex); /* Serialize open/close/setup/teardown */ |
e124a82f | 308 | |
602adf40 | 309 | static LIST_HEAD(rbd_dev_list); /* devices */ |
e124a82f AE |
310 | static DEFINE_SPINLOCK(rbd_dev_list_lock); |
311 | ||
432b8587 AE |
312 | static LIST_HEAD(rbd_client_list); /* clients */ |
313 | static DEFINE_SPINLOCK(rbd_client_list_lock); | |
602adf40 | 314 | |
304f6808 AE |
315 | static int rbd_dev_snaps_update(struct rbd_device *rbd_dev); |
316 | static int rbd_dev_snaps_register(struct rbd_device *rbd_dev); | |
317 | ||
dfc5606d | 318 | static void rbd_dev_release(struct device *dev); |
41f38c2b | 319 | static void rbd_remove_snap_dev(struct rbd_snap *snap); |
dfc5606d | 320 | |
f0f8cef5 AE |
321 | static ssize_t rbd_add(struct bus_type *bus, const char *buf, |
322 | size_t count); | |
323 | static ssize_t rbd_remove(struct bus_type *bus, const char *buf, | |
324 | size_t count); | |
325 | ||
326 | static struct bus_attribute rbd_bus_attrs[] = { | |
327 | __ATTR(add, S_IWUSR, NULL, rbd_add), | |
328 | __ATTR(remove, S_IWUSR, NULL, rbd_remove), | |
329 | __ATTR_NULL | |
330 | }; | |
331 | ||
332 | static struct bus_type rbd_bus_type = { | |
333 | .name = "rbd", | |
334 | .bus_attrs = rbd_bus_attrs, | |
335 | }; | |
336 | ||
337 | static void rbd_root_dev_release(struct device *dev) | |
338 | { | |
339 | } | |
340 | ||
341 | static struct device rbd_root_dev = { | |
342 | .init_name = "rbd", | |
343 | .release = rbd_root_dev_release, | |
344 | }; | |
345 | ||
06ecc6cb AE |
346 | static __printf(2, 3) |
347 | void rbd_warn(struct rbd_device *rbd_dev, const char *fmt, ...) | |
348 | { | |
349 | struct va_format vaf; | |
350 | va_list args; | |
351 | ||
352 | va_start(args, fmt); | |
353 | vaf.fmt = fmt; | |
354 | vaf.va = &args; | |
355 | ||
356 | if (!rbd_dev) | |
357 | printk(KERN_WARNING "%s: %pV\n", RBD_DRV_NAME, &vaf); | |
358 | else if (rbd_dev->disk) | |
359 | printk(KERN_WARNING "%s: %s: %pV\n", | |
360 | RBD_DRV_NAME, rbd_dev->disk->disk_name, &vaf); | |
361 | else if (rbd_dev->spec && rbd_dev->spec->image_name) | |
362 | printk(KERN_WARNING "%s: image %s: %pV\n", | |
363 | RBD_DRV_NAME, rbd_dev->spec->image_name, &vaf); | |
364 | else if (rbd_dev->spec && rbd_dev->spec->image_id) | |
365 | printk(KERN_WARNING "%s: id %s: %pV\n", | |
366 | RBD_DRV_NAME, rbd_dev->spec->image_id, &vaf); | |
367 | else /* punt */ | |
368 | printk(KERN_WARNING "%s: rbd_dev %p: %pV\n", | |
369 | RBD_DRV_NAME, rbd_dev, &vaf); | |
370 | va_end(args); | |
371 | } | |
372 | ||
aafb230e AE |
373 | #ifdef RBD_DEBUG |
374 | #define rbd_assert(expr) \ | |
375 | if (unlikely(!(expr))) { \ | |
376 | printk(KERN_ERR "\nAssertion failure in %s() " \ | |
377 | "at line %d:\n\n" \ | |
378 | "\trbd_assert(%s);\n\n", \ | |
379 | __func__, __LINE__, #expr); \ | |
380 | BUG(); \ | |
381 | } | |
382 | #else /* !RBD_DEBUG */ | |
383 | # define rbd_assert(expr) ((void) 0) | |
384 | #endif /* !RBD_DEBUG */ | |
dfc5606d | 385 | |
117973fb AE |
386 | static int rbd_dev_refresh(struct rbd_device *rbd_dev, u64 *hver); |
387 | static int rbd_dev_v2_refresh(struct rbd_device *rbd_dev, u64 *hver); | |
59c2be1e | 388 | |
602adf40 YS |
389 | static int rbd_open(struct block_device *bdev, fmode_t mode) |
390 | { | |
f0f8cef5 | 391 | struct rbd_device *rbd_dev = bdev->bd_disk->private_data; |
b82d167b | 392 | bool removing = false; |
602adf40 | 393 | |
f84344f3 | 394 | if ((mode & FMODE_WRITE) && rbd_dev->mapping.read_only) |
602adf40 YS |
395 | return -EROFS; |
396 | ||
a14ea269 | 397 | spin_lock_irq(&rbd_dev->lock); |
b82d167b AE |
398 | if (test_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags)) |
399 | removing = true; | |
400 | else | |
401 | rbd_dev->open_count++; | |
a14ea269 | 402 | spin_unlock_irq(&rbd_dev->lock); |
b82d167b AE |
403 | if (removing) |
404 | return -ENOENT; | |
405 | ||
42382b70 | 406 | mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING); |
c3e946ce | 407 | (void) get_device(&rbd_dev->dev); |
f84344f3 | 408 | set_device_ro(bdev, rbd_dev->mapping.read_only); |
42382b70 | 409 | mutex_unlock(&ctl_mutex); |
340c7a2b | 410 | |
602adf40 YS |
411 | return 0; |
412 | } | |
413 | ||
dfc5606d YS |
414 | static int rbd_release(struct gendisk *disk, fmode_t mode) |
415 | { | |
416 | struct rbd_device *rbd_dev = disk->private_data; | |
b82d167b AE |
417 | unsigned long open_count_before; |
418 | ||
a14ea269 | 419 | spin_lock_irq(&rbd_dev->lock); |
b82d167b | 420 | open_count_before = rbd_dev->open_count--; |
a14ea269 | 421 | spin_unlock_irq(&rbd_dev->lock); |
b82d167b | 422 | rbd_assert(open_count_before > 0); |
dfc5606d | 423 | |
42382b70 | 424 | mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING); |
c3e946ce | 425 | put_device(&rbd_dev->dev); |
42382b70 | 426 | mutex_unlock(&ctl_mutex); |
dfc5606d YS |
427 | |
428 | return 0; | |
429 | } | |
430 | ||
602adf40 YS |
431 | static const struct block_device_operations rbd_bd_ops = { |
432 | .owner = THIS_MODULE, | |
433 | .open = rbd_open, | |
dfc5606d | 434 | .release = rbd_release, |
602adf40 YS |
435 | }; |
436 | ||
437 | /* | |
438 | * Initialize an rbd client instance. | |
43ae4701 | 439 | * We own *ceph_opts. |
602adf40 | 440 | */ |
f8c38929 | 441 | static struct rbd_client *rbd_client_create(struct ceph_options *ceph_opts) |
602adf40 YS |
442 | { |
443 | struct rbd_client *rbdc; | |
444 | int ret = -ENOMEM; | |
445 | ||
37206ee5 | 446 | dout("%s:\n", __func__); |
602adf40 YS |
447 | rbdc = kmalloc(sizeof(struct rbd_client), GFP_KERNEL); |
448 | if (!rbdc) | |
449 | goto out_opt; | |
450 | ||
451 | kref_init(&rbdc->kref); | |
452 | INIT_LIST_HEAD(&rbdc->node); | |
453 | ||
bc534d86 AE |
454 | mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING); |
455 | ||
43ae4701 | 456 | rbdc->client = ceph_create_client(ceph_opts, rbdc, 0, 0); |
602adf40 | 457 | if (IS_ERR(rbdc->client)) |
bc534d86 | 458 | goto out_mutex; |
43ae4701 | 459 | ceph_opts = NULL; /* Now rbdc->client is responsible for ceph_opts */ |
602adf40 YS |
460 | |
461 | ret = ceph_open_session(rbdc->client); | |
462 | if (ret < 0) | |
463 | goto out_err; | |
464 | ||
432b8587 | 465 | spin_lock(&rbd_client_list_lock); |
602adf40 | 466 | list_add_tail(&rbdc->node, &rbd_client_list); |
432b8587 | 467 | spin_unlock(&rbd_client_list_lock); |
602adf40 | 468 | |
bc534d86 | 469 | mutex_unlock(&ctl_mutex); |
37206ee5 | 470 | dout("%s: rbdc %p\n", __func__, rbdc); |
bc534d86 | 471 | |
602adf40 YS |
472 | return rbdc; |
473 | ||
474 | out_err: | |
475 | ceph_destroy_client(rbdc->client); | |
bc534d86 AE |
476 | out_mutex: |
477 | mutex_unlock(&ctl_mutex); | |
602adf40 YS |
478 | kfree(rbdc); |
479 | out_opt: | |
43ae4701 AE |
480 | if (ceph_opts) |
481 | ceph_destroy_options(ceph_opts); | |
37206ee5 AE |
482 | dout("%s: error %d\n", __func__, ret); |
483 | ||
28f259b7 | 484 | return ERR_PTR(ret); |
602adf40 YS |
485 | } |
486 | ||
487 | /* | |
1f7ba331 AE |
488 | * Find a ceph client with specific addr and configuration. If |
489 | * found, bump its reference count. | |
602adf40 | 490 | */ |
1f7ba331 | 491 | static struct rbd_client *rbd_client_find(struct ceph_options *ceph_opts) |
602adf40 YS |
492 | { |
493 | struct rbd_client *client_node; | |
1f7ba331 | 494 | bool found = false; |
602adf40 | 495 | |
43ae4701 | 496 | if (ceph_opts->flags & CEPH_OPT_NOSHARE) |
602adf40 YS |
497 | return NULL; |
498 | ||
1f7ba331 AE |
499 | spin_lock(&rbd_client_list_lock); |
500 | list_for_each_entry(client_node, &rbd_client_list, node) { | |
501 | if (!ceph_compare_options(ceph_opts, client_node->client)) { | |
502 | kref_get(&client_node->kref); | |
503 | found = true; | |
504 | break; | |
505 | } | |
506 | } | |
507 | spin_unlock(&rbd_client_list_lock); | |
508 | ||
509 | return found ? client_node : NULL; | |
602adf40 YS |
510 | } |
511 | ||
59c2be1e YS |
512 | /* |
513 | * mount options | |
514 | */ | |
515 | enum { | |
59c2be1e YS |
516 | Opt_last_int, |
517 | /* int args above */ | |
518 | Opt_last_string, | |
519 | /* string args above */ | |
cc0538b6 AE |
520 | Opt_read_only, |
521 | Opt_read_write, | |
522 | /* Boolean args above */ | |
523 | Opt_last_bool, | |
59c2be1e YS |
524 | }; |
525 | ||
43ae4701 | 526 | static match_table_t rbd_opts_tokens = { |
59c2be1e YS |
527 | /* int args above */ |
528 | /* string args above */ | |
be466c1c | 529 | {Opt_read_only, "read_only"}, |
cc0538b6 AE |
530 | {Opt_read_only, "ro"}, /* Alternate spelling */ |
531 | {Opt_read_write, "read_write"}, | |
532 | {Opt_read_write, "rw"}, /* Alternate spelling */ | |
533 | /* Boolean args above */ | |
59c2be1e YS |
534 | {-1, NULL} |
535 | }; | |
536 | ||
98571b5a AE |
537 | struct rbd_options { |
538 | bool read_only; | |
539 | }; | |
540 | ||
541 | #define RBD_READ_ONLY_DEFAULT false | |
542 | ||
59c2be1e YS |
543 | static int parse_rbd_opts_token(char *c, void *private) |
544 | { | |
43ae4701 | 545 | struct rbd_options *rbd_opts = private; |
59c2be1e YS |
546 | substring_t argstr[MAX_OPT_ARGS]; |
547 | int token, intval, ret; | |
548 | ||
43ae4701 | 549 | token = match_token(c, rbd_opts_tokens, argstr); |
59c2be1e YS |
550 | if (token < 0) |
551 | return -EINVAL; | |
552 | ||
553 | if (token < Opt_last_int) { | |
554 | ret = match_int(&argstr[0], &intval); | |
555 | if (ret < 0) { | |
556 | pr_err("bad mount option arg (not int) " | |
557 | "at '%s'\n", c); | |
558 | return ret; | |
559 | } | |
560 | dout("got int token %d val %d\n", token, intval); | |
561 | } else if (token > Opt_last_int && token < Opt_last_string) { | |
562 | dout("got string token %d val %s\n", token, | |
563 | argstr[0].from); | |
cc0538b6 AE |
564 | } else if (token > Opt_last_string && token < Opt_last_bool) { |
565 | dout("got Boolean token %d\n", token); | |
59c2be1e YS |
566 | } else { |
567 | dout("got token %d\n", token); | |
568 | } | |
569 | ||
570 | switch (token) { | |
cc0538b6 AE |
571 | case Opt_read_only: |
572 | rbd_opts->read_only = true; | |
573 | break; | |
574 | case Opt_read_write: | |
575 | rbd_opts->read_only = false; | |
576 | break; | |
59c2be1e | 577 | default: |
aafb230e AE |
578 | rbd_assert(false); |
579 | break; | |
59c2be1e YS |
580 | } |
581 | return 0; | |
582 | } | |
583 | ||
602adf40 YS |
584 | /* |
585 | * Get a ceph client with specific addr and configuration, if one does | |
586 | * not exist create it. | |
587 | */ | |
9d3997fd | 588 | static struct rbd_client *rbd_get_client(struct ceph_options *ceph_opts) |
602adf40 | 589 | { |
f8c38929 | 590 | struct rbd_client *rbdc; |
59c2be1e | 591 | |
1f7ba331 | 592 | rbdc = rbd_client_find(ceph_opts); |
9d3997fd | 593 | if (rbdc) /* using an existing client */ |
43ae4701 | 594 | ceph_destroy_options(ceph_opts); |
9d3997fd | 595 | else |
f8c38929 | 596 | rbdc = rbd_client_create(ceph_opts); |
602adf40 | 597 | |
9d3997fd | 598 | return rbdc; |
602adf40 YS |
599 | } |
600 | ||
601 | /* | |
602 | * Destroy ceph client | |
d23a4b3f | 603 | * |
432b8587 | 604 | * Caller must hold rbd_client_list_lock. |
602adf40 YS |
605 | */ |
606 | static void rbd_client_release(struct kref *kref) | |
607 | { | |
608 | struct rbd_client *rbdc = container_of(kref, struct rbd_client, kref); | |
609 | ||
37206ee5 | 610 | dout("%s: rbdc %p\n", __func__, rbdc); |
cd9d9f5d | 611 | spin_lock(&rbd_client_list_lock); |
602adf40 | 612 | list_del(&rbdc->node); |
cd9d9f5d | 613 | spin_unlock(&rbd_client_list_lock); |
602adf40 YS |
614 | |
615 | ceph_destroy_client(rbdc->client); | |
616 | kfree(rbdc); | |
617 | } | |
618 | ||
619 | /* | |
620 | * Drop reference to ceph client node. If it's not referenced anymore, release | |
621 | * it. | |
622 | */ | |
9d3997fd | 623 | static void rbd_put_client(struct rbd_client *rbdc) |
602adf40 | 624 | { |
c53d5893 AE |
625 | if (rbdc) |
626 | kref_put(&rbdc->kref, rbd_client_release); | |
602adf40 YS |
627 | } |
628 | ||
a30b71b9 AE |
629 | static bool rbd_image_format_valid(u32 image_format) |
630 | { | |
631 | return image_format == 1 || image_format == 2; | |
632 | } | |
633 | ||
8e94af8e AE |
634 | static bool rbd_dev_ondisk_valid(struct rbd_image_header_ondisk *ondisk) |
635 | { | |
103a150f AE |
636 | size_t size; |
637 | u32 snap_count; | |
638 | ||
639 | /* The header has to start with the magic rbd header text */ | |
640 | if (memcmp(&ondisk->text, RBD_HEADER_TEXT, sizeof (RBD_HEADER_TEXT))) | |
641 | return false; | |
642 | ||
db2388b6 AE |
643 | /* The bio layer requires at least sector-sized I/O */ |
644 | ||
645 | if (ondisk->options.order < SECTOR_SHIFT) | |
646 | return false; | |
647 | ||
648 | /* If we use u64 in a few spots we may be able to loosen this */ | |
649 | ||
650 | if (ondisk->options.order > 8 * sizeof (int) - 1) | |
651 | return false; | |
652 | ||
103a150f AE |
653 | /* |
654 | * The size of a snapshot header has to fit in a size_t, and | |
655 | * that limits the number of snapshots. | |
656 | */ | |
657 | snap_count = le32_to_cpu(ondisk->snap_count); | |
658 | size = SIZE_MAX - sizeof (struct ceph_snap_context); | |
659 | if (snap_count > size / sizeof (__le64)) | |
660 | return false; | |
661 | ||
662 | /* | |
663 | * Not only that, but the size of the entire the snapshot | |
664 | * header must also be representable in a size_t. | |
665 | */ | |
666 | size -= snap_count * sizeof (__le64); | |
667 | if ((u64) size < le64_to_cpu(ondisk->snap_names_len)) | |
668 | return false; | |
669 | ||
670 | return true; | |
8e94af8e AE |
671 | } |
672 | ||
602adf40 YS |
673 | /* |
674 | * Create a new header structure, translate header format from the on-disk | |
675 | * header. | |
676 | */ | |
677 | static int rbd_header_from_disk(struct rbd_image_header *header, | |
4156d998 | 678 | struct rbd_image_header_ondisk *ondisk) |
602adf40 | 679 | { |
ccece235 | 680 | u32 snap_count; |
58c17b0e | 681 | size_t len; |
d2bb24e5 | 682 | size_t size; |
621901d6 | 683 | u32 i; |
602adf40 | 684 | |
6a52325f AE |
685 | memset(header, 0, sizeof (*header)); |
686 | ||
103a150f AE |
687 | snap_count = le32_to_cpu(ondisk->snap_count); |
688 | ||
58c17b0e AE |
689 | len = strnlen(ondisk->object_prefix, sizeof (ondisk->object_prefix)); |
690 | header->object_prefix = kmalloc(len + 1, GFP_KERNEL); | |
6a52325f | 691 | if (!header->object_prefix) |
602adf40 | 692 | return -ENOMEM; |
58c17b0e AE |
693 | memcpy(header->object_prefix, ondisk->object_prefix, len); |
694 | header->object_prefix[len] = '\0'; | |
00f1f36f | 695 | |
602adf40 | 696 | if (snap_count) { |
f785cc1d AE |
697 | u64 snap_names_len = le64_to_cpu(ondisk->snap_names_len); |
698 | ||
621901d6 AE |
699 | /* Save a copy of the snapshot names */ |
700 | ||
f785cc1d AE |
701 | if (snap_names_len > (u64) SIZE_MAX) |
702 | return -EIO; | |
703 | header->snap_names = kmalloc(snap_names_len, GFP_KERNEL); | |
602adf40 | 704 | if (!header->snap_names) |
6a52325f | 705 | goto out_err; |
f785cc1d AE |
706 | /* |
707 | * Note that rbd_dev_v1_header_read() guarantees | |
708 | * the ondisk buffer we're working with has | |
709 | * snap_names_len bytes beyond the end of the | |
710 | * snapshot id array, this memcpy() is safe. | |
711 | */ | |
712 | memcpy(header->snap_names, &ondisk->snaps[snap_count], | |
713 | snap_names_len); | |
6a52325f | 714 | |
621901d6 AE |
715 | /* Record each snapshot's size */ |
716 | ||
d2bb24e5 AE |
717 | size = snap_count * sizeof (*header->snap_sizes); |
718 | header->snap_sizes = kmalloc(size, GFP_KERNEL); | |
602adf40 | 719 | if (!header->snap_sizes) |
6a52325f | 720 | goto out_err; |
621901d6 AE |
721 | for (i = 0; i < snap_count; i++) |
722 | header->snap_sizes[i] = | |
723 | le64_to_cpu(ondisk->snaps[i].image_size); | |
602adf40 | 724 | } else { |
ccece235 | 725 | WARN_ON(ondisk->snap_names_len); |
602adf40 YS |
726 | header->snap_names = NULL; |
727 | header->snap_sizes = NULL; | |
728 | } | |
849b4260 | 729 | |
34b13184 | 730 | header->features = 0; /* No features support in v1 images */ |
602adf40 YS |
731 | header->obj_order = ondisk->options.order; |
732 | header->crypt_type = ondisk->options.crypt_type; | |
733 | header->comp_type = ondisk->options.comp_type; | |
6a52325f | 734 | |
621901d6 AE |
735 | /* Allocate and fill in the snapshot context */ |
736 | ||
f84344f3 | 737 | header->image_size = le64_to_cpu(ondisk->image_size); |
6a52325f AE |
738 | size = sizeof (struct ceph_snap_context); |
739 | size += snap_count * sizeof (header->snapc->snaps[0]); | |
740 | header->snapc = kzalloc(size, GFP_KERNEL); | |
741 | if (!header->snapc) | |
742 | goto out_err; | |
602adf40 YS |
743 | |
744 | atomic_set(&header->snapc->nref, 1); | |
505cbb9b | 745 | header->snapc->seq = le64_to_cpu(ondisk->snap_seq); |
602adf40 | 746 | header->snapc->num_snaps = snap_count; |
621901d6 AE |
747 | for (i = 0; i < snap_count; i++) |
748 | header->snapc->snaps[i] = | |
749 | le64_to_cpu(ondisk->snaps[i].id); | |
602adf40 YS |
750 | |
751 | return 0; | |
752 | ||
6a52325f | 753 | out_err: |
849b4260 | 754 | kfree(header->snap_sizes); |
ccece235 | 755 | header->snap_sizes = NULL; |
602adf40 | 756 | kfree(header->snap_names); |
ccece235 | 757 | header->snap_names = NULL; |
6a52325f AE |
758 | kfree(header->object_prefix); |
759 | header->object_prefix = NULL; | |
ccece235 | 760 | |
00f1f36f | 761 | return -ENOMEM; |
602adf40 YS |
762 | } |
763 | ||
9e15b77d AE |
764 | static const char *rbd_snap_name(struct rbd_device *rbd_dev, u64 snap_id) |
765 | { | |
766 | struct rbd_snap *snap; | |
767 | ||
768 | if (snap_id == CEPH_NOSNAP) | |
769 | return RBD_SNAP_HEAD_NAME; | |
770 | ||
771 | list_for_each_entry(snap, &rbd_dev->snaps, node) | |
772 | if (snap_id == snap->id) | |
773 | return snap->name; | |
774 | ||
775 | return NULL; | |
776 | } | |
777 | ||
8836b995 | 778 | static int snap_by_name(struct rbd_device *rbd_dev, const char *snap_name) |
602adf40 | 779 | { |
602adf40 | 780 | |
e86924a8 | 781 | struct rbd_snap *snap; |
602adf40 | 782 | |
e86924a8 AE |
783 | list_for_each_entry(snap, &rbd_dev->snaps, node) { |
784 | if (!strcmp(snap_name, snap->name)) { | |
0d7dbfce | 785 | rbd_dev->spec->snap_id = snap->id; |
e86924a8 | 786 | rbd_dev->mapping.size = snap->size; |
34b13184 | 787 | rbd_dev->mapping.features = snap->features; |
602adf40 | 788 | |
e86924a8 | 789 | return 0; |
00f1f36f | 790 | } |
00f1f36f | 791 | } |
e86924a8 | 792 | |
00f1f36f | 793 | return -ENOENT; |
602adf40 YS |
794 | } |
795 | ||
819d52bf | 796 | static int rbd_dev_set_mapping(struct rbd_device *rbd_dev) |
602adf40 | 797 | { |
78dc447d | 798 | int ret; |
602adf40 | 799 | |
0d7dbfce | 800 | if (!memcmp(rbd_dev->spec->snap_name, RBD_SNAP_HEAD_NAME, |
cc9d734c | 801 | sizeof (RBD_SNAP_HEAD_NAME))) { |
0d7dbfce | 802 | rbd_dev->spec->snap_id = CEPH_NOSNAP; |
99c1f08f | 803 | rbd_dev->mapping.size = rbd_dev->header.image_size; |
34b13184 | 804 | rbd_dev->mapping.features = rbd_dev->header.features; |
e86924a8 | 805 | ret = 0; |
602adf40 | 806 | } else { |
0d7dbfce | 807 | ret = snap_by_name(rbd_dev, rbd_dev->spec->snap_name); |
602adf40 YS |
808 | if (ret < 0) |
809 | goto done; | |
f84344f3 | 810 | rbd_dev->mapping.read_only = true; |
602adf40 | 811 | } |
6d292906 AE |
812 | set_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags); |
813 | ||
602adf40 | 814 | done: |
602adf40 YS |
815 | return ret; |
816 | } | |
817 | ||
818 | static void rbd_header_free(struct rbd_image_header *header) | |
819 | { | |
849b4260 | 820 | kfree(header->object_prefix); |
d78fd7ae | 821 | header->object_prefix = NULL; |
602adf40 | 822 | kfree(header->snap_sizes); |
d78fd7ae | 823 | header->snap_sizes = NULL; |
849b4260 | 824 | kfree(header->snap_names); |
d78fd7ae | 825 | header->snap_names = NULL; |
d1d25646 | 826 | ceph_put_snap_context(header->snapc); |
d78fd7ae | 827 | header->snapc = NULL; |
602adf40 YS |
828 | } |
829 | ||
98571b5a | 830 | static const char *rbd_segment_name(struct rbd_device *rbd_dev, u64 offset) |
602adf40 | 831 | { |
65ccfe21 AE |
832 | char *name; |
833 | u64 segment; | |
834 | int ret; | |
602adf40 | 835 | |
2fd82b9e | 836 | name = kmalloc(MAX_OBJ_NAME_SIZE + 1, GFP_NOIO); |
65ccfe21 AE |
837 | if (!name) |
838 | return NULL; | |
839 | segment = offset >> rbd_dev->header.obj_order; | |
2fd82b9e | 840 | ret = snprintf(name, MAX_OBJ_NAME_SIZE + 1, "%s.%012llx", |
65ccfe21 | 841 | rbd_dev->header.object_prefix, segment); |
2fd82b9e | 842 | if (ret < 0 || ret > MAX_OBJ_NAME_SIZE) { |
65ccfe21 AE |
843 | pr_err("error formatting segment name for #%llu (%d)\n", |
844 | segment, ret); | |
845 | kfree(name); | |
846 | name = NULL; | |
847 | } | |
602adf40 | 848 | |
65ccfe21 AE |
849 | return name; |
850 | } | |
602adf40 | 851 | |
65ccfe21 AE |
852 | static u64 rbd_segment_offset(struct rbd_device *rbd_dev, u64 offset) |
853 | { | |
854 | u64 segment_size = (u64) 1 << rbd_dev->header.obj_order; | |
602adf40 | 855 | |
65ccfe21 AE |
856 | return offset & (segment_size - 1); |
857 | } | |
858 | ||
859 | static u64 rbd_segment_length(struct rbd_device *rbd_dev, | |
860 | u64 offset, u64 length) | |
861 | { | |
862 | u64 segment_size = (u64) 1 << rbd_dev->header.obj_order; | |
863 | ||
864 | offset &= segment_size - 1; | |
865 | ||
aafb230e | 866 | rbd_assert(length <= U64_MAX - offset); |
65ccfe21 AE |
867 | if (offset + length > segment_size) |
868 | length = segment_size - offset; | |
869 | ||
870 | return length; | |
602adf40 YS |
871 | } |
872 | ||
029bcbd8 JD |
873 | /* |
874 | * returns the size of an object in the image | |
875 | */ | |
876 | static u64 rbd_obj_bytes(struct rbd_image_header *header) | |
877 | { | |
878 | return 1 << header->obj_order; | |
879 | } | |
880 | ||
602adf40 YS |
881 | /* |
882 | * bio helpers | |
883 | */ | |
884 | ||
885 | static void bio_chain_put(struct bio *chain) | |
886 | { | |
887 | struct bio *tmp; | |
888 | ||
889 | while (chain) { | |
890 | tmp = chain; | |
891 | chain = chain->bi_next; | |
892 | bio_put(tmp); | |
893 | } | |
894 | } | |
895 | ||
896 | /* | |
897 | * zeros a bio chain, starting at specific offset | |
898 | */ | |
899 | static void zero_bio_chain(struct bio *chain, int start_ofs) | |
900 | { | |
901 | struct bio_vec *bv; | |
902 | unsigned long flags; | |
903 | void *buf; | |
904 | int i; | |
905 | int pos = 0; | |
906 | ||
907 | while (chain) { | |
908 | bio_for_each_segment(bv, chain, i) { | |
909 | if (pos + bv->bv_len > start_ofs) { | |
910 | int remainder = max(start_ofs - pos, 0); | |
911 | buf = bvec_kmap_irq(bv, &flags); | |
912 | memset(buf + remainder, 0, | |
913 | bv->bv_len - remainder); | |
85b5aaa6 | 914 | bvec_kunmap_irq(buf, &flags); |
602adf40 YS |
915 | } |
916 | pos += bv->bv_len; | |
917 | } | |
918 | ||
919 | chain = chain->bi_next; | |
920 | } | |
921 | } | |
922 | ||
923 | /* | |
f7760dad AE |
924 | * Clone a portion of a bio, starting at the given byte offset |
925 | * and continuing for the number of bytes indicated. | |
602adf40 | 926 | */ |
f7760dad AE |
927 | static struct bio *bio_clone_range(struct bio *bio_src, |
928 | unsigned int offset, | |
929 | unsigned int len, | |
930 | gfp_t gfpmask) | |
602adf40 | 931 | { |
f7760dad AE |
932 | struct bio_vec *bv; |
933 | unsigned int resid; | |
934 | unsigned short idx; | |
935 | unsigned int voff; | |
936 | unsigned short end_idx; | |
937 | unsigned short vcnt; | |
938 | struct bio *bio; | |
939 | ||
940 | /* Handle the easy case for the caller */ | |
941 | ||
942 | if (!offset && len == bio_src->bi_size) | |
943 | return bio_clone(bio_src, gfpmask); | |
944 | ||
945 | if (WARN_ON_ONCE(!len)) | |
946 | return NULL; | |
947 | if (WARN_ON_ONCE(len > bio_src->bi_size)) | |
948 | return NULL; | |
949 | if (WARN_ON_ONCE(offset > bio_src->bi_size - len)) | |
950 | return NULL; | |
951 | ||
952 | /* Find first affected segment... */ | |
953 | ||
954 | resid = offset; | |
955 | __bio_for_each_segment(bv, bio_src, idx, 0) { | |
956 | if (resid < bv->bv_len) | |
957 | break; | |
958 | resid -= bv->bv_len; | |
602adf40 | 959 | } |
f7760dad | 960 | voff = resid; |
602adf40 | 961 | |
f7760dad | 962 | /* ...and the last affected segment */ |
602adf40 | 963 | |
f7760dad AE |
964 | resid += len; |
965 | __bio_for_each_segment(bv, bio_src, end_idx, idx) { | |
966 | if (resid <= bv->bv_len) | |
967 | break; | |
968 | resid -= bv->bv_len; | |
969 | } | |
970 | vcnt = end_idx - idx + 1; | |
971 | ||
972 | /* Build the clone */ | |
973 | ||
974 | bio = bio_alloc(gfpmask, (unsigned int) vcnt); | |
975 | if (!bio) | |
976 | return NULL; /* ENOMEM */ | |
602adf40 | 977 | |
f7760dad AE |
978 | bio->bi_bdev = bio_src->bi_bdev; |
979 | bio->bi_sector = bio_src->bi_sector + (offset >> SECTOR_SHIFT); | |
980 | bio->bi_rw = bio_src->bi_rw; | |
981 | bio->bi_flags |= 1 << BIO_CLONED; | |
982 | ||
983 | /* | |
984 | * Copy over our part of the bio_vec, then update the first | |
985 | * and last (or only) entries. | |
986 | */ | |
987 | memcpy(&bio->bi_io_vec[0], &bio_src->bi_io_vec[idx], | |
988 | vcnt * sizeof (struct bio_vec)); | |
989 | bio->bi_io_vec[0].bv_offset += voff; | |
990 | if (vcnt > 1) { | |
991 | bio->bi_io_vec[0].bv_len -= voff; | |
992 | bio->bi_io_vec[vcnt - 1].bv_len = resid; | |
993 | } else { | |
994 | bio->bi_io_vec[0].bv_len = len; | |
602adf40 YS |
995 | } |
996 | ||
f7760dad AE |
997 | bio->bi_vcnt = vcnt; |
998 | bio->bi_size = len; | |
999 | bio->bi_idx = 0; | |
1000 | ||
1001 | return bio; | |
1002 | } | |
1003 | ||
1004 | /* | |
1005 | * Clone a portion of a bio chain, starting at the given byte offset | |
1006 | * into the first bio in the source chain and continuing for the | |
1007 | * number of bytes indicated. The result is another bio chain of | |
1008 | * exactly the given length, or a null pointer on error. | |
1009 | * | |
1010 | * The bio_src and offset parameters are both in-out. On entry they | |
1011 | * refer to the first source bio and the offset into that bio where | |
1012 | * the start of data to be cloned is located. | |
1013 | * | |
1014 | * On return, bio_src is updated to refer to the bio in the source | |
1015 | * chain that contains first un-cloned byte, and *offset will | |
1016 | * contain the offset of that byte within that bio. | |
1017 | */ | |
1018 | static struct bio *bio_chain_clone_range(struct bio **bio_src, | |
1019 | unsigned int *offset, | |
1020 | unsigned int len, | |
1021 | gfp_t gfpmask) | |
1022 | { | |
1023 | struct bio *bi = *bio_src; | |
1024 | unsigned int off = *offset; | |
1025 | struct bio *chain = NULL; | |
1026 | struct bio **end; | |
1027 | ||
1028 | /* Build up a chain of clone bios up to the limit */ | |
1029 | ||
1030 | if (!bi || off >= bi->bi_size || !len) | |
1031 | return NULL; /* Nothing to clone */ | |
602adf40 | 1032 | |
f7760dad AE |
1033 | end = &chain; |
1034 | while (len) { | |
1035 | unsigned int bi_size; | |
1036 | struct bio *bio; | |
1037 | ||
f5400b7a AE |
1038 | if (!bi) { |
1039 | rbd_warn(NULL, "bio_chain exhausted with %u left", len); | |
f7760dad | 1040 | goto out_err; /* EINVAL; ran out of bio's */ |
f5400b7a | 1041 | } |
f7760dad AE |
1042 | bi_size = min_t(unsigned int, bi->bi_size - off, len); |
1043 | bio = bio_clone_range(bi, off, bi_size, gfpmask); | |
1044 | if (!bio) | |
1045 | goto out_err; /* ENOMEM */ | |
1046 | ||
1047 | *end = bio; | |
1048 | end = &bio->bi_next; | |
602adf40 | 1049 | |
f7760dad AE |
1050 | off += bi_size; |
1051 | if (off == bi->bi_size) { | |
1052 | bi = bi->bi_next; | |
1053 | off = 0; | |
1054 | } | |
1055 | len -= bi_size; | |
1056 | } | |
1057 | *bio_src = bi; | |
1058 | *offset = off; | |
1059 | ||
1060 | return chain; | |
1061 | out_err: | |
1062 | bio_chain_put(chain); | |
602adf40 | 1063 | |
602adf40 YS |
1064 | return NULL; |
1065 | } | |
1066 | ||
bf0d5f50 AE |
1067 | static void rbd_obj_request_get(struct rbd_obj_request *obj_request) |
1068 | { | |
37206ee5 AE |
1069 | dout("%s: obj %p (was %d)\n", __func__, obj_request, |
1070 | atomic_read(&obj_request->kref.refcount)); | |
bf0d5f50 AE |
1071 | kref_get(&obj_request->kref); |
1072 | } | |
1073 | ||
1074 | static void rbd_obj_request_destroy(struct kref *kref); | |
1075 | static void rbd_obj_request_put(struct rbd_obj_request *obj_request) | |
1076 | { | |
1077 | rbd_assert(obj_request != NULL); | |
37206ee5 AE |
1078 | dout("%s: obj %p (was %d)\n", __func__, obj_request, |
1079 | atomic_read(&obj_request->kref.refcount)); | |
bf0d5f50 AE |
1080 | kref_put(&obj_request->kref, rbd_obj_request_destroy); |
1081 | } | |
1082 | ||
1083 | static void rbd_img_request_get(struct rbd_img_request *img_request) | |
1084 | { | |
37206ee5 AE |
1085 | dout("%s: img %p (was %d)\n", __func__, img_request, |
1086 | atomic_read(&img_request->kref.refcount)); | |
bf0d5f50 AE |
1087 | kref_get(&img_request->kref); |
1088 | } | |
1089 | ||
1090 | static void rbd_img_request_destroy(struct kref *kref); | |
1091 | static void rbd_img_request_put(struct rbd_img_request *img_request) | |
1092 | { | |
1093 | rbd_assert(img_request != NULL); | |
37206ee5 AE |
1094 | dout("%s: img %p (was %d)\n", __func__, img_request, |
1095 | atomic_read(&img_request->kref.refcount)); | |
bf0d5f50 AE |
1096 | kref_put(&img_request->kref, rbd_img_request_destroy); |
1097 | } | |
1098 | ||
1099 | static inline void rbd_img_obj_request_add(struct rbd_img_request *img_request, | |
1100 | struct rbd_obj_request *obj_request) | |
1101 | { | |
25dcf954 AE |
1102 | rbd_assert(obj_request->img_request == NULL); |
1103 | ||
bf0d5f50 AE |
1104 | rbd_obj_request_get(obj_request); |
1105 | obj_request->img_request = img_request; | |
25dcf954 | 1106 | obj_request->which = img_request->obj_request_count; |
bf0d5f50 | 1107 | rbd_assert(obj_request->which != BAD_WHICH); |
25dcf954 AE |
1108 | img_request->obj_request_count++; |
1109 | list_add_tail(&obj_request->links, &img_request->obj_requests); | |
37206ee5 AE |
1110 | dout("%s: img %p obj %p w=%u\n", __func__, img_request, obj_request, |
1111 | obj_request->which); | |
bf0d5f50 AE |
1112 | } |
1113 | ||
1114 | static inline void rbd_img_obj_request_del(struct rbd_img_request *img_request, | |
1115 | struct rbd_obj_request *obj_request) | |
1116 | { | |
1117 | rbd_assert(obj_request->which != BAD_WHICH); | |
25dcf954 | 1118 | |
37206ee5 AE |
1119 | dout("%s: img %p obj %p w=%u\n", __func__, img_request, obj_request, |
1120 | obj_request->which); | |
bf0d5f50 | 1121 | list_del(&obj_request->links); |
25dcf954 AE |
1122 | rbd_assert(img_request->obj_request_count > 0); |
1123 | img_request->obj_request_count--; | |
1124 | rbd_assert(obj_request->which == img_request->obj_request_count); | |
1125 | obj_request->which = BAD_WHICH; | |
bf0d5f50 | 1126 | rbd_assert(obj_request->img_request == img_request); |
bf0d5f50 | 1127 | obj_request->img_request = NULL; |
25dcf954 | 1128 | obj_request->callback = NULL; |
bf0d5f50 AE |
1129 | rbd_obj_request_put(obj_request); |
1130 | } | |
1131 | ||
1132 | static bool obj_request_type_valid(enum obj_request_type type) | |
1133 | { | |
1134 | switch (type) { | |
9969ebc5 | 1135 | case OBJ_REQUEST_NODATA: |
bf0d5f50 | 1136 | case OBJ_REQUEST_BIO: |
788e2df3 | 1137 | case OBJ_REQUEST_PAGES: |
bf0d5f50 AE |
1138 | return true; |
1139 | default: | |
1140 | return false; | |
1141 | } | |
1142 | } | |
1143 | ||
8d23bf29 AE |
1144 | struct ceph_osd_req_op *rbd_osd_req_op_create(u16 opcode, ...) |
1145 | { | |
1146 | struct ceph_osd_req_op *op; | |
1147 | va_list args; | |
2647ba38 | 1148 | size_t size; |
8d23bf29 AE |
1149 | |
1150 | op = kzalloc(sizeof (*op), GFP_NOIO); | |
1151 | if (!op) | |
1152 | return NULL; | |
1153 | op->op = opcode; | |
1154 | va_start(args, opcode); | |
1155 | switch (opcode) { | |
1156 | case CEPH_OSD_OP_READ: | |
1157 | case CEPH_OSD_OP_WRITE: | |
1158 | /* rbd_osd_req_op_create(READ, offset, length) */ | |
1159 | /* rbd_osd_req_op_create(WRITE, offset, length) */ | |
1160 | op->extent.offset = va_arg(args, u64); | |
1161 | op->extent.length = va_arg(args, u64); | |
1162 | if (opcode == CEPH_OSD_OP_WRITE) | |
1163 | op->payload_len = op->extent.length; | |
1164 | break; | |
fbfab539 AE |
1165 | case CEPH_OSD_OP_STAT: |
1166 | break; | |
2647ba38 AE |
1167 | case CEPH_OSD_OP_CALL: |
1168 | /* rbd_osd_req_op_create(CALL, class, method, data, datalen) */ | |
1169 | op->cls.class_name = va_arg(args, char *); | |
1170 | size = strlen(op->cls.class_name); | |
1171 | rbd_assert(size <= (size_t) U8_MAX); | |
1172 | op->cls.class_len = size; | |
1173 | op->payload_len = size; | |
1174 | ||
1175 | op->cls.method_name = va_arg(args, char *); | |
1176 | size = strlen(op->cls.method_name); | |
1177 | rbd_assert(size <= (size_t) U8_MAX); | |
1178 | op->cls.method_len = size; | |
1179 | op->payload_len += size; | |
1180 | ||
1181 | op->cls.argc = 0; | |
1182 | op->cls.indata = va_arg(args, void *); | |
1183 | size = va_arg(args, size_t); | |
1184 | rbd_assert(size <= (size_t) U32_MAX); | |
1185 | op->cls.indata_len = (u32) size; | |
1186 | op->payload_len += size; | |
1187 | break; | |
5efea49a AE |
1188 | case CEPH_OSD_OP_NOTIFY_ACK: |
1189 | case CEPH_OSD_OP_WATCH: | |
1190 | /* rbd_osd_req_op_create(NOTIFY_ACK, cookie, version) */ | |
1191 | /* rbd_osd_req_op_create(WATCH, cookie, version, flag) */ | |
1192 | op->watch.cookie = va_arg(args, u64); | |
1193 | op->watch.ver = va_arg(args, u64); | |
1194 | op->watch.ver = cpu_to_le64(op->watch.ver); | |
1195 | if (opcode == CEPH_OSD_OP_WATCH && va_arg(args, int)) | |
1196 | op->watch.flag = (u8) 1; | |
1197 | break; | |
8d23bf29 AE |
1198 | default: |
1199 | rbd_warn(NULL, "unsupported opcode %hu\n", opcode); | |
1200 | kfree(op); | |
1201 | op = NULL; | |
1202 | break; | |
1203 | } | |
1204 | va_end(args); | |
1205 | ||
1206 | return op; | |
1207 | } | |
1208 | ||
1209 | static void rbd_osd_req_op_destroy(struct ceph_osd_req_op *op) | |
1210 | { | |
1211 | kfree(op); | |
1212 | } | |
1213 | ||
bf0d5f50 AE |
1214 | static int rbd_obj_request_submit(struct ceph_osd_client *osdc, |
1215 | struct rbd_obj_request *obj_request) | |
1216 | { | |
37206ee5 AE |
1217 | dout("%s: osdc %p obj %p\n", __func__, osdc, obj_request); |
1218 | ||
bf0d5f50 AE |
1219 | return ceph_osdc_start_request(osdc, obj_request->osd_req, false); |
1220 | } | |
1221 | ||
1222 | static void rbd_img_request_complete(struct rbd_img_request *img_request) | |
1223 | { | |
37206ee5 | 1224 | dout("%s: img %p\n", __func__, img_request); |
bf0d5f50 AE |
1225 | if (img_request->callback) |
1226 | img_request->callback(img_request); | |
1227 | else | |
1228 | rbd_img_request_put(img_request); | |
1229 | } | |
1230 | ||
788e2df3 AE |
1231 | /* Caller is responsible for rbd_obj_request_destroy(obj_request) */ |
1232 | ||
1233 | static int rbd_obj_request_wait(struct rbd_obj_request *obj_request) | |
1234 | { | |
37206ee5 AE |
1235 | dout("%s: obj %p\n", __func__, obj_request); |
1236 | ||
788e2df3 AE |
1237 | return wait_for_completion_interruptible(&obj_request->completion); |
1238 | } | |
1239 | ||
07741308 AE |
1240 | static void obj_request_done_init(struct rbd_obj_request *obj_request) |
1241 | { | |
1242 | atomic_set(&obj_request->done, 0); | |
1243 | smp_wmb(); | |
1244 | } | |
1245 | ||
1246 | static void obj_request_done_set(struct rbd_obj_request *obj_request) | |
1247 | { | |
632b88ca AE |
1248 | int done; |
1249 | ||
1250 | done = atomic_inc_return(&obj_request->done); | |
1251 | if (done > 1) { | |
1252 | struct rbd_img_request *img_request = obj_request->img_request; | |
1253 | struct rbd_device *rbd_dev; | |
1254 | ||
1255 | rbd_dev = img_request ? img_request->rbd_dev : NULL; | |
1256 | rbd_warn(rbd_dev, "obj_request %p was already done\n", | |
1257 | obj_request); | |
1258 | } | |
07741308 AE |
1259 | } |
1260 | ||
1261 | static bool obj_request_done_test(struct rbd_obj_request *obj_request) | |
1262 | { | |
632b88ca | 1263 | smp_mb(); |
07741308 AE |
1264 | return atomic_read(&obj_request->done) != 0; |
1265 | } | |
1266 | ||
9969ebc5 AE |
1267 | static void rbd_osd_trivial_callback(struct rbd_obj_request *obj_request, |
1268 | struct ceph_osd_op *op) | |
1269 | { | |
37206ee5 | 1270 | dout("%s: obj %p\n", __func__, obj_request); |
07741308 | 1271 | obj_request_done_set(obj_request); |
9969ebc5 AE |
1272 | } |
1273 | ||
bf0d5f50 AE |
1274 | static void rbd_obj_request_complete(struct rbd_obj_request *obj_request) |
1275 | { | |
37206ee5 AE |
1276 | dout("%s: obj %p cb %p\n", __func__, obj_request, |
1277 | obj_request->callback); | |
bf0d5f50 AE |
1278 | if (obj_request->callback) |
1279 | obj_request->callback(obj_request); | |
788e2df3 AE |
1280 | else |
1281 | complete_all(&obj_request->completion); | |
bf0d5f50 AE |
1282 | } |
1283 | ||
bf0d5f50 AE |
1284 | static void rbd_osd_read_callback(struct rbd_obj_request *obj_request, |
1285 | struct ceph_osd_op *op) | |
1286 | { | |
1287 | u64 xferred; | |
1288 | ||
1289 | /* | |
1290 | * We support a 64-bit length, but ultimately it has to be | |
1291 | * passed to blk_end_request(), which takes an unsigned int. | |
1292 | */ | |
1293 | xferred = le64_to_cpu(op->extent.length); | |
1294 | rbd_assert(xferred < (u64) UINT_MAX); | |
37206ee5 AE |
1295 | dout("%s: obj %p result %d %llu/%llu\n", __func__, obj_request, |
1296 | obj_request->result, xferred, obj_request->length); | |
bf0d5f50 AE |
1297 | if (obj_request->result == (s32) -ENOENT) { |
1298 | zero_bio_chain(obj_request->bio_list, 0); | |
1299 | obj_request->result = 0; | |
1300 | } else if (xferred < obj_request->length && !obj_request->result) { | |
1301 | zero_bio_chain(obj_request->bio_list, xferred); | |
1302 | xferred = obj_request->length; | |
1303 | } | |
1304 | obj_request->xferred = xferred; | |
07741308 | 1305 | obj_request_done_set(obj_request); |
bf0d5f50 AE |
1306 | } |
1307 | ||
1308 | static void rbd_osd_write_callback(struct rbd_obj_request *obj_request, | |
1309 | struct ceph_osd_op *op) | |
1310 | { | |
37206ee5 | 1311 | |
bf0d5f50 | 1312 | obj_request->xferred = le64_to_cpu(op->extent.length); |
37206ee5 AE |
1313 | dout("%s: obj %p result %d %llu/%llu\n", __func__, obj_request, |
1314 | obj_request->result, obj_request->xferred, obj_request->length); | |
1315 | ||
1316 | /* A short write really shouldn't occur. Warn if we see one */ | |
1317 | ||
1318 | if (obj_request->xferred != obj_request->length) { | |
1319 | struct rbd_img_request *img_request = obj_request->img_request; | |
1320 | struct rbd_device *rbd_dev; | |
1321 | ||
1322 | rbd_dev = img_request ? img_request->rbd_dev : NULL; | |
1323 | rbd_warn(rbd_dev, "wrote %llu want %llu\n", | |
1324 | obj_request->xferred, obj_request->length); | |
1325 | } | |
1326 | ||
07741308 | 1327 | obj_request_done_set(obj_request); |
bf0d5f50 AE |
1328 | } |
1329 | ||
fbfab539 AE |
1330 | /* |
1331 | * For a simple stat call there's nothing to do. We'll do more if | |
1332 | * this is part of a write sequence for a layered image. | |
1333 | */ | |
1334 | static void rbd_osd_stat_callback(struct rbd_obj_request *obj_request, | |
1335 | struct ceph_osd_op *op) | |
1336 | { | |
37206ee5 | 1337 | dout("%s: obj %p\n", __func__, obj_request); |
fbfab539 AE |
1338 | obj_request_done_set(obj_request); |
1339 | } | |
1340 | ||
bf0d5f50 AE |
1341 | static void rbd_osd_req_callback(struct ceph_osd_request *osd_req, |
1342 | struct ceph_msg *msg) | |
1343 | { | |
1344 | struct rbd_obj_request *obj_request = osd_req->r_priv; | |
1345 | struct ceph_osd_reply_head *reply_head; | |
1346 | struct ceph_osd_op *op; | |
1347 | u32 num_ops; | |
1348 | u16 opcode; | |
1349 | ||
37206ee5 | 1350 | dout("%s: osd_req %p msg %p\n", __func__, osd_req, msg); |
bf0d5f50 AE |
1351 | rbd_assert(osd_req == obj_request->osd_req); |
1352 | rbd_assert(!!obj_request->img_request ^ | |
1353 | (obj_request->which == BAD_WHICH)); | |
1354 | ||
1355 | obj_request->xferred = le32_to_cpu(msg->hdr.data_len); | |
1356 | reply_head = msg->front.iov_base; | |
1357 | obj_request->result = (s32) le32_to_cpu(reply_head->result); | |
1358 | obj_request->version = le64_to_cpu(osd_req->r_reassert_version.version); | |
1359 | ||
1360 | num_ops = le32_to_cpu(reply_head->num_ops); | |
1361 | WARN_ON(num_ops != 1); /* For now */ | |
1362 | ||
1363 | op = &reply_head->ops[0]; | |
1364 | opcode = le16_to_cpu(op->op); | |
1365 | switch (opcode) { | |
1366 | case CEPH_OSD_OP_READ: | |
1367 | rbd_osd_read_callback(obj_request, op); | |
1368 | break; | |
1369 | case CEPH_OSD_OP_WRITE: | |
1370 | rbd_osd_write_callback(obj_request, op); | |
1371 | break; | |
fbfab539 AE |
1372 | case CEPH_OSD_OP_STAT: |
1373 | rbd_osd_stat_callback(obj_request, op); | |
1374 | break; | |
36be9a76 | 1375 | case CEPH_OSD_OP_CALL: |
b8d70035 | 1376 | case CEPH_OSD_OP_NOTIFY_ACK: |
9969ebc5 AE |
1377 | case CEPH_OSD_OP_WATCH: |
1378 | rbd_osd_trivial_callback(obj_request, op); | |
1379 | break; | |
bf0d5f50 AE |
1380 | default: |
1381 | rbd_warn(NULL, "%s: unsupported op %hu\n", | |
1382 | obj_request->object_name, (unsigned short) opcode); | |
1383 | break; | |
1384 | } | |
1385 | ||
07741308 | 1386 | if (obj_request_done_test(obj_request)) |
bf0d5f50 AE |
1387 | rbd_obj_request_complete(obj_request); |
1388 | } | |
1389 | ||
1390 | static struct ceph_osd_request *rbd_osd_req_create( | |
1391 | struct rbd_device *rbd_dev, | |
1392 | bool write_request, | |
1393 | struct rbd_obj_request *obj_request, | |
1394 | struct ceph_osd_req_op *op) | |
1395 | { | |
1396 | struct rbd_img_request *img_request = obj_request->img_request; | |
1397 | struct ceph_snap_context *snapc = NULL; | |
1398 | struct ceph_osd_client *osdc; | |
1399 | struct ceph_osd_request *osd_req; | |
1400 | struct timespec now; | |
1401 | struct timespec *mtime; | |
1402 | u64 snap_id = CEPH_NOSNAP; | |
1403 | u64 offset = obj_request->offset; | |
1404 | u64 length = obj_request->length; | |
1405 | ||
1406 | if (img_request) { | |
1407 | rbd_assert(img_request->write_request == write_request); | |
1408 | if (img_request->write_request) | |
1409 | snapc = img_request->snapc; | |
1410 | else | |
1411 | snap_id = img_request->snap_id; | |
1412 | } | |
1413 | ||
1414 | /* Allocate and initialize the request, for the single op */ | |
1415 | ||
1416 | osdc = &rbd_dev->rbd_client->client->osdc; | |
1417 | osd_req = ceph_osdc_alloc_request(osdc, snapc, 1, false, GFP_ATOMIC); | |
1418 | if (!osd_req) | |
1419 | return NULL; /* ENOMEM */ | |
1420 | ||
1421 | rbd_assert(obj_request_type_valid(obj_request->type)); | |
1422 | switch (obj_request->type) { | |
9969ebc5 AE |
1423 | case OBJ_REQUEST_NODATA: |
1424 | break; /* Nothing to do */ | |
bf0d5f50 AE |
1425 | case OBJ_REQUEST_BIO: |
1426 | rbd_assert(obj_request->bio_list != NULL); | |
1427 | osd_req->r_bio = obj_request->bio_list; | |
bf0d5f50 | 1428 | break; |
788e2df3 AE |
1429 | case OBJ_REQUEST_PAGES: |
1430 | osd_req->r_pages = obj_request->pages; | |
1431 | osd_req->r_num_pages = obj_request->page_count; | |
1432 | osd_req->r_page_alignment = offset & ~PAGE_MASK; | |
1433 | break; | |
bf0d5f50 AE |
1434 | } |
1435 | ||
1436 | if (write_request) { | |
1437 | osd_req->r_flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK; | |
1438 | now = CURRENT_TIME; | |
1439 | mtime = &now; | |
1440 | } else { | |
1441 | osd_req->r_flags = CEPH_OSD_FLAG_READ; | |
1442 | mtime = NULL; /* not needed for reads */ | |
1443 | offset = 0; /* These are not used... */ | |
1444 | length = 0; /* ...for osd read requests */ | |
1445 | } | |
1446 | ||
1447 | osd_req->r_callback = rbd_osd_req_callback; | |
1448 | osd_req->r_priv = obj_request; | |
1449 | ||
1450 | osd_req->r_oid_len = strlen(obj_request->object_name); | |
1451 | rbd_assert(osd_req->r_oid_len < sizeof (osd_req->r_oid)); | |
1452 | memcpy(osd_req->r_oid, obj_request->object_name, osd_req->r_oid_len); | |
1453 | ||
1454 | osd_req->r_file_layout = rbd_dev->layout; /* struct */ | |
1455 | ||
1456 | /* osd_req will get its own reference to snapc (if non-null) */ | |
1457 | ||
1458 | ceph_osdc_build_request(osd_req, offset, length, 1, op, | |
1459 | snapc, snap_id, mtime); | |
1460 | ||
1461 | return osd_req; | |
1462 | } | |
1463 | ||
1464 | static void rbd_osd_req_destroy(struct ceph_osd_request *osd_req) | |
1465 | { | |
1466 | ceph_osdc_put_request(osd_req); | |
1467 | } | |
1468 | ||
1469 | /* object_name is assumed to be a non-null pointer and NUL-terminated */ | |
1470 | ||
1471 | static struct rbd_obj_request *rbd_obj_request_create(const char *object_name, | |
1472 | u64 offset, u64 length, | |
1473 | enum obj_request_type type) | |
1474 | { | |
1475 | struct rbd_obj_request *obj_request; | |
1476 | size_t size; | |
1477 | char *name; | |
1478 | ||
1479 | rbd_assert(obj_request_type_valid(type)); | |
1480 | ||
1481 | size = strlen(object_name) + 1; | |
1482 | obj_request = kzalloc(sizeof (*obj_request) + size, GFP_KERNEL); | |
1483 | if (!obj_request) | |
1484 | return NULL; | |
1485 | ||
1486 | name = (char *)(obj_request + 1); | |
1487 | obj_request->object_name = memcpy(name, object_name, size); | |
1488 | obj_request->offset = offset; | |
1489 | obj_request->length = length; | |
1490 | obj_request->which = BAD_WHICH; | |
1491 | obj_request->type = type; | |
1492 | INIT_LIST_HEAD(&obj_request->links); | |
07741308 | 1493 | obj_request_done_init(obj_request); |
788e2df3 | 1494 | init_completion(&obj_request->completion); |
bf0d5f50 AE |
1495 | kref_init(&obj_request->kref); |
1496 | ||
37206ee5 AE |
1497 | dout("%s: \"%s\" %llu/%llu %d -> obj %p\n", __func__, object_name, |
1498 | offset, length, (int)type, obj_request); | |
1499 | ||
bf0d5f50 AE |
1500 | return obj_request; |
1501 | } | |
1502 | ||
1503 | static void rbd_obj_request_destroy(struct kref *kref) | |
1504 | { | |
1505 | struct rbd_obj_request *obj_request; | |
1506 | ||
1507 | obj_request = container_of(kref, struct rbd_obj_request, kref); | |
1508 | ||
37206ee5 AE |
1509 | dout("%s: obj %p\n", __func__, obj_request); |
1510 | ||
bf0d5f50 AE |
1511 | rbd_assert(obj_request->img_request == NULL); |
1512 | rbd_assert(obj_request->which == BAD_WHICH); | |
1513 | ||
1514 | if (obj_request->osd_req) | |
1515 | rbd_osd_req_destroy(obj_request->osd_req); | |
1516 | ||
1517 | rbd_assert(obj_request_type_valid(obj_request->type)); | |
1518 | switch (obj_request->type) { | |
9969ebc5 AE |
1519 | case OBJ_REQUEST_NODATA: |
1520 | break; /* Nothing to do */ | |
bf0d5f50 AE |
1521 | case OBJ_REQUEST_BIO: |
1522 | if (obj_request->bio_list) | |
1523 | bio_chain_put(obj_request->bio_list); | |
1524 | break; | |
788e2df3 AE |
1525 | case OBJ_REQUEST_PAGES: |
1526 | if (obj_request->pages) | |
1527 | ceph_release_page_vector(obj_request->pages, | |
1528 | obj_request->page_count); | |
1529 | break; | |
bf0d5f50 AE |
1530 | } |
1531 | ||
1532 | kfree(obj_request); | |
1533 | } | |
1534 | ||
1535 | /* | |
1536 | * Caller is responsible for filling in the list of object requests | |
1537 | * that comprises the image request, and the Linux request pointer | |
1538 | * (if there is one). | |
1539 | */ | |
1540 | struct rbd_img_request *rbd_img_request_create(struct rbd_device *rbd_dev, | |
1541 | u64 offset, u64 length, | |
1542 | bool write_request) | |
1543 | { | |
1544 | struct rbd_img_request *img_request; | |
1545 | struct ceph_snap_context *snapc = NULL; | |
1546 | ||
1547 | img_request = kmalloc(sizeof (*img_request), GFP_ATOMIC); | |
1548 | if (!img_request) | |
1549 | return NULL; | |
1550 | ||
1551 | if (write_request) { | |
1552 | down_read(&rbd_dev->header_rwsem); | |
1553 | snapc = ceph_get_snap_context(rbd_dev->header.snapc); | |
1554 | up_read(&rbd_dev->header_rwsem); | |
1555 | if (WARN_ON(!snapc)) { | |
1556 | kfree(img_request); | |
1557 | return NULL; /* Shouldn't happen */ | |
1558 | } | |
1559 | } | |
1560 | ||
1561 | img_request->rq = NULL; | |
1562 | img_request->rbd_dev = rbd_dev; | |
1563 | img_request->offset = offset; | |
1564 | img_request->length = length; | |
1565 | img_request->write_request = write_request; | |
1566 | if (write_request) | |
1567 | img_request->snapc = snapc; | |
1568 | else | |
1569 | img_request->snap_id = rbd_dev->spec->snap_id; | |
1570 | spin_lock_init(&img_request->completion_lock); | |
1571 | img_request->next_completion = 0; | |
1572 | img_request->callback = NULL; | |
1573 | img_request->obj_request_count = 0; | |
1574 | INIT_LIST_HEAD(&img_request->obj_requests); | |
1575 | kref_init(&img_request->kref); | |
1576 | ||
1577 | rbd_img_request_get(img_request); /* Avoid a warning */ | |
1578 | rbd_img_request_put(img_request); /* TEMPORARY */ | |
1579 | ||
37206ee5 AE |
1580 | dout("%s: rbd_dev %p %s %llu/%llu -> img %p\n", __func__, rbd_dev, |
1581 | write_request ? "write" : "read", offset, length, | |
1582 | img_request); | |
1583 | ||
bf0d5f50 AE |
1584 | return img_request; |
1585 | } | |
1586 | ||
1587 | static void rbd_img_request_destroy(struct kref *kref) | |
1588 | { | |
1589 | struct rbd_img_request *img_request; | |
1590 | struct rbd_obj_request *obj_request; | |
1591 | struct rbd_obj_request *next_obj_request; | |
1592 | ||
1593 | img_request = container_of(kref, struct rbd_img_request, kref); | |
1594 | ||
37206ee5 AE |
1595 | dout("%s: img %p\n", __func__, img_request); |
1596 | ||
bf0d5f50 AE |
1597 | for_each_obj_request_safe(img_request, obj_request, next_obj_request) |
1598 | rbd_img_obj_request_del(img_request, obj_request); | |
25dcf954 | 1599 | rbd_assert(img_request->obj_request_count == 0); |
bf0d5f50 AE |
1600 | |
1601 | if (img_request->write_request) | |
1602 | ceph_put_snap_context(img_request->snapc); | |
1603 | ||
1604 | kfree(img_request); | |
1605 | } | |
1606 | ||
1607 | static int rbd_img_request_fill_bio(struct rbd_img_request *img_request, | |
1608 | struct bio *bio_list) | |
1609 | { | |
1610 | struct rbd_device *rbd_dev = img_request->rbd_dev; | |
1611 | struct rbd_obj_request *obj_request = NULL; | |
1612 | struct rbd_obj_request *next_obj_request; | |
1613 | unsigned int bio_offset; | |
1614 | u64 image_offset; | |
1615 | u64 resid; | |
1616 | u16 opcode; | |
1617 | ||
37206ee5 AE |
1618 | dout("%s: img %p bio %p\n", __func__, img_request, bio_list); |
1619 | ||
bf0d5f50 AE |
1620 | opcode = img_request->write_request ? CEPH_OSD_OP_WRITE |
1621 | : CEPH_OSD_OP_READ; | |
1622 | bio_offset = 0; | |
1623 | image_offset = img_request->offset; | |
1624 | rbd_assert(image_offset == bio_list->bi_sector << SECTOR_SHIFT); | |
1625 | resid = img_request->length; | |
4dda41d3 | 1626 | rbd_assert(resid > 0); |
bf0d5f50 AE |
1627 | while (resid) { |
1628 | const char *object_name; | |
1629 | unsigned int clone_size; | |
1630 | struct ceph_osd_req_op *op; | |
1631 | u64 offset; | |
1632 | u64 length; | |
1633 | ||
1634 | object_name = rbd_segment_name(rbd_dev, image_offset); | |
1635 | if (!object_name) | |
1636 | goto out_unwind; | |
1637 | offset = rbd_segment_offset(rbd_dev, image_offset); | |
1638 | length = rbd_segment_length(rbd_dev, image_offset, resid); | |
1639 | obj_request = rbd_obj_request_create(object_name, | |
1640 | offset, length, | |
1641 | OBJ_REQUEST_BIO); | |
1642 | kfree(object_name); /* object request has its own copy */ | |
1643 | if (!obj_request) | |
1644 | goto out_unwind; | |
1645 | ||
1646 | rbd_assert(length <= (u64) UINT_MAX); | |
1647 | clone_size = (unsigned int) length; | |
1648 | obj_request->bio_list = bio_chain_clone_range(&bio_list, | |
1649 | &bio_offset, clone_size, | |
1650 | GFP_ATOMIC); | |
1651 | if (!obj_request->bio_list) | |
1652 | goto out_partial; | |
1653 | ||
1654 | /* | |
1655 | * Build up the op to use in building the osd | |
1656 | * request. Note that the contents of the op are | |
1657 | * copied by rbd_osd_req_create(). | |
1658 | */ | |
1659 | op = rbd_osd_req_op_create(opcode, offset, length); | |
1660 | if (!op) | |
1661 | goto out_partial; | |
1662 | obj_request->osd_req = rbd_osd_req_create(rbd_dev, | |
1663 | img_request->write_request, | |
1664 | obj_request, op); | |
1665 | rbd_osd_req_op_destroy(op); | |
1666 | if (!obj_request->osd_req) | |
1667 | goto out_partial; | |
1668 | /* status and version are initially zero-filled */ | |
1669 | ||
1670 | rbd_img_obj_request_add(img_request, obj_request); | |
1671 | ||
1672 | image_offset += length; | |
1673 | resid -= length; | |
1674 | } | |
1675 | ||
1676 | return 0; | |
1677 | ||
1678 | out_partial: | |
1679 | rbd_obj_request_put(obj_request); | |
1680 | out_unwind: | |
1681 | for_each_obj_request_safe(img_request, obj_request, next_obj_request) | |
1682 | rbd_obj_request_put(obj_request); | |
1683 | ||
1684 | return -ENOMEM; | |
1685 | } | |
1686 | ||
1687 | static void rbd_img_obj_callback(struct rbd_obj_request *obj_request) | |
1688 | { | |
1689 | struct rbd_img_request *img_request; | |
1690 | u32 which = obj_request->which; | |
1691 | bool more = true; | |
1692 | ||
1693 | img_request = obj_request->img_request; | |
4dda41d3 | 1694 | |
37206ee5 | 1695 | dout("%s: img %p obj %p\n", __func__, img_request, obj_request); |
bf0d5f50 AE |
1696 | rbd_assert(img_request != NULL); |
1697 | rbd_assert(img_request->rq != NULL); | |
4dda41d3 | 1698 | rbd_assert(img_request->obj_request_count > 0); |
bf0d5f50 AE |
1699 | rbd_assert(which != BAD_WHICH); |
1700 | rbd_assert(which < img_request->obj_request_count); | |
1701 | rbd_assert(which >= img_request->next_completion); | |
1702 | ||
1703 | spin_lock_irq(&img_request->completion_lock); | |
1704 | if (which != img_request->next_completion) | |
1705 | goto out; | |
1706 | ||
1707 | for_each_obj_request_from(img_request, obj_request) { | |
1708 | unsigned int xferred; | |
1709 | int result; | |
1710 | ||
1711 | rbd_assert(more); | |
1712 | rbd_assert(which < img_request->obj_request_count); | |
1713 | ||
07741308 | 1714 | if (!obj_request_done_test(obj_request)) |
bf0d5f50 AE |
1715 | break; |
1716 | ||
1717 | rbd_assert(obj_request->xferred <= (u64) UINT_MAX); | |
1718 | xferred = (unsigned int) obj_request->xferred; | |
1719 | result = (int) obj_request->result; | |
1720 | if (result) | |
1721 | rbd_warn(NULL, "obj_request %s result %d xferred %u\n", | |
1722 | img_request->write_request ? "write" : "read", | |
1723 | result, xferred); | |
1724 | ||
1725 | more = blk_end_request(img_request->rq, result, xferred); | |
1726 | which++; | |
1727 | } | |
1728 | rbd_assert(more ^ (which == img_request->obj_request_count)); | |
1729 | img_request->next_completion = which; | |
1730 | out: | |
1731 | spin_unlock_irq(&img_request->completion_lock); | |
1732 | ||
1733 | if (!more) | |
1734 | rbd_img_request_complete(img_request); | |
1735 | } | |
1736 | ||
1737 | static int rbd_img_request_submit(struct rbd_img_request *img_request) | |
1738 | { | |
1739 | struct rbd_device *rbd_dev = img_request->rbd_dev; | |
1740 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; | |
1741 | struct rbd_obj_request *obj_request; | |
1742 | ||
37206ee5 | 1743 | dout("%s: img %p\n", __func__, img_request); |
bf0d5f50 AE |
1744 | for_each_obj_request(img_request, obj_request) { |
1745 | int ret; | |
1746 | ||
1747 | obj_request->callback = rbd_img_obj_callback; | |
1748 | ret = rbd_obj_request_submit(osdc, obj_request); | |
1749 | if (ret) | |
1750 | return ret; | |
1751 | /* | |
1752 | * The image request has its own reference to each | |
1753 | * of its object requests, so we can safely drop the | |
1754 | * initial one here. | |
1755 | */ | |
1756 | rbd_obj_request_put(obj_request); | |
1757 | } | |
1758 | ||
1759 | return 0; | |
1760 | } | |
1761 | ||
cf81b60e | 1762 | static int rbd_obj_notify_ack(struct rbd_device *rbd_dev, |
b8d70035 AE |
1763 | u64 ver, u64 notify_id) |
1764 | { | |
1765 | struct rbd_obj_request *obj_request; | |
1766 | struct ceph_osd_req_op *op; | |
1767 | struct ceph_osd_client *osdc; | |
1768 | int ret; | |
1769 | ||
1770 | obj_request = rbd_obj_request_create(rbd_dev->header_name, 0, 0, | |
1771 | OBJ_REQUEST_NODATA); | |
1772 | if (!obj_request) | |
1773 | return -ENOMEM; | |
1774 | ||
1775 | ret = -ENOMEM; | |
1776 | op = rbd_osd_req_op_create(CEPH_OSD_OP_NOTIFY_ACK, notify_id, ver); | |
1777 | if (!op) | |
1778 | goto out; | |
1779 | obj_request->osd_req = rbd_osd_req_create(rbd_dev, false, | |
1780 | obj_request, op); | |
1781 | rbd_osd_req_op_destroy(op); | |
1782 | if (!obj_request->osd_req) | |
1783 | goto out; | |
1784 | ||
1785 | osdc = &rbd_dev->rbd_client->client->osdc; | |
cf81b60e | 1786 | obj_request->callback = rbd_obj_request_put; |
b8d70035 | 1787 | ret = rbd_obj_request_submit(osdc, obj_request); |
b8d70035 | 1788 | out: |
cf81b60e AE |
1789 | if (ret) |
1790 | rbd_obj_request_put(obj_request); | |
b8d70035 AE |
1791 | |
1792 | return ret; | |
1793 | } | |
1794 | ||
1795 | static void rbd_watch_cb(u64 ver, u64 notify_id, u8 opcode, void *data) | |
1796 | { | |
1797 | struct rbd_device *rbd_dev = (struct rbd_device *)data; | |
1798 | u64 hver; | |
1799 | int rc; | |
1800 | ||
1801 | if (!rbd_dev) | |
1802 | return; | |
1803 | ||
37206ee5 | 1804 | dout("%s: \"%s\" notify_id %llu opcode %u\n", __func__, |
b8d70035 AE |
1805 | rbd_dev->header_name, (unsigned long long) notify_id, |
1806 | (unsigned int) opcode); | |
1807 | rc = rbd_dev_refresh(rbd_dev, &hver); | |
1808 | if (rc) | |
1809 | rbd_warn(rbd_dev, "got notification but failed to " | |
1810 | " update snaps: %d\n", rc); | |
1811 | ||
cf81b60e | 1812 | rbd_obj_notify_ack(rbd_dev, hver, notify_id); |
b8d70035 AE |
1813 | } |
1814 | ||
9969ebc5 AE |
1815 | /* |
1816 | * Request sync osd watch/unwatch. The value of "start" determines | |
1817 | * whether a watch request is being initiated or torn down. | |
1818 | */ | |
1819 | static int rbd_dev_header_watch_sync(struct rbd_device *rbd_dev, int start) | |
1820 | { | |
1821 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; | |
1822 | struct rbd_obj_request *obj_request; | |
1823 | struct ceph_osd_req_op *op; | |
1824 | int ret; | |
1825 | ||
1826 | rbd_assert(start ^ !!rbd_dev->watch_event); | |
1827 | rbd_assert(start ^ !!rbd_dev->watch_request); | |
1828 | ||
1829 | if (start) { | |
3c663bbd | 1830 | ret = ceph_osdc_create_event(osdc, rbd_watch_cb, rbd_dev, |
9969ebc5 AE |
1831 | &rbd_dev->watch_event); |
1832 | if (ret < 0) | |
1833 | return ret; | |
8eb87565 | 1834 | rbd_assert(rbd_dev->watch_event != NULL); |
9969ebc5 AE |
1835 | } |
1836 | ||
1837 | ret = -ENOMEM; | |
1838 | obj_request = rbd_obj_request_create(rbd_dev->header_name, 0, 0, | |
1839 | OBJ_REQUEST_NODATA); | |
1840 | if (!obj_request) | |
1841 | goto out_cancel; | |
1842 | ||
1843 | op = rbd_osd_req_op_create(CEPH_OSD_OP_WATCH, | |
1844 | rbd_dev->watch_event->cookie, | |
1845 | rbd_dev->header.obj_version, start); | |
1846 | if (!op) | |
1847 | goto out_cancel; | |
1848 | obj_request->osd_req = rbd_osd_req_create(rbd_dev, true, | |
1849 | obj_request, op); | |
1850 | rbd_osd_req_op_destroy(op); | |
1851 | if (!obj_request->osd_req) | |
1852 | goto out_cancel; | |
1853 | ||
8eb87565 | 1854 | if (start) |
975241af | 1855 | ceph_osdc_set_request_linger(osdc, obj_request->osd_req); |
8eb87565 | 1856 | else |
6977c3f9 | 1857 | ceph_osdc_unregister_linger_request(osdc, |
975241af | 1858 | rbd_dev->watch_request->osd_req); |
9969ebc5 AE |
1859 | ret = rbd_obj_request_submit(osdc, obj_request); |
1860 | if (ret) | |
1861 | goto out_cancel; | |
1862 | ret = rbd_obj_request_wait(obj_request); | |
1863 | if (ret) | |
1864 | goto out_cancel; | |
9969ebc5 AE |
1865 | ret = obj_request->result; |
1866 | if (ret) | |
1867 | goto out_cancel; | |
1868 | ||
8eb87565 AE |
1869 | /* |
1870 | * A watch request is set to linger, so the underlying osd | |
1871 | * request won't go away until we unregister it. We retain | |
1872 | * a pointer to the object request during that time (in | |
1873 | * rbd_dev->watch_request), so we'll keep a reference to | |
1874 | * it. We'll drop that reference (below) after we've | |
1875 | * unregistered it. | |
1876 | */ | |
1877 | if (start) { | |
1878 | rbd_dev->watch_request = obj_request; | |
1879 | ||
1880 | return 0; | |
1881 | } | |
1882 | ||
1883 | /* We have successfully torn down the watch request */ | |
1884 | ||
1885 | rbd_obj_request_put(rbd_dev->watch_request); | |
1886 | rbd_dev->watch_request = NULL; | |
9969ebc5 AE |
1887 | out_cancel: |
1888 | /* Cancel the event if we're tearing down, or on error */ | |
1889 | ceph_osdc_cancel_event(rbd_dev->watch_event); | |
1890 | rbd_dev->watch_event = NULL; | |
9969ebc5 AE |
1891 | if (obj_request) |
1892 | rbd_obj_request_put(obj_request); | |
1893 | ||
1894 | return ret; | |
1895 | } | |
1896 | ||
36be9a76 AE |
1897 | /* |
1898 | * Synchronous osd object method call | |
1899 | */ | |
1900 | static int rbd_obj_method_sync(struct rbd_device *rbd_dev, | |
1901 | const char *object_name, | |
1902 | const char *class_name, | |
1903 | const char *method_name, | |
1904 | const char *outbound, | |
1905 | size_t outbound_size, | |
1906 | char *inbound, | |
1907 | size_t inbound_size, | |
1908 | u64 *version) | |
1909 | { | |
1910 | struct rbd_obj_request *obj_request; | |
1911 | struct ceph_osd_client *osdc; | |
1912 | struct ceph_osd_req_op *op; | |
1913 | struct page **pages; | |
1914 | u32 page_count; | |
1915 | int ret; | |
1916 | ||
1917 | /* | |
1918 | * Method calls are ultimately read operations but they | |
1919 | * don't involve object data (so no offset or length). | |
1920 | * The result should placed into the inbound buffer | |
1921 | * provided. They also supply outbound data--parameters for | |
1922 | * the object method. Currently if this is present it will | |
1923 | * be a snapshot id. | |
1924 | */ | |
1925 | page_count = (u32) calc_pages_for(0, inbound_size); | |
1926 | pages = ceph_alloc_page_vector(page_count, GFP_KERNEL); | |
1927 | if (IS_ERR(pages)) | |
1928 | return PTR_ERR(pages); | |
1929 | ||
1930 | ret = -ENOMEM; | |
1931 | obj_request = rbd_obj_request_create(object_name, 0, 0, | |
1932 | OBJ_REQUEST_PAGES); | |
1933 | if (!obj_request) | |
1934 | goto out; | |
1935 | ||
1936 | obj_request->pages = pages; | |
1937 | obj_request->page_count = page_count; | |
1938 | ||
1939 | op = rbd_osd_req_op_create(CEPH_OSD_OP_CALL, class_name, | |
1940 | method_name, outbound, outbound_size); | |
1941 | if (!op) | |
1942 | goto out; | |
1943 | obj_request->osd_req = rbd_osd_req_create(rbd_dev, false, | |
1944 | obj_request, op); | |
1945 | rbd_osd_req_op_destroy(op); | |
1946 | if (!obj_request->osd_req) | |
1947 | goto out; | |
1948 | ||
1949 | osdc = &rbd_dev->rbd_client->client->osdc; | |
1950 | ret = rbd_obj_request_submit(osdc, obj_request); | |
1951 | if (ret) | |
1952 | goto out; | |
1953 | ret = rbd_obj_request_wait(obj_request); | |
1954 | if (ret) | |
1955 | goto out; | |
1956 | ||
1957 | ret = obj_request->result; | |
1958 | if (ret < 0) | |
1959 | goto out; | |
23ed6e13 | 1960 | ret = 0; |
903bb32e | 1961 | ceph_copy_from_page_vector(pages, inbound, 0, obj_request->xferred); |
36be9a76 AE |
1962 | if (version) |
1963 | *version = obj_request->version; | |
1964 | out: | |
1965 | if (obj_request) | |
1966 | rbd_obj_request_put(obj_request); | |
1967 | else | |
1968 | ceph_release_page_vector(pages, page_count); | |
1969 | ||
1970 | return ret; | |
1971 | } | |
1972 | ||
bf0d5f50 AE |
1973 | static void rbd_request_fn(struct request_queue *q) |
1974 | { | |
1975 | struct rbd_device *rbd_dev = q->queuedata; | |
1976 | bool read_only = rbd_dev->mapping.read_only; | |
1977 | struct request *rq; | |
1978 | int result; | |
1979 | ||
1980 | while ((rq = blk_fetch_request(q))) { | |
1981 | bool write_request = rq_data_dir(rq) == WRITE; | |
1982 | struct rbd_img_request *img_request; | |
1983 | u64 offset; | |
1984 | u64 length; | |
1985 | ||
1986 | /* Ignore any non-FS requests that filter through. */ | |
1987 | ||
1988 | if (rq->cmd_type != REQ_TYPE_FS) { | |
4dda41d3 AE |
1989 | dout("%s: non-fs request type %d\n", __func__, |
1990 | (int) rq->cmd_type); | |
1991 | __blk_end_request_all(rq, 0); | |
1992 | continue; | |
1993 | } | |
1994 | ||
1995 | /* Ignore/skip any zero-length requests */ | |
1996 | ||
1997 | offset = (u64) blk_rq_pos(rq) << SECTOR_SHIFT; | |
1998 | length = (u64) blk_rq_bytes(rq); | |
1999 | ||
2000 | if (!length) { | |
2001 | dout("%s: zero-length request\n", __func__); | |
bf0d5f50 AE |
2002 | __blk_end_request_all(rq, 0); |
2003 | continue; | |
2004 | } | |
2005 | ||
2006 | spin_unlock_irq(q->queue_lock); | |
2007 | ||
2008 | /* Disallow writes to a read-only device */ | |
2009 | ||
2010 | if (write_request) { | |
2011 | result = -EROFS; | |
2012 | if (read_only) | |
2013 | goto end_request; | |
2014 | rbd_assert(rbd_dev->spec->snap_id == CEPH_NOSNAP); | |
2015 | } | |
2016 | ||
6d292906 AE |
2017 | /* |
2018 | * Quit early if the mapped snapshot no longer | |
2019 | * exists. It's still possible the snapshot will | |
2020 | * have disappeared by the time our request arrives | |
2021 | * at the osd, but there's no sense in sending it if | |
2022 | * we already know. | |
2023 | */ | |
2024 | if (!test_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags)) { | |
bf0d5f50 AE |
2025 | dout("request for non-existent snapshot"); |
2026 | rbd_assert(rbd_dev->spec->snap_id != CEPH_NOSNAP); | |
2027 | result = -ENXIO; | |
2028 | goto end_request; | |
2029 | } | |
2030 | ||
bf0d5f50 AE |
2031 | result = -EINVAL; |
2032 | if (WARN_ON(offset && length > U64_MAX - offset + 1)) | |
2033 | goto end_request; /* Shouldn't happen */ | |
2034 | ||
2035 | result = -ENOMEM; | |
2036 | img_request = rbd_img_request_create(rbd_dev, offset, length, | |
2037 | write_request); | |
2038 | if (!img_request) | |
2039 | goto end_request; | |
2040 | ||
2041 | img_request->rq = rq; | |
2042 | ||
2043 | result = rbd_img_request_fill_bio(img_request, rq->bio); | |
2044 | if (!result) | |
2045 | result = rbd_img_request_submit(img_request); | |
2046 | if (result) | |
2047 | rbd_img_request_put(img_request); | |
2048 | end_request: | |
2049 | spin_lock_irq(q->queue_lock); | |
2050 | if (result < 0) { | |
2051 | rbd_warn(rbd_dev, "obj_request %s result %d\n", | |
2052 | write_request ? "write" : "read", result); | |
2053 | __blk_end_request_all(rq, result); | |
2054 | } | |
2055 | } | |
2056 | } | |
2057 | ||
602adf40 YS |
2058 | /* |
2059 | * a queue callback. Makes sure that we don't create a bio that spans across | |
2060 | * multiple osd objects. One exception would be with a single page bios, | |
f7760dad | 2061 | * which we handle later at bio_chain_clone_range() |
602adf40 YS |
2062 | */ |
2063 | static int rbd_merge_bvec(struct request_queue *q, struct bvec_merge_data *bmd, | |
2064 | struct bio_vec *bvec) | |
2065 | { | |
2066 | struct rbd_device *rbd_dev = q->queuedata; | |
e5cfeed2 AE |
2067 | sector_t sector_offset; |
2068 | sector_t sectors_per_obj; | |
2069 | sector_t obj_sector_offset; | |
2070 | int ret; | |
2071 | ||
2072 | /* | |
2073 | * Find how far into its rbd object the partition-relative | |
2074 | * bio start sector is to offset relative to the enclosing | |
2075 | * device. | |
2076 | */ | |
2077 | sector_offset = get_start_sect(bmd->bi_bdev) + bmd->bi_sector; | |
2078 | sectors_per_obj = 1 << (rbd_dev->header.obj_order - SECTOR_SHIFT); | |
2079 | obj_sector_offset = sector_offset & (sectors_per_obj - 1); | |
2080 | ||
2081 | /* | |
2082 | * Compute the number of bytes from that offset to the end | |
2083 | * of the object. Account for what's already used by the bio. | |
2084 | */ | |
2085 | ret = (int) (sectors_per_obj - obj_sector_offset) << SECTOR_SHIFT; | |
2086 | if (ret > bmd->bi_size) | |
2087 | ret -= bmd->bi_size; | |
2088 | else | |
2089 | ret = 0; | |
2090 | ||
2091 | /* | |
2092 | * Don't send back more than was asked for. And if the bio | |
2093 | * was empty, let the whole thing through because: "Note | |
2094 | * that a block device *must* allow a single page to be | |
2095 | * added to an empty bio." | |
2096 | */ | |
2097 | rbd_assert(bvec->bv_len <= PAGE_SIZE); | |
2098 | if (ret > (int) bvec->bv_len || !bmd->bi_size) | |
2099 | ret = (int) bvec->bv_len; | |
2100 | ||
2101 | return ret; | |
602adf40 YS |
2102 | } |
2103 | ||
2104 | static void rbd_free_disk(struct rbd_device *rbd_dev) | |
2105 | { | |
2106 | struct gendisk *disk = rbd_dev->disk; | |
2107 | ||
2108 | if (!disk) | |
2109 | return; | |
2110 | ||
602adf40 YS |
2111 | if (disk->flags & GENHD_FL_UP) |
2112 | del_gendisk(disk); | |
2113 | if (disk->queue) | |
2114 | blk_cleanup_queue(disk->queue); | |
2115 | put_disk(disk); | |
2116 | } | |
2117 | ||
788e2df3 AE |
2118 | static int rbd_obj_read_sync(struct rbd_device *rbd_dev, |
2119 | const char *object_name, | |
2120 | u64 offset, u64 length, | |
2121 | char *buf, u64 *version) | |
2122 | ||
2123 | { | |
2124 | struct ceph_osd_req_op *op; | |
2125 | struct rbd_obj_request *obj_request; | |
2126 | struct ceph_osd_client *osdc; | |
2127 | struct page **pages = NULL; | |
2128 | u32 page_count; | |
1ceae7ef | 2129 | size_t size; |
788e2df3 AE |
2130 | int ret; |
2131 | ||
2132 | page_count = (u32) calc_pages_for(offset, length); | |
2133 | pages = ceph_alloc_page_vector(page_count, GFP_KERNEL); | |
2134 | if (IS_ERR(pages)) | |
2135 | ret = PTR_ERR(pages); | |
2136 | ||
2137 | ret = -ENOMEM; | |
2138 | obj_request = rbd_obj_request_create(object_name, offset, length, | |
36be9a76 | 2139 | OBJ_REQUEST_PAGES); |
788e2df3 AE |
2140 | if (!obj_request) |
2141 | goto out; | |
2142 | ||
2143 | obj_request->pages = pages; | |
2144 | obj_request->page_count = page_count; | |
2145 | ||
2146 | op = rbd_osd_req_op_create(CEPH_OSD_OP_READ, offset, length); | |
2147 | if (!op) | |
2148 | goto out; | |
2149 | obj_request->osd_req = rbd_osd_req_create(rbd_dev, false, | |
2150 | obj_request, op); | |
2151 | rbd_osd_req_op_destroy(op); | |
2152 | if (!obj_request->osd_req) | |
2153 | goto out; | |
2154 | ||
2155 | osdc = &rbd_dev->rbd_client->client->osdc; | |
2156 | ret = rbd_obj_request_submit(osdc, obj_request); | |
2157 | if (ret) | |
2158 | goto out; | |
2159 | ret = rbd_obj_request_wait(obj_request); | |
2160 | if (ret) | |
2161 | goto out; | |
2162 | ||
2163 | ret = obj_request->result; | |
2164 | if (ret < 0) | |
2165 | goto out; | |
1ceae7ef AE |
2166 | |
2167 | rbd_assert(obj_request->xferred <= (u64) SIZE_MAX); | |
2168 | size = (size_t) obj_request->xferred; | |
903bb32e | 2169 | ceph_copy_from_page_vector(pages, buf, 0, size); |
23ed6e13 AE |
2170 | rbd_assert(size <= (size_t) INT_MAX); |
2171 | ret = (int) size; | |
788e2df3 AE |
2172 | if (version) |
2173 | *version = obj_request->version; | |
2174 | out: | |
2175 | if (obj_request) | |
2176 | rbd_obj_request_put(obj_request); | |
2177 | else | |
2178 | ceph_release_page_vector(pages, page_count); | |
2179 | ||
2180 | return ret; | |
2181 | } | |
2182 | ||
602adf40 | 2183 | /* |
4156d998 AE |
2184 | * Read the complete header for the given rbd device. |
2185 | * | |
2186 | * Returns a pointer to a dynamically-allocated buffer containing | |
2187 | * the complete and validated header. Caller can pass the address | |
2188 | * of a variable that will be filled in with the version of the | |
2189 | * header object at the time it was read. | |
2190 | * | |
2191 | * Returns a pointer-coded errno if a failure occurs. | |
602adf40 | 2192 | */ |
4156d998 AE |
2193 | static struct rbd_image_header_ondisk * |
2194 | rbd_dev_v1_header_read(struct rbd_device *rbd_dev, u64 *version) | |
602adf40 | 2195 | { |
4156d998 | 2196 | struct rbd_image_header_ondisk *ondisk = NULL; |
50f7c4c9 | 2197 | u32 snap_count = 0; |
4156d998 AE |
2198 | u64 names_size = 0; |
2199 | u32 want_count; | |
2200 | int ret; | |
602adf40 | 2201 | |
00f1f36f | 2202 | /* |
4156d998 AE |
2203 | * The complete header will include an array of its 64-bit |
2204 | * snapshot ids, followed by the names of those snapshots as | |
2205 | * a contiguous block of NUL-terminated strings. Note that | |
2206 | * the number of snapshots could change by the time we read | |
2207 | * it in, in which case we re-read it. | |
00f1f36f | 2208 | */ |
4156d998 AE |
2209 | do { |
2210 | size_t size; | |
2211 | ||
2212 | kfree(ondisk); | |
2213 | ||
2214 | size = sizeof (*ondisk); | |
2215 | size += snap_count * sizeof (struct rbd_image_snap_ondisk); | |
2216 | size += names_size; | |
2217 | ondisk = kmalloc(size, GFP_KERNEL); | |
2218 | if (!ondisk) | |
2219 | return ERR_PTR(-ENOMEM); | |
2220 | ||
788e2df3 | 2221 | ret = rbd_obj_read_sync(rbd_dev, rbd_dev->header_name, |
4156d998 AE |
2222 | 0, size, |
2223 | (char *) ondisk, version); | |
4156d998 AE |
2224 | if (ret < 0) |
2225 | goto out_err; | |
2226 | if (WARN_ON((size_t) ret < size)) { | |
2227 | ret = -ENXIO; | |
06ecc6cb AE |
2228 | rbd_warn(rbd_dev, "short header read (want %zd got %d)", |
2229 | size, ret); | |
4156d998 AE |
2230 | goto out_err; |
2231 | } | |
2232 | if (!rbd_dev_ondisk_valid(ondisk)) { | |
2233 | ret = -ENXIO; | |
06ecc6cb | 2234 | rbd_warn(rbd_dev, "invalid header"); |
4156d998 | 2235 | goto out_err; |
81e759fb | 2236 | } |
602adf40 | 2237 | |
4156d998 AE |
2238 | names_size = le64_to_cpu(ondisk->snap_names_len); |
2239 | want_count = snap_count; | |
2240 | snap_count = le32_to_cpu(ondisk->snap_count); | |
2241 | } while (snap_count != want_count); | |
00f1f36f | 2242 | |
4156d998 | 2243 | return ondisk; |
00f1f36f | 2244 | |
4156d998 AE |
2245 | out_err: |
2246 | kfree(ondisk); | |
2247 | ||
2248 | return ERR_PTR(ret); | |
2249 | } | |
2250 | ||
2251 | /* | |
2252 | * reload the ondisk the header | |
2253 | */ | |
2254 | static int rbd_read_header(struct rbd_device *rbd_dev, | |
2255 | struct rbd_image_header *header) | |
2256 | { | |
2257 | struct rbd_image_header_ondisk *ondisk; | |
2258 | u64 ver = 0; | |
2259 | int ret; | |
602adf40 | 2260 | |
4156d998 AE |
2261 | ondisk = rbd_dev_v1_header_read(rbd_dev, &ver); |
2262 | if (IS_ERR(ondisk)) | |
2263 | return PTR_ERR(ondisk); | |
2264 | ret = rbd_header_from_disk(header, ondisk); | |
2265 | if (ret >= 0) | |
2266 | header->obj_version = ver; | |
2267 | kfree(ondisk); | |
2268 | ||
2269 | return ret; | |
602adf40 YS |
2270 | } |
2271 | ||
41f38c2b | 2272 | static void rbd_remove_all_snaps(struct rbd_device *rbd_dev) |
dfc5606d YS |
2273 | { |
2274 | struct rbd_snap *snap; | |
a0593290 | 2275 | struct rbd_snap *next; |
dfc5606d | 2276 | |
a0593290 | 2277 | list_for_each_entry_safe(snap, next, &rbd_dev->snaps, node) |
41f38c2b | 2278 | rbd_remove_snap_dev(snap); |
dfc5606d YS |
2279 | } |
2280 | ||
9478554a AE |
2281 | static void rbd_update_mapping_size(struct rbd_device *rbd_dev) |
2282 | { | |
2283 | sector_t size; | |
2284 | ||
0d7dbfce | 2285 | if (rbd_dev->spec->snap_id != CEPH_NOSNAP) |
9478554a AE |
2286 | return; |
2287 | ||
2288 | size = (sector_t) rbd_dev->header.image_size / SECTOR_SIZE; | |
2289 | dout("setting size to %llu sectors", (unsigned long long) size); | |
2290 | rbd_dev->mapping.size = (u64) size; | |
2291 | set_capacity(rbd_dev->disk, size); | |
2292 | } | |
2293 | ||
602adf40 YS |
2294 | /* |
2295 | * only read the first part of the ondisk header, without the snaps info | |
2296 | */ | |
117973fb | 2297 | static int rbd_dev_v1_refresh(struct rbd_device *rbd_dev, u64 *hver) |
602adf40 YS |
2298 | { |
2299 | int ret; | |
2300 | struct rbd_image_header h; | |
602adf40 YS |
2301 | |
2302 | ret = rbd_read_header(rbd_dev, &h); | |
2303 | if (ret < 0) | |
2304 | return ret; | |
2305 | ||
a51aa0c0 JD |
2306 | down_write(&rbd_dev->header_rwsem); |
2307 | ||
9478554a AE |
2308 | /* Update image size, and check for resize of mapped image */ |
2309 | rbd_dev->header.image_size = h.image_size; | |
2310 | rbd_update_mapping_size(rbd_dev); | |
9db4b3e3 | 2311 | |
849b4260 | 2312 | /* rbd_dev->header.object_prefix shouldn't change */ |
602adf40 | 2313 | kfree(rbd_dev->header.snap_sizes); |
849b4260 | 2314 | kfree(rbd_dev->header.snap_names); |
d1d25646 JD |
2315 | /* osd requests may still refer to snapc */ |
2316 | ceph_put_snap_context(rbd_dev->header.snapc); | |
602adf40 | 2317 | |
b813623a AE |
2318 | if (hver) |
2319 | *hver = h.obj_version; | |
a71b891b | 2320 | rbd_dev->header.obj_version = h.obj_version; |
93a24e08 | 2321 | rbd_dev->header.image_size = h.image_size; |
602adf40 YS |
2322 | rbd_dev->header.snapc = h.snapc; |
2323 | rbd_dev->header.snap_names = h.snap_names; | |
2324 | rbd_dev->header.snap_sizes = h.snap_sizes; | |
849b4260 AE |
2325 | /* Free the extra copy of the object prefix */ |
2326 | WARN_ON(strcmp(rbd_dev->header.object_prefix, h.object_prefix)); | |
2327 | kfree(h.object_prefix); | |
2328 | ||
304f6808 AE |
2329 | ret = rbd_dev_snaps_update(rbd_dev); |
2330 | if (!ret) | |
2331 | ret = rbd_dev_snaps_register(rbd_dev); | |
dfc5606d | 2332 | |
c666601a | 2333 | up_write(&rbd_dev->header_rwsem); |
602adf40 | 2334 | |
dfc5606d | 2335 | return ret; |
602adf40 YS |
2336 | } |
2337 | ||
117973fb | 2338 | static int rbd_dev_refresh(struct rbd_device *rbd_dev, u64 *hver) |
1fe5e993 AE |
2339 | { |
2340 | int ret; | |
2341 | ||
117973fb | 2342 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); |
1fe5e993 | 2343 | mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING); |
117973fb AE |
2344 | if (rbd_dev->image_format == 1) |
2345 | ret = rbd_dev_v1_refresh(rbd_dev, hver); | |
2346 | else | |
2347 | ret = rbd_dev_v2_refresh(rbd_dev, hver); | |
1fe5e993 AE |
2348 | mutex_unlock(&ctl_mutex); |
2349 | ||
2350 | return ret; | |
2351 | } | |
2352 | ||
602adf40 YS |
2353 | static int rbd_init_disk(struct rbd_device *rbd_dev) |
2354 | { | |
2355 | struct gendisk *disk; | |
2356 | struct request_queue *q; | |
593a9e7b | 2357 | u64 segment_size; |
602adf40 | 2358 | |
602adf40 | 2359 | /* create gendisk info */ |
602adf40 YS |
2360 | disk = alloc_disk(RBD_MINORS_PER_MAJOR); |
2361 | if (!disk) | |
1fcdb8aa | 2362 | return -ENOMEM; |
602adf40 | 2363 | |
f0f8cef5 | 2364 | snprintf(disk->disk_name, sizeof(disk->disk_name), RBD_DRV_NAME "%d", |
de71a297 | 2365 | rbd_dev->dev_id); |
602adf40 YS |
2366 | disk->major = rbd_dev->major; |
2367 | disk->first_minor = 0; | |
2368 | disk->fops = &rbd_bd_ops; | |
2369 | disk->private_data = rbd_dev; | |
2370 | ||
bf0d5f50 | 2371 | q = blk_init_queue(rbd_request_fn, &rbd_dev->lock); |
602adf40 YS |
2372 | if (!q) |
2373 | goto out_disk; | |
029bcbd8 | 2374 | |
593a9e7b AE |
2375 | /* We use the default size, but let's be explicit about it. */ |
2376 | blk_queue_physical_block_size(q, SECTOR_SIZE); | |
2377 | ||
029bcbd8 | 2378 | /* set io sizes to object size */ |
593a9e7b AE |
2379 | segment_size = rbd_obj_bytes(&rbd_dev->header); |
2380 | blk_queue_max_hw_sectors(q, segment_size / SECTOR_SIZE); | |
2381 | blk_queue_max_segment_size(q, segment_size); | |
2382 | blk_queue_io_min(q, segment_size); | |
2383 | blk_queue_io_opt(q, segment_size); | |
029bcbd8 | 2384 | |
602adf40 YS |
2385 | blk_queue_merge_bvec(q, rbd_merge_bvec); |
2386 | disk->queue = q; | |
2387 | ||
2388 | q->queuedata = rbd_dev; | |
2389 | ||
2390 | rbd_dev->disk = disk; | |
602adf40 | 2391 | |
12f02944 AE |
2392 | set_capacity(rbd_dev->disk, rbd_dev->mapping.size / SECTOR_SIZE); |
2393 | ||
602adf40 | 2394 | return 0; |
602adf40 YS |
2395 | out_disk: |
2396 | put_disk(disk); | |
1fcdb8aa AE |
2397 | |
2398 | return -ENOMEM; | |
602adf40 YS |
2399 | } |
2400 | ||
dfc5606d YS |
2401 | /* |
2402 | sysfs | |
2403 | */ | |
2404 | ||
593a9e7b AE |
2405 | static struct rbd_device *dev_to_rbd_dev(struct device *dev) |
2406 | { | |
2407 | return container_of(dev, struct rbd_device, dev); | |
2408 | } | |
2409 | ||
dfc5606d YS |
2410 | static ssize_t rbd_size_show(struct device *dev, |
2411 | struct device_attribute *attr, char *buf) | |
2412 | { | |
593a9e7b | 2413 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
a51aa0c0 JD |
2414 | sector_t size; |
2415 | ||
2416 | down_read(&rbd_dev->header_rwsem); | |
2417 | size = get_capacity(rbd_dev->disk); | |
2418 | up_read(&rbd_dev->header_rwsem); | |
dfc5606d | 2419 | |
a51aa0c0 | 2420 | return sprintf(buf, "%llu\n", (unsigned long long) size * SECTOR_SIZE); |
dfc5606d YS |
2421 | } |
2422 | ||
34b13184 AE |
2423 | /* |
2424 | * Note this shows the features for whatever's mapped, which is not | |
2425 | * necessarily the base image. | |
2426 | */ | |
2427 | static ssize_t rbd_features_show(struct device *dev, | |
2428 | struct device_attribute *attr, char *buf) | |
2429 | { | |
2430 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
2431 | ||
2432 | return sprintf(buf, "0x%016llx\n", | |
2433 | (unsigned long long) rbd_dev->mapping.features); | |
2434 | } | |
2435 | ||
dfc5606d YS |
2436 | static ssize_t rbd_major_show(struct device *dev, |
2437 | struct device_attribute *attr, char *buf) | |
2438 | { | |
593a9e7b | 2439 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
602adf40 | 2440 | |
dfc5606d YS |
2441 | return sprintf(buf, "%d\n", rbd_dev->major); |
2442 | } | |
2443 | ||
2444 | static ssize_t rbd_client_id_show(struct device *dev, | |
2445 | struct device_attribute *attr, char *buf) | |
602adf40 | 2446 | { |
593a9e7b | 2447 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 2448 | |
1dbb4399 AE |
2449 | return sprintf(buf, "client%lld\n", |
2450 | ceph_client_id(rbd_dev->rbd_client->client)); | |
602adf40 YS |
2451 | } |
2452 | ||
dfc5606d YS |
2453 | static ssize_t rbd_pool_show(struct device *dev, |
2454 | struct device_attribute *attr, char *buf) | |
602adf40 | 2455 | { |
593a9e7b | 2456 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 2457 | |
0d7dbfce | 2458 | return sprintf(buf, "%s\n", rbd_dev->spec->pool_name); |
dfc5606d YS |
2459 | } |
2460 | ||
9bb2f334 AE |
2461 | static ssize_t rbd_pool_id_show(struct device *dev, |
2462 | struct device_attribute *attr, char *buf) | |
2463 | { | |
2464 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
2465 | ||
0d7dbfce AE |
2466 | return sprintf(buf, "%llu\n", |
2467 | (unsigned long long) rbd_dev->spec->pool_id); | |
9bb2f334 AE |
2468 | } |
2469 | ||
dfc5606d YS |
2470 | static ssize_t rbd_name_show(struct device *dev, |
2471 | struct device_attribute *attr, char *buf) | |
2472 | { | |
593a9e7b | 2473 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 2474 | |
a92ffdf8 AE |
2475 | if (rbd_dev->spec->image_name) |
2476 | return sprintf(buf, "%s\n", rbd_dev->spec->image_name); | |
2477 | ||
2478 | return sprintf(buf, "(unknown)\n"); | |
dfc5606d YS |
2479 | } |
2480 | ||
589d30e0 AE |
2481 | static ssize_t rbd_image_id_show(struct device *dev, |
2482 | struct device_attribute *attr, char *buf) | |
2483 | { | |
2484 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
2485 | ||
0d7dbfce | 2486 | return sprintf(buf, "%s\n", rbd_dev->spec->image_id); |
589d30e0 AE |
2487 | } |
2488 | ||
34b13184 AE |
2489 | /* |
2490 | * Shows the name of the currently-mapped snapshot (or | |
2491 | * RBD_SNAP_HEAD_NAME for the base image). | |
2492 | */ | |
dfc5606d YS |
2493 | static ssize_t rbd_snap_show(struct device *dev, |
2494 | struct device_attribute *attr, | |
2495 | char *buf) | |
2496 | { | |
593a9e7b | 2497 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 2498 | |
0d7dbfce | 2499 | return sprintf(buf, "%s\n", rbd_dev->spec->snap_name); |
dfc5606d YS |
2500 | } |
2501 | ||
86b00e0d AE |
2502 | /* |
2503 | * For an rbd v2 image, shows the pool id, image id, and snapshot id | |
2504 | * for the parent image. If there is no parent, simply shows | |
2505 | * "(no parent image)". | |
2506 | */ | |
2507 | static ssize_t rbd_parent_show(struct device *dev, | |
2508 | struct device_attribute *attr, | |
2509 | char *buf) | |
2510 | { | |
2511 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
2512 | struct rbd_spec *spec = rbd_dev->parent_spec; | |
2513 | int count; | |
2514 | char *bufp = buf; | |
2515 | ||
2516 | if (!spec) | |
2517 | return sprintf(buf, "(no parent image)\n"); | |
2518 | ||
2519 | count = sprintf(bufp, "pool_id %llu\npool_name %s\n", | |
2520 | (unsigned long long) spec->pool_id, spec->pool_name); | |
2521 | if (count < 0) | |
2522 | return count; | |
2523 | bufp += count; | |
2524 | ||
2525 | count = sprintf(bufp, "image_id %s\nimage_name %s\n", spec->image_id, | |
2526 | spec->image_name ? spec->image_name : "(unknown)"); | |
2527 | if (count < 0) | |
2528 | return count; | |
2529 | bufp += count; | |
2530 | ||
2531 | count = sprintf(bufp, "snap_id %llu\nsnap_name %s\n", | |
2532 | (unsigned long long) spec->snap_id, spec->snap_name); | |
2533 | if (count < 0) | |
2534 | return count; | |
2535 | bufp += count; | |
2536 | ||
2537 | count = sprintf(bufp, "overlap %llu\n", rbd_dev->parent_overlap); | |
2538 | if (count < 0) | |
2539 | return count; | |
2540 | bufp += count; | |
2541 | ||
2542 | return (ssize_t) (bufp - buf); | |
2543 | } | |
2544 | ||
dfc5606d YS |
2545 | static ssize_t rbd_image_refresh(struct device *dev, |
2546 | struct device_attribute *attr, | |
2547 | const char *buf, | |
2548 | size_t size) | |
2549 | { | |
593a9e7b | 2550 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
b813623a | 2551 | int ret; |
602adf40 | 2552 | |
117973fb | 2553 | ret = rbd_dev_refresh(rbd_dev, NULL); |
b813623a AE |
2554 | |
2555 | return ret < 0 ? ret : size; | |
dfc5606d | 2556 | } |
602adf40 | 2557 | |
dfc5606d | 2558 | static DEVICE_ATTR(size, S_IRUGO, rbd_size_show, NULL); |
34b13184 | 2559 | static DEVICE_ATTR(features, S_IRUGO, rbd_features_show, NULL); |
dfc5606d YS |
2560 | static DEVICE_ATTR(major, S_IRUGO, rbd_major_show, NULL); |
2561 | static DEVICE_ATTR(client_id, S_IRUGO, rbd_client_id_show, NULL); | |
2562 | static DEVICE_ATTR(pool, S_IRUGO, rbd_pool_show, NULL); | |
9bb2f334 | 2563 | static DEVICE_ATTR(pool_id, S_IRUGO, rbd_pool_id_show, NULL); |
dfc5606d | 2564 | static DEVICE_ATTR(name, S_IRUGO, rbd_name_show, NULL); |
589d30e0 | 2565 | static DEVICE_ATTR(image_id, S_IRUGO, rbd_image_id_show, NULL); |
dfc5606d YS |
2566 | static DEVICE_ATTR(refresh, S_IWUSR, NULL, rbd_image_refresh); |
2567 | static DEVICE_ATTR(current_snap, S_IRUGO, rbd_snap_show, NULL); | |
86b00e0d | 2568 | static DEVICE_ATTR(parent, S_IRUGO, rbd_parent_show, NULL); |
dfc5606d YS |
2569 | |
2570 | static struct attribute *rbd_attrs[] = { | |
2571 | &dev_attr_size.attr, | |
34b13184 | 2572 | &dev_attr_features.attr, |
dfc5606d YS |
2573 | &dev_attr_major.attr, |
2574 | &dev_attr_client_id.attr, | |
2575 | &dev_attr_pool.attr, | |
9bb2f334 | 2576 | &dev_attr_pool_id.attr, |
dfc5606d | 2577 | &dev_attr_name.attr, |
589d30e0 | 2578 | &dev_attr_image_id.attr, |
dfc5606d | 2579 | &dev_attr_current_snap.attr, |
86b00e0d | 2580 | &dev_attr_parent.attr, |
dfc5606d | 2581 | &dev_attr_refresh.attr, |
dfc5606d YS |
2582 | NULL |
2583 | }; | |
2584 | ||
2585 | static struct attribute_group rbd_attr_group = { | |
2586 | .attrs = rbd_attrs, | |
2587 | }; | |
2588 | ||
2589 | static const struct attribute_group *rbd_attr_groups[] = { | |
2590 | &rbd_attr_group, | |
2591 | NULL | |
2592 | }; | |
2593 | ||
2594 | static void rbd_sysfs_dev_release(struct device *dev) | |
2595 | { | |
2596 | } | |
2597 | ||
2598 | static struct device_type rbd_device_type = { | |
2599 | .name = "rbd", | |
2600 | .groups = rbd_attr_groups, | |
2601 | .release = rbd_sysfs_dev_release, | |
2602 | }; | |
2603 | ||
2604 | ||
2605 | /* | |
2606 | sysfs - snapshots | |
2607 | */ | |
2608 | ||
2609 | static ssize_t rbd_snap_size_show(struct device *dev, | |
2610 | struct device_attribute *attr, | |
2611 | char *buf) | |
2612 | { | |
2613 | struct rbd_snap *snap = container_of(dev, struct rbd_snap, dev); | |
2614 | ||
3591538f | 2615 | return sprintf(buf, "%llu\n", (unsigned long long)snap->size); |
dfc5606d YS |
2616 | } |
2617 | ||
2618 | static ssize_t rbd_snap_id_show(struct device *dev, | |
2619 | struct device_attribute *attr, | |
2620 | char *buf) | |
2621 | { | |
2622 | struct rbd_snap *snap = container_of(dev, struct rbd_snap, dev); | |
2623 | ||
3591538f | 2624 | return sprintf(buf, "%llu\n", (unsigned long long)snap->id); |
dfc5606d YS |
2625 | } |
2626 | ||
34b13184 AE |
2627 | static ssize_t rbd_snap_features_show(struct device *dev, |
2628 | struct device_attribute *attr, | |
2629 | char *buf) | |
2630 | { | |
2631 | struct rbd_snap *snap = container_of(dev, struct rbd_snap, dev); | |
2632 | ||
2633 | return sprintf(buf, "0x%016llx\n", | |
2634 | (unsigned long long) snap->features); | |
2635 | } | |
2636 | ||
dfc5606d YS |
2637 | static DEVICE_ATTR(snap_size, S_IRUGO, rbd_snap_size_show, NULL); |
2638 | static DEVICE_ATTR(snap_id, S_IRUGO, rbd_snap_id_show, NULL); | |
34b13184 | 2639 | static DEVICE_ATTR(snap_features, S_IRUGO, rbd_snap_features_show, NULL); |
dfc5606d YS |
2640 | |
2641 | static struct attribute *rbd_snap_attrs[] = { | |
2642 | &dev_attr_snap_size.attr, | |
2643 | &dev_attr_snap_id.attr, | |
34b13184 | 2644 | &dev_attr_snap_features.attr, |
dfc5606d YS |
2645 | NULL, |
2646 | }; | |
2647 | ||
2648 | static struct attribute_group rbd_snap_attr_group = { | |
2649 | .attrs = rbd_snap_attrs, | |
2650 | }; | |
2651 | ||
2652 | static void rbd_snap_dev_release(struct device *dev) | |
2653 | { | |
2654 | struct rbd_snap *snap = container_of(dev, struct rbd_snap, dev); | |
2655 | kfree(snap->name); | |
2656 | kfree(snap); | |
2657 | } | |
2658 | ||
2659 | static const struct attribute_group *rbd_snap_attr_groups[] = { | |
2660 | &rbd_snap_attr_group, | |
2661 | NULL | |
2662 | }; | |
2663 | ||
2664 | static struct device_type rbd_snap_device_type = { | |
2665 | .groups = rbd_snap_attr_groups, | |
2666 | .release = rbd_snap_dev_release, | |
2667 | }; | |
2668 | ||
8b8fb99c AE |
2669 | static struct rbd_spec *rbd_spec_get(struct rbd_spec *spec) |
2670 | { | |
2671 | kref_get(&spec->kref); | |
2672 | ||
2673 | return spec; | |
2674 | } | |
2675 | ||
2676 | static void rbd_spec_free(struct kref *kref); | |
2677 | static void rbd_spec_put(struct rbd_spec *spec) | |
2678 | { | |
2679 | if (spec) | |
2680 | kref_put(&spec->kref, rbd_spec_free); | |
2681 | } | |
2682 | ||
2683 | static struct rbd_spec *rbd_spec_alloc(void) | |
2684 | { | |
2685 | struct rbd_spec *spec; | |
2686 | ||
2687 | spec = kzalloc(sizeof (*spec), GFP_KERNEL); | |
2688 | if (!spec) | |
2689 | return NULL; | |
2690 | kref_init(&spec->kref); | |
2691 | ||
2692 | rbd_spec_put(rbd_spec_get(spec)); /* TEMPORARY */ | |
2693 | ||
2694 | return spec; | |
2695 | } | |
2696 | ||
2697 | static void rbd_spec_free(struct kref *kref) | |
2698 | { | |
2699 | struct rbd_spec *spec = container_of(kref, struct rbd_spec, kref); | |
2700 | ||
2701 | kfree(spec->pool_name); | |
2702 | kfree(spec->image_id); | |
2703 | kfree(spec->image_name); | |
2704 | kfree(spec->snap_name); | |
2705 | kfree(spec); | |
2706 | } | |
2707 | ||
c53d5893 AE |
2708 | struct rbd_device *rbd_dev_create(struct rbd_client *rbdc, |
2709 | struct rbd_spec *spec) | |
2710 | { | |
2711 | struct rbd_device *rbd_dev; | |
2712 | ||
2713 | rbd_dev = kzalloc(sizeof (*rbd_dev), GFP_KERNEL); | |
2714 | if (!rbd_dev) | |
2715 | return NULL; | |
2716 | ||
2717 | spin_lock_init(&rbd_dev->lock); | |
6d292906 | 2718 | rbd_dev->flags = 0; |
c53d5893 AE |
2719 | INIT_LIST_HEAD(&rbd_dev->node); |
2720 | INIT_LIST_HEAD(&rbd_dev->snaps); | |
2721 | init_rwsem(&rbd_dev->header_rwsem); | |
2722 | ||
2723 | rbd_dev->spec = spec; | |
2724 | rbd_dev->rbd_client = rbdc; | |
2725 | ||
0903e875 AE |
2726 | /* Initialize the layout used for all rbd requests */ |
2727 | ||
2728 | rbd_dev->layout.fl_stripe_unit = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER); | |
2729 | rbd_dev->layout.fl_stripe_count = cpu_to_le32(1); | |
2730 | rbd_dev->layout.fl_object_size = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER); | |
2731 | rbd_dev->layout.fl_pg_pool = cpu_to_le32((u32) spec->pool_id); | |
2732 | ||
c53d5893 AE |
2733 | return rbd_dev; |
2734 | } | |
2735 | ||
2736 | static void rbd_dev_destroy(struct rbd_device *rbd_dev) | |
2737 | { | |
86b00e0d | 2738 | rbd_spec_put(rbd_dev->parent_spec); |
c53d5893 AE |
2739 | kfree(rbd_dev->header_name); |
2740 | rbd_put_client(rbd_dev->rbd_client); | |
2741 | rbd_spec_put(rbd_dev->spec); | |
2742 | kfree(rbd_dev); | |
2743 | } | |
2744 | ||
304f6808 AE |
2745 | static bool rbd_snap_registered(struct rbd_snap *snap) |
2746 | { | |
2747 | bool ret = snap->dev.type == &rbd_snap_device_type; | |
2748 | bool reg = device_is_registered(&snap->dev); | |
2749 | ||
2750 | rbd_assert(!ret ^ reg); | |
2751 | ||
2752 | return ret; | |
2753 | } | |
2754 | ||
41f38c2b | 2755 | static void rbd_remove_snap_dev(struct rbd_snap *snap) |
dfc5606d YS |
2756 | { |
2757 | list_del(&snap->node); | |
304f6808 AE |
2758 | if (device_is_registered(&snap->dev)) |
2759 | device_unregister(&snap->dev); | |
dfc5606d YS |
2760 | } |
2761 | ||
14e7085d | 2762 | static int rbd_register_snap_dev(struct rbd_snap *snap, |
dfc5606d YS |
2763 | struct device *parent) |
2764 | { | |
2765 | struct device *dev = &snap->dev; | |
2766 | int ret; | |
2767 | ||
2768 | dev->type = &rbd_snap_device_type; | |
2769 | dev->parent = parent; | |
2770 | dev->release = rbd_snap_dev_release; | |
d4b125e9 | 2771 | dev_set_name(dev, "%s%s", RBD_SNAP_DEV_NAME_PREFIX, snap->name); |
304f6808 AE |
2772 | dout("%s: registering device for snapshot %s\n", __func__, snap->name); |
2773 | ||
dfc5606d YS |
2774 | ret = device_register(dev); |
2775 | ||
2776 | return ret; | |
2777 | } | |
2778 | ||
4e891e0a | 2779 | static struct rbd_snap *__rbd_add_snap_dev(struct rbd_device *rbd_dev, |
c8d18425 | 2780 | const char *snap_name, |
34b13184 AE |
2781 | u64 snap_id, u64 snap_size, |
2782 | u64 snap_features) | |
dfc5606d | 2783 | { |
4e891e0a | 2784 | struct rbd_snap *snap; |
dfc5606d | 2785 | int ret; |
4e891e0a AE |
2786 | |
2787 | snap = kzalloc(sizeof (*snap), GFP_KERNEL); | |
dfc5606d | 2788 | if (!snap) |
4e891e0a AE |
2789 | return ERR_PTR(-ENOMEM); |
2790 | ||
2791 | ret = -ENOMEM; | |
c8d18425 | 2792 | snap->name = kstrdup(snap_name, GFP_KERNEL); |
4e891e0a AE |
2793 | if (!snap->name) |
2794 | goto err; | |
2795 | ||
c8d18425 AE |
2796 | snap->id = snap_id; |
2797 | snap->size = snap_size; | |
34b13184 | 2798 | snap->features = snap_features; |
4e891e0a AE |
2799 | |
2800 | return snap; | |
2801 | ||
dfc5606d YS |
2802 | err: |
2803 | kfree(snap->name); | |
2804 | kfree(snap); | |
4e891e0a AE |
2805 | |
2806 | return ERR_PTR(ret); | |
dfc5606d YS |
2807 | } |
2808 | ||
cd892126 AE |
2809 | static char *rbd_dev_v1_snap_info(struct rbd_device *rbd_dev, u32 which, |
2810 | u64 *snap_size, u64 *snap_features) | |
2811 | { | |
2812 | char *snap_name; | |
2813 | ||
2814 | rbd_assert(which < rbd_dev->header.snapc->num_snaps); | |
2815 | ||
2816 | *snap_size = rbd_dev->header.snap_sizes[which]; | |
2817 | *snap_features = 0; /* No features for v1 */ | |
2818 | ||
2819 | /* Skip over names until we find the one we are looking for */ | |
2820 | ||
2821 | snap_name = rbd_dev->header.snap_names; | |
2822 | while (which--) | |
2823 | snap_name += strlen(snap_name) + 1; | |
2824 | ||
2825 | return snap_name; | |
2826 | } | |
2827 | ||
9d475de5 AE |
2828 | /* |
2829 | * Get the size and object order for an image snapshot, or if | |
2830 | * snap_id is CEPH_NOSNAP, gets this information for the base | |
2831 | * image. | |
2832 | */ | |
2833 | static int _rbd_dev_v2_snap_size(struct rbd_device *rbd_dev, u64 snap_id, | |
2834 | u8 *order, u64 *snap_size) | |
2835 | { | |
2836 | __le64 snapid = cpu_to_le64(snap_id); | |
2837 | int ret; | |
2838 | struct { | |
2839 | u8 order; | |
2840 | __le64 size; | |
2841 | } __attribute__ ((packed)) size_buf = { 0 }; | |
2842 | ||
36be9a76 | 2843 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
9d475de5 AE |
2844 | "rbd", "get_size", |
2845 | (char *) &snapid, sizeof (snapid), | |
07b2391f | 2846 | (char *) &size_buf, sizeof (size_buf), NULL); |
36be9a76 | 2847 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
9d475de5 AE |
2848 | if (ret < 0) |
2849 | return ret; | |
2850 | ||
2851 | *order = size_buf.order; | |
2852 | *snap_size = le64_to_cpu(size_buf.size); | |
2853 | ||
2854 | dout(" snap_id 0x%016llx order = %u, snap_size = %llu\n", | |
2855 | (unsigned long long) snap_id, (unsigned int) *order, | |
2856 | (unsigned long long) *snap_size); | |
2857 | ||
2858 | return 0; | |
2859 | } | |
2860 | ||
2861 | static int rbd_dev_v2_image_size(struct rbd_device *rbd_dev) | |
2862 | { | |
2863 | return _rbd_dev_v2_snap_size(rbd_dev, CEPH_NOSNAP, | |
2864 | &rbd_dev->header.obj_order, | |
2865 | &rbd_dev->header.image_size); | |
2866 | } | |
2867 | ||
1e130199 AE |
2868 | static int rbd_dev_v2_object_prefix(struct rbd_device *rbd_dev) |
2869 | { | |
2870 | void *reply_buf; | |
2871 | int ret; | |
2872 | void *p; | |
2873 | ||
2874 | reply_buf = kzalloc(RBD_OBJ_PREFIX_LEN_MAX, GFP_KERNEL); | |
2875 | if (!reply_buf) | |
2876 | return -ENOMEM; | |
2877 | ||
36be9a76 | 2878 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
1e130199 AE |
2879 | "rbd", "get_object_prefix", |
2880 | NULL, 0, | |
07b2391f | 2881 | reply_buf, RBD_OBJ_PREFIX_LEN_MAX, NULL); |
36be9a76 | 2882 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
1e130199 AE |
2883 | if (ret < 0) |
2884 | goto out; | |
2885 | ||
2886 | p = reply_buf; | |
2887 | rbd_dev->header.object_prefix = ceph_extract_encoded_string(&p, | |
2888 | p + RBD_OBJ_PREFIX_LEN_MAX, | |
2889 | NULL, GFP_NOIO); | |
2890 | ||
2891 | if (IS_ERR(rbd_dev->header.object_prefix)) { | |
2892 | ret = PTR_ERR(rbd_dev->header.object_prefix); | |
2893 | rbd_dev->header.object_prefix = NULL; | |
2894 | } else { | |
2895 | dout(" object_prefix = %s\n", rbd_dev->header.object_prefix); | |
2896 | } | |
2897 | ||
2898 | out: | |
2899 | kfree(reply_buf); | |
2900 | ||
2901 | return ret; | |
2902 | } | |
2903 | ||
b1b5402a AE |
2904 | static int _rbd_dev_v2_snap_features(struct rbd_device *rbd_dev, u64 snap_id, |
2905 | u64 *snap_features) | |
2906 | { | |
2907 | __le64 snapid = cpu_to_le64(snap_id); | |
2908 | struct { | |
2909 | __le64 features; | |
2910 | __le64 incompat; | |
2911 | } features_buf = { 0 }; | |
d889140c | 2912 | u64 incompat; |
b1b5402a AE |
2913 | int ret; |
2914 | ||
36be9a76 | 2915 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
b1b5402a AE |
2916 | "rbd", "get_features", |
2917 | (char *) &snapid, sizeof (snapid), | |
2918 | (char *) &features_buf, sizeof (features_buf), | |
07b2391f | 2919 | NULL); |
36be9a76 | 2920 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
b1b5402a AE |
2921 | if (ret < 0) |
2922 | return ret; | |
d889140c AE |
2923 | |
2924 | incompat = le64_to_cpu(features_buf.incompat); | |
2925 | if (incompat & ~RBD_FEATURES_ALL) | |
b8f5c6ed | 2926 | return -ENXIO; |
d889140c | 2927 | |
b1b5402a AE |
2928 | *snap_features = le64_to_cpu(features_buf.features); |
2929 | ||
2930 | dout(" snap_id 0x%016llx features = 0x%016llx incompat = 0x%016llx\n", | |
2931 | (unsigned long long) snap_id, | |
2932 | (unsigned long long) *snap_features, | |
2933 | (unsigned long long) le64_to_cpu(features_buf.incompat)); | |
2934 | ||
2935 | return 0; | |
2936 | } | |
2937 | ||
2938 | static int rbd_dev_v2_features(struct rbd_device *rbd_dev) | |
2939 | { | |
2940 | return _rbd_dev_v2_snap_features(rbd_dev, CEPH_NOSNAP, | |
2941 | &rbd_dev->header.features); | |
2942 | } | |
2943 | ||
86b00e0d AE |
2944 | static int rbd_dev_v2_parent_info(struct rbd_device *rbd_dev) |
2945 | { | |
2946 | struct rbd_spec *parent_spec; | |
2947 | size_t size; | |
2948 | void *reply_buf = NULL; | |
2949 | __le64 snapid; | |
2950 | void *p; | |
2951 | void *end; | |
2952 | char *image_id; | |
2953 | u64 overlap; | |
86b00e0d AE |
2954 | int ret; |
2955 | ||
2956 | parent_spec = rbd_spec_alloc(); | |
2957 | if (!parent_spec) | |
2958 | return -ENOMEM; | |
2959 | ||
2960 | size = sizeof (__le64) + /* pool_id */ | |
2961 | sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX + /* image_id */ | |
2962 | sizeof (__le64) + /* snap_id */ | |
2963 | sizeof (__le64); /* overlap */ | |
2964 | reply_buf = kmalloc(size, GFP_KERNEL); | |
2965 | if (!reply_buf) { | |
2966 | ret = -ENOMEM; | |
2967 | goto out_err; | |
2968 | } | |
2969 | ||
2970 | snapid = cpu_to_le64(CEPH_NOSNAP); | |
36be9a76 | 2971 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
86b00e0d AE |
2972 | "rbd", "get_parent", |
2973 | (char *) &snapid, sizeof (snapid), | |
07b2391f | 2974 | (char *) reply_buf, size, NULL); |
36be9a76 | 2975 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
86b00e0d AE |
2976 | if (ret < 0) |
2977 | goto out_err; | |
2978 | ||
2979 | ret = -ERANGE; | |
2980 | p = reply_buf; | |
2981 | end = (char *) reply_buf + size; | |
2982 | ceph_decode_64_safe(&p, end, parent_spec->pool_id, out_err); | |
2983 | if (parent_spec->pool_id == CEPH_NOPOOL) | |
2984 | goto out; /* No parent? No problem. */ | |
2985 | ||
0903e875 AE |
2986 | /* The ceph file layout needs to fit pool id in 32 bits */ |
2987 | ||
2988 | ret = -EIO; | |
2989 | if (WARN_ON(parent_spec->pool_id > (u64) U32_MAX)) | |
2990 | goto out; | |
2991 | ||
979ed480 | 2992 | image_id = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL); |
86b00e0d AE |
2993 | if (IS_ERR(image_id)) { |
2994 | ret = PTR_ERR(image_id); | |
2995 | goto out_err; | |
2996 | } | |
2997 | parent_spec->image_id = image_id; | |
2998 | ceph_decode_64_safe(&p, end, parent_spec->snap_id, out_err); | |
2999 | ceph_decode_64_safe(&p, end, overlap, out_err); | |
3000 | ||
3001 | rbd_dev->parent_overlap = overlap; | |
3002 | rbd_dev->parent_spec = parent_spec; | |
3003 | parent_spec = NULL; /* rbd_dev now owns this */ | |
3004 | out: | |
3005 | ret = 0; | |
3006 | out_err: | |
3007 | kfree(reply_buf); | |
3008 | rbd_spec_put(parent_spec); | |
3009 | ||
3010 | return ret; | |
3011 | } | |
3012 | ||
9e15b77d AE |
3013 | static char *rbd_dev_image_name(struct rbd_device *rbd_dev) |
3014 | { | |
3015 | size_t image_id_size; | |
3016 | char *image_id; | |
3017 | void *p; | |
3018 | void *end; | |
3019 | size_t size; | |
3020 | void *reply_buf = NULL; | |
3021 | size_t len = 0; | |
3022 | char *image_name = NULL; | |
3023 | int ret; | |
3024 | ||
3025 | rbd_assert(!rbd_dev->spec->image_name); | |
3026 | ||
69e7a02f AE |
3027 | len = strlen(rbd_dev->spec->image_id); |
3028 | image_id_size = sizeof (__le32) + len; | |
9e15b77d AE |
3029 | image_id = kmalloc(image_id_size, GFP_KERNEL); |
3030 | if (!image_id) | |
3031 | return NULL; | |
3032 | ||
3033 | p = image_id; | |
3034 | end = (char *) image_id + image_id_size; | |
69e7a02f | 3035 | ceph_encode_string(&p, end, rbd_dev->spec->image_id, (u32) len); |
9e15b77d AE |
3036 | |
3037 | size = sizeof (__le32) + RBD_IMAGE_NAME_LEN_MAX; | |
3038 | reply_buf = kmalloc(size, GFP_KERNEL); | |
3039 | if (!reply_buf) | |
3040 | goto out; | |
3041 | ||
36be9a76 | 3042 | ret = rbd_obj_method_sync(rbd_dev, RBD_DIRECTORY, |
9e15b77d AE |
3043 | "rbd", "dir_get_name", |
3044 | image_id, image_id_size, | |
07b2391f | 3045 | (char *) reply_buf, size, NULL); |
9e15b77d AE |
3046 | if (ret < 0) |
3047 | goto out; | |
3048 | p = reply_buf; | |
3049 | end = (char *) reply_buf + size; | |
3050 | image_name = ceph_extract_encoded_string(&p, end, &len, GFP_KERNEL); | |
3051 | if (IS_ERR(image_name)) | |
3052 | image_name = NULL; | |
3053 | else | |
3054 | dout("%s: name is %s len is %zd\n", __func__, image_name, len); | |
3055 | out: | |
3056 | kfree(reply_buf); | |
3057 | kfree(image_id); | |
3058 | ||
3059 | return image_name; | |
3060 | } | |
3061 | ||
3062 | /* | |
3063 | * When a parent image gets probed, we only have the pool, image, | |
3064 | * and snapshot ids but not the names of any of them. This call | |
3065 | * is made later to fill in those names. It has to be done after | |
3066 | * rbd_dev_snaps_update() has completed because some of the | |
3067 | * information (in particular, snapshot name) is not available | |
3068 | * until then. | |
3069 | */ | |
3070 | static int rbd_dev_probe_update_spec(struct rbd_device *rbd_dev) | |
3071 | { | |
3072 | struct ceph_osd_client *osdc; | |
3073 | const char *name; | |
3074 | void *reply_buf = NULL; | |
3075 | int ret; | |
3076 | ||
3077 | if (rbd_dev->spec->pool_name) | |
3078 | return 0; /* Already have the names */ | |
3079 | ||
3080 | /* Look up the pool name */ | |
3081 | ||
3082 | osdc = &rbd_dev->rbd_client->client->osdc; | |
3083 | name = ceph_pg_pool_name_by_id(osdc->osdmap, rbd_dev->spec->pool_id); | |
935dc89f AE |
3084 | if (!name) { |
3085 | rbd_warn(rbd_dev, "there is no pool with id %llu", | |
3086 | rbd_dev->spec->pool_id); /* Really a BUG() */ | |
3087 | return -EIO; | |
3088 | } | |
9e15b77d AE |
3089 | |
3090 | rbd_dev->spec->pool_name = kstrdup(name, GFP_KERNEL); | |
3091 | if (!rbd_dev->spec->pool_name) | |
3092 | return -ENOMEM; | |
3093 | ||
3094 | /* Fetch the image name; tolerate failure here */ | |
3095 | ||
3096 | name = rbd_dev_image_name(rbd_dev); | |
69e7a02f | 3097 | if (name) |
9e15b77d | 3098 | rbd_dev->spec->image_name = (char *) name; |
69e7a02f | 3099 | else |
06ecc6cb | 3100 | rbd_warn(rbd_dev, "unable to get image name"); |
9e15b77d AE |
3101 | |
3102 | /* Look up the snapshot name. */ | |
3103 | ||
3104 | name = rbd_snap_name(rbd_dev, rbd_dev->spec->snap_id); | |
3105 | if (!name) { | |
935dc89f AE |
3106 | rbd_warn(rbd_dev, "no snapshot with id %llu", |
3107 | rbd_dev->spec->snap_id); /* Really a BUG() */ | |
9e15b77d AE |
3108 | ret = -EIO; |
3109 | goto out_err; | |
3110 | } | |
3111 | rbd_dev->spec->snap_name = kstrdup(name, GFP_KERNEL); | |
3112 | if(!rbd_dev->spec->snap_name) | |
3113 | goto out_err; | |
3114 | ||
3115 | return 0; | |
3116 | out_err: | |
3117 | kfree(reply_buf); | |
3118 | kfree(rbd_dev->spec->pool_name); | |
3119 | rbd_dev->spec->pool_name = NULL; | |
3120 | ||
3121 | return ret; | |
3122 | } | |
3123 | ||
6e14b1a6 | 3124 | static int rbd_dev_v2_snap_context(struct rbd_device *rbd_dev, u64 *ver) |
35d489f9 AE |
3125 | { |
3126 | size_t size; | |
3127 | int ret; | |
3128 | void *reply_buf; | |
3129 | void *p; | |
3130 | void *end; | |
3131 | u64 seq; | |
3132 | u32 snap_count; | |
3133 | struct ceph_snap_context *snapc; | |
3134 | u32 i; | |
3135 | ||
3136 | /* | |
3137 | * We'll need room for the seq value (maximum snapshot id), | |
3138 | * snapshot count, and array of that many snapshot ids. | |
3139 | * For now we have a fixed upper limit on the number we're | |
3140 | * prepared to receive. | |
3141 | */ | |
3142 | size = sizeof (__le64) + sizeof (__le32) + | |
3143 | RBD_MAX_SNAP_COUNT * sizeof (__le64); | |
3144 | reply_buf = kzalloc(size, GFP_KERNEL); | |
3145 | if (!reply_buf) | |
3146 | return -ENOMEM; | |
3147 | ||
36be9a76 | 3148 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
35d489f9 AE |
3149 | "rbd", "get_snapcontext", |
3150 | NULL, 0, | |
07b2391f | 3151 | reply_buf, size, ver); |
36be9a76 | 3152 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
35d489f9 AE |
3153 | if (ret < 0) |
3154 | goto out; | |
3155 | ||
3156 | ret = -ERANGE; | |
3157 | p = reply_buf; | |
3158 | end = (char *) reply_buf + size; | |
3159 | ceph_decode_64_safe(&p, end, seq, out); | |
3160 | ceph_decode_32_safe(&p, end, snap_count, out); | |
3161 | ||
3162 | /* | |
3163 | * Make sure the reported number of snapshot ids wouldn't go | |
3164 | * beyond the end of our buffer. But before checking that, | |
3165 | * make sure the computed size of the snapshot context we | |
3166 | * allocate is representable in a size_t. | |
3167 | */ | |
3168 | if (snap_count > (SIZE_MAX - sizeof (struct ceph_snap_context)) | |
3169 | / sizeof (u64)) { | |
3170 | ret = -EINVAL; | |
3171 | goto out; | |
3172 | } | |
3173 | if (!ceph_has_room(&p, end, snap_count * sizeof (__le64))) | |
3174 | goto out; | |
3175 | ||
3176 | size = sizeof (struct ceph_snap_context) + | |
3177 | snap_count * sizeof (snapc->snaps[0]); | |
3178 | snapc = kmalloc(size, GFP_KERNEL); | |
3179 | if (!snapc) { | |
3180 | ret = -ENOMEM; | |
3181 | goto out; | |
3182 | } | |
3183 | ||
3184 | atomic_set(&snapc->nref, 1); | |
3185 | snapc->seq = seq; | |
3186 | snapc->num_snaps = snap_count; | |
3187 | for (i = 0; i < snap_count; i++) | |
3188 | snapc->snaps[i] = ceph_decode_64(&p); | |
3189 | ||
3190 | rbd_dev->header.snapc = snapc; | |
3191 | ||
3192 | dout(" snap context seq = %llu, snap_count = %u\n", | |
3193 | (unsigned long long) seq, (unsigned int) snap_count); | |
3194 | ||
3195 | out: | |
3196 | kfree(reply_buf); | |
3197 | ||
3198 | return 0; | |
3199 | } | |
3200 | ||
b8b1e2db AE |
3201 | static char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev, u32 which) |
3202 | { | |
3203 | size_t size; | |
3204 | void *reply_buf; | |
3205 | __le64 snap_id; | |
3206 | int ret; | |
3207 | void *p; | |
3208 | void *end; | |
b8b1e2db AE |
3209 | char *snap_name; |
3210 | ||
3211 | size = sizeof (__le32) + RBD_MAX_SNAP_NAME_LEN; | |
3212 | reply_buf = kmalloc(size, GFP_KERNEL); | |
3213 | if (!reply_buf) | |
3214 | return ERR_PTR(-ENOMEM); | |
3215 | ||
3216 | snap_id = cpu_to_le64(rbd_dev->header.snapc->snaps[which]); | |
36be9a76 | 3217 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
b8b1e2db AE |
3218 | "rbd", "get_snapshot_name", |
3219 | (char *) &snap_id, sizeof (snap_id), | |
07b2391f | 3220 | reply_buf, size, NULL); |
36be9a76 | 3221 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
b8b1e2db AE |
3222 | if (ret < 0) |
3223 | goto out; | |
3224 | ||
3225 | p = reply_buf; | |
3226 | end = (char *) reply_buf + size; | |
e5c35534 | 3227 | snap_name = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL); |
b8b1e2db AE |
3228 | if (IS_ERR(snap_name)) { |
3229 | ret = PTR_ERR(snap_name); | |
3230 | goto out; | |
3231 | } else { | |
3232 | dout(" snap_id 0x%016llx snap_name = %s\n", | |
3233 | (unsigned long long) le64_to_cpu(snap_id), snap_name); | |
3234 | } | |
3235 | kfree(reply_buf); | |
3236 | ||
3237 | return snap_name; | |
3238 | out: | |
3239 | kfree(reply_buf); | |
3240 | ||
3241 | return ERR_PTR(ret); | |
3242 | } | |
3243 | ||
3244 | static char *rbd_dev_v2_snap_info(struct rbd_device *rbd_dev, u32 which, | |
3245 | u64 *snap_size, u64 *snap_features) | |
3246 | { | |
e0b49868 | 3247 | u64 snap_id; |
b8b1e2db AE |
3248 | u8 order; |
3249 | int ret; | |
3250 | ||
3251 | snap_id = rbd_dev->header.snapc->snaps[which]; | |
3252 | ret = _rbd_dev_v2_snap_size(rbd_dev, snap_id, &order, snap_size); | |
3253 | if (ret) | |
3254 | return ERR_PTR(ret); | |
3255 | ret = _rbd_dev_v2_snap_features(rbd_dev, snap_id, snap_features); | |
3256 | if (ret) | |
3257 | return ERR_PTR(ret); | |
3258 | ||
3259 | return rbd_dev_v2_snap_name(rbd_dev, which); | |
3260 | } | |
3261 | ||
3262 | static char *rbd_dev_snap_info(struct rbd_device *rbd_dev, u32 which, | |
3263 | u64 *snap_size, u64 *snap_features) | |
3264 | { | |
3265 | if (rbd_dev->image_format == 1) | |
3266 | return rbd_dev_v1_snap_info(rbd_dev, which, | |
3267 | snap_size, snap_features); | |
3268 | if (rbd_dev->image_format == 2) | |
3269 | return rbd_dev_v2_snap_info(rbd_dev, which, | |
3270 | snap_size, snap_features); | |
3271 | return ERR_PTR(-EINVAL); | |
3272 | } | |
3273 | ||
117973fb AE |
3274 | static int rbd_dev_v2_refresh(struct rbd_device *rbd_dev, u64 *hver) |
3275 | { | |
3276 | int ret; | |
3277 | __u8 obj_order; | |
3278 | ||
3279 | down_write(&rbd_dev->header_rwsem); | |
3280 | ||
3281 | /* Grab old order first, to see if it changes */ | |
3282 | ||
3283 | obj_order = rbd_dev->header.obj_order, | |
3284 | ret = rbd_dev_v2_image_size(rbd_dev); | |
3285 | if (ret) | |
3286 | goto out; | |
3287 | if (rbd_dev->header.obj_order != obj_order) { | |
3288 | ret = -EIO; | |
3289 | goto out; | |
3290 | } | |
3291 | rbd_update_mapping_size(rbd_dev); | |
3292 | ||
3293 | ret = rbd_dev_v2_snap_context(rbd_dev, hver); | |
3294 | dout("rbd_dev_v2_snap_context returned %d\n", ret); | |
3295 | if (ret) | |
3296 | goto out; | |
3297 | ret = rbd_dev_snaps_update(rbd_dev); | |
3298 | dout("rbd_dev_snaps_update returned %d\n", ret); | |
3299 | if (ret) | |
3300 | goto out; | |
3301 | ret = rbd_dev_snaps_register(rbd_dev); | |
3302 | dout("rbd_dev_snaps_register returned %d\n", ret); | |
3303 | out: | |
3304 | up_write(&rbd_dev->header_rwsem); | |
3305 | ||
3306 | return ret; | |
3307 | } | |
3308 | ||
dfc5606d | 3309 | /* |
35938150 AE |
3310 | * Scan the rbd device's current snapshot list and compare it to the |
3311 | * newly-received snapshot context. Remove any existing snapshots | |
3312 | * not present in the new snapshot context. Add a new snapshot for | |
3313 | * any snaphots in the snapshot context not in the current list. | |
3314 | * And verify there are no changes to snapshots we already know | |
3315 | * about. | |
3316 | * | |
3317 | * Assumes the snapshots in the snapshot context are sorted by | |
3318 | * snapshot id, highest id first. (Snapshots in the rbd_dev's list | |
3319 | * are also maintained in that order.) | |
dfc5606d | 3320 | */ |
304f6808 | 3321 | static int rbd_dev_snaps_update(struct rbd_device *rbd_dev) |
dfc5606d | 3322 | { |
35938150 AE |
3323 | struct ceph_snap_context *snapc = rbd_dev->header.snapc; |
3324 | const u32 snap_count = snapc->num_snaps; | |
35938150 AE |
3325 | struct list_head *head = &rbd_dev->snaps; |
3326 | struct list_head *links = head->next; | |
3327 | u32 index = 0; | |
dfc5606d | 3328 | |
9fcbb800 | 3329 | dout("%s: snap count is %u\n", __func__, (unsigned int) snap_count); |
35938150 AE |
3330 | while (index < snap_count || links != head) { |
3331 | u64 snap_id; | |
3332 | struct rbd_snap *snap; | |
cd892126 AE |
3333 | char *snap_name; |
3334 | u64 snap_size = 0; | |
3335 | u64 snap_features = 0; | |
dfc5606d | 3336 | |
35938150 AE |
3337 | snap_id = index < snap_count ? snapc->snaps[index] |
3338 | : CEPH_NOSNAP; | |
3339 | snap = links != head ? list_entry(links, struct rbd_snap, node) | |
3340 | : NULL; | |
aafb230e | 3341 | rbd_assert(!snap || snap->id != CEPH_NOSNAP); |
dfc5606d | 3342 | |
35938150 AE |
3343 | if (snap_id == CEPH_NOSNAP || (snap && snap->id > snap_id)) { |
3344 | struct list_head *next = links->next; | |
dfc5606d | 3345 | |
6d292906 AE |
3346 | /* |
3347 | * A previously-existing snapshot is not in | |
3348 | * the new snap context. | |
3349 | * | |
3350 | * If the now missing snapshot is the one the | |
3351 | * image is mapped to, clear its exists flag | |
3352 | * so we can avoid sending any more requests | |
3353 | * to it. | |
3354 | */ | |
0d7dbfce | 3355 | if (rbd_dev->spec->snap_id == snap->id) |
6d292906 | 3356 | clear_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags); |
41f38c2b | 3357 | rbd_remove_snap_dev(snap); |
9fcbb800 | 3358 | dout("%ssnap id %llu has been removed\n", |
0d7dbfce AE |
3359 | rbd_dev->spec->snap_id == snap->id ? |
3360 | "mapped " : "", | |
9fcbb800 | 3361 | (unsigned long long) snap->id); |
35938150 AE |
3362 | |
3363 | /* Done with this list entry; advance */ | |
3364 | ||
3365 | links = next; | |
dfc5606d YS |
3366 | continue; |
3367 | } | |
35938150 | 3368 | |
b8b1e2db AE |
3369 | snap_name = rbd_dev_snap_info(rbd_dev, index, |
3370 | &snap_size, &snap_features); | |
cd892126 AE |
3371 | if (IS_ERR(snap_name)) |
3372 | return PTR_ERR(snap_name); | |
3373 | ||
9fcbb800 AE |
3374 | dout("entry %u: snap_id = %llu\n", (unsigned int) snap_count, |
3375 | (unsigned long long) snap_id); | |
35938150 AE |
3376 | if (!snap || (snap_id != CEPH_NOSNAP && snap->id < snap_id)) { |
3377 | struct rbd_snap *new_snap; | |
3378 | ||
3379 | /* We haven't seen this snapshot before */ | |
3380 | ||
c8d18425 | 3381 | new_snap = __rbd_add_snap_dev(rbd_dev, snap_name, |
cd892126 | 3382 | snap_id, snap_size, snap_features); |
9fcbb800 AE |
3383 | if (IS_ERR(new_snap)) { |
3384 | int err = PTR_ERR(new_snap); | |
3385 | ||
3386 | dout(" failed to add dev, error %d\n", err); | |
3387 | ||
3388 | return err; | |
3389 | } | |
35938150 AE |
3390 | |
3391 | /* New goes before existing, or at end of list */ | |
3392 | ||
9fcbb800 | 3393 | dout(" added dev%s\n", snap ? "" : " at end\n"); |
35938150 AE |
3394 | if (snap) |
3395 | list_add_tail(&new_snap->node, &snap->node); | |
3396 | else | |
523f3258 | 3397 | list_add_tail(&new_snap->node, head); |
35938150 AE |
3398 | } else { |
3399 | /* Already have this one */ | |
3400 | ||
9fcbb800 AE |
3401 | dout(" already present\n"); |
3402 | ||
cd892126 | 3403 | rbd_assert(snap->size == snap_size); |
aafb230e | 3404 | rbd_assert(!strcmp(snap->name, snap_name)); |
cd892126 | 3405 | rbd_assert(snap->features == snap_features); |
35938150 AE |
3406 | |
3407 | /* Done with this list entry; advance */ | |
3408 | ||
3409 | links = links->next; | |
dfc5606d | 3410 | } |
35938150 AE |
3411 | |
3412 | /* Advance to the next entry in the snapshot context */ | |
3413 | ||
3414 | index++; | |
dfc5606d | 3415 | } |
9fcbb800 | 3416 | dout("%s: done\n", __func__); |
dfc5606d YS |
3417 | |
3418 | return 0; | |
3419 | } | |
3420 | ||
304f6808 AE |
3421 | /* |
3422 | * Scan the list of snapshots and register the devices for any that | |
3423 | * have not already been registered. | |
3424 | */ | |
3425 | static int rbd_dev_snaps_register(struct rbd_device *rbd_dev) | |
3426 | { | |
3427 | struct rbd_snap *snap; | |
3428 | int ret = 0; | |
3429 | ||
37206ee5 | 3430 | dout("%s:\n", __func__); |
86ff77bb AE |
3431 | if (WARN_ON(!device_is_registered(&rbd_dev->dev))) |
3432 | return -EIO; | |
304f6808 AE |
3433 | |
3434 | list_for_each_entry(snap, &rbd_dev->snaps, node) { | |
3435 | if (!rbd_snap_registered(snap)) { | |
3436 | ret = rbd_register_snap_dev(snap, &rbd_dev->dev); | |
3437 | if (ret < 0) | |
3438 | break; | |
3439 | } | |
3440 | } | |
3441 | dout("%s: returning %d\n", __func__, ret); | |
3442 | ||
3443 | return ret; | |
3444 | } | |
3445 | ||
dfc5606d YS |
3446 | static int rbd_bus_add_dev(struct rbd_device *rbd_dev) |
3447 | { | |
dfc5606d | 3448 | struct device *dev; |
cd789ab9 | 3449 | int ret; |
dfc5606d YS |
3450 | |
3451 | mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING); | |
dfc5606d | 3452 | |
cd789ab9 | 3453 | dev = &rbd_dev->dev; |
dfc5606d YS |
3454 | dev->bus = &rbd_bus_type; |
3455 | dev->type = &rbd_device_type; | |
3456 | dev->parent = &rbd_root_dev; | |
3457 | dev->release = rbd_dev_release; | |
de71a297 | 3458 | dev_set_name(dev, "%d", rbd_dev->dev_id); |
dfc5606d | 3459 | ret = device_register(dev); |
dfc5606d | 3460 | |
dfc5606d | 3461 | mutex_unlock(&ctl_mutex); |
cd789ab9 | 3462 | |
dfc5606d | 3463 | return ret; |
602adf40 YS |
3464 | } |
3465 | ||
dfc5606d YS |
3466 | static void rbd_bus_del_dev(struct rbd_device *rbd_dev) |
3467 | { | |
3468 | device_unregister(&rbd_dev->dev); | |
3469 | } | |
3470 | ||
e2839308 | 3471 | static atomic64_t rbd_dev_id_max = ATOMIC64_INIT(0); |
1ddbe94e AE |
3472 | |
3473 | /* | |
499afd5b AE |
3474 | * Get a unique rbd identifier for the given new rbd_dev, and add |
3475 | * the rbd_dev to the global list. The minimum rbd id is 1. | |
1ddbe94e | 3476 | */ |
e2839308 | 3477 | static void rbd_dev_id_get(struct rbd_device *rbd_dev) |
b7f23c36 | 3478 | { |
e2839308 | 3479 | rbd_dev->dev_id = atomic64_inc_return(&rbd_dev_id_max); |
499afd5b AE |
3480 | |
3481 | spin_lock(&rbd_dev_list_lock); | |
3482 | list_add_tail(&rbd_dev->node, &rbd_dev_list); | |
3483 | spin_unlock(&rbd_dev_list_lock); | |
e2839308 AE |
3484 | dout("rbd_dev %p given dev id %llu\n", rbd_dev, |
3485 | (unsigned long long) rbd_dev->dev_id); | |
1ddbe94e | 3486 | } |
b7f23c36 | 3487 | |
1ddbe94e | 3488 | /* |
499afd5b AE |
3489 | * Remove an rbd_dev from the global list, and record that its |
3490 | * identifier is no longer in use. | |
1ddbe94e | 3491 | */ |
e2839308 | 3492 | static void rbd_dev_id_put(struct rbd_device *rbd_dev) |
1ddbe94e | 3493 | { |
d184f6bf | 3494 | struct list_head *tmp; |
de71a297 | 3495 | int rbd_id = rbd_dev->dev_id; |
d184f6bf AE |
3496 | int max_id; |
3497 | ||
aafb230e | 3498 | rbd_assert(rbd_id > 0); |
499afd5b | 3499 | |
e2839308 AE |
3500 | dout("rbd_dev %p released dev id %llu\n", rbd_dev, |
3501 | (unsigned long long) rbd_dev->dev_id); | |
499afd5b AE |
3502 | spin_lock(&rbd_dev_list_lock); |
3503 | list_del_init(&rbd_dev->node); | |
d184f6bf AE |
3504 | |
3505 | /* | |
3506 | * If the id being "put" is not the current maximum, there | |
3507 | * is nothing special we need to do. | |
3508 | */ | |
e2839308 | 3509 | if (rbd_id != atomic64_read(&rbd_dev_id_max)) { |
d184f6bf AE |
3510 | spin_unlock(&rbd_dev_list_lock); |
3511 | return; | |
3512 | } | |
3513 | ||
3514 | /* | |
3515 | * We need to update the current maximum id. Search the | |
3516 | * list to find out what it is. We're more likely to find | |
3517 | * the maximum at the end, so search the list backward. | |
3518 | */ | |
3519 | max_id = 0; | |
3520 | list_for_each_prev(tmp, &rbd_dev_list) { | |
3521 | struct rbd_device *rbd_dev; | |
3522 | ||
3523 | rbd_dev = list_entry(tmp, struct rbd_device, node); | |
b213e0b1 AE |
3524 | if (rbd_dev->dev_id > max_id) |
3525 | max_id = rbd_dev->dev_id; | |
d184f6bf | 3526 | } |
499afd5b | 3527 | spin_unlock(&rbd_dev_list_lock); |
b7f23c36 | 3528 | |
1ddbe94e | 3529 | /* |
e2839308 | 3530 | * The max id could have been updated by rbd_dev_id_get(), in |
d184f6bf AE |
3531 | * which case it now accurately reflects the new maximum. |
3532 | * Be careful not to overwrite the maximum value in that | |
3533 | * case. | |
1ddbe94e | 3534 | */ |
e2839308 AE |
3535 | atomic64_cmpxchg(&rbd_dev_id_max, rbd_id, max_id); |
3536 | dout(" max dev id has been reset\n"); | |
b7f23c36 AE |
3537 | } |
3538 | ||
e28fff26 AE |
3539 | /* |
3540 | * Skips over white space at *buf, and updates *buf to point to the | |
3541 | * first found non-space character (if any). Returns the length of | |
593a9e7b AE |
3542 | * the token (string of non-white space characters) found. Note |
3543 | * that *buf must be terminated with '\0'. | |
e28fff26 AE |
3544 | */ |
3545 | static inline size_t next_token(const char **buf) | |
3546 | { | |
3547 | /* | |
3548 | * These are the characters that produce nonzero for | |
3549 | * isspace() in the "C" and "POSIX" locales. | |
3550 | */ | |
3551 | const char *spaces = " \f\n\r\t\v"; | |
3552 | ||
3553 | *buf += strspn(*buf, spaces); /* Find start of token */ | |
3554 | ||
3555 | return strcspn(*buf, spaces); /* Return token length */ | |
3556 | } | |
3557 | ||
3558 | /* | |
3559 | * Finds the next token in *buf, and if the provided token buffer is | |
3560 | * big enough, copies the found token into it. The result, if | |
593a9e7b AE |
3561 | * copied, is guaranteed to be terminated with '\0'. Note that *buf |
3562 | * must be terminated with '\0' on entry. | |
e28fff26 AE |
3563 | * |
3564 | * Returns the length of the token found (not including the '\0'). | |
3565 | * Return value will be 0 if no token is found, and it will be >= | |
3566 | * token_size if the token would not fit. | |
3567 | * | |
593a9e7b | 3568 | * The *buf pointer will be updated to point beyond the end of the |
e28fff26 AE |
3569 | * found token. Note that this occurs even if the token buffer is |
3570 | * too small to hold it. | |
3571 | */ | |
3572 | static inline size_t copy_token(const char **buf, | |
3573 | char *token, | |
3574 | size_t token_size) | |
3575 | { | |
3576 | size_t len; | |
3577 | ||
3578 | len = next_token(buf); | |
3579 | if (len < token_size) { | |
3580 | memcpy(token, *buf, len); | |
3581 | *(token + len) = '\0'; | |
3582 | } | |
3583 | *buf += len; | |
3584 | ||
3585 | return len; | |
3586 | } | |
3587 | ||
ea3352f4 AE |
3588 | /* |
3589 | * Finds the next token in *buf, dynamically allocates a buffer big | |
3590 | * enough to hold a copy of it, and copies the token into the new | |
3591 | * buffer. The copy is guaranteed to be terminated with '\0'. Note | |
3592 | * that a duplicate buffer is created even for a zero-length token. | |
3593 | * | |
3594 | * Returns a pointer to the newly-allocated duplicate, or a null | |
3595 | * pointer if memory for the duplicate was not available. If | |
3596 | * the lenp argument is a non-null pointer, the length of the token | |
3597 | * (not including the '\0') is returned in *lenp. | |
3598 | * | |
3599 | * If successful, the *buf pointer will be updated to point beyond | |
3600 | * the end of the found token. | |
3601 | * | |
3602 | * Note: uses GFP_KERNEL for allocation. | |
3603 | */ | |
3604 | static inline char *dup_token(const char **buf, size_t *lenp) | |
3605 | { | |
3606 | char *dup; | |
3607 | size_t len; | |
3608 | ||
3609 | len = next_token(buf); | |
4caf35f9 | 3610 | dup = kmemdup(*buf, len + 1, GFP_KERNEL); |
ea3352f4 AE |
3611 | if (!dup) |
3612 | return NULL; | |
ea3352f4 AE |
3613 | *(dup + len) = '\0'; |
3614 | *buf += len; | |
3615 | ||
3616 | if (lenp) | |
3617 | *lenp = len; | |
3618 | ||
3619 | return dup; | |
3620 | } | |
3621 | ||
a725f65e | 3622 | /* |
859c31df AE |
3623 | * Parse the options provided for an "rbd add" (i.e., rbd image |
3624 | * mapping) request. These arrive via a write to /sys/bus/rbd/add, | |
3625 | * and the data written is passed here via a NUL-terminated buffer. | |
3626 | * Returns 0 if successful or an error code otherwise. | |
d22f76e7 | 3627 | * |
859c31df AE |
3628 | * The information extracted from these options is recorded in |
3629 | * the other parameters which return dynamically-allocated | |
3630 | * structures: | |
3631 | * ceph_opts | |
3632 | * The address of a pointer that will refer to a ceph options | |
3633 | * structure. Caller must release the returned pointer using | |
3634 | * ceph_destroy_options() when it is no longer needed. | |
3635 | * rbd_opts | |
3636 | * Address of an rbd options pointer. Fully initialized by | |
3637 | * this function; caller must release with kfree(). | |
3638 | * spec | |
3639 | * Address of an rbd image specification pointer. Fully | |
3640 | * initialized by this function based on parsed options. | |
3641 | * Caller must release with rbd_spec_put(). | |
3642 | * | |
3643 | * The options passed take this form: | |
3644 | * <mon_addrs> <options> <pool_name> <image_name> [<snap_id>] | |
3645 | * where: | |
3646 | * <mon_addrs> | |
3647 | * A comma-separated list of one or more monitor addresses. | |
3648 | * A monitor address is an ip address, optionally followed | |
3649 | * by a port number (separated by a colon). | |
3650 | * I.e.: ip1[:port1][,ip2[:port2]...] | |
3651 | * <options> | |
3652 | * A comma-separated list of ceph and/or rbd options. | |
3653 | * <pool_name> | |
3654 | * The name of the rados pool containing the rbd image. | |
3655 | * <image_name> | |
3656 | * The name of the image in that pool to map. | |
3657 | * <snap_id> | |
3658 | * An optional snapshot id. If provided, the mapping will | |
3659 | * present data from the image at the time that snapshot was | |
3660 | * created. The image head is used if no snapshot id is | |
3661 | * provided. Snapshot mappings are always read-only. | |
a725f65e | 3662 | */ |
859c31df | 3663 | static int rbd_add_parse_args(const char *buf, |
dc79b113 | 3664 | struct ceph_options **ceph_opts, |
859c31df AE |
3665 | struct rbd_options **opts, |
3666 | struct rbd_spec **rbd_spec) | |
e28fff26 | 3667 | { |
d22f76e7 | 3668 | size_t len; |
859c31df | 3669 | char *options; |
0ddebc0c AE |
3670 | const char *mon_addrs; |
3671 | size_t mon_addrs_size; | |
859c31df | 3672 | struct rbd_spec *spec = NULL; |
4e9afeba | 3673 | struct rbd_options *rbd_opts = NULL; |
859c31df | 3674 | struct ceph_options *copts; |
dc79b113 | 3675 | int ret; |
e28fff26 AE |
3676 | |
3677 | /* The first four tokens are required */ | |
3678 | ||
7ef3214a | 3679 | len = next_token(&buf); |
4fb5d671 AE |
3680 | if (!len) { |
3681 | rbd_warn(NULL, "no monitor address(es) provided"); | |
3682 | return -EINVAL; | |
3683 | } | |
0ddebc0c | 3684 | mon_addrs = buf; |
f28e565a | 3685 | mon_addrs_size = len + 1; |
7ef3214a | 3686 | buf += len; |
a725f65e | 3687 | |
dc79b113 | 3688 | ret = -EINVAL; |
f28e565a AE |
3689 | options = dup_token(&buf, NULL); |
3690 | if (!options) | |
dc79b113 | 3691 | return -ENOMEM; |
4fb5d671 AE |
3692 | if (!*options) { |
3693 | rbd_warn(NULL, "no options provided"); | |
3694 | goto out_err; | |
3695 | } | |
e28fff26 | 3696 | |
859c31df AE |
3697 | spec = rbd_spec_alloc(); |
3698 | if (!spec) | |
f28e565a | 3699 | goto out_mem; |
859c31df AE |
3700 | |
3701 | spec->pool_name = dup_token(&buf, NULL); | |
3702 | if (!spec->pool_name) | |
3703 | goto out_mem; | |
4fb5d671 AE |
3704 | if (!*spec->pool_name) { |
3705 | rbd_warn(NULL, "no pool name provided"); | |
3706 | goto out_err; | |
3707 | } | |
e28fff26 | 3708 | |
69e7a02f | 3709 | spec->image_name = dup_token(&buf, NULL); |
859c31df | 3710 | if (!spec->image_name) |
f28e565a | 3711 | goto out_mem; |
4fb5d671 AE |
3712 | if (!*spec->image_name) { |
3713 | rbd_warn(NULL, "no image name provided"); | |
3714 | goto out_err; | |
3715 | } | |
d4b125e9 | 3716 | |
f28e565a AE |
3717 | /* |
3718 | * Snapshot name is optional; default is to use "-" | |
3719 | * (indicating the head/no snapshot). | |
3720 | */ | |
3feeb894 | 3721 | len = next_token(&buf); |
820a5f3e | 3722 | if (!len) { |
3feeb894 AE |
3723 | buf = RBD_SNAP_HEAD_NAME; /* No snapshot supplied */ |
3724 | len = sizeof (RBD_SNAP_HEAD_NAME) - 1; | |
f28e565a | 3725 | } else if (len > RBD_MAX_SNAP_NAME_LEN) { |
dc79b113 | 3726 | ret = -ENAMETOOLONG; |
f28e565a | 3727 | goto out_err; |
849b4260 | 3728 | } |
4caf35f9 | 3729 | spec->snap_name = kmemdup(buf, len + 1, GFP_KERNEL); |
859c31df | 3730 | if (!spec->snap_name) |
f28e565a | 3731 | goto out_mem; |
859c31df | 3732 | *(spec->snap_name + len) = '\0'; |
e5c35534 | 3733 | |
0ddebc0c | 3734 | /* Initialize all rbd options to the defaults */ |
e28fff26 | 3735 | |
4e9afeba AE |
3736 | rbd_opts = kzalloc(sizeof (*rbd_opts), GFP_KERNEL); |
3737 | if (!rbd_opts) | |
3738 | goto out_mem; | |
3739 | ||
3740 | rbd_opts->read_only = RBD_READ_ONLY_DEFAULT; | |
d22f76e7 | 3741 | |
859c31df | 3742 | copts = ceph_parse_options(options, mon_addrs, |
0ddebc0c | 3743 | mon_addrs + mon_addrs_size - 1, |
4e9afeba | 3744 | parse_rbd_opts_token, rbd_opts); |
859c31df AE |
3745 | if (IS_ERR(copts)) { |
3746 | ret = PTR_ERR(copts); | |
dc79b113 AE |
3747 | goto out_err; |
3748 | } | |
859c31df AE |
3749 | kfree(options); |
3750 | ||
3751 | *ceph_opts = copts; | |
4e9afeba | 3752 | *opts = rbd_opts; |
859c31df | 3753 | *rbd_spec = spec; |
0ddebc0c | 3754 | |
dc79b113 | 3755 | return 0; |
f28e565a | 3756 | out_mem: |
dc79b113 | 3757 | ret = -ENOMEM; |
d22f76e7 | 3758 | out_err: |
859c31df AE |
3759 | kfree(rbd_opts); |
3760 | rbd_spec_put(spec); | |
f28e565a | 3761 | kfree(options); |
d22f76e7 | 3762 | |
dc79b113 | 3763 | return ret; |
a725f65e AE |
3764 | } |
3765 | ||
589d30e0 AE |
3766 | /* |
3767 | * An rbd format 2 image has a unique identifier, distinct from the | |
3768 | * name given to it by the user. Internally, that identifier is | |
3769 | * what's used to specify the names of objects related to the image. | |
3770 | * | |
3771 | * A special "rbd id" object is used to map an rbd image name to its | |
3772 | * id. If that object doesn't exist, then there is no v2 rbd image | |
3773 | * with the supplied name. | |
3774 | * | |
3775 | * This function will record the given rbd_dev's image_id field if | |
3776 | * it can be determined, and in that case will return 0. If any | |
3777 | * errors occur a negative errno will be returned and the rbd_dev's | |
3778 | * image_id field will be unchanged (and should be NULL). | |
3779 | */ | |
3780 | static int rbd_dev_image_id(struct rbd_device *rbd_dev) | |
3781 | { | |
3782 | int ret; | |
3783 | size_t size; | |
3784 | char *object_name; | |
3785 | void *response; | |
3786 | void *p; | |
3787 | ||
2c0d0a10 AE |
3788 | /* |
3789 | * When probing a parent image, the image id is already | |
3790 | * known (and the image name likely is not). There's no | |
3791 | * need to fetch the image id again in this case. | |
3792 | */ | |
3793 | if (rbd_dev->spec->image_id) | |
3794 | return 0; | |
3795 | ||
589d30e0 AE |
3796 | /* |
3797 | * First, see if the format 2 image id file exists, and if | |
3798 | * so, get the image's persistent id from it. | |
3799 | */ | |
69e7a02f | 3800 | size = sizeof (RBD_ID_PREFIX) + strlen(rbd_dev->spec->image_name); |
589d30e0 AE |
3801 | object_name = kmalloc(size, GFP_NOIO); |
3802 | if (!object_name) | |
3803 | return -ENOMEM; | |
0d7dbfce | 3804 | sprintf(object_name, "%s%s", RBD_ID_PREFIX, rbd_dev->spec->image_name); |
589d30e0 AE |
3805 | dout("rbd id object name is %s\n", object_name); |
3806 | ||
3807 | /* Response will be an encoded string, which includes a length */ | |
3808 | ||
3809 | size = sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX; | |
3810 | response = kzalloc(size, GFP_NOIO); | |
3811 | if (!response) { | |
3812 | ret = -ENOMEM; | |
3813 | goto out; | |
3814 | } | |
3815 | ||
36be9a76 | 3816 | ret = rbd_obj_method_sync(rbd_dev, object_name, |
589d30e0 AE |
3817 | "rbd", "get_id", |
3818 | NULL, 0, | |
07b2391f | 3819 | response, RBD_IMAGE_ID_LEN_MAX, NULL); |
36be9a76 | 3820 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
589d30e0 AE |
3821 | if (ret < 0) |
3822 | goto out; | |
3823 | ||
3824 | p = response; | |
0d7dbfce | 3825 | rbd_dev->spec->image_id = ceph_extract_encoded_string(&p, |
589d30e0 | 3826 | p + RBD_IMAGE_ID_LEN_MAX, |
979ed480 | 3827 | NULL, GFP_NOIO); |
0d7dbfce AE |
3828 | if (IS_ERR(rbd_dev->spec->image_id)) { |
3829 | ret = PTR_ERR(rbd_dev->spec->image_id); | |
3830 | rbd_dev->spec->image_id = NULL; | |
589d30e0 | 3831 | } else { |
0d7dbfce | 3832 | dout("image_id is %s\n", rbd_dev->spec->image_id); |
589d30e0 AE |
3833 | } |
3834 | out: | |
3835 | kfree(response); | |
3836 | kfree(object_name); | |
3837 | ||
3838 | return ret; | |
3839 | } | |
3840 | ||
a30b71b9 AE |
3841 | static int rbd_dev_v1_probe(struct rbd_device *rbd_dev) |
3842 | { | |
3843 | int ret; | |
3844 | size_t size; | |
3845 | ||
3846 | /* Version 1 images have no id; empty string is used */ | |
3847 | ||
0d7dbfce AE |
3848 | rbd_dev->spec->image_id = kstrdup("", GFP_KERNEL); |
3849 | if (!rbd_dev->spec->image_id) | |
a30b71b9 | 3850 | return -ENOMEM; |
a30b71b9 AE |
3851 | |
3852 | /* Record the header object name for this rbd image. */ | |
3853 | ||
69e7a02f | 3854 | size = strlen(rbd_dev->spec->image_name) + sizeof (RBD_SUFFIX); |
a30b71b9 AE |
3855 | rbd_dev->header_name = kmalloc(size, GFP_KERNEL); |
3856 | if (!rbd_dev->header_name) { | |
3857 | ret = -ENOMEM; | |
3858 | goto out_err; | |
3859 | } | |
0d7dbfce AE |
3860 | sprintf(rbd_dev->header_name, "%s%s", |
3861 | rbd_dev->spec->image_name, RBD_SUFFIX); | |
a30b71b9 AE |
3862 | |
3863 | /* Populate rbd image metadata */ | |
3864 | ||
3865 | ret = rbd_read_header(rbd_dev, &rbd_dev->header); | |
3866 | if (ret < 0) | |
3867 | goto out_err; | |
86b00e0d AE |
3868 | |
3869 | /* Version 1 images have no parent (no layering) */ | |
3870 | ||
3871 | rbd_dev->parent_spec = NULL; | |
3872 | rbd_dev->parent_overlap = 0; | |
3873 | ||
a30b71b9 AE |
3874 | rbd_dev->image_format = 1; |
3875 | ||
3876 | dout("discovered version 1 image, header name is %s\n", | |
3877 | rbd_dev->header_name); | |
3878 | ||
3879 | return 0; | |
3880 | ||
3881 | out_err: | |
3882 | kfree(rbd_dev->header_name); | |
3883 | rbd_dev->header_name = NULL; | |
0d7dbfce AE |
3884 | kfree(rbd_dev->spec->image_id); |
3885 | rbd_dev->spec->image_id = NULL; | |
a30b71b9 AE |
3886 | |
3887 | return ret; | |
3888 | } | |
3889 | ||
3890 | static int rbd_dev_v2_probe(struct rbd_device *rbd_dev) | |
3891 | { | |
3892 | size_t size; | |
9d475de5 | 3893 | int ret; |
6e14b1a6 | 3894 | u64 ver = 0; |
a30b71b9 AE |
3895 | |
3896 | /* | |
3897 | * Image id was filled in by the caller. Record the header | |
3898 | * object name for this rbd image. | |
3899 | */ | |
979ed480 | 3900 | size = sizeof (RBD_HEADER_PREFIX) + strlen(rbd_dev->spec->image_id); |
a30b71b9 AE |
3901 | rbd_dev->header_name = kmalloc(size, GFP_KERNEL); |
3902 | if (!rbd_dev->header_name) | |
3903 | return -ENOMEM; | |
3904 | sprintf(rbd_dev->header_name, "%s%s", | |
0d7dbfce | 3905 | RBD_HEADER_PREFIX, rbd_dev->spec->image_id); |
9d475de5 AE |
3906 | |
3907 | /* Get the size and object order for the image */ | |
3908 | ||
3909 | ret = rbd_dev_v2_image_size(rbd_dev); | |
1e130199 AE |
3910 | if (ret < 0) |
3911 | goto out_err; | |
3912 | ||
3913 | /* Get the object prefix (a.k.a. block_name) for the image */ | |
3914 | ||
3915 | ret = rbd_dev_v2_object_prefix(rbd_dev); | |
b1b5402a AE |
3916 | if (ret < 0) |
3917 | goto out_err; | |
3918 | ||
d889140c | 3919 | /* Get the and check features for the image */ |
b1b5402a AE |
3920 | |
3921 | ret = rbd_dev_v2_features(rbd_dev); | |
9d475de5 AE |
3922 | if (ret < 0) |
3923 | goto out_err; | |
35d489f9 | 3924 | |
86b00e0d AE |
3925 | /* If the image supports layering, get the parent info */ |
3926 | ||
3927 | if (rbd_dev->header.features & RBD_FEATURE_LAYERING) { | |
3928 | ret = rbd_dev_v2_parent_info(rbd_dev); | |
3929 | if (ret < 0) | |
3930 | goto out_err; | |
3931 | } | |
3932 | ||
6e14b1a6 AE |
3933 | /* crypto and compression type aren't (yet) supported for v2 images */ |
3934 | ||
3935 | rbd_dev->header.crypt_type = 0; | |
3936 | rbd_dev->header.comp_type = 0; | |
35d489f9 | 3937 | |
6e14b1a6 AE |
3938 | /* Get the snapshot context, plus the header version */ |
3939 | ||
3940 | ret = rbd_dev_v2_snap_context(rbd_dev, &ver); | |
35d489f9 AE |
3941 | if (ret) |
3942 | goto out_err; | |
6e14b1a6 AE |
3943 | rbd_dev->header.obj_version = ver; |
3944 | ||
a30b71b9 AE |
3945 | rbd_dev->image_format = 2; |
3946 | ||
3947 | dout("discovered version 2 image, header name is %s\n", | |
3948 | rbd_dev->header_name); | |
3949 | ||
35152979 | 3950 | return 0; |
9d475de5 | 3951 | out_err: |
86b00e0d AE |
3952 | rbd_dev->parent_overlap = 0; |
3953 | rbd_spec_put(rbd_dev->parent_spec); | |
3954 | rbd_dev->parent_spec = NULL; | |
9d475de5 AE |
3955 | kfree(rbd_dev->header_name); |
3956 | rbd_dev->header_name = NULL; | |
1e130199 AE |
3957 | kfree(rbd_dev->header.object_prefix); |
3958 | rbd_dev->header.object_prefix = NULL; | |
9d475de5 AE |
3959 | |
3960 | return ret; | |
a30b71b9 AE |
3961 | } |
3962 | ||
83a06263 AE |
3963 | static int rbd_dev_probe_finish(struct rbd_device *rbd_dev) |
3964 | { | |
3965 | int ret; | |
3966 | ||
3967 | /* no need to lock here, as rbd_dev is not registered yet */ | |
3968 | ret = rbd_dev_snaps_update(rbd_dev); | |
3969 | if (ret) | |
3970 | return ret; | |
3971 | ||
9e15b77d AE |
3972 | ret = rbd_dev_probe_update_spec(rbd_dev); |
3973 | if (ret) | |
3974 | goto err_out_snaps; | |
3975 | ||
83a06263 AE |
3976 | ret = rbd_dev_set_mapping(rbd_dev); |
3977 | if (ret) | |
3978 | goto err_out_snaps; | |
3979 | ||
3980 | /* generate unique id: find highest unique id, add one */ | |
3981 | rbd_dev_id_get(rbd_dev); | |
3982 | ||
3983 | /* Fill in the device name, now that we have its id. */ | |
3984 | BUILD_BUG_ON(DEV_NAME_LEN | |
3985 | < sizeof (RBD_DRV_NAME) + MAX_INT_FORMAT_WIDTH); | |
3986 | sprintf(rbd_dev->name, "%s%d", RBD_DRV_NAME, rbd_dev->dev_id); | |
3987 | ||
3988 | /* Get our block major device number. */ | |
3989 | ||
3990 | ret = register_blkdev(0, rbd_dev->name); | |
3991 | if (ret < 0) | |
3992 | goto err_out_id; | |
3993 | rbd_dev->major = ret; | |
3994 | ||
3995 | /* Set up the blkdev mapping. */ | |
3996 | ||
3997 | ret = rbd_init_disk(rbd_dev); | |
3998 | if (ret) | |
3999 | goto err_out_blkdev; | |
4000 | ||
4001 | ret = rbd_bus_add_dev(rbd_dev); | |
4002 | if (ret) | |
4003 | goto err_out_disk; | |
4004 | ||
4005 | /* | |
4006 | * At this point cleanup in the event of an error is the job | |
4007 | * of the sysfs code (initiated by rbd_bus_del_dev()). | |
4008 | */ | |
4009 | down_write(&rbd_dev->header_rwsem); | |
4010 | ret = rbd_dev_snaps_register(rbd_dev); | |
4011 | up_write(&rbd_dev->header_rwsem); | |
4012 | if (ret) | |
4013 | goto err_out_bus; | |
4014 | ||
9969ebc5 | 4015 | ret = rbd_dev_header_watch_sync(rbd_dev, 1); |
83a06263 AE |
4016 | if (ret) |
4017 | goto err_out_bus; | |
4018 | ||
4019 | /* Everything's ready. Announce the disk to the world. */ | |
4020 | ||
4021 | add_disk(rbd_dev->disk); | |
4022 | ||
4023 | pr_info("%s: added with size 0x%llx\n", rbd_dev->disk->disk_name, | |
4024 | (unsigned long long) rbd_dev->mapping.size); | |
4025 | ||
4026 | return ret; | |
4027 | err_out_bus: | |
4028 | /* this will also clean up rest of rbd_dev stuff */ | |
4029 | ||
4030 | rbd_bus_del_dev(rbd_dev); | |
4031 | ||
4032 | return ret; | |
4033 | err_out_disk: | |
4034 | rbd_free_disk(rbd_dev); | |
4035 | err_out_blkdev: | |
4036 | unregister_blkdev(rbd_dev->major, rbd_dev->name); | |
4037 | err_out_id: | |
4038 | rbd_dev_id_put(rbd_dev); | |
4039 | err_out_snaps: | |
4040 | rbd_remove_all_snaps(rbd_dev); | |
4041 | ||
4042 | return ret; | |
4043 | } | |
4044 | ||
a30b71b9 AE |
4045 | /* |
4046 | * Probe for the existence of the header object for the given rbd | |
4047 | * device. For format 2 images this includes determining the image | |
4048 | * id. | |
4049 | */ | |
4050 | static int rbd_dev_probe(struct rbd_device *rbd_dev) | |
4051 | { | |
4052 | int ret; | |
4053 | ||
4054 | /* | |
4055 | * Get the id from the image id object. If it's not a | |
4056 | * format 2 image, we'll get ENOENT back, and we'll assume | |
4057 | * it's a format 1 image. | |
4058 | */ | |
4059 | ret = rbd_dev_image_id(rbd_dev); | |
4060 | if (ret) | |
4061 | ret = rbd_dev_v1_probe(rbd_dev); | |
4062 | else | |
4063 | ret = rbd_dev_v2_probe(rbd_dev); | |
83a06263 | 4064 | if (ret) { |
a30b71b9 AE |
4065 | dout("probe failed, returning %d\n", ret); |
4066 | ||
83a06263 AE |
4067 | return ret; |
4068 | } | |
4069 | ||
4070 | ret = rbd_dev_probe_finish(rbd_dev); | |
4071 | if (ret) | |
4072 | rbd_header_free(&rbd_dev->header); | |
4073 | ||
a30b71b9 AE |
4074 | return ret; |
4075 | } | |
4076 | ||
59c2be1e YS |
4077 | static ssize_t rbd_add(struct bus_type *bus, |
4078 | const char *buf, | |
4079 | size_t count) | |
602adf40 | 4080 | { |
cb8627c7 | 4081 | struct rbd_device *rbd_dev = NULL; |
dc79b113 | 4082 | struct ceph_options *ceph_opts = NULL; |
4e9afeba | 4083 | struct rbd_options *rbd_opts = NULL; |
859c31df | 4084 | struct rbd_spec *spec = NULL; |
9d3997fd | 4085 | struct rbd_client *rbdc; |
27cc2594 AE |
4086 | struct ceph_osd_client *osdc; |
4087 | int rc = -ENOMEM; | |
602adf40 YS |
4088 | |
4089 | if (!try_module_get(THIS_MODULE)) | |
4090 | return -ENODEV; | |
4091 | ||
602adf40 | 4092 | /* parse add command */ |
859c31df | 4093 | rc = rbd_add_parse_args(buf, &ceph_opts, &rbd_opts, &spec); |
dc79b113 | 4094 | if (rc < 0) |
bd4ba655 | 4095 | goto err_out_module; |
78cea76e | 4096 | |
9d3997fd AE |
4097 | rbdc = rbd_get_client(ceph_opts); |
4098 | if (IS_ERR(rbdc)) { | |
4099 | rc = PTR_ERR(rbdc); | |
0ddebc0c | 4100 | goto err_out_args; |
9d3997fd | 4101 | } |
c53d5893 | 4102 | ceph_opts = NULL; /* rbd_dev client now owns this */ |
602adf40 | 4103 | |
602adf40 | 4104 | /* pick the pool */ |
9d3997fd | 4105 | osdc = &rbdc->client->osdc; |
859c31df | 4106 | rc = ceph_pg_poolid_by_name(osdc->osdmap, spec->pool_name); |
602adf40 YS |
4107 | if (rc < 0) |
4108 | goto err_out_client; | |
859c31df AE |
4109 | spec->pool_id = (u64) rc; |
4110 | ||
0903e875 AE |
4111 | /* The ceph file layout needs to fit pool id in 32 bits */ |
4112 | ||
4113 | if (WARN_ON(spec->pool_id > (u64) U32_MAX)) { | |
4114 | rc = -EIO; | |
4115 | goto err_out_client; | |
4116 | } | |
4117 | ||
c53d5893 | 4118 | rbd_dev = rbd_dev_create(rbdc, spec); |
bd4ba655 AE |
4119 | if (!rbd_dev) |
4120 | goto err_out_client; | |
c53d5893 AE |
4121 | rbdc = NULL; /* rbd_dev now owns this */ |
4122 | spec = NULL; /* rbd_dev now owns this */ | |
602adf40 | 4123 | |
bd4ba655 | 4124 | rbd_dev->mapping.read_only = rbd_opts->read_only; |
c53d5893 AE |
4125 | kfree(rbd_opts); |
4126 | rbd_opts = NULL; /* done with this */ | |
bd4ba655 | 4127 | |
a30b71b9 AE |
4128 | rc = rbd_dev_probe(rbd_dev); |
4129 | if (rc < 0) | |
c53d5893 | 4130 | goto err_out_rbd_dev; |
05fd6f6f | 4131 | |
602adf40 | 4132 | return count; |
c53d5893 AE |
4133 | err_out_rbd_dev: |
4134 | rbd_dev_destroy(rbd_dev); | |
bd4ba655 | 4135 | err_out_client: |
9d3997fd | 4136 | rbd_put_client(rbdc); |
0ddebc0c | 4137 | err_out_args: |
78cea76e AE |
4138 | if (ceph_opts) |
4139 | ceph_destroy_options(ceph_opts); | |
4e9afeba | 4140 | kfree(rbd_opts); |
859c31df | 4141 | rbd_spec_put(spec); |
bd4ba655 AE |
4142 | err_out_module: |
4143 | module_put(THIS_MODULE); | |
27cc2594 | 4144 | |
602adf40 | 4145 | dout("Error adding device %s\n", buf); |
27cc2594 AE |
4146 | |
4147 | return (ssize_t) rc; | |
602adf40 YS |
4148 | } |
4149 | ||
de71a297 | 4150 | static struct rbd_device *__rbd_get_dev(unsigned long dev_id) |
602adf40 YS |
4151 | { |
4152 | struct list_head *tmp; | |
4153 | struct rbd_device *rbd_dev; | |
4154 | ||
e124a82f | 4155 | spin_lock(&rbd_dev_list_lock); |
602adf40 YS |
4156 | list_for_each(tmp, &rbd_dev_list) { |
4157 | rbd_dev = list_entry(tmp, struct rbd_device, node); | |
de71a297 | 4158 | if (rbd_dev->dev_id == dev_id) { |
e124a82f | 4159 | spin_unlock(&rbd_dev_list_lock); |
602adf40 | 4160 | return rbd_dev; |
e124a82f | 4161 | } |
602adf40 | 4162 | } |
e124a82f | 4163 | spin_unlock(&rbd_dev_list_lock); |
602adf40 YS |
4164 | return NULL; |
4165 | } | |
4166 | ||
dfc5606d | 4167 | static void rbd_dev_release(struct device *dev) |
602adf40 | 4168 | { |
593a9e7b | 4169 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
602adf40 | 4170 | |
59c2be1e | 4171 | if (rbd_dev->watch_event) |
9969ebc5 | 4172 | rbd_dev_header_watch_sync(rbd_dev, 0); |
602adf40 YS |
4173 | |
4174 | /* clean up and free blkdev */ | |
4175 | rbd_free_disk(rbd_dev); | |
4176 | unregister_blkdev(rbd_dev->major, rbd_dev->name); | |
32eec68d | 4177 | |
2ac4e75d AE |
4178 | /* release allocated disk header fields */ |
4179 | rbd_header_free(&rbd_dev->header); | |
4180 | ||
32eec68d | 4181 | /* done with the id, and with the rbd_dev */ |
e2839308 | 4182 | rbd_dev_id_put(rbd_dev); |
c53d5893 AE |
4183 | rbd_assert(rbd_dev->rbd_client != NULL); |
4184 | rbd_dev_destroy(rbd_dev); | |
602adf40 YS |
4185 | |
4186 | /* release module ref */ | |
4187 | module_put(THIS_MODULE); | |
602adf40 YS |
4188 | } |
4189 | ||
dfc5606d YS |
4190 | static ssize_t rbd_remove(struct bus_type *bus, |
4191 | const char *buf, | |
4192 | size_t count) | |
602adf40 YS |
4193 | { |
4194 | struct rbd_device *rbd_dev = NULL; | |
4195 | int target_id, rc; | |
4196 | unsigned long ul; | |
4197 | int ret = count; | |
4198 | ||
4199 | rc = strict_strtoul(buf, 10, &ul); | |
4200 | if (rc) | |
4201 | return rc; | |
4202 | ||
4203 | /* convert to int; abort if we lost anything in the conversion */ | |
4204 | target_id = (int) ul; | |
4205 | if (target_id != ul) | |
4206 | return -EINVAL; | |
4207 | ||
4208 | mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING); | |
4209 | ||
4210 | rbd_dev = __rbd_get_dev(target_id); | |
4211 | if (!rbd_dev) { | |
4212 | ret = -ENOENT; | |
4213 | goto done; | |
42382b70 AE |
4214 | } |
4215 | ||
a14ea269 | 4216 | spin_lock_irq(&rbd_dev->lock); |
b82d167b | 4217 | if (rbd_dev->open_count) |
42382b70 | 4218 | ret = -EBUSY; |
b82d167b AE |
4219 | else |
4220 | set_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags); | |
a14ea269 | 4221 | spin_unlock_irq(&rbd_dev->lock); |
b82d167b | 4222 | if (ret < 0) |
42382b70 | 4223 | goto done; |
602adf40 | 4224 | |
41f38c2b | 4225 | rbd_remove_all_snaps(rbd_dev); |
dfc5606d | 4226 | rbd_bus_del_dev(rbd_dev); |
602adf40 YS |
4227 | |
4228 | done: | |
4229 | mutex_unlock(&ctl_mutex); | |
aafb230e | 4230 | |
602adf40 YS |
4231 | return ret; |
4232 | } | |
4233 | ||
602adf40 YS |
4234 | /* |
4235 | * create control files in sysfs | |
dfc5606d | 4236 | * /sys/bus/rbd/... |
602adf40 YS |
4237 | */ |
4238 | static int rbd_sysfs_init(void) | |
4239 | { | |
dfc5606d | 4240 | int ret; |
602adf40 | 4241 | |
fed4c143 | 4242 | ret = device_register(&rbd_root_dev); |
21079786 | 4243 | if (ret < 0) |
dfc5606d | 4244 | return ret; |
602adf40 | 4245 | |
fed4c143 AE |
4246 | ret = bus_register(&rbd_bus_type); |
4247 | if (ret < 0) | |
4248 | device_unregister(&rbd_root_dev); | |
602adf40 | 4249 | |
602adf40 YS |
4250 | return ret; |
4251 | } | |
4252 | ||
4253 | static void rbd_sysfs_cleanup(void) | |
4254 | { | |
dfc5606d | 4255 | bus_unregister(&rbd_bus_type); |
fed4c143 | 4256 | device_unregister(&rbd_root_dev); |
602adf40 YS |
4257 | } |
4258 | ||
4259 | int __init rbd_init(void) | |
4260 | { | |
4261 | int rc; | |
4262 | ||
1e32d34c AE |
4263 | if (!libceph_compatible(NULL)) { |
4264 | rbd_warn(NULL, "libceph incompatibility (quitting)"); | |
4265 | ||
4266 | return -EINVAL; | |
4267 | } | |
602adf40 YS |
4268 | rc = rbd_sysfs_init(); |
4269 | if (rc) | |
4270 | return rc; | |
f0f8cef5 | 4271 | pr_info("loaded " RBD_DRV_NAME_LONG "\n"); |
602adf40 YS |
4272 | return 0; |
4273 | } | |
4274 | ||
4275 | void __exit rbd_exit(void) | |
4276 | { | |
4277 | rbd_sysfs_cleanup(); | |
4278 | } | |
4279 | ||
4280 | module_init(rbd_init); | |
4281 | module_exit(rbd_exit); | |
4282 | ||
4283 | MODULE_AUTHOR("Sage Weil <sage@newdream.net>"); | |
4284 | MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>"); | |
4285 | MODULE_DESCRIPTION("rados block device"); | |
4286 | ||
4287 | /* following authorship retained from original osdblk.c */ | |
4288 | MODULE_AUTHOR("Jeff Garzik <jeff@garzik.org>"); | |
4289 | ||
4290 | MODULE_LICENSE("GPL"); |