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; | |
1b83bef2 | 199 | int result; |
bf0d5f50 AE |
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 | ||
cc344fa1 | 1144 | static struct ceph_osd_req_op *rbd_osd_req_op_create(u16 opcode, ...) |
8d23bf29 AE |
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 | ||
bf0d5f50 AE |
1267 | static void rbd_obj_request_complete(struct rbd_obj_request *obj_request) |
1268 | { | |
37206ee5 AE |
1269 | dout("%s: obj %p cb %p\n", __func__, obj_request, |
1270 | obj_request->callback); | |
bf0d5f50 AE |
1271 | if (obj_request->callback) |
1272 | obj_request->callback(obj_request); | |
788e2df3 AE |
1273 | else |
1274 | complete_all(&obj_request->completion); | |
bf0d5f50 AE |
1275 | } |
1276 | ||
c47f9371 | 1277 | static void rbd_osd_trivial_callback(struct rbd_obj_request *obj_request) |
39bf2c5d AE |
1278 | { |
1279 | dout("%s: obj %p\n", __func__, obj_request); | |
1280 | obj_request_done_set(obj_request); | |
1281 | } | |
1282 | ||
c47f9371 | 1283 | static void rbd_osd_read_callback(struct rbd_obj_request *obj_request) |
bf0d5f50 | 1284 | { |
37206ee5 | 1285 | dout("%s: obj %p result %d %llu/%llu\n", __func__, obj_request, |
c47f9371 | 1286 | obj_request->result, obj_request->xferred, obj_request->length); |
1b83bef2 SW |
1287 | /* |
1288 | * ENOENT means a hole in the object. We zero-fill the | |
1289 | * entire length of the request. A short read also implies | |
1290 | * zero-fill to the end of the request. Either way we | |
1291 | * update the xferred count to indicate the whole request | |
1292 | * was satisfied. | |
1293 | */ | |
1294 | if (obj_request->result == -ENOENT) { | |
bf0d5f50 AE |
1295 | zero_bio_chain(obj_request->bio_list, 0); |
1296 | obj_request->result = 0; | |
1b83bef2 | 1297 | obj_request->xferred = obj_request->length; |
c47f9371 AE |
1298 | } else if (obj_request->xferred < obj_request->length && |
1299 | !obj_request->result) { | |
1300 | zero_bio_chain(obj_request->bio_list, obj_request->xferred); | |
1301 | obj_request->xferred = obj_request->length; | |
bf0d5f50 | 1302 | } |
07741308 | 1303 | obj_request_done_set(obj_request); |
bf0d5f50 AE |
1304 | } |
1305 | ||
c47f9371 | 1306 | static void rbd_osd_write_callback(struct rbd_obj_request *obj_request) |
bf0d5f50 | 1307 | { |
1b83bef2 SW |
1308 | dout("%s: obj %p result %d %llu\n", __func__, obj_request, |
1309 | obj_request->result, obj_request->length); | |
1310 | /* | |
1311 | * There is no such thing as a successful short write. | |
1312 | * Our xferred value is the number of bytes transferred | |
1313 | * back. Set it to our originally-requested length. | |
1314 | */ | |
1315 | obj_request->xferred = obj_request->length; | |
07741308 | 1316 | obj_request_done_set(obj_request); |
bf0d5f50 AE |
1317 | } |
1318 | ||
fbfab539 AE |
1319 | /* |
1320 | * For a simple stat call there's nothing to do. We'll do more if | |
1321 | * this is part of a write sequence for a layered image. | |
1322 | */ | |
c47f9371 | 1323 | static void rbd_osd_stat_callback(struct rbd_obj_request *obj_request) |
fbfab539 | 1324 | { |
37206ee5 | 1325 | dout("%s: obj %p\n", __func__, obj_request); |
fbfab539 AE |
1326 | obj_request_done_set(obj_request); |
1327 | } | |
1328 | ||
bf0d5f50 AE |
1329 | static void rbd_osd_req_callback(struct ceph_osd_request *osd_req, |
1330 | struct ceph_msg *msg) | |
1331 | { | |
1332 | struct rbd_obj_request *obj_request = osd_req->r_priv; | |
bf0d5f50 AE |
1333 | u16 opcode; |
1334 | ||
37206ee5 | 1335 | dout("%s: osd_req %p msg %p\n", __func__, osd_req, msg); |
bf0d5f50 AE |
1336 | rbd_assert(osd_req == obj_request->osd_req); |
1337 | rbd_assert(!!obj_request->img_request ^ | |
1338 | (obj_request->which == BAD_WHICH)); | |
1339 | ||
1b83bef2 SW |
1340 | if (osd_req->r_result < 0) |
1341 | obj_request->result = osd_req->r_result; | |
bf0d5f50 AE |
1342 | obj_request->version = le64_to_cpu(osd_req->r_reassert_version.version); |
1343 | ||
1b83bef2 | 1344 | WARN_ON(osd_req->r_num_ops != 1); /* For now */ |
bf0d5f50 | 1345 | |
c47f9371 AE |
1346 | /* |
1347 | * We support a 64-bit length, but ultimately it has to be | |
1348 | * passed to blk_end_request(), which takes an unsigned int. | |
1349 | */ | |
1b83bef2 | 1350 | obj_request->xferred = osd_req->r_reply_op_len[0]; |
c47f9371 | 1351 | rbd_assert(obj_request->xferred < (u64) UINT_MAX); |
1b83bef2 | 1352 | opcode = osd_req->r_request_ops[0].op; |
bf0d5f50 AE |
1353 | switch (opcode) { |
1354 | case CEPH_OSD_OP_READ: | |
c47f9371 | 1355 | rbd_osd_read_callback(obj_request); |
bf0d5f50 AE |
1356 | break; |
1357 | case CEPH_OSD_OP_WRITE: | |
c47f9371 | 1358 | rbd_osd_write_callback(obj_request); |
bf0d5f50 | 1359 | break; |
fbfab539 | 1360 | case CEPH_OSD_OP_STAT: |
c47f9371 | 1361 | rbd_osd_stat_callback(obj_request); |
fbfab539 | 1362 | break; |
36be9a76 | 1363 | case CEPH_OSD_OP_CALL: |
b8d70035 | 1364 | case CEPH_OSD_OP_NOTIFY_ACK: |
9969ebc5 | 1365 | case CEPH_OSD_OP_WATCH: |
c47f9371 | 1366 | rbd_osd_trivial_callback(obj_request); |
9969ebc5 | 1367 | break; |
bf0d5f50 AE |
1368 | default: |
1369 | rbd_warn(NULL, "%s: unsupported op %hu\n", | |
1370 | obj_request->object_name, (unsigned short) opcode); | |
1371 | break; | |
1372 | } | |
1373 | ||
07741308 | 1374 | if (obj_request_done_test(obj_request)) |
bf0d5f50 AE |
1375 | rbd_obj_request_complete(obj_request); |
1376 | } | |
1377 | ||
1378 | static struct ceph_osd_request *rbd_osd_req_create( | |
1379 | struct rbd_device *rbd_dev, | |
1380 | bool write_request, | |
1381 | struct rbd_obj_request *obj_request, | |
1382 | struct ceph_osd_req_op *op) | |
1383 | { | |
1384 | struct rbd_img_request *img_request = obj_request->img_request; | |
1385 | struct ceph_snap_context *snapc = NULL; | |
1386 | struct ceph_osd_client *osdc; | |
1387 | struct ceph_osd_request *osd_req; | |
1388 | struct timespec now; | |
1389 | struct timespec *mtime; | |
1390 | u64 snap_id = CEPH_NOSNAP; | |
1391 | u64 offset = obj_request->offset; | |
1392 | u64 length = obj_request->length; | |
1393 | ||
1394 | if (img_request) { | |
1395 | rbd_assert(img_request->write_request == write_request); | |
1396 | if (img_request->write_request) | |
1397 | snapc = img_request->snapc; | |
1398 | else | |
1399 | snap_id = img_request->snap_id; | |
1400 | } | |
1401 | ||
1402 | /* Allocate and initialize the request, for the single op */ | |
1403 | ||
1404 | osdc = &rbd_dev->rbd_client->client->osdc; | |
1405 | osd_req = ceph_osdc_alloc_request(osdc, snapc, 1, false, GFP_ATOMIC); | |
1406 | if (!osd_req) | |
1407 | return NULL; /* ENOMEM */ | |
1408 | ||
1409 | rbd_assert(obj_request_type_valid(obj_request->type)); | |
1410 | switch (obj_request->type) { | |
9969ebc5 AE |
1411 | case OBJ_REQUEST_NODATA: |
1412 | break; /* Nothing to do */ | |
bf0d5f50 AE |
1413 | case OBJ_REQUEST_BIO: |
1414 | rbd_assert(obj_request->bio_list != NULL); | |
1415 | osd_req->r_bio = obj_request->bio_list; | |
bf0d5f50 | 1416 | break; |
788e2df3 AE |
1417 | case OBJ_REQUEST_PAGES: |
1418 | osd_req->r_pages = obj_request->pages; | |
1419 | osd_req->r_num_pages = obj_request->page_count; | |
1420 | osd_req->r_page_alignment = offset & ~PAGE_MASK; | |
1421 | break; | |
bf0d5f50 AE |
1422 | } |
1423 | ||
1424 | if (write_request) { | |
1425 | osd_req->r_flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK; | |
1426 | now = CURRENT_TIME; | |
1427 | mtime = &now; | |
1428 | } else { | |
1429 | osd_req->r_flags = CEPH_OSD_FLAG_READ; | |
1430 | mtime = NULL; /* not needed for reads */ | |
1431 | offset = 0; /* These are not used... */ | |
1432 | length = 0; /* ...for osd read requests */ | |
1433 | } | |
1434 | ||
1435 | osd_req->r_callback = rbd_osd_req_callback; | |
1436 | osd_req->r_priv = obj_request; | |
1437 | ||
1438 | osd_req->r_oid_len = strlen(obj_request->object_name); | |
1439 | rbd_assert(osd_req->r_oid_len < sizeof (osd_req->r_oid)); | |
1440 | memcpy(osd_req->r_oid, obj_request->object_name, osd_req->r_oid_len); | |
1441 | ||
1442 | osd_req->r_file_layout = rbd_dev->layout; /* struct */ | |
1443 | ||
1444 | /* osd_req will get its own reference to snapc (if non-null) */ | |
1445 | ||
1446 | ceph_osdc_build_request(osd_req, offset, length, 1, op, | |
1447 | snapc, snap_id, mtime); | |
1448 | ||
1449 | return osd_req; | |
1450 | } | |
1451 | ||
1452 | static void rbd_osd_req_destroy(struct ceph_osd_request *osd_req) | |
1453 | { | |
1454 | ceph_osdc_put_request(osd_req); | |
1455 | } | |
1456 | ||
1457 | /* object_name is assumed to be a non-null pointer and NUL-terminated */ | |
1458 | ||
1459 | static struct rbd_obj_request *rbd_obj_request_create(const char *object_name, | |
1460 | u64 offset, u64 length, | |
1461 | enum obj_request_type type) | |
1462 | { | |
1463 | struct rbd_obj_request *obj_request; | |
1464 | size_t size; | |
1465 | char *name; | |
1466 | ||
1467 | rbd_assert(obj_request_type_valid(type)); | |
1468 | ||
1469 | size = strlen(object_name) + 1; | |
1470 | obj_request = kzalloc(sizeof (*obj_request) + size, GFP_KERNEL); | |
1471 | if (!obj_request) | |
1472 | return NULL; | |
1473 | ||
1474 | name = (char *)(obj_request + 1); | |
1475 | obj_request->object_name = memcpy(name, object_name, size); | |
1476 | obj_request->offset = offset; | |
1477 | obj_request->length = length; | |
1478 | obj_request->which = BAD_WHICH; | |
1479 | obj_request->type = type; | |
1480 | INIT_LIST_HEAD(&obj_request->links); | |
07741308 | 1481 | obj_request_done_init(obj_request); |
788e2df3 | 1482 | init_completion(&obj_request->completion); |
bf0d5f50 AE |
1483 | kref_init(&obj_request->kref); |
1484 | ||
37206ee5 AE |
1485 | dout("%s: \"%s\" %llu/%llu %d -> obj %p\n", __func__, object_name, |
1486 | offset, length, (int)type, obj_request); | |
1487 | ||
bf0d5f50 AE |
1488 | return obj_request; |
1489 | } | |
1490 | ||
1491 | static void rbd_obj_request_destroy(struct kref *kref) | |
1492 | { | |
1493 | struct rbd_obj_request *obj_request; | |
1494 | ||
1495 | obj_request = container_of(kref, struct rbd_obj_request, kref); | |
1496 | ||
37206ee5 AE |
1497 | dout("%s: obj %p\n", __func__, obj_request); |
1498 | ||
bf0d5f50 AE |
1499 | rbd_assert(obj_request->img_request == NULL); |
1500 | rbd_assert(obj_request->which == BAD_WHICH); | |
1501 | ||
1502 | if (obj_request->osd_req) | |
1503 | rbd_osd_req_destroy(obj_request->osd_req); | |
1504 | ||
1505 | rbd_assert(obj_request_type_valid(obj_request->type)); | |
1506 | switch (obj_request->type) { | |
9969ebc5 AE |
1507 | case OBJ_REQUEST_NODATA: |
1508 | break; /* Nothing to do */ | |
bf0d5f50 AE |
1509 | case OBJ_REQUEST_BIO: |
1510 | if (obj_request->bio_list) | |
1511 | bio_chain_put(obj_request->bio_list); | |
1512 | break; | |
788e2df3 AE |
1513 | case OBJ_REQUEST_PAGES: |
1514 | if (obj_request->pages) | |
1515 | ceph_release_page_vector(obj_request->pages, | |
1516 | obj_request->page_count); | |
1517 | break; | |
bf0d5f50 AE |
1518 | } |
1519 | ||
1520 | kfree(obj_request); | |
1521 | } | |
1522 | ||
1523 | /* | |
1524 | * Caller is responsible for filling in the list of object requests | |
1525 | * that comprises the image request, and the Linux request pointer | |
1526 | * (if there is one). | |
1527 | */ | |
cc344fa1 AE |
1528 | static struct rbd_img_request *rbd_img_request_create( |
1529 | struct rbd_device *rbd_dev, | |
bf0d5f50 AE |
1530 | u64 offset, u64 length, |
1531 | bool write_request) | |
1532 | { | |
1533 | struct rbd_img_request *img_request; | |
1534 | struct ceph_snap_context *snapc = NULL; | |
1535 | ||
1536 | img_request = kmalloc(sizeof (*img_request), GFP_ATOMIC); | |
1537 | if (!img_request) | |
1538 | return NULL; | |
1539 | ||
1540 | if (write_request) { | |
1541 | down_read(&rbd_dev->header_rwsem); | |
1542 | snapc = ceph_get_snap_context(rbd_dev->header.snapc); | |
1543 | up_read(&rbd_dev->header_rwsem); | |
1544 | if (WARN_ON(!snapc)) { | |
1545 | kfree(img_request); | |
1546 | return NULL; /* Shouldn't happen */ | |
1547 | } | |
1548 | } | |
1549 | ||
1550 | img_request->rq = NULL; | |
1551 | img_request->rbd_dev = rbd_dev; | |
1552 | img_request->offset = offset; | |
1553 | img_request->length = length; | |
1554 | img_request->write_request = write_request; | |
1555 | if (write_request) | |
1556 | img_request->snapc = snapc; | |
1557 | else | |
1558 | img_request->snap_id = rbd_dev->spec->snap_id; | |
1559 | spin_lock_init(&img_request->completion_lock); | |
1560 | img_request->next_completion = 0; | |
1561 | img_request->callback = NULL; | |
1562 | img_request->obj_request_count = 0; | |
1563 | INIT_LIST_HEAD(&img_request->obj_requests); | |
1564 | kref_init(&img_request->kref); | |
1565 | ||
1566 | rbd_img_request_get(img_request); /* Avoid a warning */ | |
1567 | rbd_img_request_put(img_request); /* TEMPORARY */ | |
1568 | ||
37206ee5 AE |
1569 | dout("%s: rbd_dev %p %s %llu/%llu -> img %p\n", __func__, rbd_dev, |
1570 | write_request ? "write" : "read", offset, length, | |
1571 | img_request); | |
1572 | ||
bf0d5f50 AE |
1573 | return img_request; |
1574 | } | |
1575 | ||
1576 | static void rbd_img_request_destroy(struct kref *kref) | |
1577 | { | |
1578 | struct rbd_img_request *img_request; | |
1579 | struct rbd_obj_request *obj_request; | |
1580 | struct rbd_obj_request *next_obj_request; | |
1581 | ||
1582 | img_request = container_of(kref, struct rbd_img_request, kref); | |
1583 | ||
37206ee5 AE |
1584 | dout("%s: img %p\n", __func__, img_request); |
1585 | ||
bf0d5f50 AE |
1586 | for_each_obj_request_safe(img_request, obj_request, next_obj_request) |
1587 | rbd_img_obj_request_del(img_request, obj_request); | |
25dcf954 | 1588 | rbd_assert(img_request->obj_request_count == 0); |
bf0d5f50 AE |
1589 | |
1590 | if (img_request->write_request) | |
1591 | ceph_put_snap_context(img_request->snapc); | |
1592 | ||
1593 | kfree(img_request); | |
1594 | } | |
1595 | ||
1596 | static int rbd_img_request_fill_bio(struct rbd_img_request *img_request, | |
1597 | struct bio *bio_list) | |
1598 | { | |
1599 | struct rbd_device *rbd_dev = img_request->rbd_dev; | |
1600 | struct rbd_obj_request *obj_request = NULL; | |
1601 | struct rbd_obj_request *next_obj_request; | |
1602 | unsigned int bio_offset; | |
1603 | u64 image_offset; | |
1604 | u64 resid; | |
1605 | u16 opcode; | |
1606 | ||
37206ee5 AE |
1607 | dout("%s: img %p bio %p\n", __func__, img_request, bio_list); |
1608 | ||
bf0d5f50 AE |
1609 | opcode = img_request->write_request ? CEPH_OSD_OP_WRITE |
1610 | : CEPH_OSD_OP_READ; | |
1611 | bio_offset = 0; | |
1612 | image_offset = img_request->offset; | |
1613 | rbd_assert(image_offset == bio_list->bi_sector << SECTOR_SHIFT); | |
1614 | resid = img_request->length; | |
4dda41d3 | 1615 | rbd_assert(resid > 0); |
bf0d5f50 AE |
1616 | while (resid) { |
1617 | const char *object_name; | |
1618 | unsigned int clone_size; | |
1619 | struct ceph_osd_req_op *op; | |
1620 | u64 offset; | |
1621 | u64 length; | |
1622 | ||
1623 | object_name = rbd_segment_name(rbd_dev, image_offset); | |
1624 | if (!object_name) | |
1625 | goto out_unwind; | |
1626 | offset = rbd_segment_offset(rbd_dev, image_offset); | |
1627 | length = rbd_segment_length(rbd_dev, image_offset, resid); | |
1628 | obj_request = rbd_obj_request_create(object_name, | |
1629 | offset, length, | |
1630 | OBJ_REQUEST_BIO); | |
1631 | kfree(object_name); /* object request has its own copy */ | |
1632 | if (!obj_request) | |
1633 | goto out_unwind; | |
1634 | ||
1635 | rbd_assert(length <= (u64) UINT_MAX); | |
1636 | clone_size = (unsigned int) length; | |
1637 | obj_request->bio_list = bio_chain_clone_range(&bio_list, | |
1638 | &bio_offset, clone_size, | |
1639 | GFP_ATOMIC); | |
1640 | if (!obj_request->bio_list) | |
1641 | goto out_partial; | |
1642 | ||
1643 | /* | |
1644 | * Build up the op to use in building the osd | |
1645 | * request. Note that the contents of the op are | |
1646 | * copied by rbd_osd_req_create(). | |
1647 | */ | |
1648 | op = rbd_osd_req_op_create(opcode, offset, length); | |
1649 | if (!op) | |
1650 | goto out_partial; | |
1651 | obj_request->osd_req = rbd_osd_req_create(rbd_dev, | |
1652 | img_request->write_request, | |
1653 | obj_request, op); | |
1654 | rbd_osd_req_op_destroy(op); | |
1655 | if (!obj_request->osd_req) | |
1656 | goto out_partial; | |
1657 | /* status and version are initially zero-filled */ | |
1658 | ||
1659 | rbd_img_obj_request_add(img_request, obj_request); | |
1660 | ||
1661 | image_offset += length; | |
1662 | resid -= length; | |
1663 | } | |
1664 | ||
1665 | return 0; | |
1666 | ||
1667 | out_partial: | |
1668 | rbd_obj_request_put(obj_request); | |
1669 | out_unwind: | |
1670 | for_each_obj_request_safe(img_request, obj_request, next_obj_request) | |
1671 | rbd_obj_request_put(obj_request); | |
1672 | ||
1673 | return -ENOMEM; | |
1674 | } | |
1675 | ||
1676 | static void rbd_img_obj_callback(struct rbd_obj_request *obj_request) | |
1677 | { | |
1678 | struct rbd_img_request *img_request; | |
1679 | u32 which = obj_request->which; | |
1680 | bool more = true; | |
1681 | ||
1682 | img_request = obj_request->img_request; | |
4dda41d3 | 1683 | |
37206ee5 | 1684 | dout("%s: img %p obj %p\n", __func__, img_request, obj_request); |
bf0d5f50 AE |
1685 | rbd_assert(img_request != NULL); |
1686 | rbd_assert(img_request->rq != NULL); | |
4dda41d3 | 1687 | rbd_assert(img_request->obj_request_count > 0); |
bf0d5f50 AE |
1688 | rbd_assert(which != BAD_WHICH); |
1689 | rbd_assert(which < img_request->obj_request_count); | |
1690 | rbd_assert(which >= img_request->next_completion); | |
1691 | ||
1692 | spin_lock_irq(&img_request->completion_lock); | |
1693 | if (which != img_request->next_completion) | |
1694 | goto out; | |
1695 | ||
1696 | for_each_obj_request_from(img_request, obj_request) { | |
1697 | unsigned int xferred; | |
1698 | int result; | |
1699 | ||
1700 | rbd_assert(more); | |
1701 | rbd_assert(which < img_request->obj_request_count); | |
1702 | ||
07741308 | 1703 | if (!obj_request_done_test(obj_request)) |
bf0d5f50 AE |
1704 | break; |
1705 | ||
1706 | rbd_assert(obj_request->xferred <= (u64) UINT_MAX); | |
1707 | xferred = (unsigned int) obj_request->xferred; | |
1708 | result = (int) obj_request->result; | |
1709 | if (result) | |
1710 | rbd_warn(NULL, "obj_request %s result %d xferred %u\n", | |
1711 | img_request->write_request ? "write" : "read", | |
1712 | result, xferred); | |
1713 | ||
1714 | more = blk_end_request(img_request->rq, result, xferred); | |
1715 | which++; | |
1716 | } | |
1b83bef2 | 1717 | |
bf0d5f50 AE |
1718 | rbd_assert(more ^ (which == img_request->obj_request_count)); |
1719 | img_request->next_completion = which; | |
1720 | out: | |
1721 | spin_unlock_irq(&img_request->completion_lock); | |
1722 | ||
1723 | if (!more) | |
1724 | rbd_img_request_complete(img_request); | |
1725 | } | |
1726 | ||
1727 | static int rbd_img_request_submit(struct rbd_img_request *img_request) | |
1728 | { | |
1729 | struct rbd_device *rbd_dev = img_request->rbd_dev; | |
1730 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; | |
1731 | struct rbd_obj_request *obj_request; | |
1732 | ||
37206ee5 | 1733 | dout("%s: img %p\n", __func__, img_request); |
bf0d5f50 AE |
1734 | for_each_obj_request(img_request, obj_request) { |
1735 | int ret; | |
1736 | ||
1737 | obj_request->callback = rbd_img_obj_callback; | |
1738 | ret = rbd_obj_request_submit(osdc, obj_request); | |
1739 | if (ret) | |
1740 | return ret; | |
1741 | /* | |
1742 | * The image request has its own reference to each | |
1743 | * of its object requests, so we can safely drop the | |
1744 | * initial one here. | |
1745 | */ | |
1746 | rbd_obj_request_put(obj_request); | |
1747 | } | |
1748 | ||
1749 | return 0; | |
1750 | } | |
1751 | ||
cf81b60e | 1752 | static int rbd_obj_notify_ack(struct rbd_device *rbd_dev, |
b8d70035 AE |
1753 | u64 ver, u64 notify_id) |
1754 | { | |
1755 | struct rbd_obj_request *obj_request; | |
1756 | struct ceph_osd_req_op *op; | |
1757 | struct ceph_osd_client *osdc; | |
1758 | int ret; | |
1759 | ||
1760 | obj_request = rbd_obj_request_create(rbd_dev->header_name, 0, 0, | |
1761 | OBJ_REQUEST_NODATA); | |
1762 | if (!obj_request) | |
1763 | return -ENOMEM; | |
1764 | ||
1765 | ret = -ENOMEM; | |
1766 | op = rbd_osd_req_op_create(CEPH_OSD_OP_NOTIFY_ACK, notify_id, ver); | |
1767 | if (!op) | |
1768 | goto out; | |
1769 | obj_request->osd_req = rbd_osd_req_create(rbd_dev, false, | |
1770 | obj_request, op); | |
1771 | rbd_osd_req_op_destroy(op); | |
1772 | if (!obj_request->osd_req) | |
1773 | goto out; | |
1774 | ||
1775 | osdc = &rbd_dev->rbd_client->client->osdc; | |
cf81b60e | 1776 | obj_request->callback = rbd_obj_request_put; |
b8d70035 | 1777 | ret = rbd_obj_request_submit(osdc, obj_request); |
b8d70035 | 1778 | out: |
cf81b60e AE |
1779 | if (ret) |
1780 | rbd_obj_request_put(obj_request); | |
b8d70035 AE |
1781 | |
1782 | return ret; | |
1783 | } | |
1784 | ||
1785 | static void rbd_watch_cb(u64 ver, u64 notify_id, u8 opcode, void *data) | |
1786 | { | |
1787 | struct rbd_device *rbd_dev = (struct rbd_device *)data; | |
1788 | u64 hver; | |
1789 | int rc; | |
1790 | ||
1791 | if (!rbd_dev) | |
1792 | return; | |
1793 | ||
37206ee5 | 1794 | dout("%s: \"%s\" notify_id %llu opcode %u\n", __func__, |
b8d70035 AE |
1795 | rbd_dev->header_name, (unsigned long long) notify_id, |
1796 | (unsigned int) opcode); | |
1797 | rc = rbd_dev_refresh(rbd_dev, &hver); | |
1798 | if (rc) | |
1799 | rbd_warn(rbd_dev, "got notification but failed to " | |
1800 | " update snaps: %d\n", rc); | |
1801 | ||
cf81b60e | 1802 | rbd_obj_notify_ack(rbd_dev, hver, notify_id); |
b8d70035 AE |
1803 | } |
1804 | ||
9969ebc5 AE |
1805 | /* |
1806 | * Request sync osd watch/unwatch. The value of "start" determines | |
1807 | * whether a watch request is being initiated or torn down. | |
1808 | */ | |
1809 | static int rbd_dev_header_watch_sync(struct rbd_device *rbd_dev, int start) | |
1810 | { | |
1811 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; | |
1812 | struct rbd_obj_request *obj_request; | |
1813 | struct ceph_osd_req_op *op; | |
1814 | int ret; | |
1815 | ||
1816 | rbd_assert(start ^ !!rbd_dev->watch_event); | |
1817 | rbd_assert(start ^ !!rbd_dev->watch_request); | |
1818 | ||
1819 | if (start) { | |
3c663bbd | 1820 | ret = ceph_osdc_create_event(osdc, rbd_watch_cb, rbd_dev, |
9969ebc5 AE |
1821 | &rbd_dev->watch_event); |
1822 | if (ret < 0) | |
1823 | return ret; | |
8eb87565 | 1824 | rbd_assert(rbd_dev->watch_event != NULL); |
9969ebc5 AE |
1825 | } |
1826 | ||
1827 | ret = -ENOMEM; | |
1828 | obj_request = rbd_obj_request_create(rbd_dev->header_name, 0, 0, | |
1829 | OBJ_REQUEST_NODATA); | |
1830 | if (!obj_request) | |
1831 | goto out_cancel; | |
1832 | ||
1833 | op = rbd_osd_req_op_create(CEPH_OSD_OP_WATCH, | |
1834 | rbd_dev->watch_event->cookie, | |
1835 | rbd_dev->header.obj_version, start); | |
1836 | if (!op) | |
1837 | goto out_cancel; | |
1838 | obj_request->osd_req = rbd_osd_req_create(rbd_dev, true, | |
1839 | obj_request, op); | |
1840 | rbd_osd_req_op_destroy(op); | |
1841 | if (!obj_request->osd_req) | |
1842 | goto out_cancel; | |
1843 | ||
8eb87565 | 1844 | if (start) |
975241af | 1845 | ceph_osdc_set_request_linger(osdc, obj_request->osd_req); |
8eb87565 | 1846 | else |
6977c3f9 | 1847 | ceph_osdc_unregister_linger_request(osdc, |
975241af | 1848 | rbd_dev->watch_request->osd_req); |
9969ebc5 AE |
1849 | ret = rbd_obj_request_submit(osdc, obj_request); |
1850 | if (ret) | |
1851 | goto out_cancel; | |
1852 | ret = rbd_obj_request_wait(obj_request); | |
1853 | if (ret) | |
1854 | goto out_cancel; | |
9969ebc5 AE |
1855 | ret = obj_request->result; |
1856 | if (ret) | |
1857 | goto out_cancel; | |
1858 | ||
8eb87565 AE |
1859 | /* |
1860 | * A watch request is set to linger, so the underlying osd | |
1861 | * request won't go away until we unregister it. We retain | |
1862 | * a pointer to the object request during that time (in | |
1863 | * rbd_dev->watch_request), so we'll keep a reference to | |
1864 | * it. We'll drop that reference (below) after we've | |
1865 | * unregistered it. | |
1866 | */ | |
1867 | if (start) { | |
1868 | rbd_dev->watch_request = obj_request; | |
1869 | ||
1870 | return 0; | |
1871 | } | |
1872 | ||
1873 | /* We have successfully torn down the watch request */ | |
1874 | ||
1875 | rbd_obj_request_put(rbd_dev->watch_request); | |
1876 | rbd_dev->watch_request = NULL; | |
9969ebc5 AE |
1877 | out_cancel: |
1878 | /* Cancel the event if we're tearing down, or on error */ | |
1879 | ceph_osdc_cancel_event(rbd_dev->watch_event); | |
1880 | rbd_dev->watch_event = NULL; | |
9969ebc5 AE |
1881 | if (obj_request) |
1882 | rbd_obj_request_put(obj_request); | |
1883 | ||
1884 | return ret; | |
1885 | } | |
1886 | ||
36be9a76 AE |
1887 | /* |
1888 | * Synchronous osd object method call | |
1889 | */ | |
1890 | static int rbd_obj_method_sync(struct rbd_device *rbd_dev, | |
1891 | const char *object_name, | |
1892 | const char *class_name, | |
1893 | const char *method_name, | |
1894 | const char *outbound, | |
1895 | size_t outbound_size, | |
1896 | char *inbound, | |
1897 | size_t inbound_size, | |
1898 | u64 *version) | |
1899 | { | |
1900 | struct rbd_obj_request *obj_request; | |
1901 | struct ceph_osd_client *osdc; | |
1902 | struct ceph_osd_req_op *op; | |
1903 | struct page **pages; | |
1904 | u32 page_count; | |
1905 | int ret; | |
1906 | ||
1907 | /* | |
1908 | * Method calls are ultimately read operations but they | |
1909 | * don't involve object data (so no offset or length). | |
1910 | * The result should placed into the inbound buffer | |
1911 | * provided. They also supply outbound data--parameters for | |
1912 | * the object method. Currently if this is present it will | |
1913 | * be a snapshot id. | |
1914 | */ | |
1915 | page_count = (u32) calc_pages_for(0, inbound_size); | |
1916 | pages = ceph_alloc_page_vector(page_count, GFP_KERNEL); | |
1917 | if (IS_ERR(pages)) | |
1918 | return PTR_ERR(pages); | |
1919 | ||
1920 | ret = -ENOMEM; | |
1921 | obj_request = rbd_obj_request_create(object_name, 0, 0, | |
1922 | OBJ_REQUEST_PAGES); | |
1923 | if (!obj_request) | |
1924 | goto out; | |
1925 | ||
1926 | obj_request->pages = pages; | |
1927 | obj_request->page_count = page_count; | |
1928 | ||
1929 | op = rbd_osd_req_op_create(CEPH_OSD_OP_CALL, class_name, | |
1930 | method_name, outbound, outbound_size); | |
1931 | if (!op) | |
1932 | goto out; | |
1933 | obj_request->osd_req = rbd_osd_req_create(rbd_dev, false, | |
1934 | obj_request, op); | |
1935 | rbd_osd_req_op_destroy(op); | |
1936 | if (!obj_request->osd_req) | |
1937 | goto out; | |
1938 | ||
1939 | osdc = &rbd_dev->rbd_client->client->osdc; | |
1940 | ret = rbd_obj_request_submit(osdc, obj_request); | |
1941 | if (ret) | |
1942 | goto out; | |
1943 | ret = rbd_obj_request_wait(obj_request); | |
1944 | if (ret) | |
1945 | goto out; | |
1946 | ||
1947 | ret = obj_request->result; | |
1948 | if (ret < 0) | |
1949 | goto out; | |
23ed6e13 | 1950 | ret = 0; |
903bb32e | 1951 | ceph_copy_from_page_vector(pages, inbound, 0, obj_request->xferred); |
36be9a76 AE |
1952 | if (version) |
1953 | *version = obj_request->version; | |
1954 | out: | |
1955 | if (obj_request) | |
1956 | rbd_obj_request_put(obj_request); | |
1957 | else | |
1958 | ceph_release_page_vector(pages, page_count); | |
1959 | ||
1960 | return ret; | |
1961 | } | |
1962 | ||
bf0d5f50 | 1963 | static void rbd_request_fn(struct request_queue *q) |
cc344fa1 | 1964 | __releases(q->queue_lock) __acquires(q->queue_lock) |
bf0d5f50 AE |
1965 | { |
1966 | struct rbd_device *rbd_dev = q->queuedata; | |
1967 | bool read_only = rbd_dev->mapping.read_only; | |
1968 | struct request *rq; | |
1969 | int result; | |
1970 | ||
1971 | while ((rq = blk_fetch_request(q))) { | |
1972 | bool write_request = rq_data_dir(rq) == WRITE; | |
1973 | struct rbd_img_request *img_request; | |
1974 | u64 offset; | |
1975 | u64 length; | |
1976 | ||
1977 | /* Ignore any non-FS requests that filter through. */ | |
1978 | ||
1979 | if (rq->cmd_type != REQ_TYPE_FS) { | |
4dda41d3 AE |
1980 | dout("%s: non-fs request type %d\n", __func__, |
1981 | (int) rq->cmd_type); | |
1982 | __blk_end_request_all(rq, 0); | |
1983 | continue; | |
1984 | } | |
1985 | ||
1986 | /* Ignore/skip any zero-length requests */ | |
1987 | ||
1988 | offset = (u64) blk_rq_pos(rq) << SECTOR_SHIFT; | |
1989 | length = (u64) blk_rq_bytes(rq); | |
1990 | ||
1991 | if (!length) { | |
1992 | dout("%s: zero-length request\n", __func__); | |
bf0d5f50 AE |
1993 | __blk_end_request_all(rq, 0); |
1994 | continue; | |
1995 | } | |
1996 | ||
1997 | spin_unlock_irq(q->queue_lock); | |
1998 | ||
1999 | /* Disallow writes to a read-only device */ | |
2000 | ||
2001 | if (write_request) { | |
2002 | result = -EROFS; | |
2003 | if (read_only) | |
2004 | goto end_request; | |
2005 | rbd_assert(rbd_dev->spec->snap_id == CEPH_NOSNAP); | |
2006 | } | |
2007 | ||
6d292906 AE |
2008 | /* |
2009 | * Quit early if the mapped snapshot no longer | |
2010 | * exists. It's still possible the snapshot will | |
2011 | * have disappeared by the time our request arrives | |
2012 | * at the osd, but there's no sense in sending it if | |
2013 | * we already know. | |
2014 | */ | |
2015 | if (!test_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags)) { | |
bf0d5f50 AE |
2016 | dout("request for non-existent snapshot"); |
2017 | rbd_assert(rbd_dev->spec->snap_id != CEPH_NOSNAP); | |
2018 | result = -ENXIO; | |
2019 | goto end_request; | |
2020 | } | |
2021 | ||
bf0d5f50 AE |
2022 | result = -EINVAL; |
2023 | if (WARN_ON(offset && length > U64_MAX - offset + 1)) | |
2024 | goto end_request; /* Shouldn't happen */ | |
2025 | ||
2026 | result = -ENOMEM; | |
2027 | img_request = rbd_img_request_create(rbd_dev, offset, length, | |
2028 | write_request); | |
2029 | if (!img_request) | |
2030 | goto end_request; | |
2031 | ||
2032 | img_request->rq = rq; | |
2033 | ||
2034 | result = rbd_img_request_fill_bio(img_request, rq->bio); | |
2035 | if (!result) | |
2036 | result = rbd_img_request_submit(img_request); | |
2037 | if (result) | |
2038 | rbd_img_request_put(img_request); | |
2039 | end_request: | |
2040 | spin_lock_irq(q->queue_lock); | |
2041 | if (result < 0) { | |
2042 | rbd_warn(rbd_dev, "obj_request %s result %d\n", | |
2043 | write_request ? "write" : "read", result); | |
2044 | __blk_end_request_all(rq, result); | |
2045 | } | |
2046 | } | |
2047 | } | |
2048 | ||
602adf40 YS |
2049 | /* |
2050 | * a queue callback. Makes sure that we don't create a bio that spans across | |
2051 | * multiple osd objects. One exception would be with a single page bios, | |
f7760dad | 2052 | * which we handle later at bio_chain_clone_range() |
602adf40 YS |
2053 | */ |
2054 | static int rbd_merge_bvec(struct request_queue *q, struct bvec_merge_data *bmd, | |
2055 | struct bio_vec *bvec) | |
2056 | { | |
2057 | struct rbd_device *rbd_dev = q->queuedata; | |
e5cfeed2 AE |
2058 | sector_t sector_offset; |
2059 | sector_t sectors_per_obj; | |
2060 | sector_t obj_sector_offset; | |
2061 | int ret; | |
2062 | ||
2063 | /* | |
2064 | * Find how far into its rbd object the partition-relative | |
2065 | * bio start sector is to offset relative to the enclosing | |
2066 | * device. | |
2067 | */ | |
2068 | sector_offset = get_start_sect(bmd->bi_bdev) + bmd->bi_sector; | |
2069 | sectors_per_obj = 1 << (rbd_dev->header.obj_order - SECTOR_SHIFT); | |
2070 | obj_sector_offset = sector_offset & (sectors_per_obj - 1); | |
2071 | ||
2072 | /* | |
2073 | * Compute the number of bytes from that offset to the end | |
2074 | * of the object. Account for what's already used by the bio. | |
2075 | */ | |
2076 | ret = (int) (sectors_per_obj - obj_sector_offset) << SECTOR_SHIFT; | |
2077 | if (ret > bmd->bi_size) | |
2078 | ret -= bmd->bi_size; | |
2079 | else | |
2080 | ret = 0; | |
2081 | ||
2082 | /* | |
2083 | * Don't send back more than was asked for. And if the bio | |
2084 | * was empty, let the whole thing through because: "Note | |
2085 | * that a block device *must* allow a single page to be | |
2086 | * added to an empty bio." | |
2087 | */ | |
2088 | rbd_assert(bvec->bv_len <= PAGE_SIZE); | |
2089 | if (ret > (int) bvec->bv_len || !bmd->bi_size) | |
2090 | ret = (int) bvec->bv_len; | |
2091 | ||
2092 | return ret; | |
602adf40 YS |
2093 | } |
2094 | ||
2095 | static void rbd_free_disk(struct rbd_device *rbd_dev) | |
2096 | { | |
2097 | struct gendisk *disk = rbd_dev->disk; | |
2098 | ||
2099 | if (!disk) | |
2100 | return; | |
2101 | ||
602adf40 YS |
2102 | if (disk->flags & GENHD_FL_UP) |
2103 | del_gendisk(disk); | |
2104 | if (disk->queue) | |
2105 | blk_cleanup_queue(disk->queue); | |
2106 | put_disk(disk); | |
2107 | } | |
2108 | ||
788e2df3 AE |
2109 | static int rbd_obj_read_sync(struct rbd_device *rbd_dev, |
2110 | const char *object_name, | |
2111 | u64 offset, u64 length, | |
2112 | char *buf, u64 *version) | |
2113 | ||
2114 | { | |
2115 | struct ceph_osd_req_op *op; | |
2116 | struct rbd_obj_request *obj_request; | |
2117 | struct ceph_osd_client *osdc; | |
2118 | struct page **pages = NULL; | |
2119 | u32 page_count; | |
1ceae7ef | 2120 | size_t size; |
788e2df3 AE |
2121 | int ret; |
2122 | ||
2123 | page_count = (u32) calc_pages_for(offset, length); | |
2124 | pages = ceph_alloc_page_vector(page_count, GFP_KERNEL); | |
2125 | if (IS_ERR(pages)) | |
2126 | ret = PTR_ERR(pages); | |
2127 | ||
2128 | ret = -ENOMEM; | |
2129 | obj_request = rbd_obj_request_create(object_name, offset, length, | |
36be9a76 | 2130 | OBJ_REQUEST_PAGES); |
788e2df3 AE |
2131 | if (!obj_request) |
2132 | goto out; | |
2133 | ||
2134 | obj_request->pages = pages; | |
2135 | obj_request->page_count = page_count; | |
2136 | ||
2137 | op = rbd_osd_req_op_create(CEPH_OSD_OP_READ, offset, length); | |
2138 | if (!op) | |
2139 | goto out; | |
2140 | obj_request->osd_req = rbd_osd_req_create(rbd_dev, false, | |
2141 | obj_request, op); | |
2142 | rbd_osd_req_op_destroy(op); | |
2143 | if (!obj_request->osd_req) | |
2144 | goto out; | |
2145 | ||
2146 | osdc = &rbd_dev->rbd_client->client->osdc; | |
2147 | ret = rbd_obj_request_submit(osdc, obj_request); | |
2148 | if (ret) | |
2149 | goto out; | |
2150 | ret = rbd_obj_request_wait(obj_request); | |
2151 | if (ret) | |
2152 | goto out; | |
2153 | ||
2154 | ret = obj_request->result; | |
2155 | if (ret < 0) | |
2156 | goto out; | |
1ceae7ef AE |
2157 | |
2158 | rbd_assert(obj_request->xferred <= (u64) SIZE_MAX); | |
2159 | size = (size_t) obj_request->xferred; | |
903bb32e | 2160 | ceph_copy_from_page_vector(pages, buf, 0, size); |
23ed6e13 AE |
2161 | rbd_assert(size <= (size_t) INT_MAX); |
2162 | ret = (int) size; | |
788e2df3 AE |
2163 | if (version) |
2164 | *version = obj_request->version; | |
2165 | out: | |
2166 | if (obj_request) | |
2167 | rbd_obj_request_put(obj_request); | |
2168 | else | |
2169 | ceph_release_page_vector(pages, page_count); | |
2170 | ||
2171 | return ret; | |
2172 | } | |
2173 | ||
602adf40 | 2174 | /* |
4156d998 AE |
2175 | * Read the complete header for the given rbd device. |
2176 | * | |
2177 | * Returns a pointer to a dynamically-allocated buffer containing | |
2178 | * the complete and validated header. Caller can pass the address | |
2179 | * of a variable that will be filled in with the version of the | |
2180 | * header object at the time it was read. | |
2181 | * | |
2182 | * Returns a pointer-coded errno if a failure occurs. | |
602adf40 | 2183 | */ |
4156d998 AE |
2184 | static struct rbd_image_header_ondisk * |
2185 | rbd_dev_v1_header_read(struct rbd_device *rbd_dev, u64 *version) | |
602adf40 | 2186 | { |
4156d998 | 2187 | struct rbd_image_header_ondisk *ondisk = NULL; |
50f7c4c9 | 2188 | u32 snap_count = 0; |
4156d998 AE |
2189 | u64 names_size = 0; |
2190 | u32 want_count; | |
2191 | int ret; | |
602adf40 | 2192 | |
00f1f36f | 2193 | /* |
4156d998 AE |
2194 | * The complete header will include an array of its 64-bit |
2195 | * snapshot ids, followed by the names of those snapshots as | |
2196 | * a contiguous block of NUL-terminated strings. Note that | |
2197 | * the number of snapshots could change by the time we read | |
2198 | * it in, in which case we re-read it. | |
00f1f36f | 2199 | */ |
4156d998 AE |
2200 | do { |
2201 | size_t size; | |
2202 | ||
2203 | kfree(ondisk); | |
2204 | ||
2205 | size = sizeof (*ondisk); | |
2206 | size += snap_count * sizeof (struct rbd_image_snap_ondisk); | |
2207 | size += names_size; | |
2208 | ondisk = kmalloc(size, GFP_KERNEL); | |
2209 | if (!ondisk) | |
2210 | return ERR_PTR(-ENOMEM); | |
2211 | ||
788e2df3 | 2212 | ret = rbd_obj_read_sync(rbd_dev, rbd_dev->header_name, |
4156d998 AE |
2213 | 0, size, |
2214 | (char *) ondisk, version); | |
4156d998 AE |
2215 | if (ret < 0) |
2216 | goto out_err; | |
2217 | if (WARN_ON((size_t) ret < size)) { | |
2218 | ret = -ENXIO; | |
06ecc6cb AE |
2219 | rbd_warn(rbd_dev, "short header read (want %zd got %d)", |
2220 | size, ret); | |
4156d998 AE |
2221 | goto out_err; |
2222 | } | |
2223 | if (!rbd_dev_ondisk_valid(ondisk)) { | |
2224 | ret = -ENXIO; | |
06ecc6cb | 2225 | rbd_warn(rbd_dev, "invalid header"); |
4156d998 | 2226 | goto out_err; |
81e759fb | 2227 | } |
602adf40 | 2228 | |
4156d998 AE |
2229 | names_size = le64_to_cpu(ondisk->snap_names_len); |
2230 | want_count = snap_count; | |
2231 | snap_count = le32_to_cpu(ondisk->snap_count); | |
2232 | } while (snap_count != want_count); | |
00f1f36f | 2233 | |
4156d998 | 2234 | return ondisk; |
00f1f36f | 2235 | |
4156d998 AE |
2236 | out_err: |
2237 | kfree(ondisk); | |
2238 | ||
2239 | return ERR_PTR(ret); | |
2240 | } | |
2241 | ||
2242 | /* | |
2243 | * reload the ondisk the header | |
2244 | */ | |
2245 | static int rbd_read_header(struct rbd_device *rbd_dev, | |
2246 | struct rbd_image_header *header) | |
2247 | { | |
2248 | struct rbd_image_header_ondisk *ondisk; | |
2249 | u64 ver = 0; | |
2250 | int ret; | |
602adf40 | 2251 | |
4156d998 AE |
2252 | ondisk = rbd_dev_v1_header_read(rbd_dev, &ver); |
2253 | if (IS_ERR(ondisk)) | |
2254 | return PTR_ERR(ondisk); | |
2255 | ret = rbd_header_from_disk(header, ondisk); | |
2256 | if (ret >= 0) | |
2257 | header->obj_version = ver; | |
2258 | kfree(ondisk); | |
2259 | ||
2260 | return ret; | |
602adf40 YS |
2261 | } |
2262 | ||
41f38c2b | 2263 | static void rbd_remove_all_snaps(struct rbd_device *rbd_dev) |
dfc5606d YS |
2264 | { |
2265 | struct rbd_snap *snap; | |
a0593290 | 2266 | struct rbd_snap *next; |
dfc5606d | 2267 | |
a0593290 | 2268 | list_for_each_entry_safe(snap, next, &rbd_dev->snaps, node) |
41f38c2b | 2269 | rbd_remove_snap_dev(snap); |
dfc5606d YS |
2270 | } |
2271 | ||
9478554a AE |
2272 | static void rbd_update_mapping_size(struct rbd_device *rbd_dev) |
2273 | { | |
2274 | sector_t size; | |
2275 | ||
0d7dbfce | 2276 | if (rbd_dev->spec->snap_id != CEPH_NOSNAP) |
9478554a AE |
2277 | return; |
2278 | ||
2279 | size = (sector_t) rbd_dev->header.image_size / SECTOR_SIZE; | |
2280 | dout("setting size to %llu sectors", (unsigned long long) size); | |
2281 | rbd_dev->mapping.size = (u64) size; | |
2282 | set_capacity(rbd_dev->disk, size); | |
2283 | } | |
2284 | ||
602adf40 YS |
2285 | /* |
2286 | * only read the first part of the ondisk header, without the snaps info | |
2287 | */ | |
117973fb | 2288 | static int rbd_dev_v1_refresh(struct rbd_device *rbd_dev, u64 *hver) |
602adf40 YS |
2289 | { |
2290 | int ret; | |
2291 | struct rbd_image_header h; | |
602adf40 YS |
2292 | |
2293 | ret = rbd_read_header(rbd_dev, &h); | |
2294 | if (ret < 0) | |
2295 | return ret; | |
2296 | ||
a51aa0c0 JD |
2297 | down_write(&rbd_dev->header_rwsem); |
2298 | ||
9478554a AE |
2299 | /* Update image size, and check for resize of mapped image */ |
2300 | rbd_dev->header.image_size = h.image_size; | |
2301 | rbd_update_mapping_size(rbd_dev); | |
9db4b3e3 | 2302 | |
849b4260 | 2303 | /* rbd_dev->header.object_prefix shouldn't change */ |
602adf40 | 2304 | kfree(rbd_dev->header.snap_sizes); |
849b4260 | 2305 | kfree(rbd_dev->header.snap_names); |
d1d25646 JD |
2306 | /* osd requests may still refer to snapc */ |
2307 | ceph_put_snap_context(rbd_dev->header.snapc); | |
602adf40 | 2308 | |
b813623a AE |
2309 | if (hver) |
2310 | *hver = h.obj_version; | |
a71b891b | 2311 | rbd_dev->header.obj_version = h.obj_version; |
93a24e08 | 2312 | rbd_dev->header.image_size = h.image_size; |
602adf40 YS |
2313 | rbd_dev->header.snapc = h.snapc; |
2314 | rbd_dev->header.snap_names = h.snap_names; | |
2315 | rbd_dev->header.snap_sizes = h.snap_sizes; | |
849b4260 AE |
2316 | /* Free the extra copy of the object prefix */ |
2317 | WARN_ON(strcmp(rbd_dev->header.object_prefix, h.object_prefix)); | |
2318 | kfree(h.object_prefix); | |
2319 | ||
304f6808 AE |
2320 | ret = rbd_dev_snaps_update(rbd_dev); |
2321 | if (!ret) | |
2322 | ret = rbd_dev_snaps_register(rbd_dev); | |
dfc5606d | 2323 | |
c666601a | 2324 | up_write(&rbd_dev->header_rwsem); |
602adf40 | 2325 | |
dfc5606d | 2326 | return ret; |
602adf40 YS |
2327 | } |
2328 | ||
117973fb | 2329 | static int rbd_dev_refresh(struct rbd_device *rbd_dev, u64 *hver) |
1fe5e993 AE |
2330 | { |
2331 | int ret; | |
2332 | ||
117973fb | 2333 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); |
1fe5e993 | 2334 | mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING); |
117973fb AE |
2335 | if (rbd_dev->image_format == 1) |
2336 | ret = rbd_dev_v1_refresh(rbd_dev, hver); | |
2337 | else | |
2338 | ret = rbd_dev_v2_refresh(rbd_dev, hver); | |
1fe5e993 AE |
2339 | mutex_unlock(&ctl_mutex); |
2340 | ||
2341 | return ret; | |
2342 | } | |
2343 | ||
602adf40 YS |
2344 | static int rbd_init_disk(struct rbd_device *rbd_dev) |
2345 | { | |
2346 | struct gendisk *disk; | |
2347 | struct request_queue *q; | |
593a9e7b | 2348 | u64 segment_size; |
602adf40 | 2349 | |
602adf40 | 2350 | /* create gendisk info */ |
602adf40 YS |
2351 | disk = alloc_disk(RBD_MINORS_PER_MAJOR); |
2352 | if (!disk) | |
1fcdb8aa | 2353 | return -ENOMEM; |
602adf40 | 2354 | |
f0f8cef5 | 2355 | snprintf(disk->disk_name, sizeof(disk->disk_name), RBD_DRV_NAME "%d", |
de71a297 | 2356 | rbd_dev->dev_id); |
602adf40 YS |
2357 | disk->major = rbd_dev->major; |
2358 | disk->first_minor = 0; | |
2359 | disk->fops = &rbd_bd_ops; | |
2360 | disk->private_data = rbd_dev; | |
2361 | ||
bf0d5f50 | 2362 | q = blk_init_queue(rbd_request_fn, &rbd_dev->lock); |
602adf40 YS |
2363 | if (!q) |
2364 | goto out_disk; | |
029bcbd8 | 2365 | |
593a9e7b AE |
2366 | /* We use the default size, but let's be explicit about it. */ |
2367 | blk_queue_physical_block_size(q, SECTOR_SIZE); | |
2368 | ||
029bcbd8 | 2369 | /* set io sizes to object size */ |
593a9e7b AE |
2370 | segment_size = rbd_obj_bytes(&rbd_dev->header); |
2371 | blk_queue_max_hw_sectors(q, segment_size / SECTOR_SIZE); | |
2372 | blk_queue_max_segment_size(q, segment_size); | |
2373 | blk_queue_io_min(q, segment_size); | |
2374 | blk_queue_io_opt(q, segment_size); | |
029bcbd8 | 2375 | |
602adf40 YS |
2376 | blk_queue_merge_bvec(q, rbd_merge_bvec); |
2377 | disk->queue = q; | |
2378 | ||
2379 | q->queuedata = rbd_dev; | |
2380 | ||
2381 | rbd_dev->disk = disk; | |
602adf40 | 2382 | |
12f02944 AE |
2383 | set_capacity(rbd_dev->disk, rbd_dev->mapping.size / SECTOR_SIZE); |
2384 | ||
602adf40 | 2385 | return 0; |
602adf40 YS |
2386 | out_disk: |
2387 | put_disk(disk); | |
1fcdb8aa AE |
2388 | |
2389 | return -ENOMEM; | |
602adf40 YS |
2390 | } |
2391 | ||
dfc5606d YS |
2392 | /* |
2393 | sysfs | |
2394 | */ | |
2395 | ||
593a9e7b AE |
2396 | static struct rbd_device *dev_to_rbd_dev(struct device *dev) |
2397 | { | |
2398 | return container_of(dev, struct rbd_device, dev); | |
2399 | } | |
2400 | ||
dfc5606d YS |
2401 | static ssize_t rbd_size_show(struct device *dev, |
2402 | struct device_attribute *attr, char *buf) | |
2403 | { | |
593a9e7b | 2404 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
a51aa0c0 JD |
2405 | sector_t size; |
2406 | ||
2407 | down_read(&rbd_dev->header_rwsem); | |
2408 | size = get_capacity(rbd_dev->disk); | |
2409 | up_read(&rbd_dev->header_rwsem); | |
dfc5606d | 2410 | |
a51aa0c0 | 2411 | return sprintf(buf, "%llu\n", (unsigned long long) size * SECTOR_SIZE); |
dfc5606d YS |
2412 | } |
2413 | ||
34b13184 AE |
2414 | /* |
2415 | * Note this shows the features for whatever's mapped, which is not | |
2416 | * necessarily the base image. | |
2417 | */ | |
2418 | static ssize_t rbd_features_show(struct device *dev, | |
2419 | struct device_attribute *attr, char *buf) | |
2420 | { | |
2421 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
2422 | ||
2423 | return sprintf(buf, "0x%016llx\n", | |
2424 | (unsigned long long) rbd_dev->mapping.features); | |
2425 | } | |
2426 | ||
dfc5606d YS |
2427 | static ssize_t rbd_major_show(struct device *dev, |
2428 | struct device_attribute *attr, char *buf) | |
2429 | { | |
593a9e7b | 2430 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
602adf40 | 2431 | |
dfc5606d YS |
2432 | return sprintf(buf, "%d\n", rbd_dev->major); |
2433 | } | |
2434 | ||
2435 | static ssize_t rbd_client_id_show(struct device *dev, | |
2436 | struct device_attribute *attr, char *buf) | |
602adf40 | 2437 | { |
593a9e7b | 2438 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 2439 | |
1dbb4399 AE |
2440 | return sprintf(buf, "client%lld\n", |
2441 | ceph_client_id(rbd_dev->rbd_client->client)); | |
602adf40 YS |
2442 | } |
2443 | ||
dfc5606d YS |
2444 | static ssize_t rbd_pool_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 | |
0d7dbfce | 2449 | return sprintf(buf, "%s\n", rbd_dev->spec->pool_name); |
dfc5606d YS |
2450 | } |
2451 | ||
9bb2f334 AE |
2452 | static ssize_t rbd_pool_id_show(struct device *dev, |
2453 | struct device_attribute *attr, char *buf) | |
2454 | { | |
2455 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
2456 | ||
0d7dbfce AE |
2457 | return sprintf(buf, "%llu\n", |
2458 | (unsigned long long) rbd_dev->spec->pool_id); | |
9bb2f334 AE |
2459 | } |
2460 | ||
dfc5606d YS |
2461 | static ssize_t rbd_name_show(struct device *dev, |
2462 | struct device_attribute *attr, char *buf) | |
2463 | { | |
593a9e7b | 2464 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 2465 | |
a92ffdf8 AE |
2466 | if (rbd_dev->spec->image_name) |
2467 | return sprintf(buf, "%s\n", rbd_dev->spec->image_name); | |
2468 | ||
2469 | return sprintf(buf, "(unknown)\n"); | |
dfc5606d YS |
2470 | } |
2471 | ||
589d30e0 AE |
2472 | static ssize_t rbd_image_id_show(struct device *dev, |
2473 | struct device_attribute *attr, char *buf) | |
2474 | { | |
2475 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
2476 | ||
0d7dbfce | 2477 | return sprintf(buf, "%s\n", rbd_dev->spec->image_id); |
589d30e0 AE |
2478 | } |
2479 | ||
34b13184 AE |
2480 | /* |
2481 | * Shows the name of the currently-mapped snapshot (or | |
2482 | * RBD_SNAP_HEAD_NAME for the base image). | |
2483 | */ | |
dfc5606d YS |
2484 | static ssize_t rbd_snap_show(struct device *dev, |
2485 | struct device_attribute *attr, | |
2486 | char *buf) | |
2487 | { | |
593a9e7b | 2488 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 2489 | |
0d7dbfce | 2490 | return sprintf(buf, "%s\n", rbd_dev->spec->snap_name); |
dfc5606d YS |
2491 | } |
2492 | ||
86b00e0d AE |
2493 | /* |
2494 | * For an rbd v2 image, shows the pool id, image id, and snapshot id | |
2495 | * for the parent image. If there is no parent, simply shows | |
2496 | * "(no parent image)". | |
2497 | */ | |
2498 | static ssize_t rbd_parent_show(struct device *dev, | |
2499 | struct device_attribute *attr, | |
2500 | char *buf) | |
2501 | { | |
2502 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
2503 | struct rbd_spec *spec = rbd_dev->parent_spec; | |
2504 | int count; | |
2505 | char *bufp = buf; | |
2506 | ||
2507 | if (!spec) | |
2508 | return sprintf(buf, "(no parent image)\n"); | |
2509 | ||
2510 | count = sprintf(bufp, "pool_id %llu\npool_name %s\n", | |
2511 | (unsigned long long) spec->pool_id, spec->pool_name); | |
2512 | if (count < 0) | |
2513 | return count; | |
2514 | bufp += count; | |
2515 | ||
2516 | count = sprintf(bufp, "image_id %s\nimage_name %s\n", spec->image_id, | |
2517 | spec->image_name ? spec->image_name : "(unknown)"); | |
2518 | if (count < 0) | |
2519 | return count; | |
2520 | bufp += count; | |
2521 | ||
2522 | count = sprintf(bufp, "snap_id %llu\nsnap_name %s\n", | |
2523 | (unsigned long long) spec->snap_id, spec->snap_name); | |
2524 | if (count < 0) | |
2525 | return count; | |
2526 | bufp += count; | |
2527 | ||
2528 | count = sprintf(bufp, "overlap %llu\n", rbd_dev->parent_overlap); | |
2529 | if (count < 0) | |
2530 | return count; | |
2531 | bufp += count; | |
2532 | ||
2533 | return (ssize_t) (bufp - buf); | |
2534 | } | |
2535 | ||
dfc5606d YS |
2536 | static ssize_t rbd_image_refresh(struct device *dev, |
2537 | struct device_attribute *attr, | |
2538 | const char *buf, | |
2539 | size_t size) | |
2540 | { | |
593a9e7b | 2541 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
b813623a | 2542 | int ret; |
602adf40 | 2543 | |
117973fb | 2544 | ret = rbd_dev_refresh(rbd_dev, NULL); |
b813623a AE |
2545 | |
2546 | return ret < 0 ? ret : size; | |
dfc5606d | 2547 | } |
602adf40 | 2548 | |
dfc5606d | 2549 | static DEVICE_ATTR(size, S_IRUGO, rbd_size_show, NULL); |
34b13184 | 2550 | static DEVICE_ATTR(features, S_IRUGO, rbd_features_show, NULL); |
dfc5606d YS |
2551 | static DEVICE_ATTR(major, S_IRUGO, rbd_major_show, NULL); |
2552 | static DEVICE_ATTR(client_id, S_IRUGO, rbd_client_id_show, NULL); | |
2553 | static DEVICE_ATTR(pool, S_IRUGO, rbd_pool_show, NULL); | |
9bb2f334 | 2554 | static DEVICE_ATTR(pool_id, S_IRUGO, rbd_pool_id_show, NULL); |
dfc5606d | 2555 | static DEVICE_ATTR(name, S_IRUGO, rbd_name_show, NULL); |
589d30e0 | 2556 | static DEVICE_ATTR(image_id, S_IRUGO, rbd_image_id_show, NULL); |
dfc5606d YS |
2557 | static DEVICE_ATTR(refresh, S_IWUSR, NULL, rbd_image_refresh); |
2558 | static DEVICE_ATTR(current_snap, S_IRUGO, rbd_snap_show, NULL); | |
86b00e0d | 2559 | static DEVICE_ATTR(parent, S_IRUGO, rbd_parent_show, NULL); |
dfc5606d YS |
2560 | |
2561 | static struct attribute *rbd_attrs[] = { | |
2562 | &dev_attr_size.attr, | |
34b13184 | 2563 | &dev_attr_features.attr, |
dfc5606d YS |
2564 | &dev_attr_major.attr, |
2565 | &dev_attr_client_id.attr, | |
2566 | &dev_attr_pool.attr, | |
9bb2f334 | 2567 | &dev_attr_pool_id.attr, |
dfc5606d | 2568 | &dev_attr_name.attr, |
589d30e0 | 2569 | &dev_attr_image_id.attr, |
dfc5606d | 2570 | &dev_attr_current_snap.attr, |
86b00e0d | 2571 | &dev_attr_parent.attr, |
dfc5606d | 2572 | &dev_attr_refresh.attr, |
dfc5606d YS |
2573 | NULL |
2574 | }; | |
2575 | ||
2576 | static struct attribute_group rbd_attr_group = { | |
2577 | .attrs = rbd_attrs, | |
2578 | }; | |
2579 | ||
2580 | static const struct attribute_group *rbd_attr_groups[] = { | |
2581 | &rbd_attr_group, | |
2582 | NULL | |
2583 | }; | |
2584 | ||
2585 | static void rbd_sysfs_dev_release(struct device *dev) | |
2586 | { | |
2587 | } | |
2588 | ||
2589 | static struct device_type rbd_device_type = { | |
2590 | .name = "rbd", | |
2591 | .groups = rbd_attr_groups, | |
2592 | .release = rbd_sysfs_dev_release, | |
2593 | }; | |
2594 | ||
2595 | ||
2596 | /* | |
2597 | sysfs - snapshots | |
2598 | */ | |
2599 | ||
2600 | static ssize_t rbd_snap_size_show(struct device *dev, | |
2601 | struct device_attribute *attr, | |
2602 | char *buf) | |
2603 | { | |
2604 | struct rbd_snap *snap = container_of(dev, struct rbd_snap, dev); | |
2605 | ||
3591538f | 2606 | return sprintf(buf, "%llu\n", (unsigned long long)snap->size); |
dfc5606d YS |
2607 | } |
2608 | ||
2609 | static ssize_t rbd_snap_id_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->id); |
dfc5606d YS |
2616 | } |
2617 | ||
34b13184 AE |
2618 | static ssize_t rbd_snap_features_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 | ||
2624 | return sprintf(buf, "0x%016llx\n", | |
2625 | (unsigned long long) snap->features); | |
2626 | } | |
2627 | ||
dfc5606d YS |
2628 | static DEVICE_ATTR(snap_size, S_IRUGO, rbd_snap_size_show, NULL); |
2629 | static DEVICE_ATTR(snap_id, S_IRUGO, rbd_snap_id_show, NULL); | |
34b13184 | 2630 | static DEVICE_ATTR(snap_features, S_IRUGO, rbd_snap_features_show, NULL); |
dfc5606d YS |
2631 | |
2632 | static struct attribute *rbd_snap_attrs[] = { | |
2633 | &dev_attr_snap_size.attr, | |
2634 | &dev_attr_snap_id.attr, | |
34b13184 | 2635 | &dev_attr_snap_features.attr, |
dfc5606d YS |
2636 | NULL, |
2637 | }; | |
2638 | ||
2639 | static struct attribute_group rbd_snap_attr_group = { | |
2640 | .attrs = rbd_snap_attrs, | |
2641 | }; | |
2642 | ||
2643 | static void rbd_snap_dev_release(struct device *dev) | |
2644 | { | |
2645 | struct rbd_snap *snap = container_of(dev, struct rbd_snap, dev); | |
2646 | kfree(snap->name); | |
2647 | kfree(snap); | |
2648 | } | |
2649 | ||
2650 | static const struct attribute_group *rbd_snap_attr_groups[] = { | |
2651 | &rbd_snap_attr_group, | |
2652 | NULL | |
2653 | }; | |
2654 | ||
2655 | static struct device_type rbd_snap_device_type = { | |
2656 | .groups = rbd_snap_attr_groups, | |
2657 | .release = rbd_snap_dev_release, | |
2658 | }; | |
2659 | ||
8b8fb99c AE |
2660 | static struct rbd_spec *rbd_spec_get(struct rbd_spec *spec) |
2661 | { | |
2662 | kref_get(&spec->kref); | |
2663 | ||
2664 | return spec; | |
2665 | } | |
2666 | ||
2667 | static void rbd_spec_free(struct kref *kref); | |
2668 | static void rbd_spec_put(struct rbd_spec *spec) | |
2669 | { | |
2670 | if (spec) | |
2671 | kref_put(&spec->kref, rbd_spec_free); | |
2672 | } | |
2673 | ||
2674 | static struct rbd_spec *rbd_spec_alloc(void) | |
2675 | { | |
2676 | struct rbd_spec *spec; | |
2677 | ||
2678 | spec = kzalloc(sizeof (*spec), GFP_KERNEL); | |
2679 | if (!spec) | |
2680 | return NULL; | |
2681 | kref_init(&spec->kref); | |
2682 | ||
2683 | rbd_spec_put(rbd_spec_get(spec)); /* TEMPORARY */ | |
2684 | ||
2685 | return spec; | |
2686 | } | |
2687 | ||
2688 | static void rbd_spec_free(struct kref *kref) | |
2689 | { | |
2690 | struct rbd_spec *spec = container_of(kref, struct rbd_spec, kref); | |
2691 | ||
2692 | kfree(spec->pool_name); | |
2693 | kfree(spec->image_id); | |
2694 | kfree(spec->image_name); | |
2695 | kfree(spec->snap_name); | |
2696 | kfree(spec); | |
2697 | } | |
2698 | ||
cc344fa1 | 2699 | static struct rbd_device *rbd_dev_create(struct rbd_client *rbdc, |
c53d5893 AE |
2700 | struct rbd_spec *spec) |
2701 | { | |
2702 | struct rbd_device *rbd_dev; | |
2703 | ||
2704 | rbd_dev = kzalloc(sizeof (*rbd_dev), GFP_KERNEL); | |
2705 | if (!rbd_dev) | |
2706 | return NULL; | |
2707 | ||
2708 | spin_lock_init(&rbd_dev->lock); | |
6d292906 | 2709 | rbd_dev->flags = 0; |
c53d5893 AE |
2710 | INIT_LIST_HEAD(&rbd_dev->node); |
2711 | INIT_LIST_HEAD(&rbd_dev->snaps); | |
2712 | init_rwsem(&rbd_dev->header_rwsem); | |
2713 | ||
2714 | rbd_dev->spec = spec; | |
2715 | rbd_dev->rbd_client = rbdc; | |
2716 | ||
0903e875 AE |
2717 | /* Initialize the layout used for all rbd requests */ |
2718 | ||
2719 | rbd_dev->layout.fl_stripe_unit = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER); | |
2720 | rbd_dev->layout.fl_stripe_count = cpu_to_le32(1); | |
2721 | rbd_dev->layout.fl_object_size = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER); | |
2722 | rbd_dev->layout.fl_pg_pool = cpu_to_le32((u32) spec->pool_id); | |
2723 | ||
c53d5893 AE |
2724 | return rbd_dev; |
2725 | } | |
2726 | ||
2727 | static void rbd_dev_destroy(struct rbd_device *rbd_dev) | |
2728 | { | |
86b00e0d | 2729 | rbd_spec_put(rbd_dev->parent_spec); |
c53d5893 AE |
2730 | kfree(rbd_dev->header_name); |
2731 | rbd_put_client(rbd_dev->rbd_client); | |
2732 | rbd_spec_put(rbd_dev->spec); | |
2733 | kfree(rbd_dev); | |
2734 | } | |
2735 | ||
304f6808 AE |
2736 | static bool rbd_snap_registered(struct rbd_snap *snap) |
2737 | { | |
2738 | bool ret = snap->dev.type == &rbd_snap_device_type; | |
2739 | bool reg = device_is_registered(&snap->dev); | |
2740 | ||
2741 | rbd_assert(!ret ^ reg); | |
2742 | ||
2743 | return ret; | |
2744 | } | |
2745 | ||
41f38c2b | 2746 | static void rbd_remove_snap_dev(struct rbd_snap *snap) |
dfc5606d YS |
2747 | { |
2748 | list_del(&snap->node); | |
304f6808 AE |
2749 | if (device_is_registered(&snap->dev)) |
2750 | device_unregister(&snap->dev); | |
dfc5606d YS |
2751 | } |
2752 | ||
14e7085d | 2753 | static int rbd_register_snap_dev(struct rbd_snap *snap, |
dfc5606d YS |
2754 | struct device *parent) |
2755 | { | |
2756 | struct device *dev = &snap->dev; | |
2757 | int ret; | |
2758 | ||
2759 | dev->type = &rbd_snap_device_type; | |
2760 | dev->parent = parent; | |
2761 | dev->release = rbd_snap_dev_release; | |
d4b125e9 | 2762 | dev_set_name(dev, "%s%s", RBD_SNAP_DEV_NAME_PREFIX, snap->name); |
304f6808 AE |
2763 | dout("%s: registering device for snapshot %s\n", __func__, snap->name); |
2764 | ||
dfc5606d YS |
2765 | ret = device_register(dev); |
2766 | ||
2767 | return ret; | |
2768 | } | |
2769 | ||
4e891e0a | 2770 | static struct rbd_snap *__rbd_add_snap_dev(struct rbd_device *rbd_dev, |
c8d18425 | 2771 | const char *snap_name, |
34b13184 AE |
2772 | u64 snap_id, u64 snap_size, |
2773 | u64 snap_features) | |
dfc5606d | 2774 | { |
4e891e0a | 2775 | struct rbd_snap *snap; |
dfc5606d | 2776 | int ret; |
4e891e0a AE |
2777 | |
2778 | snap = kzalloc(sizeof (*snap), GFP_KERNEL); | |
dfc5606d | 2779 | if (!snap) |
4e891e0a AE |
2780 | return ERR_PTR(-ENOMEM); |
2781 | ||
2782 | ret = -ENOMEM; | |
c8d18425 | 2783 | snap->name = kstrdup(snap_name, GFP_KERNEL); |
4e891e0a AE |
2784 | if (!snap->name) |
2785 | goto err; | |
2786 | ||
c8d18425 AE |
2787 | snap->id = snap_id; |
2788 | snap->size = snap_size; | |
34b13184 | 2789 | snap->features = snap_features; |
4e891e0a AE |
2790 | |
2791 | return snap; | |
2792 | ||
dfc5606d YS |
2793 | err: |
2794 | kfree(snap->name); | |
2795 | kfree(snap); | |
4e891e0a AE |
2796 | |
2797 | return ERR_PTR(ret); | |
dfc5606d YS |
2798 | } |
2799 | ||
cd892126 AE |
2800 | static char *rbd_dev_v1_snap_info(struct rbd_device *rbd_dev, u32 which, |
2801 | u64 *snap_size, u64 *snap_features) | |
2802 | { | |
2803 | char *snap_name; | |
2804 | ||
2805 | rbd_assert(which < rbd_dev->header.snapc->num_snaps); | |
2806 | ||
2807 | *snap_size = rbd_dev->header.snap_sizes[which]; | |
2808 | *snap_features = 0; /* No features for v1 */ | |
2809 | ||
2810 | /* Skip over names until we find the one we are looking for */ | |
2811 | ||
2812 | snap_name = rbd_dev->header.snap_names; | |
2813 | while (which--) | |
2814 | snap_name += strlen(snap_name) + 1; | |
2815 | ||
2816 | return snap_name; | |
2817 | } | |
2818 | ||
9d475de5 AE |
2819 | /* |
2820 | * Get the size and object order for an image snapshot, or if | |
2821 | * snap_id is CEPH_NOSNAP, gets this information for the base | |
2822 | * image. | |
2823 | */ | |
2824 | static int _rbd_dev_v2_snap_size(struct rbd_device *rbd_dev, u64 snap_id, | |
2825 | u8 *order, u64 *snap_size) | |
2826 | { | |
2827 | __le64 snapid = cpu_to_le64(snap_id); | |
2828 | int ret; | |
2829 | struct { | |
2830 | u8 order; | |
2831 | __le64 size; | |
2832 | } __attribute__ ((packed)) size_buf = { 0 }; | |
2833 | ||
36be9a76 | 2834 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
9d475de5 AE |
2835 | "rbd", "get_size", |
2836 | (char *) &snapid, sizeof (snapid), | |
07b2391f | 2837 | (char *) &size_buf, sizeof (size_buf), NULL); |
36be9a76 | 2838 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
9d475de5 AE |
2839 | if (ret < 0) |
2840 | return ret; | |
2841 | ||
2842 | *order = size_buf.order; | |
2843 | *snap_size = le64_to_cpu(size_buf.size); | |
2844 | ||
2845 | dout(" snap_id 0x%016llx order = %u, snap_size = %llu\n", | |
2846 | (unsigned long long) snap_id, (unsigned int) *order, | |
2847 | (unsigned long long) *snap_size); | |
2848 | ||
2849 | return 0; | |
2850 | } | |
2851 | ||
2852 | static int rbd_dev_v2_image_size(struct rbd_device *rbd_dev) | |
2853 | { | |
2854 | return _rbd_dev_v2_snap_size(rbd_dev, CEPH_NOSNAP, | |
2855 | &rbd_dev->header.obj_order, | |
2856 | &rbd_dev->header.image_size); | |
2857 | } | |
2858 | ||
1e130199 AE |
2859 | static int rbd_dev_v2_object_prefix(struct rbd_device *rbd_dev) |
2860 | { | |
2861 | void *reply_buf; | |
2862 | int ret; | |
2863 | void *p; | |
2864 | ||
2865 | reply_buf = kzalloc(RBD_OBJ_PREFIX_LEN_MAX, GFP_KERNEL); | |
2866 | if (!reply_buf) | |
2867 | return -ENOMEM; | |
2868 | ||
36be9a76 | 2869 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
1e130199 AE |
2870 | "rbd", "get_object_prefix", |
2871 | NULL, 0, | |
07b2391f | 2872 | reply_buf, RBD_OBJ_PREFIX_LEN_MAX, NULL); |
36be9a76 | 2873 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
1e130199 AE |
2874 | if (ret < 0) |
2875 | goto out; | |
2876 | ||
2877 | p = reply_buf; | |
2878 | rbd_dev->header.object_prefix = ceph_extract_encoded_string(&p, | |
2879 | p + RBD_OBJ_PREFIX_LEN_MAX, | |
2880 | NULL, GFP_NOIO); | |
2881 | ||
2882 | if (IS_ERR(rbd_dev->header.object_prefix)) { | |
2883 | ret = PTR_ERR(rbd_dev->header.object_prefix); | |
2884 | rbd_dev->header.object_prefix = NULL; | |
2885 | } else { | |
2886 | dout(" object_prefix = %s\n", rbd_dev->header.object_prefix); | |
2887 | } | |
2888 | ||
2889 | out: | |
2890 | kfree(reply_buf); | |
2891 | ||
2892 | return ret; | |
2893 | } | |
2894 | ||
b1b5402a AE |
2895 | static int _rbd_dev_v2_snap_features(struct rbd_device *rbd_dev, u64 snap_id, |
2896 | u64 *snap_features) | |
2897 | { | |
2898 | __le64 snapid = cpu_to_le64(snap_id); | |
2899 | struct { | |
2900 | __le64 features; | |
2901 | __le64 incompat; | |
2902 | } features_buf = { 0 }; | |
d889140c | 2903 | u64 incompat; |
b1b5402a AE |
2904 | int ret; |
2905 | ||
36be9a76 | 2906 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
b1b5402a AE |
2907 | "rbd", "get_features", |
2908 | (char *) &snapid, sizeof (snapid), | |
2909 | (char *) &features_buf, sizeof (features_buf), | |
07b2391f | 2910 | NULL); |
36be9a76 | 2911 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
b1b5402a AE |
2912 | if (ret < 0) |
2913 | return ret; | |
d889140c AE |
2914 | |
2915 | incompat = le64_to_cpu(features_buf.incompat); | |
2916 | if (incompat & ~RBD_FEATURES_ALL) | |
b8f5c6ed | 2917 | return -ENXIO; |
d889140c | 2918 | |
b1b5402a AE |
2919 | *snap_features = le64_to_cpu(features_buf.features); |
2920 | ||
2921 | dout(" snap_id 0x%016llx features = 0x%016llx incompat = 0x%016llx\n", | |
2922 | (unsigned long long) snap_id, | |
2923 | (unsigned long long) *snap_features, | |
2924 | (unsigned long long) le64_to_cpu(features_buf.incompat)); | |
2925 | ||
2926 | return 0; | |
2927 | } | |
2928 | ||
2929 | static int rbd_dev_v2_features(struct rbd_device *rbd_dev) | |
2930 | { | |
2931 | return _rbd_dev_v2_snap_features(rbd_dev, CEPH_NOSNAP, | |
2932 | &rbd_dev->header.features); | |
2933 | } | |
2934 | ||
86b00e0d AE |
2935 | static int rbd_dev_v2_parent_info(struct rbd_device *rbd_dev) |
2936 | { | |
2937 | struct rbd_spec *parent_spec; | |
2938 | size_t size; | |
2939 | void *reply_buf = NULL; | |
2940 | __le64 snapid; | |
2941 | void *p; | |
2942 | void *end; | |
2943 | char *image_id; | |
2944 | u64 overlap; | |
86b00e0d AE |
2945 | int ret; |
2946 | ||
2947 | parent_spec = rbd_spec_alloc(); | |
2948 | if (!parent_spec) | |
2949 | return -ENOMEM; | |
2950 | ||
2951 | size = sizeof (__le64) + /* pool_id */ | |
2952 | sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX + /* image_id */ | |
2953 | sizeof (__le64) + /* snap_id */ | |
2954 | sizeof (__le64); /* overlap */ | |
2955 | reply_buf = kmalloc(size, GFP_KERNEL); | |
2956 | if (!reply_buf) { | |
2957 | ret = -ENOMEM; | |
2958 | goto out_err; | |
2959 | } | |
2960 | ||
2961 | snapid = cpu_to_le64(CEPH_NOSNAP); | |
36be9a76 | 2962 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
86b00e0d AE |
2963 | "rbd", "get_parent", |
2964 | (char *) &snapid, sizeof (snapid), | |
07b2391f | 2965 | (char *) reply_buf, size, NULL); |
36be9a76 | 2966 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
86b00e0d AE |
2967 | if (ret < 0) |
2968 | goto out_err; | |
2969 | ||
2970 | ret = -ERANGE; | |
2971 | p = reply_buf; | |
2972 | end = (char *) reply_buf + size; | |
2973 | ceph_decode_64_safe(&p, end, parent_spec->pool_id, out_err); | |
2974 | if (parent_spec->pool_id == CEPH_NOPOOL) | |
2975 | goto out; /* No parent? No problem. */ | |
2976 | ||
0903e875 AE |
2977 | /* The ceph file layout needs to fit pool id in 32 bits */ |
2978 | ||
2979 | ret = -EIO; | |
2980 | if (WARN_ON(parent_spec->pool_id > (u64) U32_MAX)) | |
2981 | goto out; | |
2982 | ||
979ed480 | 2983 | image_id = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL); |
86b00e0d AE |
2984 | if (IS_ERR(image_id)) { |
2985 | ret = PTR_ERR(image_id); | |
2986 | goto out_err; | |
2987 | } | |
2988 | parent_spec->image_id = image_id; | |
2989 | ceph_decode_64_safe(&p, end, parent_spec->snap_id, out_err); | |
2990 | ceph_decode_64_safe(&p, end, overlap, out_err); | |
2991 | ||
2992 | rbd_dev->parent_overlap = overlap; | |
2993 | rbd_dev->parent_spec = parent_spec; | |
2994 | parent_spec = NULL; /* rbd_dev now owns this */ | |
2995 | out: | |
2996 | ret = 0; | |
2997 | out_err: | |
2998 | kfree(reply_buf); | |
2999 | rbd_spec_put(parent_spec); | |
3000 | ||
3001 | return ret; | |
3002 | } | |
3003 | ||
9e15b77d AE |
3004 | static char *rbd_dev_image_name(struct rbd_device *rbd_dev) |
3005 | { | |
3006 | size_t image_id_size; | |
3007 | char *image_id; | |
3008 | void *p; | |
3009 | void *end; | |
3010 | size_t size; | |
3011 | void *reply_buf = NULL; | |
3012 | size_t len = 0; | |
3013 | char *image_name = NULL; | |
3014 | int ret; | |
3015 | ||
3016 | rbd_assert(!rbd_dev->spec->image_name); | |
3017 | ||
69e7a02f AE |
3018 | len = strlen(rbd_dev->spec->image_id); |
3019 | image_id_size = sizeof (__le32) + len; | |
9e15b77d AE |
3020 | image_id = kmalloc(image_id_size, GFP_KERNEL); |
3021 | if (!image_id) | |
3022 | return NULL; | |
3023 | ||
3024 | p = image_id; | |
3025 | end = (char *) image_id + image_id_size; | |
69e7a02f | 3026 | ceph_encode_string(&p, end, rbd_dev->spec->image_id, (u32) len); |
9e15b77d AE |
3027 | |
3028 | size = sizeof (__le32) + RBD_IMAGE_NAME_LEN_MAX; | |
3029 | reply_buf = kmalloc(size, GFP_KERNEL); | |
3030 | if (!reply_buf) | |
3031 | goto out; | |
3032 | ||
36be9a76 | 3033 | ret = rbd_obj_method_sync(rbd_dev, RBD_DIRECTORY, |
9e15b77d AE |
3034 | "rbd", "dir_get_name", |
3035 | image_id, image_id_size, | |
07b2391f | 3036 | (char *) reply_buf, size, NULL); |
9e15b77d AE |
3037 | if (ret < 0) |
3038 | goto out; | |
3039 | p = reply_buf; | |
3040 | end = (char *) reply_buf + size; | |
3041 | image_name = ceph_extract_encoded_string(&p, end, &len, GFP_KERNEL); | |
3042 | if (IS_ERR(image_name)) | |
3043 | image_name = NULL; | |
3044 | else | |
3045 | dout("%s: name is %s len is %zd\n", __func__, image_name, len); | |
3046 | out: | |
3047 | kfree(reply_buf); | |
3048 | kfree(image_id); | |
3049 | ||
3050 | return image_name; | |
3051 | } | |
3052 | ||
3053 | /* | |
3054 | * When a parent image gets probed, we only have the pool, image, | |
3055 | * and snapshot ids but not the names of any of them. This call | |
3056 | * is made later to fill in those names. It has to be done after | |
3057 | * rbd_dev_snaps_update() has completed because some of the | |
3058 | * information (in particular, snapshot name) is not available | |
3059 | * until then. | |
3060 | */ | |
3061 | static int rbd_dev_probe_update_spec(struct rbd_device *rbd_dev) | |
3062 | { | |
3063 | struct ceph_osd_client *osdc; | |
3064 | const char *name; | |
3065 | void *reply_buf = NULL; | |
3066 | int ret; | |
3067 | ||
3068 | if (rbd_dev->spec->pool_name) | |
3069 | return 0; /* Already have the names */ | |
3070 | ||
3071 | /* Look up the pool name */ | |
3072 | ||
3073 | osdc = &rbd_dev->rbd_client->client->osdc; | |
3074 | name = ceph_pg_pool_name_by_id(osdc->osdmap, rbd_dev->spec->pool_id); | |
935dc89f AE |
3075 | if (!name) { |
3076 | rbd_warn(rbd_dev, "there is no pool with id %llu", | |
3077 | rbd_dev->spec->pool_id); /* Really a BUG() */ | |
3078 | return -EIO; | |
3079 | } | |
9e15b77d AE |
3080 | |
3081 | rbd_dev->spec->pool_name = kstrdup(name, GFP_KERNEL); | |
3082 | if (!rbd_dev->spec->pool_name) | |
3083 | return -ENOMEM; | |
3084 | ||
3085 | /* Fetch the image name; tolerate failure here */ | |
3086 | ||
3087 | name = rbd_dev_image_name(rbd_dev); | |
69e7a02f | 3088 | if (name) |
9e15b77d | 3089 | rbd_dev->spec->image_name = (char *) name; |
69e7a02f | 3090 | else |
06ecc6cb | 3091 | rbd_warn(rbd_dev, "unable to get image name"); |
9e15b77d AE |
3092 | |
3093 | /* Look up the snapshot name. */ | |
3094 | ||
3095 | name = rbd_snap_name(rbd_dev, rbd_dev->spec->snap_id); | |
3096 | if (!name) { | |
935dc89f AE |
3097 | rbd_warn(rbd_dev, "no snapshot with id %llu", |
3098 | rbd_dev->spec->snap_id); /* Really a BUG() */ | |
9e15b77d AE |
3099 | ret = -EIO; |
3100 | goto out_err; | |
3101 | } | |
3102 | rbd_dev->spec->snap_name = kstrdup(name, GFP_KERNEL); | |
3103 | if(!rbd_dev->spec->snap_name) | |
3104 | goto out_err; | |
3105 | ||
3106 | return 0; | |
3107 | out_err: | |
3108 | kfree(reply_buf); | |
3109 | kfree(rbd_dev->spec->pool_name); | |
3110 | rbd_dev->spec->pool_name = NULL; | |
3111 | ||
3112 | return ret; | |
3113 | } | |
3114 | ||
6e14b1a6 | 3115 | static int rbd_dev_v2_snap_context(struct rbd_device *rbd_dev, u64 *ver) |
35d489f9 AE |
3116 | { |
3117 | size_t size; | |
3118 | int ret; | |
3119 | void *reply_buf; | |
3120 | void *p; | |
3121 | void *end; | |
3122 | u64 seq; | |
3123 | u32 snap_count; | |
3124 | struct ceph_snap_context *snapc; | |
3125 | u32 i; | |
3126 | ||
3127 | /* | |
3128 | * We'll need room for the seq value (maximum snapshot id), | |
3129 | * snapshot count, and array of that many snapshot ids. | |
3130 | * For now we have a fixed upper limit on the number we're | |
3131 | * prepared to receive. | |
3132 | */ | |
3133 | size = sizeof (__le64) + sizeof (__le32) + | |
3134 | RBD_MAX_SNAP_COUNT * sizeof (__le64); | |
3135 | reply_buf = kzalloc(size, GFP_KERNEL); | |
3136 | if (!reply_buf) | |
3137 | return -ENOMEM; | |
3138 | ||
36be9a76 | 3139 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
35d489f9 AE |
3140 | "rbd", "get_snapcontext", |
3141 | NULL, 0, | |
07b2391f | 3142 | reply_buf, size, ver); |
36be9a76 | 3143 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
35d489f9 AE |
3144 | if (ret < 0) |
3145 | goto out; | |
3146 | ||
3147 | ret = -ERANGE; | |
3148 | p = reply_buf; | |
3149 | end = (char *) reply_buf + size; | |
3150 | ceph_decode_64_safe(&p, end, seq, out); | |
3151 | ceph_decode_32_safe(&p, end, snap_count, out); | |
3152 | ||
3153 | /* | |
3154 | * Make sure the reported number of snapshot ids wouldn't go | |
3155 | * beyond the end of our buffer. But before checking that, | |
3156 | * make sure the computed size of the snapshot context we | |
3157 | * allocate is representable in a size_t. | |
3158 | */ | |
3159 | if (snap_count > (SIZE_MAX - sizeof (struct ceph_snap_context)) | |
3160 | / sizeof (u64)) { | |
3161 | ret = -EINVAL; | |
3162 | goto out; | |
3163 | } | |
3164 | if (!ceph_has_room(&p, end, snap_count * sizeof (__le64))) | |
3165 | goto out; | |
3166 | ||
3167 | size = sizeof (struct ceph_snap_context) + | |
3168 | snap_count * sizeof (snapc->snaps[0]); | |
3169 | snapc = kmalloc(size, GFP_KERNEL); | |
3170 | if (!snapc) { | |
3171 | ret = -ENOMEM; | |
3172 | goto out; | |
3173 | } | |
3174 | ||
3175 | atomic_set(&snapc->nref, 1); | |
3176 | snapc->seq = seq; | |
3177 | snapc->num_snaps = snap_count; | |
3178 | for (i = 0; i < snap_count; i++) | |
3179 | snapc->snaps[i] = ceph_decode_64(&p); | |
3180 | ||
3181 | rbd_dev->header.snapc = snapc; | |
3182 | ||
3183 | dout(" snap context seq = %llu, snap_count = %u\n", | |
3184 | (unsigned long long) seq, (unsigned int) snap_count); | |
3185 | ||
3186 | out: | |
3187 | kfree(reply_buf); | |
3188 | ||
3189 | return 0; | |
3190 | } | |
3191 | ||
b8b1e2db AE |
3192 | static char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev, u32 which) |
3193 | { | |
3194 | size_t size; | |
3195 | void *reply_buf; | |
3196 | __le64 snap_id; | |
3197 | int ret; | |
3198 | void *p; | |
3199 | void *end; | |
b8b1e2db AE |
3200 | char *snap_name; |
3201 | ||
3202 | size = sizeof (__le32) + RBD_MAX_SNAP_NAME_LEN; | |
3203 | reply_buf = kmalloc(size, GFP_KERNEL); | |
3204 | if (!reply_buf) | |
3205 | return ERR_PTR(-ENOMEM); | |
3206 | ||
3207 | snap_id = cpu_to_le64(rbd_dev->header.snapc->snaps[which]); | |
36be9a76 | 3208 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
b8b1e2db AE |
3209 | "rbd", "get_snapshot_name", |
3210 | (char *) &snap_id, sizeof (snap_id), | |
07b2391f | 3211 | reply_buf, size, NULL); |
36be9a76 | 3212 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
b8b1e2db AE |
3213 | if (ret < 0) |
3214 | goto out; | |
3215 | ||
3216 | p = reply_buf; | |
3217 | end = (char *) reply_buf + size; | |
e5c35534 | 3218 | snap_name = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL); |
b8b1e2db AE |
3219 | if (IS_ERR(snap_name)) { |
3220 | ret = PTR_ERR(snap_name); | |
3221 | goto out; | |
3222 | } else { | |
3223 | dout(" snap_id 0x%016llx snap_name = %s\n", | |
3224 | (unsigned long long) le64_to_cpu(snap_id), snap_name); | |
3225 | } | |
3226 | kfree(reply_buf); | |
3227 | ||
3228 | return snap_name; | |
3229 | out: | |
3230 | kfree(reply_buf); | |
3231 | ||
3232 | return ERR_PTR(ret); | |
3233 | } | |
3234 | ||
3235 | static char *rbd_dev_v2_snap_info(struct rbd_device *rbd_dev, u32 which, | |
3236 | u64 *snap_size, u64 *snap_features) | |
3237 | { | |
e0b49868 | 3238 | u64 snap_id; |
b8b1e2db AE |
3239 | u8 order; |
3240 | int ret; | |
3241 | ||
3242 | snap_id = rbd_dev->header.snapc->snaps[which]; | |
3243 | ret = _rbd_dev_v2_snap_size(rbd_dev, snap_id, &order, snap_size); | |
3244 | if (ret) | |
3245 | return ERR_PTR(ret); | |
3246 | ret = _rbd_dev_v2_snap_features(rbd_dev, snap_id, snap_features); | |
3247 | if (ret) | |
3248 | return ERR_PTR(ret); | |
3249 | ||
3250 | return rbd_dev_v2_snap_name(rbd_dev, which); | |
3251 | } | |
3252 | ||
3253 | static char *rbd_dev_snap_info(struct rbd_device *rbd_dev, u32 which, | |
3254 | u64 *snap_size, u64 *snap_features) | |
3255 | { | |
3256 | if (rbd_dev->image_format == 1) | |
3257 | return rbd_dev_v1_snap_info(rbd_dev, which, | |
3258 | snap_size, snap_features); | |
3259 | if (rbd_dev->image_format == 2) | |
3260 | return rbd_dev_v2_snap_info(rbd_dev, which, | |
3261 | snap_size, snap_features); | |
3262 | return ERR_PTR(-EINVAL); | |
3263 | } | |
3264 | ||
117973fb AE |
3265 | static int rbd_dev_v2_refresh(struct rbd_device *rbd_dev, u64 *hver) |
3266 | { | |
3267 | int ret; | |
3268 | __u8 obj_order; | |
3269 | ||
3270 | down_write(&rbd_dev->header_rwsem); | |
3271 | ||
3272 | /* Grab old order first, to see if it changes */ | |
3273 | ||
3274 | obj_order = rbd_dev->header.obj_order, | |
3275 | ret = rbd_dev_v2_image_size(rbd_dev); | |
3276 | if (ret) | |
3277 | goto out; | |
3278 | if (rbd_dev->header.obj_order != obj_order) { | |
3279 | ret = -EIO; | |
3280 | goto out; | |
3281 | } | |
3282 | rbd_update_mapping_size(rbd_dev); | |
3283 | ||
3284 | ret = rbd_dev_v2_snap_context(rbd_dev, hver); | |
3285 | dout("rbd_dev_v2_snap_context returned %d\n", ret); | |
3286 | if (ret) | |
3287 | goto out; | |
3288 | ret = rbd_dev_snaps_update(rbd_dev); | |
3289 | dout("rbd_dev_snaps_update returned %d\n", ret); | |
3290 | if (ret) | |
3291 | goto out; | |
3292 | ret = rbd_dev_snaps_register(rbd_dev); | |
3293 | dout("rbd_dev_snaps_register returned %d\n", ret); | |
3294 | out: | |
3295 | up_write(&rbd_dev->header_rwsem); | |
3296 | ||
3297 | return ret; | |
3298 | } | |
3299 | ||
dfc5606d | 3300 | /* |
35938150 AE |
3301 | * Scan the rbd device's current snapshot list and compare it to the |
3302 | * newly-received snapshot context. Remove any existing snapshots | |
3303 | * not present in the new snapshot context. Add a new snapshot for | |
3304 | * any snaphots in the snapshot context not in the current list. | |
3305 | * And verify there are no changes to snapshots we already know | |
3306 | * about. | |
3307 | * | |
3308 | * Assumes the snapshots in the snapshot context are sorted by | |
3309 | * snapshot id, highest id first. (Snapshots in the rbd_dev's list | |
3310 | * are also maintained in that order.) | |
dfc5606d | 3311 | */ |
304f6808 | 3312 | static int rbd_dev_snaps_update(struct rbd_device *rbd_dev) |
dfc5606d | 3313 | { |
35938150 AE |
3314 | struct ceph_snap_context *snapc = rbd_dev->header.snapc; |
3315 | const u32 snap_count = snapc->num_snaps; | |
35938150 AE |
3316 | struct list_head *head = &rbd_dev->snaps; |
3317 | struct list_head *links = head->next; | |
3318 | u32 index = 0; | |
dfc5606d | 3319 | |
9fcbb800 | 3320 | dout("%s: snap count is %u\n", __func__, (unsigned int) snap_count); |
35938150 AE |
3321 | while (index < snap_count || links != head) { |
3322 | u64 snap_id; | |
3323 | struct rbd_snap *snap; | |
cd892126 AE |
3324 | char *snap_name; |
3325 | u64 snap_size = 0; | |
3326 | u64 snap_features = 0; | |
dfc5606d | 3327 | |
35938150 AE |
3328 | snap_id = index < snap_count ? snapc->snaps[index] |
3329 | : CEPH_NOSNAP; | |
3330 | snap = links != head ? list_entry(links, struct rbd_snap, node) | |
3331 | : NULL; | |
aafb230e | 3332 | rbd_assert(!snap || snap->id != CEPH_NOSNAP); |
dfc5606d | 3333 | |
35938150 AE |
3334 | if (snap_id == CEPH_NOSNAP || (snap && snap->id > snap_id)) { |
3335 | struct list_head *next = links->next; | |
dfc5606d | 3336 | |
6d292906 AE |
3337 | /* |
3338 | * A previously-existing snapshot is not in | |
3339 | * the new snap context. | |
3340 | * | |
3341 | * If the now missing snapshot is the one the | |
3342 | * image is mapped to, clear its exists flag | |
3343 | * so we can avoid sending any more requests | |
3344 | * to it. | |
3345 | */ | |
0d7dbfce | 3346 | if (rbd_dev->spec->snap_id == snap->id) |
6d292906 | 3347 | clear_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags); |
41f38c2b | 3348 | rbd_remove_snap_dev(snap); |
9fcbb800 | 3349 | dout("%ssnap id %llu has been removed\n", |
0d7dbfce AE |
3350 | rbd_dev->spec->snap_id == snap->id ? |
3351 | "mapped " : "", | |
9fcbb800 | 3352 | (unsigned long long) snap->id); |
35938150 AE |
3353 | |
3354 | /* Done with this list entry; advance */ | |
3355 | ||
3356 | links = next; | |
dfc5606d YS |
3357 | continue; |
3358 | } | |
35938150 | 3359 | |
b8b1e2db AE |
3360 | snap_name = rbd_dev_snap_info(rbd_dev, index, |
3361 | &snap_size, &snap_features); | |
cd892126 AE |
3362 | if (IS_ERR(snap_name)) |
3363 | return PTR_ERR(snap_name); | |
3364 | ||
9fcbb800 AE |
3365 | dout("entry %u: snap_id = %llu\n", (unsigned int) snap_count, |
3366 | (unsigned long long) snap_id); | |
35938150 AE |
3367 | if (!snap || (snap_id != CEPH_NOSNAP && snap->id < snap_id)) { |
3368 | struct rbd_snap *new_snap; | |
3369 | ||
3370 | /* We haven't seen this snapshot before */ | |
3371 | ||
c8d18425 | 3372 | new_snap = __rbd_add_snap_dev(rbd_dev, snap_name, |
cd892126 | 3373 | snap_id, snap_size, snap_features); |
9fcbb800 AE |
3374 | if (IS_ERR(new_snap)) { |
3375 | int err = PTR_ERR(new_snap); | |
3376 | ||
3377 | dout(" failed to add dev, error %d\n", err); | |
3378 | ||
3379 | return err; | |
3380 | } | |
35938150 AE |
3381 | |
3382 | /* New goes before existing, or at end of list */ | |
3383 | ||
9fcbb800 | 3384 | dout(" added dev%s\n", snap ? "" : " at end\n"); |
35938150 AE |
3385 | if (snap) |
3386 | list_add_tail(&new_snap->node, &snap->node); | |
3387 | else | |
523f3258 | 3388 | list_add_tail(&new_snap->node, head); |
35938150 AE |
3389 | } else { |
3390 | /* Already have this one */ | |
3391 | ||
9fcbb800 AE |
3392 | dout(" already present\n"); |
3393 | ||
cd892126 | 3394 | rbd_assert(snap->size == snap_size); |
aafb230e | 3395 | rbd_assert(!strcmp(snap->name, snap_name)); |
cd892126 | 3396 | rbd_assert(snap->features == snap_features); |
35938150 AE |
3397 | |
3398 | /* Done with this list entry; advance */ | |
3399 | ||
3400 | links = links->next; | |
dfc5606d | 3401 | } |
35938150 AE |
3402 | |
3403 | /* Advance to the next entry in the snapshot context */ | |
3404 | ||
3405 | index++; | |
dfc5606d | 3406 | } |
9fcbb800 | 3407 | dout("%s: done\n", __func__); |
dfc5606d YS |
3408 | |
3409 | return 0; | |
3410 | } | |
3411 | ||
304f6808 AE |
3412 | /* |
3413 | * Scan the list of snapshots and register the devices for any that | |
3414 | * have not already been registered. | |
3415 | */ | |
3416 | static int rbd_dev_snaps_register(struct rbd_device *rbd_dev) | |
3417 | { | |
3418 | struct rbd_snap *snap; | |
3419 | int ret = 0; | |
3420 | ||
37206ee5 | 3421 | dout("%s:\n", __func__); |
86ff77bb AE |
3422 | if (WARN_ON(!device_is_registered(&rbd_dev->dev))) |
3423 | return -EIO; | |
304f6808 AE |
3424 | |
3425 | list_for_each_entry(snap, &rbd_dev->snaps, node) { | |
3426 | if (!rbd_snap_registered(snap)) { | |
3427 | ret = rbd_register_snap_dev(snap, &rbd_dev->dev); | |
3428 | if (ret < 0) | |
3429 | break; | |
3430 | } | |
3431 | } | |
3432 | dout("%s: returning %d\n", __func__, ret); | |
3433 | ||
3434 | return ret; | |
3435 | } | |
3436 | ||
dfc5606d YS |
3437 | static int rbd_bus_add_dev(struct rbd_device *rbd_dev) |
3438 | { | |
dfc5606d | 3439 | struct device *dev; |
cd789ab9 | 3440 | int ret; |
dfc5606d YS |
3441 | |
3442 | mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING); | |
dfc5606d | 3443 | |
cd789ab9 | 3444 | dev = &rbd_dev->dev; |
dfc5606d YS |
3445 | dev->bus = &rbd_bus_type; |
3446 | dev->type = &rbd_device_type; | |
3447 | dev->parent = &rbd_root_dev; | |
3448 | dev->release = rbd_dev_release; | |
de71a297 | 3449 | dev_set_name(dev, "%d", rbd_dev->dev_id); |
dfc5606d | 3450 | ret = device_register(dev); |
dfc5606d | 3451 | |
dfc5606d | 3452 | mutex_unlock(&ctl_mutex); |
cd789ab9 | 3453 | |
dfc5606d | 3454 | return ret; |
602adf40 YS |
3455 | } |
3456 | ||
dfc5606d YS |
3457 | static void rbd_bus_del_dev(struct rbd_device *rbd_dev) |
3458 | { | |
3459 | device_unregister(&rbd_dev->dev); | |
3460 | } | |
3461 | ||
e2839308 | 3462 | static atomic64_t rbd_dev_id_max = ATOMIC64_INIT(0); |
1ddbe94e AE |
3463 | |
3464 | /* | |
499afd5b AE |
3465 | * Get a unique rbd identifier for the given new rbd_dev, and add |
3466 | * the rbd_dev to the global list. The minimum rbd id is 1. | |
1ddbe94e | 3467 | */ |
e2839308 | 3468 | static void rbd_dev_id_get(struct rbd_device *rbd_dev) |
b7f23c36 | 3469 | { |
e2839308 | 3470 | rbd_dev->dev_id = atomic64_inc_return(&rbd_dev_id_max); |
499afd5b AE |
3471 | |
3472 | spin_lock(&rbd_dev_list_lock); | |
3473 | list_add_tail(&rbd_dev->node, &rbd_dev_list); | |
3474 | spin_unlock(&rbd_dev_list_lock); | |
e2839308 AE |
3475 | dout("rbd_dev %p given dev id %llu\n", rbd_dev, |
3476 | (unsigned long long) rbd_dev->dev_id); | |
1ddbe94e | 3477 | } |
b7f23c36 | 3478 | |
1ddbe94e | 3479 | /* |
499afd5b AE |
3480 | * Remove an rbd_dev from the global list, and record that its |
3481 | * identifier is no longer in use. | |
1ddbe94e | 3482 | */ |
e2839308 | 3483 | static void rbd_dev_id_put(struct rbd_device *rbd_dev) |
1ddbe94e | 3484 | { |
d184f6bf | 3485 | struct list_head *tmp; |
de71a297 | 3486 | int rbd_id = rbd_dev->dev_id; |
d184f6bf AE |
3487 | int max_id; |
3488 | ||
aafb230e | 3489 | rbd_assert(rbd_id > 0); |
499afd5b | 3490 | |
e2839308 AE |
3491 | dout("rbd_dev %p released dev id %llu\n", rbd_dev, |
3492 | (unsigned long long) rbd_dev->dev_id); | |
499afd5b AE |
3493 | spin_lock(&rbd_dev_list_lock); |
3494 | list_del_init(&rbd_dev->node); | |
d184f6bf AE |
3495 | |
3496 | /* | |
3497 | * If the id being "put" is not the current maximum, there | |
3498 | * is nothing special we need to do. | |
3499 | */ | |
e2839308 | 3500 | if (rbd_id != atomic64_read(&rbd_dev_id_max)) { |
d184f6bf AE |
3501 | spin_unlock(&rbd_dev_list_lock); |
3502 | return; | |
3503 | } | |
3504 | ||
3505 | /* | |
3506 | * We need to update the current maximum id. Search the | |
3507 | * list to find out what it is. We're more likely to find | |
3508 | * the maximum at the end, so search the list backward. | |
3509 | */ | |
3510 | max_id = 0; | |
3511 | list_for_each_prev(tmp, &rbd_dev_list) { | |
3512 | struct rbd_device *rbd_dev; | |
3513 | ||
3514 | rbd_dev = list_entry(tmp, struct rbd_device, node); | |
b213e0b1 AE |
3515 | if (rbd_dev->dev_id > max_id) |
3516 | max_id = rbd_dev->dev_id; | |
d184f6bf | 3517 | } |
499afd5b | 3518 | spin_unlock(&rbd_dev_list_lock); |
b7f23c36 | 3519 | |
1ddbe94e | 3520 | /* |
e2839308 | 3521 | * The max id could have been updated by rbd_dev_id_get(), in |
d184f6bf AE |
3522 | * which case it now accurately reflects the new maximum. |
3523 | * Be careful not to overwrite the maximum value in that | |
3524 | * case. | |
1ddbe94e | 3525 | */ |
e2839308 AE |
3526 | atomic64_cmpxchg(&rbd_dev_id_max, rbd_id, max_id); |
3527 | dout(" max dev id has been reset\n"); | |
b7f23c36 AE |
3528 | } |
3529 | ||
e28fff26 AE |
3530 | /* |
3531 | * Skips over white space at *buf, and updates *buf to point to the | |
3532 | * first found non-space character (if any). Returns the length of | |
593a9e7b AE |
3533 | * the token (string of non-white space characters) found. Note |
3534 | * that *buf must be terminated with '\0'. | |
e28fff26 AE |
3535 | */ |
3536 | static inline size_t next_token(const char **buf) | |
3537 | { | |
3538 | /* | |
3539 | * These are the characters that produce nonzero for | |
3540 | * isspace() in the "C" and "POSIX" locales. | |
3541 | */ | |
3542 | const char *spaces = " \f\n\r\t\v"; | |
3543 | ||
3544 | *buf += strspn(*buf, spaces); /* Find start of token */ | |
3545 | ||
3546 | return strcspn(*buf, spaces); /* Return token length */ | |
3547 | } | |
3548 | ||
3549 | /* | |
3550 | * Finds the next token in *buf, and if the provided token buffer is | |
3551 | * big enough, copies the found token into it. The result, if | |
593a9e7b AE |
3552 | * copied, is guaranteed to be terminated with '\0'. Note that *buf |
3553 | * must be terminated with '\0' on entry. | |
e28fff26 AE |
3554 | * |
3555 | * Returns the length of the token found (not including the '\0'). | |
3556 | * Return value will be 0 if no token is found, and it will be >= | |
3557 | * token_size if the token would not fit. | |
3558 | * | |
593a9e7b | 3559 | * The *buf pointer will be updated to point beyond the end of the |
e28fff26 AE |
3560 | * found token. Note that this occurs even if the token buffer is |
3561 | * too small to hold it. | |
3562 | */ | |
3563 | static inline size_t copy_token(const char **buf, | |
3564 | char *token, | |
3565 | size_t token_size) | |
3566 | { | |
3567 | size_t len; | |
3568 | ||
3569 | len = next_token(buf); | |
3570 | if (len < token_size) { | |
3571 | memcpy(token, *buf, len); | |
3572 | *(token + len) = '\0'; | |
3573 | } | |
3574 | *buf += len; | |
3575 | ||
3576 | return len; | |
3577 | } | |
3578 | ||
ea3352f4 AE |
3579 | /* |
3580 | * Finds the next token in *buf, dynamically allocates a buffer big | |
3581 | * enough to hold a copy of it, and copies the token into the new | |
3582 | * buffer. The copy is guaranteed to be terminated with '\0'. Note | |
3583 | * that a duplicate buffer is created even for a zero-length token. | |
3584 | * | |
3585 | * Returns a pointer to the newly-allocated duplicate, or a null | |
3586 | * pointer if memory for the duplicate was not available. If | |
3587 | * the lenp argument is a non-null pointer, the length of the token | |
3588 | * (not including the '\0') is returned in *lenp. | |
3589 | * | |
3590 | * If successful, the *buf pointer will be updated to point beyond | |
3591 | * the end of the found token. | |
3592 | * | |
3593 | * Note: uses GFP_KERNEL for allocation. | |
3594 | */ | |
3595 | static inline char *dup_token(const char **buf, size_t *lenp) | |
3596 | { | |
3597 | char *dup; | |
3598 | size_t len; | |
3599 | ||
3600 | len = next_token(buf); | |
4caf35f9 | 3601 | dup = kmemdup(*buf, len + 1, GFP_KERNEL); |
ea3352f4 AE |
3602 | if (!dup) |
3603 | return NULL; | |
ea3352f4 AE |
3604 | *(dup + len) = '\0'; |
3605 | *buf += len; | |
3606 | ||
3607 | if (lenp) | |
3608 | *lenp = len; | |
3609 | ||
3610 | return dup; | |
3611 | } | |
3612 | ||
a725f65e | 3613 | /* |
859c31df AE |
3614 | * Parse the options provided for an "rbd add" (i.e., rbd image |
3615 | * mapping) request. These arrive via a write to /sys/bus/rbd/add, | |
3616 | * and the data written is passed here via a NUL-terminated buffer. | |
3617 | * Returns 0 if successful or an error code otherwise. | |
d22f76e7 | 3618 | * |
859c31df AE |
3619 | * The information extracted from these options is recorded in |
3620 | * the other parameters which return dynamically-allocated | |
3621 | * structures: | |
3622 | * ceph_opts | |
3623 | * The address of a pointer that will refer to a ceph options | |
3624 | * structure. Caller must release the returned pointer using | |
3625 | * ceph_destroy_options() when it is no longer needed. | |
3626 | * rbd_opts | |
3627 | * Address of an rbd options pointer. Fully initialized by | |
3628 | * this function; caller must release with kfree(). | |
3629 | * spec | |
3630 | * Address of an rbd image specification pointer. Fully | |
3631 | * initialized by this function based on parsed options. | |
3632 | * Caller must release with rbd_spec_put(). | |
3633 | * | |
3634 | * The options passed take this form: | |
3635 | * <mon_addrs> <options> <pool_name> <image_name> [<snap_id>] | |
3636 | * where: | |
3637 | * <mon_addrs> | |
3638 | * A comma-separated list of one or more monitor addresses. | |
3639 | * A monitor address is an ip address, optionally followed | |
3640 | * by a port number (separated by a colon). | |
3641 | * I.e.: ip1[:port1][,ip2[:port2]...] | |
3642 | * <options> | |
3643 | * A comma-separated list of ceph and/or rbd options. | |
3644 | * <pool_name> | |
3645 | * The name of the rados pool containing the rbd image. | |
3646 | * <image_name> | |
3647 | * The name of the image in that pool to map. | |
3648 | * <snap_id> | |
3649 | * An optional snapshot id. If provided, the mapping will | |
3650 | * present data from the image at the time that snapshot was | |
3651 | * created. The image head is used if no snapshot id is | |
3652 | * provided. Snapshot mappings are always read-only. | |
a725f65e | 3653 | */ |
859c31df | 3654 | static int rbd_add_parse_args(const char *buf, |
dc79b113 | 3655 | struct ceph_options **ceph_opts, |
859c31df AE |
3656 | struct rbd_options **opts, |
3657 | struct rbd_spec **rbd_spec) | |
e28fff26 | 3658 | { |
d22f76e7 | 3659 | size_t len; |
859c31df | 3660 | char *options; |
0ddebc0c AE |
3661 | const char *mon_addrs; |
3662 | size_t mon_addrs_size; | |
859c31df | 3663 | struct rbd_spec *spec = NULL; |
4e9afeba | 3664 | struct rbd_options *rbd_opts = NULL; |
859c31df | 3665 | struct ceph_options *copts; |
dc79b113 | 3666 | int ret; |
e28fff26 AE |
3667 | |
3668 | /* The first four tokens are required */ | |
3669 | ||
7ef3214a | 3670 | len = next_token(&buf); |
4fb5d671 AE |
3671 | if (!len) { |
3672 | rbd_warn(NULL, "no monitor address(es) provided"); | |
3673 | return -EINVAL; | |
3674 | } | |
0ddebc0c | 3675 | mon_addrs = buf; |
f28e565a | 3676 | mon_addrs_size = len + 1; |
7ef3214a | 3677 | buf += len; |
a725f65e | 3678 | |
dc79b113 | 3679 | ret = -EINVAL; |
f28e565a AE |
3680 | options = dup_token(&buf, NULL); |
3681 | if (!options) | |
dc79b113 | 3682 | return -ENOMEM; |
4fb5d671 AE |
3683 | if (!*options) { |
3684 | rbd_warn(NULL, "no options provided"); | |
3685 | goto out_err; | |
3686 | } | |
e28fff26 | 3687 | |
859c31df AE |
3688 | spec = rbd_spec_alloc(); |
3689 | if (!spec) | |
f28e565a | 3690 | goto out_mem; |
859c31df AE |
3691 | |
3692 | spec->pool_name = dup_token(&buf, NULL); | |
3693 | if (!spec->pool_name) | |
3694 | goto out_mem; | |
4fb5d671 AE |
3695 | if (!*spec->pool_name) { |
3696 | rbd_warn(NULL, "no pool name provided"); | |
3697 | goto out_err; | |
3698 | } | |
e28fff26 | 3699 | |
69e7a02f | 3700 | spec->image_name = dup_token(&buf, NULL); |
859c31df | 3701 | if (!spec->image_name) |
f28e565a | 3702 | goto out_mem; |
4fb5d671 AE |
3703 | if (!*spec->image_name) { |
3704 | rbd_warn(NULL, "no image name provided"); | |
3705 | goto out_err; | |
3706 | } | |
d4b125e9 | 3707 | |
f28e565a AE |
3708 | /* |
3709 | * Snapshot name is optional; default is to use "-" | |
3710 | * (indicating the head/no snapshot). | |
3711 | */ | |
3feeb894 | 3712 | len = next_token(&buf); |
820a5f3e | 3713 | if (!len) { |
3feeb894 AE |
3714 | buf = RBD_SNAP_HEAD_NAME; /* No snapshot supplied */ |
3715 | len = sizeof (RBD_SNAP_HEAD_NAME) - 1; | |
f28e565a | 3716 | } else if (len > RBD_MAX_SNAP_NAME_LEN) { |
dc79b113 | 3717 | ret = -ENAMETOOLONG; |
f28e565a | 3718 | goto out_err; |
849b4260 | 3719 | } |
4caf35f9 | 3720 | spec->snap_name = kmemdup(buf, len + 1, GFP_KERNEL); |
859c31df | 3721 | if (!spec->snap_name) |
f28e565a | 3722 | goto out_mem; |
859c31df | 3723 | *(spec->snap_name + len) = '\0'; |
e5c35534 | 3724 | |
0ddebc0c | 3725 | /* Initialize all rbd options to the defaults */ |
e28fff26 | 3726 | |
4e9afeba AE |
3727 | rbd_opts = kzalloc(sizeof (*rbd_opts), GFP_KERNEL); |
3728 | if (!rbd_opts) | |
3729 | goto out_mem; | |
3730 | ||
3731 | rbd_opts->read_only = RBD_READ_ONLY_DEFAULT; | |
d22f76e7 | 3732 | |
859c31df | 3733 | copts = ceph_parse_options(options, mon_addrs, |
0ddebc0c | 3734 | mon_addrs + mon_addrs_size - 1, |
4e9afeba | 3735 | parse_rbd_opts_token, rbd_opts); |
859c31df AE |
3736 | if (IS_ERR(copts)) { |
3737 | ret = PTR_ERR(copts); | |
dc79b113 AE |
3738 | goto out_err; |
3739 | } | |
859c31df AE |
3740 | kfree(options); |
3741 | ||
3742 | *ceph_opts = copts; | |
4e9afeba | 3743 | *opts = rbd_opts; |
859c31df | 3744 | *rbd_spec = spec; |
0ddebc0c | 3745 | |
dc79b113 | 3746 | return 0; |
f28e565a | 3747 | out_mem: |
dc79b113 | 3748 | ret = -ENOMEM; |
d22f76e7 | 3749 | out_err: |
859c31df AE |
3750 | kfree(rbd_opts); |
3751 | rbd_spec_put(spec); | |
f28e565a | 3752 | kfree(options); |
d22f76e7 | 3753 | |
dc79b113 | 3754 | return ret; |
a725f65e AE |
3755 | } |
3756 | ||
589d30e0 AE |
3757 | /* |
3758 | * An rbd format 2 image has a unique identifier, distinct from the | |
3759 | * name given to it by the user. Internally, that identifier is | |
3760 | * what's used to specify the names of objects related to the image. | |
3761 | * | |
3762 | * A special "rbd id" object is used to map an rbd image name to its | |
3763 | * id. If that object doesn't exist, then there is no v2 rbd image | |
3764 | * with the supplied name. | |
3765 | * | |
3766 | * This function will record the given rbd_dev's image_id field if | |
3767 | * it can be determined, and in that case will return 0. If any | |
3768 | * errors occur a negative errno will be returned and the rbd_dev's | |
3769 | * image_id field will be unchanged (and should be NULL). | |
3770 | */ | |
3771 | static int rbd_dev_image_id(struct rbd_device *rbd_dev) | |
3772 | { | |
3773 | int ret; | |
3774 | size_t size; | |
3775 | char *object_name; | |
3776 | void *response; | |
3777 | void *p; | |
3778 | ||
2c0d0a10 AE |
3779 | /* |
3780 | * When probing a parent image, the image id is already | |
3781 | * known (and the image name likely is not). There's no | |
3782 | * need to fetch the image id again in this case. | |
3783 | */ | |
3784 | if (rbd_dev->spec->image_id) | |
3785 | return 0; | |
3786 | ||
589d30e0 AE |
3787 | /* |
3788 | * First, see if the format 2 image id file exists, and if | |
3789 | * so, get the image's persistent id from it. | |
3790 | */ | |
69e7a02f | 3791 | size = sizeof (RBD_ID_PREFIX) + strlen(rbd_dev->spec->image_name); |
589d30e0 AE |
3792 | object_name = kmalloc(size, GFP_NOIO); |
3793 | if (!object_name) | |
3794 | return -ENOMEM; | |
0d7dbfce | 3795 | sprintf(object_name, "%s%s", RBD_ID_PREFIX, rbd_dev->spec->image_name); |
589d30e0 AE |
3796 | dout("rbd id object name is %s\n", object_name); |
3797 | ||
3798 | /* Response will be an encoded string, which includes a length */ | |
3799 | ||
3800 | size = sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX; | |
3801 | response = kzalloc(size, GFP_NOIO); | |
3802 | if (!response) { | |
3803 | ret = -ENOMEM; | |
3804 | goto out; | |
3805 | } | |
3806 | ||
36be9a76 | 3807 | ret = rbd_obj_method_sync(rbd_dev, object_name, |
589d30e0 AE |
3808 | "rbd", "get_id", |
3809 | NULL, 0, | |
07b2391f | 3810 | response, RBD_IMAGE_ID_LEN_MAX, NULL); |
36be9a76 | 3811 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
589d30e0 AE |
3812 | if (ret < 0) |
3813 | goto out; | |
3814 | ||
3815 | p = response; | |
0d7dbfce | 3816 | rbd_dev->spec->image_id = ceph_extract_encoded_string(&p, |
589d30e0 | 3817 | p + RBD_IMAGE_ID_LEN_MAX, |
979ed480 | 3818 | NULL, GFP_NOIO); |
0d7dbfce AE |
3819 | if (IS_ERR(rbd_dev->spec->image_id)) { |
3820 | ret = PTR_ERR(rbd_dev->spec->image_id); | |
3821 | rbd_dev->spec->image_id = NULL; | |
589d30e0 | 3822 | } else { |
0d7dbfce | 3823 | dout("image_id is %s\n", rbd_dev->spec->image_id); |
589d30e0 AE |
3824 | } |
3825 | out: | |
3826 | kfree(response); | |
3827 | kfree(object_name); | |
3828 | ||
3829 | return ret; | |
3830 | } | |
3831 | ||
a30b71b9 AE |
3832 | static int rbd_dev_v1_probe(struct rbd_device *rbd_dev) |
3833 | { | |
3834 | int ret; | |
3835 | size_t size; | |
3836 | ||
3837 | /* Version 1 images have no id; empty string is used */ | |
3838 | ||
0d7dbfce AE |
3839 | rbd_dev->spec->image_id = kstrdup("", GFP_KERNEL); |
3840 | if (!rbd_dev->spec->image_id) | |
a30b71b9 | 3841 | return -ENOMEM; |
a30b71b9 AE |
3842 | |
3843 | /* Record the header object name for this rbd image. */ | |
3844 | ||
69e7a02f | 3845 | size = strlen(rbd_dev->spec->image_name) + sizeof (RBD_SUFFIX); |
a30b71b9 AE |
3846 | rbd_dev->header_name = kmalloc(size, GFP_KERNEL); |
3847 | if (!rbd_dev->header_name) { | |
3848 | ret = -ENOMEM; | |
3849 | goto out_err; | |
3850 | } | |
0d7dbfce AE |
3851 | sprintf(rbd_dev->header_name, "%s%s", |
3852 | rbd_dev->spec->image_name, RBD_SUFFIX); | |
a30b71b9 AE |
3853 | |
3854 | /* Populate rbd image metadata */ | |
3855 | ||
3856 | ret = rbd_read_header(rbd_dev, &rbd_dev->header); | |
3857 | if (ret < 0) | |
3858 | goto out_err; | |
86b00e0d AE |
3859 | |
3860 | /* Version 1 images have no parent (no layering) */ | |
3861 | ||
3862 | rbd_dev->parent_spec = NULL; | |
3863 | rbd_dev->parent_overlap = 0; | |
3864 | ||
a30b71b9 AE |
3865 | rbd_dev->image_format = 1; |
3866 | ||
3867 | dout("discovered version 1 image, header name is %s\n", | |
3868 | rbd_dev->header_name); | |
3869 | ||
3870 | return 0; | |
3871 | ||
3872 | out_err: | |
3873 | kfree(rbd_dev->header_name); | |
3874 | rbd_dev->header_name = NULL; | |
0d7dbfce AE |
3875 | kfree(rbd_dev->spec->image_id); |
3876 | rbd_dev->spec->image_id = NULL; | |
a30b71b9 AE |
3877 | |
3878 | return ret; | |
3879 | } | |
3880 | ||
3881 | static int rbd_dev_v2_probe(struct rbd_device *rbd_dev) | |
3882 | { | |
3883 | size_t size; | |
9d475de5 | 3884 | int ret; |
6e14b1a6 | 3885 | u64 ver = 0; |
a30b71b9 AE |
3886 | |
3887 | /* | |
3888 | * Image id was filled in by the caller. Record the header | |
3889 | * object name for this rbd image. | |
3890 | */ | |
979ed480 | 3891 | size = sizeof (RBD_HEADER_PREFIX) + strlen(rbd_dev->spec->image_id); |
a30b71b9 AE |
3892 | rbd_dev->header_name = kmalloc(size, GFP_KERNEL); |
3893 | if (!rbd_dev->header_name) | |
3894 | return -ENOMEM; | |
3895 | sprintf(rbd_dev->header_name, "%s%s", | |
0d7dbfce | 3896 | RBD_HEADER_PREFIX, rbd_dev->spec->image_id); |
9d475de5 AE |
3897 | |
3898 | /* Get the size and object order for the image */ | |
3899 | ||
3900 | ret = rbd_dev_v2_image_size(rbd_dev); | |
1e130199 AE |
3901 | if (ret < 0) |
3902 | goto out_err; | |
3903 | ||
3904 | /* Get the object prefix (a.k.a. block_name) for the image */ | |
3905 | ||
3906 | ret = rbd_dev_v2_object_prefix(rbd_dev); | |
b1b5402a AE |
3907 | if (ret < 0) |
3908 | goto out_err; | |
3909 | ||
d889140c | 3910 | /* Get the and check features for the image */ |
b1b5402a AE |
3911 | |
3912 | ret = rbd_dev_v2_features(rbd_dev); | |
9d475de5 AE |
3913 | if (ret < 0) |
3914 | goto out_err; | |
35d489f9 | 3915 | |
86b00e0d AE |
3916 | /* If the image supports layering, get the parent info */ |
3917 | ||
3918 | if (rbd_dev->header.features & RBD_FEATURE_LAYERING) { | |
3919 | ret = rbd_dev_v2_parent_info(rbd_dev); | |
3920 | if (ret < 0) | |
3921 | goto out_err; | |
3922 | } | |
3923 | ||
6e14b1a6 AE |
3924 | /* crypto and compression type aren't (yet) supported for v2 images */ |
3925 | ||
3926 | rbd_dev->header.crypt_type = 0; | |
3927 | rbd_dev->header.comp_type = 0; | |
35d489f9 | 3928 | |
6e14b1a6 AE |
3929 | /* Get the snapshot context, plus the header version */ |
3930 | ||
3931 | ret = rbd_dev_v2_snap_context(rbd_dev, &ver); | |
35d489f9 AE |
3932 | if (ret) |
3933 | goto out_err; | |
6e14b1a6 AE |
3934 | rbd_dev->header.obj_version = ver; |
3935 | ||
a30b71b9 AE |
3936 | rbd_dev->image_format = 2; |
3937 | ||
3938 | dout("discovered version 2 image, header name is %s\n", | |
3939 | rbd_dev->header_name); | |
3940 | ||
35152979 | 3941 | return 0; |
9d475de5 | 3942 | out_err: |
86b00e0d AE |
3943 | rbd_dev->parent_overlap = 0; |
3944 | rbd_spec_put(rbd_dev->parent_spec); | |
3945 | rbd_dev->parent_spec = NULL; | |
9d475de5 AE |
3946 | kfree(rbd_dev->header_name); |
3947 | rbd_dev->header_name = NULL; | |
1e130199 AE |
3948 | kfree(rbd_dev->header.object_prefix); |
3949 | rbd_dev->header.object_prefix = NULL; | |
9d475de5 AE |
3950 | |
3951 | return ret; | |
a30b71b9 AE |
3952 | } |
3953 | ||
83a06263 AE |
3954 | static int rbd_dev_probe_finish(struct rbd_device *rbd_dev) |
3955 | { | |
3956 | int ret; | |
3957 | ||
3958 | /* no need to lock here, as rbd_dev is not registered yet */ | |
3959 | ret = rbd_dev_snaps_update(rbd_dev); | |
3960 | if (ret) | |
3961 | return ret; | |
3962 | ||
9e15b77d AE |
3963 | ret = rbd_dev_probe_update_spec(rbd_dev); |
3964 | if (ret) | |
3965 | goto err_out_snaps; | |
3966 | ||
83a06263 AE |
3967 | ret = rbd_dev_set_mapping(rbd_dev); |
3968 | if (ret) | |
3969 | goto err_out_snaps; | |
3970 | ||
3971 | /* generate unique id: find highest unique id, add one */ | |
3972 | rbd_dev_id_get(rbd_dev); | |
3973 | ||
3974 | /* Fill in the device name, now that we have its id. */ | |
3975 | BUILD_BUG_ON(DEV_NAME_LEN | |
3976 | < sizeof (RBD_DRV_NAME) + MAX_INT_FORMAT_WIDTH); | |
3977 | sprintf(rbd_dev->name, "%s%d", RBD_DRV_NAME, rbd_dev->dev_id); | |
3978 | ||
3979 | /* Get our block major device number. */ | |
3980 | ||
3981 | ret = register_blkdev(0, rbd_dev->name); | |
3982 | if (ret < 0) | |
3983 | goto err_out_id; | |
3984 | rbd_dev->major = ret; | |
3985 | ||
3986 | /* Set up the blkdev mapping. */ | |
3987 | ||
3988 | ret = rbd_init_disk(rbd_dev); | |
3989 | if (ret) | |
3990 | goto err_out_blkdev; | |
3991 | ||
3992 | ret = rbd_bus_add_dev(rbd_dev); | |
3993 | if (ret) | |
3994 | goto err_out_disk; | |
3995 | ||
3996 | /* | |
3997 | * At this point cleanup in the event of an error is the job | |
3998 | * of the sysfs code (initiated by rbd_bus_del_dev()). | |
3999 | */ | |
4000 | down_write(&rbd_dev->header_rwsem); | |
4001 | ret = rbd_dev_snaps_register(rbd_dev); | |
4002 | up_write(&rbd_dev->header_rwsem); | |
4003 | if (ret) | |
4004 | goto err_out_bus; | |
4005 | ||
9969ebc5 | 4006 | ret = rbd_dev_header_watch_sync(rbd_dev, 1); |
83a06263 AE |
4007 | if (ret) |
4008 | goto err_out_bus; | |
4009 | ||
4010 | /* Everything's ready. Announce the disk to the world. */ | |
4011 | ||
4012 | add_disk(rbd_dev->disk); | |
4013 | ||
4014 | pr_info("%s: added with size 0x%llx\n", rbd_dev->disk->disk_name, | |
4015 | (unsigned long long) rbd_dev->mapping.size); | |
4016 | ||
4017 | return ret; | |
4018 | err_out_bus: | |
4019 | /* this will also clean up rest of rbd_dev stuff */ | |
4020 | ||
4021 | rbd_bus_del_dev(rbd_dev); | |
4022 | ||
4023 | return ret; | |
4024 | err_out_disk: | |
4025 | rbd_free_disk(rbd_dev); | |
4026 | err_out_blkdev: | |
4027 | unregister_blkdev(rbd_dev->major, rbd_dev->name); | |
4028 | err_out_id: | |
4029 | rbd_dev_id_put(rbd_dev); | |
4030 | err_out_snaps: | |
4031 | rbd_remove_all_snaps(rbd_dev); | |
4032 | ||
4033 | return ret; | |
4034 | } | |
4035 | ||
a30b71b9 AE |
4036 | /* |
4037 | * Probe for the existence of the header object for the given rbd | |
4038 | * device. For format 2 images this includes determining the image | |
4039 | * id. | |
4040 | */ | |
4041 | static int rbd_dev_probe(struct rbd_device *rbd_dev) | |
4042 | { | |
4043 | int ret; | |
4044 | ||
4045 | /* | |
4046 | * Get the id from the image id object. If it's not a | |
4047 | * format 2 image, we'll get ENOENT back, and we'll assume | |
4048 | * it's a format 1 image. | |
4049 | */ | |
4050 | ret = rbd_dev_image_id(rbd_dev); | |
4051 | if (ret) | |
4052 | ret = rbd_dev_v1_probe(rbd_dev); | |
4053 | else | |
4054 | ret = rbd_dev_v2_probe(rbd_dev); | |
83a06263 | 4055 | if (ret) { |
a30b71b9 AE |
4056 | dout("probe failed, returning %d\n", ret); |
4057 | ||
83a06263 AE |
4058 | return ret; |
4059 | } | |
4060 | ||
4061 | ret = rbd_dev_probe_finish(rbd_dev); | |
4062 | if (ret) | |
4063 | rbd_header_free(&rbd_dev->header); | |
4064 | ||
a30b71b9 AE |
4065 | return ret; |
4066 | } | |
4067 | ||
59c2be1e YS |
4068 | static ssize_t rbd_add(struct bus_type *bus, |
4069 | const char *buf, | |
4070 | size_t count) | |
602adf40 | 4071 | { |
cb8627c7 | 4072 | struct rbd_device *rbd_dev = NULL; |
dc79b113 | 4073 | struct ceph_options *ceph_opts = NULL; |
4e9afeba | 4074 | struct rbd_options *rbd_opts = NULL; |
859c31df | 4075 | struct rbd_spec *spec = NULL; |
9d3997fd | 4076 | struct rbd_client *rbdc; |
27cc2594 AE |
4077 | struct ceph_osd_client *osdc; |
4078 | int rc = -ENOMEM; | |
602adf40 YS |
4079 | |
4080 | if (!try_module_get(THIS_MODULE)) | |
4081 | return -ENODEV; | |
4082 | ||
602adf40 | 4083 | /* parse add command */ |
859c31df | 4084 | rc = rbd_add_parse_args(buf, &ceph_opts, &rbd_opts, &spec); |
dc79b113 | 4085 | if (rc < 0) |
bd4ba655 | 4086 | goto err_out_module; |
78cea76e | 4087 | |
9d3997fd AE |
4088 | rbdc = rbd_get_client(ceph_opts); |
4089 | if (IS_ERR(rbdc)) { | |
4090 | rc = PTR_ERR(rbdc); | |
0ddebc0c | 4091 | goto err_out_args; |
9d3997fd | 4092 | } |
c53d5893 | 4093 | ceph_opts = NULL; /* rbd_dev client now owns this */ |
602adf40 | 4094 | |
602adf40 | 4095 | /* pick the pool */ |
9d3997fd | 4096 | osdc = &rbdc->client->osdc; |
859c31df | 4097 | rc = ceph_pg_poolid_by_name(osdc->osdmap, spec->pool_name); |
602adf40 YS |
4098 | if (rc < 0) |
4099 | goto err_out_client; | |
859c31df AE |
4100 | spec->pool_id = (u64) rc; |
4101 | ||
0903e875 AE |
4102 | /* The ceph file layout needs to fit pool id in 32 bits */ |
4103 | ||
4104 | if (WARN_ON(spec->pool_id > (u64) U32_MAX)) { | |
4105 | rc = -EIO; | |
4106 | goto err_out_client; | |
4107 | } | |
4108 | ||
c53d5893 | 4109 | rbd_dev = rbd_dev_create(rbdc, spec); |
bd4ba655 AE |
4110 | if (!rbd_dev) |
4111 | goto err_out_client; | |
c53d5893 AE |
4112 | rbdc = NULL; /* rbd_dev now owns this */ |
4113 | spec = NULL; /* rbd_dev now owns this */ | |
602adf40 | 4114 | |
bd4ba655 | 4115 | rbd_dev->mapping.read_only = rbd_opts->read_only; |
c53d5893 AE |
4116 | kfree(rbd_opts); |
4117 | rbd_opts = NULL; /* done with this */ | |
bd4ba655 | 4118 | |
a30b71b9 AE |
4119 | rc = rbd_dev_probe(rbd_dev); |
4120 | if (rc < 0) | |
c53d5893 | 4121 | goto err_out_rbd_dev; |
05fd6f6f | 4122 | |
602adf40 | 4123 | return count; |
c53d5893 AE |
4124 | err_out_rbd_dev: |
4125 | rbd_dev_destroy(rbd_dev); | |
bd4ba655 | 4126 | err_out_client: |
9d3997fd | 4127 | rbd_put_client(rbdc); |
0ddebc0c | 4128 | err_out_args: |
78cea76e AE |
4129 | if (ceph_opts) |
4130 | ceph_destroy_options(ceph_opts); | |
4e9afeba | 4131 | kfree(rbd_opts); |
859c31df | 4132 | rbd_spec_put(spec); |
bd4ba655 AE |
4133 | err_out_module: |
4134 | module_put(THIS_MODULE); | |
27cc2594 | 4135 | |
602adf40 | 4136 | dout("Error adding device %s\n", buf); |
27cc2594 AE |
4137 | |
4138 | return (ssize_t) rc; | |
602adf40 YS |
4139 | } |
4140 | ||
de71a297 | 4141 | static struct rbd_device *__rbd_get_dev(unsigned long dev_id) |
602adf40 YS |
4142 | { |
4143 | struct list_head *tmp; | |
4144 | struct rbd_device *rbd_dev; | |
4145 | ||
e124a82f | 4146 | spin_lock(&rbd_dev_list_lock); |
602adf40 YS |
4147 | list_for_each(tmp, &rbd_dev_list) { |
4148 | rbd_dev = list_entry(tmp, struct rbd_device, node); | |
de71a297 | 4149 | if (rbd_dev->dev_id == dev_id) { |
e124a82f | 4150 | spin_unlock(&rbd_dev_list_lock); |
602adf40 | 4151 | return rbd_dev; |
e124a82f | 4152 | } |
602adf40 | 4153 | } |
e124a82f | 4154 | spin_unlock(&rbd_dev_list_lock); |
602adf40 YS |
4155 | return NULL; |
4156 | } | |
4157 | ||
dfc5606d | 4158 | static void rbd_dev_release(struct device *dev) |
602adf40 | 4159 | { |
593a9e7b | 4160 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
602adf40 | 4161 | |
59c2be1e | 4162 | if (rbd_dev->watch_event) |
9969ebc5 | 4163 | rbd_dev_header_watch_sync(rbd_dev, 0); |
602adf40 YS |
4164 | |
4165 | /* clean up and free blkdev */ | |
4166 | rbd_free_disk(rbd_dev); | |
4167 | unregister_blkdev(rbd_dev->major, rbd_dev->name); | |
32eec68d | 4168 | |
2ac4e75d AE |
4169 | /* release allocated disk header fields */ |
4170 | rbd_header_free(&rbd_dev->header); | |
4171 | ||
32eec68d | 4172 | /* done with the id, and with the rbd_dev */ |
e2839308 | 4173 | rbd_dev_id_put(rbd_dev); |
c53d5893 AE |
4174 | rbd_assert(rbd_dev->rbd_client != NULL); |
4175 | rbd_dev_destroy(rbd_dev); | |
602adf40 YS |
4176 | |
4177 | /* release module ref */ | |
4178 | module_put(THIS_MODULE); | |
602adf40 YS |
4179 | } |
4180 | ||
dfc5606d YS |
4181 | static ssize_t rbd_remove(struct bus_type *bus, |
4182 | const char *buf, | |
4183 | size_t count) | |
602adf40 YS |
4184 | { |
4185 | struct rbd_device *rbd_dev = NULL; | |
4186 | int target_id, rc; | |
4187 | unsigned long ul; | |
4188 | int ret = count; | |
4189 | ||
4190 | rc = strict_strtoul(buf, 10, &ul); | |
4191 | if (rc) | |
4192 | return rc; | |
4193 | ||
4194 | /* convert to int; abort if we lost anything in the conversion */ | |
4195 | target_id = (int) ul; | |
4196 | if (target_id != ul) | |
4197 | return -EINVAL; | |
4198 | ||
4199 | mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING); | |
4200 | ||
4201 | rbd_dev = __rbd_get_dev(target_id); | |
4202 | if (!rbd_dev) { | |
4203 | ret = -ENOENT; | |
4204 | goto done; | |
42382b70 AE |
4205 | } |
4206 | ||
a14ea269 | 4207 | spin_lock_irq(&rbd_dev->lock); |
b82d167b | 4208 | if (rbd_dev->open_count) |
42382b70 | 4209 | ret = -EBUSY; |
b82d167b AE |
4210 | else |
4211 | set_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags); | |
a14ea269 | 4212 | spin_unlock_irq(&rbd_dev->lock); |
b82d167b | 4213 | if (ret < 0) |
42382b70 | 4214 | goto done; |
602adf40 | 4215 | |
41f38c2b | 4216 | rbd_remove_all_snaps(rbd_dev); |
dfc5606d | 4217 | rbd_bus_del_dev(rbd_dev); |
602adf40 YS |
4218 | |
4219 | done: | |
4220 | mutex_unlock(&ctl_mutex); | |
aafb230e | 4221 | |
602adf40 YS |
4222 | return ret; |
4223 | } | |
4224 | ||
602adf40 YS |
4225 | /* |
4226 | * create control files in sysfs | |
dfc5606d | 4227 | * /sys/bus/rbd/... |
602adf40 YS |
4228 | */ |
4229 | static int rbd_sysfs_init(void) | |
4230 | { | |
dfc5606d | 4231 | int ret; |
602adf40 | 4232 | |
fed4c143 | 4233 | ret = device_register(&rbd_root_dev); |
21079786 | 4234 | if (ret < 0) |
dfc5606d | 4235 | return ret; |
602adf40 | 4236 | |
fed4c143 AE |
4237 | ret = bus_register(&rbd_bus_type); |
4238 | if (ret < 0) | |
4239 | device_unregister(&rbd_root_dev); | |
602adf40 | 4240 | |
602adf40 YS |
4241 | return ret; |
4242 | } | |
4243 | ||
4244 | static void rbd_sysfs_cleanup(void) | |
4245 | { | |
dfc5606d | 4246 | bus_unregister(&rbd_bus_type); |
fed4c143 | 4247 | device_unregister(&rbd_root_dev); |
602adf40 YS |
4248 | } |
4249 | ||
cc344fa1 | 4250 | static int __init rbd_init(void) |
602adf40 YS |
4251 | { |
4252 | int rc; | |
4253 | ||
1e32d34c AE |
4254 | if (!libceph_compatible(NULL)) { |
4255 | rbd_warn(NULL, "libceph incompatibility (quitting)"); | |
4256 | ||
4257 | return -EINVAL; | |
4258 | } | |
602adf40 YS |
4259 | rc = rbd_sysfs_init(); |
4260 | if (rc) | |
4261 | return rc; | |
f0f8cef5 | 4262 | pr_info("loaded " RBD_DRV_NAME_LONG "\n"); |
602adf40 YS |
4263 | return 0; |
4264 | } | |
4265 | ||
cc344fa1 | 4266 | static void __exit rbd_exit(void) |
602adf40 YS |
4267 | { |
4268 | rbd_sysfs_cleanup(); | |
4269 | } | |
4270 | ||
4271 | module_init(rbd_init); | |
4272 | module_exit(rbd_exit); | |
4273 | ||
4274 | MODULE_AUTHOR("Sage Weil <sage@newdream.net>"); | |
4275 | MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>"); | |
4276 | MODULE_DESCRIPTION("rados block device"); | |
4277 | ||
4278 | /* following authorship retained from original osdblk.c */ | |
4279 | MODULE_AUTHOR("Jeff Garzik <jeff@garzik.org>"); | |
4280 | ||
4281 | MODULE_LICENSE("GPL"); |