2 rbd.c -- Export ceph rados objects as a Linux block device
5 based on drivers/block/osdblk.c:
7 Copyright 2009 Red Hat, Inc.
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.
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.
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.
24 For usage instructions, please refer to:
26 Documentation/ABI/testing/sysfs-bus-rbd
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>
34 #include <linux/parser.h>
36 #include <linux/kernel.h>
37 #include <linux/device.h>
38 #include <linux/module.h>
40 #include <linux/blkdev.h>
42 #include "rbd_types.h"
44 #define RBD_DEBUG /* Activate rbd_assert() calls */
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.
52 #define SECTOR_SHIFT 9
53 #define SECTOR_SIZE (1ULL << SECTOR_SHIFT)
55 /* It might be useful to have these defined elsewhere */
57 #define U8_MAX ((u8) (~0U))
58 #define U16_MAX ((u16) (~0U))
59 #define U32_MAX ((u32) (~0U))
60 #define U64_MAX ((u64) (~0ULL))
62 #define RBD_DRV_NAME "rbd"
63 #define RBD_DRV_NAME_LONG "rbd (rados block device)"
65 #define RBD_MINORS_PER_MAJOR 256 /* max minors per blkdev */
67 #define RBD_SNAP_DEV_NAME_PREFIX "snap_"
68 #define RBD_MAX_SNAP_NAME_LEN \
69 (NAME_MAX - (sizeof (RBD_SNAP_DEV_NAME_PREFIX) - 1))
71 #define RBD_MAX_SNAP_COUNT 510 /* allows max snapc to fit in 4KB */
73 #define RBD_SNAP_HEAD_NAME "-"
75 /* This allows a single page to hold an image name sent by OSD */
76 #define RBD_IMAGE_NAME_LEN_MAX (PAGE_SIZE - sizeof (__le32) - 1)
77 #define RBD_IMAGE_ID_LEN_MAX 64
79 #define RBD_OBJ_PREFIX_LEN_MAX 64
83 #define RBD_FEATURE_LAYERING 1
85 /* Features supported by this (client software) implementation. */
87 #define RBD_FEATURES_ALL (0)
90 * An RBD device name will be "rbd#", where the "rbd" comes from
91 * RBD_DRV_NAME above, and # is a unique integer identifier.
92 * MAX_INT_FORMAT_WIDTH is used in ensuring DEV_NAME_LEN is big
93 * enough to hold all possible device names.
95 #define DEV_NAME_LEN 32
96 #define MAX_INT_FORMAT_WIDTH ((5 * sizeof (int)) / 2 + 1)
99 * block device image metadata (in-memory version)
101 struct rbd_image_header
{
102 /* These four fields never change for a given rbd image */
109 /* The remaining fields need to be updated occasionally */
111 struct ceph_snap_context
*snapc
;
119 * An rbd image specification.
121 * The tuple (pool_id, image_id, snap_id) is sufficient to uniquely
122 * identify an image. Each rbd_dev structure includes a pointer to
123 * an rbd_spec structure that encapsulates this identity.
125 * Each of the id's in an rbd_spec has an associated name. For a
126 * user-mapped image, the names are supplied and the id's associated
127 * with them are looked up. For a layered image, a parent image is
128 * defined by the tuple, and the names are looked up.
130 * An rbd_dev structure contains a parent_spec pointer which is
131 * non-null if the image it represents is a child in a layered
132 * image. This pointer will refer to the rbd_spec structure used
133 * by the parent rbd_dev for its own identity (i.e., the structure
134 * is shared between the parent and child).
136 * Since these structures are populated once, during the discovery
137 * phase of image construction, they are effectively immutable so
138 * we make no effort to synchronize access to them.
140 * Note that code herein does not assume the image name is known (it
141 * could be a null pointer).
157 * an instance of the client. multiple devices may share an rbd client.
160 struct ceph_client
*client
;
162 struct list_head node
;
165 struct rbd_img_request
;
166 typedef void (*rbd_img_callback_t
)(struct rbd_img_request
*);
168 #define BAD_WHICH U32_MAX /* Good which or bad which, which? */
170 struct rbd_obj_request
;
171 typedef void (*rbd_obj_callback_t
)(struct rbd_obj_request
*);
173 enum obj_request_type
{
174 OBJ_REQUEST_NODATA
, OBJ_REQUEST_BIO
, OBJ_REQUEST_PAGES
177 struct rbd_obj_request
{
178 const char *object_name
;
179 u64 offset
; /* object start byte */
180 u64 length
; /* bytes from offset */
182 struct rbd_img_request
*img_request
;
183 struct list_head links
; /* img_request->obj_requests */
184 u32 which
; /* posn image request list */
186 enum obj_request_type type
;
188 struct bio
*bio_list
;
195 struct ceph_osd_request
*osd_req
;
197 u64 xferred
; /* bytes transferred */
202 rbd_obj_callback_t callback
;
203 struct completion completion
;
208 struct rbd_img_request
{
210 struct rbd_device
*rbd_dev
;
211 u64 offset
; /* starting image byte offset */
212 u64 length
; /* byte count from offset */
213 bool write_request
; /* false for read */
215 struct ceph_snap_context
*snapc
; /* for writes */
216 u64 snap_id
; /* for reads */
218 spinlock_t completion_lock
;/* protects next_completion */
220 rbd_img_callback_t callback
;
222 u32 obj_request_count
;
223 struct list_head obj_requests
; /* rbd_obj_request structs */
228 #define for_each_obj_request(ireq, oreq) \
229 list_for_each_entry(oreq, &ireq->obj_requests, links)
230 #define for_each_obj_request_from(ireq, oreq) \
231 list_for_each_entry_from(oreq, &ireq->obj_requests, links)
232 #define for_each_obj_request_safe(ireq, oreq, n) \
233 list_for_each_entry_safe_reverse(oreq, n, &ireq->obj_requests, links)
239 struct list_head node
;
254 int dev_id
; /* blkdev unique id */
256 int major
; /* blkdev assigned major */
257 struct gendisk
*disk
; /* blkdev's gendisk and rq */
259 u32 image_format
; /* Either 1 or 2 */
260 struct rbd_client
*rbd_client
;
262 char name
[DEV_NAME_LEN
]; /* blkdev name, e.g. rbd3 */
264 spinlock_t lock
; /* queue, flags, open_count */
266 struct rbd_image_header header
;
267 unsigned long flags
; /* possibly lock protected */
268 struct rbd_spec
*spec
;
272 struct ceph_file_layout layout
;
274 struct ceph_osd_event
*watch_event
;
275 struct rbd_obj_request
*watch_request
;
277 struct rbd_spec
*parent_spec
;
280 /* protects updating the header */
281 struct rw_semaphore header_rwsem
;
283 struct rbd_mapping mapping
;
285 struct list_head node
;
287 /* list of snapshots */
288 struct list_head snaps
;
292 unsigned long open_count
; /* protected by lock */
296 * Flag bits for rbd_dev->flags. If atomicity is required,
297 * rbd_dev->lock is used to protect access.
299 * Currently, only the "removing" flag (which is coupled with the
300 * "open_count" field) requires atomic access.
303 RBD_DEV_FLAG_EXISTS
, /* mapped snapshot has not been deleted */
304 RBD_DEV_FLAG_REMOVING
, /* this mapping is being removed */
307 static DEFINE_MUTEX(ctl_mutex
); /* Serialize open/close/setup/teardown */
309 static LIST_HEAD(rbd_dev_list
); /* devices */
310 static DEFINE_SPINLOCK(rbd_dev_list_lock
);
312 static LIST_HEAD(rbd_client_list
); /* clients */
313 static DEFINE_SPINLOCK(rbd_client_list_lock
);
315 static int rbd_dev_snaps_update(struct rbd_device
*rbd_dev
);
316 static int rbd_dev_snaps_register(struct rbd_device
*rbd_dev
);
318 static void rbd_dev_release(struct device
*dev
);
319 static void rbd_remove_snap_dev(struct rbd_snap
*snap
);
321 static ssize_t
rbd_add(struct bus_type
*bus
, const char *buf
,
323 static ssize_t
rbd_remove(struct bus_type
*bus
, const char *buf
,
326 static struct bus_attribute rbd_bus_attrs
[] = {
327 __ATTR(add
, S_IWUSR
, NULL
, rbd_add
),
328 __ATTR(remove
, S_IWUSR
, NULL
, rbd_remove
),
332 static struct bus_type rbd_bus_type
= {
334 .bus_attrs
= rbd_bus_attrs
,
337 static void rbd_root_dev_release(struct device
*dev
)
341 static struct device rbd_root_dev
= {
343 .release
= rbd_root_dev_release
,
346 static __printf(2, 3)
347 void rbd_warn(struct rbd_device
*rbd_dev
, const char *fmt
, ...)
349 struct va_format vaf
;
357 printk(KERN_WARNING
"%s: %pV\n", RBD_DRV_NAME
, &vaf
);
358 else if (rbd_dev
->disk
)
359 printk(KERN_WARNING
"%s: %s: %pV\n",
360 RBD_DRV_NAME
, rbd_dev
->disk
->disk_name
, &vaf
);
361 else if (rbd_dev
->spec
&& rbd_dev
->spec
->image_name
)
362 printk(KERN_WARNING
"%s: image %s: %pV\n",
363 RBD_DRV_NAME
, rbd_dev
->spec
->image_name
, &vaf
);
364 else if (rbd_dev
->spec
&& rbd_dev
->spec
->image_id
)
365 printk(KERN_WARNING
"%s: id %s: %pV\n",
366 RBD_DRV_NAME
, rbd_dev
->spec
->image_id
, &vaf
);
368 printk(KERN_WARNING
"%s: rbd_dev %p: %pV\n",
369 RBD_DRV_NAME
, rbd_dev
, &vaf
);
374 #define rbd_assert(expr) \
375 if (unlikely(!(expr))) { \
376 printk(KERN_ERR "\nAssertion failure in %s() " \
378 "\trbd_assert(%s);\n\n", \
379 __func__, __LINE__, #expr); \
382 #else /* !RBD_DEBUG */
383 # define rbd_assert(expr) ((void) 0)
384 #endif /* !RBD_DEBUG */
386 static int rbd_dev_refresh(struct rbd_device
*rbd_dev
, u64
*hver
);
387 static int rbd_dev_v2_refresh(struct rbd_device
*rbd_dev
, u64
*hver
);
389 static int rbd_open(struct block_device
*bdev
, fmode_t mode
)
391 struct rbd_device
*rbd_dev
= bdev
->bd_disk
->private_data
;
392 bool removing
= false;
394 if ((mode
& FMODE_WRITE
) && rbd_dev
->mapping
.read_only
)
397 spin_lock(&rbd_dev
->lock
);
398 if (test_bit(RBD_DEV_FLAG_REMOVING
, &rbd_dev
->flags
))
401 rbd_dev
->open_count
++;
402 spin_unlock(&rbd_dev
->lock
);
406 mutex_lock_nested(&ctl_mutex
, SINGLE_DEPTH_NESTING
);
407 (void) get_device(&rbd_dev
->dev
);
408 set_device_ro(bdev
, rbd_dev
->mapping
.read_only
);
409 mutex_unlock(&ctl_mutex
);
414 static int rbd_release(struct gendisk
*disk
, fmode_t mode
)
416 struct rbd_device
*rbd_dev
= disk
->private_data
;
417 unsigned long open_count_before
;
419 spin_lock(&rbd_dev
->lock
);
420 open_count_before
= rbd_dev
->open_count
--;
421 spin_unlock(&rbd_dev
->lock
);
422 rbd_assert(open_count_before
> 0);
424 mutex_lock_nested(&ctl_mutex
, SINGLE_DEPTH_NESTING
);
425 put_device(&rbd_dev
->dev
);
426 mutex_unlock(&ctl_mutex
);
431 static const struct block_device_operations rbd_bd_ops
= {
432 .owner
= THIS_MODULE
,
434 .release
= rbd_release
,
438 * Initialize an rbd client instance.
441 static struct rbd_client
*rbd_client_create(struct ceph_options
*ceph_opts
)
443 struct rbd_client
*rbdc
;
446 dout("rbd_client_create\n");
447 rbdc
= kmalloc(sizeof(struct rbd_client
), GFP_KERNEL
);
451 kref_init(&rbdc
->kref
);
452 INIT_LIST_HEAD(&rbdc
->node
);
454 mutex_lock_nested(&ctl_mutex
, SINGLE_DEPTH_NESTING
);
456 rbdc
->client
= ceph_create_client(ceph_opts
, rbdc
, 0, 0);
457 if (IS_ERR(rbdc
->client
))
459 ceph_opts
= NULL
; /* Now rbdc->client is responsible for ceph_opts */
461 ret
= ceph_open_session(rbdc
->client
);
465 spin_lock(&rbd_client_list_lock
);
466 list_add_tail(&rbdc
->node
, &rbd_client_list
);
467 spin_unlock(&rbd_client_list_lock
);
469 mutex_unlock(&ctl_mutex
);
471 dout("rbd_client_create created %p\n", rbdc
);
475 ceph_destroy_client(rbdc
->client
);
477 mutex_unlock(&ctl_mutex
);
481 ceph_destroy_options(ceph_opts
);
486 * Find a ceph client with specific addr and configuration. If
487 * found, bump its reference count.
489 static struct rbd_client
*rbd_client_find(struct ceph_options
*ceph_opts
)
491 struct rbd_client
*client_node
;
494 if (ceph_opts
->flags
& CEPH_OPT_NOSHARE
)
497 spin_lock(&rbd_client_list_lock
);
498 list_for_each_entry(client_node
, &rbd_client_list
, node
) {
499 if (!ceph_compare_options(ceph_opts
, client_node
->client
)) {
500 kref_get(&client_node
->kref
);
505 spin_unlock(&rbd_client_list_lock
);
507 return found
? client_node
: NULL
;
517 /* string args above */
520 /* Boolean args above */
524 static match_table_t rbd_opts_tokens
= {
526 /* string args above */
527 {Opt_read_only
, "read_only"},
528 {Opt_read_only
, "ro"}, /* Alternate spelling */
529 {Opt_read_write
, "read_write"},
530 {Opt_read_write
, "rw"}, /* Alternate spelling */
531 /* Boolean args above */
539 #define RBD_READ_ONLY_DEFAULT false
541 static int parse_rbd_opts_token(char *c
, void *private)
543 struct rbd_options
*rbd_opts
= private;
544 substring_t argstr
[MAX_OPT_ARGS
];
545 int token
, intval
, ret
;
547 token
= match_token(c
, rbd_opts_tokens
, argstr
);
551 if (token
< Opt_last_int
) {
552 ret
= match_int(&argstr
[0], &intval
);
554 pr_err("bad mount option arg (not int) "
558 dout("got int token %d val %d\n", token
, intval
);
559 } else if (token
> Opt_last_int
&& token
< Opt_last_string
) {
560 dout("got string token %d val %s\n", token
,
562 } else if (token
> Opt_last_string
&& token
< Opt_last_bool
) {
563 dout("got Boolean token %d\n", token
);
565 dout("got token %d\n", token
);
570 rbd_opts
->read_only
= true;
573 rbd_opts
->read_only
= false;
583 * Get a ceph client with specific addr and configuration, if one does
584 * not exist create it.
586 static struct rbd_client
*rbd_get_client(struct ceph_options
*ceph_opts
)
588 struct rbd_client
*rbdc
;
590 rbdc
= rbd_client_find(ceph_opts
);
591 if (rbdc
) /* using an existing client */
592 ceph_destroy_options(ceph_opts
);
594 rbdc
= rbd_client_create(ceph_opts
);
600 * Destroy ceph client
602 * Caller must hold rbd_client_list_lock.
604 static void rbd_client_release(struct kref
*kref
)
606 struct rbd_client
*rbdc
= container_of(kref
, struct rbd_client
, kref
);
608 dout("rbd_release_client %p\n", rbdc
);
609 spin_lock(&rbd_client_list_lock
);
610 list_del(&rbdc
->node
);
611 spin_unlock(&rbd_client_list_lock
);
613 ceph_destroy_client(rbdc
->client
);
618 * Drop reference to ceph client node. If it's not referenced anymore, release
621 static void rbd_put_client(struct rbd_client
*rbdc
)
624 kref_put(&rbdc
->kref
, rbd_client_release
);
627 static bool rbd_image_format_valid(u32 image_format
)
629 return image_format
== 1 || image_format
== 2;
632 static bool rbd_dev_ondisk_valid(struct rbd_image_header_ondisk
*ondisk
)
637 /* The header has to start with the magic rbd header text */
638 if (memcmp(&ondisk
->text
, RBD_HEADER_TEXT
, sizeof (RBD_HEADER_TEXT
)))
641 /* The bio layer requires at least sector-sized I/O */
643 if (ondisk
->options
.order
< SECTOR_SHIFT
)
646 /* If we use u64 in a few spots we may be able to loosen this */
648 if (ondisk
->options
.order
> 8 * sizeof (int) - 1)
652 * The size of a snapshot header has to fit in a size_t, and
653 * that limits the number of snapshots.
655 snap_count
= le32_to_cpu(ondisk
->snap_count
);
656 size
= SIZE_MAX
- sizeof (struct ceph_snap_context
);
657 if (snap_count
> size
/ sizeof (__le64
))
661 * Not only that, but the size of the entire the snapshot
662 * header must also be representable in a size_t.
664 size
-= snap_count
* sizeof (__le64
);
665 if ((u64
) size
< le64_to_cpu(ondisk
->snap_names_len
))
672 * Create a new header structure, translate header format from the on-disk
675 static int rbd_header_from_disk(struct rbd_image_header
*header
,
676 struct rbd_image_header_ondisk
*ondisk
)
683 memset(header
, 0, sizeof (*header
));
685 snap_count
= le32_to_cpu(ondisk
->snap_count
);
687 len
= strnlen(ondisk
->object_prefix
, sizeof (ondisk
->object_prefix
));
688 header
->object_prefix
= kmalloc(len
+ 1, GFP_KERNEL
);
689 if (!header
->object_prefix
)
691 memcpy(header
->object_prefix
, ondisk
->object_prefix
, len
);
692 header
->object_prefix
[len
] = '\0';
695 u64 snap_names_len
= le64_to_cpu(ondisk
->snap_names_len
);
697 /* Save a copy of the snapshot names */
699 if (snap_names_len
> (u64
) SIZE_MAX
)
701 header
->snap_names
= kmalloc(snap_names_len
, GFP_KERNEL
);
702 if (!header
->snap_names
)
705 * Note that rbd_dev_v1_header_read() guarantees
706 * the ondisk buffer we're working with has
707 * snap_names_len bytes beyond the end of the
708 * snapshot id array, this memcpy() is safe.
710 memcpy(header
->snap_names
, &ondisk
->snaps
[snap_count
],
713 /* Record each snapshot's size */
715 size
= snap_count
* sizeof (*header
->snap_sizes
);
716 header
->snap_sizes
= kmalloc(size
, GFP_KERNEL
);
717 if (!header
->snap_sizes
)
719 for (i
= 0; i
< snap_count
; i
++)
720 header
->snap_sizes
[i
] =
721 le64_to_cpu(ondisk
->snaps
[i
].image_size
);
723 WARN_ON(ondisk
->snap_names_len
);
724 header
->snap_names
= NULL
;
725 header
->snap_sizes
= NULL
;
728 header
->features
= 0; /* No features support in v1 images */
729 header
->obj_order
= ondisk
->options
.order
;
730 header
->crypt_type
= ondisk
->options
.crypt_type
;
731 header
->comp_type
= ondisk
->options
.comp_type
;
733 /* Allocate and fill in the snapshot context */
735 header
->image_size
= le64_to_cpu(ondisk
->image_size
);
736 size
= sizeof (struct ceph_snap_context
);
737 size
+= snap_count
* sizeof (header
->snapc
->snaps
[0]);
738 header
->snapc
= kzalloc(size
, GFP_KERNEL
);
742 atomic_set(&header
->snapc
->nref
, 1);
743 header
->snapc
->seq
= le64_to_cpu(ondisk
->snap_seq
);
744 header
->snapc
->num_snaps
= snap_count
;
745 for (i
= 0; i
< snap_count
; i
++)
746 header
->snapc
->snaps
[i
] =
747 le64_to_cpu(ondisk
->snaps
[i
].id
);
752 kfree(header
->snap_sizes
);
753 header
->snap_sizes
= NULL
;
754 kfree(header
->snap_names
);
755 header
->snap_names
= NULL
;
756 kfree(header
->object_prefix
);
757 header
->object_prefix
= NULL
;
762 static const char *rbd_snap_name(struct rbd_device
*rbd_dev
, u64 snap_id
)
764 struct rbd_snap
*snap
;
766 if (snap_id
== CEPH_NOSNAP
)
767 return RBD_SNAP_HEAD_NAME
;
769 list_for_each_entry(snap
, &rbd_dev
->snaps
, node
)
770 if (snap_id
== snap
->id
)
776 static int snap_by_name(struct rbd_device
*rbd_dev
, const char *snap_name
)
779 struct rbd_snap
*snap
;
781 list_for_each_entry(snap
, &rbd_dev
->snaps
, node
) {
782 if (!strcmp(snap_name
, snap
->name
)) {
783 rbd_dev
->spec
->snap_id
= snap
->id
;
784 rbd_dev
->mapping
.size
= snap
->size
;
785 rbd_dev
->mapping
.features
= snap
->features
;
794 static int rbd_dev_set_mapping(struct rbd_device
*rbd_dev
)
798 if (!memcmp(rbd_dev
->spec
->snap_name
, RBD_SNAP_HEAD_NAME
,
799 sizeof (RBD_SNAP_HEAD_NAME
))) {
800 rbd_dev
->spec
->snap_id
= CEPH_NOSNAP
;
801 rbd_dev
->mapping
.size
= rbd_dev
->header
.image_size
;
802 rbd_dev
->mapping
.features
= rbd_dev
->header
.features
;
805 ret
= snap_by_name(rbd_dev
, rbd_dev
->spec
->snap_name
);
808 rbd_dev
->mapping
.read_only
= true;
810 set_bit(RBD_DEV_FLAG_EXISTS
, &rbd_dev
->flags
);
816 static void rbd_header_free(struct rbd_image_header
*header
)
818 kfree(header
->object_prefix
);
819 header
->object_prefix
= NULL
;
820 kfree(header
->snap_sizes
);
821 header
->snap_sizes
= NULL
;
822 kfree(header
->snap_names
);
823 header
->snap_names
= NULL
;
824 ceph_put_snap_context(header
->snapc
);
825 header
->snapc
= NULL
;
828 static const char *rbd_segment_name(struct rbd_device
*rbd_dev
, u64 offset
)
834 name
= kmalloc(MAX_OBJ_NAME_SIZE
+ 1, GFP_NOIO
);
837 segment
= offset
>> rbd_dev
->header
.obj_order
;
838 ret
= snprintf(name
, MAX_OBJ_NAME_SIZE
+ 1, "%s.%012llx",
839 rbd_dev
->header
.object_prefix
, segment
);
840 if (ret
< 0 || ret
> MAX_OBJ_NAME_SIZE
) {
841 pr_err("error formatting segment name for #%llu (%d)\n",
850 static u64
rbd_segment_offset(struct rbd_device
*rbd_dev
, u64 offset
)
852 u64 segment_size
= (u64
) 1 << rbd_dev
->header
.obj_order
;
854 return offset
& (segment_size
- 1);
857 static u64
rbd_segment_length(struct rbd_device
*rbd_dev
,
858 u64 offset
, u64 length
)
860 u64 segment_size
= (u64
) 1 << rbd_dev
->header
.obj_order
;
862 offset
&= segment_size
- 1;
864 rbd_assert(length
<= U64_MAX
- offset
);
865 if (offset
+ length
> segment_size
)
866 length
= segment_size
- offset
;
872 * returns the size of an object in the image
874 static u64
rbd_obj_bytes(struct rbd_image_header
*header
)
876 return 1 << header
->obj_order
;
883 static void bio_chain_put(struct bio
*chain
)
889 chain
= chain
->bi_next
;
895 * zeros a bio chain, starting at specific offset
897 static void zero_bio_chain(struct bio
*chain
, int start_ofs
)
906 bio_for_each_segment(bv
, chain
, i
) {
907 if (pos
+ bv
->bv_len
> start_ofs
) {
908 int remainder
= max(start_ofs
- pos
, 0);
909 buf
= bvec_kmap_irq(bv
, &flags
);
910 memset(buf
+ remainder
, 0,
911 bv
->bv_len
- remainder
);
912 bvec_kunmap_irq(buf
, &flags
);
917 chain
= chain
->bi_next
;
922 * Clone a portion of a bio, starting at the given byte offset
923 * and continuing for the number of bytes indicated.
925 static struct bio
*bio_clone_range(struct bio
*bio_src
,
934 unsigned short end_idx
;
938 /* Handle the easy case for the caller */
940 if (!offset
&& len
== bio_src
->bi_size
)
941 return bio_clone(bio_src
, gfpmask
);
943 if (WARN_ON_ONCE(!len
))
945 if (WARN_ON_ONCE(len
> bio_src
->bi_size
))
947 if (WARN_ON_ONCE(offset
> bio_src
->bi_size
- len
))
950 /* Find first affected segment... */
953 __bio_for_each_segment(bv
, bio_src
, idx
, 0) {
954 if (resid
< bv
->bv_len
)
960 /* ...and the last affected segment */
963 __bio_for_each_segment(bv
, bio_src
, end_idx
, idx
) {
964 if (resid
<= bv
->bv_len
)
968 vcnt
= end_idx
- idx
+ 1;
970 /* Build the clone */
972 bio
= bio_alloc(gfpmask
, (unsigned int) vcnt
);
974 return NULL
; /* ENOMEM */
976 bio
->bi_bdev
= bio_src
->bi_bdev
;
977 bio
->bi_sector
= bio_src
->bi_sector
+ (offset
>> SECTOR_SHIFT
);
978 bio
->bi_rw
= bio_src
->bi_rw
;
979 bio
->bi_flags
|= 1 << BIO_CLONED
;
982 * Copy over our part of the bio_vec, then update the first
983 * and last (or only) entries.
985 memcpy(&bio
->bi_io_vec
[0], &bio_src
->bi_io_vec
[idx
],
986 vcnt
* sizeof (struct bio_vec
));
987 bio
->bi_io_vec
[0].bv_offset
+= voff
;
989 bio
->bi_io_vec
[0].bv_len
-= voff
;
990 bio
->bi_io_vec
[vcnt
- 1].bv_len
= resid
;
992 bio
->bi_io_vec
[0].bv_len
= len
;
1003 * Clone a portion of a bio chain, starting at the given byte offset
1004 * into the first bio in the source chain and continuing for the
1005 * number of bytes indicated. The result is another bio chain of
1006 * exactly the given length, or a null pointer on error.
1008 * The bio_src and offset parameters are both in-out. On entry they
1009 * refer to the first source bio and the offset into that bio where
1010 * the start of data to be cloned is located.
1012 * On return, bio_src is updated to refer to the bio in the source
1013 * chain that contains first un-cloned byte, and *offset will
1014 * contain the offset of that byte within that bio.
1016 static struct bio
*bio_chain_clone_range(struct bio
**bio_src
,
1017 unsigned int *offset
,
1021 struct bio
*bi
= *bio_src
;
1022 unsigned int off
= *offset
;
1023 struct bio
*chain
= NULL
;
1026 /* Build up a chain of clone bios up to the limit */
1028 if (!bi
|| off
>= bi
->bi_size
|| !len
)
1029 return NULL
; /* Nothing to clone */
1033 unsigned int bi_size
;
1037 rbd_warn(NULL
, "bio_chain exhausted with %u left", len
);
1038 goto out_err
; /* EINVAL; ran out of bio's */
1040 bi_size
= min_t(unsigned int, bi
->bi_size
- off
, len
);
1041 bio
= bio_clone_range(bi
, off
, bi_size
, gfpmask
);
1043 goto out_err
; /* ENOMEM */
1046 end
= &bio
->bi_next
;
1049 if (off
== bi
->bi_size
) {
1060 bio_chain_put(chain
);
1065 static void rbd_obj_request_get(struct rbd_obj_request
*obj_request
)
1067 kref_get(&obj_request
->kref
);
1070 static void rbd_obj_request_destroy(struct kref
*kref
);
1071 static void rbd_obj_request_put(struct rbd_obj_request
*obj_request
)
1073 rbd_assert(obj_request
!= NULL
);
1074 kref_put(&obj_request
->kref
, rbd_obj_request_destroy
);
1077 static void rbd_img_request_get(struct rbd_img_request
*img_request
)
1079 kref_get(&img_request
->kref
);
1082 static void rbd_img_request_destroy(struct kref
*kref
);
1083 static void rbd_img_request_put(struct rbd_img_request
*img_request
)
1085 rbd_assert(img_request
!= NULL
);
1086 kref_put(&img_request
->kref
, rbd_img_request_destroy
);
1089 static inline void rbd_img_obj_request_add(struct rbd_img_request
*img_request
,
1090 struct rbd_obj_request
*obj_request
)
1092 rbd_assert(obj_request
->img_request
== NULL
);
1094 rbd_obj_request_get(obj_request
);
1095 obj_request
->img_request
= img_request
;
1096 obj_request
->which
= img_request
->obj_request_count
;
1097 rbd_assert(obj_request
->which
!= BAD_WHICH
);
1098 img_request
->obj_request_count
++;
1099 list_add_tail(&obj_request
->links
, &img_request
->obj_requests
);
1102 static inline void rbd_img_obj_request_del(struct rbd_img_request
*img_request
,
1103 struct rbd_obj_request
*obj_request
)
1105 rbd_assert(obj_request
->which
!= BAD_WHICH
);
1107 list_del(&obj_request
->links
);
1108 rbd_assert(img_request
->obj_request_count
> 0);
1109 img_request
->obj_request_count
--;
1110 rbd_assert(obj_request
->which
== img_request
->obj_request_count
);
1111 obj_request
->which
= BAD_WHICH
;
1112 rbd_assert(obj_request
->img_request
== img_request
);
1113 obj_request
->img_request
= NULL
;
1114 obj_request
->callback
= NULL
;
1115 rbd_obj_request_put(obj_request
);
1118 static bool obj_request_type_valid(enum obj_request_type type
)
1121 case OBJ_REQUEST_NODATA
:
1122 case OBJ_REQUEST_BIO
:
1123 case OBJ_REQUEST_PAGES
:
1130 struct ceph_osd_req_op
*rbd_osd_req_op_create(u16 opcode
, ...)
1132 struct ceph_osd_req_op
*op
;
1136 op
= kzalloc(sizeof (*op
), GFP_NOIO
);
1140 va_start(args
, opcode
);
1142 case CEPH_OSD_OP_READ
:
1143 case CEPH_OSD_OP_WRITE
:
1144 /* rbd_osd_req_op_create(READ, offset, length) */
1145 /* rbd_osd_req_op_create(WRITE, offset, length) */
1146 op
->extent
.offset
= va_arg(args
, u64
);
1147 op
->extent
.length
= va_arg(args
, u64
);
1148 if (opcode
== CEPH_OSD_OP_WRITE
)
1149 op
->payload_len
= op
->extent
.length
;
1151 case CEPH_OSD_OP_CALL
:
1152 /* rbd_osd_req_op_create(CALL, class, method, data, datalen) */
1153 op
->cls
.class_name
= va_arg(args
, char *);
1154 size
= strlen(op
->cls
.class_name
);
1155 rbd_assert(size
<= (size_t) U8_MAX
);
1156 op
->cls
.class_len
= size
;
1157 op
->payload_len
= size
;
1159 op
->cls
.method_name
= va_arg(args
, char *);
1160 size
= strlen(op
->cls
.method_name
);
1161 rbd_assert(size
<= (size_t) U8_MAX
);
1162 op
->cls
.method_len
= size
;
1163 op
->payload_len
+= size
;
1166 op
->cls
.indata
= va_arg(args
, void *);
1167 size
= va_arg(args
, size_t);
1168 rbd_assert(size
<= (size_t) U32_MAX
);
1169 op
->cls
.indata_len
= (u32
) size
;
1170 op
->payload_len
+= size
;
1172 case CEPH_OSD_OP_NOTIFY_ACK
:
1173 case CEPH_OSD_OP_WATCH
:
1174 /* rbd_osd_req_op_create(NOTIFY_ACK, cookie, version) */
1175 /* rbd_osd_req_op_create(WATCH, cookie, version, flag) */
1176 op
->watch
.cookie
= va_arg(args
, u64
);
1177 op
->watch
.ver
= va_arg(args
, u64
);
1178 op
->watch
.ver
= cpu_to_le64(op
->watch
.ver
);
1179 if (opcode
== CEPH_OSD_OP_WATCH
&& va_arg(args
, int))
1180 op
->watch
.flag
= (u8
) 1;
1183 rbd_warn(NULL
, "unsupported opcode %hu\n", opcode
);
1193 static void rbd_osd_req_op_destroy(struct ceph_osd_req_op
*op
)
1198 static int rbd_obj_request_submit(struct ceph_osd_client
*osdc
,
1199 struct rbd_obj_request
*obj_request
)
1201 return ceph_osdc_start_request(osdc
, obj_request
->osd_req
, false);
1204 static void rbd_img_request_complete(struct rbd_img_request
*img_request
)
1206 if (img_request
->callback
)
1207 img_request
->callback(img_request
);
1209 rbd_img_request_put(img_request
);
1212 /* Caller is responsible for rbd_obj_request_destroy(obj_request) */
1214 static int rbd_obj_request_wait(struct rbd_obj_request
*obj_request
)
1216 return wait_for_completion_interruptible(&obj_request
->completion
);
1219 static void rbd_osd_trivial_callback(struct rbd_obj_request
*obj_request
,
1220 struct ceph_osd_op
*op
)
1222 atomic_set(&obj_request
->done
, 1);
1225 static void rbd_obj_request_complete(struct rbd_obj_request
*obj_request
)
1227 if (obj_request
->callback
)
1228 obj_request
->callback(obj_request
);
1230 complete_all(&obj_request
->completion
);
1233 static void rbd_osd_read_callback(struct rbd_obj_request
*obj_request
,
1234 struct ceph_osd_op
*op
)
1239 * We support a 64-bit length, but ultimately it has to be
1240 * passed to blk_end_request(), which takes an unsigned int.
1242 xferred
= le64_to_cpu(op
->extent
.length
);
1243 rbd_assert(xferred
< (u64
) UINT_MAX
);
1244 if (obj_request
->result
== (s32
) -ENOENT
) {
1245 zero_bio_chain(obj_request
->bio_list
, 0);
1246 obj_request
->result
= 0;
1247 } else if (xferred
< obj_request
->length
&& !obj_request
->result
) {
1248 zero_bio_chain(obj_request
->bio_list
, xferred
);
1249 xferred
= obj_request
->length
;
1251 obj_request
->xferred
= xferred
;
1252 atomic_set(&obj_request
->done
, 1);
1255 static void rbd_osd_write_callback(struct rbd_obj_request
*obj_request
,
1256 struct ceph_osd_op
*op
)
1258 obj_request
->xferred
= le64_to_cpu(op
->extent
.length
);
1259 atomic_set(&obj_request
->done
, 1);
1262 static void rbd_osd_req_callback(struct ceph_osd_request
*osd_req
,
1263 struct ceph_msg
*msg
)
1265 struct rbd_obj_request
*obj_request
= osd_req
->r_priv
;
1266 struct ceph_osd_reply_head
*reply_head
;
1267 struct ceph_osd_op
*op
;
1271 rbd_assert(osd_req
== obj_request
->osd_req
);
1272 rbd_assert(!!obj_request
->img_request
^
1273 (obj_request
->which
== BAD_WHICH
));
1275 obj_request
->xferred
= le32_to_cpu(msg
->hdr
.data_len
);
1276 reply_head
= msg
->front
.iov_base
;
1277 obj_request
->result
= (s32
) le32_to_cpu(reply_head
->result
);
1278 obj_request
->version
= le64_to_cpu(osd_req
->r_reassert_version
.version
);
1280 num_ops
= le32_to_cpu(reply_head
->num_ops
);
1281 WARN_ON(num_ops
!= 1); /* For now */
1283 op
= &reply_head
->ops
[0];
1284 opcode
= le16_to_cpu(op
->op
);
1286 case CEPH_OSD_OP_READ
:
1287 rbd_osd_read_callback(obj_request
, op
);
1289 case CEPH_OSD_OP_WRITE
:
1290 rbd_osd_write_callback(obj_request
, op
);
1292 case CEPH_OSD_OP_CALL
:
1293 case CEPH_OSD_OP_NOTIFY_ACK
:
1294 case CEPH_OSD_OP_WATCH
:
1295 rbd_osd_trivial_callback(obj_request
, op
);
1298 rbd_warn(NULL
, "%s: unsupported op %hu\n",
1299 obj_request
->object_name
, (unsigned short) opcode
);
1303 if (atomic_read(&obj_request
->done
))
1304 rbd_obj_request_complete(obj_request
);
1307 static struct ceph_osd_request
*rbd_osd_req_create(
1308 struct rbd_device
*rbd_dev
,
1310 struct rbd_obj_request
*obj_request
,
1311 struct ceph_osd_req_op
*op
)
1313 struct rbd_img_request
*img_request
= obj_request
->img_request
;
1314 struct ceph_snap_context
*snapc
= NULL
;
1315 struct ceph_osd_client
*osdc
;
1316 struct ceph_osd_request
*osd_req
;
1317 struct timespec now
;
1318 struct timespec
*mtime
;
1319 u64 snap_id
= CEPH_NOSNAP
;
1320 u64 offset
= obj_request
->offset
;
1321 u64 length
= obj_request
->length
;
1324 rbd_assert(img_request
->write_request
== write_request
);
1325 if (img_request
->write_request
)
1326 snapc
= img_request
->snapc
;
1328 snap_id
= img_request
->snap_id
;
1331 /* Allocate and initialize the request, for the single op */
1333 osdc
= &rbd_dev
->rbd_client
->client
->osdc
;
1334 osd_req
= ceph_osdc_alloc_request(osdc
, snapc
, 1, false, GFP_ATOMIC
);
1336 return NULL
; /* ENOMEM */
1338 rbd_assert(obj_request_type_valid(obj_request
->type
));
1339 switch (obj_request
->type
) {
1340 case OBJ_REQUEST_NODATA
:
1341 break; /* Nothing to do */
1342 case OBJ_REQUEST_BIO
:
1343 rbd_assert(obj_request
->bio_list
!= NULL
);
1344 osd_req
->r_bio
= obj_request
->bio_list
;
1346 case OBJ_REQUEST_PAGES
:
1347 osd_req
->r_pages
= obj_request
->pages
;
1348 osd_req
->r_num_pages
= obj_request
->page_count
;
1349 osd_req
->r_page_alignment
= offset
& ~PAGE_MASK
;
1353 if (write_request
) {
1354 osd_req
->r_flags
= CEPH_OSD_FLAG_WRITE
| CEPH_OSD_FLAG_ONDISK
;
1358 osd_req
->r_flags
= CEPH_OSD_FLAG_READ
;
1359 mtime
= NULL
; /* not needed for reads */
1360 offset
= 0; /* These are not used... */
1361 length
= 0; /* ...for osd read requests */
1364 osd_req
->r_callback
= rbd_osd_req_callback
;
1365 osd_req
->r_priv
= obj_request
;
1367 osd_req
->r_oid_len
= strlen(obj_request
->object_name
);
1368 rbd_assert(osd_req
->r_oid_len
< sizeof (osd_req
->r_oid
));
1369 memcpy(osd_req
->r_oid
, obj_request
->object_name
, osd_req
->r_oid_len
);
1371 osd_req
->r_file_layout
= rbd_dev
->layout
; /* struct */
1373 /* osd_req will get its own reference to snapc (if non-null) */
1375 ceph_osdc_build_request(osd_req
, offset
, length
, 1, op
,
1376 snapc
, snap_id
, mtime
);
1381 static void rbd_osd_req_destroy(struct ceph_osd_request
*osd_req
)
1383 ceph_osdc_put_request(osd_req
);
1386 /* object_name is assumed to be a non-null pointer and NUL-terminated */
1388 static struct rbd_obj_request
*rbd_obj_request_create(const char *object_name
,
1389 u64 offset
, u64 length
,
1390 enum obj_request_type type
)
1392 struct rbd_obj_request
*obj_request
;
1396 rbd_assert(obj_request_type_valid(type
));
1398 size
= strlen(object_name
) + 1;
1399 obj_request
= kzalloc(sizeof (*obj_request
) + size
, GFP_KERNEL
);
1403 name
= (char *)(obj_request
+ 1);
1404 obj_request
->object_name
= memcpy(name
, object_name
, size
);
1405 obj_request
->offset
= offset
;
1406 obj_request
->length
= length
;
1407 obj_request
->which
= BAD_WHICH
;
1408 obj_request
->type
= type
;
1409 INIT_LIST_HEAD(&obj_request
->links
);
1410 atomic_set(&obj_request
->done
, 0);
1411 init_completion(&obj_request
->completion
);
1412 kref_init(&obj_request
->kref
);
1417 static void rbd_obj_request_destroy(struct kref
*kref
)
1419 struct rbd_obj_request
*obj_request
;
1421 obj_request
= container_of(kref
, struct rbd_obj_request
, kref
);
1423 rbd_assert(obj_request
->img_request
== NULL
);
1424 rbd_assert(obj_request
->which
== BAD_WHICH
);
1426 if (obj_request
->osd_req
)
1427 rbd_osd_req_destroy(obj_request
->osd_req
);
1429 rbd_assert(obj_request_type_valid(obj_request
->type
));
1430 switch (obj_request
->type
) {
1431 case OBJ_REQUEST_NODATA
:
1432 break; /* Nothing to do */
1433 case OBJ_REQUEST_BIO
:
1434 if (obj_request
->bio_list
)
1435 bio_chain_put(obj_request
->bio_list
);
1437 case OBJ_REQUEST_PAGES
:
1438 if (obj_request
->pages
)
1439 ceph_release_page_vector(obj_request
->pages
,
1440 obj_request
->page_count
);
1448 * Caller is responsible for filling in the list of object requests
1449 * that comprises the image request, and the Linux request pointer
1450 * (if there is one).
1452 struct rbd_img_request
*rbd_img_request_create(struct rbd_device
*rbd_dev
,
1453 u64 offset
, u64 length
,
1456 struct rbd_img_request
*img_request
;
1457 struct ceph_snap_context
*snapc
= NULL
;
1459 img_request
= kmalloc(sizeof (*img_request
), GFP_ATOMIC
);
1463 if (write_request
) {
1464 down_read(&rbd_dev
->header_rwsem
);
1465 snapc
= ceph_get_snap_context(rbd_dev
->header
.snapc
);
1466 up_read(&rbd_dev
->header_rwsem
);
1467 if (WARN_ON(!snapc
)) {
1469 return NULL
; /* Shouldn't happen */
1473 img_request
->rq
= NULL
;
1474 img_request
->rbd_dev
= rbd_dev
;
1475 img_request
->offset
= offset
;
1476 img_request
->length
= length
;
1477 img_request
->write_request
= write_request
;
1479 img_request
->snapc
= snapc
;
1481 img_request
->snap_id
= rbd_dev
->spec
->snap_id
;
1482 spin_lock_init(&img_request
->completion_lock
);
1483 img_request
->next_completion
= 0;
1484 img_request
->callback
= NULL
;
1485 img_request
->obj_request_count
= 0;
1486 INIT_LIST_HEAD(&img_request
->obj_requests
);
1487 kref_init(&img_request
->kref
);
1489 rbd_img_request_get(img_request
); /* Avoid a warning */
1490 rbd_img_request_put(img_request
); /* TEMPORARY */
1495 static void rbd_img_request_destroy(struct kref
*kref
)
1497 struct rbd_img_request
*img_request
;
1498 struct rbd_obj_request
*obj_request
;
1499 struct rbd_obj_request
*next_obj_request
;
1501 img_request
= container_of(kref
, struct rbd_img_request
, kref
);
1503 for_each_obj_request_safe(img_request
, obj_request
, next_obj_request
)
1504 rbd_img_obj_request_del(img_request
, obj_request
);
1505 rbd_assert(img_request
->obj_request_count
== 0);
1507 if (img_request
->write_request
)
1508 ceph_put_snap_context(img_request
->snapc
);
1513 static int rbd_img_request_fill_bio(struct rbd_img_request
*img_request
,
1514 struct bio
*bio_list
)
1516 struct rbd_device
*rbd_dev
= img_request
->rbd_dev
;
1517 struct rbd_obj_request
*obj_request
= NULL
;
1518 struct rbd_obj_request
*next_obj_request
;
1519 unsigned int bio_offset
;
1524 opcode
= img_request
->write_request
? CEPH_OSD_OP_WRITE
1527 image_offset
= img_request
->offset
;
1528 rbd_assert(image_offset
== bio_list
->bi_sector
<< SECTOR_SHIFT
);
1529 resid
= img_request
->length
;
1531 const char *object_name
;
1532 unsigned int clone_size
;
1533 struct ceph_osd_req_op
*op
;
1537 object_name
= rbd_segment_name(rbd_dev
, image_offset
);
1540 offset
= rbd_segment_offset(rbd_dev
, image_offset
);
1541 length
= rbd_segment_length(rbd_dev
, image_offset
, resid
);
1542 obj_request
= rbd_obj_request_create(object_name
,
1545 kfree(object_name
); /* object request has its own copy */
1549 rbd_assert(length
<= (u64
) UINT_MAX
);
1550 clone_size
= (unsigned int) length
;
1551 obj_request
->bio_list
= bio_chain_clone_range(&bio_list
,
1552 &bio_offset
, clone_size
,
1554 if (!obj_request
->bio_list
)
1558 * Build up the op to use in building the osd
1559 * request. Note that the contents of the op are
1560 * copied by rbd_osd_req_create().
1562 op
= rbd_osd_req_op_create(opcode
, offset
, length
);
1565 obj_request
->osd_req
= rbd_osd_req_create(rbd_dev
,
1566 img_request
->write_request
,
1568 rbd_osd_req_op_destroy(op
);
1569 if (!obj_request
->osd_req
)
1571 /* status and version are initially zero-filled */
1573 rbd_img_obj_request_add(img_request
, obj_request
);
1575 image_offset
+= length
;
1582 rbd_obj_request_put(obj_request
);
1584 for_each_obj_request_safe(img_request
, obj_request
, next_obj_request
)
1585 rbd_obj_request_put(obj_request
);
1590 static void rbd_img_obj_callback(struct rbd_obj_request
*obj_request
)
1592 struct rbd_img_request
*img_request
;
1593 u32 which
= obj_request
->which
;
1596 img_request
= obj_request
->img_request
;
1597 rbd_assert(img_request
!= NULL
);
1598 rbd_assert(img_request
->rq
!= NULL
);
1599 rbd_assert(which
!= BAD_WHICH
);
1600 rbd_assert(which
< img_request
->obj_request_count
);
1601 rbd_assert(which
>= img_request
->next_completion
);
1603 spin_lock_irq(&img_request
->completion_lock
);
1604 if (which
!= img_request
->next_completion
)
1607 for_each_obj_request_from(img_request
, obj_request
) {
1608 unsigned int xferred
;
1612 rbd_assert(which
< img_request
->obj_request_count
);
1614 if (!atomic_read(&obj_request
->done
))
1617 rbd_assert(obj_request
->xferred
<= (u64
) UINT_MAX
);
1618 xferred
= (unsigned int) obj_request
->xferred
;
1619 result
= (int) obj_request
->result
;
1621 rbd_warn(NULL
, "obj_request %s result %d xferred %u\n",
1622 img_request
->write_request
? "write" : "read",
1625 more
= blk_end_request(img_request
->rq
, result
, xferred
);
1628 rbd_assert(more
^ (which
== img_request
->obj_request_count
));
1629 img_request
->next_completion
= which
;
1631 spin_unlock_irq(&img_request
->completion_lock
);
1634 rbd_img_request_complete(img_request
);
1637 static int rbd_img_request_submit(struct rbd_img_request
*img_request
)
1639 struct rbd_device
*rbd_dev
= img_request
->rbd_dev
;
1640 struct ceph_osd_client
*osdc
= &rbd_dev
->rbd_client
->client
->osdc
;
1641 struct rbd_obj_request
*obj_request
;
1643 for_each_obj_request(img_request
, obj_request
) {
1646 obj_request
->callback
= rbd_img_obj_callback
;
1647 ret
= rbd_obj_request_submit(osdc
, obj_request
);
1651 * The image request has its own reference to each
1652 * of its object requests, so we can safely drop the
1655 rbd_obj_request_put(obj_request
);
1661 static int rbd_obj_notify_ack(struct rbd_device
*rbd_dev
,
1662 u64 ver
, u64 notify_id
)
1664 struct rbd_obj_request
*obj_request
;
1665 struct ceph_osd_req_op
*op
;
1666 struct ceph_osd_client
*osdc
;
1669 obj_request
= rbd_obj_request_create(rbd_dev
->header_name
, 0, 0,
1670 OBJ_REQUEST_NODATA
);
1675 op
= rbd_osd_req_op_create(CEPH_OSD_OP_NOTIFY_ACK
, notify_id
, ver
);
1678 obj_request
->osd_req
= rbd_osd_req_create(rbd_dev
, false,
1680 rbd_osd_req_op_destroy(op
);
1681 if (!obj_request
->osd_req
)
1684 osdc
= &rbd_dev
->rbd_client
->client
->osdc
;
1685 obj_request
->callback
= rbd_obj_request_put
;
1686 ret
= rbd_obj_request_submit(osdc
, obj_request
);
1689 rbd_obj_request_put(obj_request
);
1694 static void rbd_watch_cb(u64 ver
, u64 notify_id
, u8 opcode
, void *data
)
1696 struct rbd_device
*rbd_dev
= (struct rbd_device
*)data
;
1703 dout("rbd_watch_cb %s notify_id=%llu opcode=%u\n",
1704 rbd_dev
->header_name
, (unsigned long long) notify_id
,
1705 (unsigned int) opcode
);
1706 rc
= rbd_dev_refresh(rbd_dev
, &hver
);
1708 rbd_warn(rbd_dev
, "got notification but failed to "
1709 " update snaps: %d\n", rc
);
1711 rbd_obj_notify_ack(rbd_dev
, hver
, notify_id
);
1715 * Request sync osd watch/unwatch. The value of "start" determines
1716 * whether a watch request is being initiated or torn down.
1718 static int rbd_dev_header_watch_sync(struct rbd_device
*rbd_dev
, int start
)
1720 struct ceph_osd_client
*osdc
= &rbd_dev
->rbd_client
->client
->osdc
;
1721 struct rbd_obj_request
*obj_request
;
1722 struct ceph_osd_req_op
*op
;
1725 rbd_assert(start
^ !!rbd_dev
->watch_event
);
1726 rbd_assert(start
^ !!rbd_dev
->watch_request
);
1729 ret
= ceph_osdc_create_event(osdc
, rbd_watch_cb
, 0, rbd_dev
,
1730 &rbd_dev
->watch_event
);
1733 rbd_assert(rbd_dev
->watch_event
!= NULL
);
1737 obj_request
= rbd_obj_request_create(rbd_dev
->header_name
, 0, 0,
1738 OBJ_REQUEST_NODATA
);
1742 op
= rbd_osd_req_op_create(CEPH_OSD_OP_WATCH
,
1743 rbd_dev
->watch_event
->cookie
,
1744 rbd_dev
->header
.obj_version
, start
);
1747 obj_request
->osd_req
= rbd_osd_req_create(rbd_dev
, true,
1749 rbd_osd_req_op_destroy(op
);
1750 if (!obj_request
->osd_req
)
1754 ceph_osdc_set_request_linger(osdc
, obj_request
->osd_req
);
1756 ceph_osdc_unregister_linger_request(osdc
,
1757 rbd_dev
->watch_request
->osd_req
);
1758 ret
= rbd_obj_request_submit(osdc
, obj_request
);
1761 ret
= rbd_obj_request_wait(obj_request
);
1764 ret
= obj_request
->result
;
1769 * A watch request is set to linger, so the underlying osd
1770 * request won't go away until we unregister it. We retain
1771 * a pointer to the object request during that time (in
1772 * rbd_dev->watch_request), so we'll keep a reference to
1773 * it. We'll drop that reference (below) after we've
1777 rbd_dev
->watch_request
= obj_request
;
1782 /* We have successfully torn down the watch request */
1784 rbd_obj_request_put(rbd_dev
->watch_request
);
1785 rbd_dev
->watch_request
= NULL
;
1787 /* Cancel the event if we're tearing down, or on error */
1788 ceph_osdc_cancel_event(rbd_dev
->watch_event
);
1789 rbd_dev
->watch_event
= NULL
;
1791 rbd_obj_request_put(obj_request
);
1797 * Synchronous osd object method call
1799 static int rbd_obj_method_sync(struct rbd_device
*rbd_dev
,
1800 const char *object_name
,
1801 const char *class_name
,
1802 const char *method_name
,
1803 const char *outbound
,
1804 size_t outbound_size
,
1806 size_t inbound_size
,
1809 struct rbd_obj_request
*obj_request
;
1810 struct ceph_osd_client
*osdc
;
1811 struct ceph_osd_req_op
*op
;
1812 struct page
**pages
;
1817 * Method calls are ultimately read operations but they
1818 * don't involve object data (so no offset or length).
1819 * The result should placed into the inbound buffer
1820 * provided. They also supply outbound data--parameters for
1821 * the object method. Currently if this is present it will
1824 page_count
= (u32
) calc_pages_for(0, inbound_size
);
1825 pages
= ceph_alloc_page_vector(page_count
, GFP_KERNEL
);
1827 return PTR_ERR(pages
);
1830 obj_request
= rbd_obj_request_create(object_name
, 0, 0,
1835 obj_request
->pages
= pages
;
1836 obj_request
->page_count
= page_count
;
1838 op
= rbd_osd_req_op_create(CEPH_OSD_OP_CALL
, class_name
,
1839 method_name
, outbound
, outbound_size
);
1842 obj_request
->osd_req
= rbd_osd_req_create(rbd_dev
, false,
1844 rbd_osd_req_op_destroy(op
);
1845 if (!obj_request
->osd_req
)
1848 osdc
= &rbd_dev
->rbd_client
->client
->osdc
;
1849 ret
= rbd_obj_request_submit(osdc
, obj_request
);
1852 ret
= rbd_obj_request_wait(obj_request
);
1856 ret
= obj_request
->result
;
1859 ret
= ceph_copy_from_page_vector(pages
, inbound
, 0,
1860 obj_request
->xferred
);
1862 *version
= obj_request
->version
;
1865 rbd_obj_request_put(obj_request
);
1867 ceph_release_page_vector(pages
, page_count
);
1872 static void rbd_request_fn(struct request_queue
*q
)
1874 struct rbd_device
*rbd_dev
= q
->queuedata
;
1875 bool read_only
= rbd_dev
->mapping
.read_only
;
1879 while ((rq
= blk_fetch_request(q
))) {
1880 bool write_request
= rq_data_dir(rq
) == WRITE
;
1881 struct rbd_img_request
*img_request
;
1885 /* Ignore any non-FS requests that filter through. */
1887 if (rq
->cmd_type
!= REQ_TYPE_FS
) {
1888 __blk_end_request_all(rq
, 0);
1892 spin_unlock_irq(q
->queue_lock
);
1894 /* Disallow writes to a read-only device */
1896 if (write_request
) {
1900 rbd_assert(rbd_dev
->spec
->snap_id
== CEPH_NOSNAP
);
1904 * Quit early if the mapped snapshot no longer
1905 * exists. It's still possible the snapshot will
1906 * have disappeared by the time our request arrives
1907 * at the osd, but there's no sense in sending it if
1910 if (!test_bit(RBD_DEV_FLAG_EXISTS
, &rbd_dev
->flags
)) {
1911 dout("request for non-existent snapshot");
1912 rbd_assert(rbd_dev
->spec
->snap_id
!= CEPH_NOSNAP
);
1917 offset
= (u64
) blk_rq_pos(rq
) << SECTOR_SHIFT
;
1918 length
= (u64
) blk_rq_bytes(rq
);
1921 if (WARN_ON(offset
&& length
> U64_MAX
- offset
+ 1))
1922 goto end_request
; /* Shouldn't happen */
1925 img_request
= rbd_img_request_create(rbd_dev
, offset
, length
,
1930 img_request
->rq
= rq
;
1932 result
= rbd_img_request_fill_bio(img_request
, rq
->bio
);
1934 result
= rbd_img_request_submit(img_request
);
1936 rbd_img_request_put(img_request
);
1938 spin_lock_irq(q
->queue_lock
);
1940 rbd_warn(rbd_dev
, "obj_request %s result %d\n",
1941 write_request
? "write" : "read", result
);
1942 __blk_end_request_all(rq
, result
);
1948 * a queue callback. Makes sure that we don't create a bio that spans across
1949 * multiple osd objects. One exception would be with a single page bios,
1950 * which we handle later at bio_chain_clone_range()
1952 static int rbd_merge_bvec(struct request_queue
*q
, struct bvec_merge_data
*bmd
,
1953 struct bio_vec
*bvec
)
1955 struct rbd_device
*rbd_dev
= q
->queuedata
;
1956 sector_t sector_offset
;
1957 sector_t sectors_per_obj
;
1958 sector_t obj_sector_offset
;
1962 * Find how far into its rbd object the partition-relative
1963 * bio start sector is to offset relative to the enclosing
1966 sector_offset
= get_start_sect(bmd
->bi_bdev
) + bmd
->bi_sector
;
1967 sectors_per_obj
= 1 << (rbd_dev
->header
.obj_order
- SECTOR_SHIFT
);
1968 obj_sector_offset
= sector_offset
& (sectors_per_obj
- 1);
1971 * Compute the number of bytes from that offset to the end
1972 * of the object. Account for what's already used by the bio.
1974 ret
= (int) (sectors_per_obj
- obj_sector_offset
) << SECTOR_SHIFT
;
1975 if (ret
> bmd
->bi_size
)
1976 ret
-= bmd
->bi_size
;
1981 * Don't send back more than was asked for. And if the bio
1982 * was empty, let the whole thing through because: "Note
1983 * that a block device *must* allow a single page to be
1984 * added to an empty bio."
1986 rbd_assert(bvec
->bv_len
<= PAGE_SIZE
);
1987 if (ret
> (int) bvec
->bv_len
|| !bmd
->bi_size
)
1988 ret
= (int) bvec
->bv_len
;
1993 static void rbd_free_disk(struct rbd_device
*rbd_dev
)
1995 struct gendisk
*disk
= rbd_dev
->disk
;
2000 if (disk
->flags
& GENHD_FL_UP
)
2003 blk_cleanup_queue(disk
->queue
);
2007 static int rbd_obj_read_sync(struct rbd_device
*rbd_dev
,
2008 const char *object_name
,
2009 u64 offset
, u64 length
,
2010 char *buf
, u64
*version
)
2013 struct ceph_osd_req_op
*op
;
2014 struct rbd_obj_request
*obj_request
;
2015 struct ceph_osd_client
*osdc
;
2016 struct page
**pages
= NULL
;
2020 page_count
= (u32
) calc_pages_for(offset
, length
);
2021 pages
= ceph_alloc_page_vector(page_count
, GFP_KERNEL
);
2023 ret
= PTR_ERR(pages
);
2026 obj_request
= rbd_obj_request_create(object_name
, offset
, length
,
2031 obj_request
->pages
= pages
;
2032 obj_request
->page_count
= page_count
;
2034 op
= rbd_osd_req_op_create(CEPH_OSD_OP_READ
, offset
, length
);
2037 obj_request
->osd_req
= rbd_osd_req_create(rbd_dev
, false,
2039 rbd_osd_req_op_destroy(op
);
2040 if (!obj_request
->osd_req
)
2043 osdc
= &rbd_dev
->rbd_client
->client
->osdc
;
2044 ret
= rbd_obj_request_submit(osdc
, obj_request
);
2047 ret
= rbd_obj_request_wait(obj_request
);
2051 ret
= obj_request
->result
;
2054 ret
= ceph_copy_from_page_vector(pages
, buf
, 0, obj_request
->xferred
);
2056 *version
= obj_request
->version
;
2059 rbd_obj_request_put(obj_request
);
2061 ceph_release_page_vector(pages
, page_count
);
2067 * Read the complete header for the given rbd device.
2069 * Returns a pointer to a dynamically-allocated buffer containing
2070 * the complete and validated header. Caller can pass the address
2071 * of a variable that will be filled in with the version of the
2072 * header object at the time it was read.
2074 * Returns a pointer-coded errno if a failure occurs.
2076 static struct rbd_image_header_ondisk
*
2077 rbd_dev_v1_header_read(struct rbd_device
*rbd_dev
, u64
*version
)
2079 struct rbd_image_header_ondisk
*ondisk
= NULL
;
2086 * The complete header will include an array of its 64-bit
2087 * snapshot ids, followed by the names of those snapshots as
2088 * a contiguous block of NUL-terminated strings. Note that
2089 * the number of snapshots could change by the time we read
2090 * it in, in which case we re-read it.
2097 size
= sizeof (*ondisk
);
2098 size
+= snap_count
* sizeof (struct rbd_image_snap_ondisk
);
2100 ondisk
= kmalloc(size
, GFP_KERNEL
);
2102 return ERR_PTR(-ENOMEM
);
2104 ret
= rbd_obj_read_sync(rbd_dev
, rbd_dev
->header_name
,
2106 (char *) ondisk
, version
);
2110 if (WARN_ON((size_t) ret
< size
)) {
2112 rbd_warn(rbd_dev
, "short header read (want %zd got %d)",
2116 if (!rbd_dev_ondisk_valid(ondisk
)) {
2118 rbd_warn(rbd_dev
, "invalid header");
2122 names_size
= le64_to_cpu(ondisk
->snap_names_len
);
2123 want_count
= snap_count
;
2124 snap_count
= le32_to_cpu(ondisk
->snap_count
);
2125 } while (snap_count
!= want_count
);
2132 return ERR_PTR(ret
);
2136 * reload the ondisk the header
2138 static int rbd_read_header(struct rbd_device
*rbd_dev
,
2139 struct rbd_image_header
*header
)
2141 struct rbd_image_header_ondisk
*ondisk
;
2145 ondisk
= rbd_dev_v1_header_read(rbd_dev
, &ver
);
2147 return PTR_ERR(ondisk
);
2148 ret
= rbd_header_from_disk(header
, ondisk
);
2150 header
->obj_version
= ver
;
2156 static void rbd_remove_all_snaps(struct rbd_device
*rbd_dev
)
2158 struct rbd_snap
*snap
;
2159 struct rbd_snap
*next
;
2161 list_for_each_entry_safe(snap
, next
, &rbd_dev
->snaps
, node
)
2162 rbd_remove_snap_dev(snap
);
2165 static void rbd_update_mapping_size(struct rbd_device
*rbd_dev
)
2169 if (rbd_dev
->spec
->snap_id
!= CEPH_NOSNAP
)
2172 size
= (sector_t
) rbd_dev
->header
.image_size
/ SECTOR_SIZE
;
2173 dout("setting size to %llu sectors", (unsigned long long) size
);
2174 rbd_dev
->mapping
.size
= (u64
) size
;
2175 set_capacity(rbd_dev
->disk
, size
);
2179 * only read the first part of the ondisk header, without the snaps info
2181 static int rbd_dev_v1_refresh(struct rbd_device
*rbd_dev
, u64
*hver
)
2184 struct rbd_image_header h
;
2186 ret
= rbd_read_header(rbd_dev
, &h
);
2190 down_write(&rbd_dev
->header_rwsem
);
2192 /* Update image size, and check for resize of mapped image */
2193 rbd_dev
->header
.image_size
= h
.image_size
;
2194 rbd_update_mapping_size(rbd_dev
);
2196 /* rbd_dev->header.object_prefix shouldn't change */
2197 kfree(rbd_dev
->header
.snap_sizes
);
2198 kfree(rbd_dev
->header
.snap_names
);
2199 /* osd requests may still refer to snapc */
2200 ceph_put_snap_context(rbd_dev
->header
.snapc
);
2203 *hver
= h
.obj_version
;
2204 rbd_dev
->header
.obj_version
= h
.obj_version
;
2205 rbd_dev
->header
.image_size
= h
.image_size
;
2206 rbd_dev
->header
.snapc
= h
.snapc
;
2207 rbd_dev
->header
.snap_names
= h
.snap_names
;
2208 rbd_dev
->header
.snap_sizes
= h
.snap_sizes
;
2209 /* Free the extra copy of the object prefix */
2210 WARN_ON(strcmp(rbd_dev
->header
.object_prefix
, h
.object_prefix
));
2211 kfree(h
.object_prefix
);
2213 ret
= rbd_dev_snaps_update(rbd_dev
);
2215 ret
= rbd_dev_snaps_register(rbd_dev
);
2217 up_write(&rbd_dev
->header_rwsem
);
2222 static int rbd_dev_refresh(struct rbd_device
*rbd_dev
, u64
*hver
)
2226 rbd_assert(rbd_image_format_valid(rbd_dev
->image_format
));
2227 mutex_lock_nested(&ctl_mutex
, SINGLE_DEPTH_NESTING
);
2228 if (rbd_dev
->image_format
== 1)
2229 ret
= rbd_dev_v1_refresh(rbd_dev
, hver
);
2231 ret
= rbd_dev_v2_refresh(rbd_dev
, hver
);
2232 mutex_unlock(&ctl_mutex
);
2237 static int rbd_init_disk(struct rbd_device
*rbd_dev
)
2239 struct gendisk
*disk
;
2240 struct request_queue
*q
;
2243 /* create gendisk info */
2244 disk
= alloc_disk(RBD_MINORS_PER_MAJOR
);
2248 snprintf(disk
->disk_name
, sizeof(disk
->disk_name
), RBD_DRV_NAME
"%d",
2250 disk
->major
= rbd_dev
->major
;
2251 disk
->first_minor
= 0;
2252 disk
->fops
= &rbd_bd_ops
;
2253 disk
->private_data
= rbd_dev
;
2255 q
= blk_init_queue(rbd_request_fn
, &rbd_dev
->lock
);
2259 /* We use the default size, but let's be explicit about it. */
2260 blk_queue_physical_block_size(q
, SECTOR_SIZE
);
2262 /* set io sizes to object size */
2263 segment_size
= rbd_obj_bytes(&rbd_dev
->header
);
2264 blk_queue_max_hw_sectors(q
, segment_size
/ SECTOR_SIZE
);
2265 blk_queue_max_segment_size(q
, segment_size
);
2266 blk_queue_io_min(q
, segment_size
);
2267 blk_queue_io_opt(q
, segment_size
);
2269 blk_queue_merge_bvec(q
, rbd_merge_bvec
);
2272 q
->queuedata
= rbd_dev
;
2274 rbd_dev
->disk
= disk
;
2276 set_capacity(rbd_dev
->disk
, rbd_dev
->mapping
.size
/ SECTOR_SIZE
);
2289 static struct rbd_device
*dev_to_rbd_dev(struct device
*dev
)
2291 return container_of(dev
, struct rbd_device
, dev
);
2294 static ssize_t
rbd_size_show(struct device
*dev
,
2295 struct device_attribute
*attr
, char *buf
)
2297 struct rbd_device
*rbd_dev
= dev_to_rbd_dev(dev
);
2300 down_read(&rbd_dev
->header_rwsem
);
2301 size
= get_capacity(rbd_dev
->disk
);
2302 up_read(&rbd_dev
->header_rwsem
);
2304 return sprintf(buf
, "%llu\n", (unsigned long long) size
* SECTOR_SIZE
);
2308 * Note this shows the features for whatever's mapped, which is not
2309 * necessarily the base image.
2311 static ssize_t
rbd_features_show(struct device
*dev
,
2312 struct device_attribute
*attr
, char *buf
)
2314 struct rbd_device
*rbd_dev
= dev_to_rbd_dev(dev
);
2316 return sprintf(buf
, "0x%016llx\n",
2317 (unsigned long long) rbd_dev
->mapping
.features
);
2320 static ssize_t
rbd_major_show(struct device
*dev
,
2321 struct device_attribute
*attr
, char *buf
)
2323 struct rbd_device
*rbd_dev
= dev_to_rbd_dev(dev
);
2325 return sprintf(buf
, "%d\n", rbd_dev
->major
);
2328 static ssize_t
rbd_client_id_show(struct device
*dev
,
2329 struct device_attribute
*attr
, char *buf
)
2331 struct rbd_device
*rbd_dev
= dev_to_rbd_dev(dev
);
2333 return sprintf(buf
, "client%lld\n",
2334 ceph_client_id(rbd_dev
->rbd_client
->client
));
2337 static ssize_t
rbd_pool_show(struct device
*dev
,
2338 struct device_attribute
*attr
, char *buf
)
2340 struct rbd_device
*rbd_dev
= dev_to_rbd_dev(dev
);
2342 return sprintf(buf
, "%s\n", rbd_dev
->spec
->pool_name
);
2345 static ssize_t
rbd_pool_id_show(struct device
*dev
,
2346 struct device_attribute
*attr
, char *buf
)
2348 struct rbd_device
*rbd_dev
= dev_to_rbd_dev(dev
);
2350 return sprintf(buf
, "%llu\n",
2351 (unsigned long long) rbd_dev
->spec
->pool_id
);
2354 static ssize_t
rbd_name_show(struct device
*dev
,
2355 struct device_attribute
*attr
, char *buf
)
2357 struct rbd_device
*rbd_dev
= dev_to_rbd_dev(dev
);
2359 if (rbd_dev
->spec
->image_name
)
2360 return sprintf(buf
, "%s\n", rbd_dev
->spec
->image_name
);
2362 return sprintf(buf
, "(unknown)\n");
2365 static ssize_t
rbd_image_id_show(struct device
*dev
,
2366 struct device_attribute
*attr
, char *buf
)
2368 struct rbd_device
*rbd_dev
= dev_to_rbd_dev(dev
);
2370 return sprintf(buf
, "%s\n", rbd_dev
->spec
->image_id
);
2374 * Shows the name of the currently-mapped snapshot (or
2375 * RBD_SNAP_HEAD_NAME for the base image).
2377 static ssize_t
rbd_snap_show(struct device
*dev
,
2378 struct device_attribute
*attr
,
2381 struct rbd_device
*rbd_dev
= dev_to_rbd_dev(dev
);
2383 return sprintf(buf
, "%s\n", rbd_dev
->spec
->snap_name
);
2387 * For an rbd v2 image, shows the pool id, image id, and snapshot id
2388 * for the parent image. If there is no parent, simply shows
2389 * "(no parent image)".
2391 static ssize_t
rbd_parent_show(struct device
*dev
,
2392 struct device_attribute
*attr
,
2395 struct rbd_device
*rbd_dev
= dev_to_rbd_dev(dev
);
2396 struct rbd_spec
*spec
= rbd_dev
->parent_spec
;
2401 return sprintf(buf
, "(no parent image)\n");
2403 count
= sprintf(bufp
, "pool_id %llu\npool_name %s\n",
2404 (unsigned long long) spec
->pool_id
, spec
->pool_name
);
2409 count
= sprintf(bufp
, "image_id %s\nimage_name %s\n", spec
->image_id
,
2410 spec
->image_name
? spec
->image_name
: "(unknown)");
2415 count
= sprintf(bufp
, "snap_id %llu\nsnap_name %s\n",
2416 (unsigned long long) spec
->snap_id
, spec
->snap_name
);
2421 count
= sprintf(bufp
, "overlap %llu\n", rbd_dev
->parent_overlap
);
2426 return (ssize_t
) (bufp
- buf
);
2429 static ssize_t
rbd_image_refresh(struct device
*dev
,
2430 struct device_attribute
*attr
,
2434 struct rbd_device
*rbd_dev
= dev_to_rbd_dev(dev
);
2437 ret
= rbd_dev_refresh(rbd_dev
, NULL
);
2439 return ret
< 0 ? ret
: size
;
2442 static DEVICE_ATTR(size
, S_IRUGO
, rbd_size_show
, NULL
);
2443 static DEVICE_ATTR(features
, S_IRUGO
, rbd_features_show
, NULL
);
2444 static DEVICE_ATTR(major
, S_IRUGO
, rbd_major_show
, NULL
);
2445 static DEVICE_ATTR(client_id
, S_IRUGO
, rbd_client_id_show
, NULL
);
2446 static DEVICE_ATTR(pool
, S_IRUGO
, rbd_pool_show
, NULL
);
2447 static DEVICE_ATTR(pool_id
, S_IRUGO
, rbd_pool_id_show
, NULL
);
2448 static DEVICE_ATTR(name
, S_IRUGO
, rbd_name_show
, NULL
);
2449 static DEVICE_ATTR(image_id
, S_IRUGO
, rbd_image_id_show
, NULL
);
2450 static DEVICE_ATTR(refresh
, S_IWUSR
, NULL
, rbd_image_refresh
);
2451 static DEVICE_ATTR(current_snap
, S_IRUGO
, rbd_snap_show
, NULL
);
2452 static DEVICE_ATTR(parent
, S_IRUGO
, rbd_parent_show
, NULL
);
2454 static struct attribute
*rbd_attrs
[] = {
2455 &dev_attr_size
.attr
,
2456 &dev_attr_features
.attr
,
2457 &dev_attr_major
.attr
,
2458 &dev_attr_client_id
.attr
,
2459 &dev_attr_pool
.attr
,
2460 &dev_attr_pool_id
.attr
,
2461 &dev_attr_name
.attr
,
2462 &dev_attr_image_id
.attr
,
2463 &dev_attr_current_snap
.attr
,
2464 &dev_attr_parent
.attr
,
2465 &dev_attr_refresh
.attr
,
2469 static struct attribute_group rbd_attr_group
= {
2473 static const struct attribute_group
*rbd_attr_groups
[] = {
2478 static void rbd_sysfs_dev_release(struct device
*dev
)
2482 static struct device_type rbd_device_type
= {
2484 .groups
= rbd_attr_groups
,
2485 .release
= rbd_sysfs_dev_release
,
2493 static ssize_t
rbd_snap_size_show(struct device
*dev
,
2494 struct device_attribute
*attr
,
2497 struct rbd_snap
*snap
= container_of(dev
, struct rbd_snap
, dev
);
2499 return sprintf(buf
, "%llu\n", (unsigned long long)snap
->size
);
2502 static ssize_t
rbd_snap_id_show(struct device
*dev
,
2503 struct device_attribute
*attr
,
2506 struct rbd_snap
*snap
= container_of(dev
, struct rbd_snap
, dev
);
2508 return sprintf(buf
, "%llu\n", (unsigned long long)snap
->id
);
2511 static ssize_t
rbd_snap_features_show(struct device
*dev
,
2512 struct device_attribute
*attr
,
2515 struct rbd_snap
*snap
= container_of(dev
, struct rbd_snap
, dev
);
2517 return sprintf(buf
, "0x%016llx\n",
2518 (unsigned long long) snap
->features
);
2521 static DEVICE_ATTR(snap_size
, S_IRUGO
, rbd_snap_size_show
, NULL
);
2522 static DEVICE_ATTR(snap_id
, S_IRUGO
, rbd_snap_id_show
, NULL
);
2523 static DEVICE_ATTR(snap_features
, S_IRUGO
, rbd_snap_features_show
, NULL
);
2525 static struct attribute
*rbd_snap_attrs
[] = {
2526 &dev_attr_snap_size
.attr
,
2527 &dev_attr_snap_id
.attr
,
2528 &dev_attr_snap_features
.attr
,
2532 static struct attribute_group rbd_snap_attr_group
= {
2533 .attrs
= rbd_snap_attrs
,
2536 static void rbd_snap_dev_release(struct device
*dev
)
2538 struct rbd_snap
*snap
= container_of(dev
, struct rbd_snap
, dev
);
2543 static const struct attribute_group
*rbd_snap_attr_groups
[] = {
2544 &rbd_snap_attr_group
,
2548 static struct device_type rbd_snap_device_type
= {
2549 .groups
= rbd_snap_attr_groups
,
2550 .release
= rbd_snap_dev_release
,
2553 static struct rbd_spec
*rbd_spec_get(struct rbd_spec
*spec
)
2555 kref_get(&spec
->kref
);
2560 static void rbd_spec_free(struct kref
*kref
);
2561 static void rbd_spec_put(struct rbd_spec
*spec
)
2564 kref_put(&spec
->kref
, rbd_spec_free
);
2567 static struct rbd_spec
*rbd_spec_alloc(void)
2569 struct rbd_spec
*spec
;
2571 spec
= kzalloc(sizeof (*spec
), GFP_KERNEL
);
2574 kref_init(&spec
->kref
);
2576 rbd_spec_put(rbd_spec_get(spec
)); /* TEMPORARY */
2581 static void rbd_spec_free(struct kref
*kref
)
2583 struct rbd_spec
*spec
= container_of(kref
, struct rbd_spec
, kref
);
2585 kfree(spec
->pool_name
);
2586 kfree(spec
->image_id
);
2587 kfree(spec
->image_name
);
2588 kfree(spec
->snap_name
);
2592 struct rbd_device
*rbd_dev_create(struct rbd_client
*rbdc
,
2593 struct rbd_spec
*spec
)
2595 struct rbd_device
*rbd_dev
;
2597 rbd_dev
= kzalloc(sizeof (*rbd_dev
), GFP_KERNEL
);
2601 spin_lock_init(&rbd_dev
->lock
);
2603 INIT_LIST_HEAD(&rbd_dev
->node
);
2604 INIT_LIST_HEAD(&rbd_dev
->snaps
);
2605 init_rwsem(&rbd_dev
->header_rwsem
);
2607 rbd_dev
->spec
= spec
;
2608 rbd_dev
->rbd_client
= rbdc
;
2610 /* Initialize the layout used for all rbd requests */
2612 rbd_dev
->layout
.fl_stripe_unit
= cpu_to_le32(1 << RBD_MAX_OBJ_ORDER
);
2613 rbd_dev
->layout
.fl_stripe_count
= cpu_to_le32(1);
2614 rbd_dev
->layout
.fl_object_size
= cpu_to_le32(1 << RBD_MAX_OBJ_ORDER
);
2615 rbd_dev
->layout
.fl_pg_pool
= cpu_to_le32((u32
) spec
->pool_id
);
2620 static void rbd_dev_destroy(struct rbd_device
*rbd_dev
)
2622 rbd_spec_put(rbd_dev
->parent_spec
);
2623 kfree(rbd_dev
->header_name
);
2624 rbd_put_client(rbd_dev
->rbd_client
);
2625 rbd_spec_put(rbd_dev
->spec
);
2629 static bool rbd_snap_registered(struct rbd_snap
*snap
)
2631 bool ret
= snap
->dev
.type
== &rbd_snap_device_type
;
2632 bool reg
= device_is_registered(&snap
->dev
);
2634 rbd_assert(!ret
^ reg
);
2639 static void rbd_remove_snap_dev(struct rbd_snap
*snap
)
2641 list_del(&snap
->node
);
2642 if (device_is_registered(&snap
->dev
))
2643 device_unregister(&snap
->dev
);
2646 static int rbd_register_snap_dev(struct rbd_snap
*snap
,
2647 struct device
*parent
)
2649 struct device
*dev
= &snap
->dev
;
2652 dev
->type
= &rbd_snap_device_type
;
2653 dev
->parent
= parent
;
2654 dev
->release
= rbd_snap_dev_release
;
2655 dev_set_name(dev
, "%s%s", RBD_SNAP_DEV_NAME_PREFIX
, snap
->name
);
2656 dout("%s: registering device for snapshot %s\n", __func__
, snap
->name
);
2658 ret
= device_register(dev
);
2663 static struct rbd_snap
*__rbd_add_snap_dev(struct rbd_device
*rbd_dev
,
2664 const char *snap_name
,
2665 u64 snap_id
, u64 snap_size
,
2668 struct rbd_snap
*snap
;
2671 snap
= kzalloc(sizeof (*snap
), GFP_KERNEL
);
2673 return ERR_PTR(-ENOMEM
);
2676 snap
->name
= kstrdup(snap_name
, GFP_KERNEL
);
2681 snap
->size
= snap_size
;
2682 snap
->features
= snap_features
;
2690 return ERR_PTR(ret
);
2693 static char *rbd_dev_v1_snap_info(struct rbd_device
*rbd_dev
, u32 which
,
2694 u64
*snap_size
, u64
*snap_features
)
2698 rbd_assert(which
< rbd_dev
->header
.snapc
->num_snaps
);
2700 *snap_size
= rbd_dev
->header
.snap_sizes
[which
];
2701 *snap_features
= 0; /* No features for v1 */
2703 /* Skip over names until we find the one we are looking for */
2705 snap_name
= rbd_dev
->header
.snap_names
;
2707 snap_name
+= strlen(snap_name
) + 1;
2713 * Get the size and object order for an image snapshot, or if
2714 * snap_id is CEPH_NOSNAP, gets this information for the base
2717 static int _rbd_dev_v2_snap_size(struct rbd_device
*rbd_dev
, u64 snap_id
,
2718 u8
*order
, u64
*snap_size
)
2720 __le64 snapid
= cpu_to_le64(snap_id
);
2725 } __attribute__ ((packed
)) size_buf
= { 0 };
2727 ret
= rbd_obj_method_sync(rbd_dev
, rbd_dev
->header_name
,
2729 (char *) &snapid
, sizeof (snapid
),
2730 (char *) &size_buf
, sizeof (size_buf
), NULL
);
2731 dout("%s: rbd_obj_method_sync returned %d\n", __func__
, ret
);
2735 *order
= size_buf
.order
;
2736 *snap_size
= le64_to_cpu(size_buf
.size
);
2738 dout(" snap_id 0x%016llx order = %u, snap_size = %llu\n",
2739 (unsigned long long) snap_id
, (unsigned int) *order
,
2740 (unsigned long long) *snap_size
);
2745 static int rbd_dev_v2_image_size(struct rbd_device
*rbd_dev
)
2747 return _rbd_dev_v2_snap_size(rbd_dev
, CEPH_NOSNAP
,
2748 &rbd_dev
->header
.obj_order
,
2749 &rbd_dev
->header
.image_size
);
2752 static int rbd_dev_v2_object_prefix(struct rbd_device
*rbd_dev
)
2758 reply_buf
= kzalloc(RBD_OBJ_PREFIX_LEN_MAX
, GFP_KERNEL
);
2762 ret
= rbd_obj_method_sync(rbd_dev
, rbd_dev
->header_name
,
2763 "rbd", "get_object_prefix",
2765 reply_buf
, RBD_OBJ_PREFIX_LEN_MAX
, NULL
);
2766 dout("%s: rbd_obj_method_sync returned %d\n", __func__
, ret
);
2769 ret
= 0; /* rbd_obj_method_sync() can return positive */
2772 rbd_dev
->header
.object_prefix
= ceph_extract_encoded_string(&p
,
2773 p
+ RBD_OBJ_PREFIX_LEN_MAX
,
2776 if (IS_ERR(rbd_dev
->header
.object_prefix
)) {
2777 ret
= PTR_ERR(rbd_dev
->header
.object_prefix
);
2778 rbd_dev
->header
.object_prefix
= NULL
;
2780 dout(" object_prefix = %s\n", rbd_dev
->header
.object_prefix
);
2789 static int _rbd_dev_v2_snap_features(struct rbd_device
*rbd_dev
, u64 snap_id
,
2792 __le64 snapid
= cpu_to_le64(snap_id
);
2796 } features_buf
= { 0 };
2800 ret
= rbd_obj_method_sync(rbd_dev
, rbd_dev
->header_name
,
2801 "rbd", "get_features",
2802 (char *) &snapid
, sizeof (snapid
),
2803 (char *) &features_buf
, sizeof (features_buf
),
2805 dout("%s: rbd_obj_method_sync returned %d\n", __func__
, ret
);
2809 incompat
= le64_to_cpu(features_buf
.incompat
);
2810 if (incompat
& ~RBD_FEATURES_ALL
)
2813 *snap_features
= le64_to_cpu(features_buf
.features
);
2815 dout(" snap_id 0x%016llx features = 0x%016llx incompat = 0x%016llx\n",
2816 (unsigned long long) snap_id
,
2817 (unsigned long long) *snap_features
,
2818 (unsigned long long) le64_to_cpu(features_buf
.incompat
));
2823 static int rbd_dev_v2_features(struct rbd_device
*rbd_dev
)
2825 return _rbd_dev_v2_snap_features(rbd_dev
, CEPH_NOSNAP
,
2826 &rbd_dev
->header
.features
);
2829 static int rbd_dev_v2_parent_info(struct rbd_device
*rbd_dev
)
2831 struct rbd_spec
*parent_spec
;
2833 void *reply_buf
= NULL
;
2841 parent_spec
= rbd_spec_alloc();
2845 size
= sizeof (__le64
) + /* pool_id */
2846 sizeof (__le32
) + RBD_IMAGE_ID_LEN_MAX
+ /* image_id */
2847 sizeof (__le64
) + /* snap_id */
2848 sizeof (__le64
); /* overlap */
2849 reply_buf
= kmalloc(size
, GFP_KERNEL
);
2855 snapid
= cpu_to_le64(CEPH_NOSNAP
);
2856 ret
= rbd_obj_method_sync(rbd_dev
, rbd_dev
->header_name
,
2857 "rbd", "get_parent",
2858 (char *) &snapid
, sizeof (snapid
),
2859 (char *) reply_buf
, size
, NULL
);
2860 dout("%s: rbd_obj_method_sync returned %d\n", __func__
, ret
);
2866 end
= (char *) reply_buf
+ size
;
2867 ceph_decode_64_safe(&p
, end
, parent_spec
->pool_id
, out_err
);
2868 if (parent_spec
->pool_id
== CEPH_NOPOOL
)
2869 goto out
; /* No parent? No problem. */
2871 /* The ceph file layout needs to fit pool id in 32 bits */
2874 if (WARN_ON(parent_spec
->pool_id
> (u64
) U32_MAX
))
2877 image_id
= ceph_extract_encoded_string(&p
, end
, NULL
, GFP_KERNEL
);
2878 if (IS_ERR(image_id
)) {
2879 ret
= PTR_ERR(image_id
);
2882 parent_spec
->image_id
= image_id
;
2883 ceph_decode_64_safe(&p
, end
, parent_spec
->snap_id
, out_err
);
2884 ceph_decode_64_safe(&p
, end
, overlap
, out_err
);
2886 rbd_dev
->parent_overlap
= overlap
;
2887 rbd_dev
->parent_spec
= parent_spec
;
2888 parent_spec
= NULL
; /* rbd_dev now owns this */
2893 rbd_spec_put(parent_spec
);
2898 static char *rbd_dev_image_name(struct rbd_device
*rbd_dev
)
2900 size_t image_id_size
;
2905 void *reply_buf
= NULL
;
2907 char *image_name
= NULL
;
2910 rbd_assert(!rbd_dev
->spec
->image_name
);
2912 len
= strlen(rbd_dev
->spec
->image_id
);
2913 image_id_size
= sizeof (__le32
) + len
;
2914 image_id
= kmalloc(image_id_size
, GFP_KERNEL
);
2919 end
= (char *) image_id
+ image_id_size
;
2920 ceph_encode_string(&p
, end
, rbd_dev
->spec
->image_id
, (u32
) len
);
2922 size
= sizeof (__le32
) + RBD_IMAGE_NAME_LEN_MAX
;
2923 reply_buf
= kmalloc(size
, GFP_KERNEL
);
2927 ret
= rbd_obj_method_sync(rbd_dev
, RBD_DIRECTORY
,
2928 "rbd", "dir_get_name",
2929 image_id
, image_id_size
,
2930 (char *) reply_buf
, size
, NULL
);
2934 end
= (char *) reply_buf
+ size
;
2935 image_name
= ceph_extract_encoded_string(&p
, end
, &len
, GFP_KERNEL
);
2936 if (IS_ERR(image_name
))
2939 dout("%s: name is %s len is %zd\n", __func__
, image_name
, len
);
2948 * When a parent image gets probed, we only have the pool, image,
2949 * and snapshot ids but not the names of any of them. This call
2950 * is made later to fill in those names. It has to be done after
2951 * rbd_dev_snaps_update() has completed because some of the
2952 * information (in particular, snapshot name) is not available
2955 static int rbd_dev_probe_update_spec(struct rbd_device
*rbd_dev
)
2957 struct ceph_osd_client
*osdc
;
2959 void *reply_buf
= NULL
;
2962 if (rbd_dev
->spec
->pool_name
)
2963 return 0; /* Already have the names */
2965 /* Look up the pool name */
2967 osdc
= &rbd_dev
->rbd_client
->client
->osdc
;
2968 name
= ceph_pg_pool_name_by_id(osdc
->osdmap
, rbd_dev
->spec
->pool_id
);
2970 rbd_warn(rbd_dev
, "there is no pool with id %llu",
2971 rbd_dev
->spec
->pool_id
); /* Really a BUG() */
2975 rbd_dev
->spec
->pool_name
= kstrdup(name
, GFP_KERNEL
);
2976 if (!rbd_dev
->spec
->pool_name
)
2979 /* Fetch the image name; tolerate failure here */
2981 name
= rbd_dev_image_name(rbd_dev
);
2983 rbd_dev
->spec
->image_name
= (char *) name
;
2985 rbd_warn(rbd_dev
, "unable to get image name");
2987 /* Look up the snapshot name. */
2989 name
= rbd_snap_name(rbd_dev
, rbd_dev
->spec
->snap_id
);
2991 rbd_warn(rbd_dev
, "no snapshot with id %llu",
2992 rbd_dev
->spec
->snap_id
); /* Really a BUG() */
2996 rbd_dev
->spec
->snap_name
= kstrdup(name
, GFP_KERNEL
);
2997 if(!rbd_dev
->spec
->snap_name
)
3003 kfree(rbd_dev
->spec
->pool_name
);
3004 rbd_dev
->spec
->pool_name
= NULL
;
3009 static int rbd_dev_v2_snap_context(struct rbd_device
*rbd_dev
, u64
*ver
)
3018 struct ceph_snap_context
*snapc
;
3022 * We'll need room for the seq value (maximum snapshot id),
3023 * snapshot count, and array of that many snapshot ids.
3024 * For now we have a fixed upper limit on the number we're
3025 * prepared to receive.
3027 size
= sizeof (__le64
) + sizeof (__le32
) +
3028 RBD_MAX_SNAP_COUNT
* sizeof (__le64
);
3029 reply_buf
= kzalloc(size
, GFP_KERNEL
);
3033 ret
= rbd_obj_method_sync(rbd_dev
, rbd_dev
->header_name
,
3034 "rbd", "get_snapcontext",
3036 reply_buf
, size
, ver
);
3037 dout("%s: rbd_obj_method_sync returned %d\n", __func__
, ret
);
3043 end
= (char *) reply_buf
+ size
;
3044 ceph_decode_64_safe(&p
, end
, seq
, out
);
3045 ceph_decode_32_safe(&p
, end
, snap_count
, out
);
3048 * Make sure the reported number of snapshot ids wouldn't go
3049 * beyond the end of our buffer. But before checking that,
3050 * make sure the computed size of the snapshot context we
3051 * allocate is representable in a size_t.
3053 if (snap_count
> (SIZE_MAX
- sizeof (struct ceph_snap_context
))
3058 if (!ceph_has_room(&p
, end
, snap_count
* sizeof (__le64
)))
3061 size
= sizeof (struct ceph_snap_context
) +
3062 snap_count
* sizeof (snapc
->snaps
[0]);
3063 snapc
= kmalloc(size
, GFP_KERNEL
);
3069 atomic_set(&snapc
->nref
, 1);
3071 snapc
->num_snaps
= snap_count
;
3072 for (i
= 0; i
< snap_count
; i
++)
3073 snapc
->snaps
[i
] = ceph_decode_64(&p
);
3075 rbd_dev
->header
.snapc
= snapc
;
3077 dout(" snap context seq = %llu, snap_count = %u\n",
3078 (unsigned long long) seq
, (unsigned int) snap_count
);
3086 static char *rbd_dev_v2_snap_name(struct rbd_device
*rbd_dev
, u32 which
)
3096 size
= sizeof (__le32
) + RBD_MAX_SNAP_NAME_LEN
;
3097 reply_buf
= kmalloc(size
, GFP_KERNEL
);
3099 return ERR_PTR(-ENOMEM
);
3101 snap_id
= cpu_to_le64(rbd_dev
->header
.snapc
->snaps
[which
]);
3102 ret
= rbd_obj_method_sync(rbd_dev
, rbd_dev
->header_name
,
3103 "rbd", "get_snapshot_name",
3104 (char *) &snap_id
, sizeof (snap_id
),
3105 reply_buf
, size
, NULL
);
3106 dout("%s: rbd_obj_method_sync returned %d\n", __func__
, ret
);
3111 end
= (char *) reply_buf
+ size
;
3112 snap_name
= ceph_extract_encoded_string(&p
, end
, NULL
, GFP_KERNEL
);
3113 if (IS_ERR(snap_name
)) {
3114 ret
= PTR_ERR(snap_name
);
3117 dout(" snap_id 0x%016llx snap_name = %s\n",
3118 (unsigned long long) le64_to_cpu(snap_id
), snap_name
);
3126 return ERR_PTR(ret
);
3129 static char *rbd_dev_v2_snap_info(struct rbd_device
*rbd_dev
, u32 which
,
3130 u64
*snap_size
, u64
*snap_features
)
3136 snap_id
= rbd_dev
->header
.snapc
->snaps
[which
];
3137 ret
= _rbd_dev_v2_snap_size(rbd_dev
, snap_id
, &order
, snap_size
);
3139 return ERR_PTR(ret
);
3140 ret
= _rbd_dev_v2_snap_features(rbd_dev
, snap_id
, snap_features
);
3142 return ERR_PTR(ret
);
3144 return rbd_dev_v2_snap_name(rbd_dev
, which
);
3147 static char *rbd_dev_snap_info(struct rbd_device
*rbd_dev
, u32 which
,
3148 u64
*snap_size
, u64
*snap_features
)
3150 if (rbd_dev
->image_format
== 1)
3151 return rbd_dev_v1_snap_info(rbd_dev
, which
,
3152 snap_size
, snap_features
);
3153 if (rbd_dev
->image_format
== 2)
3154 return rbd_dev_v2_snap_info(rbd_dev
, which
,
3155 snap_size
, snap_features
);
3156 return ERR_PTR(-EINVAL
);
3159 static int rbd_dev_v2_refresh(struct rbd_device
*rbd_dev
, u64
*hver
)
3164 down_write(&rbd_dev
->header_rwsem
);
3166 /* Grab old order first, to see if it changes */
3168 obj_order
= rbd_dev
->header
.obj_order
,
3169 ret
= rbd_dev_v2_image_size(rbd_dev
);
3172 if (rbd_dev
->header
.obj_order
!= obj_order
) {
3176 rbd_update_mapping_size(rbd_dev
);
3178 ret
= rbd_dev_v2_snap_context(rbd_dev
, hver
);
3179 dout("rbd_dev_v2_snap_context returned %d\n", ret
);
3182 ret
= rbd_dev_snaps_update(rbd_dev
);
3183 dout("rbd_dev_snaps_update returned %d\n", ret
);
3186 ret
= rbd_dev_snaps_register(rbd_dev
);
3187 dout("rbd_dev_snaps_register returned %d\n", ret
);
3189 up_write(&rbd_dev
->header_rwsem
);
3195 * Scan the rbd device's current snapshot list and compare it to the
3196 * newly-received snapshot context. Remove any existing snapshots
3197 * not present in the new snapshot context. Add a new snapshot for
3198 * any snaphots in the snapshot context not in the current list.
3199 * And verify there are no changes to snapshots we already know
3202 * Assumes the snapshots in the snapshot context are sorted by
3203 * snapshot id, highest id first. (Snapshots in the rbd_dev's list
3204 * are also maintained in that order.)
3206 static int rbd_dev_snaps_update(struct rbd_device
*rbd_dev
)
3208 struct ceph_snap_context
*snapc
= rbd_dev
->header
.snapc
;
3209 const u32 snap_count
= snapc
->num_snaps
;
3210 struct list_head
*head
= &rbd_dev
->snaps
;
3211 struct list_head
*links
= head
->next
;
3214 dout("%s: snap count is %u\n", __func__
, (unsigned int) snap_count
);
3215 while (index
< snap_count
|| links
!= head
) {
3217 struct rbd_snap
*snap
;
3220 u64 snap_features
= 0;
3222 snap_id
= index
< snap_count
? snapc
->snaps
[index
]
3224 snap
= links
!= head
? list_entry(links
, struct rbd_snap
, node
)
3226 rbd_assert(!snap
|| snap
->id
!= CEPH_NOSNAP
);
3228 if (snap_id
== CEPH_NOSNAP
|| (snap
&& snap
->id
> snap_id
)) {
3229 struct list_head
*next
= links
->next
;
3232 * A previously-existing snapshot is not in
3233 * the new snap context.
3235 * If the now missing snapshot is the one the
3236 * image is mapped to, clear its exists flag
3237 * so we can avoid sending any more requests
3240 if (rbd_dev
->spec
->snap_id
== snap
->id
)
3241 clear_bit(RBD_DEV_FLAG_EXISTS
, &rbd_dev
->flags
);
3242 rbd_remove_snap_dev(snap
);
3243 dout("%ssnap id %llu has been removed\n",
3244 rbd_dev
->spec
->snap_id
== snap
->id
?
3246 (unsigned long long) snap
->id
);
3248 /* Done with this list entry; advance */
3254 snap_name
= rbd_dev_snap_info(rbd_dev
, index
,
3255 &snap_size
, &snap_features
);
3256 if (IS_ERR(snap_name
))
3257 return PTR_ERR(snap_name
);
3259 dout("entry %u: snap_id = %llu\n", (unsigned int) snap_count
,
3260 (unsigned long long) snap_id
);
3261 if (!snap
|| (snap_id
!= CEPH_NOSNAP
&& snap
->id
< snap_id
)) {
3262 struct rbd_snap
*new_snap
;
3264 /* We haven't seen this snapshot before */
3266 new_snap
= __rbd_add_snap_dev(rbd_dev
, snap_name
,
3267 snap_id
, snap_size
, snap_features
);
3268 if (IS_ERR(new_snap
)) {
3269 int err
= PTR_ERR(new_snap
);
3271 dout(" failed to add dev, error %d\n", err
);
3276 /* New goes before existing, or at end of list */
3278 dout(" added dev%s\n", snap
? "" : " at end\n");
3280 list_add_tail(&new_snap
->node
, &snap
->node
);
3282 list_add_tail(&new_snap
->node
, head
);
3284 /* Already have this one */
3286 dout(" already present\n");
3288 rbd_assert(snap
->size
== snap_size
);
3289 rbd_assert(!strcmp(snap
->name
, snap_name
));
3290 rbd_assert(snap
->features
== snap_features
);
3292 /* Done with this list entry; advance */
3294 links
= links
->next
;
3297 /* Advance to the next entry in the snapshot context */
3301 dout("%s: done\n", __func__
);
3307 * Scan the list of snapshots and register the devices for any that
3308 * have not already been registered.
3310 static int rbd_dev_snaps_register(struct rbd_device
*rbd_dev
)
3312 struct rbd_snap
*snap
;
3315 dout("%s called\n", __func__
);
3316 if (WARN_ON(!device_is_registered(&rbd_dev
->dev
)))
3319 list_for_each_entry(snap
, &rbd_dev
->snaps
, node
) {
3320 if (!rbd_snap_registered(snap
)) {
3321 ret
= rbd_register_snap_dev(snap
, &rbd_dev
->dev
);
3326 dout("%s: returning %d\n", __func__
, ret
);
3331 static int rbd_bus_add_dev(struct rbd_device
*rbd_dev
)
3336 mutex_lock_nested(&ctl_mutex
, SINGLE_DEPTH_NESTING
);
3338 dev
= &rbd_dev
->dev
;
3339 dev
->bus
= &rbd_bus_type
;
3340 dev
->type
= &rbd_device_type
;
3341 dev
->parent
= &rbd_root_dev
;
3342 dev
->release
= rbd_dev_release
;
3343 dev_set_name(dev
, "%d", rbd_dev
->dev_id
);
3344 ret
= device_register(dev
);
3346 mutex_unlock(&ctl_mutex
);
3351 static void rbd_bus_del_dev(struct rbd_device
*rbd_dev
)
3353 device_unregister(&rbd_dev
->dev
);
3356 static atomic64_t rbd_dev_id_max
= ATOMIC64_INIT(0);
3359 * Get a unique rbd identifier for the given new rbd_dev, and add
3360 * the rbd_dev to the global list. The minimum rbd id is 1.
3362 static void rbd_dev_id_get(struct rbd_device
*rbd_dev
)
3364 rbd_dev
->dev_id
= atomic64_inc_return(&rbd_dev_id_max
);
3366 spin_lock(&rbd_dev_list_lock
);
3367 list_add_tail(&rbd_dev
->node
, &rbd_dev_list
);
3368 spin_unlock(&rbd_dev_list_lock
);
3369 dout("rbd_dev %p given dev id %llu\n", rbd_dev
,
3370 (unsigned long long) rbd_dev
->dev_id
);
3374 * Remove an rbd_dev from the global list, and record that its
3375 * identifier is no longer in use.
3377 static void rbd_dev_id_put(struct rbd_device
*rbd_dev
)
3379 struct list_head
*tmp
;
3380 int rbd_id
= rbd_dev
->dev_id
;
3383 rbd_assert(rbd_id
> 0);
3385 dout("rbd_dev %p released dev id %llu\n", rbd_dev
,
3386 (unsigned long long) rbd_dev
->dev_id
);
3387 spin_lock(&rbd_dev_list_lock
);
3388 list_del_init(&rbd_dev
->node
);
3391 * If the id being "put" is not the current maximum, there
3392 * is nothing special we need to do.
3394 if (rbd_id
!= atomic64_read(&rbd_dev_id_max
)) {
3395 spin_unlock(&rbd_dev_list_lock
);
3400 * We need to update the current maximum id. Search the
3401 * list to find out what it is. We're more likely to find
3402 * the maximum at the end, so search the list backward.
3405 list_for_each_prev(tmp
, &rbd_dev_list
) {
3406 struct rbd_device
*rbd_dev
;
3408 rbd_dev
= list_entry(tmp
, struct rbd_device
, node
);
3409 if (rbd_dev
->dev_id
> max_id
)
3410 max_id
= rbd_dev
->dev_id
;
3412 spin_unlock(&rbd_dev_list_lock
);
3415 * The max id could have been updated by rbd_dev_id_get(), in
3416 * which case it now accurately reflects the new maximum.
3417 * Be careful not to overwrite the maximum value in that
3420 atomic64_cmpxchg(&rbd_dev_id_max
, rbd_id
, max_id
);
3421 dout(" max dev id has been reset\n");
3425 * Skips over white space at *buf, and updates *buf to point to the
3426 * first found non-space character (if any). Returns the length of
3427 * the token (string of non-white space characters) found. Note
3428 * that *buf must be terminated with '\0'.
3430 static inline size_t next_token(const char **buf
)
3433 * These are the characters that produce nonzero for
3434 * isspace() in the "C" and "POSIX" locales.
3436 const char *spaces
= " \f\n\r\t\v";
3438 *buf
+= strspn(*buf
, spaces
); /* Find start of token */
3440 return strcspn(*buf
, spaces
); /* Return token length */
3444 * Finds the next token in *buf, and if the provided token buffer is
3445 * big enough, copies the found token into it. The result, if
3446 * copied, is guaranteed to be terminated with '\0'. Note that *buf
3447 * must be terminated with '\0' on entry.
3449 * Returns the length of the token found (not including the '\0').
3450 * Return value will be 0 if no token is found, and it will be >=
3451 * token_size if the token would not fit.
3453 * The *buf pointer will be updated to point beyond the end of the
3454 * found token. Note that this occurs even if the token buffer is
3455 * too small to hold it.
3457 static inline size_t copy_token(const char **buf
,
3463 len
= next_token(buf
);
3464 if (len
< token_size
) {
3465 memcpy(token
, *buf
, len
);
3466 *(token
+ len
) = '\0';
3474 * Finds the next token in *buf, dynamically allocates a buffer big
3475 * enough to hold a copy of it, and copies the token into the new
3476 * buffer. The copy is guaranteed to be terminated with '\0'. Note
3477 * that a duplicate buffer is created even for a zero-length token.
3479 * Returns a pointer to the newly-allocated duplicate, or a null
3480 * pointer if memory for the duplicate was not available. If
3481 * the lenp argument is a non-null pointer, the length of the token
3482 * (not including the '\0') is returned in *lenp.
3484 * If successful, the *buf pointer will be updated to point beyond
3485 * the end of the found token.
3487 * Note: uses GFP_KERNEL for allocation.
3489 static inline char *dup_token(const char **buf
, size_t *lenp
)
3494 len
= next_token(buf
);
3495 dup
= kmemdup(*buf
, len
+ 1, GFP_KERNEL
);
3498 *(dup
+ len
) = '\0';
3508 * Parse the options provided for an "rbd add" (i.e., rbd image
3509 * mapping) request. These arrive via a write to /sys/bus/rbd/add,
3510 * and the data written is passed here via a NUL-terminated buffer.
3511 * Returns 0 if successful or an error code otherwise.
3513 * The information extracted from these options is recorded in
3514 * the other parameters which return dynamically-allocated
3517 * The address of a pointer that will refer to a ceph options
3518 * structure. Caller must release the returned pointer using
3519 * ceph_destroy_options() when it is no longer needed.
3521 * Address of an rbd options pointer. Fully initialized by
3522 * this function; caller must release with kfree().
3524 * Address of an rbd image specification pointer. Fully
3525 * initialized by this function based on parsed options.
3526 * Caller must release with rbd_spec_put().
3528 * The options passed take this form:
3529 * <mon_addrs> <options> <pool_name> <image_name> [<snap_id>]
3532 * A comma-separated list of one or more monitor addresses.
3533 * A monitor address is an ip address, optionally followed
3534 * by a port number (separated by a colon).
3535 * I.e.: ip1[:port1][,ip2[:port2]...]
3537 * A comma-separated list of ceph and/or rbd options.
3539 * The name of the rados pool containing the rbd image.
3541 * The name of the image in that pool to map.
3543 * An optional snapshot id. If provided, the mapping will
3544 * present data from the image at the time that snapshot was
3545 * created. The image head is used if no snapshot id is
3546 * provided. Snapshot mappings are always read-only.
3548 static int rbd_add_parse_args(const char *buf
,
3549 struct ceph_options
**ceph_opts
,
3550 struct rbd_options
**opts
,
3551 struct rbd_spec
**rbd_spec
)
3555 const char *mon_addrs
;
3556 size_t mon_addrs_size
;
3557 struct rbd_spec
*spec
= NULL
;
3558 struct rbd_options
*rbd_opts
= NULL
;
3559 struct ceph_options
*copts
;
3562 /* The first four tokens are required */
3564 len
= next_token(&buf
);
3566 rbd_warn(NULL
, "no monitor address(es) provided");
3570 mon_addrs_size
= len
+ 1;
3574 options
= dup_token(&buf
, NULL
);
3578 rbd_warn(NULL
, "no options provided");
3582 spec
= rbd_spec_alloc();
3586 spec
->pool_name
= dup_token(&buf
, NULL
);
3587 if (!spec
->pool_name
)
3589 if (!*spec
->pool_name
) {
3590 rbd_warn(NULL
, "no pool name provided");
3594 spec
->image_name
= dup_token(&buf
, NULL
);
3595 if (!spec
->image_name
)
3597 if (!*spec
->image_name
) {
3598 rbd_warn(NULL
, "no image name provided");
3603 * Snapshot name is optional; default is to use "-"
3604 * (indicating the head/no snapshot).
3606 len
= next_token(&buf
);
3608 buf
= RBD_SNAP_HEAD_NAME
; /* No snapshot supplied */
3609 len
= sizeof (RBD_SNAP_HEAD_NAME
) - 1;
3610 } else if (len
> RBD_MAX_SNAP_NAME_LEN
) {
3611 ret
= -ENAMETOOLONG
;
3614 spec
->snap_name
= kmemdup(buf
, len
+ 1, GFP_KERNEL
);
3615 if (!spec
->snap_name
)
3617 *(spec
->snap_name
+ len
) = '\0';
3619 /* Initialize all rbd options to the defaults */
3621 rbd_opts
= kzalloc(sizeof (*rbd_opts
), GFP_KERNEL
);
3625 rbd_opts
->read_only
= RBD_READ_ONLY_DEFAULT
;
3627 copts
= ceph_parse_options(options
, mon_addrs
,
3628 mon_addrs
+ mon_addrs_size
- 1,
3629 parse_rbd_opts_token
, rbd_opts
);
3630 if (IS_ERR(copts
)) {
3631 ret
= PTR_ERR(copts
);
3652 * An rbd format 2 image has a unique identifier, distinct from the
3653 * name given to it by the user. Internally, that identifier is
3654 * what's used to specify the names of objects related to the image.
3656 * A special "rbd id" object is used to map an rbd image name to its
3657 * id. If that object doesn't exist, then there is no v2 rbd image
3658 * with the supplied name.
3660 * This function will record the given rbd_dev's image_id field if
3661 * it can be determined, and in that case will return 0. If any
3662 * errors occur a negative errno will be returned and the rbd_dev's
3663 * image_id field will be unchanged (and should be NULL).
3665 static int rbd_dev_image_id(struct rbd_device
*rbd_dev
)
3674 * When probing a parent image, the image id is already
3675 * known (and the image name likely is not). There's no
3676 * need to fetch the image id again in this case.
3678 if (rbd_dev
->spec
->image_id
)
3682 * First, see if the format 2 image id file exists, and if
3683 * so, get the image's persistent id from it.
3685 size
= sizeof (RBD_ID_PREFIX
) + strlen(rbd_dev
->spec
->image_name
);
3686 object_name
= kmalloc(size
, GFP_NOIO
);
3689 sprintf(object_name
, "%s%s", RBD_ID_PREFIX
, rbd_dev
->spec
->image_name
);
3690 dout("rbd id object name is %s\n", object_name
);
3692 /* Response will be an encoded string, which includes a length */
3694 size
= sizeof (__le32
) + RBD_IMAGE_ID_LEN_MAX
;
3695 response
= kzalloc(size
, GFP_NOIO
);
3701 ret
= rbd_obj_method_sync(rbd_dev
, object_name
,
3704 response
, RBD_IMAGE_ID_LEN_MAX
, NULL
);
3705 dout("%s: rbd_obj_method_sync returned %d\n", __func__
, ret
);
3708 ret
= 0; /* rbd_obj_method_sync() can return positive */
3711 rbd_dev
->spec
->image_id
= ceph_extract_encoded_string(&p
,
3712 p
+ RBD_IMAGE_ID_LEN_MAX
,
3714 if (IS_ERR(rbd_dev
->spec
->image_id
)) {
3715 ret
= PTR_ERR(rbd_dev
->spec
->image_id
);
3716 rbd_dev
->spec
->image_id
= NULL
;
3718 dout("image_id is %s\n", rbd_dev
->spec
->image_id
);
3727 static int rbd_dev_v1_probe(struct rbd_device
*rbd_dev
)
3732 /* Version 1 images have no id; empty string is used */
3734 rbd_dev
->spec
->image_id
= kstrdup("", GFP_KERNEL
);
3735 if (!rbd_dev
->spec
->image_id
)
3738 /* Record the header object name for this rbd image. */
3740 size
= strlen(rbd_dev
->spec
->image_name
) + sizeof (RBD_SUFFIX
);
3741 rbd_dev
->header_name
= kmalloc(size
, GFP_KERNEL
);
3742 if (!rbd_dev
->header_name
) {
3746 sprintf(rbd_dev
->header_name
, "%s%s",
3747 rbd_dev
->spec
->image_name
, RBD_SUFFIX
);
3749 /* Populate rbd image metadata */
3751 ret
= rbd_read_header(rbd_dev
, &rbd_dev
->header
);
3755 /* Version 1 images have no parent (no layering) */
3757 rbd_dev
->parent_spec
= NULL
;
3758 rbd_dev
->parent_overlap
= 0;
3760 rbd_dev
->image_format
= 1;
3762 dout("discovered version 1 image, header name is %s\n",
3763 rbd_dev
->header_name
);
3768 kfree(rbd_dev
->header_name
);
3769 rbd_dev
->header_name
= NULL
;
3770 kfree(rbd_dev
->spec
->image_id
);
3771 rbd_dev
->spec
->image_id
= NULL
;
3776 static int rbd_dev_v2_probe(struct rbd_device
*rbd_dev
)
3783 * Image id was filled in by the caller. Record the header
3784 * object name for this rbd image.
3786 size
= sizeof (RBD_HEADER_PREFIX
) + strlen(rbd_dev
->spec
->image_id
);
3787 rbd_dev
->header_name
= kmalloc(size
, GFP_KERNEL
);
3788 if (!rbd_dev
->header_name
)
3790 sprintf(rbd_dev
->header_name
, "%s%s",
3791 RBD_HEADER_PREFIX
, rbd_dev
->spec
->image_id
);
3793 /* Get the size and object order for the image */
3795 ret
= rbd_dev_v2_image_size(rbd_dev
);
3799 /* Get the object prefix (a.k.a. block_name) for the image */
3801 ret
= rbd_dev_v2_object_prefix(rbd_dev
);
3805 /* Get the and check features for the image */
3807 ret
= rbd_dev_v2_features(rbd_dev
);
3811 /* If the image supports layering, get the parent info */
3813 if (rbd_dev
->header
.features
& RBD_FEATURE_LAYERING
) {
3814 ret
= rbd_dev_v2_parent_info(rbd_dev
);
3819 /* crypto and compression type aren't (yet) supported for v2 images */
3821 rbd_dev
->header
.crypt_type
= 0;
3822 rbd_dev
->header
.comp_type
= 0;
3824 /* Get the snapshot context, plus the header version */
3826 ret
= rbd_dev_v2_snap_context(rbd_dev
, &ver
);
3829 rbd_dev
->header
.obj_version
= ver
;
3831 rbd_dev
->image_format
= 2;
3833 dout("discovered version 2 image, header name is %s\n",
3834 rbd_dev
->header_name
);
3838 rbd_dev
->parent_overlap
= 0;
3839 rbd_spec_put(rbd_dev
->parent_spec
);
3840 rbd_dev
->parent_spec
= NULL
;
3841 kfree(rbd_dev
->header_name
);
3842 rbd_dev
->header_name
= NULL
;
3843 kfree(rbd_dev
->header
.object_prefix
);
3844 rbd_dev
->header
.object_prefix
= NULL
;
3849 static int rbd_dev_probe_finish(struct rbd_device
*rbd_dev
)
3853 /* no need to lock here, as rbd_dev is not registered yet */
3854 ret
= rbd_dev_snaps_update(rbd_dev
);
3858 ret
= rbd_dev_probe_update_spec(rbd_dev
);
3862 ret
= rbd_dev_set_mapping(rbd_dev
);
3866 /* generate unique id: find highest unique id, add one */
3867 rbd_dev_id_get(rbd_dev
);
3869 /* Fill in the device name, now that we have its id. */
3870 BUILD_BUG_ON(DEV_NAME_LEN
3871 < sizeof (RBD_DRV_NAME
) + MAX_INT_FORMAT_WIDTH
);
3872 sprintf(rbd_dev
->name
, "%s%d", RBD_DRV_NAME
, rbd_dev
->dev_id
);
3874 /* Get our block major device number. */
3876 ret
= register_blkdev(0, rbd_dev
->name
);
3879 rbd_dev
->major
= ret
;
3881 /* Set up the blkdev mapping. */
3883 ret
= rbd_init_disk(rbd_dev
);
3885 goto err_out_blkdev
;
3887 ret
= rbd_bus_add_dev(rbd_dev
);
3892 * At this point cleanup in the event of an error is the job
3893 * of the sysfs code (initiated by rbd_bus_del_dev()).
3895 down_write(&rbd_dev
->header_rwsem
);
3896 ret
= rbd_dev_snaps_register(rbd_dev
);
3897 up_write(&rbd_dev
->header_rwsem
);
3901 ret
= rbd_dev_header_watch_sync(rbd_dev
, 1);
3905 /* Everything's ready. Announce the disk to the world. */
3907 add_disk(rbd_dev
->disk
);
3909 pr_info("%s: added with size 0x%llx\n", rbd_dev
->disk
->disk_name
,
3910 (unsigned long long) rbd_dev
->mapping
.size
);
3914 /* this will also clean up rest of rbd_dev stuff */
3916 rbd_bus_del_dev(rbd_dev
);
3920 rbd_free_disk(rbd_dev
);
3922 unregister_blkdev(rbd_dev
->major
, rbd_dev
->name
);
3924 rbd_dev_id_put(rbd_dev
);
3926 rbd_remove_all_snaps(rbd_dev
);
3932 * Probe for the existence of the header object for the given rbd
3933 * device. For format 2 images this includes determining the image
3936 static int rbd_dev_probe(struct rbd_device
*rbd_dev
)
3941 * Get the id from the image id object. If it's not a
3942 * format 2 image, we'll get ENOENT back, and we'll assume
3943 * it's a format 1 image.
3945 ret
= rbd_dev_image_id(rbd_dev
);
3947 ret
= rbd_dev_v1_probe(rbd_dev
);
3949 ret
= rbd_dev_v2_probe(rbd_dev
);
3951 dout("probe failed, returning %d\n", ret
);
3956 ret
= rbd_dev_probe_finish(rbd_dev
);
3958 rbd_header_free(&rbd_dev
->header
);
3963 static ssize_t
rbd_add(struct bus_type
*bus
,
3967 struct rbd_device
*rbd_dev
= NULL
;
3968 struct ceph_options
*ceph_opts
= NULL
;
3969 struct rbd_options
*rbd_opts
= NULL
;
3970 struct rbd_spec
*spec
= NULL
;
3971 struct rbd_client
*rbdc
;
3972 struct ceph_osd_client
*osdc
;
3975 if (!try_module_get(THIS_MODULE
))
3978 /* parse add command */
3979 rc
= rbd_add_parse_args(buf
, &ceph_opts
, &rbd_opts
, &spec
);
3981 goto err_out_module
;
3983 rbdc
= rbd_get_client(ceph_opts
);
3988 ceph_opts
= NULL
; /* rbd_dev client now owns this */
3991 osdc
= &rbdc
->client
->osdc
;
3992 rc
= ceph_pg_poolid_by_name(osdc
->osdmap
, spec
->pool_name
);
3994 goto err_out_client
;
3995 spec
->pool_id
= (u64
) rc
;
3997 /* The ceph file layout needs to fit pool id in 32 bits */
3999 if (WARN_ON(spec
->pool_id
> (u64
) U32_MAX
)) {
4001 goto err_out_client
;
4004 rbd_dev
= rbd_dev_create(rbdc
, spec
);
4006 goto err_out_client
;
4007 rbdc
= NULL
; /* rbd_dev now owns this */
4008 spec
= NULL
; /* rbd_dev now owns this */
4010 rbd_dev
->mapping
.read_only
= rbd_opts
->read_only
;
4012 rbd_opts
= NULL
; /* done with this */
4014 rc
= rbd_dev_probe(rbd_dev
);
4016 goto err_out_rbd_dev
;
4020 rbd_dev_destroy(rbd_dev
);
4022 rbd_put_client(rbdc
);
4025 ceph_destroy_options(ceph_opts
);
4029 module_put(THIS_MODULE
);
4031 dout("Error adding device %s\n", buf
);
4033 return (ssize_t
) rc
;
4036 static struct rbd_device
*__rbd_get_dev(unsigned long dev_id
)
4038 struct list_head
*tmp
;
4039 struct rbd_device
*rbd_dev
;
4041 spin_lock(&rbd_dev_list_lock
);
4042 list_for_each(tmp
, &rbd_dev_list
) {
4043 rbd_dev
= list_entry(tmp
, struct rbd_device
, node
);
4044 if (rbd_dev
->dev_id
== dev_id
) {
4045 spin_unlock(&rbd_dev_list_lock
);
4049 spin_unlock(&rbd_dev_list_lock
);
4053 static void rbd_dev_release(struct device
*dev
)
4055 struct rbd_device
*rbd_dev
= dev_to_rbd_dev(dev
);
4057 if (rbd_dev
->watch_event
)
4058 rbd_dev_header_watch_sync(rbd_dev
, 0);
4060 /* clean up and free blkdev */
4061 rbd_free_disk(rbd_dev
);
4062 unregister_blkdev(rbd_dev
->major
, rbd_dev
->name
);
4064 /* release allocated disk header fields */
4065 rbd_header_free(&rbd_dev
->header
);
4067 /* done with the id, and with the rbd_dev */
4068 rbd_dev_id_put(rbd_dev
);
4069 rbd_assert(rbd_dev
->rbd_client
!= NULL
);
4070 rbd_dev_destroy(rbd_dev
);
4072 /* release module ref */
4073 module_put(THIS_MODULE
);
4076 static ssize_t
rbd_remove(struct bus_type
*bus
,
4080 struct rbd_device
*rbd_dev
= NULL
;
4085 rc
= strict_strtoul(buf
, 10, &ul
);
4089 /* convert to int; abort if we lost anything in the conversion */
4090 target_id
= (int) ul
;
4091 if (target_id
!= ul
)
4094 mutex_lock_nested(&ctl_mutex
, SINGLE_DEPTH_NESTING
);
4096 rbd_dev
= __rbd_get_dev(target_id
);
4102 spin_lock(&rbd_dev
->lock
);
4103 if (rbd_dev
->open_count
)
4106 set_bit(RBD_DEV_FLAG_REMOVING
, &rbd_dev
->flags
);
4107 spin_unlock(&rbd_dev
->lock
);
4111 rbd_remove_all_snaps(rbd_dev
);
4112 rbd_bus_del_dev(rbd_dev
);
4115 mutex_unlock(&ctl_mutex
);
4121 * create control files in sysfs
4124 static int rbd_sysfs_init(void)
4128 ret
= device_register(&rbd_root_dev
);
4132 ret
= bus_register(&rbd_bus_type
);
4134 device_unregister(&rbd_root_dev
);
4139 static void rbd_sysfs_cleanup(void)
4141 bus_unregister(&rbd_bus_type
);
4142 device_unregister(&rbd_root_dev
);
4145 int __init
rbd_init(void)
4149 if (!libceph_compatible(NULL
)) {
4150 rbd_warn(NULL
, "libceph incompatibility (quitting)");
4154 rc
= rbd_sysfs_init();
4157 pr_info("loaded " RBD_DRV_NAME_LONG
"\n");
4161 void __exit
rbd_exit(void)
4163 rbd_sysfs_cleanup();
4166 module_init(rbd_init
);
4167 module_exit(rbd_exit
);
4169 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
4170 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
4171 MODULE_DESCRIPTION("rados block device");
4173 /* following authorship retained from original osdblk.c */
4174 MODULE_AUTHOR("Jeff Garzik <jeff@garzik.org>");
4176 MODULE_LICENSE("GPL");