]>
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 | ||
f0f8cef5 AE |
55 | #define RBD_DRV_NAME "rbd" |
56 | #define RBD_DRV_NAME_LONG "rbd (rados block device)" | |
602adf40 YS |
57 | |
58 | #define RBD_MINORS_PER_MAJOR 256 /* max minors per blkdev */ | |
59 | ||
d4b125e9 AE |
60 | #define RBD_SNAP_DEV_NAME_PREFIX "snap_" |
61 | #define RBD_MAX_SNAP_NAME_LEN \ | |
62 | (NAME_MAX - (sizeof (RBD_SNAP_DEV_NAME_PREFIX) - 1)) | |
63 | ||
35d489f9 | 64 | #define RBD_MAX_SNAP_COUNT 510 /* allows max snapc to fit in 4KB */ |
602adf40 YS |
65 | |
66 | #define RBD_SNAP_HEAD_NAME "-" | |
67 | ||
9e15b77d AE |
68 | /* This allows a single page to hold an image name sent by OSD */ |
69 | #define RBD_IMAGE_NAME_LEN_MAX (PAGE_SIZE - sizeof (__le32) - 1) | |
1e130199 | 70 | #define RBD_IMAGE_ID_LEN_MAX 64 |
9e15b77d | 71 | |
1e130199 | 72 | #define RBD_OBJ_PREFIX_LEN_MAX 64 |
589d30e0 | 73 | |
d889140c AE |
74 | /* Feature bits */ |
75 | ||
76 | #define RBD_FEATURE_LAYERING 1 | |
77 | ||
78 | /* Features supported by this (client software) implementation. */ | |
79 | ||
80 | #define RBD_FEATURES_ALL (0) | |
81 | ||
81a89793 AE |
82 | /* |
83 | * An RBD device name will be "rbd#", where the "rbd" comes from | |
84 | * RBD_DRV_NAME above, and # is a unique integer identifier. | |
85 | * MAX_INT_FORMAT_WIDTH is used in ensuring DEV_NAME_LEN is big | |
86 | * enough to hold all possible device names. | |
87 | */ | |
602adf40 | 88 | #define DEV_NAME_LEN 32 |
81a89793 | 89 | #define MAX_INT_FORMAT_WIDTH ((5 * sizeof (int)) / 2 + 1) |
602adf40 YS |
90 | |
91 | /* | |
92 | * block device image metadata (in-memory version) | |
93 | */ | |
94 | struct rbd_image_header { | |
f84344f3 | 95 | /* These four fields never change for a given rbd image */ |
849b4260 | 96 | char *object_prefix; |
34b13184 | 97 | u64 features; |
602adf40 YS |
98 | __u8 obj_order; |
99 | __u8 crypt_type; | |
100 | __u8 comp_type; | |
602adf40 | 101 | |
f84344f3 AE |
102 | /* The remaining fields need to be updated occasionally */ |
103 | u64 image_size; | |
104 | struct ceph_snap_context *snapc; | |
602adf40 YS |
105 | char *snap_names; |
106 | u64 *snap_sizes; | |
59c2be1e YS |
107 | |
108 | u64 obj_version; | |
109 | }; | |
110 | ||
0d7dbfce AE |
111 | /* |
112 | * An rbd image specification. | |
113 | * | |
114 | * The tuple (pool_id, image_id, snap_id) is sufficient to uniquely | |
c66c6e0c AE |
115 | * identify an image. Each rbd_dev structure includes a pointer to |
116 | * an rbd_spec structure that encapsulates this identity. | |
117 | * | |
118 | * Each of the id's in an rbd_spec has an associated name. For a | |
119 | * user-mapped image, the names are supplied and the id's associated | |
120 | * with them are looked up. For a layered image, a parent image is | |
121 | * defined by the tuple, and the names are looked up. | |
122 | * | |
123 | * An rbd_dev structure contains a parent_spec pointer which is | |
124 | * non-null if the image it represents is a child in a layered | |
125 | * image. This pointer will refer to the rbd_spec structure used | |
126 | * by the parent rbd_dev for its own identity (i.e., the structure | |
127 | * is shared between the parent and child). | |
128 | * | |
129 | * Since these structures are populated once, during the discovery | |
130 | * phase of image construction, they are effectively immutable so | |
131 | * we make no effort to synchronize access to them. | |
132 | * | |
133 | * Note that code herein does not assume the image name is known (it | |
134 | * could be a null pointer). | |
0d7dbfce AE |
135 | */ |
136 | struct rbd_spec { | |
137 | u64 pool_id; | |
138 | char *pool_name; | |
139 | ||
140 | char *image_id; | |
0d7dbfce | 141 | char *image_name; |
0d7dbfce AE |
142 | |
143 | u64 snap_id; | |
144 | char *snap_name; | |
145 | ||
146 | struct kref kref; | |
147 | }; | |
148 | ||
602adf40 | 149 | /* |
f0f8cef5 | 150 | * an instance of the client. multiple devices may share an rbd client. |
602adf40 YS |
151 | */ |
152 | struct rbd_client { | |
153 | struct ceph_client *client; | |
154 | struct kref kref; | |
155 | struct list_head node; | |
156 | }; | |
157 | ||
bf0d5f50 AE |
158 | struct rbd_img_request; |
159 | typedef void (*rbd_img_callback_t)(struct rbd_img_request *); | |
160 | ||
161 | #define BAD_WHICH U32_MAX /* Good which or bad which, which? */ | |
162 | ||
163 | struct rbd_obj_request; | |
164 | typedef void (*rbd_obj_callback_t)(struct rbd_obj_request *); | |
165 | ||
9969ebc5 AE |
166 | enum obj_request_type { |
167 | OBJ_REQUEST_NODATA, OBJ_REQUEST_BIO, OBJ_REQUEST_PAGES | |
168 | }; | |
bf0d5f50 AE |
169 | |
170 | struct rbd_obj_request { | |
171 | const char *object_name; | |
172 | u64 offset; /* object start byte */ | |
173 | u64 length; /* bytes from offset */ | |
174 | ||
175 | struct rbd_img_request *img_request; | |
176 | struct list_head links; /* img_request->obj_requests */ | |
177 | u32 which; /* posn image request list */ | |
178 | ||
179 | enum obj_request_type type; | |
788e2df3 AE |
180 | union { |
181 | struct bio *bio_list; | |
182 | struct { | |
183 | struct page **pages; | |
184 | u32 page_count; | |
185 | }; | |
186 | }; | |
bf0d5f50 AE |
187 | |
188 | struct ceph_osd_request *osd_req; | |
189 | ||
190 | u64 xferred; /* bytes transferred */ | |
191 | u64 version; | |
1b83bef2 | 192 | int result; |
bf0d5f50 AE |
193 | atomic_t done; |
194 | ||
195 | rbd_obj_callback_t callback; | |
788e2df3 | 196 | struct completion completion; |
bf0d5f50 AE |
197 | |
198 | struct kref kref; | |
199 | }; | |
200 | ||
201 | struct rbd_img_request { | |
202 | struct request *rq; | |
203 | struct rbd_device *rbd_dev; | |
204 | u64 offset; /* starting image byte offset */ | |
205 | u64 length; /* byte count from offset */ | |
206 | bool write_request; /* false for read */ | |
207 | union { | |
208 | struct ceph_snap_context *snapc; /* for writes */ | |
209 | u64 snap_id; /* for reads */ | |
210 | }; | |
211 | spinlock_t completion_lock;/* protects next_completion */ | |
212 | u32 next_completion; | |
213 | rbd_img_callback_t callback; | |
214 | ||
215 | u32 obj_request_count; | |
216 | struct list_head obj_requests; /* rbd_obj_request structs */ | |
217 | ||
218 | struct kref kref; | |
219 | }; | |
220 | ||
221 | #define for_each_obj_request(ireq, oreq) \ | |
ef06f4d3 | 222 | list_for_each_entry(oreq, &(ireq)->obj_requests, links) |
bf0d5f50 | 223 | #define for_each_obj_request_from(ireq, oreq) \ |
ef06f4d3 | 224 | list_for_each_entry_from(oreq, &(ireq)->obj_requests, links) |
bf0d5f50 | 225 | #define for_each_obj_request_safe(ireq, oreq, n) \ |
ef06f4d3 | 226 | list_for_each_entry_safe_reverse(oreq, n, &(ireq)->obj_requests, links) |
bf0d5f50 | 227 | |
dfc5606d YS |
228 | struct rbd_snap { |
229 | struct device dev; | |
230 | const char *name; | |
3591538f | 231 | u64 size; |
dfc5606d YS |
232 | struct list_head node; |
233 | u64 id; | |
34b13184 | 234 | u64 features; |
dfc5606d YS |
235 | }; |
236 | ||
f84344f3 | 237 | struct rbd_mapping { |
99c1f08f | 238 | u64 size; |
34b13184 | 239 | u64 features; |
f84344f3 AE |
240 | bool read_only; |
241 | }; | |
242 | ||
602adf40 YS |
243 | /* |
244 | * a single device | |
245 | */ | |
246 | struct rbd_device { | |
de71a297 | 247 | int dev_id; /* blkdev unique id */ |
602adf40 YS |
248 | |
249 | int major; /* blkdev assigned major */ | |
250 | struct gendisk *disk; /* blkdev's gendisk and rq */ | |
602adf40 | 251 | |
a30b71b9 | 252 | u32 image_format; /* Either 1 or 2 */ |
602adf40 YS |
253 | struct rbd_client *rbd_client; |
254 | ||
255 | char name[DEV_NAME_LEN]; /* blkdev name, e.g. rbd3 */ | |
256 | ||
b82d167b | 257 | spinlock_t lock; /* queue, flags, open_count */ |
602adf40 YS |
258 | |
259 | struct rbd_image_header header; | |
b82d167b | 260 | unsigned long flags; /* possibly lock protected */ |
0d7dbfce | 261 | struct rbd_spec *spec; |
602adf40 | 262 | |
0d7dbfce | 263 | char *header_name; |
971f839a | 264 | |
0903e875 AE |
265 | struct ceph_file_layout layout; |
266 | ||
59c2be1e | 267 | struct ceph_osd_event *watch_event; |
975241af | 268 | struct rbd_obj_request *watch_request; |
59c2be1e | 269 | |
86b00e0d AE |
270 | struct rbd_spec *parent_spec; |
271 | u64 parent_overlap; | |
272 | ||
c666601a JD |
273 | /* protects updating the header */ |
274 | struct rw_semaphore header_rwsem; | |
f84344f3 AE |
275 | |
276 | struct rbd_mapping mapping; | |
602adf40 YS |
277 | |
278 | struct list_head node; | |
dfc5606d YS |
279 | |
280 | /* list of snapshots */ | |
281 | struct list_head snaps; | |
282 | ||
283 | /* sysfs related */ | |
284 | struct device dev; | |
b82d167b | 285 | unsigned long open_count; /* protected by lock */ |
dfc5606d YS |
286 | }; |
287 | ||
b82d167b AE |
288 | /* |
289 | * Flag bits for rbd_dev->flags. If atomicity is required, | |
290 | * rbd_dev->lock is used to protect access. | |
291 | * | |
292 | * Currently, only the "removing" flag (which is coupled with the | |
293 | * "open_count" field) requires atomic access. | |
294 | */ | |
6d292906 AE |
295 | enum rbd_dev_flags { |
296 | RBD_DEV_FLAG_EXISTS, /* mapped snapshot has not been deleted */ | |
b82d167b | 297 | RBD_DEV_FLAG_REMOVING, /* this mapping is being removed */ |
6d292906 AE |
298 | }; |
299 | ||
602adf40 | 300 | static DEFINE_MUTEX(ctl_mutex); /* Serialize open/close/setup/teardown */ |
e124a82f | 301 | |
602adf40 | 302 | static LIST_HEAD(rbd_dev_list); /* devices */ |
e124a82f AE |
303 | static DEFINE_SPINLOCK(rbd_dev_list_lock); |
304 | ||
432b8587 AE |
305 | static LIST_HEAD(rbd_client_list); /* clients */ |
306 | static DEFINE_SPINLOCK(rbd_client_list_lock); | |
602adf40 | 307 | |
304f6808 AE |
308 | static int rbd_dev_snaps_update(struct rbd_device *rbd_dev); |
309 | static int rbd_dev_snaps_register(struct rbd_device *rbd_dev); | |
310 | ||
dfc5606d | 311 | static void rbd_dev_release(struct device *dev); |
41f38c2b | 312 | static void rbd_remove_snap_dev(struct rbd_snap *snap); |
dfc5606d | 313 | |
f0f8cef5 AE |
314 | static ssize_t rbd_add(struct bus_type *bus, const char *buf, |
315 | size_t count); | |
316 | static ssize_t rbd_remove(struct bus_type *bus, const char *buf, | |
317 | size_t count); | |
318 | ||
319 | static struct bus_attribute rbd_bus_attrs[] = { | |
320 | __ATTR(add, S_IWUSR, NULL, rbd_add), | |
321 | __ATTR(remove, S_IWUSR, NULL, rbd_remove), | |
322 | __ATTR_NULL | |
323 | }; | |
324 | ||
325 | static struct bus_type rbd_bus_type = { | |
326 | .name = "rbd", | |
327 | .bus_attrs = rbd_bus_attrs, | |
328 | }; | |
329 | ||
330 | static void rbd_root_dev_release(struct device *dev) | |
331 | { | |
332 | } | |
333 | ||
334 | static struct device rbd_root_dev = { | |
335 | .init_name = "rbd", | |
336 | .release = rbd_root_dev_release, | |
337 | }; | |
338 | ||
06ecc6cb AE |
339 | static __printf(2, 3) |
340 | void rbd_warn(struct rbd_device *rbd_dev, const char *fmt, ...) | |
341 | { | |
342 | struct va_format vaf; | |
343 | va_list args; | |
344 | ||
345 | va_start(args, fmt); | |
346 | vaf.fmt = fmt; | |
347 | vaf.va = &args; | |
348 | ||
349 | if (!rbd_dev) | |
350 | printk(KERN_WARNING "%s: %pV\n", RBD_DRV_NAME, &vaf); | |
351 | else if (rbd_dev->disk) | |
352 | printk(KERN_WARNING "%s: %s: %pV\n", | |
353 | RBD_DRV_NAME, rbd_dev->disk->disk_name, &vaf); | |
354 | else if (rbd_dev->spec && rbd_dev->spec->image_name) | |
355 | printk(KERN_WARNING "%s: image %s: %pV\n", | |
356 | RBD_DRV_NAME, rbd_dev->spec->image_name, &vaf); | |
357 | else if (rbd_dev->spec && rbd_dev->spec->image_id) | |
358 | printk(KERN_WARNING "%s: id %s: %pV\n", | |
359 | RBD_DRV_NAME, rbd_dev->spec->image_id, &vaf); | |
360 | else /* punt */ | |
361 | printk(KERN_WARNING "%s: rbd_dev %p: %pV\n", | |
362 | RBD_DRV_NAME, rbd_dev, &vaf); | |
363 | va_end(args); | |
364 | } | |
365 | ||
aafb230e AE |
366 | #ifdef RBD_DEBUG |
367 | #define rbd_assert(expr) \ | |
368 | if (unlikely(!(expr))) { \ | |
369 | printk(KERN_ERR "\nAssertion failure in %s() " \ | |
370 | "at line %d:\n\n" \ | |
371 | "\trbd_assert(%s);\n\n", \ | |
372 | __func__, __LINE__, #expr); \ | |
373 | BUG(); \ | |
374 | } | |
375 | #else /* !RBD_DEBUG */ | |
376 | # define rbd_assert(expr) ((void) 0) | |
377 | #endif /* !RBD_DEBUG */ | |
dfc5606d | 378 | |
117973fb AE |
379 | static int rbd_dev_refresh(struct rbd_device *rbd_dev, u64 *hver); |
380 | static int rbd_dev_v2_refresh(struct rbd_device *rbd_dev, u64 *hver); | |
59c2be1e | 381 | |
602adf40 YS |
382 | static int rbd_open(struct block_device *bdev, fmode_t mode) |
383 | { | |
f0f8cef5 | 384 | struct rbd_device *rbd_dev = bdev->bd_disk->private_data; |
b82d167b | 385 | bool removing = false; |
602adf40 | 386 | |
f84344f3 | 387 | if ((mode & FMODE_WRITE) && rbd_dev->mapping.read_only) |
602adf40 YS |
388 | return -EROFS; |
389 | ||
a14ea269 | 390 | spin_lock_irq(&rbd_dev->lock); |
b82d167b AE |
391 | if (test_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags)) |
392 | removing = true; | |
393 | else | |
394 | rbd_dev->open_count++; | |
a14ea269 | 395 | spin_unlock_irq(&rbd_dev->lock); |
b82d167b AE |
396 | if (removing) |
397 | return -ENOENT; | |
398 | ||
42382b70 | 399 | mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING); |
c3e946ce | 400 | (void) get_device(&rbd_dev->dev); |
f84344f3 | 401 | set_device_ro(bdev, rbd_dev->mapping.read_only); |
42382b70 | 402 | mutex_unlock(&ctl_mutex); |
340c7a2b | 403 | |
602adf40 YS |
404 | return 0; |
405 | } | |
406 | ||
dfc5606d YS |
407 | static int rbd_release(struct gendisk *disk, fmode_t mode) |
408 | { | |
409 | struct rbd_device *rbd_dev = disk->private_data; | |
b82d167b AE |
410 | unsigned long open_count_before; |
411 | ||
a14ea269 | 412 | spin_lock_irq(&rbd_dev->lock); |
b82d167b | 413 | open_count_before = rbd_dev->open_count--; |
a14ea269 | 414 | spin_unlock_irq(&rbd_dev->lock); |
b82d167b | 415 | rbd_assert(open_count_before > 0); |
dfc5606d | 416 | |
42382b70 | 417 | mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING); |
c3e946ce | 418 | put_device(&rbd_dev->dev); |
42382b70 | 419 | mutex_unlock(&ctl_mutex); |
dfc5606d YS |
420 | |
421 | return 0; | |
422 | } | |
423 | ||
602adf40 YS |
424 | static const struct block_device_operations rbd_bd_ops = { |
425 | .owner = THIS_MODULE, | |
426 | .open = rbd_open, | |
dfc5606d | 427 | .release = rbd_release, |
602adf40 YS |
428 | }; |
429 | ||
430 | /* | |
431 | * Initialize an rbd client instance. | |
43ae4701 | 432 | * We own *ceph_opts. |
602adf40 | 433 | */ |
f8c38929 | 434 | static struct rbd_client *rbd_client_create(struct ceph_options *ceph_opts) |
602adf40 YS |
435 | { |
436 | struct rbd_client *rbdc; | |
437 | int ret = -ENOMEM; | |
438 | ||
37206ee5 | 439 | dout("%s:\n", __func__); |
602adf40 YS |
440 | rbdc = kmalloc(sizeof(struct rbd_client), GFP_KERNEL); |
441 | if (!rbdc) | |
442 | goto out_opt; | |
443 | ||
444 | kref_init(&rbdc->kref); | |
445 | INIT_LIST_HEAD(&rbdc->node); | |
446 | ||
bc534d86 AE |
447 | mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING); |
448 | ||
43ae4701 | 449 | rbdc->client = ceph_create_client(ceph_opts, rbdc, 0, 0); |
602adf40 | 450 | if (IS_ERR(rbdc->client)) |
bc534d86 | 451 | goto out_mutex; |
43ae4701 | 452 | ceph_opts = NULL; /* Now rbdc->client is responsible for ceph_opts */ |
602adf40 YS |
453 | |
454 | ret = ceph_open_session(rbdc->client); | |
455 | if (ret < 0) | |
456 | goto out_err; | |
457 | ||
432b8587 | 458 | spin_lock(&rbd_client_list_lock); |
602adf40 | 459 | list_add_tail(&rbdc->node, &rbd_client_list); |
432b8587 | 460 | spin_unlock(&rbd_client_list_lock); |
602adf40 | 461 | |
bc534d86 | 462 | mutex_unlock(&ctl_mutex); |
37206ee5 | 463 | dout("%s: rbdc %p\n", __func__, rbdc); |
bc534d86 | 464 | |
602adf40 YS |
465 | return rbdc; |
466 | ||
467 | out_err: | |
468 | ceph_destroy_client(rbdc->client); | |
bc534d86 AE |
469 | out_mutex: |
470 | mutex_unlock(&ctl_mutex); | |
602adf40 YS |
471 | kfree(rbdc); |
472 | out_opt: | |
43ae4701 AE |
473 | if (ceph_opts) |
474 | ceph_destroy_options(ceph_opts); | |
37206ee5 AE |
475 | dout("%s: error %d\n", __func__, ret); |
476 | ||
28f259b7 | 477 | return ERR_PTR(ret); |
602adf40 YS |
478 | } |
479 | ||
480 | /* | |
1f7ba331 AE |
481 | * Find a ceph client with specific addr and configuration. If |
482 | * found, bump its reference count. | |
602adf40 | 483 | */ |
1f7ba331 | 484 | static struct rbd_client *rbd_client_find(struct ceph_options *ceph_opts) |
602adf40 YS |
485 | { |
486 | struct rbd_client *client_node; | |
1f7ba331 | 487 | bool found = false; |
602adf40 | 488 | |
43ae4701 | 489 | if (ceph_opts->flags & CEPH_OPT_NOSHARE) |
602adf40 YS |
490 | return NULL; |
491 | ||
1f7ba331 AE |
492 | spin_lock(&rbd_client_list_lock); |
493 | list_for_each_entry(client_node, &rbd_client_list, node) { | |
494 | if (!ceph_compare_options(ceph_opts, client_node->client)) { | |
495 | kref_get(&client_node->kref); | |
496 | found = true; | |
497 | break; | |
498 | } | |
499 | } | |
500 | spin_unlock(&rbd_client_list_lock); | |
501 | ||
502 | return found ? client_node : NULL; | |
602adf40 YS |
503 | } |
504 | ||
59c2be1e YS |
505 | /* |
506 | * mount options | |
507 | */ | |
508 | enum { | |
59c2be1e YS |
509 | Opt_last_int, |
510 | /* int args above */ | |
511 | Opt_last_string, | |
512 | /* string args above */ | |
cc0538b6 AE |
513 | Opt_read_only, |
514 | Opt_read_write, | |
515 | /* Boolean args above */ | |
516 | Opt_last_bool, | |
59c2be1e YS |
517 | }; |
518 | ||
43ae4701 | 519 | static match_table_t rbd_opts_tokens = { |
59c2be1e YS |
520 | /* int args above */ |
521 | /* string args above */ | |
be466c1c | 522 | {Opt_read_only, "read_only"}, |
cc0538b6 AE |
523 | {Opt_read_only, "ro"}, /* Alternate spelling */ |
524 | {Opt_read_write, "read_write"}, | |
525 | {Opt_read_write, "rw"}, /* Alternate spelling */ | |
526 | /* Boolean args above */ | |
59c2be1e YS |
527 | {-1, NULL} |
528 | }; | |
529 | ||
98571b5a AE |
530 | struct rbd_options { |
531 | bool read_only; | |
532 | }; | |
533 | ||
534 | #define RBD_READ_ONLY_DEFAULT false | |
535 | ||
59c2be1e YS |
536 | static int parse_rbd_opts_token(char *c, void *private) |
537 | { | |
43ae4701 | 538 | struct rbd_options *rbd_opts = private; |
59c2be1e YS |
539 | substring_t argstr[MAX_OPT_ARGS]; |
540 | int token, intval, ret; | |
541 | ||
43ae4701 | 542 | token = match_token(c, rbd_opts_tokens, argstr); |
59c2be1e YS |
543 | if (token < 0) |
544 | return -EINVAL; | |
545 | ||
546 | if (token < Opt_last_int) { | |
547 | ret = match_int(&argstr[0], &intval); | |
548 | if (ret < 0) { | |
549 | pr_err("bad mount option arg (not int) " | |
550 | "at '%s'\n", c); | |
551 | return ret; | |
552 | } | |
553 | dout("got int token %d val %d\n", token, intval); | |
554 | } else if (token > Opt_last_int && token < Opt_last_string) { | |
555 | dout("got string token %d val %s\n", token, | |
556 | argstr[0].from); | |
cc0538b6 AE |
557 | } else if (token > Opt_last_string && token < Opt_last_bool) { |
558 | dout("got Boolean token %d\n", token); | |
59c2be1e YS |
559 | } else { |
560 | dout("got token %d\n", token); | |
561 | } | |
562 | ||
563 | switch (token) { | |
cc0538b6 AE |
564 | case Opt_read_only: |
565 | rbd_opts->read_only = true; | |
566 | break; | |
567 | case Opt_read_write: | |
568 | rbd_opts->read_only = false; | |
569 | break; | |
59c2be1e | 570 | default: |
aafb230e AE |
571 | rbd_assert(false); |
572 | break; | |
59c2be1e YS |
573 | } |
574 | return 0; | |
575 | } | |
576 | ||
602adf40 YS |
577 | /* |
578 | * Get a ceph client with specific addr and configuration, if one does | |
579 | * not exist create it. | |
580 | */ | |
9d3997fd | 581 | static struct rbd_client *rbd_get_client(struct ceph_options *ceph_opts) |
602adf40 | 582 | { |
f8c38929 | 583 | struct rbd_client *rbdc; |
59c2be1e | 584 | |
1f7ba331 | 585 | rbdc = rbd_client_find(ceph_opts); |
9d3997fd | 586 | if (rbdc) /* using an existing client */ |
43ae4701 | 587 | ceph_destroy_options(ceph_opts); |
9d3997fd | 588 | else |
f8c38929 | 589 | rbdc = rbd_client_create(ceph_opts); |
602adf40 | 590 | |
9d3997fd | 591 | return rbdc; |
602adf40 YS |
592 | } |
593 | ||
594 | /* | |
595 | * Destroy ceph client | |
d23a4b3f | 596 | * |
432b8587 | 597 | * Caller must hold rbd_client_list_lock. |
602adf40 YS |
598 | */ |
599 | static void rbd_client_release(struct kref *kref) | |
600 | { | |
601 | struct rbd_client *rbdc = container_of(kref, struct rbd_client, kref); | |
602 | ||
37206ee5 | 603 | dout("%s: rbdc %p\n", __func__, rbdc); |
cd9d9f5d | 604 | spin_lock(&rbd_client_list_lock); |
602adf40 | 605 | list_del(&rbdc->node); |
cd9d9f5d | 606 | spin_unlock(&rbd_client_list_lock); |
602adf40 YS |
607 | |
608 | ceph_destroy_client(rbdc->client); | |
609 | kfree(rbdc); | |
610 | } | |
611 | ||
612 | /* | |
613 | * Drop reference to ceph client node. If it's not referenced anymore, release | |
614 | * it. | |
615 | */ | |
9d3997fd | 616 | static void rbd_put_client(struct rbd_client *rbdc) |
602adf40 | 617 | { |
c53d5893 AE |
618 | if (rbdc) |
619 | kref_put(&rbdc->kref, rbd_client_release); | |
602adf40 YS |
620 | } |
621 | ||
a30b71b9 AE |
622 | static bool rbd_image_format_valid(u32 image_format) |
623 | { | |
624 | return image_format == 1 || image_format == 2; | |
625 | } | |
626 | ||
8e94af8e AE |
627 | static bool rbd_dev_ondisk_valid(struct rbd_image_header_ondisk *ondisk) |
628 | { | |
103a150f AE |
629 | size_t size; |
630 | u32 snap_count; | |
631 | ||
632 | /* The header has to start with the magic rbd header text */ | |
633 | if (memcmp(&ondisk->text, RBD_HEADER_TEXT, sizeof (RBD_HEADER_TEXT))) | |
634 | return false; | |
635 | ||
db2388b6 AE |
636 | /* The bio layer requires at least sector-sized I/O */ |
637 | ||
638 | if (ondisk->options.order < SECTOR_SHIFT) | |
639 | return false; | |
640 | ||
641 | /* If we use u64 in a few spots we may be able to loosen this */ | |
642 | ||
643 | if (ondisk->options.order > 8 * sizeof (int) - 1) | |
644 | return false; | |
645 | ||
103a150f AE |
646 | /* |
647 | * The size of a snapshot header has to fit in a size_t, and | |
648 | * that limits the number of snapshots. | |
649 | */ | |
650 | snap_count = le32_to_cpu(ondisk->snap_count); | |
651 | size = SIZE_MAX - sizeof (struct ceph_snap_context); | |
652 | if (snap_count > size / sizeof (__le64)) | |
653 | return false; | |
654 | ||
655 | /* | |
656 | * Not only that, but the size of the entire the snapshot | |
657 | * header must also be representable in a size_t. | |
658 | */ | |
659 | size -= snap_count * sizeof (__le64); | |
660 | if ((u64) size < le64_to_cpu(ondisk->snap_names_len)) | |
661 | return false; | |
662 | ||
663 | return true; | |
8e94af8e AE |
664 | } |
665 | ||
602adf40 YS |
666 | /* |
667 | * Create a new header structure, translate header format from the on-disk | |
668 | * header. | |
669 | */ | |
670 | static int rbd_header_from_disk(struct rbd_image_header *header, | |
4156d998 | 671 | struct rbd_image_header_ondisk *ondisk) |
602adf40 | 672 | { |
ccece235 | 673 | u32 snap_count; |
58c17b0e | 674 | size_t len; |
d2bb24e5 | 675 | size_t size; |
621901d6 | 676 | u32 i; |
602adf40 | 677 | |
6a52325f AE |
678 | memset(header, 0, sizeof (*header)); |
679 | ||
103a150f AE |
680 | snap_count = le32_to_cpu(ondisk->snap_count); |
681 | ||
58c17b0e AE |
682 | len = strnlen(ondisk->object_prefix, sizeof (ondisk->object_prefix)); |
683 | header->object_prefix = kmalloc(len + 1, GFP_KERNEL); | |
6a52325f | 684 | if (!header->object_prefix) |
602adf40 | 685 | return -ENOMEM; |
58c17b0e AE |
686 | memcpy(header->object_prefix, ondisk->object_prefix, len); |
687 | header->object_prefix[len] = '\0'; | |
00f1f36f | 688 | |
602adf40 | 689 | if (snap_count) { |
f785cc1d AE |
690 | u64 snap_names_len = le64_to_cpu(ondisk->snap_names_len); |
691 | ||
621901d6 AE |
692 | /* Save a copy of the snapshot names */ |
693 | ||
f785cc1d AE |
694 | if (snap_names_len > (u64) SIZE_MAX) |
695 | return -EIO; | |
696 | header->snap_names = kmalloc(snap_names_len, GFP_KERNEL); | |
602adf40 | 697 | if (!header->snap_names) |
6a52325f | 698 | goto out_err; |
f785cc1d AE |
699 | /* |
700 | * Note that rbd_dev_v1_header_read() guarantees | |
701 | * the ondisk buffer we're working with has | |
702 | * snap_names_len bytes beyond the end of the | |
703 | * snapshot id array, this memcpy() is safe. | |
704 | */ | |
705 | memcpy(header->snap_names, &ondisk->snaps[snap_count], | |
706 | snap_names_len); | |
6a52325f | 707 | |
621901d6 AE |
708 | /* Record each snapshot's size */ |
709 | ||
d2bb24e5 AE |
710 | size = snap_count * sizeof (*header->snap_sizes); |
711 | header->snap_sizes = kmalloc(size, GFP_KERNEL); | |
602adf40 | 712 | if (!header->snap_sizes) |
6a52325f | 713 | goto out_err; |
621901d6 AE |
714 | for (i = 0; i < snap_count; i++) |
715 | header->snap_sizes[i] = | |
716 | le64_to_cpu(ondisk->snaps[i].image_size); | |
602adf40 | 717 | } else { |
ccece235 | 718 | WARN_ON(ondisk->snap_names_len); |
602adf40 YS |
719 | header->snap_names = NULL; |
720 | header->snap_sizes = NULL; | |
721 | } | |
849b4260 | 722 | |
34b13184 | 723 | header->features = 0; /* No features support in v1 images */ |
602adf40 YS |
724 | header->obj_order = ondisk->options.order; |
725 | header->crypt_type = ondisk->options.crypt_type; | |
726 | header->comp_type = ondisk->options.comp_type; | |
6a52325f | 727 | |
621901d6 AE |
728 | /* Allocate and fill in the snapshot context */ |
729 | ||
f84344f3 | 730 | header->image_size = le64_to_cpu(ondisk->image_size); |
6a52325f AE |
731 | size = sizeof (struct ceph_snap_context); |
732 | size += snap_count * sizeof (header->snapc->snaps[0]); | |
733 | header->snapc = kzalloc(size, GFP_KERNEL); | |
734 | if (!header->snapc) | |
735 | goto out_err; | |
602adf40 YS |
736 | |
737 | atomic_set(&header->snapc->nref, 1); | |
505cbb9b | 738 | header->snapc->seq = le64_to_cpu(ondisk->snap_seq); |
602adf40 | 739 | header->snapc->num_snaps = snap_count; |
621901d6 AE |
740 | for (i = 0; i < snap_count; i++) |
741 | header->snapc->snaps[i] = | |
742 | le64_to_cpu(ondisk->snaps[i].id); | |
602adf40 YS |
743 | |
744 | return 0; | |
745 | ||
6a52325f | 746 | out_err: |
849b4260 | 747 | kfree(header->snap_sizes); |
ccece235 | 748 | header->snap_sizes = NULL; |
602adf40 | 749 | kfree(header->snap_names); |
ccece235 | 750 | header->snap_names = NULL; |
6a52325f AE |
751 | kfree(header->object_prefix); |
752 | header->object_prefix = NULL; | |
ccece235 | 753 | |
00f1f36f | 754 | return -ENOMEM; |
602adf40 YS |
755 | } |
756 | ||
9e15b77d AE |
757 | static const char *rbd_snap_name(struct rbd_device *rbd_dev, u64 snap_id) |
758 | { | |
759 | struct rbd_snap *snap; | |
760 | ||
761 | if (snap_id == CEPH_NOSNAP) | |
762 | return RBD_SNAP_HEAD_NAME; | |
763 | ||
764 | list_for_each_entry(snap, &rbd_dev->snaps, node) | |
765 | if (snap_id == snap->id) | |
766 | return snap->name; | |
767 | ||
768 | return NULL; | |
769 | } | |
770 | ||
8836b995 | 771 | static int snap_by_name(struct rbd_device *rbd_dev, const char *snap_name) |
602adf40 | 772 | { |
602adf40 | 773 | |
e86924a8 | 774 | struct rbd_snap *snap; |
602adf40 | 775 | |
e86924a8 AE |
776 | list_for_each_entry(snap, &rbd_dev->snaps, node) { |
777 | if (!strcmp(snap_name, snap->name)) { | |
0d7dbfce | 778 | rbd_dev->spec->snap_id = snap->id; |
e86924a8 | 779 | rbd_dev->mapping.size = snap->size; |
34b13184 | 780 | rbd_dev->mapping.features = snap->features; |
602adf40 | 781 | |
e86924a8 | 782 | return 0; |
00f1f36f | 783 | } |
00f1f36f | 784 | } |
e86924a8 | 785 | |
00f1f36f | 786 | return -ENOENT; |
602adf40 YS |
787 | } |
788 | ||
819d52bf | 789 | static int rbd_dev_set_mapping(struct rbd_device *rbd_dev) |
602adf40 | 790 | { |
78dc447d | 791 | int ret; |
602adf40 | 792 | |
0d7dbfce | 793 | if (!memcmp(rbd_dev->spec->snap_name, RBD_SNAP_HEAD_NAME, |
cc9d734c | 794 | sizeof (RBD_SNAP_HEAD_NAME))) { |
0d7dbfce | 795 | rbd_dev->spec->snap_id = CEPH_NOSNAP; |
99c1f08f | 796 | rbd_dev->mapping.size = rbd_dev->header.image_size; |
34b13184 | 797 | rbd_dev->mapping.features = rbd_dev->header.features; |
e86924a8 | 798 | ret = 0; |
602adf40 | 799 | } else { |
0d7dbfce | 800 | ret = snap_by_name(rbd_dev, rbd_dev->spec->snap_name); |
602adf40 YS |
801 | if (ret < 0) |
802 | goto done; | |
f84344f3 | 803 | rbd_dev->mapping.read_only = true; |
602adf40 | 804 | } |
6d292906 AE |
805 | set_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags); |
806 | ||
602adf40 | 807 | done: |
602adf40 YS |
808 | return ret; |
809 | } | |
810 | ||
811 | static void rbd_header_free(struct rbd_image_header *header) | |
812 | { | |
849b4260 | 813 | kfree(header->object_prefix); |
d78fd7ae | 814 | header->object_prefix = NULL; |
602adf40 | 815 | kfree(header->snap_sizes); |
d78fd7ae | 816 | header->snap_sizes = NULL; |
849b4260 | 817 | kfree(header->snap_names); |
d78fd7ae | 818 | header->snap_names = NULL; |
d1d25646 | 819 | ceph_put_snap_context(header->snapc); |
d78fd7ae | 820 | header->snapc = NULL; |
602adf40 YS |
821 | } |
822 | ||
98571b5a | 823 | static const char *rbd_segment_name(struct rbd_device *rbd_dev, u64 offset) |
602adf40 | 824 | { |
65ccfe21 AE |
825 | char *name; |
826 | u64 segment; | |
827 | int ret; | |
602adf40 | 828 | |
2fd82b9e | 829 | name = kmalloc(MAX_OBJ_NAME_SIZE + 1, GFP_NOIO); |
65ccfe21 AE |
830 | if (!name) |
831 | return NULL; | |
832 | segment = offset >> rbd_dev->header.obj_order; | |
2fd82b9e | 833 | ret = snprintf(name, MAX_OBJ_NAME_SIZE + 1, "%s.%012llx", |
65ccfe21 | 834 | rbd_dev->header.object_prefix, segment); |
2fd82b9e | 835 | if (ret < 0 || ret > MAX_OBJ_NAME_SIZE) { |
65ccfe21 AE |
836 | pr_err("error formatting segment name for #%llu (%d)\n", |
837 | segment, ret); | |
838 | kfree(name); | |
839 | name = NULL; | |
840 | } | |
602adf40 | 841 | |
65ccfe21 AE |
842 | return name; |
843 | } | |
602adf40 | 844 | |
65ccfe21 AE |
845 | static u64 rbd_segment_offset(struct rbd_device *rbd_dev, u64 offset) |
846 | { | |
847 | u64 segment_size = (u64) 1 << rbd_dev->header.obj_order; | |
602adf40 | 848 | |
65ccfe21 AE |
849 | return offset & (segment_size - 1); |
850 | } | |
851 | ||
852 | static u64 rbd_segment_length(struct rbd_device *rbd_dev, | |
853 | u64 offset, u64 length) | |
854 | { | |
855 | u64 segment_size = (u64) 1 << rbd_dev->header.obj_order; | |
856 | ||
857 | offset &= segment_size - 1; | |
858 | ||
aafb230e | 859 | rbd_assert(length <= U64_MAX - offset); |
65ccfe21 AE |
860 | if (offset + length > segment_size) |
861 | length = segment_size - offset; | |
862 | ||
863 | return length; | |
602adf40 YS |
864 | } |
865 | ||
029bcbd8 JD |
866 | /* |
867 | * returns the size of an object in the image | |
868 | */ | |
869 | static u64 rbd_obj_bytes(struct rbd_image_header *header) | |
870 | { | |
871 | return 1 << header->obj_order; | |
872 | } | |
873 | ||
602adf40 YS |
874 | /* |
875 | * bio helpers | |
876 | */ | |
877 | ||
878 | static void bio_chain_put(struct bio *chain) | |
879 | { | |
880 | struct bio *tmp; | |
881 | ||
882 | while (chain) { | |
883 | tmp = chain; | |
884 | chain = chain->bi_next; | |
885 | bio_put(tmp); | |
886 | } | |
887 | } | |
888 | ||
889 | /* | |
890 | * zeros a bio chain, starting at specific offset | |
891 | */ | |
892 | static void zero_bio_chain(struct bio *chain, int start_ofs) | |
893 | { | |
894 | struct bio_vec *bv; | |
895 | unsigned long flags; | |
896 | void *buf; | |
897 | int i; | |
898 | int pos = 0; | |
899 | ||
900 | while (chain) { | |
901 | bio_for_each_segment(bv, chain, i) { | |
902 | if (pos + bv->bv_len > start_ofs) { | |
903 | int remainder = max(start_ofs - pos, 0); | |
904 | buf = bvec_kmap_irq(bv, &flags); | |
905 | memset(buf + remainder, 0, | |
906 | bv->bv_len - remainder); | |
85b5aaa6 | 907 | bvec_kunmap_irq(buf, &flags); |
602adf40 YS |
908 | } |
909 | pos += bv->bv_len; | |
910 | } | |
911 | ||
912 | chain = chain->bi_next; | |
913 | } | |
914 | } | |
915 | ||
916 | /* | |
f7760dad AE |
917 | * Clone a portion of a bio, starting at the given byte offset |
918 | * and continuing for the number of bytes indicated. | |
602adf40 | 919 | */ |
f7760dad AE |
920 | static struct bio *bio_clone_range(struct bio *bio_src, |
921 | unsigned int offset, | |
922 | unsigned int len, | |
923 | gfp_t gfpmask) | |
602adf40 | 924 | { |
f7760dad AE |
925 | struct bio_vec *bv; |
926 | unsigned int resid; | |
927 | unsigned short idx; | |
928 | unsigned int voff; | |
929 | unsigned short end_idx; | |
930 | unsigned short vcnt; | |
931 | struct bio *bio; | |
932 | ||
933 | /* Handle the easy case for the caller */ | |
934 | ||
935 | if (!offset && len == bio_src->bi_size) | |
936 | return bio_clone(bio_src, gfpmask); | |
937 | ||
938 | if (WARN_ON_ONCE(!len)) | |
939 | return NULL; | |
940 | if (WARN_ON_ONCE(len > bio_src->bi_size)) | |
941 | return NULL; | |
942 | if (WARN_ON_ONCE(offset > bio_src->bi_size - len)) | |
943 | return NULL; | |
944 | ||
945 | /* Find first affected segment... */ | |
946 | ||
947 | resid = offset; | |
948 | __bio_for_each_segment(bv, bio_src, idx, 0) { | |
949 | if (resid < bv->bv_len) | |
950 | break; | |
951 | resid -= bv->bv_len; | |
602adf40 | 952 | } |
f7760dad | 953 | voff = resid; |
602adf40 | 954 | |
f7760dad | 955 | /* ...and the last affected segment */ |
602adf40 | 956 | |
f7760dad AE |
957 | resid += len; |
958 | __bio_for_each_segment(bv, bio_src, end_idx, idx) { | |
959 | if (resid <= bv->bv_len) | |
960 | break; | |
961 | resid -= bv->bv_len; | |
962 | } | |
963 | vcnt = end_idx - idx + 1; | |
964 | ||
965 | /* Build the clone */ | |
966 | ||
967 | bio = bio_alloc(gfpmask, (unsigned int) vcnt); | |
968 | if (!bio) | |
969 | return NULL; /* ENOMEM */ | |
602adf40 | 970 | |
f7760dad AE |
971 | bio->bi_bdev = bio_src->bi_bdev; |
972 | bio->bi_sector = bio_src->bi_sector + (offset >> SECTOR_SHIFT); | |
973 | bio->bi_rw = bio_src->bi_rw; | |
974 | bio->bi_flags |= 1 << BIO_CLONED; | |
975 | ||
976 | /* | |
977 | * Copy over our part of the bio_vec, then update the first | |
978 | * and last (or only) entries. | |
979 | */ | |
980 | memcpy(&bio->bi_io_vec[0], &bio_src->bi_io_vec[idx], | |
981 | vcnt * sizeof (struct bio_vec)); | |
982 | bio->bi_io_vec[0].bv_offset += voff; | |
983 | if (vcnt > 1) { | |
984 | bio->bi_io_vec[0].bv_len -= voff; | |
985 | bio->bi_io_vec[vcnt - 1].bv_len = resid; | |
986 | } else { | |
987 | bio->bi_io_vec[0].bv_len = len; | |
602adf40 YS |
988 | } |
989 | ||
f7760dad AE |
990 | bio->bi_vcnt = vcnt; |
991 | bio->bi_size = len; | |
992 | bio->bi_idx = 0; | |
993 | ||
994 | return bio; | |
995 | } | |
996 | ||
997 | /* | |
998 | * Clone a portion of a bio chain, starting at the given byte offset | |
999 | * into the first bio in the source chain and continuing for the | |
1000 | * number of bytes indicated. The result is another bio chain of | |
1001 | * exactly the given length, or a null pointer on error. | |
1002 | * | |
1003 | * The bio_src and offset parameters are both in-out. On entry they | |
1004 | * refer to the first source bio and the offset into that bio where | |
1005 | * the start of data to be cloned is located. | |
1006 | * | |
1007 | * On return, bio_src is updated to refer to the bio in the source | |
1008 | * chain that contains first un-cloned byte, and *offset will | |
1009 | * contain the offset of that byte within that bio. | |
1010 | */ | |
1011 | static struct bio *bio_chain_clone_range(struct bio **bio_src, | |
1012 | unsigned int *offset, | |
1013 | unsigned int len, | |
1014 | gfp_t gfpmask) | |
1015 | { | |
1016 | struct bio *bi = *bio_src; | |
1017 | unsigned int off = *offset; | |
1018 | struct bio *chain = NULL; | |
1019 | struct bio **end; | |
1020 | ||
1021 | /* Build up a chain of clone bios up to the limit */ | |
1022 | ||
1023 | if (!bi || off >= bi->bi_size || !len) | |
1024 | return NULL; /* Nothing to clone */ | |
602adf40 | 1025 | |
f7760dad AE |
1026 | end = &chain; |
1027 | while (len) { | |
1028 | unsigned int bi_size; | |
1029 | struct bio *bio; | |
1030 | ||
f5400b7a AE |
1031 | if (!bi) { |
1032 | rbd_warn(NULL, "bio_chain exhausted with %u left", len); | |
f7760dad | 1033 | goto out_err; /* EINVAL; ran out of bio's */ |
f5400b7a | 1034 | } |
f7760dad AE |
1035 | bi_size = min_t(unsigned int, bi->bi_size - off, len); |
1036 | bio = bio_clone_range(bi, off, bi_size, gfpmask); | |
1037 | if (!bio) | |
1038 | goto out_err; /* ENOMEM */ | |
1039 | ||
1040 | *end = bio; | |
1041 | end = &bio->bi_next; | |
602adf40 | 1042 | |
f7760dad AE |
1043 | off += bi_size; |
1044 | if (off == bi->bi_size) { | |
1045 | bi = bi->bi_next; | |
1046 | off = 0; | |
1047 | } | |
1048 | len -= bi_size; | |
1049 | } | |
1050 | *bio_src = bi; | |
1051 | *offset = off; | |
1052 | ||
1053 | return chain; | |
1054 | out_err: | |
1055 | bio_chain_put(chain); | |
602adf40 | 1056 | |
602adf40 YS |
1057 | return NULL; |
1058 | } | |
1059 | ||
bf0d5f50 AE |
1060 | static void rbd_obj_request_get(struct rbd_obj_request *obj_request) |
1061 | { | |
37206ee5 AE |
1062 | dout("%s: obj %p (was %d)\n", __func__, obj_request, |
1063 | atomic_read(&obj_request->kref.refcount)); | |
bf0d5f50 AE |
1064 | kref_get(&obj_request->kref); |
1065 | } | |
1066 | ||
1067 | static void rbd_obj_request_destroy(struct kref *kref); | |
1068 | static void rbd_obj_request_put(struct rbd_obj_request *obj_request) | |
1069 | { | |
1070 | rbd_assert(obj_request != NULL); | |
37206ee5 AE |
1071 | dout("%s: obj %p (was %d)\n", __func__, obj_request, |
1072 | atomic_read(&obj_request->kref.refcount)); | |
bf0d5f50 AE |
1073 | kref_put(&obj_request->kref, rbd_obj_request_destroy); |
1074 | } | |
1075 | ||
1076 | static void rbd_img_request_get(struct rbd_img_request *img_request) | |
1077 | { | |
37206ee5 AE |
1078 | dout("%s: img %p (was %d)\n", __func__, img_request, |
1079 | atomic_read(&img_request->kref.refcount)); | |
bf0d5f50 AE |
1080 | kref_get(&img_request->kref); |
1081 | } | |
1082 | ||
1083 | static void rbd_img_request_destroy(struct kref *kref); | |
1084 | static void rbd_img_request_put(struct rbd_img_request *img_request) | |
1085 | { | |
1086 | rbd_assert(img_request != NULL); | |
37206ee5 AE |
1087 | dout("%s: img %p (was %d)\n", __func__, img_request, |
1088 | atomic_read(&img_request->kref.refcount)); | |
bf0d5f50 AE |
1089 | kref_put(&img_request->kref, rbd_img_request_destroy); |
1090 | } | |
1091 | ||
1092 | static inline void rbd_img_obj_request_add(struct rbd_img_request *img_request, | |
1093 | struct rbd_obj_request *obj_request) | |
1094 | { | |
25dcf954 AE |
1095 | rbd_assert(obj_request->img_request == NULL); |
1096 | ||
bf0d5f50 AE |
1097 | rbd_obj_request_get(obj_request); |
1098 | obj_request->img_request = img_request; | |
25dcf954 | 1099 | obj_request->which = img_request->obj_request_count; |
bf0d5f50 | 1100 | rbd_assert(obj_request->which != BAD_WHICH); |
25dcf954 AE |
1101 | img_request->obj_request_count++; |
1102 | list_add_tail(&obj_request->links, &img_request->obj_requests); | |
37206ee5 AE |
1103 | dout("%s: img %p obj %p w=%u\n", __func__, img_request, obj_request, |
1104 | obj_request->which); | |
bf0d5f50 AE |
1105 | } |
1106 | ||
1107 | static inline void rbd_img_obj_request_del(struct rbd_img_request *img_request, | |
1108 | struct rbd_obj_request *obj_request) | |
1109 | { | |
1110 | rbd_assert(obj_request->which != BAD_WHICH); | |
25dcf954 | 1111 | |
37206ee5 AE |
1112 | dout("%s: img %p obj %p w=%u\n", __func__, img_request, obj_request, |
1113 | obj_request->which); | |
bf0d5f50 | 1114 | list_del(&obj_request->links); |
25dcf954 AE |
1115 | rbd_assert(img_request->obj_request_count > 0); |
1116 | img_request->obj_request_count--; | |
1117 | rbd_assert(obj_request->which == img_request->obj_request_count); | |
1118 | obj_request->which = BAD_WHICH; | |
bf0d5f50 | 1119 | rbd_assert(obj_request->img_request == img_request); |
bf0d5f50 | 1120 | obj_request->img_request = NULL; |
25dcf954 | 1121 | obj_request->callback = NULL; |
bf0d5f50 AE |
1122 | rbd_obj_request_put(obj_request); |
1123 | } | |
1124 | ||
1125 | static bool obj_request_type_valid(enum obj_request_type type) | |
1126 | { | |
1127 | switch (type) { | |
9969ebc5 | 1128 | case OBJ_REQUEST_NODATA: |
bf0d5f50 | 1129 | case OBJ_REQUEST_BIO: |
788e2df3 | 1130 | case OBJ_REQUEST_PAGES: |
bf0d5f50 AE |
1131 | return true; |
1132 | default: | |
1133 | return false; | |
1134 | } | |
1135 | } | |
1136 | ||
bf0d5f50 AE |
1137 | static int rbd_obj_request_submit(struct ceph_osd_client *osdc, |
1138 | struct rbd_obj_request *obj_request) | |
1139 | { | |
37206ee5 AE |
1140 | dout("%s: osdc %p obj %p\n", __func__, osdc, obj_request); |
1141 | ||
bf0d5f50 AE |
1142 | return ceph_osdc_start_request(osdc, obj_request->osd_req, false); |
1143 | } | |
1144 | ||
1145 | static void rbd_img_request_complete(struct rbd_img_request *img_request) | |
1146 | { | |
37206ee5 | 1147 | dout("%s: img %p\n", __func__, img_request); |
bf0d5f50 AE |
1148 | if (img_request->callback) |
1149 | img_request->callback(img_request); | |
1150 | else | |
1151 | rbd_img_request_put(img_request); | |
1152 | } | |
1153 | ||
788e2df3 AE |
1154 | /* Caller is responsible for rbd_obj_request_destroy(obj_request) */ |
1155 | ||
1156 | static int rbd_obj_request_wait(struct rbd_obj_request *obj_request) | |
1157 | { | |
37206ee5 AE |
1158 | dout("%s: obj %p\n", __func__, obj_request); |
1159 | ||
788e2df3 AE |
1160 | return wait_for_completion_interruptible(&obj_request->completion); |
1161 | } | |
1162 | ||
07741308 AE |
1163 | static void obj_request_done_init(struct rbd_obj_request *obj_request) |
1164 | { | |
1165 | atomic_set(&obj_request->done, 0); | |
1166 | smp_wmb(); | |
1167 | } | |
1168 | ||
1169 | static void obj_request_done_set(struct rbd_obj_request *obj_request) | |
1170 | { | |
632b88ca AE |
1171 | int done; |
1172 | ||
1173 | done = atomic_inc_return(&obj_request->done); | |
1174 | if (done > 1) { | |
1175 | struct rbd_img_request *img_request = obj_request->img_request; | |
1176 | struct rbd_device *rbd_dev; | |
1177 | ||
1178 | rbd_dev = img_request ? img_request->rbd_dev : NULL; | |
1179 | rbd_warn(rbd_dev, "obj_request %p was already done\n", | |
1180 | obj_request); | |
1181 | } | |
07741308 AE |
1182 | } |
1183 | ||
1184 | static bool obj_request_done_test(struct rbd_obj_request *obj_request) | |
1185 | { | |
632b88ca | 1186 | smp_mb(); |
07741308 AE |
1187 | return atomic_read(&obj_request->done) != 0; |
1188 | } | |
1189 | ||
6e2a4505 AE |
1190 | static void |
1191 | rbd_img_obj_request_read_callback(struct rbd_obj_request *obj_request) | |
1192 | { | |
1193 | dout("%s: obj %p img %p result %d %llu/%llu\n", __func__, | |
1194 | obj_request, obj_request->img_request, obj_request->result, | |
1195 | obj_request->xferred, obj_request->length); | |
1196 | /* | |
1197 | * ENOENT means a hole in the image. We zero-fill the | |
1198 | * entire length of the request. A short read also implies | |
1199 | * zero-fill to the end of the request. Either way we | |
1200 | * update the xferred count to indicate the whole request | |
1201 | * was satisfied. | |
1202 | */ | |
1203 | BUG_ON(obj_request->type != OBJ_REQUEST_BIO); | |
1204 | if (obj_request->result == -ENOENT) { | |
1205 | zero_bio_chain(obj_request->bio_list, 0); | |
1206 | obj_request->result = 0; | |
1207 | obj_request->xferred = obj_request->length; | |
1208 | } else if (obj_request->xferred < obj_request->length && | |
1209 | !obj_request->result) { | |
1210 | zero_bio_chain(obj_request->bio_list, obj_request->xferred); | |
1211 | obj_request->xferred = obj_request->length; | |
1212 | } | |
1213 | obj_request_done_set(obj_request); | |
1214 | } | |
1215 | ||
bf0d5f50 AE |
1216 | static void rbd_obj_request_complete(struct rbd_obj_request *obj_request) |
1217 | { | |
37206ee5 AE |
1218 | dout("%s: obj %p cb %p\n", __func__, obj_request, |
1219 | obj_request->callback); | |
bf0d5f50 AE |
1220 | if (obj_request->callback) |
1221 | obj_request->callback(obj_request); | |
788e2df3 AE |
1222 | else |
1223 | complete_all(&obj_request->completion); | |
bf0d5f50 AE |
1224 | } |
1225 | ||
c47f9371 | 1226 | static void rbd_osd_trivial_callback(struct rbd_obj_request *obj_request) |
39bf2c5d AE |
1227 | { |
1228 | dout("%s: obj %p\n", __func__, obj_request); | |
1229 | obj_request_done_set(obj_request); | |
1230 | } | |
1231 | ||
c47f9371 | 1232 | static void rbd_osd_read_callback(struct rbd_obj_request *obj_request) |
bf0d5f50 | 1233 | { |
37206ee5 | 1234 | dout("%s: obj %p result %d %llu/%llu\n", __func__, obj_request, |
c47f9371 | 1235 | obj_request->result, obj_request->xferred, obj_request->length); |
6e2a4505 AE |
1236 | if (obj_request->img_request) |
1237 | rbd_img_obj_request_read_callback(obj_request); | |
1238 | else | |
1239 | obj_request_done_set(obj_request); | |
bf0d5f50 AE |
1240 | } |
1241 | ||
c47f9371 | 1242 | static void rbd_osd_write_callback(struct rbd_obj_request *obj_request) |
bf0d5f50 | 1243 | { |
1b83bef2 SW |
1244 | dout("%s: obj %p result %d %llu\n", __func__, obj_request, |
1245 | obj_request->result, obj_request->length); | |
1246 | /* | |
1247 | * There is no such thing as a successful short write. | |
1248 | * Our xferred value is the number of bytes transferred | |
1249 | * back. Set it to our originally-requested length. | |
1250 | */ | |
1251 | obj_request->xferred = obj_request->length; | |
07741308 | 1252 | obj_request_done_set(obj_request); |
bf0d5f50 AE |
1253 | } |
1254 | ||
fbfab539 AE |
1255 | /* |
1256 | * For a simple stat call there's nothing to do. We'll do more if | |
1257 | * this is part of a write sequence for a layered image. | |
1258 | */ | |
c47f9371 | 1259 | static void rbd_osd_stat_callback(struct rbd_obj_request *obj_request) |
fbfab539 | 1260 | { |
37206ee5 | 1261 | dout("%s: obj %p\n", __func__, obj_request); |
fbfab539 AE |
1262 | obj_request_done_set(obj_request); |
1263 | } | |
1264 | ||
bf0d5f50 AE |
1265 | static void rbd_osd_req_callback(struct ceph_osd_request *osd_req, |
1266 | struct ceph_msg *msg) | |
1267 | { | |
1268 | struct rbd_obj_request *obj_request = osd_req->r_priv; | |
bf0d5f50 AE |
1269 | u16 opcode; |
1270 | ||
37206ee5 | 1271 | dout("%s: osd_req %p msg %p\n", __func__, osd_req, msg); |
bf0d5f50 AE |
1272 | rbd_assert(osd_req == obj_request->osd_req); |
1273 | rbd_assert(!!obj_request->img_request ^ | |
1274 | (obj_request->which == BAD_WHICH)); | |
1275 | ||
1b83bef2 SW |
1276 | if (osd_req->r_result < 0) |
1277 | obj_request->result = osd_req->r_result; | |
bf0d5f50 AE |
1278 | obj_request->version = le64_to_cpu(osd_req->r_reassert_version.version); |
1279 | ||
1b83bef2 | 1280 | WARN_ON(osd_req->r_num_ops != 1); /* For now */ |
bf0d5f50 | 1281 | |
c47f9371 AE |
1282 | /* |
1283 | * We support a 64-bit length, but ultimately it has to be | |
1284 | * passed to blk_end_request(), which takes an unsigned int. | |
1285 | */ | |
1b83bef2 | 1286 | obj_request->xferred = osd_req->r_reply_op_len[0]; |
c47f9371 | 1287 | rbd_assert(obj_request->xferred < (u64) UINT_MAX); |
79528734 | 1288 | opcode = osd_req->r_ops[0].op; |
bf0d5f50 AE |
1289 | switch (opcode) { |
1290 | case CEPH_OSD_OP_READ: | |
c47f9371 | 1291 | rbd_osd_read_callback(obj_request); |
bf0d5f50 AE |
1292 | break; |
1293 | case CEPH_OSD_OP_WRITE: | |
c47f9371 | 1294 | rbd_osd_write_callback(obj_request); |
bf0d5f50 | 1295 | break; |
fbfab539 | 1296 | case CEPH_OSD_OP_STAT: |
c47f9371 | 1297 | rbd_osd_stat_callback(obj_request); |
fbfab539 | 1298 | break; |
36be9a76 | 1299 | case CEPH_OSD_OP_CALL: |
b8d70035 | 1300 | case CEPH_OSD_OP_NOTIFY_ACK: |
9969ebc5 | 1301 | case CEPH_OSD_OP_WATCH: |
c47f9371 | 1302 | rbd_osd_trivial_callback(obj_request); |
9969ebc5 | 1303 | break; |
bf0d5f50 AE |
1304 | default: |
1305 | rbd_warn(NULL, "%s: unsupported op %hu\n", | |
1306 | obj_request->object_name, (unsigned short) opcode); | |
1307 | break; | |
1308 | } | |
1309 | ||
07741308 | 1310 | if (obj_request_done_test(obj_request)) |
bf0d5f50 AE |
1311 | rbd_obj_request_complete(obj_request); |
1312 | } | |
1313 | ||
2fa12320 | 1314 | static void rbd_osd_req_format(struct rbd_obj_request *obj_request, |
79528734 | 1315 | bool write_request) |
430c28c3 AE |
1316 | { |
1317 | struct rbd_img_request *img_request = obj_request->img_request; | |
8c042b0d | 1318 | struct ceph_osd_request *osd_req = obj_request->osd_req; |
430c28c3 AE |
1319 | struct ceph_snap_context *snapc = NULL; |
1320 | u64 snap_id = CEPH_NOSNAP; | |
1321 | struct timespec *mtime = NULL; | |
1322 | struct timespec now; | |
1323 | ||
8c042b0d | 1324 | rbd_assert(osd_req != NULL); |
430c28c3 AE |
1325 | |
1326 | if (write_request) { | |
1327 | now = CURRENT_TIME; | |
1328 | mtime = &now; | |
1329 | if (img_request) | |
1330 | snapc = img_request->snapc; | |
2fa12320 AE |
1331 | } else if (img_request) { |
1332 | snap_id = img_request->snap_id; | |
8c042b0d AE |
1333 | } |
1334 | ceph_osdc_build_request(osd_req, obj_request->offset, | |
79528734 | 1335 | snapc, snap_id, mtime); |
430c28c3 AE |
1336 | } |
1337 | ||
bf0d5f50 AE |
1338 | static struct ceph_osd_request *rbd_osd_req_create( |
1339 | struct rbd_device *rbd_dev, | |
1340 | bool write_request, | |
430c28c3 | 1341 | struct rbd_obj_request *obj_request) |
bf0d5f50 AE |
1342 | { |
1343 | struct rbd_img_request *img_request = obj_request->img_request; | |
1344 | struct ceph_snap_context *snapc = NULL; | |
1345 | struct ceph_osd_client *osdc; | |
1346 | struct ceph_osd_request *osd_req; | |
bf0d5f50 AE |
1347 | |
1348 | if (img_request) { | |
1349 | rbd_assert(img_request->write_request == write_request); | |
1350 | if (img_request->write_request) | |
1351 | snapc = img_request->snapc; | |
bf0d5f50 AE |
1352 | } |
1353 | ||
1354 | /* Allocate and initialize the request, for the single op */ | |
1355 | ||
1356 | osdc = &rbd_dev->rbd_client->client->osdc; | |
1357 | osd_req = ceph_osdc_alloc_request(osdc, snapc, 1, false, GFP_ATOMIC); | |
1358 | if (!osd_req) | |
1359 | return NULL; /* ENOMEM */ | |
bf0d5f50 | 1360 | |
430c28c3 | 1361 | if (write_request) |
bf0d5f50 | 1362 | osd_req->r_flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK; |
430c28c3 | 1363 | else |
bf0d5f50 | 1364 | osd_req->r_flags = CEPH_OSD_FLAG_READ; |
bf0d5f50 AE |
1365 | |
1366 | osd_req->r_callback = rbd_osd_req_callback; | |
1367 | osd_req->r_priv = obj_request; | |
1368 | ||
1369 | osd_req->r_oid_len = strlen(obj_request->object_name); | |
1370 | rbd_assert(osd_req->r_oid_len < sizeof (osd_req->r_oid)); | |
1371 | memcpy(osd_req->r_oid, obj_request->object_name, osd_req->r_oid_len); | |
1372 | ||
1373 | osd_req->r_file_layout = rbd_dev->layout; /* struct */ | |
1374 | ||
bf0d5f50 AE |
1375 | return osd_req; |
1376 | } | |
1377 | ||
1378 | static void rbd_osd_req_destroy(struct ceph_osd_request *osd_req) | |
1379 | { | |
1380 | ceph_osdc_put_request(osd_req); | |
1381 | } | |
1382 | ||
1383 | /* object_name is assumed to be a non-null pointer and NUL-terminated */ | |
1384 | ||
1385 | static struct rbd_obj_request *rbd_obj_request_create(const char *object_name, | |
1386 | u64 offset, u64 length, | |
1387 | enum obj_request_type type) | |
1388 | { | |
1389 | struct rbd_obj_request *obj_request; | |
1390 | size_t size; | |
1391 | char *name; | |
1392 | ||
1393 | rbd_assert(obj_request_type_valid(type)); | |
1394 | ||
1395 | size = strlen(object_name) + 1; | |
1396 | obj_request = kzalloc(sizeof (*obj_request) + size, GFP_KERNEL); | |
1397 | if (!obj_request) | |
1398 | return NULL; | |
1399 | ||
1400 | name = (char *)(obj_request + 1); | |
1401 | obj_request->object_name = memcpy(name, object_name, size); | |
1402 | obj_request->offset = offset; | |
1403 | obj_request->length = length; | |
1404 | obj_request->which = BAD_WHICH; | |
1405 | obj_request->type = type; | |
1406 | INIT_LIST_HEAD(&obj_request->links); | |
07741308 | 1407 | obj_request_done_init(obj_request); |
788e2df3 | 1408 | init_completion(&obj_request->completion); |
bf0d5f50 AE |
1409 | kref_init(&obj_request->kref); |
1410 | ||
37206ee5 AE |
1411 | dout("%s: \"%s\" %llu/%llu %d -> obj %p\n", __func__, object_name, |
1412 | offset, length, (int)type, obj_request); | |
1413 | ||
bf0d5f50 AE |
1414 | return obj_request; |
1415 | } | |
1416 | ||
1417 | static void rbd_obj_request_destroy(struct kref *kref) | |
1418 | { | |
1419 | struct rbd_obj_request *obj_request; | |
1420 | ||
1421 | obj_request = container_of(kref, struct rbd_obj_request, kref); | |
1422 | ||
37206ee5 AE |
1423 | dout("%s: obj %p\n", __func__, obj_request); |
1424 | ||
bf0d5f50 AE |
1425 | rbd_assert(obj_request->img_request == NULL); |
1426 | rbd_assert(obj_request->which == BAD_WHICH); | |
1427 | ||
1428 | if (obj_request->osd_req) | |
1429 | rbd_osd_req_destroy(obj_request->osd_req); | |
1430 | ||
1431 | rbd_assert(obj_request_type_valid(obj_request->type)); | |
1432 | switch (obj_request->type) { | |
9969ebc5 AE |
1433 | case OBJ_REQUEST_NODATA: |
1434 | break; /* Nothing to do */ | |
bf0d5f50 AE |
1435 | case OBJ_REQUEST_BIO: |
1436 | if (obj_request->bio_list) | |
1437 | bio_chain_put(obj_request->bio_list); | |
1438 | break; | |
788e2df3 AE |
1439 | case OBJ_REQUEST_PAGES: |
1440 | if (obj_request->pages) | |
1441 | ceph_release_page_vector(obj_request->pages, | |
1442 | obj_request->page_count); | |
1443 | break; | |
bf0d5f50 AE |
1444 | } |
1445 | ||
1446 | kfree(obj_request); | |
1447 | } | |
1448 | ||
1449 | /* | |
1450 | * Caller is responsible for filling in the list of object requests | |
1451 | * that comprises the image request, and the Linux request pointer | |
1452 | * (if there is one). | |
1453 | */ | |
cc344fa1 AE |
1454 | static struct rbd_img_request *rbd_img_request_create( |
1455 | struct rbd_device *rbd_dev, | |
bf0d5f50 AE |
1456 | u64 offset, u64 length, |
1457 | bool write_request) | |
1458 | { | |
1459 | struct rbd_img_request *img_request; | |
1460 | struct ceph_snap_context *snapc = NULL; | |
1461 | ||
1462 | img_request = kmalloc(sizeof (*img_request), GFP_ATOMIC); | |
1463 | if (!img_request) | |
1464 | return NULL; | |
1465 | ||
1466 | if (write_request) { | |
1467 | down_read(&rbd_dev->header_rwsem); | |
1468 | snapc = ceph_get_snap_context(rbd_dev->header.snapc); | |
1469 | up_read(&rbd_dev->header_rwsem); | |
1470 | if (WARN_ON(!snapc)) { | |
1471 | kfree(img_request); | |
1472 | return NULL; /* Shouldn't happen */ | |
1473 | } | |
1474 | } | |
1475 | ||
1476 | img_request->rq = NULL; | |
1477 | img_request->rbd_dev = rbd_dev; | |
1478 | img_request->offset = offset; | |
1479 | img_request->length = length; | |
1480 | img_request->write_request = write_request; | |
1481 | if (write_request) | |
1482 | img_request->snapc = snapc; | |
1483 | else | |
1484 | img_request->snap_id = rbd_dev->spec->snap_id; | |
1485 | spin_lock_init(&img_request->completion_lock); | |
1486 | img_request->next_completion = 0; | |
1487 | img_request->callback = NULL; | |
1488 | img_request->obj_request_count = 0; | |
1489 | INIT_LIST_HEAD(&img_request->obj_requests); | |
1490 | kref_init(&img_request->kref); | |
1491 | ||
1492 | rbd_img_request_get(img_request); /* Avoid a warning */ | |
1493 | rbd_img_request_put(img_request); /* TEMPORARY */ | |
1494 | ||
37206ee5 AE |
1495 | dout("%s: rbd_dev %p %s %llu/%llu -> img %p\n", __func__, rbd_dev, |
1496 | write_request ? "write" : "read", offset, length, | |
1497 | img_request); | |
1498 | ||
bf0d5f50 AE |
1499 | return img_request; |
1500 | } | |
1501 | ||
1502 | static void rbd_img_request_destroy(struct kref *kref) | |
1503 | { | |
1504 | struct rbd_img_request *img_request; | |
1505 | struct rbd_obj_request *obj_request; | |
1506 | struct rbd_obj_request *next_obj_request; | |
1507 | ||
1508 | img_request = container_of(kref, struct rbd_img_request, kref); | |
1509 | ||
37206ee5 AE |
1510 | dout("%s: img %p\n", __func__, img_request); |
1511 | ||
bf0d5f50 AE |
1512 | for_each_obj_request_safe(img_request, obj_request, next_obj_request) |
1513 | rbd_img_obj_request_del(img_request, obj_request); | |
25dcf954 | 1514 | rbd_assert(img_request->obj_request_count == 0); |
bf0d5f50 AE |
1515 | |
1516 | if (img_request->write_request) | |
1517 | ceph_put_snap_context(img_request->snapc); | |
1518 | ||
1519 | kfree(img_request); | |
1520 | } | |
1521 | ||
1522 | static int rbd_img_request_fill_bio(struct rbd_img_request *img_request, | |
1523 | struct bio *bio_list) | |
1524 | { | |
1525 | struct rbd_device *rbd_dev = img_request->rbd_dev; | |
1526 | struct rbd_obj_request *obj_request = NULL; | |
1527 | struct rbd_obj_request *next_obj_request; | |
430c28c3 | 1528 | bool write_request = img_request->write_request; |
bf0d5f50 AE |
1529 | unsigned int bio_offset; |
1530 | u64 image_offset; | |
1531 | u64 resid; | |
1532 | u16 opcode; | |
1533 | ||
37206ee5 AE |
1534 | dout("%s: img %p bio %p\n", __func__, img_request, bio_list); |
1535 | ||
430c28c3 | 1536 | opcode = write_request ? CEPH_OSD_OP_WRITE : CEPH_OSD_OP_READ; |
bf0d5f50 AE |
1537 | bio_offset = 0; |
1538 | image_offset = img_request->offset; | |
1539 | rbd_assert(image_offset == bio_list->bi_sector << SECTOR_SHIFT); | |
1540 | resid = img_request->length; | |
4dda41d3 | 1541 | rbd_assert(resid > 0); |
bf0d5f50 | 1542 | while (resid) { |
2fa12320 AE |
1543 | struct ceph_osd_request *osd_req; |
1544 | struct ceph_osd_data *osd_data; | |
bf0d5f50 AE |
1545 | const char *object_name; |
1546 | unsigned int clone_size; | |
bf0d5f50 AE |
1547 | u64 offset; |
1548 | u64 length; | |
1549 | ||
1550 | object_name = rbd_segment_name(rbd_dev, image_offset); | |
1551 | if (!object_name) | |
1552 | goto out_unwind; | |
1553 | offset = rbd_segment_offset(rbd_dev, image_offset); | |
1554 | length = rbd_segment_length(rbd_dev, image_offset, resid); | |
1555 | obj_request = rbd_obj_request_create(object_name, | |
1556 | offset, length, | |
1557 | OBJ_REQUEST_BIO); | |
1558 | kfree(object_name); /* object request has its own copy */ | |
1559 | if (!obj_request) | |
1560 | goto out_unwind; | |
1561 | ||
1562 | rbd_assert(length <= (u64) UINT_MAX); | |
1563 | clone_size = (unsigned int) length; | |
1564 | obj_request->bio_list = bio_chain_clone_range(&bio_list, | |
1565 | &bio_offset, clone_size, | |
1566 | GFP_ATOMIC); | |
1567 | if (!obj_request->bio_list) | |
1568 | goto out_partial; | |
1569 | ||
2fa12320 AE |
1570 | osd_req = rbd_osd_req_create(rbd_dev, write_request, |
1571 | obj_request); | |
1572 | if (!osd_req) | |
bf0d5f50 | 1573 | goto out_partial; |
2fa12320 | 1574 | obj_request->osd_req = osd_req; |
430c28c3 | 1575 | |
2fa12320 AE |
1576 | osd_data = write_request ? &osd_req->r_data_out |
1577 | : &osd_req->r_data_in; | |
1578 | osd_req_op_extent_init(osd_req, 0, opcode, offset, length, | |
1579 | 0, 0); | |
44cd188d AE |
1580 | ceph_osd_data_bio_init(osd_data, obj_request->bio_list, |
1581 | obj_request->length); | |
2fa12320 AE |
1582 | osd_req_op_extent_osd_data(osd_req, 0, osd_data); |
1583 | rbd_osd_req_format(obj_request, write_request); | |
430c28c3 | 1584 | |
bf0d5f50 AE |
1585 | /* status and version are initially zero-filled */ |
1586 | ||
1587 | rbd_img_obj_request_add(img_request, obj_request); | |
1588 | ||
1589 | image_offset += length; | |
1590 | resid -= length; | |
1591 | } | |
1592 | ||
1593 | return 0; | |
1594 | ||
1595 | out_partial: | |
1596 | rbd_obj_request_put(obj_request); | |
1597 | out_unwind: | |
1598 | for_each_obj_request_safe(img_request, obj_request, next_obj_request) | |
1599 | rbd_obj_request_put(obj_request); | |
1600 | ||
1601 | return -ENOMEM; | |
1602 | } | |
1603 | ||
1604 | static void rbd_img_obj_callback(struct rbd_obj_request *obj_request) | |
1605 | { | |
1606 | struct rbd_img_request *img_request; | |
1607 | u32 which = obj_request->which; | |
1608 | bool more = true; | |
1609 | ||
1610 | img_request = obj_request->img_request; | |
4dda41d3 | 1611 | |
37206ee5 | 1612 | dout("%s: img %p obj %p\n", __func__, img_request, obj_request); |
bf0d5f50 AE |
1613 | rbd_assert(img_request != NULL); |
1614 | rbd_assert(img_request->rq != NULL); | |
4dda41d3 | 1615 | rbd_assert(img_request->obj_request_count > 0); |
bf0d5f50 AE |
1616 | rbd_assert(which != BAD_WHICH); |
1617 | rbd_assert(which < img_request->obj_request_count); | |
1618 | rbd_assert(which >= img_request->next_completion); | |
1619 | ||
1620 | spin_lock_irq(&img_request->completion_lock); | |
1621 | if (which != img_request->next_completion) | |
1622 | goto out; | |
1623 | ||
1624 | for_each_obj_request_from(img_request, obj_request) { | |
1625 | unsigned int xferred; | |
1626 | int result; | |
1627 | ||
1628 | rbd_assert(more); | |
1629 | rbd_assert(which < img_request->obj_request_count); | |
1630 | ||
07741308 | 1631 | if (!obj_request_done_test(obj_request)) |
bf0d5f50 AE |
1632 | break; |
1633 | ||
1634 | rbd_assert(obj_request->xferred <= (u64) UINT_MAX); | |
1635 | xferred = (unsigned int) obj_request->xferred; | |
1636 | result = (int) obj_request->result; | |
1637 | if (result) | |
1638 | rbd_warn(NULL, "obj_request %s result %d xferred %u\n", | |
1639 | img_request->write_request ? "write" : "read", | |
1640 | result, xferred); | |
1641 | ||
1642 | more = blk_end_request(img_request->rq, result, xferred); | |
1643 | which++; | |
1644 | } | |
1b83bef2 | 1645 | |
bf0d5f50 AE |
1646 | rbd_assert(more ^ (which == img_request->obj_request_count)); |
1647 | img_request->next_completion = which; | |
1648 | out: | |
1649 | spin_unlock_irq(&img_request->completion_lock); | |
1650 | ||
1651 | if (!more) | |
1652 | rbd_img_request_complete(img_request); | |
1653 | } | |
1654 | ||
1655 | static int rbd_img_request_submit(struct rbd_img_request *img_request) | |
1656 | { | |
1657 | struct rbd_device *rbd_dev = img_request->rbd_dev; | |
1658 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; | |
1659 | struct rbd_obj_request *obj_request; | |
46faeed4 | 1660 | struct rbd_obj_request *next_obj_request; |
bf0d5f50 | 1661 | |
37206ee5 | 1662 | dout("%s: img %p\n", __func__, img_request); |
46faeed4 | 1663 | for_each_obj_request_safe(img_request, obj_request, next_obj_request) { |
bf0d5f50 AE |
1664 | int ret; |
1665 | ||
1666 | obj_request->callback = rbd_img_obj_callback; | |
1667 | ret = rbd_obj_request_submit(osdc, obj_request); | |
1668 | if (ret) | |
1669 | return ret; | |
1670 | /* | |
1671 | * The image request has its own reference to each | |
1672 | * of its object requests, so we can safely drop the | |
1673 | * initial one here. | |
1674 | */ | |
1675 | rbd_obj_request_put(obj_request); | |
1676 | } | |
1677 | ||
1678 | return 0; | |
1679 | } | |
1680 | ||
cf81b60e | 1681 | static int rbd_obj_notify_ack(struct rbd_device *rbd_dev, |
b8d70035 AE |
1682 | u64 ver, u64 notify_id) |
1683 | { | |
1684 | struct rbd_obj_request *obj_request; | |
b8d70035 AE |
1685 | struct ceph_osd_client *osdc; |
1686 | int ret; | |
1687 | ||
1688 | obj_request = rbd_obj_request_create(rbd_dev->header_name, 0, 0, | |
1689 | OBJ_REQUEST_NODATA); | |
1690 | if (!obj_request) | |
1691 | return -ENOMEM; | |
1692 | ||
1693 | ret = -ENOMEM; | |
430c28c3 | 1694 | obj_request->osd_req = rbd_osd_req_create(rbd_dev, false, obj_request); |
b8d70035 AE |
1695 | if (!obj_request->osd_req) |
1696 | goto out; | |
1697 | ||
c99d2d4a AE |
1698 | osd_req_op_watch_init(obj_request->osd_req, 0, CEPH_OSD_OP_NOTIFY_ACK, |
1699 | notify_id, ver, 0); | |
2fa12320 | 1700 | rbd_osd_req_format(obj_request, false); |
430c28c3 | 1701 | |
b8d70035 | 1702 | osdc = &rbd_dev->rbd_client->client->osdc; |
cf81b60e | 1703 | obj_request->callback = rbd_obj_request_put; |
b8d70035 | 1704 | ret = rbd_obj_request_submit(osdc, obj_request); |
b8d70035 | 1705 | out: |
cf81b60e AE |
1706 | if (ret) |
1707 | rbd_obj_request_put(obj_request); | |
b8d70035 AE |
1708 | |
1709 | return ret; | |
1710 | } | |
1711 | ||
1712 | static void rbd_watch_cb(u64 ver, u64 notify_id, u8 opcode, void *data) | |
1713 | { | |
1714 | struct rbd_device *rbd_dev = (struct rbd_device *)data; | |
1715 | u64 hver; | |
1716 | int rc; | |
1717 | ||
1718 | if (!rbd_dev) | |
1719 | return; | |
1720 | ||
37206ee5 | 1721 | dout("%s: \"%s\" notify_id %llu opcode %u\n", __func__, |
b8d70035 AE |
1722 | rbd_dev->header_name, (unsigned long long) notify_id, |
1723 | (unsigned int) opcode); | |
1724 | rc = rbd_dev_refresh(rbd_dev, &hver); | |
1725 | if (rc) | |
1726 | rbd_warn(rbd_dev, "got notification but failed to " | |
1727 | " update snaps: %d\n", rc); | |
1728 | ||
cf81b60e | 1729 | rbd_obj_notify_ack(rbd_dev, hver, notify_id); |
b8d70035 AE |
1730 | } |
1731 | ||
9969ebc5 AE |
1732 | /* |
1733 | * Request sync osd watch/unwatch. The value of "start" determines | |
1734 | * whether a watch request is being initiated or torn down. | |
1735 | */ | |
1736 | static int rbd_dev_header_watch_sync(struct rbd_device *rbd_dev, int start) | |
1737 | { | |
1738 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; | |
1739 | struct rbd_obj_request *obj_request; | |
9969ebc5 AE |
1740 | int ret; |
1741 | ||
1742 | rbd_assert(start ^ !!rbd_dev->watch_event); | |
1743 | rbd_assert(start ^ !!rbd_dev->watch_request); | |
1744 | ||
1745 | if (start) { | |
3c663bbd | 1746 | ret = ceph_osdc_create_event(osdc, rbd_watch_cb, rbd_dev, |
9969ebc5 AE |
1747 | &rbd_dev->watch_event); |
1748 | if (ret < 0) | |
1749 | return ret; | |
8eb87565 | 1750 | rbd_assert(rbd_dev->watch_event != NULL); |
9969ebc5 AE |
1751 | } |
1752 | ||
1753 | ret = -ENOMEM; | |
1754 | obj_request = rbd_obj_request_create(rbd_dev->header_name, 0, 0, | |
1755 | OBJ_REQUEST_NODATA); | |
1756 | if (!obj_request) | |
1757 | goto out_cancel; | |
1758 | ||
430c28c3 AE |
1759 | obj_request->osd_req = rbd_osd_req_create(rbd_dev, true, obj_request); |
1760 | if (!obj_request->osd_req) | |
1761 | goto out_cancel; | |
1762 | ||
c99d2d4a | 1763 | osd_req_op_watch_init(obj_request->osd_req, 0, CEPH_OSD_OP_WATCH, |
9969ebc5 AE |
1764 | rbd_dev->watch_event->cookie, |
1765 | rbd_dev->header.obj_version, start); | |
2fa12320 | 1766 | rbd_osd_req_format(obj_request, true); |
9969ebc5 | 1767 | |
8eb87565 | 1768 | if (start) |
975241af | 1769 | ceph_osdc_set_request_linger(osdc, obj_request->osd_req); |
8eb87565 | 1770 | else |
6977c3f9 | 1771 | ceph_osdc_unregister_linger_request(osdc, |
975241af | 1772 | rbd_dev->watch_request->osd_req); |
9969ebc5 AE |
1773 | ret = rbd_obj_request_submit(osdc, obj_request); |
1774 | if (ret) | |
1775 | goto out_cancel; | |
1776 | ret = rbd_obj_request_wait(obj_request); | |
1777 | if (ret) | |
1778 | goto out_cancel; | |
9969ebc5 AE |
1779 | ret = obj_request->result; |
1780 | if (ret) | |
1781 | goto out_cancel; | |
1782 | ||
8eb87565 AE |
1783 | /* |
1784 | * A watch request is set to linger, so the underlying osd | |
1785 | * request won't go away until we unregister it. We retain | |
1786 | * a pointer to the object request during that time (in | |
1787 | * rbd_dev->watch_request), so we'll keep a reference to | |
1788 | * it. We'll drop that reference (below) after we've | |
1789 | * unregistered it. | |
1790 | */ | |
1791 | if (start) { | |
1792 | rbd_dev->watch_request = obj_request; | |
1793 | ||
1794 | return 0; | |
1795 | } | |
1796 | ||
1797 | /* We have successfully torn down the watch request */ | |
1798 | ||
1799 | rbd_obj_request_put(rbd_dev->watch_request); | |
1800 | rbd_dev->watch_request = NULL; | |
9969ebc5 AE |
1801 | out_cancel: |
1802 | /* Cancel the event if we're tearing down, or on error */ | |
1803 | ceph_osdc_cancel_event(rbd_dev->watch_event); | |
1804 | rbd_dev->watch_event = NULL; | |
9969ebc5 AE |
1805 | if (obj_request) |
1806 | rbd_obj_request_put(obj_request); | |
1807 | ||
1808 | return ret; | |
1809 | } | |
1810 | ||
36be9a76 AE |
1811 | /* |
1812 | * Synchronous osd object method call | |
1813 | */ | |
1814 | static int rbd_obj_method_sync(struct rbd_device *rbd_dev, | |
1815 | const char *object_name, | |
1816 | const char *class_name, | |
1817 | const char *method_name, | |
1818 | const char *outbound, | |
1819 | size_t outbound_size, | |
1820 | char *inbound, | |
1821 | size_t inbound_size, | |
1822 | u64 *version) | |
1823 | { | |
1824 | struct rbd_obj_request *obj_request; | |
2fa12320 | 1825 | struct ceph_osd_data *osd_data; |
36be9a76 | 1826 | struct ceph_osd_client *osdc; |
36be9a76 AE |
1827 | struct page **pages; |
1828 | u32 page_count; | |
1829 | int ret; | |
1830 | ||
1831 | /* | |
6010a451 AE |
1832 | * Method calls are ultimately read operations. The result |
1833 | * should placed into the inbound buffer provided. They | |
1834 | * also supply outbound data--parameters for the object | |
1835 | * method. Currently if this is present it will be a | |
1836 | * snapshot id. | |
36be9a76 AE |
1837 | */ |
1838 | page_count = (u32) calc_pages_for(0, inbound_size); | |
1839 | pages = ceph_alloc_page_vector(page_count, GFP_KERNEL); | |
1840 | if (IS_ERR(pages)) | |
1841 | return PTR_ERR(pages); | |
1842 | ||
1843 | ret = -ENOMEM; | |
6010a451 | 1844 | obj_request = rbd_obj_request_create(object_name, 0, inbound_size, |
36be9a76 AE |
1845 | OBJ_REQUEST_PAGES); |
1846 | if (!obj_request) | |
1847 | goto out; | |
1848 | ||
1849 | obj_request->pages = pages; | |
1850 | obj_request->page_count = page_count; | |
1851 | ||
430c28c3 | 1852 | obj_request->osd_req = rbd_osd_req_create(rbd_dev, false, obj_request); |
36be9a76 AE |
1853 | if (!obj_request->osd_req) |
1854 | goto out; | |
1855 | ||
2fa12320 | 1856 | osd_data = &obj_request->osd_req->r_data_in; |
c99d2d4a AE |
1857 | osd_req_op_cls_init(obj_request->osd_req, 0, CEPH_OSD_OP_CALL, |
1858 | class_name, method_name, | |
430c28c3 | 1859 | outbound, outbound_size); |
44cd188d AE |
1860 | ceph_osd_data_pages_init(osd_data, obj_request->pages, inbound_size, |
1861 | 0, false, false); | |
2fa12320 AE |
1862 | osd_req_op_cls_response_data(obj_request->osd_req, 0, osd_data); |
1863 | rbd_osd_req_format(obj_request, false); | |
430c28c3 | 1864 | |
36be9a76 AE |
1865 | osdc = &rbd_dev->rbd_client->client->osdc; |
1866 | ret = rbd_obj_request_submit(osdc, obj_request); | |
1867 | if (ret) | |
1868 | goto out; | |
1869 | ret = rbd_obj_request_wait(obj_request); | |
1870 | if (ret) | |
1871 | goto out; | |
1872 | ||
1873 | ret = obj_request->result; | |
1874 | if (ret < 0) | |
1875 | goto out; | |
23ed6e13 | 1876 | ret = 0; |
903bb32e | 1877 | ceph_copy_from_page_vector(pages, inbound, 0, obj_request->xferred); |
36be9a76 AE |
1878 | if (version) |
1879 | *version = obj_request->version; | |
1880 | out: | |
1881 | if (obj_request) | |
1882 | rbd_obj_request_put(obj_request); | |
1883 | else | |
1884 | ceph_release_page_vector(pages, page_count); | |
1885 | ||
1886 | return ret; | |
1887 | } | |
1888 | ||
bf0d5f50 | 1889 | static void rbd_request_fn(struct request_queue *q) |
cc344fa1 | 1890 | __releases(q->queue_lock) __acquires(q->queue_lock) |
bf0d5f50 AE |
1891 | { |
1892 | struct rbd_device *rbd_dev = q->queuedata; | |
1893 | bool read_only = rbd_dev->mapping.read_only; | |
1894 | struct request *rq; | |
1895 | int result; | |
1896 | ||
1897 | while ((rq = blk_fetch_request(q))) { | |
1898 | bool write_request = rq_data_dir(rq) == WRITE; | |
1899 | struct rbd_img_request *img_request; | |
1900 | u64 offset; | |
1901 | u64 length; | |
1902 | ||
1903 | /* Ignore any non-FS requests that filter through. */ | |
1904 | ||
1905 | if (rq->cmd_type != REQ_TYPE_FS) { | |
4dda41d3 AE |
1906 | dout("%s: non-fs request type %d\n", __func__, |
1907 | (int) rq->cmd_type); | |
1908 | __blk_end_request_all(rq, 0); | |
1909 | continue; | |
1910 | } | |
1911 | ||
1912 | /* Ignore/skip any zero-length requests */ | |
1913 | ||
1914 | offset = (u64) blk_rq_pos(rq) << SECTOR_SHIFT; | |
1915 | length = (u64) blk_rq_bytes(rq); | |
1916 | ||
1917 | if (!length) { | |
1918 | dout("%s: zero-length request\n", __func__); | |
bf0d5f50 AE |
1919 | __blk_end_request_all(rq, 0); |
1920 | continue; | |
1921 | } | |
1922 | ||
1923 | spin_unlock_irq(q->queue_lock); | |
1924 | ||
1925 | /* Disallow writes to a read-only device */ | |
1926 | ||
1927 | if (write_request) { | |
1928 | result = -EROFS; | |
1929 | if (read_only) | |
1930 | goto end_request; | |
1931 | rbd_assert(rbd_dev->spec->snap_id == CEPH_NOSNAP); | |
1932 | } | |
1933 | ||
6d292906 AE |
1934 | /* |
1935 | * Quit early if the mapped snapshot no longer | |
1936 | * exists. It's still possible the snapshot will | |
1937 | * have disappeared by the time our request arrives | |
1938 | * at the osd, but there's no sense in sending it if | |
1939 | * we already know. | |
1940 | */ | |
1941 | if (!test_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags)) { | |
bf0d5f50 AE |
1942 | dout("request for non-existent snapshot"); |
1943 | rbd_assert(rbd_dev->spec->snap_id != CEPH_NOSNAP); | |
1944 | result = -ENXIO; | |
1945 | goto end_request; | |
1946 | } | |
1947 | ||
bf0d5f50 AE |
1948 | result = -EINVAL; |
1949 | if (WARN_ON(offset && length > U64_MAX - offset + 1)) | |
1950 | goto end_request; /* Shouldn't happen */ | |
1951 | ||
1952 | result = -ENOMEM; | |
1953 | img_request = rbd_img_request_create(rbd_dev, offset, length, | |
1954 | write_request); | |
1955 | if (!img_request) | |
1956 | goto end_request; | |
1957 | ||
1958 | img_request->rq = rq; | |
1959 | ||
1960 | result = rbd_img_request_fill_bio(img_request, rq->bio); | |
1961 | if (!result) | |
1962 | result = rbd_img_request_submit(img_request); | |
1963 | if (result) | |
1964 | rbd_img_request_put(img_request); | |
1965 | end_request: | |
1966 | spin_lock_irq(q->queue_lock); | |
1967 | if (result < 0) { | |
1968 | rbd_warn(rbd_dev, "obj_request %s result %d\n", | |
1969 | write_request ? "write" : "read", result); | |
1970 | __blk_end_request_all(rq, result); | |
1971 | } | |
1972 | } | |
1973 | } | |
1974 | ||
602adf40 YS |
1975 | /* |
1976 | * a queue callback. Makes sure that we don't create a bio that spans across | |
1977 | * multiple osd objects. One exception would be with a single page bios, | |
f7760dad | 1978 | * which we handle later at bio_chain_clone_range() |
602adf40 YS |
1979 | */ |
1980 | static int rbd_merge_bvec(struct request_queue *q, struct bvec_merge_data *bmd, | |
1981 | struct bio_vec *bvec) | |
1982 | { | |
1983 | struct rbd_device *rbd_dev = q->queuedata; | |
e5cfeed2 AE |
1984 | sector_t sector_offset; |
1985 | sector_t sectors_per_obj; | |
1986 | sector_t obj_sector_offset; | |
1987 | int ret; | |
1988 | ||
1989 | /* | |
1990 | * Find how far into its rbd object the partition-relative | |
1991 | * bio start sector is to offset relative to the enclosing | |
1992 | * device. | |
1993 | */ | |
1994 | sector_offset = get_start_sect(bmd->bi_bdev) + bmd->bi_sector; | |
1995 | sectors_per_obj = 1 << (rbd_dev->header.obj_order - SECTOR_SHIFT); | |
1996 | obj_sector_offset = sector_offset & (sectors_per_obj - 1); | |
1997 | ||
1998 | /* | |
1999 | * Compute the number of bytes from that offset to the end | |
2000 | * of the object. Account for what's already used by the bio. | |
2001 | */ | |
2002 | ret = (int) (sectors_per_obj - obj_sector_offset) << SECTOR_SHIFT; | |
2003 | if (ret > bmd->bi_size) | |
2004 | ret -= bmd->bi_size; | |
2005 | else | |
2006 | ret = 0; | |
2007 | ||
2008 | /* | |
2009 | * Don't send back more than was asked for. And if the bio | |
2010 | * was empty, let the whole thing through because: "Note | |
2011 | * that a block device *must* allow a single page to be | |
2012 | * added to an empty bio." | |
2013 | */ | |
2014 | rbd_assert(bvec->bv_len <= PAGE_SIZE); | |
2015 | if (ret > (int) bvec->bv_len || !bmd->bi_size) | |
2016 | ret = (int) bvec->bv_len; | |
2017 | ||
2018 | return ret; | |
602adf40 YS |
2019 | } |
2020 | ||
2021 | static void rbd_free_disk(struct rbd_device *rbd_dev) | |
2022 | { | |
2023 | struct gendisk *disk = rbd_dev->disk; | |
2024 | ||
2025 | if (!disk) | |
2026 | return; | |
2027 | ||
602adf40 YS |
2028 | if (disk->flags & GENHD_FL_UP) |
2029 | del_gendisk(disk); | |
2030 | if (disk->queue) | |
2031 | blk_cleanup_queue(disk->queue); | |
2032 | put_disk(disk); | |
2033 | } | |
2034 | ||
788e2df3 AE |
2035 | static int rbd_obj_read_sync(struct rbd_device *rbd_dev, |
2036 | const char *object_name, | |
2037 | u64 offset, u64 length, | |
2038 | char *buf, u64 *version) | |
2039 | ||
2040 | { | |
788e2df3 | 2041 | struct rbd_obj_request *obj_request; |
2fa12320 | 2042 | struct ceph_osd_data *osd_data; |
788e2df3 AE |
2043 | struct ceph_osd_client *osdc; |
2044 | struct page **pages = NULL; | |
2045 | u32 page_count; | |
1ceae7ef | 2046 | size_t size; |
788e2df3 AE |
2047 | int ret; |
2048 | ||
2049 | page_count = (u32) calc_pages_for(offset, length); | |
2050 | pages = ceph_alloc_page_vector(page_count, GFP_KERNEL); | |
2051 | if (IS_ERR(pages)) | |
2052 | ret = PTR_ERR(pages); | |
2053 | ||
2054 | ret = -ENOMEM; | |
2055 | obj_request = rbd_obj_request_create(object_name, offset, length, | |
36be9a76 | 2056 | OBJ_REQUEST_PAGES); |
788e2df3 AE |
2057 | if (!obj_request) |
2058 | goto out; | |
2059 | ||
2060 | obj_request->pages = pages; | |
2061 | obj_request->page_count = page_count; | |
2062 | ||
430c28c3 | 2063 | obj_request->osd_req = rbd_osd_req_create(rbd_dev, false, obj_request); |
788e2df3 AE |
2064 | if (!obj_request->osd_req) |
2065 | goto out; | |
2066 | ||
2fa12320 | 2067 | osd_data = &obj_request->osd_req->r_data_in; |
c99d2d4a AE |
2068 | osd_req_op_extent_init(obj_request->osd_req, 0, CEPH_OSD_OP_READ, |
2069 | offset, length, 0, 0); | |
44cd188d AE |
2070 | ceph_osd_data_pages_init(osd_data, obj_request->pages, |
2071 | obj_request->length, | |
2072 | obj_request->offset & ~PAGE_MASK, | |
2073 | false, false); | |
2fa12320 AE |
2074 | osd_req_op_extent_osd_data(obj_request->osd_req, 0, osd_data); |
2075 | rbd_osd_req_format(obj_request, false); | |
430c28c3 | 2076 | |
788e2df3 AE |
2077 | osdc = &rbd_dev->rbd_client->client->osdc; |
2078 | ret = rbd_obj_request_submit(osdc, obj_request); | |
2079 | if (ret) | |
2080 | goto out; | |
2081 | ret = rbd_obj_request_wait(obj_request); | |
2082 | if (ret) | |
2083 | goto out; | |
2084 | ||
2085 | ret = obj_request->result; | |
2086 | if (ret < 0) | |
2087 | goto out; | |
1ceae7ef AE |
2088 | |
2089 | rbd_assert(obj_request->xferred <= (u64) SIZE_MAX); | |
2090 | size = (size_t) obj_request->xferred; | |
903bb32e | 2091 | ceph_copy_from_page_vector(pages, buf, 0, size); |
23ed6e13 AE |
2092 | rbd_assert(size <= (size_t) INT_MAX); |
2093 | ret = (int) size; | |
788e2df3 AE |
2094 | if (version) |
2095 | *version = obj_request->version; | |
2096 | out: | |
2097 | if (obj_request) | |
2098 | rbd_obj_request_put(obj_request); | |
2099 | else | |
2100 | ceph_release_page_vector(pages, page_count); | |
2101 | ||
2102 | return ret; | |
2103 | } | |
2104 | ||
602adf40 | 2105 | /* |
4156d998 AE |
2106 | * Read the complete header for the given rbd device. |
2107 | * | |
2108 | * Returns a pointer to a dynamically-allocated buffer containing | |
2109 | * the complete and validated header. Caller can pass the address | |
2110 | * of a variable that will be filled in with the version of the | |
2111 | * header object at the time it was read. | |
2112 | * | |
2113 | * Returns a pointer-coded errno if a failure occurs. | |
602adf40 | 2114 | */ |
4156d998 AE |
2115 | static struct rbd_image_header_ondisk * |
2116 | rbd_dev_v1_header_read(struct rbd_device *rbd_dev, u64 *version) | |
602adf40 | 2117 | { |
4156d998 | 2118 | struct rbd_image_header_ondisk *ondisk = NULL; |
50f7c4c9 | 2119 | u32 snap_count = 0; |
4156d998 AE |
2120 | u64 names_size = 0; |
2121 | u32 want_count; | |
2122 | int ret; | |
602adf40 | 2123 | |
00f1f36f | 2124 | /* |
4156d998 AE |
2125 | * The complete header will include an array of its 64-bit |
2126 | * snapshot ids, followed by the names of those snapshots as | |
2127 | * a contiguous block of NUL-terminated strings. Note that | |
2128 | * the number of snapshots could change by the time we read | |
2129 | * it in, in which case we re-read it. | |
00f1f36f | 2130 | */ |
4156d998 AE |
2131 | do { |
2132 | size_t size; | |
2133 | ||
2134 | kfree(ondisk); | |
2135 | ||
2136 | size = sizeof (*ondisk); | |
2137 | size += snap_count * sizeof (struct rbd_image_snap_ondisk); | |
2138 | size += names_size; | |
2139 | ondisk = kmalloc(size, GFP_KERNEL); | |
2140 | if (!ondisk) | |
2141 | return ERR_PTR(-ENOMEM); | |
2142 | ||
788e2df3 | 2143 | ret = rbd_obj_read_sync(rbd_dev, rbd_dev->header_name, |
4156d998 AE |
2144 | 0, size, |
2145 | (char *) ondisk, version); | |
4156d998 AE |
2146 | if (ret < 0) |
2147 | goto out_err; | |
2148 | if (WARN_ON((size_t) ret < size)) { | |
2149 | ret = -ENXIO; | |
06ecc6cb AE |
2150 | rbd_warn(rbd_dev, "short header read (want %zd got %d)", |
2151 | size, ret); | |
4156d998 AE |
2152 | goto out_err; |
2153 | } | |
2154 | if (!rbd_dev_ondisk_valid(ondisk)) { | |
2155 | ret = -ENXIO; | |
06ecc6cb | 2156 | rbd_warn(rbd_dev, "invalid header"); |
4156d998 | 2157 | goto out_err; |
81e759fb | 2158 | } |
602adf40 | 2159 | |
4156d998 AE |
2160 | names_size = le64_to_cpu(ondisk->snap_names_len); |
2161 | want_count = snap_count; | |
2162 | snap_count = le32_to_cpu(ondisk->snap_count); | |
2163 | } while (snap_count != want_count); | |
00f1f36f | 2164 | |
4156d998 | 2165 | return ondisk; |
00f1f36f | 2166 | |
4156d998 AE |
2167 | out_err: |
2168 | kfree(ondisk); | |
2169 | ||
2170 | return ERR_PTR(ret); | |
2171 | } | |
2172 | ||
2173 | /* | |
2174 | * reload the ondisk the header | |
2175 | */ | |
2176 | static int rbd_read_header(struct rbd_device *rbd_dev, | |
2177 | struct rbd_image_header *header) | |
2178 | { | |
2179 | struct rbd_image_header_ondisk *ondisk; | |
2180 | u64 ver = 0; | |
2181 | int ret; | |
602adf40 | 2182 | |
4156d998 AE |
2183 | ondisk = rbd_dev_v1_header_read(rbd_dev, &ver); |
2184 | if (IS_ERR(ondisk)) | |
2185 | return PTR_ERR(ondisk); | |
2186 | ret = rbd_header_from_disk(header, ondisk); | |
2187 | if (ret >= 0) | |
2188 | header->obj_version = ver; | |
2189 | kfree(ondisk); | |
2190 | ||
2191 | return ret; | |
602adf40 YS |
2192 | } |
2193 | ||
41f38c2b | 2194 | static void rbd_remove_all_snaps(struct rbd_device *rbd_dev) |
dfc5606d YS |
2195 | { |
2196 | struct rbd_snap *snap; | |
a0593290 | 2197 | struct rbd_snap *next; |
dfc5606d | 2198 | |
a0593290 | 2199 | list_for_each_entry_safe(snap, next, &rbd_dev->snaps, node) |
41f38c2b | 2200 | rbd_remove_snap_dev(snap); |
dfc5606d YS |
2201 | } |
2202 | ||
9478554a AE |
2203 | static void rbd_update_mapping_size(struct rbd_device *rbd_dev) |
2204 | { | |
2205 | sector_t size; | |
2206 | ||
0d7dbfce | 2207 | if (rbd_dev->spec->snap_id != CEPH_NOSNAP) |
9478554a AE |
2208 | return; |
2209 | ||
2210 | size = (sector_t) rbd_dev->header.image_size / SECTOR_SIZE; | |
2211 | dout("setting size to %llu sectors", (unsigned long long) size); | |
2212 | rbd_dev->mapping.size = (u64) size; | |
2213 | set_capacity(rbd_dev->disk, size); | |
2214 | } | |
2215 | ||
602adf40 YS |
2216 | /* |
2217 | * only read the first part of the ondisk header, without the snaps info | |
2218 | */ | |
117973fb | 2219 | static int rbd_dev_v1_refresh(struct rbd_device *rbd_dev, u64 *hver) |
602adf40 YS |
2220 | { |
2221 | int ret; | |
2222 | struct rbd_image_header h; | |
602adf40 YS |
2223 | |
2224 | ret = rbd_read_header(rbd_dev, &h); | |
2225 | if (ret < 0) | |
2226 | return ret; | |
2227 | ||
a51aa0c0 JD |
2228 | down_write(&rbd_dev->header_rwsem); |
2229 | ||
9478554a AE |
2230 | /* Update image size, and check for resize of mapped image */ |
2231 | rbd_dev->header.image_size = h.image_size; | |
2232 | rbd_update_mapping_size(rbd_dev); | |
9db4b3e3 | 2233 | |
849b4260 | 2234 | /* rbd_dev->header.object_prefix shouldn't change */ |
602adf40 | 2235 | kfree(rbd_dev->header.snap_sizes); |
849b4260 | 2236 | kfree(rbd_dev->header.snap_names); |
d1d25646 JD |
2237 | /* osd requests may still refer to snapc */ |
2238 | ceph_put_snap_context(rbd_dev->header.snapc); | |
602adf40 | 2239 | |
b813623a AE |
2240 | if (hver) |
2241 | *hver = h.obj_version; | |
a71b891b | 2242 | rbd_dev->header.obj_version = h.obj_version; |
93a24e08 | 2243 | rbd_dev->header.image_size = h.image_size; |
602adf40 YS |
2244 | rbd_dev->header.snapc = h.snapc; |
2245 | rbd_dev->header.snap_names = h.snap_names; | |
2246 | rbd_dev->header.snap_sizes = h.snap_sizes; | |
849b4260 AE |
2247 | /* Free the extra copy of the object prefix */ |
2248 | WARN_ON(strcmp(rbd_dev->header.object_prefix, h.object_prefix)); | |
2249 | kfree(h.object_prefix); | |
2250 | ||
304f6808 AE |
2251 | ret = rbd_dev_snaps_update(rbd_dev); |
2252 | if (!ret) | |
2253 | ret = rbd_dev_snaps_register(rbd_dev); | |
dfc5606d | 2254 | |
c666601a | 2255 | up_write(&rbd_dev->header_rwsem); |
602adf40 | 2256 | |
dfc5606d | 2257 | return ret; |
602adf40 YS |
2258 | } |
2259 | ||
117973fb | 2260 | static int rbd_dev_refresh(struct rbd_device *rbd_dev, u64 *hver) |
1fe5e993 AE |
2261 | { |
2262 | int ret; | |
2263 | ||
117973fb | 2264 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); |
1fe5e993 | 2265 | mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING); |
117973fb AE |
2266 | if (rbd_dev->image_format == 1) |
2267 | ret = rbd_dev_v1_refresh(rbd_dev, hver); | |
2268 | else | |
2269 | ret = rbd_dev_v2_refresh(rbd_dev, hver); | |
1fe5e993 AE |
2270 | mutex_unlock(&ctl_mutex); |
2271 | ||
2272 | return ret; | |
2273 | } | |
2274 | ||
602adf40 YS |
2275 | static int rbd_init_disk(struct rbd_device *rbd_dev) |
2276 | { | |
2277 | struct gendisk *disk; | |
2278 | struct request_queue *q; | |
593a9e7b | 2279 | u64 segment_size; |
602adf40 | 2280 | |
602adf40 | 2281 | /* create gendisk info */ |
602adf40 YS |
2282 | disk = alloc_disk(RBD_MINORS_PER_MAJOR); |
2283 | if (!disk) | |
1fcdb8aa | 2284 | return -ENOMEM; |
602adf40 | 2285 | |
f0f8cef5 | 2286 | snprintf(disk->disk_name, sizeof(disk->disk_name), RBD_DRV_NAME "%d", |
de71a297 | 2287 | rbd_dev->dev_id); |
602adf40 YS |
2288 | disk->major = rbd_dev->major; |
2289 | disk->first_minor = 0; | |
2290 | disk->fops = &rbd_bd_ops; | |
2291 | disk->private_data = rbd_dev; | |
2292 | ||
bf0d5f50 | 2293 | q = blk_init_queue(rbd_request_fn, &rbd_dev->lock); |
602adf40 YS |
2294 | if (!q) |
2295 | goto out_disk; | |
029bcbd8 | 2296 | |
593a9e7b AE |
2297 | /* We use the default size, but let's be explicit about it. */ |
2298 | blk_queue_physical_block_size(q, SECTOR_SIZE); | |
2299 | ||
029bcbd8 | 2300 | /* set io sizes to object size */ |
593a9e7b AE |
2301 | segment_size = rbd_obj_bytes(&rbd_dev->header); |
2302 | blk_queue_max_hw_sectors(q, segment_size / SECTOR_SIZE); | |
2303 | blk_queue_max_segment_size(q, segment_size); | |
2304 | blk_queue_io_min(q, segment_size); | |
2305 | blk_queue_io_opt(q, segment_size); | |
029bcbd8 | 2306 | |
602adf40 YS |
2307 | blk_queue_merge_bvec(q, rbd_merge_bvec); |
2308 | disk->queue = q; | |
2309 | ||
2310 | q->queuedata = rbd_dev; | |
2311 | ||
2312 | rbd_dev->disk = disk; | |
602adf40 | 2313 | |
12f02944 AE |
2314 | set_capacity(rbd_dev->disk, rbd_dev->mapping.size / SECTOR_SIZE); |
2315 | ||
602adf40 | 2316 | return 0; |
602adf40 YS |
2317 | out_disk: |
2318 | put_disk(disk); | |
1fcdb8aa AE |
2319 | |
2320 | return -ENOMEM; | |
602adf40 YS |
2321 | } |
2322 | ||
dfc5606d YS |
2323 | /* |
2324 | sysfs | |
2325 | */ | |
2326 | ||
593a9e7b AE |
2327 | static struct rbd_device *dev_to_rbd_dev(struct device *dev) |
2328 | { | |
2329 | return container_of(dev, struct rbd_device, dev); | |
2330 | } | |
2331 | ||
dfc5606d YS |
2332 | static ssize_t rbd_size_show(struct device *dev, |
2333 | struct device_attribute *attr, char *buf) | |
2334 | { | |
593a9e7b | 2335 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
a51aa0c0 JD |
2336 | sector_t size; |
2337 | ||
2338 | down_read(&rbd_dev->header_rwsem); | |
2339 | size = get_capacity(rbd_dev->disk); | |
2340 | up_read(&rbd_dev->header_rwsem); | |
dfc5606d | 2341 | |
a51aa0c0 | 2342 | return sprintf(buf, "%llu\n", (unsigned long long) size * SECTOR_SIZE); |
dfc5606d YS |
2343 | } |
2344 | ||
34b13184 AE |
2345 | /* |
2346 | * Note this shows the features for whatever's mapped, which is not | |
2347 | * necessarily the base image. | |
2348 | */ | |
2349 | static ssize_t rbd_features_show(struct device *dev, | |
2350 | struct device_attribute *attr, char *buf) | |
2351 | { | |
2352 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
2353 | ||
2354 | return sprintf(buf, "0x%016llx\n", | |
2355 | (unsigned long long) rbd_dev->mapping.features); | |
2356 | } | |
2357 | ||
dfc5606d YS |
2358 | static ssize_t rbd_major_show(struct device *dev, |
2359 | struct device_attribute *attr, char *buf) | |
2360 | { | |
593a9e7b | 2361 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
602adf40 | 2362 | |
dfc5606d YS |
2363 | return sprintf(buf, "%d\n", rbd_dev->major); |
2364 | } | |
2365 | ||
2366 | static ssize_t rbd_client_id_show(struct device *dev, | |
2367 | struct device_attribute *attr, char *buf) | |
602adf40 | 2368 | { |
593a9e7b | 2369 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 2370 | |
1dbb4399 AE |
2371 | return sprintf(buf, "client%lld\n", |
2372 | ceph_client_id(rbd_dev->rbd_client->client)); | |
602adf40 YS |
2373 | } |
2374 | ||
dfc5606d YS |
2375 | static ssize_t rbd_pool_show(struct device *dev, |
2376 | struct device_attribute *attr, char *buf) | |
602adf40 | 2377 | { |
593a9e7b | 2378 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 2379 | |
0d7dbfce | 2380 | return sprintf(buf, "%s\n", rbd_dev->spec->pool_name); |
dfc5606d YS |
2381 | } |
2382 | ||
9bb2f334 AE |
2383 | static ssize_t rbd_pool_id_show(struct device *dev, |
2384 | struct device_attribute *attr, char *buf) | |
2385 | { | |
2386 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
2387 | ||
0d7dbfce AE |
2388 | return sprintf(buf, "%llu\n", |
2389 | (unsigned long long) rbd_dev->spec->pool_id); | |
9bb2f334 AE |
2390 | } |
2391 | ||
dfc5606d YS |
2392 | static ssize_t rbd_name_show(struct device *dev, |
2393 | struct device_attribute *attr, char *buf) | |
2394 | { | |
593a9e7b | 2395 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 2396 | |
a92ffdf8 AE |
2397 | if (rbd_dev->spec->image_name) |
2398 | return sprintf(buf, "%s\n", rbd_dev->spec->image_name); | |
2399 | ||
2400 | return sprintf(buf, "(unknown)\n"); | |
dfc5606d YS |
2401 | } |
2402 | ||
589d30e0 AE |
2403 | static ssize_t rbd_image_id_show(struct device *dev, |
2404 | struct device_attribute *attr, char *buf) | |
2405 | { | |
2406 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
2407 | ||
0d7dbfce | 2408 | return sprintf(buf, "%s\n", rbd_dev->spec->image_id); |
589d30e0 AE |
2409 | } |
2410 | ||
34b13184 AE |
2411 | /* |
2412 | * Shows the name of the currently-mapped snapshot (or | |
2413 | * RBD_SNAP_HEAD_NAME for the base image). | |
2414 | */ | |
dfc5606d YS |
2415 | static ssize_t rbd_snap_show(struct device *dev, |
2416 | struct device_attribute *attr, | |
2417 | char *buf) | |
2418 | { | |
593a9e7b | 2419 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 2420 | |
0d7dbfce | 2421 | return sprintf(buf, "%s\n", rbd_dev->spec->snap_name); |
dfc5606d YS |
2422 | } |
2423 | ||
86b00e0d AE |
2424 | /* |
2425 | * For an rbd v2 image, shows the pool id, image id, and snapshot id | |
2426 | * for the parent image. If there is no parent, simply shows | |
2427 | * "(no parent image)". | |
2428 | */ | |
2429 | static ssize_t rbd_parent_show(struct device *dev, | |
2430 | struct device_attribute *attr, | |
2431 | char *buf) | |
2432 | { | |
2433 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
2434 | struct rbd_spec *spec = rbd_dev->parent_spec; | |
2435 | int count; | |
2436 | char *bufp = buf; | |
2437 | ||
2438 | if (!spec) | |
2439 | return sprintf(buf, "(no parent image)\n"); | |
2440 | ||
2441 | count = sprintf(bufp, "pool_id %llu\npool_name %s\n", | |
2442 | (unsigned long long) spec->pool_id, spec->pool_name); | |
2443 | if (count < 0) | |
2444 | return count; | |
2445 | bufp += count; | |
2446 | ||
2447 | count = sprintf(bufp, "image_id %s\nimage_name %s\n", spec->image_id, | |
2448 | spec->image_name ? spec->image_name : "(unknown)"); | |
2449 | if (count < 0) | |
2450 | return count; | |
2451 | bufp += count; | |
2452 | ||
2453 | count = sprintf(bufp, "snap_id %llu\nsnap_name %s\n", | |
2454 | (unsigned long long) spec->snap_id, spec->snap_name); | |
2455 | if (count < 0) | |
2456 | return count; | |
2457 | bufp += count; | |
2458 | ||
2459 | count = sprintf(bufp, "overlap %llu\n", rbd_dev->parent_overlap); | |
2460 | if (count < 0) | |
2461 | return count; | |
2462 | bufp += count; | |
2463 | ||
2464 | return (ssize_t) (bufp - buf); | |
2465 | } | |
2466 | ||
dfc5606d YS |
2467 | static ssize_t rbd_image_refresh(struct device *dev, |
2468 | struct device_attribute *attr, | |
2469 | const char *buf, | |
2470 | size_t size) | |
2471 | { | |
593a9e7b | 2472 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
b813623a | 2473 | int ret; |
602adf40 | 2474 | |
117973fb | 2475 | ret = rbd_dev_refresh(rbd_dev, NULL); |
b813623a AE |
2476 | |
2477 | return ret < 0 ? ret : size; | |
dfc5606d | 2478 | } |
602adf40 | 2479 | |
dfc5606d | 2480 | static DEVICE_ATTR(size, S_IRUGO, rbd_size_show, NULL); |
34b13184 | 2481 | static DEVICE_ATTR(features, S_IRUGO, rbd_features_show, NULL); |
dfc5606d YS |
2482 | static DEVICE_ATTR(major, S_IRUGO, rbd_major_show, NULL); |
2483 | static DEVICE_ATTR(client_id, S_IRUGO, rbd_client_id_show, NULL); | |
2484 | static DEVICE_ATTR(pool, S_IRUGO, rbd_pool_show, NULL); | |
9bb2f334 | 2485 | static DEVICE_ATTR(pool_id, S_IRUGO, rbd_pool_id_show, NULL); |
dfc5606d | 2486 | static DEVICE_ATTR(name, S_IRUGO, rbd_name_show, NULL); |
589d30e0 | 2487 | static DEVICE_ATTR(image_id, S_IRUGO, rbd_image_id_show, NULL); |
dfc5606d YS |
2488 | static DEVICE_ATTR(refresh, S_IWUSR, NULL, rbd_image_refresh); |
2489 | static DEVICE_ATTR(current_snap, S_IRUGO, rbd_snap_show, NULL); | |
86b00e0d | 2490 | static DEVICE_ATTR(parent, S_IRUGO, rbd_parent_show, NULL); |
dfc5606d YS |
2491 | |
2492 | static struct attribute *rbd_attrs[] = { | |
2493 | &dev_attr_size.attr, | |
34b13184 | 2494 | &dev_attr_features.attr, |
dfc5606d YS |
2495 | &dev_attr_major.attr, |
2496 | &dev_attr_client_id.attr, | |
2497 | &dev_attr_pool.attr, | |
9bb2f334 | 2498 | &dev_attr_pool_id.attr, |
dfc5606d | 2499 | &dev_attr_name.attr, |
589d30e0 | 2500 | &dev_attr_image_id.attr, |
dfc5606d | 2501 | &dev_attr_current_snap.attr, |
86b00e0d | 2502 | &dev_attr_parent.attr, |
dfc5606d | 2503 | &dev_attr_refresh.attr, |
dfc5606d YS |
2504 | NULL |
2505 | }; | |
2506 | ||
2507 | static struct attribute_group rbd_attr_group = { | |
2508 | .attrs = rbd_attrs, | |
2509 | }; | |
2510 | ||
2511 | static const struct attribute_group *rbd_attr_groups[] = { | |
2512 | &rbd_attr_group, | |
2513 | NULL | |
2514 | }; | |
2515 | ||
2516 | static void rbd_sysfs_dev_release(struct device *dev) | |
2517 | { | |
2518 | } | |
2519 | ||
2520 | static struct device_type rbd_device_type = { | |
2521 | .name = "rbd", | |
2522 | .groups = rbd_attr_groups, | |
2523 | .release = rbd_sysfs_dev_release, | |
2524 | }; | |
2525 | ||
2526 | ||
2527 | /* | |
2528 | sysfs - snapshots | |
2529 | */ | |
2530 | ||
2531 | static ssize_t rbd_snap_size_show(struct device *dev, | |
2532 | struct device_attribute *attr, | |
2533 | char *buf) | |
2534 | { | |
2535 | struct rbd_snap *snap = container_of(dev, struct rbd_snap, dev); | |
2536 | ||
3591538f | 2537 | return sprintf(buf, "%llu\n", (unsigned long long)snap->size); |
dfc5606d YS |
2538 | } |
2539 | ||
2540 | static ssize_t rbd_snap_id_show(struct device *dev, | |
2541 | struct device_attribute *attr, | |
2542 | char *buf) | |
2543 | { | |
2544 | struct rbd_snap *snap = container_of(dev, struct rbd_snap, dev); | |
2545 | ||
3591538f | 2546 | return sprintf(buf, "%llu\n", (unsigned long long)snap->id); |
dfc5606d YS |
2547 | } |
2548 | ||
34b13184 AE |
2549 | static ssize_t rbd_snap_features_show(struct device *dev, |
2550 | struct device_attribute *attr, | |
2551 | char *buf) | |
2552 | { | |
2553 | struct rbd_snap *snap = container_of(dev, struct rbd_snap, dev); | |
2554 | ||
2555 | return sprintf(buf, "0x%016llx\n", | |
2556 | (unsigned long long) snap->features); | |
2557 | } | |
2558 | ||
dfc5606d YS |
2559 | static DEVICE_ATTR(snap_size, S_IRUGO, rbd_snap_size_show, NULL); |
2560 | static DEVICE_ATTR(snap_id, S_IRUGO, rbd_snap_id_show, NULL); | |
34b13184 | 2561 | static DEVICE_ATTR(snap_features, S_IRUGO, rbd_snap_features_show, NULL); |
dfc5606d YS |
2562 | |
2563 | static struct attribute *rbd_snap_attrs[] = { | |
2564 | &dev_attr_snap_size.attr, | |
2565 | &dev_attr_snap_id.attr, | |
34b13184 | 2566 | &dev_attr_snap_features.attr, |
dfc5606d YS |
2567 | NULL, |
2568 | }; | |
2569 | ||
2570 | static struct attribute_group rbd_snap_attr_group = { | |
2571 | .attrs = rbd_snap_attrs, | |
2572 | }; | |
2573 | ||
2574 | static void rbd_snap_dev_release(struct device *dev) | |
2575 | { | |
2576 | struct rbd_snap *snap = container_of(dev, struct rbd_snap, dev); | |
2577 | kfree(snap->name); | |
2578 | kfree(snap); | |
2579 | } | |
2580 | ||
2581 | static const struct attribute_group *rbd_snap_attr_groups[] = { | |
2582 | &rbd_snap_attr_group, | |
2583 | NULL | |
2584 | }; | |
2585 | ||
2586 | static struct device_type rbd_snap_device_type = { | |
2587 | .groups = rbd_snap_attr_groups, | |
2588 | .release = rbd_snap_dev_release, | |
2589 | }; | |
2590 | ||
8b8fb99c AE |
2591 | static struct rbd_spec *rbd_spec_get(struct rbd_spec *spec) |
2592 | { | |
2593 | kref_get(&spec->kref); | |
2594 | ||
2595 | return spec; | |
2596 | } | |
2597 | ||
2598 | static void rbd_spec_free(struct kref *kref); | |
2599 | static void rbd_spec_put(struct rbd_spec *spec) | |
2600 | { | |
2601 | if (spec) | |
2602 | kref_put(&spec->kref, rbd_spec_free); | |
2603 | } | |
2604 | ||
2605 | static struct rbd_spec *rbd_spec_alloc(void) | |
2606 | { | |
2607 | struct rbd_spec *spec; | |
2608 | ||
2609 | spec = kzalloc(sizeof (*spec), GFP_KERNEL); | |
2610 | if (!spec) | |
2611 | return NULL; | |
2612 | kref_init(&spec->kref); | |
2613 | ||
2614 | rbd_spec_put(rbd_spec_get(spec)); /* TEMPORARY */ | |
2615 | ||
2616 | return spec; | |
2617 | } | |
2618 | ||
2619 | static void rbd_spec_free(struct kref *kref) | |
2620 | { | |
2621 | struct rbd_spec *spec = container_of(kref, struct rbd_spec, kref); | |
2622 | ||
2623 | kfree(spec->pool_name); | |
2624 | kfree(spec->image_id); | |
2625 | kfree(spec->image_name); | |
2626 | kfree(spec->snap_name); | |
2627 | kfree(spec); | |
2628 | } | |
2629 | ||
cc344fa1 | 2630 | static struct rbd_device *rbd_dev_create(struct rbd_client *rbdc, |
c53d5893 AE |
2631 | struct rbd_spec *spec) |
2632 | { | |
2633 | struct rbd_device *rbd_dev; | |
2634 | ||
2635 | rbd_dev = kzalloc(sizeof (*rbd_dev), GFP_KERNEL); | |
2636 | if (!rbd_dev) | |
2637 | return NULL; | |
2638 | ||
2639 | spin_lock_init(&rbd_dev->lock); | |
6d292906 | 2640 | rbd_dev->flags = 0; |
c53d5893 AE |
2641 | INIT_LIST_HEAD(&rbd_dev->node); |
2642 | INIT_LIST_HEAD(&rbd_dev->snaps); | |
2643 | init_rwsem(&rbd_dev->header_rwsem); | |
2644 | ||
2645 | rbd_dev->spec = spec; | |
2646 | rbd_dev->rbd_client = rbdc; | |
2647 | ||
0903e875 AE |
2648 | /* Initialize the layout used for all rbd requests */ |
2649 | ||
2650 | rbd_dev->layout.fl_stripe_unit = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER); | |
2651 | rbd_dev->layout.fl_stripe_count = cpu_to_le32(1); | |
2652 | rbd_dev->layout.fl_object_size = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER); | |
2653 | rbd_dev->layout.fl_pg_pool = cpu_to_le32((u32) spec->pool_id); | |
2654 | ||
c53d5893 AE |
2655 | return rbd_dev; |
2656 | } | |
2657 | ||
2658 | static void rbd_dev_destroy(struct rbd_device *rbd_dev) | |
2659 | { | |
86b00e0d | 2660 | rbd_spec_put(rbd_dev->parent_spec); |
c53d5893 AE |
2661 | kfree(rbd_dev->header_name); |
2662 | rbd_put_client(rbd_dev->rbd_client); | |
2663 | rbd_spec_put(rbd_dev->spec); | |
2664 | kfree(rbd_dev); | |
2665 | } | |
2666 | ||
304f6808 AE |
2667 | static bool rbd_snap_registered(struct rbd_snap *snap) |
2668 | { | |
2669 | bool ret = snap->dev.type == &rbd_snap_device_type; | |
2670 | bool reg = device_is_registered(&snap->dev); | |
2671 | ||
2672 | rbd_assert(!ret ^ reg); | |
2673 | ||
2674 | return ret; | |
2675 | } | |
2676 | ||
41f38c2b | 2677 | static void rbd_remove_snap_dev(struct rbd_snap *snap) |
dfc5606d YS |
2678 | { |
2679 | list_del(&snap->node); | |
304f6808 AE |
2680 | if (device_is_registered(&snap->dev)) |
2681 | device_unregister(&snap->dev); | |
dfc5606d YS |
2682 | } |
2683 | ||
14e7085d | 2684 | static int rbd_register_snap_dev(struct rbd_snap *snap, |
dfc5606d YS |
2685 | struct device *parent) |
2686 | { | |
2687 | struct device *dev = &snap->dev; | |
2688 | int ret; | |
2689 | ||
2690 | dev->type = &rbd_snap_device_type; | |
2691 | dev->parent = parent; | |
2692 | dev->release = rbd_snap_dev_release; | |
d4b125e9 | 2693 | dev_set_name(dev, "%s%s", RBD_SNAP_DEV_NAME_PREFIX, snap->name); |
304f6808 AE |
2694 | dout("%s: registering device for snapshot %s\n", __func__, snap->name); |
2695 | ||
dfc5606d YS |
2696 | ret = device_register(dev); |
2697 | ||
2698 | return ret; | |
2699 | } | |
2700 | ||
4e891e0a | 2701 | static struct rbd_snap *__rbd_add_snap_dev(struct rbd_device *rbd_dev, |
c8d18425 | 2702 | const char *snap_name, |
34b13184 AE |
2703 | u64 snap_id, u64 snap_size, |
2704 | u64 snap_features) | |
dfc5606d | 2705 | { |
4e891e0a | 2706 | struct rbd_snap *snap; |
dfc5606d | 2707 | int ret; |
4e891e0a AE |
2708 | |
2709 | snap = kzalloc(sizeof (*snap), GFP_KERNEL); | |
dfc5606d | 2710 | if (!snap) |
4e891e0a AE |
2711 | return ERR_PTR(-ENOMEM); |
2712 | ||
2713 | ret = -ENOMEM; | |
c8d18425 | 2714 | snap->name = kstrdup(snap_name, GFP_KERNEL); |
4e891e0a AE |
2715 | if (!snap->name) |
2716 | goto err; | |
2717 | ||
c8d18425 AE |
2718 | snap->id = snap_id; |
2719 | snap->size = snap_size; | |
34b13184 | 2720 | snap->features = snap_features; |
4e891e0a AE |
2721 | |
2722 | return snap; | |
2723 | ||
dfc5606d YS |
2724 | err: |
2725 | kfree(snap->name); | |
2726 | kfree(snap); | |
4e891e0a AE |
2727 | |
2728 | return ERR_PTR(ret); | |
dfc5606d YS |
2729 | } |
2730 | ||
cd892126 AE |
2731 | static char *rbd_dev_v1_snap_info(struct rbd_device *rbd_dev, u32 which, |
2732 | u64 *snap_size, u64 *snap_features) | |
2733 | { | |
2734 | char *snap_name; | |
2735 | ||
2736 | rbd_assert(which < rbd_dev->header.snapc->num_snaps); | |
2737 | ||
2738 | *snap_size = rbd_dev->header.snap_sizes[which]; | |
2739 | *snap_features = 0; /* No features for v1 */ | |
2740 | ||
2741 | /* Skip over names until we find the one we are looking for */ | |
2742 | ||
2743 | snap_name = rbd_dev->header.snap_names; | |
2744 | while (which--) | |
2745 | snap_name += strlen(snap_name) + 1; | |
2746 | ||
2747 | return snap_name; | |
2748 | } | |
2749 | ||
9d475de5 AE |
2750 | /* |
2751 | * Get the size and object order for an image snapshot, or if | |
2752 | * snap_id is CEPH_NOSNAP, gets this information for the base | |
2753 | * image. | |
2754 | */ | |
2755 | static int _rbd_dev_v2_snap_size(struct rbd_device *rbd_dev, u64 snap_id, | |
2756 | u8 *order, u64 *snap_size) | |
2757 | { | |
2758 | __le64 snapid = cpu_to_le64(snap_id); | |
2759 | int ret; | |
2760 | struct { | |
2761 | u8 order; | |
2762 | __le64 size; | |
2763 | } __attribute__ ((packed)) size_buf = { 0 }; | |
2764 | ||
36be9a76 | 2765 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
9d475de5 AE |
2766 | "rbd", "get_size", |
2767 | (char *) &snapid, sizeof (snapid), | |
07b2391f | 2768 | (char *) &size_buf, sizeof (size_buf), NULL); |
36be9a76 | 2769 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
9d475de5 AE |
2770 | if (ret < 0) |
2771 | return ret; | |
2772 | ||
2773 | *order = size_buf.order; | |
2774 | *snap_size = le64_to_cpu(size_buf.size); | |
2775 | ||
2776 | dout(" snap_id 0x%016llx order = %u, snap_size = %llu\n", | |
2777 | (unsigned long long) snap_id, (unsigned int) *order, | |
2778 | (unsigned long long) *snap_size); | |
2779 | ||
2780 | return 0; | |
2781 | } | |
2782 | ||
2783 | static int rbd_dev_v2_image_size(struct rbd_device *rbd_dev) | |
2784 | { | |
2785 | return _rbd_dev_v2_snap_size(rbd_dev, CEPH_NOSNAP, | |
2786 | &rbd_dev->header.obj_order, | |
2787 | &rbd_dev->header.image_size); | |
2788 | } | |
2789 | ||
1e130199 AE |
2790 | static int rbd_dev_v2_object_prefix(struct rbd_device *rbd_dev) |
2791 | { | |
2792 | void *reply_buf; | |
2793 | int ret; | |
2794 | void *p; | |
2795 | ||
2796 | reply_buf = kzalloc(RBD_OBJ_PREFIX_LEN_MAX, GFP_KERNEL); | |
2797 | if (!reply_buf) | |
2798 | return -ENOMEM; | |
2799 | ||
36be9a76 | 2800 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
1e130199 AE |
2801 | "rbd", "get_object_prefix", |
2802 | NULL, 0, | |
07b2391f | 2803 | reply_buf, RBD_OBJ_PREFIX_LEN_MAX, NULL); |
36be9a76 | 2804 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
1e130199 AE |
2805 | if (ret < 0) |
2806 | goto out; | |
2807 | ||
2808 | p = reply_buf; | |
2809 | rbd_dev->header.object_prefix = ceph_extract_encoded_string(&p, | |
2810 | p + RBD_OBJ_PREFIX_LEN_MAX, | |
2811 | NULL, GFP_NOIO); | |
2812 | ||
2813 | if (IS_ERR(rbd_dev->header.object_prefix)) { | |
2814 | ret = PTR_ERR(rbd_dev->header.object_prefix); | |
2815 | rbd_dev->header.object_prefix = NULL; | |
2816 | } else { | |
2817 | dout(" object_prefix = %s\n", rbd_dev->header.object_prefix); | |
2818 | } | |
2819 | ||
2820 | out: | |
2821 | kfree(reply_buf); | |
2822 | ||
2823 | return ret; | |
2824 | } | |
2825 | ||
b1b5402a AE |
2826 | static int _rbd_dev_v2_snap_features(struct rbd_device *rbd_dev, u64 snap_id, |
2827 | u64 *snap_features) | |
2828 | { | |
2829 | __le64 snapid = cpu_to_le64(snap_id); | |
2830 | struct { | |
2831 | __le64 features; | |
2832 | __le64 incompat; | |
2833 | } features_buf = { 0 }; | |
d889140c | 2834 | u64 incompat; |
b1b5402a AE |
2835 | int ret; |
2836 | ||
36be9a76 | 2837 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
b1b5402a AE |
2838 | "rbd", "get_features", |
2839 | (char *) &snapid, sizeof (snapid), | |
2840 | (char *) &features_buf, sizeof (features_buf), | |
07b2391f | 2841 | NULL); |
36be9a76 | 2842 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
b1b5402a AE |
2843 | if (ret < 0) |
2844 | return ret; | |
d889140c AE |
2845 | |
2846 | incompat = le64_to_cpu(features_buf.incompat); | |
2847 | if (incompat & ~RBD_FEATURES_ALL) | |
b8f5c6ed | 2848 | return -ENXIO; |
d889140c | 2849 | |
b1b5402a AE |
2850 | *snap_features = le64_to_cpu(features_buf.features); |
2851 | ||
2852 | dout(" snap_id 0x%016llx features = 0x%016llx incompat = 0x%016llx\n", | |
2853 | (unsigned long long) snap_id, | |
2854 | (unsigned long long) *snap_features, | |
2855 | (unsigned long long) le64_to_cpu(features_buf.incompat)); | |
2856 | ||
2857 | return 0; | |
2858 | } | |
2859 | ||
2860 | static int rbd_dev_v2_features(struct rbd_device *rbd_dev) | |
2861 | { | |
2862 | return _rbd_dev_v2_snap_features(rbd_dev, CEPH_NOSNAP, | |
2863 | &rbd_dev->header.features); | |
2864 | } | |
2865 | ||
86b00e0d AE |
2866 | static int rbd_dev_v2_parent_info(struct rbd_device *rbd_dev) |
2867 | { | |
2868 | struct rbd_spec *parent_spec; | |
2869 | size_t size; | |
2870 | void *reply_buf = NULL; | |
2871 | __le64 snapid; | |
2872 | void *p; | |
2873 | void *end; | |
2874 | char *image_id; | |
2875 | u64 overlap; | |
86b00e0d AE |
2876 | int ret; |
2877 | ||
2878 | parent_spec = rbd_spec_alloc(); | |
2879 | if (!parent_spec) | |
2880 | return -ENOMEM; | |
2881 | ||
2882 | size = sizeof (__le64) + /* pool_id */ | |
2883 | sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX + /* image_id */ | |
2884 | sizeof (__le64) + /* snap_id */ | |
2885 | sizeof (__le64); /* overlap */ | |
2886 | reply_buf = kmalloc(size, GFP_KERNEL); | |
2887 | if (!reply_buf) { | |
2888 | ret = -ENOMEM; | |
2889 | goto out_err; | |
2890 | } | |
2891 | ||
2892 | snapid = cpu_to_le64(CEPH_NOSNAP); | |
36be9a76 | 2893 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
86b00e0d AE |
2894 | "rbd", "get_parent", |
2895 | (char *) &snapid, sizeof (snapid), | |
07b2391f | 2896 | (char *) reply_buf, size, NULL); |
36be9a76 | 2897 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
86b00e0d AE |
2898 | if (ret < 0) |
2899 | goto out_err; | |
2900 | ||
2901 | ret = -ERANGE; | |
2902 | p = reply_buf; | |
2903 | end = (char *) reply_buf + size; | |
2904 | ceph_decode_64_safe(&p, end, parent_spec->pool_id, out_err); | |
2905 | if (parent_spec->pool_id == CEPH_NOPOOL) | |
2906 | goto out; /* No parent? No problem. */ | |
2907 | ||
0903e875 AE |
2908 | /* The ceph file layout needs to fit pool id in 32 bits */ |
2909 | ||
2910 | ret = -EIO; | |
2911 | if (WARN_ON(parent_spec->pool_id > (u64) U32_MAX)) | |
2912 | goto out; | |
2913 | ||
979ed480 | 2914 | image_id = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL); |
86b00e0d AE |
2915 | if (IS_ERR(image_id)) { |
2916 | ret = PTR_ERR(image_id); | |
2917 | goto out_err; | |
2918 | } | |
2919 | parent_spec->image_id = image_id; | |
2920 | ceph_decode_64_safe(&p, end, parent_spec->snap_id, out_err); | |
2921 | ceph_decode_64_safe(&p, end, overlap, out_err); | |
2922 | ||
2923 | rbd_dev->parent_overlap = overlap; | |
2924 | rbd_dev->parent_spec = parent_spec; | |
2925 | parent_spec = NULL; /* rbd_dev now owns this */ | |
2926 | out: | |
2927 | ret = 0; | |
2928 | out_err: | |
2929 | kfree(reply_buf); | |
2930 | rbd_spec_put(parent_spec); | |
2931 | ||
2932 | return ret; | |
2933 | } | |
2934 | ||
9e15b77d AE |
2935 | static char *rbd_dev_image_name(struct rbd_device *rbd_dev) |
2936 | { | |
2937 | size_t image_id_size; | |
2938 | char *image_id; | |
2939 | void *p; | |
2940 | void *end; | |
2941 | size_t size; | |
2942 | void *reply_buf = NULL; | |
2943 | size_t len = 0; | |
2944 | char *image_name = NULL; | |
2945 | int ret; | |
2946 | ||
2947 | rbd_assert(!rbd_dev->spec->image_name); | |
2948 | ||
69e7a02f AE |
2949 | len = strlen(rbd_dev->spec->image_id); |
2950 | image_id_size = sizeof (__le32) + len; | |
9e15b77d AE |
2951 | image_id = kmalloc(image_id_size, GFP_KERNEL); |
2952 | if (!image_id) | |
2953 | return NULL; | |
2954 | ||
2955 | p = image_id; | |
2956 | end = (char *) image_id + image_id_size; | |
69e7a02f | 2957 | ceph_encode_string(&p, end, rbd_dev->spec->image_id, (u32) len); |
9e15b77d AE |
2958 | |
2959 | size = sizeof (__le32) + RBD_IMAGE_NAME_LEN_MAX; | |
2960 | reply_buf = kmalloc(size, GFP_KERNEL); | |
2961 | if (!reply_buf) | |
2962 | goto out; | |
2963 | ||
36be9a76 | 2964 | ret = rbd_obj_method_sync(rbd_dev, RBD_DIRECTORY, |
9e15b77d AE |
2965 | "rbd", "dir_get_name", |
2966 | image_id, image_id_size, | |
07b2391f | 2967 | (char *) reply_buf, size, NULL); |
9e15b77d AE |
2968 | if (ret < 0) |
2969 | goto out; | |
2970 | p = reply_buf; | |
2971 | end = (char *) reply_buf + size; | |
2972 | image_name = ceph_extract_encoded_string(&p, end, &len, GFP_KERNEL); | |
2973 | if (IS_ERR(image_name)) | |
2974 | image_name = NULL; | |
2975 | else | |
2976 | dout("%s: name is %s len is %zd\n", __func__, image_name, len); | |
2977 | out: | |
2978 | kfree(reply_buf); | |
2979 | kfree(image_id); | |
2980 | ||
2981 | return image_name; | |
2982 | } | |
2983 | ||
2984 | /* | |
2985 | * When a parent image gets probed, we only have the pool, image, | |
2986 | * and snapshot ids but not the names of any of them. This call | |
2987 | * is made later to fill in those names. It has to be done after | |
2988 | * rbd_dev_snaps_update() has completed because some of the | |
2989 | * information (in particular, snapshot name) is not available | |
2990 | * until then. | |
2991 | */ | |
2992 | static int rbd_dev_probe_update_spec(struct rbd_device *rbd_dev) | |
2993 | { | |
2994 | struct ceph_osd_client *osdc; | |
2995 | const char *name; | |
2996 | void *reply_buf = NULL; | |
2997 | int ret; | |
2998 | ||
2999 | if (rbd_dev->spec->pool_name) | |
3000 | return 0; /* Already have the names */ | |
3001 | ||
3002 | /* Look up the pool name */ | |
3003 | ||
3004 | osdc = &rbd_dev->rbd_client->client->osdc; | |
3005 | name = ceph_pg_pool_name_by_id(osdc->osdmap, rbd_dev->spec->pool_id); | |
935dc89f AE |
3006 | if (!name) { |
3007 | rbd_warn(rbd_dev, "there is no pool with id %llu", | |
3008 | rbd_dev->spec->pool_id); /* Really a BUG() */ | |
3009 | return -EIO; | |
3010 | } | |
9e15b77d AE |
3011 | |
3012 | rbd_dev->spec->pool_name = kstrdup(name, GFP_KERNEL); | |
3013 | if (!rbd_dev->spec->pool_name) | |
3014 | return -ENOMEM; | |
3015 | ||
3016 | /* Fetch the image name; tolerate failure here */ | |
3017 | ||
3018 | name = rbd_dev_image_name(rbd_dev); | |
69e7a02f | 3019 | if (name) |
9e15b77d | 3020 | rbd_dev->spec->image_name = (char *) name; |
69e7a02f | 3021 | else |
06ecc6cb | 3022 | rbd_warn(rbd_dev, "unable to get image name"); |
9e15b77d AE |
3023 | |
3024 | /* Look up the snapshot name. */ | |
3025 | ||
3026 | name = rbd_snap_name(rbd_dev, rbd_dev->spec->snap_id); | |
3027 | if (!name) { | |
935dc89f AE |
3028 | rbd_warn(rbd_dev, "no snapshot with id %llu", |
3029 | rbd_dev->spec->snap_id); /* Really a BUG() */ | |
9e15b77d AE |
3030 | ret = -EIO; |
3031 | goto out_err; | |
3032 | } | |
3033 | rbd_dev->spec->snap_name = kstrdup(name, GFP_KERNEL); | |
3034 | if(!rbd_dev->spec->snap_name) | |
3035 | goto out_err; | |
3036 | ||
3037 | return 0; | |
3038 | out_err: | |
3039 | kfree(reply_buf); | |
3040 | kfree(rbd_dev->spec->pool_name); | |
3041 | rbd_dev->spec->pool_name = NULL; | |
3042 | ||
3043 | return ret; | |
3044 | } | |
3045 | ||
6e14b1a6 | 3046 | static int rbd_dev_v2_snap_context(struct rbd_device *rbd_dev, u64 *ver) |
35d489f9 AE |
3047 | { |
3048 | size_t size; | |
3049 | int ret; | |
3050 | void *reply_buf; | |
3051 | void *p; | |
3052 | void *end; | |
3053 | u64 seq; | |
3054 | u32 snap_count; | |
3055 | struct ceph_snap_context *snapc; | |
3056 | u32 i; | |
3057 | ||
3058 | /* | |
3059 | * We'll need room for the seq value (maximum snapshot id), | |
3060 | * snapshot count, and array of that many snapshot ids. | |
3061 | * For now we have a fixed upper limit on the number we're | |
3062 | * prepared to receive. | |
3063 | */ | |
3064 | size = sizeof (__le64) + sizeof (__le32) + | |
3065 | RBD_MAX_SNAP_COUNT * sizeof (__le64); | |
3066 | reply_buf = kzalloc(size, GFP_KERNEL); | |
3067 | if (!reply_buf) | |
3068 | return -ENOMEM; | |
3069 | ||
36be9a76 | 3070 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
35d489f9 AE |
3071 | "rbd", "get_snapcontext", |
3072 | NULL, 0, | |
07b2391f | 3073 | reply_buf, size, ver); |
36be9a76 | 3074 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
35d489f9 AE |
3075 | if (ret < 0) |
3076 | goto out; | |
3077 | ||
3078 | ret = -ERANGE; | |
3079 | p = reply_buf; | |
3080 | end = (char *) reply_buf + size; | |
3081 | ceph_decode_64_safe(&p, end, seq, out); | |
3082 | ceph_decode_32_safe(&p, end, snap_count, out); | |
3083 | ||
3084 | /* | |
3085 | * Make sure the reported number of snapshot ids wouldn't go | |
3086 | * beyond the end of our buffer. But before checking that, | |
3087 | * make sure the computed size of the snapshot context we | |
3088 | * allocate is representable in a size_t. | |
3089 | */ | |
3090 | if (snap_count > (SIZE_MAX - sizeof (struct ceph_snap_context)) | |
3091 | / sizeof (u64)) { | |
3092 | ret = -EINVAL; | |
3093 | goto out; | |
3094 | } | |
3095 | if (!ceph_has_room(&p, end, snap_count * sizeof (__le64))) | |
3096 | goto out; | |
3097 | ||
3098 | size = sizeof (struct ceph_snap_context) + | |
3099 | snap_count * sizeof (snapc->snaps[0]); | |
3100 | snapc = kmalloc(size, GFP_KERNEL); | |
3101 | if (!snapc) { | |
3102 | ret = -ENOMEM; | |
3103 | goto out; | |
3104 | } | |
3105 | ||
3106 | atomic_set(&snapc->nref, 1); | |
3107 | snapc->seq = seq; | |
3108 | snapc->num_snaps = snap_count; | |
3109 | for (i = 0; i < snap_count; i++) | |
3110 | snapc->snaps[i] = ceph_decode_64(&p); | |
3111 | ||
3112 | rbd_dev->header.snapc = snapc; | |
3113 | ||
3114 | dout(" snap context seq = %llu, snap_count = %u\n", | |
3115 | (unsigned long long) seq, (unsigned int) snap_count); | |
3116 | ||
3117 | out: | |
3118 | kfree(reply_buf); | |
3119 | ||
3120 | return 0; | |
3121 | } | |
3122 | ||
b8b1e2db AE |
3123 | static char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev, u32 which) |
3124 | { | |
3125 | size_t size; | |
3126 | void *reply_buf; | |
3127 | __le64 snap_id; | |
3128 | int ret; | |
3129 | void *p; | |
3130 | void *end; | |
b8b1e2db AE |
3131 | char *snap_name; |
3132 | ||
3133 | size = sizeof (__le32) + RBD_MAX_SNAP_NAME_LEN; | |
3134 | reply_buf = kmalloc(size, GFP_KERNEL); | |
3135 | if (!reply_buf) | |
3136 | return ERR_PTR(-ENOMEM); | |
3137 | ||
3138 | snap_id = cpu_to_le64(rbd_dev->header.snapc->snaps[which]); | |
36be9a76 | 3139 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
b8b1e2db AE |
3140 | "rbd", "get_snapshot_name", |
3141 | (char *) &snap_id, sizeof (snap_id), | |
07b2391f | 3142 | reply_buf, size, NULL); |
36be9a76 | 3143 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
b8b1e2db AE |
3144 | if (ret < 0) |
3145 | goto out; | |
3146 | ||
3147 | p = reply_buf; | |
3148 | end = (char *) reply_buf + size; | |
e5c35534 | 3149 | snap_name = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL); |
b8b1e2db AE |
3150 | if (IS_ERR(snap_name)) { |
3151 | ret = PTR_ERR(snap_name); | |
3152 | goto out; | |
3153 | } else { | |
3154 | dout(" snap_id 0x%016llx snap_name = %s\n", | |
3155 | (unsigned long long) le64_to_cpu(snap_id), snap_name); | |
3156 | } | |
3157 | kfree(reply_buf); | |
3158 | ||
3159 | return snap_name; | |
3160 | out: | |
3161 | kfree(reply_buf); | |
3162 | ||
3163 | return ERR_PTR(ret); | |
3164 | } | |
3165 | ||
3166 | static char *rbd_dev_v2_snap_info(struct rbd_device *rbd_dev, u32 which, | |
3167 | u64 *snap_size, u64 *snap_features) | |
3168 | { | |
e0b49868 | 3169 | u64 snap_id; |
b8b1e2db AE |
3170 | u8 order; |
3171 | int ret; | |
3172 | ||
3173 | snap_id = rbd_dev->header.snapc->snaps[which]; | |
3174 | ret = _rbd_dev_v2_snap_size(rbd_dev, snap_id, &order, snap_size); | |
3175 | if (ret) | |
3176 | return ERR_PTR(ret); | |
3177 | ret = _rbd_dev_v2_snap_features(rbd_dev, snap_id, snap_features); | |
3178 | if (ret) | |
3179 | return ERR_PTR(ret); | |
3180 | ||
3181 | return rbd_dev_v2_snap_name(rbd_dev, which); | |
3182 | } | |
3183 | ||
3184 | static char *rbd_dev_snap_info(struct rbd_device *rbd_dev, u32 which, | |
3185 | u64 *snap_size, u64 *snap_features) | |
3186 | { | |
3187 | if (rbd_dev->image_format == 1) | |
3188 | return rbd_dev_v1_snap_info(rbd_dev, which, | |
3189 | snap_size, snap_features); | |
3190 | if (rbd_dev->image_format == 2) | |
3191 | return rbd_dev_v2_snap_info(rbd_dev, which, | |
3192 | snap_size, snap_features); | |
3193 | return ERR_PTR(-EINVAL); | |
3194 | } | |
3195 | ||
117973fb AE |
3196 | static int rbd_dev_v2_refresh(struct rbd_device *rbd_dev, u64 *hver) |
3197 | { | |
3198 | int ret; | |
3199 | __u8 obj_order; | |
3200 | ||
3201 | down_write(&rbd_dev->header_rwsem); | |
3202 | ||
3203 | /* Grab old order first, to see if it changes */ | |
3204 | ||
3205 | obj_order = rbd_dev->header.obj_order, | |
3206 | ret = rbd_dev_v2_image_size(rbd_dev); | |
3207 | if (ret) | |
3208 | goto out; | |
3209 | if (rbd_dev->header.obj_order != obj_order) { | |
3210 | ret = -EIO; | |
3211 | goto out; | |
3212 | } | |
3213 | rbd_update_mapping_size(rbd_dev); | |
3214 | ||
3215 | ret = rbd_dev_v2_snap_context(rbd_dev, hver); | |
3216 | dout("rbd_dev_v2_snap_context returned %d\n", ret); | |
3217 | if (ret) | |
3218 | goto out; | |
3219 | ret = rbd_dev_snaps_update(rbd_dev); | |
3220 | dout("rbd_dev_snaps_update returned %d\n", ret); | |
3221 | if (ret) | |
3222 | goto out; | |
3223 | ret = rbd_dev_snaps_register(rbd_dev); | |
3224 | dout("rbd_dev_snaps_register returned %d\n", ret); | |
3225 | out: | |
3226 | up_write(&rbd_dev->header_rwsem); | |
3227 | ||
3228 | return ret; | |
3229 | } | |
3230 | ||
dfc5606d | 3231 | /* |
35938150 AE |
3232 | * Scan the rbd device's current snapshot list and compare it to the |
3233 | * newly-received snapshot context. Remove any existing snapshots | |
3234 | * not present in the new snapshot context. Add a new snapshot for | |
3235 | * any snaphots in the snapshot context not in the current list. | |
3236 | * And verify there are no changes to snapshots we already know | |
3237 | * about. | |
3238 | * | |
3239 | * Assumes the snapshots in the snapshot context are sorted by | |
3240 | * snapshot id, highest id first. (Snapshots in the rbd_dev's list | |
3241 | * are also maintained in that order.) | |
dfc5606d | 3242 | */ |
304f6808 | 3243 | static int rbd_dev_snaps_update(struct rbd_device *rbd_dev) |
dfc5606d | 3244 | { |
35938150 AE |
3245 | struct ceph_snap_context *snapc = rbd_dev->header.snapc; |
3246 | const u32 snap_count = snapc->num_snaps; | |
35938150 AE |
3247 | struct list_head *head = &rbd_dev->snaps; |
3248 | struct list_head *links = head->next; | |
3249 | u32 index = 0; | |
dfc5606d | 3250 | |
9fcbb800 | 3251 | dout("%s: snap count is %u\n", __func__, (unsigned int) snap_count); |
35938150 AE |
3252 | while (index < snap_count || links != head) { |
3253 | u64 snap_id; | |
3254 | struct rbd_snap *snap; | |
cd892126 AE |
3255 | char *snap_name; |
3256 | u64 snap_size = 0; | |
3257 | u64 snap_features = 0; | |
dfc5606d | 3258 | |
35938150 AE |
3259 | snap_id = index < snap_count ? snapc->snaps[index] |
3260 | : CEPH_NOSNAP; | |
3261 | snap = links != head ? list_entry(links, struct rbd_snap, node) | |
3262 | : NULL; | |
aafb230e | 3263 | rbd_assert(!snap || snap->id != CEPH_NOSNAP); |
dfc5606d | 3264 | |
35938150 AE |
3265 | if (snap_id == CEPH_NOSNAP || (snap && snap->id > snap_id)) { |
3266 | struct list_head *next = links->next; | |
dfc5606d | 3267 | |
6d292906 AE |
3268 | /* |
3269 | * A previously-existing snapshot is not in | |
3270 | * the new snap context. | |
3271 | * | |
3272 | * If the now missing snapshot is the one the | |
3273 | * image is mapped to, clear its exists flag | |
3274 | * so we can avoid sending any more requests | |
3275 | * to it. | |
3276 | */ | |
0d7dbfce | 3277 | if (rbd_dev->spec->snap_id == snap->id) |
6d292906 | 3278 | clear_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags); |
41f38c2b | 3279 | rbd_remove_snap_dev(snap); |
9fcbb800 | 3280 | dout("%ssnap id %llu has been removed\n", |
0d7dbfce AE |
3281 | rbd_dev->spec->snap_id == snap->id ? |
3282 | "mapped " : "", | |
9fcbb800 | 3283 | (unsigned long long) snap->id); |
35938150 AE |
3284 | |
3285 | /* Done with this list entry; advance */ | |
3286 | ||
3287 | links = next; | |
dfc5606d YS |
3288 | continue; |
3289 | } | |
35938150 | 3290 | |
b8b1e2db AE |
3291 | snap_name = rbd_dev_snap_info(rbd_dev, index, |
3292 | &snap_size, &snap_features); | |
cd892126 AE |
3293 | if (IS_ERR(snap_name)) |
3294 | return PTR_ERR(snap_name); | |
3295 | ||
9fcbb800 AE |
3296 | dout("entry %u: snap_id = %llu\n", (unsigned int) snap_count, |
3297 | (unsigned long long) snap_id); | |
35938150 AE |
3298 | if (!snap || (snap_id != CEPH_NOSNAP && snap->id < snap_id)) { |
3299 | struct rbd_snap *new_snap; | |
3300 | ||
3301 | /* We haven't seen this snapshot before */ | |
3302 | ||
c8d18425 | 3303 | new_snap = __rbd_add_snap_dev(rbd_dev, snap_name, |
cd892126 | 3304 | snap_id, snap_size, snap_features); |
9fcbb800 AE |
3305 | if (IS_ERR(new_snap)) { |
3306 | int err = PTR_ERR(new_snap); | |
3307 | ||
3308 | dout(" failed to add dev, error %d\n", err); | |
3309 | ||
3310 | return err; | |
3311 | } | |
35938150 AE |
3312 | |
3313 | /* New goes before existing, or at end of list */ | |
3314 | ||
9fcbb800 | 3315 | dout(" added dev%s\n", snap ? "" : " at end\n"); |
35938150 AE |
3316 | if (snap) |
3317 | list_add_tail(&new_snap->node, &snap->node); | |
3318 | else | |
523f3258 | 3319 | list_add_tail(&new_snap->node, head); |
35938150 AE |
3320 | } else { |
3321 | /* Already have this one */ | |
3322 | ||
9fcbb800 AE |
3323 | dout(" already present\n"); |
3324 | ||
cd892126 | 3325 | rbd_assert(snap->size == snap_size); |
aafb230e | 3326 | rbd_assert(!strcmp(snap->name, snap_name)); |
cd892126 | 3327 | rbd_assert(snap->features == snap_features); |
35938150 AE |
3328 | |
3329 | /* Done with this list entry; advance */ | |
3330 | ||
3331 | links = links->next; | |
dfc5606d | 3332 | } |
35938150 AE |
3333 | |
3334 | /* Advance to the next entry in the snapshot context */ | |
3335 | ||
3336 | index++; | |
dfc5606d | 3337 | } |
9fcbb800 | 3338 | dout("%s: done\n", __func__); |
dfc5606d YS |
3339 | |
3340 | return 0; | |
3341 | } | |
3342 | ||
304f6808 AE |
3343 | /* |
3344 | * Scan the list of snapshots and register the devices for any that | |
3345 | * have not already been registered. | |
3346 | */ | |
3347 | static int rbd_dev_snaps_register(struct rbd_device *rbd_dev) | |
3348 | { | |
3349 | struct rbd_snap *snap; | |
3350 | int ret = 0; | |
3351 | ||
37206ee5 | 3352 | dout("%s:\n", __func__); |
86ff77bb AE |
3353 | if (WARN_ON(!device_is_registered(&rbd_dev->dev))) |
3354 | return -EIO; | |
304f6808 AE |
3355 | |
3356 | list_for_each_entry(snap, &rbd_dev->snaps, node) { | |
3357 | if (!rbd_snap_registered(snap)) { | |
3358 | ret = rbd_register_snap_dev(snap, &rbd_dev->dev); | |
3359 | if (ret < 0) | |
3360 | break; | |
3361 | } | |
3362 | } | |
3363 | dout("%s: returning %d\n", __func__, ret); | |
3364 | ||
3365 | return ret; | |
3366 | } | |
3367 | ||
dfc5606d YS |
3368 | static int rbd_bus_add_dev(struct rbd_device *rbd_dev) |
3369 | { | |
dfc5606d | 3370 | struct device *dev; |
cd789ab9 | 3371 | int ret; |
dfc5606d YS |
3372 | |
3373 | mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING); | |
dfc5606d | 3374 | |
cd789ab9 | 3375 | dev = &rbd_dev->dev; |
dfc5606d YS |
3376 | dev->bus = &rbd_bus_type; |
3377 | dev->type = &rbd_device_type; | |
3378 | dev->parent = &rbd_root_dev; | |
3379 | dev->release = rbd_dev_release; | |
de71a297 | 3380 | dev_set_name(dev, "%d", rbd_dev->dev_id); |
dfc5606d | 3381 | ret = device_register(dev); |
dfc5606d | 3382 | |
dfc5606d | 3383 | mutex_unlock(&ctl_mutex); |
cd789ab9 | 3384 | |
dfc5606d | 3385 | return ret; |
602adf40 YS |
3386 | } |
3387 | ||
dfc5606d YS |
3388 | static void rbd_bus_del_dev(struct rbd_device *rbd_dev) |
3389 | { | |
3390 | device_unregister(&rbd_dev->dev); | |
3391 | } | |
3392 | ||
e2839308 | 3393 | static atomic64_t rbd_dev_id_max = ATOMIC64_INIT(0); |
1ddbe94e AE |
3394 | |
3395 | /* | |
499afd5b AE |
3396 | * Get a unique rbd identifier for the given new rbd_dev, and add |
3397 | * the rbd_dev to the global list. The minimum rbd id is 1. | |
1ddbe94e | 3398 | */ |
e2839308 | 3399 | static void rbd_dev_id_get(struct rbd_device *rbd_dev) |
b7f23c36 | 3400 | { |
e2839308 | 3401 | rbd_dev->dev_id = atomic64_inc_return(&rbd_dev_id_max); |
499afd5b AE |
3402 | |
3403 | spin_lock(&rbd_dev_list_lock); | |
3404 | list_add_tail(&rbd_dev->node, &rbd_dev_list); | |
3405 | spin_unlock(&rbd_dev_list_lock); | |
e2839308 AE |
3406 | dout("rbd_dev %p given dev id %llu\n", rbd_dev, |
3407 | (unsigned long long) rbd_dev->dev_id); | |
1ddbe94e | 3408 | } |
b7f23c36 | 3409 | |
1ddbe94e | 3410 | /* |
499afd5b AE |
3411 | * Remove an rbd_dev from the global list, and record that its |
3412 | * identifier is no longer in use. | |
1ddbe94e | 3413 | */ |
e2839308 | 3414 | static void rbd_dev_id_put(struct rbd_device *rbd_dev) |
1ddbe94e | 3415 | { |
d184f6bf | 3416 | struct list_head *tmp; |
de71a297 | 3417 | int rbd_id = rbd_dev->dev_id; |
d184f6bf AE |
3418 | int max_id; |
3419 | ||
aafb230e | 3420 | rbd_assert(rbd_id > 0); |
499afd5b | 3421 | |
e2839308 AE |
3422 | dout("rbd_dev %p released dev id %llu\n", rbd_dev, |
3423 | (unsigned long long) rbd_dev->dev_id); | |
499afd5b AE |
3424 | spin_lock(&rbd_dev_list_lock); |
3425 | list_del_init(&rbd_dev->node); | |
d184f6bf AE |
3426 | |
3427 | /* | |
3428 | * If the id being "put" is not the current maximum, there | |
3429 | * is nothing special we need to do. | |
3430 | */ | |
e2839308 | 3431 | if (rbd_id != atomic64_read(&rbd_dev_id_max)) { |
d184f6bf AE |
3432 | spin_unlock(&rbd_dev_list_lock); |
3433 | return; | |
3434 | } | |
3435 | ||
3436 | /* | |
3437 | * We need to update the current maximum id. Search the | |
3438 | * list to find out what it is. We're more likely to find | |
3439 | * the maximum at the end, so search the list backward. | |
3440 | */ | |
3441 | max_id = 0; | |
3442 | list_for_each_prev(tmp, &rbd_dev_list) { | |
3443 | struct rbd_device *rbd_dev; | |
3444 | ||
3445 | rbd_dev = list_entry(tmp, struct rbd_device, node); | |
b213e0b1 AE |
3446 | if (rbd_dev->dev_id > max_id) |
3447 | max_id = rbd_dev->dev_id; | |
d184f6bf | 3448 | } |
499afd5b | 3449 | spin_unlock(&rbd_dev_list_lock); |
b7f23c36 | 3450 | |
1ddbe94e | 3451 | /* |
e2839308 | 3452 | * The max id could have been updated by rbd_dev_id_get(), in |
d184f6bf AE |
3453 | * which case it now accurately reflects the new maximum. |
3454 | * Be careful not to overwrite the maximum value in that | |
3455 | * case. | |
1ddbe94e | 3456 | */ |
e2839308 AE |
3457 | atomic64_cmpxchg(&rbd_dev_id_max, rbd_id, max_id); |
3458 | dout(" max dev id has been reset\n"); | |
b7f23c36 AE |
3459 | } |
3460 | ||
e28fff26 AE |
3461 | /* |
3462 | * Skips over white space at *buf, and updates *buf to point to the | |
3463 | * first found non-space character (if any). Returns the length of | |
593a9e7b AE |
3464 | * the token (string of non-white space characters) found. Note |
3465 | * that *buf must be terminated with '\0'. | |
e28fff26 AE |
3466 | */ |
3467 | static inline size_t next_token(const char **buf) | |
3468 | { | |
3469 | /* | |
3470 | * These are the characters that produce nonzero for | |
3471 | * isspace() in the "C" and "POSIX" locales. | |
3472 | */ | |
3473 | const char *spaces = " \f\n\r\t\v"; | |
3474 | ||
3475 | *buf += strspn(*buf, spaces); /* Find start of token */ | |
3476 | ||
3477 | return strcspn(*buf, spaces); /* Return token length */ | |
3478 | } | |
3479 | ||
3480 | /* | |
3481 | * Finds the next token in *buf, and if the provided token buffer is | |
3482 | * big enough, copies the found token into it. The result, if | |
593a9e7b AE |
3483 | * copied, is guaranteed to be terminated with '\0'. Note that *buf |
3484 | * must be terminated with '\0' on entry. | |
e28fff26 AE |
3485 | * |
3486 | * Returns the length of the token found (not including the '\0'). | |
3487 | * Return value will be 0 if no token is found, and it will be >= | |
3488 | * token_size if the token would not fit. | |
3489 | * | |
593a9e7b | 3490 | * The *buf pointer will be updated to point beyond the end of the |
e28fff26 AE |
3491 | * found token. Note that this occurs even if the token buffer is |
3492 | * too small to hold it. | |
3493 | */ | |
3494 | static inline size_t copy_token(const char **buf, | |
3495 | char *token, | |
3496 | size_t token_size) | |
3497 | { | |
3498 | size_t len; | |
3499 | ||
3500 | len = next_token(buf); | |
3501 | if (len < token_size) { | |
3502 | memcpy(token, *buf, len); | |
3503 | *(token + len) = '\0'; | |
3504 | } | |
3505 | *buf += len; | |
3506 | ||
3507 | return len; | |
3508 | } | |
3509 | ||
ea3352f4 AE |
3510 | /* |
3511 | * Finds the next token in *buf, dynamically allocates a buffer big | |
3512 | * enough to hold a copy of it, and copies the token into the new | |
3513 | * buffer. The copy is guaranteed to be terminated with '\0'. Note | |
3514 | * that a duplicate buffer is created even for a zero-length token. | |
3515 | * | |
3516 | * Returns a pointer to the newly-allocated duplicate, or a null | |
3517 | * pointer if memory for the duplicate was not available. If | |
3518 | * the lenp argument is a non-null pointer, the length of the token | |
3519 | * (not including the '\0') is returned in *lenp. | |
3520 | * | |
3521 | * If successful, the *buf pointer will be updated to point beyond | |
3522 | * the end of the found token. | |
3523 | * | |
3524 | * Note: uses GFP_KERNEL for allocation. | |
3525 | */ | |
3526 | static inline char *dup_token(const char **buf, size_t *lenp) | |
3527 | { | |
3528 | char *dup; | |
3529 | size_t len; | |
3530 | ||
3531 | len = next_token(buf); | |
4caf35f9 | 3532 | dup = kmemdup(*buf, len + 1, GFP_KERNEL); |
ea3352f4 AE |
3533 | if (!dup) |
3534 | return NULL; | |
ea3352f4 AE |
3535 | *(dup + len) = '\0'; |
3536 | *buf += len; | |
3537 | ||
3538 | if (lenp) | |
3539 | *lenp = len; | |
3540 | ||
3541 | return dup; | |
3542 | } | |
3543 | ||
a725f65e | 3544 | /* |
859c31df AE |
3545 | * Parse the options provided for an "rbd add" (i.e., rbd image |
3546 | * mapping) request. These arrive via a write to /sys/bus/rbd/add, | |
3547 | * and the data written is passed here via a NUL-terminated buffer. | |
3548 | * Returns 0 if successful or an error code otherwise. | |
d22f76e7 | 3549 | * |
859c31df AE |
3550 | * The information extracted from these options is recorded in |
3551 | * the other parameters which return dynamically-allocated | |
3552 | * structures: | |
3553 | * ceph_opts | |
3554 | * The address of a pointer that will refer to a ceph options | |
3555 | * structure. Caller must release the returned pointer using | |
3556 | * ceph_destroy_options() when it is no longer needed. | |
3557 | * rbd_opts | |
3558 | * Address of an rbd options pointer. Fully initialized by | |
3559 | * this function; caller must release with kfree(). | |
3560 | * spec | |
3561 | * Address of an rbd image specification pointer. Fully | |
3562 | * initialized by this function based on parsed options. | |
3563 | * Caller must release with rbd_spec_put(). | |
3564 | * | |
3565 | * The options passed take this form: | |
3566 | * <mon_addrs> <options> <pool_name> <image_name> [<snap_id>] | |
3567 | * where: | |
3568 | * <mon_addrs> | |
3569 | * A comma-separated list of one or more monitor addresses. | |
3570 | * A monitor address is an ip address, optionally followed | |
3571 | * by a port number (separated by a colon). | |
3572 | * I.e.: ip1[:port1][,ip2[:port2]...] | |
3573 | * <options> | |
3574 | * A comma-separated list of ceph and/or rbd options. | |
3575 | * <pool_name> | |
3576 | * The name of the rados pool containing the rbd image. | |
3577 | * <image_name> | |
3578 | * The name of the image in that pool to map. | |
3579 | * <snap_id> | |
3580 | * An optional snapshot id. If provided, the mapping will | |
3581 | * present data from the image at the time that snapshot was | |
3582 | * created. The image head is used if no snapshot id is | |
3583 | * provided. Snapshot mappings are always read-only. | |
a725f65e | 3584 | */ |
859c31df | 3585 | static int rbd_add_parse_args(const char *buf, |
dc79b113 | 3586 | struct ceph_options **ceph_opts, |
859c31df AE |
3587 | struct rbd_options **opts, |
3588 | struct rbd_spec **rbd_spec) | |
e28fff26 | 3589 | { |
d22f76e7 | 3590 | size_t len; |
859c31df | 3591 | char *options; |
0ddebc0c AE |
3592 | const char *mon_addrs; |
3593 | size_t mon_addrs_size; | |
859c31df | 3594 | struct rbd_spec *spec = NULL; |
4e9afeba | 3595 | struct rbd_options *rbd_opts = NULL; |
859c31df | 3596 | struct ceph_options *copts; |
dc79b113 | 3597 | int ret; |
e28fff26 AE |
3598 | |
3599 | /* The first four tokens are required */ | |
3600 | ||
7ef3214a | 3601 | len = next_token(&buf); |
4fb5d671 AE |
3602 | if (!len) { |
3603 | rbd_warn(NULL, "no monitor address(es) provided"); | |
3604 | return -EINVAL; | |
3605 | } | |
0ddebc0c | 3606 | mon_addrs = buf; |
f28e565a | 3607 | mon_addrs_size = len + 1; |
7ef3214a | 3608 | buf += len; |
a725f65e | 3609 | |
dc79b113 | 3610 | ret = -EINVAL; |
f28e565a AE |
3611 | options = dup_token(&buf, NULL); |
3612 | if (!options) | |
dc79b113 | 3613 | return -ENOMEM; |
4fb5d671 AE |
3614 | if (!*options) { |
3615 | rbd_warn(NULL, "no options provided"); | |
3616 | goto out_err; | |
3617 | } | |
e28fff26 | 3618 | |
859c31df AE |
3619 | spec = rbd_spec_alloc(); |
3620 | if (!spec) | |
f28e565a | 3621 | goto out_mem; |
859c31df AE |
3622 | |
3623 | spec->pool_name = dup_token(&buf, NULL); | |
3624 | if (!spec->pool_name) | |
3625 | goto out_mem; | |
4fb5d671 AE |
3626 | if (!*spec->pool_name) { |
3627 | rbd_warn(NULL, "no pool name provided"); | |
3628 | goto out_err; | |
3629 | } | |
e28fff26 | 3630 | |
69e7a02f | 3631 | spec->image_name = dup_token(&buf, NULL); |
859c31df | 3632 | if (!spec->image_name) |
f28e565a | 3633 | goto out_mem; |
4fb5d671 AE |
3634 | if (!*spec->image_name) { |
3635 | rbd_warn(NULL, "no image name provided"); | |
3636 | goto out_err; | |
3637 | } | |
d4b125e9 | 3638 | |
f28e565a AE |
3639 | /* |
3640 | * Snapshot name is optional; default is to use "-" | |
3641 | * (indicating the head/no snapshot). | |
3642 | */ | |
3feeb894 | 3643 | len = next_token(&buf); |
820a5f3e | 3644 | if (!len) { |
3feeb894 AE |
3645 | buf = RBD_SNAP_HEAD_NAME; /* No snapshot supplied */ |
3646 | len = sizeof (RBD_SNAP_HEAD_NAME) - 1; | |
f28e565a | 3647 | } else if (len > RBD_MAX_SNAP_NAME_LEN) { |
dc79b113 | 3648 | ret = -ENAMETOOLONG; |
f28e565a | 3649 | goto out_err; |
849b4260 | 3650 | } |
4caf35f9 | 3651 | spec->snap_name = kmemdup(buf, len + 1, GFP_KERNEL); |
859c31df | 3652 | if (!spec->snap_name) |
f28e565a | 3653 | goto out_mem; |
859c31df | 3654 | *(spec->snap_name + len) = '\0'; |
e5c35534 | 3655 | |
0ddebc0c | 3656 | /* Initialize all rbd options to the defaults */ |
e28fff26 | 3657 | |
4e9afeba AE |
3658 | rbd_opts = kzalloc(sizeof (*rbd_opts), GFP_KERNEL); |
3659 | if (!rbd_opts) | |
3660 | goto out_mem; | |
3661 | ||
3662 | rbd_opts->read_only = RBD_READ_ONLY_DEFAULT; | |
d22f76e7 | 3663 | |
859c31df | 3664 | copts = ceph_parse_options(options, mon_addrs, |
0ddebc0c | 3665 | mon_addrs + mon_addrs_size - 1, |
4e9afeba | 3666 | parse_rbd_opts_token, rbd_opts); |
859c31df AE |
3667 | if (IS_ERR(copts)) { |
3668 | ret = PTR_ERR(copts); | |
dc79b113 AE |
3669 | goto out_err; |
3670 | } | |
859c31df AE |
3671 | kfree(options); |
3672 | ||
3673 | *ceph_opts = copts; | |
4e9afeba | 3674 | *opts = rbd_opts; |
859c31df | 3675 | *rbd_spec = spec; |
0ddebc0c | 3676 | |
dc79b113 | 3677 | return 0; |
f28e565a | 3678 | out_mem: |
dc79b113 | 3679 | ret = -ENOMEM; |
d22f76e7 | 3680 | out_err: |
859c31df AE |
3681 | kfree(rbd_opts); |
3682 | rbd_spec_put(spec); | |
f28e565a | 3683 | kfree(options); |
d22f76e7 | 3684 | |
dc79b113 | 3685 | return ret; |
a725f65e AE |
3686 | } |
3687 | ||
589d30e0 AE |
3688 | /* |
3689 | * An rbd format 2 image has a unique identifier, distinct from the | |
3690 | * name given to it by the user. Internally, that identifier is | |
3691 | * what's used to specify the names of objects related to the image. | |
3692 | * | |
3693 | * A special "rbd id" object is used to map an rbd image name to its | |
3694 | * id. If that object doesn't exist, then there is no v2 rbd image | |
3695 | * with the supplied name. | |
3696 | * | |
3697 | * This function will record the given rbd_dev's image_id field if | |
3698 | * it can be determined, and in that case will return 0. If any | |
3699 | * errors occur a negative errno will be returned and the rbd_dev's | |
3700 | * image_id field will be unchanged (and should be NULL). | |
3701 | */ | |
3702 | static int rbd_dev_image_id(struct rbd_device *rbd_dev) | |
3703 | { | |
3704 | int ret; | |
3705 | size_t size; | |
3706 | char *object_name; | |
3707 | void *response; | |
3708 | void *p; | |
3709 | ||
2c0d0a10 AE |
3710 | /* |
3711 | * When probing a parent image, the image id is already | |
3712 | * known (and the image name likely is not). There's no | |
3713 | * need to fetch the image id again in this case. | |
3714 | */ | |
3715 | if (rbd_dev->spec->image_id) | |
3716 | return 0; | |
3717 | ||
589d30e0 AE |
3718 | /* |
3719 | * First, see if the format 2 image id file exists, and if | |
3720 | * so, get the image's persistent id from it. | |
3721 | */ | |
69e7a02f | 3722 | size = sizeof (RBD_ID_PREFIX) + strlen(rbd_dev->spec->image_name); |
589d30e0 AE |
3723 | object_name = kmalloc(size, GFP_NOIO); |
3724 | if (!object_name) | |
3725 | return -ENOMEM; | |
0d7dbfce | 3726 | sprintf(object_name, "%s%s", RBD_ID_PREFIX, rbd_dev->spec->image_name); |
589d30e0 AE |
3727 | dout("rbd id object name is %s\n", object_name); |
3728 | ||
3729 | /* Response will be an encoded string, which includes a length */ | |
3730 | ||
3731 | size = sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX; | |
3732 | response = kzalloc(size, GFP_NOIO); | |
3733 | if (!response) { | |
3734 | ret = -ENOMEM; | |
3735 | goto out; | |
3736 | } | |
3737 | ||
36be9a76 | 3738 | ret = rbd_obj_method_sync(rbd_dev, object_name, |
589d30e0 AE |
3739 | "rbd", "get_id", |
3740 | NULL, 0, | |
07b2391f | 3741 | response, RBD_IMAGE_ID_LEN_MAX, NULL); |
36be9a76 | 3742 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
589d30e0 AE |
3743 | if (ret < 0) |
3744 | goto out; | |
3745 | ||
3746 | p = response; | |
0d7dbfce | 3747 | rbd_dev->spec->image_id = ceph_extract_encoded_string(&p, |
589d30e0 | 3748 | p + RBD_IMAGE_ID_LEN_MAX, |
979ed480 | 3749 | NULL, GFP_NOIO); |
0d7dbfce AE |
3750 | if (IS_ERR(rbd_dev->spec->image_id)) { |
3751 | ret = PTR_ERR(rbd_dev->spec->image_id); | |
3752 | rbd_dev->spec->image_id = NULL; | |
589d30e0 | 3753 | } else { |
0d7dbfce | 3754 | dout("image_id is %s\n", rbd_dev->spec->image_id); |
589d30e0 AE |
3755 | } |
3756 | out: | |
3757 | kfree(response); | |
3758 | kfree(object_name); | |
3759 | ||
3760 | return ret; | |
3761 | } | |
3762 | ||
a30b71b9 AE |
3763 | static int rbd_dev_v1_probe(struct rbd_device *rbd_dev) |
3764 | { | |
3765 | int ret; | |
3766 | size_t size; | |
3767 | ||
3768 | /* Version 1 images have no id; empty string is used */ | |
3769 | ||
0d7dbfce AE |
3770 | rbd_dev->spec->image_id = kstrdup("", GFP_KERNEL); |
3771 | if (!rbd_dev->spec->image_id) | |
a30b71b9 | 3772 | return -ENOMEM; |
a30b71b9 AE |
3773 | |
3774 | /* Record the header object name for this rbd image. */ | |
3775 | ||
69e7a02f | 3776 | size = strlen(rbd_dev->spec->image_name) + sizeof (RBD_SUFFIX); |
a30b71b9 AE |
3777 | rbd_dev->header_name = kmalloc(size, GFP_KERNEL); |
3778 | if (!rbd_dev->header_name) { | |
3779 | ret = -ENOMEM; | |
3780 | goto out_err; | |
3781 | } | |
0d7dbfce AE |
3782 | sprintf(rbd_dev->header_name, "%s%s", |
3783 | rbd_dev->spec->image_name, RBD_SUFFIX); | |
a30b71b9 AE |
3784 | |
3785 | /* Populate rbd image metadata */ | |
3786 | ||
3787 | ret = rbd_read_header(rbd_dev, &rbd_dev->header); | |
3788 | if (ret < 0) | |
3789 | goto out_err; | |
86b00e0d AE |
3790 | |
3791 | /* Version 1 images have no parent (no layering) */ | |
3792 | ||
3793 | rbd_dev->parent_spec = NULL; | |
3794 | rbd_dev->parent_overlap = 0; | |
3795 | ||
a30b71b9 AE |
3796 | rbd_dev->image_format = 1; |
3797 | ||
3798 | dout("discovered version 1 image, header name is %s\n", | |
3799 | rbd_dev->header_name); | |
3800 | ||
3801 | return 0; | |
3802 | ||
3803 | out_err: | |
3804 | kfree(rbd_dev->header_name); | |
3805 | rbd_dev->header_name = NULL; | |
0d7dbfce AE |
3806 | kfree(rbd_dev->spec->image_id); |
3807 | rbd_dev->spec->image_id = NULL; | |
a30b71b9 AE |
3808 | |
3809 | return ret; | |
3810 | } | |
3811 | ||
3812 | static int rbd_dev_v2_probe(struct rbd_device *rbd_dev) | |
3813 | { | |
3814 | size_t size; | |
9d475de5 | 3815 | int ret; |
6e14b1a6 | 3816 | u64 ver = 0; |
a30b71b9 AE |
3817 | |
3818 | /* | |
3819 | * Image id was filled in by the caller. Record the header | |
3820 | * object name for this rbd image. | |
3821 | */ | |
979ed480 | 3822 | size = sizeof (RBD_HEADER_PREFIX) + strlen(rbd_dev->spec->image_id); |
a30b71b9 AE |
3823 | rbd_dev->header_name = kmalloc(size, GFP_KERNEL); |
3824 | if (!rbd_dev->header_name) | |
3825 | return -ENOMEM; | |
3826 | sprintf(rbd_dev->header_name, "%s%s", | |
0d7dbfce | 3827 | RBD_HEADER_PREFIX, rbd_dev->spec->image_id); |
9d475de5 AE |
3828 | |
3829 | /* Get the size and object order for the image */ | |
3830 | ||
3831 | ret = rbd_dev_v2_image_size(rbd_dev); | |
1e130199 AE |
3832 | if (ret < 0) |
3833 | goto out_err; | |
3834 | ||
3835 | /* Get the object prefix (a.k.a. block_name) for the image */ | |
3836 | ||
3837 | ret = rbd_dev_v2_object_prefix(rbd_dev); | |
b1b5402a AE |
3838 | if (ret < 0) |
3839 | goto out_err; | |
3840 | ||
d889140c | 3841 | /* Get the and check features for the image */ |
b1b5402a AE |
3842 | |
3843 | ret = rbd_dev_v2_features(rbd_dev); | |
9d475de5 AE |
3844 | if (ret < 0) |
3845 | goto out_err; | |
35d489f9 | 3846 | |
86b00e0d AE |
3847 | /* If the image supports layering, get the parent info */ |
3848 | ||
3849 | if (rbd_dev->header.features & RBD_FEATURE_LAYERING) { | |
3850 | ret = rbd_dev_v2_parent_info(rbd_dev); | |
3851 | if (ret < 0) | |
3852 | goto out_err; | |
3853 | } | |
3854 | ||
6e14b1a6 AE |
3855 | /* crypto and compression type aren't (yet) supported for v2 images */ |
3856 | ||
3857 | rbd_dev->header.crypt_type = 0; | |
3858 | rbd_dev->header.comp_type = 0; | |
35d489f9 | 3859 | |
6e14b1a6 AE |
3860 | /* Get the snapshot context, plus the header version */ |
3861 | ||
3862 | ret = rbd_dev_v2_snap_context(rbd_dev, &ver); | |
35d489f9 AE |
3863 | if (ret) |
3864 | goto out_err; | |
6e14b1a6 AE |
3865 | rbd_dev->header.obj_version = ver; |
3866 | ||
a30b71b9 AE |
3867 | rbd_dev->image_format = 2; |
3868 | ||
3869 | dout("discovered version 2 image, header name is %s\n", | |
3870 | rbd_dev->header_name); | |
3871 | ||
35152979 | 3872 | return 0; |
9d475de5 | 3873 | out_err: |
86b00e0d AE |
3874 | rbd_dev->parent_overlap = 0; |
3875 | rbd_spec_put(rbd_dev->parent_spec); | |
3876 | rbd_dev->parent_spec = NULL; | |
9d475de5 AE |
3877 | kfree(rbd_dev->header_name); |
3878 | rbd_dev->header_name = NULL; | |
1e130199 AE |
3879 | kfree(rbd_dev->header.object_prefix); |
3880 | rbd_dev->header.object_prefix = NULL; | |
9d475de5 AE |
3881 | |
3882 | return ret; | |
a30b71b9 AE |
3883 | } |
3884 | ||
83a06263 AE |
3885 | static int rbd_dev_probe_finish(struct rbd_device *rbd_dev) |
3886 | { | |
3887 | int ret; | |
3888 | ||
3889 | /* no need to lock here, as rbd_dev is not registered yet */ | |
3890 | ret = rbd_dev_snaps_update(rbd_dev); | |
3891 | if (ret) | |
3892 | return ret; | |
3893 | ||
9e15b77d AE |
3894 | ret = rbd_dev_probe_update_spec(rbd_dev); |
3895 | if (ret) | |
3896 | goto err_out_snaps; | |
3897 | ||
83a06263 AE |
3898 | ret = rbd_dev_set_mapping(rbd_dev); |
3899 | if (ret) | |
3900 | goto err_out_snaps; | |
3901 | ||
3902 | /* generate unique id: find highest unique id, add one */ | |
3903 | rbd_dev_id_get(rbd_dev); | |
3904 | ||
3905 | /* Fill in the device name, now that we have its id. */ | |
3906 | BUILD_BUG_ON(DEV_NAME_LEN | |
3907 | < sizeof (RBD_DRV_NAME) + MAX_INT_FORMAT_WIDTH); | |
3908 | sprintf(rbd_dev->name, "%s%d", RBD_DRV_NAME, rbd_dev->dev_id); | |
3909 | ||
3910 | /* Get our block major device number. */ | |
3911 | ||
3912 | ret = register_blkdev(0, rbd_dev->name); | |
3913 | if (ret < 0) | |
3914 | goto err_out_id; | |
3915 | rbd_dev->major = ret; | |
3916 | ||
3917 | /* Set up the blkdev mapping. */ | |
3918 | ||
3919 | ret = rbd_init_disk(rbd_dev); | |
3920 | if (ret) | |
3921 | goto err_out_blkdev; | |
3922 | ||
3923 | ret = rbd_bus_add_dev(rbd_dev); | |
3924 | if (ret) | |
3925 | goto err_out_disk; | |
3926 | ||
3927 | /* | |
3928 | * At this point cleanup in the event of an error is the job | |
3929 | * of the sysfs code (initiated by rbd_bus_del_dev()). | |
3930 | */ | |
3931 | down_write(&rbd_dev->header_rwsem); | |
3932 | ret = rbd_dev_snaps_register(rbd_dev); | |
3933 | up_write(&rbd_dev->header_rwsem); | |
3934 | if (ret) | |
3935 | goto err_out_bus; | |
3936 | ||
9969ebc5 | 3937 | ret = rbd_dev_header_watch_sync(rbd_dev, 1); |
83a06263 AE |
3938 | if (ret) |
3939 | goto err_out_bus; | |
3940 | ||
3941 | /* Everything's ready. Announce the disk to the world. */ | |
3942 | ||
3943 | add_disk(rbd_dev->disk); | |
3944 | ||
3945 | pr_info("%s: added with size 0x%llx\n", rbd_dev->disk->disk_name, | |
3946 | (unsigned long long) rbd_dev->mapping.size); | |
3947 | ||
3948 | return ret; | |
3949 | err_out_bus: | |
3950 | /* this will also clean up rest of rbd_dev stuff */ | |
3951 | ||
3952 | rbd_bus_del_dev(rbd_dev); | |
3953 | ||
3954 | return ret; | |
3955 | err_out_disk: | |
3956 | rbd_free_disk(rbd_dev); | |
3957 | err_out_blkdev: | |
3958 | unregister_blkdev(rbd_dev->major, rbd_dev->name); | |
3959 | err_out_id: | |
3960 | rbd_dev_id_put(rbd_dev); | |
3961 | err_out_snaps: | |
3962 | rbd_remove_all_snaps(rbd_dev); | |
3963 | ||
3964 | return ret; | |
3965 | } | |
3966 | ||
a30b71b9 AE |
3967 | /* |
3968 | * Probe for the existence of the header object for the given rbd | |
3969 | * device. For format 2 images this includes determining the image | |
3970 | * id. | |
3971 | */ | |
3972 | static int rbd_dev_probe(struct rbd_device *rbd_dev) | |
3973 | { | |
3974 | int ret; | |
3975 | ||
3976 | /* | |
3977 | * Get the id from the image id object. If it's not a | |
3978 | * format 2 image, we'll get ENOENT back, and we'll assume | |
3979 | * it's a format 1 image. | |
3980 | */ | |
3981 | ret = rbd_dev_image_id(rbd_dev); | |
3982 | if (ret) | |
3983 | ret = rbd_dev_v1_probe(rbd_dev); | |
3984 | else | |
3985 | ret = rbd_dev_v2_probe(rbd_dev); | |
83a06263 | 3986 | if (ret) { |
a30b71b9 AE |
3987 | dout("probe failed, returning %d\n", ret); |
3988 | ||
83a06263 AE |
3989 | return ret; |
3990 | } | |
3991 | ||
3992 | ret = rbd_dev_probe_finish(rbd_dev); | |
3993 | if (ret) | |
3994 | rbd_header_free(&rbd_dev->header); | |
3995 | ||
a30b71b9 AE |
3996 | return ret; |
3997 | } | |
3998 | ||
59c2be1e YS |
3999 | static ssize_t rbd_add(struct bus_type *bus, |
4000 | const char *buf, | |
4001 | size_t count) | |
602adf40 | 4002 | { |
cb8627c7 | 4003 | struct rbd_device *rbd_dev = NULL; |
dc79b113 | 4004 | struct ceph_options *ceph_opts = NULL; |
4e9afeba | 4005 | struct rbd_options *rbd_opts = NULL; |
859c31df | 4006 | struct rbd_spec *spec = NULL; |
9d3997fd | 4007 | struct rbd_client *rbdc; |
27cc2594 AE |
4008 | struct ceph_osd_client *osdc; |
4009 | int rc = -ENOMEM; | |
602adf40 YS |
4010 | |
4011 | if (!try_module_get(THIS_MODULE)) | |
4012 | return -ENODEV; | |
4013 | ||
602adf40 | 4014 | /* parse add command */ |
859c31df | 4015 | rc = rbd_add_parse_args(buf, &ceph_opts, &rbd_opts, &spec); |
dc79b113 | 4016 | if (rc < 0) |
bd4ba655 | 4017 | goto err_out_module; |
78cea76e | 4018 | |
9d3997fd AE |
4019 | rbdc = rbd_get_client(ceph_opts); |
4020 | if (IS_ERR(rbdc)) { | |
4021 | rc = PTR_ERR(rbdc); | |
0ddebc0c | 4022 | goto err_out_args; |
9d3997fd | 4023 | } |
c53d5893 | 4024 | ceph_opts = NULL; /* rbd_dev client now owns this */ |
602adf40 | 4025 | |
602adf40 | 4026 | /* pick the pool */ |
9d3997fd | 4027 | osdc = &rbdc->client->osdc; |
859c31df | 4028 | rc = ceph_pg_poolid_by_name(osdc->osdmap, spec->pool_name); |
602adf40 YS |
4029 | if (rc < 0) |
4030 | goto err_out_client; | |
859c31df AE |
4031 | spec->pool_id = (u64) rc; |
4032 | ||
0903e875 AE |
4033 | /* The ceph file layout needs to fit pool id in 32 bits */ |
4034 | ||
4035 | if (WARN_ON(spec->pool_id > (u64) U32_MAX)) { | |
4036 | rc = -EIO; | |
4037 | goto err_out_client; | |
4038 | } | |
4039 | ||
c53d5893 | 4040 | rbd_dev = rbd_dev_create(rbdc, spec); |
bd4ba655 AE |
4041 | if (!rbd_dev) |
4042 | goto err_out_client; | |
c53d5893 AE |
4043 | rbdc = NULL; /* rbd_dev now owns this */ |
4044 | spec = NULL; /* rbd_dev now owns this */ | |
602adf40 | 4045 | |
bd4ba655 | 4046 | rbd_dev->mapping.read_only = rbd_opts->read_only; |
c53d5893 AE |
4047 | kfree(rbd_opts); |
4048 | rbd_opts = NULL; /* done with this */ | |
bd4ba655 | 4049 | |
a30b71b9 AE |
4050 | rc = rbd_dev_probe(rbd_dev); |
4051 | if (rc < 0) | |
c53d5893 | 4052 | goto err_out_rbd_dev; |
05fd6f6f | 4053 | |
602adf40 | 4054 | return count; |
c53d5893 AE |
4055 | err_out_rbd_dev: |
4056 | rbd_dev_destroy(rbd_dev); | |
bd4ba655 | 4057 | err_out_client: |
9d3997fd | 4058 | rbd_put_client(rbdc); |
0ddebc0c | 4059 | err_out_args: |
78cea76e AE |
4060 | if (ceph_opts) |
4061 | ceph_destroy_options(ceph_opts); | |
4e9afeba | 4062 | kfree(rbd_opts); |
859c31df | 4063 | rbd_spec_put(spec); |
bd4ba655 AE |
4064 | err_out_module: |
4065 | module_put(THIS_MODULE); | |
27cc2594 | 4066 | |
602adf40 | 4067 | dout("Error adding device %s\n", buf); |
27cc2594 AE |
4068 | |
4069 | return (ssize_t) rc; | |
602adf40 YS |
4070 | } |
4071 | ||
de71a297 | 4072 | static struct rbd_device *__rbd_get_dev(unsigned long dev_id) |
602adf40 YS |
4073 | { |
4074 | struct list_head *tmp; | |
4075 | struct rbd_device *rbd_dev; | |
4076 | ||
e124a82f | 4077 | spin_lock(&rbd_dev_list_lock); |
602adf40 YS |
4078 | list_for_each(tmp, &rbd_dev_list) { |
4079 | rbd_dev = list_entry(tmp, struct rbd_device, node); | |
de71a297 | 4080 | if (rbd_dev->dev_id == dev_id) { |
e124a82f | 4081 | spin_unlock(&rbd_dev_list_lock); |
602adf40 | 4082 | return rbd_dev; |
e124a82f | 4083 | } |
602adf40 | 4084 | } |
e124a82f | 4085 | spin_unlock(&rbd_dev_list_lock); |
602adf40 YS |
4086 | return NULL; |
4087 | } | |
4088 | ||
dfc5606d | 4089 | static void rbd_dev_release(struct device *dev) |
602adf40 | 4090 | { |
593a9e7b | 4091 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
602adf40 | 4092 | |
59c2be1e | 4093 | if (rbd_dev->watch_event) |
9969ebc5 | 4094 | rbd_dev_header_watch_sync(rbd_dev, 0); |
602adf40 YS |
4095 | |
4096 | /* clean up and free blkdev */ | |
4097 | rbd_free_disk(rbd_dev); | |
4098 | unregister_blkdev(rbd_dev->major, rbd_dev->name); | |
32eec68d | 4099 | |
2ac4e75d AE |
4100 | /* release allocated disk header fields */ |
4101 | rbd_header_free(&rbd_dev->header); | |
4102 | ||
32eec68d | 4103 | /* done with the id, and with the rbd_dev */ |
e2839308 | 4104 | rbd_dev_id_put(rbd_dev); |
c53d5893 AE |
4105 | rbd_assert(rbd_dev->rbd_client != NULL); |
4106 | rbd_dev_destroy(rbd_dev); | |
602adf40 YS |
4107 | |
4108 | /* release module ref */ | |
4109 | module_put(THIS_MODULE); | |
602adf40 YS |
4110 | } |
4111 | ||
dfc5606d YS |
4112 | static ssize_t rbd_remove(struct bus_type *bus, |
4113 | const char *buf, | |
4114 | size_t count) | |
602adf40 YS |
4115 | { |
4116 | struct rbd_device *rbd_dev = NULL; | |
4117 | int target_id, rc; | |
4118 | unsigned long ul; | |
4119 | int ret = count; | |
4120 | ||
4121 | rc = strict_strtoul(buf, 10, &ul); | |
4122 | if (rc) | |
4123 | return rc; | |
4124 | ||
4125 | /* convert to int; abort if we lost anything in the conversion */ | |
4126 | target_id = (int) ul; | |
4127 | if (target_id != ul) | |
4128 | return -EINVAL; | |
4129 | ||
4130 | mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING); | |
4131 | ||
4132 | rbd_dev = __rbd_get_dev(target_id); | |
4133 | if (!rbd_dev) { | |
4134 | ret = -ENOENT; | |
4135 | goto done; | |
42382b70 AE |
4136 | } |
4137 | ||
a14ea269 | 4138 | spin_lock_irq(&rbd_dev->lock); |
b82d167b | 4139 | if (rbd_dev->open_count) |
42382b70 | 4140 | ret = -EBUSY; |
b82d167b AE |
4141 | else |
4142 | set_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags); | |
a14ea269 | 4143 | spin_unlock_irq(&rbd_dev->lock); |
b82d167b | 4144 | if (ret < 0) |
42382b70 | 4145 | goto done; |
602adf40 | 4146 | |
41f38c2b | 4147 | rbd_remove_all_snaps(rbd_dev); |
dfc5606d | 4148 | rbd_bus_del_dev(rbd_dev); |
602adf40 YS |
4149 | |
4150 | done: | |
4151 | mutex_unlock(&ctl_mutex); | |
aafb230e | 4152 | |
602adf40 YS |
4153 | return ret; |
4154 | } | |
4155 | ||
602adf40 YS |
4156 | /* |
4157 | * create control files in sysfs | |
dfc5606d | 4158 | * /sys/bus/rbd/... |
602adf40 YS |
4159 | */ |
4160 | static int rbd_sysfs_init(void) | |
4161 | { | |
dfc5606d | 4162 | int ret; |
602adf40 | 4163 | |
fed4c143 | 4164 | ret = device_register(&rbd_root_dev); |
21079786 | 4165 | if (ret < 0) |
dfc5606d | 4166 | return ret; |
602adf40 | 4167 | |
fed4c143 AE |
4168 | ret = bus_register(&rbd_bus_type); |
4169 | if (ret < 0) | |
4170 | device_unregister(&rbd_root_dev); | |
602adf40 | 4171 | |
602adf40 YS |
4172 | return ret; |
4173 | } | |
4174 | ||
4175 | static void rbd_sysfs_cleanup(void) | |
4176 | { | |
dfc5606d | 4177 | bus_unregister(&rbd_bus_type); |
fed4c143 | 4178 | device_unregister(&rbd_root_dev); |
602adf40 YS |
4179 | } |
4180 | ||
cc344fa1 | 4181 | static int __init rbd_init(void) |
602adf40 YS |
4182 | { |
4183 | int rc; | |
4184 | ||
1e32d34c AE |
4185 | if (!libceph_compatible(NULL)) { |
4186 | rbd_warn(NULL, "libceph incompatibility (quitting)"); | |
4187 | ||
4188 | return -EINVAL; | |
4189 | } | |
602adf40 YS |
4190 | rc = rbd_sysfs_init(); |
4191 | if (rc) | |
4192 | return rc; | |
f0f8cef5 | 4193 | pr_info("loaded " RBD_DRV_NAME_LONG "\n"); |
602adf40 YS |
4194 | return 0; |
4195 | } | |
4196 | ||
cc344fa1 | 4197 | static void __exit rbd_exit(void) |
602adf40 YS |
4198 | { |
4199 | rbd_sysfs_cleanup(); | |
4200 | } | |
4201 | ||
4202 | module_init(rbd_init); | |
4203 | module_exit(rbd_exit); | |
4204 | ||
4205 | MODULE_AUTHOR("Sage Weil <sage@newdream.net>"); | |
4206 | MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>"); | |
4207 | MODULE_DESCRIPTION("rados block device"); | |
4208 | ||
4209 | /* following authorship retained from original osdblk.c */ | |
4210 | MODULE_AUTHOR("Jeff Garzik <jeff@garzik.org>"); | |
4211 | ||
4212 | MODULE_LICENSE("GPL"); |