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 this defined elsewhere too */
57 #define U64_MAX ((u64) (~0ULL))
59 #define RBD_DRV_NAME "rbd"
60 #define RBD_DRV_NAME_LONG "rbd (rados block device)"
62 #define RBD_MINORS_PER_MAJOR 256 /* max minors per blkdev */
64 #define RBD_MAX_SNAP_NAME_LEN 32
65 #define RBD_MAX_OPT_LEN 1024
67 #define RBD_SNAP_HEAD_NAME "-"
69 #define RBD_IMAGE_ID_LEN_MAX 64
72 * An RBD device name will be "rbd#", where the "rbd" comes from
73 * RBD_DRV_NAME above, and # is a unique integer identifier.
74 * MAX_INT_FORMAT_WIDTH is used in ensuring DEV_NAME_LEN is big
75 * enough to hold all possible device names.
77 #define DEV_NAME_LEN 32
78 #define MAX_INT_FORMAT_WIDTH ((5 * sizeof (int)) / 2 + 1)
80 #define RBD_READ_ONLY_DEFAULT false
83 * block device image metadata (in-memory version)
85 struct rbd_image_header
{
86 /* These four fields never change for a given rbd image */
92 /* The remaining fields need to be updated occasionally */
94 struct ceph_snap_context
*snapc
;
106 * an instance of the client. multiple devices may share an rbd client.
109 struct ceph_client
*client
;
111 struct list_head node
;
115 * a request completion status
117 struct rbd_req_status
{
124 * a collection of requests
126 struct rbd_req_coll
{
130 struct rbd_req_status status
[0];
134 * a single io request
137 struct request
*rq
; /* blk layer request */
138 struct bio
*bio
; /* cloned bio */
139 struct page
**pages
; /* list of used pages */
142 struct rbd_req_coll
*coll
;
149 struct list_head node
;
165 int dev_id
; /* blkdev unique id */
167 int major
; /* blkdev assigned major */
168 struct gendisk
*disk
; /* blkdev's gendisk and rq */
170 struct rbd_options rbd_opts
;
171 struct rbd_client
*rbd_client
;
173 char name
[DEV_NAME_LEN
]; /* blkdev name, e.g. rbd3 */
175 spinlock_t lock
; /* queue lock */
177 struct rbd_image_header header
;
181 size_t image_name_len
;
186 struct ceph_osd_event
*watch_event
;
187 struct ceph_osd_request
*watch_request
;
189 /* protects updating the header */
190 struct rw_semaphore header_rwsem
;
192 struct rbd_mapping mapping
;
194 struct list_head node
;
196 /* list of snapshots */
197 struct list_head snaps
;
203 static DEFINE_MUTEX(ctl_mutex
); /* Serialize open/close/setup/teardown */
205 static LIST_HEAD(rbd_dev_list
); /* devices */
206 static DEFINE_SPINLOCK(rbd_dev_list_lock
);
208 static LIST_HEAD(rbd_client_list
); /* clients */
209 static DEFINE_SPINLOCK(rbd_client_list_lock
);
211 static int rbd_dev_snaps_update(struct rbd_device
*rbd_dev
);
212 static int rbd_dev_snaps_register(struct rbd_device
*rbd_dev
);
214 static void rbd_dev_release(struct device
*dev
);
215 static void __rbd_remove_snap_dev(struct rbd_snap
*snap
);
217 static ssize_t
rbd_add(struct bus_type
*bus
, const char *buf
,
219 static ssize_t
rbd_remove(struct bus_type
*bus
, const char *buf
,
222 static struct bus_attribute rbd_bus_attrs
[] = {
223 __ATTR(add
, S_IWUSR
, NULL
, rbd_add
),
224 __ATTR(remove
, S_IWUSR
, NULL
, rbd_remove
),
228 static struct bus_type rbd_bus_type
= {
230 .bus_attrs
= rbd_bus_attrs
,
233 static void rbd_root_dev_release(struct device
*dev
)
237 static struct device rbd_root_dev
= {
239 .release
= rbd_root_dev_release
,
243 #define rbd_assert(expr) \
244 if (unlikely(!(expr))) { \
245 printk(KERN_ERR "\nAssertion failure in %s() " \
247 "\trbd_assert(%s);\n\n", \
248 __func__, __LINE__, #expr); \
251 #else /* !RBD_DEBUG */
252 # define rbd_assert(expr) ((void) 0)
253 #endif /* !RBD_DEBUG */
255 static struct device
*rbd_get_dev(struct rbd_device
*rbd_dev
)
257 return get_device(&rbd_dev
->dev
);
260 static void rbd_put_dev(struct rbd_device
*rbd_dev
)
262 put_device(&rbd_dev
->dev
);
265 static int rbd_refresh_header(struct rbd_device
*rbd_dev
, u64
*hver
);
267 static int rbd_open(struct block_device
*bdev
, fmode_t mode
)
269 struct rbd_device
*rbd_dev
= bdev
->bd_disk
->private_data
;
271 if ((mode
& FMODE_WRITE
) && rbd_dev
->mapping
.read_only
)
274 rbd_get_dev(rbd_dev
);
275 set_device_ro(bdev
, rbd_dev
->mapping
.read_only
);
280 static int rbd_release(struct gendisk
*disk
, fmode_t mode
)
282 struct rbd_device
*rbd_dev
= disk
->private_data
;
284 rbd_put_dev(rbd_dev
);
289 static const struct block_device_operations rbd_bd_ops
= {
290 .owner
= THIS_MODULE
,
292 .release
= rbd_release
,
296 * Initialize an rbd client instance.
299 static struct rbd_client
*rbd_client_create(struct ceph_options
*ceph_opts
)
301 struct rbd_client
*rbdc
;
304 dout("rbd_client_create\n");
305 rbdc
= kmalloc(sizeof(struct rbd_client
), GFP_KERNEL
);
309 kref_init(&rbdc
->kref
);
310 INIT_LIST_HEAD(&rbdc
->node
);
312 mutex_lock_nested(&ctl_mutex
, SINGLE_DEPTH_NESTING
);
314 rbdc
->client
= ceph_create_client(ceph_opts
, rbdc
, 0, 0);
315 if (IS_ERR(rbdc
->client
))
317 ceph_opts
= NULL
; /* Now rbdc->client is responsible for ceph_opts */
319 ret
= ceph_open_session(rbdc
->client
);
323 spin_lock(&rbd_client_list_lock
);
324 list_add_tail(&rbdc
->node
, &rbd_client_list
);
325 spin_unlock(&rbd_client_list_lock
);
327 mutex_unlock(&ctl_mutex
);
329 dout("rbd_client_create created %p\n", rbdc
);
333 ceph_destroy_client(rbdc
->client
);
335 mutex_unlock(&ctl_mutex
);
339 ceph_destroy_options(ceph_opts
);
344 * Find a ceph client with specific addr and configuration. If
345 * found, bump its reference count.
347 static struct rbd_client
*rbd_client_find(struct ceph_options
*ceph_opts
)
349 struct rbd_client
*client_node
;
352 if (ceph_opts
->flags
& CEPH_OPT_NOSHARE
)
355 spin_lock(&rbd_client_list_lock
);
356 list_for_each_entry(client_node
, &rbd_client_list
, node
) {
357 if (!ceph_compare_options(ceph_opts
, client_node
->client
)) {
358 kref_get(&client_node
->kref
);
363 spin_unlock(&rbd_client_list_lock
);
365 return found
? client_node
: NULL
;
375 /* string args above */
378 /* Boolean args above */
382 static match_table_t rbd_opts_tokens
= {
384 /* string args above */
385 {Opt_read_only
, "mapping.read_only"},
386 {Opt_read_only
, "ro"}, /* Alternate spelling */
387 {Opt_read_write
, "read_write"},
388 {Opt_read_write
, "rw"}, /* Alternate spelling */
389 /* Boolean args above */
393 static int parse_rbd_opts_token(char *c
, void *private)
395 struct rbd_options
*rbd_opts
= private;
396 substring_t argstr
[MAX_OPT_ARGS
];
397 int token
, intval
, ret
;
399 token
= match_token(c
, rbd_opts_tokens
, argstr
);
403 if (token
< Opt_last_int
) {
404 ret
= match_int(&argstr
[0], &intval
);
406 pr_err("bad mount option arg (not int) "
410 dout("got int token %d val %d\n", token
, intval
);
411 } else if (token
> Opt_last_int
&& token
< Opt_last_string
) {
412 dout("got string token %d val %s\n", token
,
414 } else if (token
> Opt_last_string
&& token
< Opt_last_bool
) {
415 dout("got Boolean token %d\n", token
);
417 dout("got token %d\n", token
);
422 rbd_opts
->read_only
= true;
425 rbd_opts
->read_only
= false;
435 * Get a ceph client with specific addr and configuration, if one does
436 * not exist create it.
438 static int rbd_get_client(struct rbd_device
*rbd_dev
, const char *mon_addr
,
439 size_t mon_addr_len
, char *options
)
441 struct rbd_options
*rbd_opts
= &rbd_dev
->rbd_opts
;
442 struct ceph_options
*ceph_opts
;
443 struct rbd_client
*rbdc
;
445 rbd_opts
->read_only
= RBD_READ_ONLY_DEFAULT
;
447 ceph_opts
= ceph_parse_options(options
, mon_addr
,
448 mon_addr
+ mon_addr_len
,
449 parse_rbd_opts_token
, rbd_opts
);
450 if (IS_ERR(ceph_opts
))
451 return PTR_ERR(ceph_opts
);
453 rbdc
= rbd_client_find(ceph_opts
);
455 /* using an existing client */
456 ceph_destroy_options(ceph_opts
);
458 rbdc
= rbd_client_create(ceph_opts
);
460 return PTR_ERR(rbdc
);
462 rbd_dev
->rbd_client
= rbdc
;
468 * Destroy ceph client
470 * Caller must hold rbd_client_list_lock.
472 static void rbd_client_release(struct kref
*kref
)
474 struct rbd_client
*rbdc
= container_of(kref
, struct rbd_client
, kref
);
476 dout("rbd_release_client %p\n", rbdc
);
477 spin_lock(&rbd_client_list_lock
);
478 list_del(&rbdc
->node
);
479 spin_unlock(&rbd_client_list_lock
);
481 ceph_destroy_client(rbdc
->client
);
486 * Drop reference to ceph client node. If it's not referenced anymore, release
489 static void rbd_put_client(struct rbd_device
*rbd_dev
)
491 kref_put(&rbd_dev
->rbd_client
->kref
, rbd_client_release
);
492 rbd_dev
->rbd_client
= NULL
;
496 * Destroy requests collection
498 static void rbd_coll_release(struct kref
*kref
)
500 struct rbd_req_coll
*coll
=
501 container_of(kref
, struct rbd_req_coll
, kref
);
503 dout("rbd_coll_release %p\n", coll
);
507 static bool rbd_dev_ondisk_valid(struct rbd_image_header_ondisk
*ondisk
)
512 /* The header has to start with the magic rbd header text */
513 if (memcmp(&ondisk
->text
, RBD_HEADER_TEXT
, sizeof (RBD_HEADER_TEXT
)))
517 * The size of a snapshot header has to fit in a size_t, and
518 * that limits the number of snapshots.
520 snap_count
= le32_to_cpu(ondisk
->snap_count
);
521 size
= SIZE_MAX
- sizeof (struct ceph_snap_context
);
522 if (snap_count
> size
/ sizeof (__le64
))
526 * Not only that, but the size of the entire the snapshot
527 * header must also be representable in a size_t.
529 size
-= snap_count
* sizeof (__le64
);
530 if ((u64
) size
< le64_to_cpu(ondisk
->snap_names_len
))
537 * Create a new header structure, translate header format from the on-disk
540 static int rbd_header_from_disk(struct rbd_image_header
*header
,
541 struct rbd_image_header_ondisk
*ondisk
)
548 memset(header
, 0, sizeof (*header
));
550 snap_count
= le32_to_cpu(ondisk
->snap_count
);
552 len
= strnlen(ondisk
->object_prefix
, sizeof (ondisk
->object_prefix
));
553 header
->object_prefix
= kmalloc(len
+ 1, GFP_KERNEL
);
554 if (!header
->object_prefix
)
556 memcpy(header
->object_prefix
, ondisk
->object_prefix
, len
);
557 header
->object_prefix
[len
] = '\0';
560 u64 snap_names_len
= le64_to_cpu(ondisk
->snap_names_len
);
562 /* Save a copy of the snapshot names */
564 if (snap_names_len
> (u64
) SIZE_MAX
)
566 header
->snap_names
= kmalloc(snap_names_len
, GFP_KERNEL
);
567 if (!header
->snap_names
)
570 * Note that rbd_dev_v1_header_read() guarantees
571 * the ondisk buffer we're working with has
572 * snap_names_len bytes beyond the end of the
573 * snapshot id array, this memcpy() is safe.
575 memcpy(header
->snap_names
, &ondisk
->snaps
[snap_count
],
578 /* Record each snapshot's size */
580 size
= snap_count
* sizeof (*header
->snap_sizes
);
581 header
->snap_sizes
= kmalloc(size
, GFP_KERNEL
);
582 if (!header
->snap_sizes
)
584 for (i
= 0; i
< snap_count
; i
++)
585 header
->snap_sizes
[i
] =
586 le64_to_cpu(ondisk
->snaps
[i
].image_size
);
588 WARN_ON(ondisk
->snap_names_len
);
589 header
->snap_names
= NULL
;
590 header
->snap_sizes
= NULL
;
593 header
->obj_order
= ondisk
->options
.order
;
594 header
->crypt_type
= ondisk
->options
.crypt_type
;
595 header
->comp_type
= ondisk
->options
.comp_type
;
597 /* Allocate and fill in the snapshot context */
599 header
->image_size
= le64_to_cpu(ondisk
->image_size
);
600 size
= sizeof (struct ceph_snap_context
);
601 size
+= snap_count
* sizeof (header
->snapc
->snaps
[0]);
602 header
->snapc
= kzalloc(size
, GFP_KERNEL
);
606 atomic_set(&header
->snapc
->nref
, 1);
607 header
->snapc
->seq
= le64_to_cpu(ondisk
->snap_seq
);
608 header
->snapc
->num_snaps
= snap_count
;
609 for (i
= 0; i
< snap_count
; i
++)
610 header
->snapc
->snaps
[i
] =
611 le64_to_cpu(ondisk
->snaps
[i
].id
);
616 kfree(header
->snap_sizes
);
617 header
->snap_sizes
= NULL
;
618 kfree(header
->snap_names
);
619 header
->snap_names
= NULL
;
620 kfree(header
->object_prefix
);
621 header
->object_prefix
= NULL
;
626 static int snap_by_name(struct rbd_device
*rbd_dev
, const char *snap_name
)
629 struct rbd_snap
*snap
;
631 list_for_each_entry(snap
, &rbd_dev
->snaps
, node
) {
632 if (!strcmp(snap_name
, snap
->name
)) {
633 rbd_dev
->mapping
.snap_id
= snap
->id
;
634 rbd_dev
->mapping
.size
= snap
->size
;
643 static int rbd_dev_set_mapping(struct rbd_device
*rbd_dev
, char *snap_name
)
647 if (!memcmp(snap_name
, RBD_SNAP_HEAD_NAME
,
648 sizeof (RBD_SNAP_HEAD_NAME
))) {
649 rbd_dev
->mapping
.snap_id
= CEPH_NOSNAP
;
650 rbd_dev
->mapping
.size
= rbd_dev
->header
.image_size
;
651 rbd_dev
->mapping
.snap_exists
= false;
652 rbd_dev
->mapping
.read_only
= rbd_dev
->rbd_opts
.read_only
;
655 ret
= snap_by_name(rbd_dev
, snap_name
);
658 rbd_dev
->mapping
.snap_exists
= true;
659 rbd_dev
->mapping
.read_only
= true;
661 rbd_dev
->mapping
.snap_name
= snap_name
;
666 static void rbd_header_free(struct rbd_image_header
*header
)
668 kfree(header
->object_prefix
);
669 header
->object_prefix
= NULL
;
670 kfree(header
->snap_sizes
);
671 header
->snap_sizes
= NULL
;
672 kfree(header
->snap_names
);
673 header
->snap_names
= NULL
;
674 ceph_put_snap_context(header
->snapc
);
675 header
->snapc
= NULL
;
678 static char *rbd_segment_name(struct rbd_device
*rbd_dev
, u64 offset
)
684 name
= kmalloc(RBD_MAX_SEG_NAME_LEN
+ 1, GFP_NOIO
);
687 segment
= offset
>> rbd_dev
->header
.obj_order
;
688 ret
= snprintf(name
, RBD_MAX_SEG_NAME_LEN
, "%s.%012llx",
689 rbd_dev
->header
.object_prefix
, segment
);
690 if (ret
< 0 || ret
>= RBD_MAX_SEG_NAME_LEN
) {
691 pr_err("error formatting segment name for #%llu (%d)\n",
700 static u64
rbd_segment_offset(struct rbd_device
*rbd_dev
, u64 offset
)
702 u64 segment_size
= (u64
) 1 << rbd_dev
->header
.obj_order
;
704 return offset
& (segment_size
- 1);
707 static u64
rbd_segment_length(struct rbd_device
*rbd_dev
,
708 u64 offset
, u64 length
)
710 u64 segment_size
= (u64
) 1 << rbd_dev
->header
.obj_order
;
712 offset
&= segment_size
- 1;
714 rbd_assert(length
<= U64_MAX
- offset
);
715 if (offset
+ length
> segment_size
)
716 length
= segment_size
- offset
;
721 static int rbd_get_num_segments(struct rbd_image_header
*header
,
729 if (len
- 1 > U64_MAX
- ofs
)
732 start_seg
= ofs
>> header
->obj_order
;
733 end_seg
= (ofs
+ len
- 1) >> header
->obj_order
;
735 return end_seg
- start_seg
+ 1;
739 * returns the size of an object in the image
741 static u64
rbd_obj_bytes(struct rbd_image_header
*header
)
743 return 1 << header
->obj_order
;
750 static void bio_chain_put(struct bio
*chain
)
756 chain
= chain
->bi_next
;
762 * zeros a bio chain, starting at specific offset
764 static void zero_bio_chain(struct bio
*chain
, int start_ofs
)
773 bio_for_each_segment(bv
, chain
, i
) {
774 if (pos
+ bv
->bv_len
> start_ofs
) {
775 int remainder
= max(start_ofs
- pos
, 0);
776 buf
= bvec_kmap_irq(bv
, &flags
);
777 memset(buf
+ remainder
, 0,
778 bv
->bv_len
- remainder
);
779 bvec_kunmap_irq(buf
, &flags
);
784 chain
= chain
->bi_next
;
789 * bio_chain_clone - clone a chain of bios up to a certain length.
790 * might return a bio_pair that will need to be released.
792 static struct bio
*bio_chain_clone(struct bio
**old
, struct bio
**next
,
793 struct bio_pair
**bp
,
794 int len
, gfp_t gfpmask
)
796 struct bio
*old_chain
= *old
;
797 struct bio
*new_chain
= NULL
;
802 bio_pair_release(*bp
);
806 while (old_chain
&& (total
< len
)) {
809 tmp
= bio_kmalloc(gfpmask
, old_chain
->bi_max_vecs
);
812 gfpmask
&= ~__GFP_WAIT
; /* can't wait after the first */
814 if (total
+ old_chain
->bi_size
> len
) {
818 * this split can only happen with a single paged bio,
819 * split_bio will BUG_ON if this is not the case
821 dout("bio_chain_clone split! total=%d remaining=%d"
823 total
, len
- total
, old_chain
->bi_size
);
825 /* split the bio. We'll release it either in the next
826 call, or it will have to be released outside */
827 bp
= bio_split(old_chain
, (len
- total
) / SECTOR_SIZE
);
831 __bio_clone(tmp
, &bp
->bio1
);
835 __bio_clone(tmp
, old_chain
);
836 *next
= old_chain
->bi_next
;
846 old_chain
= old_chain
->bi_next
;
848 total
+= tmp
->bi_size
;
851 rbd_assert(total
== len
);
858 dout("bio_chain_clone with err\n");
859 bio_chain_put(new_chain
);
864 * helpers for osd request op vectors.
866 static struct ceph_osd_req_op
*rbd_create_rw_ops(int num_ops
,
867 int opcode
, u32 payload_len
)
869 struct ceph_osd_req_op
*ops
;
871 ops
= kzalloc(sizeof (*ops
) * (num_ops
+ 1), GFP_NOIO
);
878 * op extent offset and length will be set later on
879 * in calc_raw_layout()
881 ops
[0].payload_len
= payload_len
;
886 static void rbd_destroy_ops(struct ceph_osd_req_op
*ops
)
891 static void rbd_coll_end_req_index(struct request
*rq
,
892 struct rbd_req_coll
*coll
,
896 struct request_queue
*q
;
899 dout("rbd_coll_end_req_index %p index %d ret %d len %llu\n",
900 coll
, index
, ret
, (unsigned long long) len
);
906 blk_end_request(rq
, ret
, len
);
912 spin_lock_irq(q
->queue_lock
);
913 coll
->status
[index
].done
= 1;
914 coll
->status
[index
].rc
= ret
;
915 coll
->status
[index
].bytes
= len
;
916 max
= min
= coll
->num_done
;
917 while (max
< coll
->total
&& coll
->status
[max
].done
)
920 for (i
= min
; i
<max
; i
++) {
921 __blk_end_request(rq
, coll
->status
[i
].rc
,
922 coll
->status
[i
].bytes
);
924 kref_put(&coll
->kref
, rbd_coll_release
);
926 spin_unlock_irq(q
->queue_lock
);
929 static void rbd_coll_end_req(struct rbd_request
*req
,
932 rbd_coll_end_req_index(req
->rq
, req
->coll
, req
->coll_index
, ret
, len
);
936 * Send ceph osd request
938 static int rbd_do_request(struct request
*rq
,
939 struct rbd_device
*rbd_dev
,
940 struct ceph_snap_context
*snapc
,
942 const char *object_name
, u64 ofs
, u64 len
,
947 struct ceph_osd_req_op
*ops
,
948 struct rbd_req_coll
*coll
,
950 void (*rbd_cb
)(struct ceph_osd_request
*req
,
951 struct ceph_msg
*msg
),
952 struct ceph_osd_request
**linger_req
,
955 struct ceph_osd_request
*req
;
956 struct ceph_file_layout
*layout
;
959 struct timespec mtime
= CURRENT_TIME
;
960 struct rbd_request
*req_data
;
961 struct ceph_osd_request_head
*reqhead
;
962 struct ceph_osd_client
*osdc
;
964 req_data
= kzalloc(sizeof(*req_data
), GFP_NOIO
);
967 rbd_coll_end_req_index(rq
, coll
, coll_index
,
973 req_data
->coll
= coll
;
974 req_data
->coll_index
= coll_index
;
977 dout("rbd_do_request object_name=%s ofs=%llu len=%llu\n", object_name
,
978 (unsigned long long) ofs
, (unsigned long long) len
);
980 osdc
= &rbd_dev
->rbd_client
->client
->osdc
;
981 req
= ceph_osdc_alloc_request(osdc
, flags
, snapc
, ops
,
982 false, GFP_NOIO
, pages
, bio
);
988 req
->r_callback
= rbd_cb
;
992 req_data
->pages
= pages
;
995 req
->r_priv
= req_data
;
997 reqhead
= req
->r_request
->front
.iov_base
;
998 reqhead
->snapid
= cpu_to_le64(CEPH_NOSNAP
);
1000 strncpy(req
->r_oid
, object_name
, sizeof(req
->r_oid
));
1001 req
->r_oid_len
= strlen(req
->r_oid
);
1003 layout
= &req
->r_file_layout
;
1004 memset(layout
, 0, sizeof(*layout
));
1005 layout
->fl_stripe_unit
= cpu_to_le32(1 << RBD_MAX_OBJ_ORDER
);
1006 layout
->fl_stripe_count
= cpu_to_le32(1);
1007 layout
->fl_object_size
= cpu_to_le32(1 << RBD_MAX_OBJ_ORDER
);
1008 layout
->fl_pg_pool
= cpu_to_le32(rbd_dev
->pool_id
);
1009 ceph_calc_raw_layout(osdc
, layout
, snapid
, ofs
, &len
, &bno
,
1012 ceph_osdc_build_request(req
, ofs
, &len
,
1016 req
->r_oid
, req
->r_oid_len
);
1019 ceph_osdc_set_request_linger(osdc
, req
);
1023 ret
= ceph_osdc_start_request(osdc
, req
, false);
1028 ret
= ceph_osdc_wait_request(osdc
, req
);
1030 *ver
= le64_to_cpu(req
->r_reassert_version
.version
);
1031 dout("reassert_ver=%llu\n",
1032 (unsigned long long)
1033 le64_to_cpu(req
->r_reassert_version
.version
));
1034 ceph_osdc_put_request(req
);
1039 bio_chain_put(req_data
->bio
);
1040 ceph_osdc_put_request(req
);
1042 rbd_coll_end_req(req_data
, ret
, len
);
1048 * Ceph osd op callback
1050 static void rbd_req_cb(struct ceph_osd_request
*req
, struct ceph_msg
*msg
)
1052 struct rbd_request
*req_data
= req
->r_priv
;
1053 struct ceph_osd_reply_head
*replyhead
;
1054 struct ceph_osd_op
*op
;
1060 replyhead
= msg
->front
.iov_base
;
1061 WARN_ON(le32_to_cpu(replyhead
->num_ops
) == 0);
1062 op
= (void *)(replyhead
+ 1);
1063 rc
= le32_to_cpu(replyhead
->result
);
1064 bytes
= le64_to_cpu(op
->extent
.length
);
1065 read_op
= (le16_to_cpu(op
->op
) == CEPH_OSD_OP_READ
);
1067 dout("rbd_req_cb bytes=%llu readop=%d rc=%d\n",
1068 (unsigned long long) bytes
, read_op
, (int) rc
);
1070 if (rc
== -ENOENT
&& read_op
) {
1071 zero_bio_chain(req_data
->bio
, 0);
1073 } else if (rc
== 0 && read_op
&& bytes
< req_data
->len
) {
1074 zero_bio_chain(req_data
->bio
, bytes
);
1075 bytes
= req_data
->len
;
1078 rbd_coll_end_req(req_data
, rc
, bytes
);
1081 bio_chain_put(req_data
->bio
);
1083 ceph_osdc_put_request(req
);
1087 static void rbd_simple_req_cb(struct ceph_osd_request
*req
, struct ceph_msg
*msg
)
1089 ceph_osdc_put_request(req
);
1093 * Do a synchronous ceph osd operation
1095 static int rbd_req_sync_op(struct rbd_device
*rbd_dev
,
1096 struct ceph_snap_context
*snapc
,
1099 struct ceph_osd_req_op
*ops
,
1100 const char *object_name
,
1101 u64 ofs
, u64 inbound_size
,
1103 struct ceph_osd_request
**linger_req
,
1107 struct page
**pages
;
1110 rbd_assert(ops
!= NULL
);
1112 num_pages
= calc_pages_for(ofs
, inbound_size
);
1113 pages
= ceph_alloc_page_vector(num_pages
, GFP_KERNEL
);
1115 return PTR_ERR(pages
);
1117 ret
= rbd_do_request(NULL
, rbd_dev
, snapc
, snapid
,
1118 object_name
, ofs
, inbound_size
, NULL
,
1128 if ((flags
& CEPH_OSD_FLAG_READ
) && inbound
)
1129 ret
= ceph_copy_from_page_vector(pages
, inbound
, ofs
, ret
);
1132 ceph_release_page_vector(pages
, num_pages
);
1137 * Do an asynchronous ceph osd operation
1139 static int rbd_do_op(struct request
*rq
,
1140 struct rbd_device
*rbd_dev
,
1141 struct ceph_snap_context
*snapc
,
1143 int opcode
, int flags
,
1146 struct rbd_req_coll
*coll
,
1153 struct ceph_osd_req_op
*ops
;
1156 seg_name
= rbd_segment_name(rbd_dev
, ofs
);
1159 seg_len
= rbd_segment_length(rbd_dev
, ofs
, len
);
1160 seg_ofs
= rbd_segment_offset(rbd_dev
, ofs
);
1162 payload_len
= (flags
& CEPH_OSD_FLAG_WRITE
? seg_len
: 0);
1165 ops
= rbd_create_rw_ops(1, opcode
, payload_len
);
1169 /* we've taken care of segment sizes earlier when we
1170 cloned the bios. We should never have a segment
1171 truncated at this point */
1172 rbd_assert(seg_len
== len
);
1174 ret
= rbd_do_request(rq
, rbd_dev
, snapc
, snapid
,
1175 seg_name
, seg_ofs
, seg_len
,
1181 rbd_req_cb
, 0, NULL
);
1183 rbd_destroy_ops(ops
);
1190 * Request async osd write
1192 static int rbd_req_write(struct request
*rq
,
1193 struct rbd_device
*rbd_dev
,
1194 struct ceph_snap_context
*snapc
,
1197 struct rbd_req_coll
*coll
,
1200 return rbd_do_op(rq
, rbd_dev
, snapc
, CEPH_NOSNAP
,
1202 CEPH_OSD_FLAG_WRITE
| CEPH_OSD_FLAG_ONDISK
,
1203 ofs
, len
, bio
, coll
, coll_index
);
1207 * Request async osd read
1209 static int rbd_req_read(struct request
*rq
,
1210 struct rbd_device
*rbd_dev
,
1214 struct rbd_req_coll
*coll
,
1217 return rbd_do_op(rq
, rbd_dev
, NULL
,
1221 ofs
, len
, bio
, coll
, coll_index
);
1225 * Request sync osd read
1227 static int rbd_req_sync_read(struct rbd_device
*rbd_dev
,
1229 const char *object_name
,
1234 struct ceph_osd_req_op
*ops
;
1237 ops
= rbd_create_rw_ops(1, CEPH_OSD_OP_READ
, 0);
1241 ret
= rbd_req_sync_op(rbd_dev
, NULL
,
1244 ops
, object_name
, ofs
, len
, buf
, NULL
, ver
);
1245 rbd_destroy_ops(ops
);
1251 * Request sync osd watch
1253 static int rbd_req_sync_notify_ack(struct rbd_device
*rbd_dev
,
1257 struct ceph_osd_req_op
*ops
;
1260 ops
= rbd_create_rw_ops(1, CEPH_OSD_OP_NOTIFY_ACK
, 0);
1264 ops
[0].watch
.ver
= cpu_to_le64(ver
);
1265 ops
[0].watch
.cookie
= notify_id
;
1266 ops
[0].watch
.flag
= 0;
1268 ret
= rbd_do_request(NULL
, rbd_dev
, NULL
, CEPH_NOSNAP
,
1269 rbd_dev
->header_name
, 0, 0, NULL
,
1274 rbd_simple_req_cb
, 0, NULL
);
1276 rbd_destroy_ops(ops
);
1280 static void rbd_watch_cb(u64 ver
, u64 notify_id
, u8 opcode
, void *data
)
1282 struct rbd_device
*rbd_dev
= (struct rbd_device
*)data
;
1289 dout("rbd_watch_cb %s notify_id=%llu opcode=%u\n",
1290 rbd_dev
->header_name
, (unsigned long long) notify_id
,
1291 (unsigned int) opcode
);
1292 rc
= rbd_refresh_header(rbd_dev
, &hver
);
1294 pr_warning(RBD_DRV_NAME
"%d got notification but failed to "
1295 " update snaps: %d\n", rbd_dev
->major
, rc
);
1297 rbd_req_sync_notify_ack(rbd_dev
, hver
, notify_id
);
1301 * Request sync osd watch
1303 static int rbd_req_sync_watch(struct rbd_device
*rbd_dev
)
1305 struct ceph_osd_req_op
*ops
;
1306 struct ceph_osd_client
*osdc
= &rbd_dev
->rbd_client
->client
->osdc
;
1309 ops
= rbd_create_rw_ops(1, CEPH_OSD_OP_WATCH
, 0);
1313 ret
= ceph_osdc_create_event(osdc
, rbd_watch_cb
, 0,
1314 (void *)rbd_dev
, &rbd_dev
->watch_event
);
1318 ops
[0].watch
.ver
= cpu_to_le64(rbd_dev
->header
.obj_version
);
1319 ops
[0].watch
.cookie
= cpu_to_le64(rbd_dev
->watch_event
->cookie
);
1320 ops
[0].watch
.flag
= 1;
1322 ret
= rbd_req_sync_op(rbd_dev
, NULL
,
1324 CEPH_OSD_FLAG_WRITE
| CEPH_OSD_FLAG_ONDISK
,
1326 rbd_dev
->header_name
,
1328 &rbd_dev
->watch_request
, NULL
);
1333 rbd_destroy_ops(ops
);
1337 ceph_osdc_cancel_event(rbd_dev
->watch_event
);
1338 rbd_dev
->watch_event
= NULL
;
1340 rbd_destroy_ops(ops
);
1345 * Request sync osd unwatch
1347 static int rbd_req_sync_unwatch(struct rbd_device
*rbd_dev
)
1349 struct ceph_osd_req_op
*ops
;
1352 ops
= rbd_create_rw_ops(1, CEPH_OSD_OP_WATCH
, 0);
1356 ops
[0].watch
.ver
= 0;
1357 ops
[0].watch
.cookie
= cpu_to_le64(rbd_dev
->watch_event
->cookie
);
1358 ops
[0].watch
.flag
= 0;
1360 ret
= rbd_req_sync_op(rbd_dev
, NULL
,
1362 CEPH_OSD_FLAG_WRITE
| CEPH_OSD_FLAG_ONDISK
,
1364 rbd_dev
->header_name
,
1365 0, 0, NULL
, NULL
, NULL
);
1368 rbd_destroy_ops(ops
);
1369 ceph_osdc_cancel_event(rbd_dev
->watch_event
);
1370 rbd_dev
->watch_event
= NULL
;
1375 * Synchronous osd object method call
1377 static int rbd_req_sync_exec(struct rbd_device
*rbd_dev
,
1378 const char *object_name
,
1379 const char *class_name
,
1380 const char *method_name
,
1381 const char *outbound
,
1382 size_t outbound_size
,
1384 size_t inbound_size
,
1388 struct ceph_osd_req_op
*ops
;
1389 int class_name_len
= strlen(class_name
);
1390 int method_name_len
= strlen(method_name
);
1395 * Any input parameters required by the method we're calling
1396 * will be sent along with the class and method names as
1397 * part of the message payload. That data and its size are
1398 * supplied via the indata and indata_len fields (named from
1399 * the perspective of the server side) in the OSD request
1402 payload_size
= class_name_len
+ method_name_len
+ outbound_size
;
1403 ops
= rbd_create_rw_ops(1, CEPH_OSD_OP_CALL
, payload_size
);
1407 ops
[0].cls
.class_name
= class_name
;
1408 ops
[0].cls
.class_len
= (__u8
) class_name_len
;
1409 ops
[0].cls
.method_name
= method_name
;
1410 ops
[0].cls
.method_len
= (__u8
) method_name_len
;
1411 ops
[0].cls
.argc
= 0;
1412 ops
[0].cls
.indata
= outbound
;
1413 ops
[0].cls
.indata_len
= outbound_size
;
1415 ret
= rbd_req_sync_op(rbd_dev
, NULL
,
1418 object_name
, 0, inbound_size
, inbound
,
1421 rbd_destroy_ops(ops
);
1423 dout("cls_exec returned %d\n", ret
);
1427 static struct rbd_req_coll
*rbd_alloc_coll(int num_reqs
)
1429 struct rbd_req_coll
*coll
=
1430 kzalloc(sizeof(struct rbd_req_coll
) +
1431 sizeof(struct rbd_req_status
) * num_reqs
,
1436 coll
->total
= num_reqs
;
1437 kref_init(&coll
->kref
);
1442 * block device queue callback
1444 static void rbd_rq_fn(struct request_queue
*q
)
1446 struct rbd_device
*rbd_dev
= q
->queuedata
;
1448 struct bio_pair
*bp
= NULL
;
1450 while ((rq
= blk_fetch_request(q
))) {
1452 struct bio
*rq_bio
, *next_bio
= NULL
;
1457 int num_segs
, cur_seg
= 0;
1458 struct rbd_req_coll
*coll
;
1459 struct ceph_snap_context
*snapc
;
1461 dout("fetched request\n");
1463 /* filter out block requests we don't understand */
1464 if ((rq
->cmd_type
!= REQ_TYPE_FS
)) {
1465 __blk_end_request_all(rq
, 0);
1469 /* deduce our operation (read, write) */
1470 do_write
= (rq_data_dir(rq
) == WRITE
);
1472 size
= blk_rq_bytes(rq
);
1473 ofs
= blk_rq_pos(rq
) * SECTOR_SIZE
;
1475 if (do_write
&& rbd_dev
->mapping
.read_only
) {
1476 __blk_end_request_all(rq
, -EROFS
);
1480 spin_unlock_irq(q
->queue_lock
);
1482 down_read(&rbd_dev
->header_rwsem
);
1484 if (rbd_dev
->mapping
.snap_id
!= CEPH_NOSNAP
&&
1485 !rbd_dev
->mapping
.snap_exists
) {
1486 up_read(&rbd_dev
->header_rwsem
);
1487 dout("request for non-existent snapshot");
1488 spin_lock_irq(q
->queue_lock
);
1489 __blk_end_request_all(rq
, -ENXIO
);
1493 snapc
= ceph_get_snap_context(rbd_dev
->header
.snapc
);
1495 up_read(&rbd_dev
->header_rwsem
);
1497 dout("%s 0x%x bytes at 0x%llx\n",
1498 do_write
? "write" : "read",
1499 size
, (unsigned long long) blk_rq_pos(rq
) * SECTOR_SIZE
);
1501 num_segs
= rbd_get_num_segments(&rbd_dev
->header
, ofs
, size
);
1502 if (num_segs
<= 0) {
1503 spin_lock_irq(q
->queue_lock
);
1504 __blk_end_request_all(rq
, num_segs
);
1505 ceph_put_snap_context(snapc
);
1508 coll
= rbd_alloc_coll(num_segs
);
1510 spin_lock_irq(q
->queue_lock
);
1511 __blk_end_request_all(rq
, -ENOMEM
);
1512 ceph_put_snap_context(snapc
);
1517 /* a bio clone to be passed down to OSD req */
1518 dout("rq->bio->bi_vcnt=%hu\n", rq
->bio
->bi_vcnt
);
1519 op_size
= rbd_segment_length(rbd_dev
, ofs
, size
);
1520 kref_get(&coll
->kref
);
1521 bio
= bio_chain_clone(&rq_bio
, &next_bio
, &bp
,
1522 op_size
, GFP_ATOMIC
);
1524 rbd_coll_end_req_index(rq
, coll
, cur_seg
,
1530 /* init OSD command: write or read */
1532 rbd_req_write(rq
, rbd_dev
,
1538 rbd_req_read(rq
, rbd_dev
,
1539 rbd_dev
->mapping
.snap_id
,
1551 kref_put(&coll
->kref
, rbd_coll_release
);
1554 bio_pair_release(bp
);
1555 spin_lock_irq(q
->queue_lock
);
1557 ceph_put_snap_context(snapc
);
1562 * a queue callback. Makes sure that we don't create a bio that spans across
1563 * multiple osd objects. One exception would be with a single page bios,
1564 * which we handle later at bio_chain_clone
1566 static int rbd_merge_bvec(struct request_queue
*q
, struct bvec_merge_data
*bmd
,
1567 struct bio_vec
*bvec
)
1569 struct rbd_device
*rbd_dev
= q
->queuedata
;
1570 unsigned int chunk_sectors
;
1572 unsigned int bio_sectors
;
1575 chunk_sectors
= 1 << (rbd_dev
->header
.obj_order
- SECTOR_SHIFT
);
1576 sector
= bmd
->bi_sector
+ get_start_sect(bmd
->bi_bdev
);
1577 bio_sectors
= bmd
->bi_size
>> SECTOR_SHIFT
;
1579 max
= (chunk_sectors
- ((sector
& (chunk_sectors
- 1))
1580 + bio_sectors
)) << SECTOR_SHIFT
;
1582 max
= 0; /* bio_add cannot handle a negative return */
1583 if (max
<= bvec
->bv_len
&& bio_sectors
== 0)
1584 return bvec
->bv_len
;
1588 static void rbd_free_disk(struct rbd_device
*rbd_dev
)
1590 struct gendisk
*disk
= rbd_dev
->disk
;
1595 if (disk
->flags
& GENHD_FL_UP
)
1598 blk_cleanup_queue(disk
->queue
);
1603 * Read the complete header for the given rbd device.
1605 * Returns a pointer to a dynamically-allocated buffer containing
1606 * the complete and validated header. Caller can pass the address
1607 * of a variable that will be filled in with the version of the
1608 * header object at the time it was read.
1610 * Returns a pointer-coded errno if a failure occurs.
1612 static struct rbd_image_header_ondisk
*
1613 rbd_dev_v1_header_read(struct rbd_device
*rbd_dev
, u64
*version
)
1615 struct rbd_image_header_ondisk
*ondisk
= NULL
;
1622 * The complete header will include an array of its 64-bit
1623 * snapshot ids, followed by the names of those snapshots as
1624 * a contiguous block of NUL-terminated strings. Note that
1625 * the number of snapshots could change by the time we read
1626 * it in, in which case we re-read it.
1633 size
= sizeof (*ondisk
);
1634 size
+= snap_count
* sizeof (struct rbd_image_snap_ondisk
);
1636 ondisk
= kmalloc(size
, GFP_KERNEL
);
1638 return ERR_PTR(-ENOMEM
);
1640 ret
= rbd_req_sync_read(rbd_dev
, CEPH_NOSNAP
,
1641 rbd_dev
->header_name
,
1643 (char *) ondisk
, version
);
1647 if (WARN_ON((size_t) ret
< size
)) {
1649 pr_warning("short header read for image %s"
1650 " (want %zd got %d)\n",
1651 rbd_dev
->image_name
, size
, ret
);
1654 if (!rbd_dev_ondisk_valid(ondisk
)) {
1656 pr_warning("invalid header for image %s\n",
1657 rbd_dev
->image_name
);
1661 names_size
= le64_to_cpu(ondisk
->snap_names_len
);
1662 want_count
= snap_count
;
1663 snap_count
= le32_to_cpu(ondisk
->snap_count
);
1664 } while (snap_count
!= want_count
);
1671 return ERR_PTR(ret
);
1675 * reload the ondisk the header
1677 static int rbd_read_header(struct rbd_device
*rbd_dev
,
1678 struct rbd_image_header
*header
)
1680 struct rbd_image_header_ondisk
*ondisk
;
1684 ondisk
= rbd_dev_v1_header_read(rbd_dev
, &ver
);
1686 return PTR_ERR(ondisk
);
1687 ret
= rbd_header_from_disk(header
, ondisk
);
1689 header
->obj_version
= ver
;
1695 static void __rbd_remove_all_snaps(struct rbd_device
*rbd_dev
)
1697 struct rbd_snap
*snap
;
1698 struct rbd_snap
*next
;
1700 list_for_each_entry_safe(snap
, next
, &rbd_dev
->snaps
, node
)
1701 __rbd_remove_snap_dev(snap
);
1705 * only read the first part of the ondisk header, without the snaps info
1707 static int __rbd_refresh_header(struct rbd_device
*rbd_dev
, u64
*hver
)
1710 struct rbd_image_header h
;
1712 ret
= rbd_read_header(rbd_dev
, &h
);
1716 down_write(&rbd_dev
->header_rwsem
);
1719 if (rbd_dev
->mapping
.snap_id
== CEPH_NOSNAP
) {
1720 sector_t size
= (sector_t
) h
.image_size
/ SECTOR_SIZE
;
1722 if (size
!= (sector_t
) rbd_dev
->mapping
.size
) {
1723 dout("setting size to %llu sectors",
1724 (unsigned long long) size
);
1725 rbd_dev
->mapping
.size
= (u64
) size
;
1726 set_capacity(rbd_dev
->disk
, size
);
1730 /* rbd_dev->header.object_prefix shouldn't change */
1731 kfree(rbd_dev
->header
.snap_sizes
);
1732 kfree(rbd_dev
->header
.snap_names
);
1733 /* osd requests may still refer to snapc */
1734 ceph_put_snap_context(rbd_dev
->header
.snapc
);
1737 *hver
= h
.obj_version
;
1738 rbd_dev
->header
.obj_version
= h
.obj_version
;
1739 rbd_dev
->header
.image_size
= h
.image_size
;
1740 rbd_dev
->header
.snapc
= h
.snapc
;
1741 rbd_dev
->header
.snap_names
= h
.snap_names
;
1742 rbd_dev
->header
.snap_sizes
= h
.snap_sizes
;
1743 /* Free the extra copy of the object prefix */
1744 WARN_ON(strcmp(rbd_dev
->header
.object_prefix
, h
.object_prefix
));
1745 kfree(h
.object_prefix
);
1747 ret
= rbd_dev_snaps_update(rbd_dev
);
1749 ret
= rbd_dev_snaps_register(rbd_dev
);
1751 up_write(&rbd_dev
->header_rwsem
);
1756 static int rbd_refresh_header(struct rbd_device
*rbd_dev
, u64
*hver
)
1760 mutex_lock_nested(&ctl_mutex
, SINGLE_DEPTH_NESTING
);
1761 ret
= __rbd_refresh_header(rbd_dev
, hver
);
1762 mutex_unlock(&ctl_mutex
);
1767 static int rbd_init_disk(struct rbd_device
*rbd_dev
)
1769 struct gendisk
*disk
;
1770 struct request_queue
*q
;
1773 /* create gendisk info */
1774 disk
= alloc_disk(RBD_MINORS_PER_MAJOR
);
1778 snprintf(disk
->disk_name
, sizeof(disk
->disk_name
), RBD_DRV_NAME
"%d",
1780 disk
->major
= rbd_dev
->major
;
1781 disk
->first_minor
= 0;
1782 disk
->fops
= &rbd_bd_ops
;
1783 disk
->private_data
= rbd_dev
;
1786 q
= blk_init_queue(rbd_rq_fn
, &rbd_dev
->lock
);
1790 /* We use the default size, but let's be explicit about it. */
1791 blk_queue_physical_block_size(q
, SECTOR_SIZE
);
1793 /* set io sizes to object size */
1794 segment_size
= rbd_obj_bytes(&rbd_dev
->header
);
1795 blk_queue_max_hw_sectors(q
, segment_size
/ SECTOR_SIZE
);
1796 blk_queue_max_segment_size(q
, segment_size
);
1797 blk_queue_io_min(q
, segment_size
);
1798 blk_queue_io_opt(q
, segment_size
);
1800 blk_queue_merge_bvec(q
, rbd_merge_bvec
);
1803 q
->queuedata
= rbd_dev
;
1805 rbd_dev
->disk
= disk
;
1807 set_capacity(rbd_dev
->disk
, rbd_dev
->mapping
.size
/ SECTOR_SIZE
);
1820 static struct rbd_device
*dev_to_rbd_dev(struct device
*dev
)
1822 return container_of(dev
, struct rbd_device
, dev
);
1825 static ssize_t
rbd_size_show(struct device
*dev
,
1826 struct device_attribute
*attr
, char *buf
)
1828 struct rbd_device
*rbd_dev
= dev_to_rbd_dev(dev
);
1831 down_read(&rbd_dev
->header_rwsem
);
1832 size
= get_capacity(rbd_dev
->disk
);
1833 up_read(&rbd_dev
->header_rwsem
);
1835 return sprintf(buf
, "%llu\n", (unsigned long long) size
* SECTOR_SIZE
);
1838 static ssize_t
rbd_major_show(struct device
*dev
,
1839 struct device_attribute
*attr
, char *buf
)
1841 struct rbd_device
*rbd_dev
= dev_to_rbd_dev(dev
);
1843 return sprintf(buf
, "%d\n", rbd_dev
->major
);
1846 static ssize_t
rbd_client_id_show(struct device
*dev
,
1847 struct device_attribute
*attr
, char *buf
)
1849 struct rbd_device
*rbd_dev
= dev_to_rbd_dev(dev
);
1851 return sprintf(buf
, "client%lld\n",
1852 ceph_client_id(rbd_dev
->rbd_client
->client
));
1855 static ssize_t
rbd_pool_show(struct device
*dev
,
1856 struct device_attribute
*attr
, char *buf
)
1858 struct rbd_device
*rbd_dev
= dev_to_rbd_dev(dev
);
1860 return sprintf(buf
, "%s\n", rbd_dev
->pool_name
);
1863 static ssize_t
rbd_pool_id_show(struct device
*dev
,
1864 struct device_attribute
*attr
, char *buf
)
1866 struct rbd_device
*rbd_dev
= dev_to_rbd_dev(dev
);
1868 return sprintf(buf
, "%d\n", rbd_dev
->pool_id
);
1871 static ssize_t
rbd_name_show(struct device
*dev
,
1872 struct device_attribute
*attr
, char *buf
)
1874 struct rbd_device
*rbd_dev
= dev_to_rbd_dev(dev
);
1876 return sprintf(buf
, "%s\n", rbd_dev
->image_name
);
1879 static ssize_t
rbd_image_id_show(struct device
*dev
,
1880 struct device_attribute
*attr
, char *buf
)
1882 struct rbd_device
*rbd_dev
= dev_to_rbd_dev(dev
);
1884 return sprintf(buf
, "%s\n", rbd_dev
->image_id
);
1887 static ssize_t
rbd_snap_show(struct device
*dev
,
1888 struct device_attribute
*attr
,
1891 struct rbd_device
*rbd_dev
= dev_to_rbd_dev(dev
);
1893 return sprintf(buf
, "%s\n", rbd_dev
->mapping
.snap_name
);
1896 static ssize_t
rbd_image_refresh(struct device
*dev
,
1897 struct device_attribute
*attr
,
1901 struct rbd_device
*rbd_dev
= dev_to_rbd_dev(dev
);
1904 ret
= rbd_refresh_header(rbd_dev
, NULL
);
1906 return ret
< 0 ? ret
: size
;
1909 static DEVICE_ATTR(size
, S_IRUGO
, rbd_size_show
, NULL
);
1910 static DEVICE_ATTR(major
, S_IRUGO
, rbd_major_show
, NULL
);
1911 static DEVICE_ATTR(client_id
, S_IRUGO
, rbd_client_id_show
, NULL
);
1912 static DEVICE_ATTR(pool
, S_IRUGO
, rbd_pool_show
, NULL
);
1913 static DEVICE_ATTR(pool_id
, S_IRUGO
, rbd_pool_id_show
, NULL
);
1914 static DEVICE_ATTR(name
, S_IRUGO
, rbd_name_show
, NULL
);
1915 static DEVICE_ATTR(image_id
, S_IRUGO
, rbd_image_id_show
, NULL
);
1916 static DEVICE_ATTR(refresh
, S_IWUSR
, NULL
, rbd_image_refresh
);
1917 static DEVICE_ATTR(current_snap
, S_IRUGO
, rbd_snap_show
, NULL
);
1919 static struct attribute
*rbd_attrs
[] = {
1920 &dev_attr_size
.attr
,
1921 &dev_attr_major
.attr
,
1922 &dev_attr_client_id
.attr
,
1923 &dev_attr_pool
.attr
,
1924 &dev_attr_pool_id
.attr
,
1925 &dev_attr_name
.attr
,
1926 &dev_attr_image_id
.attr
,
1927 &dev_attr_current_snap
.attr
,
1928 &dev_attr_refresh
.attr
,
1932 static struct attribute_group rbd_attr_group
= {
1936 static const struct attribute_group
*rbd_attr_groups
[] = {
1941 static void rbd_sysfs_dev_release(struct device
*dev
)
1945 static struct device_type rbd_device_type
= {
1947 .groups
= rbd_attr_groups
,
1948 .release
= rbd_sysfs_dev_release
,
1956 static ssize_t
rbd_snap_size_show(struct device
*dev
,
1957 struct device_attribute
*attr
,
1960 struct rbd_snap
*snap
= container_of(dev
, struct rbd_snap
, dev
);
1962 return sprintf(buf
, "%llu\n", (unsigned long long)snap
->size
);
1965 static ssize_t
rbd_snap_id_show(struct device
*dev
,
1966 struct device_attribute
*attr
,
1969 struct rbd_snap
*snap
= container_of(dev
, struct rbd_snap
, dev
);
1971 return sprintf(buf
, "%llu\n", (unsigned long long)snap
->id
);
1974 static DEVICE_ATTR(snap_size
, S_IRUGO
, rbd_snap_size_show
, NULL
);
1975 static DEVICE_ATTR(snap_id
, S_IRUGO
, rbd_snap_id_show
, NULL
);
1977 static struct attribute
*rbd_snap_attrs
[] = {
1978 &dev_attr_snap_size
.attr
,
1979 &dev_attr_snap_id
.attr
,
1983 static struct attribute_group rbd_snap_attr_group
= {
1984 .attrs
= rbd_snap_attrs
,
1987 static void rbd_snap_dev_release(struct device
*dev
)
1989 struct rbd_snap
*snap
= container_of(dev
, struct rbd_snap
, dev
);
1994 static const struct attribute_group
*rbd_snap_attr_groups
[] = {
1995 &rbd_snap_attr_group
,
1999 static struct device_type rbd_snap_device_type
= {
2000 .groups
= rbd_snap_attr_groups
,
2001 .release
= rbd_snap_dev_release
,
2004 static bool rbd_snap_registered(struct rbd_snap
*snap
)
2006 bool ret
= snap
->dev
.type
== &rbd_snap_device_type
;
2007 bool reg
= device_is_registered(&snap
->dev
);
2009 rbd_assert(!ret
^ reg
);
2014 static void __rbd_remove_snap_dev(struct rbd_snap
*snap
)
2016 list_del(&snap
->node
);
2017 if (device_is_registered(&snap
->dev
))
2018 device_unregister(&snap
->dev
);
2021 static int rbd_register_snap_dev(struct rbd_snap
*snap
,
2022 struct device
*parent
)
2024 struct device
*dev
= &snap
->dev
;
2027 dev
->type
= &rbd_snap_device_type
;
2028 dev
->parent
= parent
;
2029 dev
->release
= rbd_snap_dev_release
;
2030 dev_set_name(dev
, "snap_%s", snap
->name
);
2031 dout("%s: registering device for snapshot %s\n", __func__
, snap
->name
);
2033 ret
= device_register(dev
);
2038 static struct rbd_snap
*__rbd_add_snap_dev(struct rbd_device
*rbd_dev
,
2039 int i
, const char *name
)
2041 struct rbd_snap
*snap
;
2044 snap
= kzalloc(sizeof (*snap
), GFP_KERNEL
);
2046 return ERR_PTR(-ENOMEM
);
2049 snap
->name
= kstrdup(name
, GFP_KERNEL
);
2053 snap
->size
= rbd_dev
->header
.snap_sizes
[i
];
2054 snap
->id
= rbd_dev
->header
.snapc
->snaps
[i
];
2062 return ERR_PTR(ret
);
2066 * Scan the rbd device's current snapshot list and compare it to the
2067 * newly-received snapshot context. Remove any existing snapshots
2068 * not present in the new snapshot context. Add a new snapshot for
2069 * any snaphots in the snapshot context not in the current list.
2070 * And verify there are no changes to snapshots we already know
2073 * Assumes the snapshots in the snapshot context are sorted by
2074 * snapshot id, highest id first. (Snapshots in the rbd_dev's list
2075 * are also maintained in that order.)
2077 static int rbd_dev_snaps_update(struct rbd_device
*rbd_dev
)
2079 struct ceph_snap_context
*snapc
= rbd_dev
->header
.snapc
;
2080 const u32 snap_count
= snapc
->num_snaps
;
2081 char *snap_name
= rbd_dev
->header
.snap_names
;
2082 struct list_head
*head
= &rbd_dev
->snaps
;
2083 struct list_head
*links
= head
->next
;
2086 dout("%s: snap count is %u\n", __func__
, (unsigned int) snap_count
);
2087 while (index
< snap_count
|| links
!= head
) {
2089 struct rbd_snap
*snap
;
2091 snap_id
= index
< snap_count
? snapc
->snaps
[index
]
2093 snap
= links
!= head
? list_entry(links
, struct rbd_snap
, node
)
2095 rbd_assert(!snap
|| snap
->id
!= CEPH_NOSNAP
);
2097 if (snap_id
== CEPH_NOSNAP
|| (snap
&& snap
->id
> snap_id
)) {
2098 struct list_head
*next
= links
->next
;
2100 /* Existing snapshot not in the new snap context */
2102 if (rbd_dev
->mapping
.snap_id
== snap
->id
)
2103 rbd_dev
->mapping
.snap_exists
= false;
2104 __rbd_remove_snap_dev(snap
);
2105 dout("%ssnap id %llu has been removed\n",
2106 rbd_dev
->mapping
.snap_id
== snap
->id
?
2108 (unsigned long long) snap
->id
);
2110 /* Done with this list entry; advance */
2116 dout("entry %u: snap_id = %llu\n", (unsigned int) snap_count
,
2117 (unsigned long long) snap_id
);
2118 if (!snap
|| (snap_id
!= CEPH_NOSNAP
&& snap
->id
< snap_id
)) {
2119 struct rbd_snap
*new_snap
;
2121 /* We haven't seen this snapshot before */
2123 new_snap
= __rbd_add_snap_dev(rbd_dev
, index
,
2125 if (IS_ERR(new_snap
)) {
2126 int err
= PTR_ERR(new_snap
);
2128 dout(" failed to add dev, error %d\n", err
);
2133 /* New goes before existing, or at end of list */
2135 dout(" added dev%s\n", snap
? "" : " at end\n");
2137 list_add_tail(&new_snap
->node
, &snap
->node
);
2139 list_add_tail(&new_snap
->node
, head
);
2141 /* Already have this one */
2143 dout(" already present\n");
2145 rbd_assert(snap
->size
==
2146 rbd_dev
->header
.snap_sizes
[index
]);
2147 rbd_assert(!strcmp(snap
->name
, snap_name
));
2149 /* Done with this list entry; advance */
2151 links
= links
->next
;
2154 /* Advance to the next entry in the snapshot context */
2157 snap_name
+= strlen(snap_name
) + 1;
2159 dout("%s: done\n", __func__
);
2165 * Scan the list of snapshots and register the devices for any that
2166 * have not already been registered.
2168 static int rbd_dev_snaps_register(struct rbd_device
*rbd_dev
)
2170 struct rbd_snap
*snap
;
2173 dout("%s called\n", __func__
);
2174 if (WARN_ON(!device_is_registered(&rbd_dev
->dev
)))
2177 list_for_each_entry(snap
, &rbd_dev
->snaps
, node
) {
2178 if (!rbd_snap_registered(snap
)) {
2179 ret
= rbd_register_snap_dev(snap
, &rbd_dev
->dev
);
2184 dout("%s: returning %d\n", __func__
, ret
);
2189 static int rbd_bus_add_dev(struct rbd_device
*rbd_dev
)
2194 mutex_lock_nested(&ctl_mutex
, SINGLE_DEPTH_NESTING
);
2196 dev
= &rbd_dev
->dev
;
2197 dev
->bus
= &rbd_bus_type
;
2198 dev
->type
= &rbd_device_type
;
2199 dev
->parent
= &rbd_root_dev
;
2200 dev
->release
= rbd_dev_release
;
2201 dev_set_name(dev
, "%d", rbd_dev
->dev_id
);
2202 ret
= device_register(dev
);
2204 mutex_unlock(&ctl_mutex
);
2209 static void rbd_bus_del_dev(struct rbd_device
*rbd_dev
)
2211 device_unregister(&rbd_dev
->dev
);
2214 static int rbd_init_watch_dev(struct rbd_device
*rbd_dev
)
2219 ret
= rbd_req_sync_watch(rbd_dev
);
2220 if (ret
== -ERANGE
) {
2221 rc
= rbd_refresh_header(rbd_dev
, NULL
);
2225 } while (ret
== -ERANGE
);
2230 static atomic64_t rbd_dev_id_max
= ATOMIC64_INIT(0);
2233 * Get a unique rbd identifier for the given new rbd_dev, and add
2234 * the rbd_dev to the global list. The minimum rbd id is 1.
2236 static void rbd_dev_id_get(struct rbd_device
*rbd_dev
)
2238 rbd_dev
->dev_id
= atomic64_inc_return(&rbd_dev_id_max
);
2240 spin_lock(&rbd_dev_list_lock
);
2241 list_add_tail(&rbd_dev
->node
, &rbd_dev_list
);
2242 spin_unlock(&rbd_dev_list_lock
);
2243 dout("rbd_dev %p given dev id %llu\n", rbd_dev
,
2244 (unsigned long long) rbd_dev
->dev_id
);
2248 * Remove an rbd_dev from the global list, and record that its
2249 * identifier is no longer in use.
2251 static void rbd_dev_id_put(struct rbd_device
*rbd_dev
)
2253 struct list_head
*tmp
;
2254 int rbd_id
= rbd_dev
->dev_id
;
2257 rbd_assert(rbd_id
> 0);
2259 dout("rbd_dev %p released dev id %llu\n", rbd_dev
,
2260 (unsigned long long) rbd_dev
->dev_id
);
2261 spin_lock(&rbd_dev_list_lock
);
2262 list_del_init(&rbd_dev
->node
);
2265 * If the id being "put" is not the current maximum, there
2266 * is nothing special we need to do.
2268 if (rbd_id
!= atomic64_read(&rbd_dev_id_max
)) {
2269 spin_unlock(&rbd_dev_list_lock
);
2274 * We need to update the current maximum id. Search the
2275 * list to find out what it is. We're more likely to find
2276 * the maximum at the end, so search the list backward.
2279 list_for_each_prev(tmp
, &rbd_dev_list
) {
2280 struct rbd_device
*rbd_dev
;
2282 rbd_dev
= list_entry(tmp
, struct rbd_device
, node
);
2283 if (rbd_id
> max_id
)
2286 spin_unlock(&rbd_dev_list_lock
);
2289 * The max id could have been updated by rbd_dev_id_get(), in
2290 * which case it now accurately reflects the new maximum.
2291 * Be careful not to overwrite the maximum value in that
2294 atomic64_cmpxchg(&rbd_dev_id_max
, rbd_id
, max_id
);
2295 dout(" max dev id has been reset\n");
2299 * Skips over white space at *buf, and updates *buf to point to the
2300 * first found non-space character (if any). Returns the length of
2301 * the token (string of non-white space characters) found. Note
2302 * that *buf must be terminated with '\0'.
2304 static inline size_t next_token(const char **buf
)
2307 * These are the characters that produce nonzero for
2308 * isspace() in the "C" and "POSIX" locales.
2310 const char *spaces
= " \f\n\r\t\v";
2312 *buf
+= strspn(*buf
, spaces
); /* Find start of token */
2314 return strcspn(*buf
, spaces
); /* Return token length */
2318 * Finds the next token in *buf, and if the provided token buffer is
2319 * big enough, copies the found token into it. The result, if
2320 * copied, is guaranteed to be terminated with '\0'. Note that *buf
2321 * must be terminated with '\0' on entry.
2323 * Returns the length of the token found (not including the '\0').
2324 * Return value will be 0 if no token is found, and it will be >=
2325 * token_size if the token would not fit.
2327 * The *buf pointer will be updated to point beyond the end of the
2328 * found token. Note that this occurs even if the token buffer is
2329 * too small to hold it.
2331 static inline size_t copy_token(const char **buf
,
2337 len
= next_token(buf
);
2338 if (len
< token_size
) {
2339 memcpy(token
, *buf
, len
);
2340 *(token
+ len
) = '\0';
2348 * Finds the next token in *buf, dynamically allocates a buffer big
2349 * enough to hold a copy of it, and copies the token into the new
2350 * buffer. The copy is guaranteed to be terminated with '\0'. Note
2351 * that a duplicate buffer is created even for a zero-length token.
2353 * Returns a pointer to the newly-allocated duplicate, or a null
2354 * pointer if memory for the duplicate was not available. If
2355 * the lenp argument is a non-null pointer, the length of the token
2356 * (not including the '\0') is returned in *lenp.
2358 * If successful, the *buf pointer will be updated to point beyond
2359 * the end of the found token.
2361 * Note: uses GFP_KERNEL for allocation.
2363 static inline char *dup_token(const char **buf
, size_t *lenp
)
2368 len
= next_token(buf
);
2369 dup
= kmalloc(len
+ 1, GFP_KERNEL
);
2373 memcpy(dup
, *buf
, len
);
2374 *(dup
+ len
) = '\0';
2384 * This fills in the pool_name, image_name, image_name_len, rbd_dev,
2385 * rbd_md_name, and name fields of the given rbd_dev, based on the
2386 * list of monitor addresses and other options provided via
2387 * /sys/bus/rbd/add. Returns a pointer to a dynamically-allocated
2388 * copy of the snapshot name to map if successful, or a
2389 * pointer-coded error otherwise.
2391 * Note: rbd_dev is assumed to have been initially zero-filled.
2393 static char *rbd_add_parse_args(struct rbd_device
*rbd_dev
,
2395 const char **mon_addrs
,
2396 size_t *mon_addrs_size
,
2398 size_t options_size
)
2401 char *err_ptr
= ERR_PTR(-EINVAL
);
2404 /* The first four tokens are required */
2406 len
= next_token(&buf
);
2409 *mon_addrs_size
= len
+ 1;
2414 len
= copy_token(&buf
, options
, options_size
);
2415 if (!len
|| len
>= options_size
)
2418 err_ptr
= ERR_PTR(-ENOMEM
);
2419 rbd_dev
->pool_name
= dup_token(&buf
, NULL
);
2420 if (!rbd_dev
->pool_name
)
2423 rbd_dev
->image_name
= dup_token(&buf
, &rbd_dev
->image_name_len
);
2424 if (!rbd_dev
->image_name
)
2427 /* Snapshot name is optional */
2428 len
= next_token(&buf
);
2430 buf
= RBD_SNAP_HEAD_NAME
; /* No snapshot supplied */
2431 len
= sizeof (RBD_SNAP_HEAD_NAME
) - 1;
2433 snap_name
= kmalloc(len
+ 1, GFP_KERNEL
);
2436 memcpy(snap_name
, buf
, len
);
2437 *(snap_name
+ len
) = '\0';
2439 dout(" SNAP_NAME is <%s>, len is %zd\n", snap_name
, len
);
2444 kfree(rbd_dev
->image_name
);
2445 rbd_dev
->image_name
= NULL
;
2446 rbd_dev
->image_name_len
= 0;
2447 kfree(rbd_dev
->pool_name
);
2448 rbd_dev
->pool_name
= NULL
;
2454 * An rbd format 2 image has a unique identifier, distinct from the
2455 * name given to it by the user. Internally, that identifier is
2456 * what's used to specify the names of objects related to the image.
2458 * A special "rbd id" object is used to map an rbd image name to its
2459 * id. If that object doesn't exist, then there is no v2 rbd image
2460 * with the supplied name.
2462 * This function will record the given rbd_dev's image_id field if
2463 * it can be determined, and in that case will return 0. If any
2464 * errors occur a negative errno will be returned and the rbd_dev's
2465 * image_id field will be unchanged (and should be NULL).
2467 static int rbd_dev_image_id(struct rbd_device
*rbd_dev
)
2476 * First, see if the format 2 image id file exists, and if
2477 * so, get the image's persistent id from it.
2479 size
= sizeof (RBD_ID_PREFIX
) + rbd_dev
->image_name_len
;
2480 object_name
= kmalloc(size
, GFP_NOIO
);
2483 sprintf(object_name
, "%s%s", RBD_ID_PREFIX
, rbd_dev
->image_name
);
2484 dout("rbd id object name is %s\n", object_name
);
2486 /* Response will be an encoded string, which includes a length */
2488 size
= sizeof (__le32
) + RBD_IMAGE_ID_LEN_MAX
;
2489 response
= kzalloc(size
, GFP_NOIO
);
2495 ret
= rbd_req_sync_exec(rbd_dev
, object_name
,
2498 response
, RBD_IMAGE_ID_LEN_MAX
,
2499 CEPH_OSD_FLAG_READ
, NULL
);
2500 dout("%s: rbd_req_sync_exec returned %d\n", __func__
, ret
);
2505 rbd_dev
->image_id
= ceph_extract_encoded_string(&p
,
2506 p
+ RBD_IMAGE_ID_LEN_MAX
,
2507 &rbd_dev
->image_id_len
,
2509 if (IS_ERR(rbd_dev
->image_id
)) {
2510 ret
= PTR_ERR(rbd_dev
->image_id
);
2511 rbd_dev
->image_id
= NULL
;
2513 dout("image_id is %s\n", rbd_dev
->image_id
);
2522 static ssize_t
rbd_add(struct bus_type
*bus
,
2527 struct rbd_device
*rbd_dev
= NULL
;
2528 const char *mon_addrs
= NULL
;
2529 size_t mon_addrs_size
= 0;
2530 struct ceph_osd_client
*osdc
;
2534 if (!try_module_get(THIS_MODULE
))
2537 options
= kmalloc(count
, GFP_KERNEL
);
2540 rbd_dev
= kzalloc(sizeof(*rbd_dev
), GFP_KERNEL
);
2544 /* static rbd_device initialization */
2545 spin_lock_init(&rbd_dev
->lock
);
2546 INIT_LIST_HEAD(&rbd_dev
->node
);
2547 INIT_LIST_HEAD(&rbd_dev
->snaps
);
2548 init_rwsem(&rbd_dev
->header_rwsem
);
2550 /* parse add command */
2551 snap_name
= rbd_add_parse_args(rbd_dev
, buf
,
2552 &mon_addrs
, &mon_addrs_size
, options
, count
);
2553 if (IS_ERR(snap_name
)) {
2554 rc
= PTR_ERR(snap_name
);
2558 rc
= rbd_get_client(rbd_dev
, mon_addrs
, mon_addrs_size
- 1, options
);
2563 osdc
= &rbd_dev
->rbd_client
->client
->osdc
;
2564 rc
= ceph_pg_poolid_by_name(osdc
->osdmap
, rbd_dev
->pool_name
);
2566 goto err_out_client
;
2567 rbd_dev
->pool_id
= rc
;
2569 rc
= rbd_dev_image_id(rbd_dev
);
2571 rc
= -ENOTSUPP
; /* Not actually supporting format 2 yet */
2572 goto err_out_client
;
2575 /* Version 1 images have no id; empty string is used */
2577 rbd_dev
->image_id
= kstrdup("", GFP_KERNEL
);
2578 if (!rbd_dev
->image_id
) {
2580 goto err_out_client
;
2582 rbd_dev
->image_id_len
= 0;
2584 /* Create the name of the header object */
2586 rbd_dev
->header_name
= kmalloc(rbd_dev
->image_name_len
2587 + sizeof (RBD_SUFFIX
),
2589 if (!rbd_dev
->header_name
)
2590 goto err_out_client
;
2591 sprintf(rbd_dev
->header_name
, "%s%s", rbd_dev
->image_name
, RBD_SUFFIX
);
2593 /* Get information about the image being mapped */
2595 rc
= rbd_read_header(rbd_dev
, &rbd_dev
->header
);
2597 goto err_out_client
;
2599 /* no need to lock here, as rbd_dev is not registered yet */
2600 rc
= rbd_dev_snaps_update(rbd_dev
);
2602 goto err_out_header
;
2604 rc
= rbd_dev_set_mapping(rbd_dev
, snap_name
);
2606 goto err_out_header
;
2608 /* generate unique id: find highest unique id, add one */
2609 rbd_dev_id_get(rbd_dev
);
2611 /* Fill in the device name, now that we have its id. */
2612 BUILD_BUG_ON(DEV_NAME_LEN
2613 < sizeof (RBD_DRV_NAME
) + MAX_INT_FORMAT_WIDTH
);
2614 sprintf(rbd_dev
->name
, "%s%d", RBD_DRV_NAME
, rbd_dev
->dev_id
);
2616 /* Get our block major device number. */
2618 rc
= register_blkdev(0, rbd_dev
->name
);
2621 rbd_dev
->major
= rc
;
2623 /* Set up the blkdev mapping. */
2625 rc
= rbd_init_disk(rbd_dev
);
2627 goto err_out_blkdev
;
2629 rc
= rbd_bus_add_dev(rbd_dev
);
2634 * At this point cleanup in the event of an error is the job
2635 * of the sysfs code (initiated by rbd_bus_del_dev()).
2638 down_write(&rbd_dev
->header_rwsem
);
2639 rc
= rbd_dev_snaps_register(rbd_dev
);
2640 up_write(&rbd_dev
->header_rwsem
);
2644 rc
= rbd_init_watch_dev(rbd_dev
);
2648 /* Everything's ready. Announce the disk to the world. */
2650 add_disk(rbd_dev
->disk
);
2652 pr_info("%s: added with size 0x%llx\n", rbd_dev
->disk
->disk_name
,
2653 (unsigned long long) rbd_dev
->mapping
.size
);
2658 /* this will also clean up rest of rbd_dev stuff */
2660 rbd_bus_del_dev(rbd_dev
);
2665 rbd_free_disk(rbd_dev
);
2667 unregister_blkdev(rbd_dev
->major
, rbd_dev
->name
);
2669 rbd_dev_id_put(rbd_dev
);
2671 rbd_header_free(&rbd_dev
->header
);
2673 kfree(rbd_dev
->header_name
);
2674 rbd_put_client(rbd_dev
);
2675 kfree(rbd_dev
->image_id
);
2677 kfree(rbd_dev
->mapping
.snap_name
);
2678 kfree(rbd_dev
->image_name
);
2679 kfree(rbd_dev
->pool_name
);
2684 dout("Error adding device %s\n", buf
);
2685 module_put(THIS_MODULE
);
2687 return (ssize_t
) rc
;
2690 static struct rbd_device
*__rbd_get_dev(unsigned long dev_id
)
2692 struct list_head
*tmp
;
2693 struct rbd_device
*rbd_dev
;
2695 spin_lock(&rbd_dev_list_lock
);
2696 list_for_each(tmp
, &rbd_dev_list
) {
2697 rbd_dev
= list_entry(tmp
, struct rbd_device
, node
);
2698 if (rbd_dev
->dev_id
== dev_id
) {
2699 spin_unlock(&rbd_dev_list_lock
);
2703 spin_unlock(&rbd_dev_list_lock
);
2707 static void rbd_dev_release(struct device
*dev
)
2709 struct rbd_device
*rbd_dev
= dev_to_rbd_dev(dev
);
2711 if (rbd_dev
->watch_request
) {
2712 struct ceph_client
*client
= rbd_dev
->rbd_client
->client
;
2714 ceph_osdc_unregister_linger_request(&client
->osdc
,
2715 rbd_dev
->watch_request
);
2717 if (rbd_dev
->watch_event
)
2718 rbd_req_sync_unwatch(rbd_dev
);
2720 rbd_put_client(rbd_dev
);
2722 /* clean up and free blkdev */
2723 rbd_free_disk(rbd_dev
);
2724 unregister_blkdev(rbd_dev
->major
, rbd_dev
->name
);
2726 /* release allocated disk header fields */
2727 rbd_header_free(&rbd_dev
->header
);
2729 /* done with the id, and with the rbd_dev */
2730 kfree(rbd_dev
->mapping
.snap_name
);
2731 kfree(rbd_dev
->image_id
);
2732 kfree(rbd_dev
->header_name
);
2733 kfree(rbd_dev
->pool_name
);
2734 kfree(rbd_dev
->image_name
);
2735 rbd_dev_id_put(rbd_dev
);
2738 /* release module ref */
2739 module_put(THIS_MODULE
);
2742 static ssize_t
rbd_remove(struct bus_type
*bus
,
2746 struct rbd_device
*rbd_dev
= NULL
;
2751 rc
= strict_strtoul(buf
, 10, &ul
);
2755 /* convert to int; abort if we lost anything in the conversion */
2756 target_id
= (int) ul
;
2757 if (target_id
!= ul
)
2760 mutex_lock_nested(&ctl_mutex
, SINGLE_DEPTH_NESTING
);
2762 rbd_dev
= __rbd_get_dev(target_id
);
2768 __rbd_remove_all_snaps(rbd_dev
);
2769 rbd_bus_del_dev(rbd_dev
);
2772 mutex_unlock(&ctl_mutex
);
2778 * create control files in sysfs
2781 static int rbd_sysfs_init(void)
2785 ret
= device_register(&rbd_root_dev
);
2789 ret
= bus_register(&rbd_bus_type
);
2791 device_unregister(&rbd_root_dev
);
2796 static void rbd_sysfs_cleanup(void)
2798 bus_unregister(&rbd_bus_type
);
2799 device_unregister(&rbd_root_dev
);
2802 int __init
rbd_init(void)
2806 rc
= rbd_sysfs_init();
2809 pr_info("loaded " RBD_DRV_NAME_LONG
"\n");
2813 void __exit
rbd_exit(void)
2815 rbd_sysfs_cleanup();
2818 module_init(rbd_init
);
2819 module_exit(rbd_exit
);
2821 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
2822 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
2823 MODULE_DESCRIPTION("rados block device");
2825 /* following authorship retained from original osdblk.c */
2826 MODULE_AUTHOR("Jeff Garzik <jeff@garzik.org>");
2828 MODULE_LICENSE("GPL");