3 rbd.c -- Export ceph rados objects as a Linux block device
6 based on drivers/block/osdblk.c:
8 Copyright 2009 Red Hat, Inc.
10 This program is free software; you can redistribute it and/or modify
11 it under the terms of the GNU General Public License as published by
12 the Free Software Foundation.
14 This program is distributed in the hope that it will be useful,
15 but WITHOUT ANY WARRANTY; without even the implied warranty of
16 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
17 GNU General Public License for more details.
19 You should have received a copy of the GNU General Public License
20 along with this program; see the file COPYING. If not, write to
21 the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA.
25 For usage instructions, please refer to:
27 Documentation/ABI/testing/sysfs-bus-rbd
31 #include <linux/ceph/libceph.h>
32 #include <linux/ceph/osd_client.h>
33 #include <linux/ceph/mon_client.h>
34 #include <linux/ceph/decode.h>
35 #include <linux/parser.h>
37 #include <linux/kernel.h>
38 #include <linux/device.h>
39 #include <linux/module.h>
41 #include <linux/blkdev.h>
43 #include "rbd_types.h"
45 #define RBD_DEBUG /* Activate rbd_assert() calls */
48 * The basic unit of block I/O is a sector. It is interpreted in a
49 * number of contexts in Linux (blk, bio, genhd), but the default is
50 * universally 512 bytes. These symbols are just slightly more
51 * meaningful than the bare numbers they represent.
53 #define SECTOR_SHIFT 9
54 #define SECTOR_SIZE (1ULL << SECTOR_SHIFT)
56 #define RBD_DRV_NAME "rbd"
57 #define RBD_DRV_NAME_LONG "rbd (rados block device)"
59 #define RBD_MINORS_PER_MAJOR 256 /* max minors per blkdev */
61 #define RBD_SNAP_DEV_NAME_PREFIX "snap_"
62 #define RBD_MAX_SNAP_NAME_LEN \
63 (NAME_MAX - (sizeof (RBD_SNAP_DEV_NAME_PREFIX) - 1))
65 #define RBD_MAX_SNAP_COUNT 510 /* allows max snapc to fit in 4KB */
67 #define RBD_SNAP_HEAD_NAME "-"
69 /* This allows a single page to hold an image name sent by OSD */
70 #define RBD_IMAGE_NAME_LEN_MAX (PAGE_SIZE - sizeof (__le32) - 1)
71 #define RBD_IMAGE_ID_LEN_MAX 64
73 #define RBD_OBJ_PREFIX_LEN_MAX 64
77 #define RBD_FEATURE_LAYERING (1<<0)
78 #define RBD_FEATURE_STRIPINGV2 (1<<1)
79 #define RBD_FEATURES_ALL \
80 (RBD_FEATURE_LAYERING | RBD_FEATURE_STRIPINGV2)
82 /* Features supported by this (client software) implementation. */
84 #define RBD_FEATURES_SUPPORTED (RBD_FEATURES_ALL)
87 * An RBD device name will be "rbd#", where the "rbd" comes from
88 * RBD_DRV_NAME above, and # is a unique integer identifier.
89 * MAX_INT_FORMAT_WIDTH is used in ensuring DEV_NAME_LEN is big
90 * enough to hold all possible device names.
92 #define DEV_NAME_LEN 32
93 #define MAX_INT_FORMAT_WIDTH ((5 * sizeof (int)) / 2 + 1)
96 * block device image metadata (in-memory version)
98 struct rbd_image_header
{
99 /* These four fields never change for a given rbd image */
106 /* The remaining fields need to be updated occasionally */
108 struct ceph_snap_context
*snapc
;
117 * An rbd image specification.
119 * The tuple (pool_id, image_id, snap_id) is sufficient to uniquely
120 * identify an image. Each rbd_dev structure includes a pointer to
121 * an rbd_spec structure that encapsulates this identity.
123 * Each of the id's in an rbd_spec has an associated name. For a
124 * user-mapped image, the names are supplied and the id's associated
125 * with them are looked up. For a layered image, a parent image is
126 * defined by the tuple, and the names are looked up.
128 * An rbd_dev structure contains a parent_spec pointer which is
129 * non-null if the image it represents is a child in a layered
130 * image. This pointer will refer to the rbd_spec structure used
131 * by the parent rbd_dev for its own identity (i.e., the structure
132 * is shared between the parent and child).
134 * Since these structures are populated once, during the discovery
135 * phase of image construction, they are effectively immutable so
136 * we make no effort to synchronize access to them.
138 * Note that code herein does not assume the image name is known (it
139 * could be a null pointer).
143 const char *pool_name
;
145 const char *image_id
;
146 const char *image_name
;
149 const char *snap_name
;
155 * an instance of the client. multiple devices may share an rbd client.
158 struct ceph_client
*client
;
160 struct list_head node
;
163 struct rbd_img_request
;
164 typedef void (*rbd_img_callback_t
)(struct rbd_img_request
*);
166 #define BAD_WHICH U32_MAX /* Good which or bad which, which? */
168 struct rbd_obj_request
;
169 typedef void (*rbd_obj_callback_t
)(struct rbd_obj_request
*);
171 enum obj_request_type
{
172 OBJ_REQUEST_NODATA
, OBJ_REQUEST_BIO
, OBJ_REQUEST_PAGES
176 OBJ_REQ_DONE
, /* completion flag: not done = 0, done = 1 */
177 OBJ_REQ_IMG_DATA
, /* object usage: standalone = 0, image = 1 */
178 OBJ_REQ_KNOWN
, /* EXISTS flag valid: no = 0, yes = 1 */
179 OBJ_REQ_EXISTS
, /* target exists: no = 0, yes = 1 */
182 struct rbd_obj_request
{
183 const char *object_name
;
184 u64 offset
; /* object start byte */
185 u64 length
; /* bytes from offset */
189 * An object request associated with an image will have its
190 * img_data flag set; a standalone object request will not.
192 * A standalone object request will have which == BAD_WHICH
193 * and a null obj_request pointer.
195 * An object request initiated in support of a layered image
196 * object (to check for its existence before a write) will
197 * have which == BAD_WHICH and a non-null obj_request pointer.
199 * Finally, an object request for rbd image data will have
200 * which != BAD_WHICH, and will have a non-null img_request
201 * pointer. The value of which will be in the range
202 * 0..(img_request->obj_request_count-1).
205 struct rbd_obj_request
*obj_request
; /* STAT op */
207 struct rbd_img_request
*img_request
;
209 /* links for img_request->obj_requests list */
210 struct list_head links
;
213 u32 which
; /* posn image request list */
215 enum obj_request_type type
;
217 struct bio
*bio_list
;
223 struct page
**copyup_pages
;
225 struct ceph_osd_request
*osd_req
;
227 u64 xferred
; /* bytes transferred */
230 rbd_obj_callback_t callback
;
231 struct completion completion
;
237 IMG_REQ_WRITE
, /* I/O direction: read = 0, write = 1 */
238 IMG_REQ_CHILD
, /* initiator: block = 0, child image = 1 */
239 IMG_REQ_LAYERED
, /* ENOENT handling: normal = 0, layered = 1 */
242 struct rbd_img_request
{
243 struct rbd_device
*rbd_dev
;
244 u64 offset
; /* starting image byte offset */
245 u64 length
; /* byte count from offset */
248 u64 snap_id
; /* for reads */
249 struct ceph_snap_context
*snapc
; /* for writes */
252 struct request
*rq
; /* block request */
253 struct rbd_obj_request
*obj_request
; /* obj req initiator */
255 struct page
**copyup_pages
;
256 spinlock_t completion_lock
;/* protects next_completion */
258 rbd_img_callback_t callback
;
259 u64 xferred
;/* aggregate bytes transferred */
260 int result
; /* first nonzero obj_request result */
262 u32 obj_request_count
;
263 struct list_head obj_requests
; /* rbd_obj_request structs */
268 #define for_each_obj_request(ireq, oreq) \
269 list_for_each_entry(oreq, &(ireq)->obj_requests, links)
270 #define for_each_obj_request_from(ireq, oreq) \
271 list_for_each_entry_from(oreq, &(ireq)->obj_requests, links)
272 #define for_each_obj_request_safe(ireq, oreq, n) \
273 list_for_each_entry_safe_reverse(oreq, n, &(ireq)->obj_requests, links)
278 struct list_head node
;
293 int dev_id
; /* blkdev unique id */
295 int major
; /* blkdev assigned major */
296 struct gendisk
*disk
; /* blkdev's gendisk and rq */
298 u32 image_format
; /* Either 1 or 2 */
299 struct rbd_client
*rbd_client
;
301 char name
[DEV_NAME_LEN
]; /* blkdev name, e.g. rbd3 */
303 spinlock_t lock
; /* queue, flags, open_count */
305 struct rbd_image_header header
;
306 unsigned long flags
; /* possibly lock protected */
307 struct rbd_spec
*spec
;
311 struct ceph_file_layout layout
;
313 struct ceph_osd_event
*watch_event
;
314 struct rbd_obj_request
*watch_request
;
316 struct rbd_spec
*parent_spec
;
318 struct rbd_device
*parent
;
320 /* protects updating the header */
321 struct rw_semaphore header_rwsem
;
323 struct rbd_mapping mapping
;
325 struct list_head node
;
327 /* list of snapshots */
328 struct list_head snaps
;
332 unsigned long open_count
; /* protected by lock */
336 * Flag bits for rbd_dev->flags. If atomicity is required,
337 * rbd_dev->lock is used to protect access.
339 * Currently, only the "removing" flag (which is coupled with the
340 * "open_count" field) requires atomic access.
343 RBD_DEV_FLAG_EXISTS
, /* mapped snapshot has not been deleted */
344 RBD_DEV_FLAG_REMOVING
, /* this mapping is being removed */
347 static DEFINE_MUTEX(ctl_mutex
); /* Serialize open/close/setup/teardown */
349 static LIST_HEAD(rbd_dev_list
); /* devices */
350 static DEFINE_SPINLOCK(rbd_dev_list_lock
);
352 static LIST_HEAD(rbd_client_list
); /* clients */
353 static DEFINE_SPINLOCK(rbd_client_list_lock
);
355 static int rbd_img_request_submit(struct rbd_img_request
*img_request
);
357 static int rbd_dev_snaps_update(struct rbd_device
*rbd_dev
);
359 static void rbd_dev_device_release(struct device
*dev
);
360 static void rbd_snap_destroy(struct rbd_snap
*snap
);
362 static ssize_t
rbd_add(struct bus_type
*bus
, const char *buf
,
364 static ssize_t
rbd_remove(struct bus_type
*bus
, const char *buf
,
366 static int rbd_dev_image_probe(struct rbd_device
*rbd_dev
);
368 static struct bus_attribute rbd_bus_attrs
[] = {
369 __ATTR(add
, S_IWUSR
, NULL
, rbd_add
),
370 __ATTR(remove
, S_IWUSR
, NULL
, rbd_remove
),
374 static struct bus_type rbd_bus_type
= {
376 .bus_attrs
= rbd_bus_attrs
,
379 static void rbd_root_dev_release(struct device
*dev
)
383 static struct device rbd_root_dev
= {
385 .release
= rbd_root_dev_release
,
388 static __printf(2, 3)
389 void rbd_warn(struct rbd_device
*rbd_dev
, const char *fmt
, ...)
391 struct va_format vaf
;
399 printk(KERN_WARNING
"%s: %pV\n", RBD_DRV_NAME
, &vaf
);
400 else if (rbd_dev
->disk
)
401 printk(KERN_WARNING
"%s: %s: %pV\n",
402 RBD_DRV_NAME
, rbd_dev
->disk
->disk_name
, &vaf
);
403 else if (rbd_dev
->spec
&& rbd_dev
->spec
->image_name
)
404 printk(KERN_WARNING
"%s: image %s: %pV\n",
405 RBD_DRV_NAME
, rbd_dev
->spec
->image_name
, &vaf
);
406 else if (rbd_dev
->spec
&& rbd_dev
->spec
->image_id
)
407 printk(KERN_WARNING
"%s: id %s: %pV\n",
408 RBD_DRV_NAME
, rbd_dev
->spec
->image_id
, &vaf
);
410 printk(KERN_WARNING
"%s: rbd_dev %p: %pV\n",
411 RBD_DRV_NAME
, rbd_dev
, &vaf
);
416 #define rbd_assert(expr) \
417 if (unlikely(!(expr))) { \
418 printk(KERN_ERR "\nAssertion failure in %s() " \
420 "\trbd_assert(%s);\n\n", \
421 __func__, __LINE__, #expr); \
424 #else /* !RBD_DEBUG */
425 # define rbd_assert(expr) ((void) 0)
426 #endif /* !RBD_DEBUG */
428 static int rbd_img_obj_request_submit(struct rbd_obj_request
*obj_request
);
429 static void rbd_img_parent_read(struct rbd_obj_request
*obj_request
);
430 static void rbd_dev_remove_parent(struct rbd_device
*rbd_dev
);
432 static int rbd_dev_refresh(struct rbd_device
*rbd_dev
);
433 static int rbd_dev_v2_refresh(struct rbd_device
*rbd_dev
);
435 static int rbd_open(struct block_device
*bdev
, fmode_t mode
)
437 struct rbd_device
*rbd_dev
= bdev
->bd_disk
->private_data
;
438 bool removing
= false;
440 if ((mode
& FMODE_WRITE
) && rbd_dev
->mapping
.read_only
)
443 spin_lock_irq(&rbd_dev
->lock
);
444 if (test_bit(RBD_DEV_FLAG_REMOVING
, &rbd_dev
->flags
))
447 rbd_dev
->open_count
++;
448 spin_unlock_irq(&rbd_dev
->lock
);
452 mutex_lock_nested(&ctl_mutex
, SINGLE_DEPTH_NESTING
);
453 (void) get_device(&rbd_dev
->dev
);
454 set_device_ro(bdev
, rbd_dev
->mapping
.read_only
);
455 mutex_unlock(&ctl_mutex
);
460 static int rbd_release(struct gendisk
*disk
, fmode_t mode
)
462 struct rbd_device
*rbd_dev
= disk
->private_data
;
463 unsigned long open_count_before
;
465 spin_lock_irq(&rbd_dev
->lock
);
466 open_count_before
= rbd_dev
->open_count
--;
467 spin_unlock_irq(&rbd_dev
->lock
);
468 rbd_assert(open_count_before
> 0);
470 mutex_lock_nested(&ctl_mutex
, SINGLE_DEPTH_NESTING
);
471 put_device(&rbd_dev
->dev
);
472 mutex_unlock(&ctl_mutex
);
477 static const struct block_device_operations rbd_bd_ops
= {
478 .owner
= THIS_MODULE
,
480 .release
= rbd_release
,
484 * Initialize an rbd client instance.
487 static struct rbd_client
*rbd_client_create(struct ceph_options
*ceph_opts
)
489 struct rbd_client
*rbdc
;
492 dout("%s:\n", __func__
);
493 rbdc
= kmalloc(sizeof(struct rbd_client
), GFP_KERNEL
);
497 kref_init(&rbdc
->kref
);
498 INIT_LIST_HEAD(&rbdc
->node
);
500 mutex_lock_nested(&ctl_mutex
, SINGLE_DEPTH_NESTING
);
502 rbdc
->client
= ceph_create_client(ceph_opts
, rbdc
, 0, 0);
503 if (IS_ERR(rbdc
->client
))
505 ceph_opts
= NULL
; /* Now rbdc->client is responsible for ceph_opts */
507 ret
= ceph_open_session(rbdc
->client
);
511 spin_lock(&rbd_client_list_lock
);
512 list_add_tail(&rbdc
->node
, &rbd_client_list
);
513 spin_unlock(&rbd_client_list_lock
);
515 mutex_unlock(&ctl_mutex
);
516 dout("%s: rbdc %p\n", __func__
, rbdc
);
521 ceph_destroy_client(rbdc
->client
);
523 mutex_unlock(&ctl_mutex
);
527 ceph_destroy_options(ceph_opts
);
528 dout("%s: error %d\n", __func__
, ret
);
533 static struct rbd_client
*__rbd_get_client(struct rbd_client
*rbdc
)
535 kref_get(&rbdc
->kref
);
541 * Find a ceph client with specific addr and configuration. If
542 * found, bump its reference count.
544 static struct rbd_client
*rbd_client_find(struct ceph_options
*ceph_opts
)
546 struct rbd_client
*client_node
;
549 if (ceph_opts
->flags
& CEPH_OPT_NOSHARE
)
552 spin_lock(&rbd_client_list_lock
);
553 list_for_each_entry(client_node
, &rbd_client_list
, node
) {
554 if (!ceph_compare_options(ceph_opts
, client_node
->client
)) {
555 __rbd_get_client(client_node
);
561 spin_unlock(&rbd_client_list_lock
);
563 return found
? client_node
: NULL
;
573 /* string args above */
576 /* Boolean args above */
580 static match_table_t rbd_opts_tokens
= {
582 /* string args above */
583 {Opt_read_only
, "read_only"},
584 {Opt_read_only
, "ro"}, /* Alternate spelling */
585 {Opt_read_write
, "read_write"},
586 {Opt_read_write
, "rw"}, /* Alternate spelling */
587 /* Boolean args above */
595 #define RBD_READ_ONLY_DEFAULT false
597 static int parse_rbd_opts_token(char *c
, void *private)
599 struct rbd_options
*rbd_opts
= private;
600 substring_t argstr
[MAX_OPT_ARGS
];
601 int token
, intval
, ret
;
603 token
= match_token(c
, rbd_opts_tokens
, argstr
);
607 if (token
< Opt_last_int
) {
608 ret
= match_int(&argstr
[0], &intval
);
610 pr_err("bad mount option arg (not int) "
614 dout("got int token %d val %d\n", token
, intval
);
615 } else if (token
> Opt_last_int
&& token
< Opt_last_string
) {
616 dout("got string token %d val %s\n", token
,
618 } else if (token
> Opt_last_string
&& token
< Opt_last_bool
) {
619 dout("got Boolean token %d\n", token
);
621 dout("got token %d\n", token
);
626 rbd_opts
->read_only
= true;
629 rbd_opts
->read_only
= false;
639 * Get a ceph client with specific addr and configuration, if one does
640 * not exist create it.
642 static struct rbd_client
*rbd_get_client(struct ceph_options
*ceph_opts
)
644 struct rbd_client
*rbdc
;
646 rbdc
= rbd_client_find(ceph_opts
);
647 if (rbdc
) /* using an existing client */
648 ceph_destroy_options(ceph_opts
);
650 rbdc
= rbd_client_create(ceph_opts
);
656 * Destroy ceph client
658 * Caller must hold rbd_client_list_lock.
660 static void rbd_client_release(struct kref
*kref
)
662 struct rbd_client
*rbdc
= container_of(kref
, struct rbd_client
, kref
);
664 dout("%s: rbdc %p\n", __func__
, rbdc
);
665 spin_lock(&rbd_client_list_lock
);
666 list_del(&rbdc
->node
);
667 spin_unlock(&rbd_client_list_lock
);
669 ceph_destroy_client(rbdc
->client
);
674 * Drop reference to ceph client node. If it's not referenced anymore, release
677 static void rbd_put_client(struct rbd_client
*rbdc
)
680 kref_put(&rbdc
->kref
, rbd_client_release
);
683 static bool rbd_image_format_valid(u32 image_format
)
685 return image_format
== 1 || image_format
== 2;
688 static bool rbd_dev_ondisk_valid(struct rbd_image_header_ondisk
*ondisk
)
693 /* The header has to start with the magic rbd header text */
694 if (memcmp(&ondisk
->text
, RBD_HEADER_TEXT
, sizeof (RBD_HEADER_TEXT
)))
697 /* The bio layer requires at least sector-sized I/O */
699 if (ondisk
->options
.order
< SECTOR_SHIFT
)
702 /* If we use u64 in a few spots we may be able to loosen this */
704 if (ondisk
->options
.order
> 8 * sizeof (int) - 1)
708 * The size of a snapshot header has to fit in a size_t, and
709 * that limits the number of snapshots.
711 snap_count
= le32_to_cpu(ondisk
->snap_count
);
712 size
= SIZE_MAX
- sizeof (struct ceph_snap_context
);
713 if (snap_count
> size
/ sizeof (__le64
))
717 * Not only that, but the size of the entire the snapshot
718 * header must also be representable in a size_t.
720 size
-= snap_count
* sizeof (__le64
);
721 if ((u64
) size
< le64_to_cpu(ondisk
->snap_names_len
))
728 * Create a new header structure, translate header format from the on-disk
731 static int rbd_header_from_disk(struct rbd_image_header
*header
,
732 struct rbd_image_header_ondisk
*ondisk
)
739 memset(header
, 0, sizeof (*header
));
741 snap_count
= le32_to_cpu(ondisk
->snap_count
);
743 len
= strnlen(ondisk
->object_prefix
, sizeof (ondisk
->object_prefix
));
744 header
->object_prefix
= kmalloc(len
+ 1, GFP_KERNEL
);
745 if (!header
->object_prefix
)
747 memcpy(header
->object_prefix
, ondisk
->object_prefix
, len
);
748 header
->object_prefix
[len
] = '\0';
751 u64 snap_names_len
= le64_to_cpu(ondisk
->snap_names_len
);
753 /* Save a copy of the snapshot names */
755 if (snap_names_len
> (u64
) SIZE_MAX
)
757 header
->snap_names
= kmalloc(snap_names_len
, GFP_KERNEL
);
758 if (!header
->snap_names
)
761 * Note that rbd_dev_v1_header_read() guarantees
762 * the ondisk buffer we're working with has
763 * snap_names_len bytes beyond the end of the
764 * snapshot id array, this memcpy() is safe.
766 memcpy(header
->snap_names
, &ondisk
->snaps
[snap_count
],
769 /* Record each snapshot's size */
771 size
= snap_count
* sizeof (*header
->snap_sizes
);
772 header
->snap_sizes
= kmalloc(size
, GFP_KERNEL
);
773 if (!header
->snap_sizes
)
775 for (i
= 0; i
< snap_count
; i
++)
776 header
->snap_sizes
[i
] =
777 le64_to_cpu(ondisk
->snaps
[i
].image_size
);
779 header
->snap_names
= NULL
;
780 header
->snap_sizes
= NULL
;
783 header
->features
= 0; /* No features support in v1 images */
784 header
->obj_order
= ondisk
->options
.order
;
785 header
->crypt_type
= ondisk
->options
.crypt_type
;
786 header
->comp_type
= ondisk
->options
.comp_type
;
788 /* Allocate and fill in the snapshot context */
790 header
->image_size
= le64_to_cpu(ondisk
->image_size
);
792 header
->snapc
= ceph_create_snap_context(snap_count
, GFP_KERNEL
);
795 header
->snapc
->seq
= le64_to_cpu(ondisk
->snap_seq
);
796 for (i
= 0; i
< snap_count
; i
++)
797 header
->snapc
->snaps
[i
] = le64_to_cpu(ondisk
->snaps
[i
].id
);
802 kfree(header
->snap_sizes
);
803 header
->snap_sizes
= NULL
;
804 kfree(header
->snap_names
);
805 header
->snap_names
= NULL
;
806 kfree(header
->object_prefix
);
807 header
->object_prefix
= NULL
;
812 static const char *rbd_snap_name(struct rbd_device
*rbd_dev
, u64 snap_id
)
814 struct rbd_snap
*snap
;
816 if (snap_id
== CEPH_NOSNAP
)
817 return RBD_SNAP_HEAD_NAME
;
819 list_for_each_entry(snap
, &rbd_dev
->snaps
, node
)
820 if (snap_id
== snap
->id
)
826 static struct rbd_snap
*snap_by_name(struct rbd_device
*rbd_dev
,
827 const char *snap_name
)
829 struct rbd_snap
*snap
;
831 list_for_each_entry(snap
, &rbd_dev
->snaps
, node
)
832 if (!strcmp(snap_name
, snap
->name
))
838 static int rbd_dev_mapping_set(struct rbd_device
*rbd_dev
)
840 if (!memcmp(rbd_dev
->spec
->snap_name
, RBD_SNAP_HEAD_NAME
,
841 sizeof (RBD_SNAP_HEAD_NAME
))) {
842 rbd_dev
->mapping
.size
= rbd_dev
->header
.image_size
;
843 rbd_dev
->mapping
.features
= rbd_dev
->header
.features
;
845 struct rbd_snap
*snap
;
847 snap
= snap_by_name(rbd_dev
, rbd_dev
->spec
->snap_name
);
850 rbd_dev
->mapping
.size
= snap
->size
;
851 rbd_dev
->mapping
.features
= snap
->features
;
852 rbd_dev
->mapping
.read_only
= true;
858 static void rbd_dev_mapping_clear(struct rbd_device
*rbd_dev
)
860 rbd_dev
->mapping
.size
= 0;
861 rbd_dev
->mapping
.features
= 0;
862 rbd_dev
->mapping
.read_only
= true;
865 static void rbd_dev_clear_mapping(struct rbd_device
*rbd_dev
)
867 rbd_dev
->mapping
.size
= 0;
868 rbd_dev
->mapping
.features
= 0;
869 rbd_dev
->mapping
.read_only
= true;
872 static const char *rbd_segment_name(struct rbd_device
*rbd_dev
, u64 offset
)
878 name
= kmalloc(MAX_OBJ_NAME_SIZE
+ 1, GFP_NOIO
);
881 segment
= offset
>> rbd_dev
->header
.obj_order
;
882 ret
= snprintf(name
, MAX_OBJ_NAME_SIZE
+ 1, "%s.%012llx",
883 rbd_dev
->header
.object_prefix
, segment
);
884 if (ret
< 0 || ret
> MAX_OBJ_NAME_SIZE
) {
885 pr_err("error formatting segment name for #%llu (%d)\n",
894 static u64
rbd_segment_offset(struct rbd_device
*rbd_dev
, u64 offset
)
896 u64 segment_size
= (u64
) 1 << rbd_dev
->header
.obj_order
;
898 return offset
& (segment_size
- 1);
901 static u64
rbd_segment_length(struct rbd_device
*rbd_dev
,
902 u64 offset
, u64 length
)
904 u64 segment_size
= (u64
) 1 << rbd_dev
->header
.obj_order
;
906 offset
&= segment_size
- 1;
908 rbd_assert(length
<= U64_MAX
- offset
);
909 if (offset
+ length
> segment_size
)
910 length
= segment_size
- offset
;
916 * returns the size of an object in the image
918 static u64
rbd_obj_bytes(struct rbd_image_header
*header
)
920 return 1 << header
->obj_order
;
927 static void bio_chain_put(struct bio
*chain
)
933 chain
= chain
->bi_next
;
939 * zeros a bio chain, starting at specific offset
941 static void zero_bio_chain(struct bio
*chain
, int start_ofs
)
950 bio_for_each_segment(bv
, chain
, i
) {
951 if (pos
+ bv
->bv_len
> start_ofs
) {
952 int remainder
= max(start_ofs
- pos
, 0);
953 buf
= bvec_kmap_irq(bv
, &flags
);
954 memset(buf
+ remainder
, 0,
955 bv
->bv_len
- remainder
);
956 bvec_kunmap_irq(buf
, &flags
);
961 chain
= chain
->bi_next
;
966 * similar to zero_bio_chain(), zeros data defined by a page array,
967 * starting at the given byte offset from the start of the array and
968 * continuing up to the given end offset. The pages array is
969 * assumed to be big enough to hold all bytes up to the end.
971 static void zero_pages(struct page
**pages
, u64 offset
, u64 end
)
973 struct page
**page
= &pages
[offset
>> PAGE_SHIFT
];
975 rbd_assert(end
> offset
);
976 rbd_assert(end
- offset
<= (u64
)SIZE_MAX
);
977 while (offset
< end
) {
983 page_offset
= (size_t)(offset
& ~PAGE_MASK
);
984 length
= min(PAGE_SIZE
- page_offset
, (size_t)(end
- offset
));
985 local_irq_save(flags
);
986 kaddr
= kmap_atomic(*page
);
987 memset(kaddr
+ page_offset
, 0, length
);
988 kunmap_atomic(kaddr
);
989 local_irq_restore(flags
);
997 * Clone a portion of a bio, starting at the given byte offset
998 * and continuing for the number of bytes indicated.
1000 static struct bio
*bio_clone_range(struct bio
*bio_src
,
1001 unsigned int offset
,
1009 unsigned short end_idx
;
1010 unsigned short vcnt
;
1013 /* Handle the easy case for the caller */
1015 if (!offset
&& len
== bio_src
->bi_size
)
1016 return bio_clone(bio_src
, gfpmask
);
1018 if (WARN_ON_ONCE(!len
))
1020 if (WARN_ON_ONCE(len
> bio_src
->bi_size
))
1022 if (WARN_ON_ONCE(offset
> bio_src
->bi_size
- len
))
1025 /* Find first affected segment... */
1028 __bio_for_each_segment(bv
, bio_src
, idx
, 0) {
1029 if (resid
< bv
->bv_len
)
1031 resid
-= bv
->bv_len
;
1035 /* ...and the last affected segment */
1038 __bio_for_each_segment(bv
, bio_src
, end_idx
, idx
) {
1039 if (resid
<= bv
->bv_len
)
1041 resid
-= bv
->bv_len
;
1043 vcnt
= end_idx
- idx
+ 1;
1045 /* Build the clone */
1047 bio
= bio_alloc(gfpmask
, (unsigned int) vcnt
);
1049 return NULL
; /* ENOMEM */
1051 bio
->bi_bdev
= bio_src
->bi_bdev
;
1052 bio
->bi_sector
= bio_src
->bi_sector
+ (offset
>> SECTOR_SHIFT
);
1053 bio
->bi_rw
= bio_src
->bi_rw
;
1054 bio
->bi_flags
|= 1 << BIO_CLONED
;
1057 * Copy over our part of the bio_vec, then update the first
1058 * and last (or only) entries.
1060 memcpy(&bio
->bi_io_vec
[0], &bio_src
->bi_io_vec
[idx
],
1061 vcnt
* sizeof (struct bio_vec
));
1062 bio
->bi_io_vec
[0].bv_offset
+= voff
;
1064 bio
->bi_io_vec
[0].bv_len
-= voff
;
1065 bio
->bi_io_vec
[vcnt
- 1].bv_len
= resid
;
1067 bio
->bi_io_vec
[0].bv_len
= len
;
1070 bio
->bi_vcnt
= vcnt
;
1078 * Clone a portion of a bio chain, starting at the given byte offset
1079 * into the first bio in the source chain and continuing for the
1080 * number of bytes indicated. The result is another bio chain of
1081 * exactly the given length, or a null pointer on error.
1083 * The bio_src and offset parameters are both in-out. On entry they
1084 * refer to the first source bio and the offset into that bio where
1085 * the start of data to be cloned is located.
1087 * On return, bio_src is updated to refer to the bio in the source
1088 * chain that contains first un-cloned byte, and *offset will
1089 * contain the offset of that byte within that bio.
1091 static struct bio
*bio_chain_clone_range(struct bio
**bio_src
,
1092 unsigned int *offset
,
1096 struct bio
*bi
= *bio_src
;
1097 unsigned int off
= *offset
;
1098 struct bio
*chain
= NULL
;
1101 /* Build up a chain of clone bios up to the limit */
1103 if (!bi
|| off
>= bi
->bi_size
|| !len
)
1104 return NULL
; /* Nothing to clone */
1108 unsigned int bi_size
;
1112 rbd_warn(NULL
, "bio_chain exhausted with %u left", len
);
1113 goto out_err
; /* EINVAL; ran out of bio's */
1115 bi_size
= min_t(unsigned int, bi
->bi_size
- off
, len
);
1116 bio
= bio_clone_range(bi
, off
, bi_size
, gfpmask
);
1118 goto out_err
; /* ENOMEM */
1121 end
= &bio
->bi_next
;
1124 if (off
== bi
->bi_size
) {
1135 bio_chain_put(chain
);
1141 * The default/initial value for all object request flags is 0. For
1142 * each flag, once its value is set to 1 it is never reset to 0
1145 static void obj_request_img_data_set(struct rbd_obj_request
*obj_request
)
1147 if (test_and_set_bit(OBJ_REQ_IMG_DATA
, &obj_request
->flags
)) {
1148 struct rbd_device
*rbd_dev
;
1150 rbd_dev
= obj_request
->img_request
->rbd_dev
;
1151 rbd_warn(rbd_dev
, "obj_request %p already marked img_data\n",
1156 static bool obj_request_img_data_test(struct rbd_obj_request
*obj_request
)
1159 return test_bit(OBJ_REQ_IMG_DATA
, &obj_request
->flags
) != 0;
1162 static void obj_request_done_set(struct rbd_obj_request
*obj_request
)
1164 if (test_and_set_bit(OBJ_REQ_DONE
, &obj_request
->flags
)) {
1165 struct rbd_device
*rbd_dev
= NULL
;
1167 if (obj_request_img_data_test(obj_request
))
1168 rbd_dev
= obj_request
->img_request
->rbd_dev
;
1169 rbd_warn(rbd_dev
, "obj_request %p already marked done\n",
1174 static bool obj_request_done_test(struct rbd_obj_request
*obj_request
)
1177 return test_bit(OBJ_REQ_DONE
, &obj_request
->flags
) != 0;
1181 * This sets the KNOWN flag after (possibly) setting the EXISTS
1182 * flag. The latter is set based on the "exists" value provided.
1184 * Note that for our purposes once an object exists it never goes
1185 * away again. It's possible that the response from two existence
1186 * checks are separated by the creation of the target object, and
1187 * the first ("doesn't exist") response arrives *after* the second
1188 * ("does exist"). In that case we ignore the second one.
1190 static void obj_request_existence_set(struct rbd_obj_request
*obj_request
,
1194 set_bit(OBJ_REQ_EXISTS
, &obj_request
->flags
);
1195 set_bit(OBJ_REQ_KNOWN
, &obj_request
->flags
);
1199 static bool obj_request_known_test(struct rbd_obj_request
*obj_request
)
1202 return test_bit(OBJ_REQ_KNOWN
, &obj_request
->flags
) != 0;
1205 static bool obj_request_exists_test(struct rbd_obj_request
*obj_request
)
1208 return test_bit(OBJ_REQ_EXISTS
, &obj_request
->flags
) != 0;
1211 static void rbd_obj_request_get(struct rbd_obj_request
*obj_request
)
1213 dout("%s: obj %p (was %d)\n", __func__
, obj_request
,
1214 atomic_read(&obj_request
->kref
.refcount
));
1215 kref_get(&obj_request
->kref
);
1218 static void rbd_obj_request_destroy(struct kref
*kref
);
1219 static void rbd_obj_request_put(struct rbd_obj_request
*obj_request
)
1221 rbd_assert(obj_request
!= NULL
);
1222 dout("%s: obj %p (was %d)\n", __func__
, obj_request
,
1223 atomic_read(&obj_request
->kref
.refcount
));
1224 kref_put(&obj_request
->kref
, rbd_obj_request_destroy
);
1227 static void rbd_img_request_get(struct rbd_img_request
*img_request
)
1229 dout("%s: img %p (was %d)\n", __func__
, img_request
,
1230 atomic_read(&img_request
->kref
.refcount
));
1231 kref_get(&img_request
->kref
);
1234 static void rbd_img_request_destroy(struct kref
*kref
);
1235 static void rbd_img_request_put(struct rbd_img_request
*img_request
)
1237 rbd_assert(img_request
!= NULL
);
1238 dout("%s: img %p (was %d)\n", __func__
, img_request
,
1239 atomic_read(&img_request
->kref
.refcount
));
1240 kref_put(&img_request
->kref
, rbd_img_request_destroy
);
1243 static inline void rbd_img_obj_request_add(struct rbd_img_request
*img_request
,
1244 struct rbd_obj_request
*obj_request
)
1246 rbd_assert(obj_request
->img_request
== NULL
);
1248 /* Image request now owns object's original reference */
1249 obj_request
->img_request
= img_request
;
1250 obj_request
->which
= img_request
->obj_request_count
;
1251 rbd_assert(!obj_request_img_data_test(obj_request
));
1252 obj_request_img_data_set(obj_request
);
1253 rbd_assert(obj_request
->which
!= BAD_WHICH
);
1254 img_request
->obj_request_count
++;
1255 list_add_tail(&obj_request
->links
, &img_request
->obj_requests
);
1256 dout("%s: img %p obj %p w=%u\n", __func__
, img_request
, obj_request
,
1257 obj_request
->which
);
1260 static inline void rbd_img_obj_request_del(struct rbd_img_request
*img_request
,
1261 struct rbd_obj_request
*obj_request
)
1263 rbd_assert(obj_request
->which
!= BAD_WHICH
);
1265 dout("%s: img %p obj %p w=%u\n", __func__
, img_request
, obj_request
,
1266 obj_request
->which
);
1267 list_del(&obj_request
->links
);
1268 rbd_assert(img_request
->obj_request_count
> 0);
1269 img_request
->obj_request_count
--;
1270 rbd_assert(obj_request
->which
== img_request
->obj_request_count
);
1271 obj_request
->which
= BAD_WHICH
;
1272 rbd_assert(obj_request_img_data_test(obj_request
));
1273 rbd_assert(obj_request
->img_request
== img_request
);
1274 obj_request
->img_request
= NULL
;
1275 obj_request
->callback
= NULL
;
1276 rbd_obj_request_put(obj_request
);
1279 static bool obj_request_type_valid(enum obj_request_type type
)
1282 case OBJ_REQUEST_NODATA
:
1283 case OBJ_REQUEST_BIO
:
1284 case OBJ_REQUEST_PAGES
:
1291 static int rbd_obj_request_submit(struct ceph_osd_client
*osdc
,
1292 struct rbd_obj_request
*obj_request
)
1294 dout("%s: osdc %p obj %p\n", __func__
, osdc
, obj_request
);
1296 return ceph_osdc_start_request(osdc
, obj_request
->osd_req
, false);
1299 static void rbd_img_request_complete(struct rbd_img_request
*img_request
)
1302 dout("%s: img %p\n", __func__
, img_request
);
1305 * If no error occurred, compute the aggregate transfer
1306 * count for the image request. We could instead use
1307 * atomic64_cmpxchg() to update it as each object request
1308 * completes; not clear which way is better off hand.
1310 if (!img_request
->result
) {
1311 struct rbd_obj_request
*obj_request
;
1314 for_each_obj_request(img_request
, obj_request
)
1315 xferred
+= obj_request
->xferred
;
1316 img_request
->xferred
= xferred
;
1319 if (img_request
->callback
)
1320 img_request
->callback(img_request
);
1322 rbd_img_request_put(img_request
);
1325 /* Caller is responsible for rbd_obj_request_destroy(obj_request) */
1327 static int rbd_obj_request_wait(struct rbd_obj_request
*obj_request
)
1329 dout("%s: obj %p\n", __func__
, obj_request
);
1331 return wait_for_completion_interruptible(&obj_request
->completion
);
1335 * The default/initial value for all image request flags is 0. Each
1336 * is conditionally set to 1 at image request initialization time
1337 * and currently never change thereafter.
1339 static void img_request_write_set(struct rbd_img_request
*img_request
)
1341 set_bit(IMG_REQ_WRITE
, &img_request
->flags
);
1345 static bool img_request_write_test(struct rbd_img_request
*img_request
)
1348 return test_bit(IMG_REQ_WRITE
, &img_request
->flags
) != 0;
1351 static void img_request_child_set(struct rbd_img_request
*img_request
)
1353 set_bit(IMG_REQ_CHILD
, &img_request
->flags
);
1357 static bool img_request_child_test(struct rbd_img_request
*img_request
)
1360 return test_bit(IMG_REQ_CHILD
, &img_request
->flags
) != 0;
1363 static void img_request_layered_set(struct rbd_img_request
*img_request
)
1365 set_bit(IMG_REQ_LAYERED
, &img_request
->flags
);
1369 static bool img_request_layered_test(struct rbd_img_request
*img_request
)
1372 return test_bit(IMG_REQ_LAYERED
, &img_request
->flags
) != 0;
1376 rbd_img_obj_request_read_callback(struct rbd_obj_request
*obj_request
)
1378 u64 xferred
= obj_request
->xferred
;
1379 u64 length
= obj_request
->length
;
1381 dout("%s: obj %p img %p result %d %llu/%llu\n", __func__
,
1382 obj_request
, obj_request
->img_request
, obj_request
->result
,
1385 * ENOENT means a hole in the image. We zero-fill the
1386 * entire length of the request. A short read also implies
1387 * zero-fill to the end of the request. Either way we
1388 * update the xferred count to indicate the whole request
1391 rbd_assert(obj_request
->type
!= OBJ_REQUEST_NODATA
);
1392 if (obj_request
->result
== -ENOENT
) {
1393 if (obj_request
->type
== OBJ_REQUEST_BIO
)
1394 zero_bio_chain(obj_request
->bio_list
, 0);
1396 zero_pages(obj_request
->pages
, 0, length
);
1397 obj_request
->result
= 0;
1398 obj_request
->xferred
= length
;
1399 } else if (xferred
< length
&& !obj_request
->result
) {
1400 if (obj_request
->type
== OBJ_REQUEST_BIO
)
1401 zero_bio_chain(obj_request
->bio_list
, xferred
);
1403 zero_pages(obj_request
->pages
, xferred
, length
);
1404 obj_request
->xferred
= length
;
1406 obj_request_done_set(obj_request
);
1409 static void rbd_obj_request_complete(struct rbd_obj_request
*obj_request
)
1411 dout("%s: obj %p cb %p\n", __func__
, obj_request
,
1412 obj_request
->callback
);
1413 if (obj_request
->callback
)
1414 obj_request
->callback(obj_request
);
1416 complete_all(&obj_request
->completion
);
1419 static void rbd_osd_trivial_callback(struct rbd_obj_request
*obj_request
)
1421 dout("%s: obj %p\n", __func__
, obj_request
);
1422 obj_request_done_set(obj_request
);
1425 static void rbd_osd_read_callback(struct rbd_obj_request
*obj_request
)
1427 struct rbd_img_request
*img_request
= NULL
;
1428 struct rbd_device
*rbd_dev
= NULL
;
1429 bool layered
= false;
1431 if (obj_request_img_data_test(obj_request
)) {
1432 img_request
= obj_request
->img_request
;
1433 layered
= img_request
&& img_request_layered_test(img_request
);
1434 rbd_dev
= img_request
->rbd_dev
;
1437 dout("%s: obj %p img %p result %d %llu/%llu\n", __func__
,
1438 obj_request
, img_request
, obj_request
->result
,
1439 obj_request
->xferred
, obj_request
->length
);
1440 if (layered
&& obj_request
->result
== -ENOENT
&&
1441 obj_request
->img_offset
< rbd_dev
->parent_overlap
)
1442 rbd_img_parent_read(obj_request
);
1443 else if (img_request
)
1444 rbd_img_obj_request_read_callback(obj_request
);
1446 obj_request_done_set(obj_request
);
1449 static void rbd_osd_write_callback(struct rbd_obj_request
*obj_request
)
1451 dout("%s: obj %p result %d %llu\n", __func__
, obj_request
,
1452 obj_request
->result
, obj_request
->length
);
1454 * There is no such thing as a successful short write. Set
1455 * it to our originally-requested length.
1457 obj_request
->xferred
= obj_request
->length
;
1458 obj_request_done_set(obj_request
);
1462 * For a simple stat call there's nothing to do. We'll do more if
1463 * this is part of a write sequence for a layered image.
1465 static void rbd_osd_stat_callback(struct rbd_obj_request
*obj_request
)
1467 dout("%s: obj %p\n", __func__
, obj_request
);
1468 obj_request_done_set(obj_request
);
1471 static void rbd_osd_req_callback(struct ceph_osd_request
*osd_req
,
1472 struct ceph_msg
*msg
)
1474 struct rbd_obj_request
*obj_request
= osd_req
->r_priv
;
1477 dout("%s: osd_req %p msg %p\n", __func__
, osd_req
, msg
);
1478 rbd_assert(osd_req
== obj_request
->osd_req
);
1479 if (obj_request_img_data_test(obj_request
)) {
1480 rbd_assert(obj_request
->img_request
);
1481 rbd_assert(obj_request
->which
!= BAD_WHICH
);
1483 rbd_assert(obj_request
->which
== BAD_WHICH
);
1486 if (osd_req
->r_result
< 0)
1487 obj_request
->result
= osd_req
->r_result
;
1489 BUG_ON(osd_req
->r_num_ops
> 2);
1492 * We support a 64-bit length, but ultimately it has to be
1493 * passed to blk_end_request(), which takes an unsigned int.
1495 obj_request
->xferred
= osd_req
->r_reply_op_len
[0];
1496 rbd_assert(obj_request
->xferred
< (u64
)UINT_MAX
);
1497 opcode
= osd_req
->r_ops
[0].op
;
1499 case CEPH_OSD_OP_READ
:
1500 rbd_osd_read_callback(obj_request
);
1502 case CEPH_OSD_OP_WRITE
:
1503 rbd_osd_write_callback(obj_request
);
1505 case CEPH_OSD_OP_STAT
:
1506 rbd_osd_stat_callback(obj_request
);
1508 case CEPH_OSD_OP_CALL
:
1509 case CEPH_OSD_OP_NOTIFY_ACK
:
1510 case CEPH_OSD_OP_WATCH
:
1511 rbd_osd_trivial_callback(obj_request
);
1514 rbd_warn(NULL
, "%s: unsupported op %hu\n",
1515 obj_request
->object_name
, (unsigned short) opcode
);
1519 if (obj_request_done_test(obj_request
))
1520 rbd_obj_request_complete(obj_request
);
1523 static void rbd_osd_req_format_read(struct rbd_obj_request
*obj_request
)
1525 struct rbd_img_request
*img_request
= obj_request
->img_request
;
1526 struct ceph_osd_request
*osd_req
= obj_request
->osd_req
;
1529 rbd_assert(osd_req
!= NULL
);
1531 snap_id
= img_request
? img_request
->snap_id
: CEPH_NOSNAP
;
1532 ceph_osdc_build_request(osd_req
, obj_request
->offset
,
1533 NULL
, snap_id
, NULL
);
1536 static void rbd_osd_req_format_write(struct rbd_obj_request
*obj_request
)
1538 struct rbd_img_request
*img_request
= obj_request
->img_request
;
1539 struct ceph_osd_request
*osd_req
= obj_request
->osd_req
;
1540 struct ceph_snap_context
*snapc
;
1541 struct timespec mtime
= CURRENT_TIME
;
1543 rbd_assert(osd_req
!= NULL
);
1545 snapc
= img_request
? img_request
->snapc
: NULL
;
1546 ceph_osdc_build_request(osd_req
, obj_request
->offset
,
1547 snapc
, CEPH_NOSNAP
, &mtime
);
1550 static struct ceph_osd_request
*rbd_osd_req_create(
1551 struct rbd_device
*rbd_dev
,
1553 struct rbd_obj_request
*obj_request
)
1555 struct ceph_snap_context
*snapc
= NULL
;
1556 struct ceph_osd_client
*osdc
;
1557 struct ceph_osd_request
*osd_req
;
1559 if (obj_request_img_data_test(obj_request
)) {
1560 struct rbd_img_request
*img_request
= obj_request
->img_request
;
1562 rbd_assert(write_request
==
1563 img_request_write_test(img_request
));
1565 snapc
= img_request
->snapc
;
1568 /* Allocate and initialize the request, for the single op */
1570 osdc
= &rbd_dev
->rbd_client
->client
->osdc
;
1571 osd_req
= ceph_osdc_alloc_request(osdc
, snapc
, 1, false, GFP_ATOMIC
);
1573 return NULL
; /* ENOMEM */
1576 osd_req
->r_flags
= CEPH_OSD_FLAG_WRITE
| CEPH_OSD_FLAG_ONDISK
;
1578 osd_req
->r_flags
= CEPH_OSD_FLAG_READ
;
1580 osd_req
->r_callback
= rbd_osd_req_callback
;
1581 osd_req
->r_priv
= obj_request
;
1583 osd_req
->r_oid_len
= strlen(obj_request
->object_name
);
1584 rbd_assert(osd_req
->r_oid_len
< sizeof (osd_req
->r_oid
));
1585 memcpy(osd_req
->r_oid
, obj_request
->object_name
, osd_req
->r_oid_len
);
1587 osd_req
->r_file_layout
= rbd_dev
->layout
; /* struct */
1593 * Create a copyup osd request based on the information in the
1594 * object request supplied. A copyup request has two osd ops,
1595 * a copyup method call, and a "normal" write request.
1597 static struct ceph_osd_request
*
1598 rbd_osd_req_create_copyup(struct rbd_obj_request
*obj_request
)
1600 struct rbd_img_request
*img_request
;
1601 struct ceph_snap_context
*snapc
;
1602 struct rbd_device
*rbd_dev
;
1603 struct ceph_osd_client
*osdc
;
1604 struct ceph_osd_request
*osd_req
;
1606 rbd_assert(obj_request_img_data_test(obj_request
));
1607 img_request
= obj_request
->img_request
;
1608 rbd_assert(img_request
);
1609 rbd_assert(img_request_write_test(img_request
));
1611 /* Allocate and initialize the request, for the two ops */
1613 snapc
= img_request
->snapc
;
1614 rbd_dev
= img_request
->rbd_dev
;
1615 osdc
= &rbd_dev
->rbd_client
->client
->osdc
;
1616 osd_req
= ceph_osdc_alloc_request(osdc
, snapc
, 2, false, GFP_ATOMIC
);
1618 return NULL
; /* ENOMEM */
1620 osd_req
->r_flags
= CEPH_OSD_FLAG_WRITE
| CEPH_OSD_FLAG_ONDISK
;
1621 osd_req
->r_callback
= rbd_osd_req_callback
;
1622 osd_req
->r_priv
= obj_request
;
1624 osd_req
->r_oid_len
= strlen(obj_request
->object_name
);
1625 rbd_assert(osd_req
->r_oid_len
< sizeof (osd_req
->r_oid
));
1626 memcpy(osd_req
->r_oid
, obj_request
->object_name
, osd_req
->r_oid_len
);
1628 osd_req
->r_file_layout
= rbd_dev
->layout
; /* struct */
1634 static void rbd_osd_req_destroy(struct ceph_osd_request
*osd_req
)
1636 ceph_osdc_put_request(osd_req
);
1639 /* object_name is assumed to be a non-null pointer and NUL-terminated */
1641 static struct rbd_obj_request
*rbd_obj_request_create(const char *object_name
,
1642 u64 offset
, u64 length
,
1643 enum obj_request_type type
)
1645 struct rbd_obj_request
*obj_request
;
1649 rbd_assert(obj_request_type_valid(type
));
1651 size
= strlen(object_name
) + 1;
1652 obj_request
= kzalloc(sizeof (*obj_request
) + size
, GFP_KERNEL
);
1656 name
= (char *)(obj_request
+ 1);
1657 obj_request
->object_name
= memcpy(name
, object_name
, size
);
1658 obj_request
->offset
= offset
;
1659 obj_request
->length
= length
;
1660 obj_request
->flags
= 0;
1661 obj_request
->which
= BAD_WHICH
;
1662 obj_request
->type
= type
;
1663 INIT_LIST_HEAD(&obj_request
->links
);
1664 init_completion(&obj_request
->completion
);
1665 kref_init(&obj_request
->kref
);
1667 dout("%s: \"%s\" %llu/%llu %d -> obj %p\n", __func__
, object_name
,
1668 offset
, length
, (int)type
, obj_request
);
1673 static void rbd_obj_request_destroy(struct kref
*kref
)
1675 struct rbd_obj_request
*obj_request
;
1677 obj_request
= container_of(kref
, struct rbd_obj_request
, kref
);
1679 dout("%s: obj %p\n", __func__
, obj_request
);
1681 rbd_assert(obj_request
->img_request
== NULL
);
1682 rbd_assert(obj_request
->which
== BAD_WHICH
);
1684 if (obj_request
->osd_req
)
1685 rbd_osd_req_destroy(obj_request
->osd_req
);
1687 rbd_assert(obj_request_type_valid(obj_request
->type
));
1688 switch (obj_request
->type
) {
1689 case OBJ_REQUEST_NODATA
:
1690 break; /* Nothing to do */
1691 case OBJ_REQUEST_BIO
:
1692 if (obj_request
->bio_list
)
1693 bio_chain_put(obj_request
->bio_list
);
1695 case OBJ_REQUEST_PAGES
:
1696 if (obj_request
->pages
)
1697 ceph_release_page_vector(obj_request
->pages
,
1698 obj_request
->page_count
);
1706 * Caller is responsible for filling in the list of object requests
1707 * that comprises the image request, and the Linux request pointer
1708 * (if there is one).
1710 static struct rbd_img_request
*rbd_img_request_create(
1711 struct rbd_device
*rbd_dev
,
1712 u64 offset
, u64 length
,
1716 struct rbd_img_request
*img_request
;
1718 img_request
= kmalloc(sizeof (*img_request
), GFP_ATOMIC
);
1722 if (write_request
) {
1723 down_read(&rbd_dev
->header_rwsem
);
1724 ceph_get_snap_context(rbd_dev
->header
.snapc
);
1725 up_read(&rbd_dev
->header_rwsem
);
1728 img_request
->rq
= NULL
;
1729 img_request
->rbd_dev
= rbd_dev
;
1730 img_request
->offset
= offset
;
1731 img_request
->length
= length
;
1732 img_request
->flags
= 0;
1733 if (write_request
) {
1734 img_request_write_set(img_request
);
1735 img_request
->snapc
= rbd_dev
->header
.snapc
;
1737 img_request
->snap_id
= rbd_dev
->spec
->snap_id
;
1740 img_request_child_set(img_request
);
1741 if (rbd_dev
->parent_spec
)
1742 img_request_layered_set(img_request
);
1743 spin_lock_init(&img_request
->completion_lock
);
1744 img_request
->next_completion
= 0;
1745 img_request
->callback
= NULL
;
1746 img_request
->result
= 0;
1747 img_request
->obj_request_count
= 0;
1748 INIT_LIST_HEAD(&img_request
->obj_requests
);
1749 kref_init(&img_request
->kref
);
1751 rbd_img_request_get(img_request
); /* Avoid a warning */
1752 rbd_img_request_put(img_request
); /* TEMPORARY */
1754 dout("%s: rbd_dev %p %s %llu/%llu -> img %p\n", __func__
, rbd_dev
,
1755 write_request
? "write" : "read", offset
, length
,
1761 static void rbd_img_request_destroy(struct kref
*kref
)
1763 struct rbd_img_request
*img_request
;
1764 struct rbd_obj_request
*obj_request
;
1765 struct rbd_obj_request
*next_obj_request
;
1767 img_request
= container_of(kref
, struct rbd_img_request
, kref
);
1769 dout("%s: img %p\n", __func__
, img_request
);
1771 for_each_obj_request_safe(img_request
, obj_request
, next_obj_request
)
1772 rbd_img_obj_request_del(img_request
, obj_request
);
1773 rbd_assert(img_request
->obj_request_count
== 0);
1775 if (img_request_write_test(img_request
))
1776 ceph_put_snap_context(img_request
->snapc
);
1778 if (img_request_child_test(img_request
))
1779 rbd_obj_request_put(img_request
->obj_request
);
1784 static bool rbd_img_obj_end_request(struct rbd_obj_request
*obj_request
)
1786 struct rbd_img_request
*img_request
;
1787 unsigned int xferred
;
1791 rbd_assert(obj_request_img_data_test(obj_request
));
1792 img_request
= obj_request
->img_request
;
1794 rbd_assert(obj_request
->xferred
<= (u64
)UINT_MAX
);
1795 xferred
= (unsigned int)obj_request
->xferred
;
1796 result
= obj_request
->result
;
1798 struct rbd_device
*rbd_dev
= img_request
->rbd_dev
;
1800 rbd_warn(rbd_dev
, "%s %llx at %llx (%llx)\n",
1801 img_request_write_test(img_request
) ? "write" : "read",
1802 obj_request
->length
, obj_request
->img_offset
,
1803 obj_request
->offset
);
1804 rbd_warn(rbd_dev
, " result %d xferred %x\n",
1806 if (!img_request
->result
)
1807 img_request
->result
= result
;
1810 /* Image object requests don't own their page array */
1812 if (obj_request
->type
== OBJ_REQUEST_PAGES
) {
1813 obj_request
->pages
= NULL
;
1814 obj_request
->page_count
= 0;
1817 if (img_request_child_test(img_request
)) {
1818 rbd_assert(img_request
->obj_request
!= NULL
);
1819 more
= obj_request
->which
< img_request
->obj_request_count
- 1;
1821 rbd_assert(img_request
->rq
!= NULL
);
1822 more
= blk_end_request(img_request
->rq
, result
, xferred
);
1828 static void rbd_img_obj_callback(struct rbd_obj_request
*obj_request
)
1830 struct rbd_img_request
*img_request
;
1831 u32 which
= obj_request
->which
;
1834 rbd_assert(obj_request_img_data_test(obj_request
));
1835 img_request
= obj_request
->img_request
;
1837 dout("%s: img %p obj %p\n", __func__
, img_request
, obj_request
);
1838 rbd_assert(img_request
!= NULL
);
1839 rbd_assert(img_request
->obj_request_count
> 0);
1840 rbd_assert(which
!= BAD_WHICH
);
1841 rbd_assert(which
< img_request
->obj_request_count
);
1842 rbd_assert(which
>= img_request
->next_completion
);
1844 spin_lock_irq(&img_request
->completion_lock
);
1845 if (which
!= img_request
->next_completion
)
1848 for_each_obj_request_from(img_request
, obj_request
) {
1850 rbd_assert(which
< img_request
->obj_request_count
);
1852 if (!obj_request_done_test(obj_request
))
1854 more
= rbd_img_obj_end_request(obj_request
);
1858 rbd_assert(more
^ (which
== img_request
->obj_request_count
));
1859 img_request
->next_completion
= which
;
1861 spin_unlock_irq(&img_request
->completion_lock
);
1864 rbd_img_request_complete(img_request
);
1868 * Split up an image request into one or more object requests, each
1869 * to a different object. The "type" parameter indicates whether
1870 * "data_desc" is the pointer to the head of a list of bio
1871 * structures, or the base of a page array. In either case this
1872 * function assumes data_desc describes memory sufficient to hold
1873 * all data described by the image request.
1875 static int rbd_img_request_fill(struct rbd_img_request
*img_request
,
1876 enum obj_request_type type
,
1879 struct rbd_device
*rbd_dev
= img_request
->rbd_dev
;
1880 struct rbd_obj_request
*obj_request
= NULL
;
1881 struct rbd_obj_request
*next_obj_request
;
1882 bool write_request
= img_request_write_test(img_request
);
1883 struct bio
*bio_list
;
1884 unsigned int bio_offset
= 0;
1885 struct page
**pages
;
1890 dout("%s: img %p type %d data_desc %p\n", __func__
, img_request
,
1891 (int)type
, data_desc
);
1893 opcode
= write_request
? CEPH_OSD_OP_WRITE
: CEPH_OSD_OP_READ
;
1894 img_offset
= img_request
->offset
;
1895 resid
= img_request
->length
;
1896 rbd_assert(resid
> 0);
1898 if (type
== OBJ_REQUEST_BIO
) {
1899 bio_list
= data_desc
;
1900 rbd_assert(img_offset
== bio_list
->bi_sector
<< SECTOR_SHIFT
);
1902 rbd_assert(type
== OBJ_REQUEST_PAGES
);
1907 struct ceph_osd_request
*osd_req
;
1908 const char *object_name
;
1912 object_name
= rbd_segment_name(rbd_dev
, img_offset
);
1915 offset
= rbd_segment_offset(rbd_dev
, img_offset
);
1916 length
= rbd_segment_length(rbd_dev
, img_offset
, resid
);
1917 obj_request
= rbd_obj_request_create(object_name
,
1918 offset
, length
, type
);
1919 kfree(object_name
); /* object request has its own copy */
1923 if (type
== OBJ_REQUEST_BIO
) {
1924 unsigned int clone_size
;
1926 rbd_assert(length
<= (u64
)UINT_MAX
);
1927 clone_size
= (unsigned int)length
;
1928 obj_request
->bio_list
=
1929 bio_chain_clone_range(&bio_list
,
1933 if (!obj_request
->bio_list
)
1936 unsigned int page_count
;
1938 obj_request
->pages
= pages
;
1939 page_count
= (u32
)calc_pages_for(offset
, length
);
1940 obj_request
->page_count
= page_count
;
1941 if ((offset
+ length
) & ~PAGE_MASK
)
1942 page_count
--; /* more on last page */
1943 pages
+= page_count
;
1946 osd_req
= rbd_osd_req_create(rbd_dev
, write_request
,
1950 obj_request
->osd_req
= osd_req
;
1951 obj_request
->callback
= rbd_img_obj_callback
;
1953 osd_req_op_extent_init(osd_req
, 0, opcode
, offset
, length
,
1955 if (type
== OBJ_REQUEST_BIO
)
1956 osd_req_op_extent_osd_data_bio(osd_req
, 0,
1957 obj_request
->bio_list
, length
);
1959 osd_req_op_extent_osd_data_pages(osd_req
, 0,
1960 obj_request
->pages
, length
,
1961 offset
& ~PAGE_MASK
, false, false);
1964 rbd_osd_req_format_write(obj_request
);
1966 rbd_osd_req_format_read(obj_request
);
1968 obj_request
->img_offset
= img_offset
;
1969 rbd_img_obj_request_add(img_request
, obj_request
);
1971 img_offset
+= length
;
1978 rbd_obj_request_put(obj_request
);
1980 for_each_obj_request_safe(img_request
, obj_request
, next_obj_request
)
1981 rbd_obj_request_put(obj_request
);
1987 rbd_img_obj_copyup_callback(struct rbd_obj_request
*obj_request
)
1989 struct rbd_img_request
*img_request
;
1990 struct rbd_device
*rbd_dev
;
1994 rbd_assert(obj_request
->type
== OBJ_REQUEST_BIO
);
1995 rbd_assert(obj_request_img_data_test(obj_request
));
1996 img_request
= obj_request
->img_request
;
1997 rbd_assert(img_request
);
1999 rbd_dev
= img_request
->rbd_dev
;
2000 rbd_assert(rbd_dev
);
2001 length
= (u64
)1 << rbd_dev
->header
.obj_order
;
2002 page_count
= (u32
)calc_pages_for(0, length
);
2004 rbd_assert(obj_request
->copyup_pages
);
2005 ceph_release_page_vector(obj_request
->copyup_pages
, page_count
);
2006 obj_request
->copyup_pages
= NULL
;
2009 * We want the transfer count to reflect the size of the
2010 * original write request. There is no such thing as a
2011 * successful short write, so if the request was successful
2012 * we can just set it to the originally-requested length.
2014 if (!obj_request
->result
)
2015 obj_request
->xferred
= obj_request
->length
;
2017 /* Finish up with the normal image object callback */
2019 rbd_img_obj_callback(obj_request
);
2023 rbd_img_obj_parent_read_full_callback(struct rbd_img_request
*img_request
)
2025 struct rbd_obj_request
*orig_request
;
2026 struct ceph_osd_request
*osd_req
;
2027 struct ceph_osd_client
*osdc
;
2028 struct rbd_device
*rbd_dev
;
2029 struct page
**pages
;
2034 rbd_assert(img_request_child_test(img_request
));
2036 /* First get what we need from the image request */
2038 pages
= img_request
->copyup_pages
;
2039 rbd_assert(pages
!= NULL
);
2040 img_request
->copyup_pages
= NULL
;
2042 orig_request
= img_request
->obj_request
;
2043 rbd_assert(orig_request
!= NULL
);
2044 rbd_assert(orig_request
->type
== OBJ_REQUEST_BIO
);
2045 result
= img_request
->result
;
2046 obj_size
= img_request
->length
;
2047 xferred
= img_request
->xferred
;
2049 rbd_dev
= img_request
->rbd_dev
;
2050 rbd_assert(rbd_dev
);
2051 rbd_assert(obj_size
== (u64
)1 << rbd_dev
->header
.obj_order
);
2053 rbd_img_request_put(img_request
);
2058 /* Allocate the new copyup osd request for the original request */
2061 rbd_assert(!orig_request
->osd_req
);
2062 osd_req
= rbd_osd_req_create_copyup(orig_request
);
2065 orig_request
->osd_req
= osd_req
;
2066 orig_request
->copyup_pages
= pages
;
2068 /* Initialize the copyup op */
2070 osd_req_op_cls_init(osd_req
, 0, CEPH_OSD_OP_CALL
, "rbd", "copyup");
2071 osd_req_op_cls_request_data_pages(osd_req
, 0, pages
, obj_size
, 0,
2074 /* Then the original write request op */
2076 osd_req_op_extent_init(osd_req
, 1, CEPH_OSD_OP_WRITE
,
2077 orig_request
->offset
,
2078 orig_request
->length
, 0, 0);
2079 osd_req_op_extent_osd_data_bio(osd_req
, 1, orig_request
->bio_list
,
2080 orig_request
->length
);
2082 rbd_osd_req_format_write(orig_request
);
2084 /* All set, send it off. */
2086 orig_request
->callback
= rbd_img_obj_copyup_callback
;
2087 osdc
= &rbd_dev
->rbd_client
->client
->osdc
;
2088 result
= rbd_obj_request_submit(osdc
, orig_request
);
2092 /* Record the error code and complete the request */
2094 orig_request
->result
= result
;
2095 orig_request
->xferred
= 0;
2096 obj_request_done_set(orig_request
);
2097 rbd_obj_request_complete(orig_request
);
2101 * Read from the parent image the range of data that covers the
2102 * entire target of the given object request. This is used for
2103 * satisfying a layered image write request when the target of an
2104 * object request from the image request does not exist.
2106 * A page array big enough to hold the returned data is allocated
2107 * and supplied to rbd_img_request_fill() as the "data descriptor."
2108 * When the read completes, this page array will be transferred to
2109 * the original object request for the copyup operation.
2111 * If an error occurs, record it as the result of the original
2112 * object request and mark it done so it gets completed.
2114 static int rbd_img_obj_parent_read_full(struct rbd_obj_request
*obj_request
)
2116 struct rbd_img_request
*img_request
= NULL
;
2117 struct rbd_img_request
*parent_request
= NULL
;
2118 struct rbd_device
*rbd_dev
;
2121 struct page
**pages
= NULL
;
2125 rbd_assert(obj_request_img_data_test(obj_request
));
2126 rbd_assert(obj_request
->type
== OBJ_REQUEST_BIO
);
2128 img_request
= obj_request
->img_request
;
2129 rbd_assert(img_request
!= NULL
);
2130 rbd_dev
= img_request
->rbd_dev
;
2131 rbd_assert(rbd_dev
->parent
!= NULL
);
2134 * First things first. The original osd request is of no
2135 * use to use any more, we'll need a new one that can hold
2136 * the two ops in a copyup request. We'll get that later,
2137 * but for now we can release the old one.
2139 rbd_osd_req_destroy(obj_request
->osd_req
);
2140 obj_request
->osd_req
= NULL
;
2143 * Determine the byte range covered by the object in the
2144 * child image to which the original request was to be sent.
2146 img_offset
= obj_request
->img_offset
- obj_request
->offset
;
2147 length
= (u64
)1 << rbd_dev
->header
.obj_order
;
2150 * There is no defined parent data beyond the parent
2151 * overlap, so limit what we read at that boundary if
2154 if (img_offset
+ length
> rbd_dev
->parent_overlap
) {
2155 rbd_assert(img_offset
< rbd_dev
->parent_overlap
);
2156 length
= rbd_dev
->parent_overlap
- img_offset
;
2160 * Allocate a page array big enough to receive the data read
2163 page_count
= (u32
)calc_pages_for(0, length
);
2164 pages
= ceph_alloc_page_vector(page_count
, GFP_KERNEL
);
2165 if (IS_ERR(pages
)) {
2166 result
= PTR_ERR(pages
);
2172 parent_request
= rbd_img_request_create(rbd_dev
->parent
,
2175 if (!parent_request
)
2177 rbd_obj_request_get(obj_request
);
2178 parent_request
->obj_request
= obj_request
;
2180 result
= rbd_img_request_fill(parent_request
, OBJ_REQUEST_PAGES
, pages
);
2183 parent_request
->copyup_pages
= pages
;
2185 parent_request
->callback
= rbd_img_obj_parent_read_full_callback
;
2186 result
= rbd_img_request_submit(parent_request
);
2190 parent_request
->copyup_pages
= NULL
;
2191 parent_request
->obj_request
= NULL
;
2192 rbd_obj_request_put(obj_request
);
2195 ceph_release_page_vector(pages
, page_count
);
2197 rbd_img_request_put(parent_request
);
2198 obj_request
->result
= result
;
2199 obj_request
->xferred
= 0;
2200 obj_request_done_set(obj_request
);
2205 static void rbd_img_obj_exists_callback(struct rbd_obj_request
*obj_request
)
2207 struct rbd_obj_request
*orig_request
;
2210 rbd_assert(!obj_request_img_data_test(obj_request
));
2213 * All we need from the object request is the original
2214 * request and the result of the STAT op. Grab those, then
2215 * we're done with the request.
2217 orig_request
= obj_request
->obj_request
;
2218 obj_request
->obj_request
= NULL
;
2219 rbd_assert(orig_request
);
2220 rbd_assert(orig_request
->img_request
);
2222 result
= obj_request
->result
;
2223 obj_request
->result
= 0;
2225 dout("%s: obj %p for obj %p result %d %llu/%llu\n", __func__
,
2226 obj_request
, orig_request
, result
,
2227 obj_request
->xferred
, obj_request
->length
);
2228 rbd_obj_request_put(obj_request
);
2230 rbd_assert(orig_request
);
2231 rbd_assert(orig_request
->img_request
);
2234 * Our only purpose here is to determine whether the object
2235 * exists, and we don't want to treat the non-existence as
2236 * an error. If something else comes back, transfer the
2237 * error to the original request and complete it now.
2240 obj_request_existence_set(orig_request
, true);
2241 } else if (result
== -ENOENT
) {
2242 obj_request_existence_set(orig_request
, false);
2243 } else if (result
) {
2244 orig_request
->result
= result
;
2249 * Resubmit the original request now that we have recorded
2250 * whether the target object exists.
2252 orig_request
->result
= rbd_img_obj_request_submit(orig_request
);
2254 if (orig_request
->result
)
2255 rbd_obj_request_complete(orig_request
);
2256 rbd_obj_request_put(orig_request
);
2259 static int rbd_img_obj_exists_submit(struct rbd_obj_request
*obj_request
)
2261 struct rbd_obj_request
*stat_request
;
2262 struct rbd_device
*rbd_dev
;
2263 struct ceph_osd_client
*osdc
;
2264 struct page
**pages
= NULL
;
2270 * The response data for a STAT call consists of:
2277 size
= sizeof (__le64
) + sizeof (__le32
) + sizeof (__le32
);
2278 page_count
= (u32
)calc_pages_for(0, size
);
2279 pages
= ceph_alloc_page_vector(page_count
, GFP_KERNEL
);
2281 return PTR_ERR(pages
);
2284 stat_request
= rbd_obj_request_create(obj_request
->object_name
, 0, 0,
2289 rbd_obj_request_get(obj_request
);
2290 stat_request
->obj_request
= obj_request
;
2291 stat_request
->pages
= pages
;
2292 stat_request
->page_count
= page_count
;
2294 rbd_assert(obj_request
->img_request
);
2295 rbd_dev
= obj_request
->img_request
->rbd_dev
;
2296 stat_request
->osd_req
= rbd_osd_req_create(rbd_dev
, false,
2298 if (!stat_request
->osd_req
)
2300 stat_request
->callback
= rbd_img_obj_exists_callback
;
2302 osd_req_op_init(stat_request
->osd_req
, 0, CEPH_OSD_OP_STAT
);
2303 osd_req_op_raw_data_in_pages(stat_request
->osd_req
, 0, pages
, size
, 0,
2305 rbd_osd_req_format_read(stat_request
);
2307 osdc
= &rbd_dev
->rbd_client
->client
->osdc
;
2308 ret
= rbd_obj_request_submit(osdc
, stat_request
);
2311 rbd_obj_request_put(obj_request
);
2316 static int rbd_img_obj_request_submit(struct rbd_obj_request
*obj_request
)
2318 struct rbd_img_request
*img_request
;
2319 struct rbd_device
*rbd_dev
;
2322 rbd_assert(obj_request_img_data_test(obj_request
));
2324 img_request
= obj_request
->img_request
;
2325 rbd_assert(img_request
);
2326 rbd_dev
= img_request
->rbd_dev
;
2329 * Only writes to layered images need special handling.
2330 * Reads and non-layered writes are simple object requests.
2331 * Layered writes that start beyond the end of the overlap
2332 * with the parent have no parent data, so they too are
2333 * simple object requests. Finally, if the target object is
2334 * known to already exist, its parent data has already been
2335 * copied, so a write to the object can also be handled as a
2336 * simple object request.
2338 if (!img_request_write_test(img_request
) ||
2339 !img_request_layered_test(img_request
) ||
2340 rbd_dev
->parent_overlap
<= obj_request
->img_offset
||
2341 ((known
= obj_request_known_test(obj_request
)) &&
2342 obj_request_exists_test(obj_request
))) {
2344 struct rbd_device
*rbd_dev
;
2345 struct ceph_osd_client
*osdc
;
2347 rbd_dev
= obj_request
->img_request
->rbd_dev
;
2348 osdc
= &rbd_dev
->rbd_client
->client
->osdc
;
2350 return rbd_obj_request_submit(osdc
, obj_request
);
2354 * It's a layered write. The target object might exist but
2355 * we may not know that yet. If we know it doesn't exist,
2356 * start by reading the data for the full target object from
2357 * the parent so we can use it for a copyup to the target.
2360 return rbd_img_obj_parent_read_full(obj_request
);
2362 /* We don't know whether the target exists. Go find out. */
2364 return rbd_img_obj_exists_submit(obj_request
);
2367 static int rbd_img_request_submit(struct rbd_img_request
*img_request
)
2369 struct rbd_obj_request
*obj_request
;
2370 struct rbd_obj_request
*next_obj_request
;
2372 dout("%s: img %p\n", __func__
, img_request
);
2373 for_each_obj_request_safe(img_request
, obj_request
, next_obj_request
) {
2376 ret
= rbd_img_obj_request_submit(obj_request
);
2384 static void rbd_img_parent_read_callback(struct rbd_img_request
*img_request
)
2386 struct rbd_obj_request
*obj_request
;
2387 struct rbd_device
*rbd_dev
;
2390 rbd_assert(img_request_child_test(img_request
));
2392 obj_request
= img_request
->obj_request
;
2393 rbd_assert(obj_request
);
2394 rbd_assert(obj_request
->img_request
);
2396 obj_request
->result
= img_request
->result
;
2397 if (obj_request
->result
)
2401 * We need to zero anything beyond the parent overlap
2402 * boundary. Since rbd_img_obj_request_read_callback()
2403 * will zero anything beyond the end of a short read, an
2404 * easy way to do this is to pretend the data from the
2405 * parent came up short--ending at the overlap boundary.
2407 rbd_assert(obj_request
->img_offset
< U64_MAX
- obj_request
->length
);
2408 obj_end
= obj_request
->img_offset
+ obj_request
->length
;
2409 rbd_dev
= obj_request
->img_request
->rbd_dev
;
2410 if (obj_end
> rbd_dev
->parent_overlap
) {
2413 if (obj_request
->img_offset
< rbd_dev
->parent_overlap
)
2414 xferred
= rbd_dev
->parent_overlap
-
2415 obj_request
->img_offset
;
2417 obj_request
->xferred
= min(img_request
->xferred
, xferred
);
2419 obj_request
->xferred
= img_request
->xferred
;
2422 rbd_img_obj_request_read_callback(obj_request
);
2423 rbd_obj_request_complete(obj_request
);
2426 static void rbd_img_parent_read(struct rbd_obj_request
*obj_request
)
2428 struct rbd_device
*rbd_dev
;
2429 struct rbd_img_request
*img_request
;
2432 rbd_assert(obj_request_img_data_test(obj_request
));
2433 rbd_assert(obj_request
->img_request
!= NULL
);
2434 rbd_assert(obj_request
->result
== (s32
) -ENOENT
);
2435 rbd_assert(obj_request
->type
== OBJ_REQUEST_BIO
);
2437 rbd_dev
= obj_request
->img_request
->rbd_dev
;
2438 rbd_assert(rbd_dev
->parent
!= NULL
);
2439 /* rbd_read_finish(obj_request, obj_request->length); */
2440 img_request
= rbd_img_request_create(rbd_dev
->parent
,
2441 obj_request
->img_offset
,
2442 obj_request
->length
,
2448 rbd_obj_request_get(obj_request
);
2449 img_request
->obj_request
= obj_request
;
2451 result
= rbd_img_request_fill(img_request
, OBJ_REQUEST_BIO
,
2452 obj_request
->bio_list
);
2456 img_request
->callback
= rbd_img_parent_read_callback
;
2457 result
= rbd_img_request_submit(img_request
);
2464 rbd_img_request_put(img_request
);
2465 obj_request
->result
= result
;
2466 obj_request
->xferred
= 0;
2467 obj_request_done_set(obj_request
);
2470 static int rbd_obj_notify_ack(struct rbd_device
*rbd_dev
, u64 notify_id
)
2472 struct rbd_obj_request
*obj_request
;
2473 struct ceph_osd_client
*osdc
= &rbd_dev
->rbd_client
->client
->osdc
;
2476 obj_request
= rbd_obj_request_create(rbd_dev
->header_name
, 0, 0,
2477 OBJ_REQUEST_NODATA
);
2482 obj_request
->osd_req
= rbd_osd_req_create(rbd_dev
, false, obj_request
);
2483 if (!obj_request
->osd_req
)
2485 obj_request
->callback
= rbd_obj_request_put
;
2487 osd_req_op_watch_init(obj_request
->osd_req
, 0, CEPH_OSD_OP_NOTIFY_ACK
,
2489 rbd_osd_req_format_read(obj_request
);
2491 ret
= rbd_obj_request_submit(osdc
, obj_request
);
2494 rbd_obj_request_put(obj_request
);
2499 static void rbd_watch_cb(u64 ver
, u64 notify_id
, u8 opcode
, void *data
)
2501 struct rbd_device
*rbd_dev
= (struct rbd_device
*)data
;
2506 dout("%s: \"%s\" notify_id %llu opcode %u\n", __func__
,
2507 rbd_dev
->header_name
, (unsigned long long)notify_id
,
2508 (unsigned int)opcode
);
2509 (void)rbd_dev_refresh(rbd_dev
);
2511 rbd_obj_notify_ack(rbd_dev
, notify_id
);
2515 * Request sync osd watch/unwatch. The value of "start" determines
2516 * whether a watch request is being initiated or torn down.
2518 static int rbd_dev_header_watch_sync(struct rbd_device
*rbd_dev
, int start
)
2520 struct ceph_osd_client
*osdc
= &rbd_dev
->rbd_client
->client
->osdc
;
2521 struct rbd_obj_request
*obj_request
;
2524 rbd_assert(start
^ !!rbd_dev
->watch_event
);
2525 rbd_assert(start
^ !!rbd_dev
->watch_request
);
2528 ret
= ceph_osdc_create_event(osdc
, rbd_watch_cb
, rbd_dev
,
2529 &rbd_dev
->watch_event
);
2532 rbd_assert(rbd_dev
->watch_event
!= NULL
);
2536 obj_request
= rbd_obj_request_create(rbd_dev
->header_name
, 0, 0,
2537 OBJ_REQUEST_NODATA
);
2541 obj_request
->osd_req
= rbd_osd_req_create(rbd_dev
, true, obj_request
);
2542 if (!obj_request
->osd_req
)
2546 ceph_osdc_set_request_linger(osdc
, obj_request
->osd_req
);
2548 ceph_osdc_unregister_linger_request(osdc
,
2549 rbd_dev
->watch_request
->osd_req
);
2551 osd_req_op_watch_init(obj_request
->osd_req
, 0, CEPH_OSD_OP_WATCH
,
2552 rbd_dev
->watch_event
->cookie
, 0, start
);
2553 rbd_osd_req_format_write(obj_request
);
2555 ret
= rbd_obj_request_submit(osdc
, obj_request
);
2558 ret
= rbd_obj_request_wait(obj_request
);
2561 ret
= obj_request
->result
;
2566 * A watch request is set to linger, so the underlying osd
2567 * request won't go away until we unregister it. We retain
2568 * a pointer to the object request during that time (in
2569 * rbd_dev->watch_request), so we'll keep a reference to
2570 * it. We'll drop that reference (below) after we've
2574 rbd_dev
->watch_request
= obj_request
;
2579 /* We have successfully torn down the watch request */
2581 rbd_obj_request_put(rbd_dev
->watch_request
);
2582 rbd_dev
->watch_request
= NULL
;
2584 /* Cancel the event if we're tearing down, or on error */
2585 ceph_osdc_cancel_event(rbd_dev
->watch_event
);
2586 rbd_dev
->watch_event
= NULL
;
2588 rbd_obj_request_put(obj_request
);
2594 * Synchronous osd object method call. Returns the number of bytes
2595 * returned in the outbound buffer, or a negative error code.
2597 static int rbd_obj_method_sync(struct rbd_device
*rbd_dev
,
2598 const char *object_name
,
2599 const char *class_name
,
2600 const char *method_name
,
2601 const void *outbound
,
2602 size_t outbound_size
,
2604 size_t inbound_size
)
2606 struct ceph_osd_client
*osdc
= &rbd_dev
->rbd_client
->client
->osdc
;
2607 struct rbd_obj_request
*obj_request
;
2608 struct page
**pages
;
2613 * Method calls are ultimately read operations. The result
2614 * should placed into the inbound buffer provided. They
2615 * also supply outbound data--parameters for the object
2616 * method. Currently if this is present it will be a
2619 page_count
= (u32
)calc_pages_for(0, inbound_size
);
2620 pages
= ceph_alloc_page_vector(page_count
, GFP_KERNEL
);
2622 return PTR_ERR(pages
);
2625 obj_request
= rbd_obj_request_create(object_name
, 0, inbound_size
,
2630 obj_request
->pages
= pages
;
2631 obj_request
->page_count
= page_count
;
2633 obj_request
->osd_req
= rbd_osd_req_create(rbd_dev
, false, obj_request
);
2634 if (!obj_request
->osd_req
)
2637 osd_req_op_cls_init(obj_request
->osd_req
, 0, CEPH_OSD_OP_CALL
,
2638 class_name
, method_name
);
2639 if (outbound_size
) {
2640 struct ceph_pagelist
*pagelist
;
2642 pagelist
= kmalloc(sizeof (*pagelist
), GFP_NOFS
);
2646 ceph_pagelist_init(pagelist
);
2647 ceph_pagelist_append(pagelist
, outbound
, outbound_size
);
2648 osd_req_op_cls_request_data_pagelist(obj_request
->osd_req
, 0,
2651 osd_req_op_cls_response_data_pages(obj_request
->osd_req
, 0,
2652 obj_request
->pages
, inbound_size
,
2654 rbd_osd_req_format_read(obj_request
);
2656 ret
= rbd_obj_request_submit(osdc
, obj_request
);
2659 ret
= rbd_obj_request_wait(obj_request
);
2663 ret
= obj_request
->result
;
2667 rbd_assert(obj_request
->xferred
< (u64
)INT_MAX
);
2668 ret
= (int)obj_request
->xferred
;
2669 ceph_copy_from_page_vector(pages
, inbound
, 0, obj_request
->xferred
);
2672 rbd_obj_request_put(obj_request
);
2674 ceph_release_page_vector(pages
, page_count
);
2679 static void rbd_request_fn(struct request_queue
*q
)
2680 __releases(q
->queue_lock
) __acquires(q
->queue_lock
)
2682 struct rbd_device
*rbd_dev
= q
->queuedata
;
2683 bool read_only
= rbd_dev
->mapping
.read_only
;
2687 while ((rq
= blk_fetch_request(q
))) {
2688 bool write_request
= rq_data_dir(rq
) == WRITE
;
2689 struct rbd_img_request
*img_request
;
2693 /* Ignore any non-FS requests that filter through. */
2695 if (rq
->cmd_type
!= REQ_TYPE_FS
) {
2696 dout("%s: non-fs request type %d\n", __func__
,
2697 (int) rq
->cmd_type
);
2698 __blk_end_request_all(rq
, 0);
2702 /* Ignore/skip any zero-length requests */
2704 offset
= (u64
) blk_rq_pos(rq
) << SECTOR_SHIFT
;
2705 length
= (u64
) blk_rq_bytes(rq
);
2708 dout("%s: zero-length request\n", __func__
);
2709 __blk_end_request_all(rq
, 0);
2713 spin_unlock_irq(q
->queue_lock
);
2715 /* Disallow writes to a read-only device */
2717 if (write_request
) {
2721 rbd_assert(rbd_dev
->spec
->snap_id
== CEPH_NOSNAP
);
2725 * Quit early if the mapped snapshot no longer
2726 * exists. It's still possible the snapshot will
2727 * have disappeared by the time our request arrives
2728 * at the osd, but there's no sense in sending it if
2731 if (!test_bit(RBD_DEV_FLAG_EXISTS
, &rbd_dev
->flags
)) {
2732 dout("request for non-existent snapshot");
2733 rbd_assert(rbd_dev
->spec
->snap_id
!= CEPH_NOSNAP
);
2739 if (offset
&& length
> U64_MAX
- offset
+ 1) {
2740 rbd_warn(rbd_dev
, "bad request range (%llu~%llu)\n",
2742 goto end_request
; /* Shouldn't happen */
2746 img_request
= rbd_img_request_create(rbd_dev
, offset
, length
,
2747 write_request
, false);
2751 img_request
->rq
= rq
;
2753 result
= rbd_img_request_fill(img_request
, OBJ_REQUEST_BIO
,
2756 result
= rbd_img_request_submit(img_request
);
2758 rbd_img_request_put(img_request
);
2760 spin_lock_irq(q
->queue_lock
);
2762 rbd_warn(rbd_dev
, "%s %llx at %llx result %d\n",
2763 write_request
? "write" : "read",
2764 length
, offset
, result
);
2766 __blk_end_request_all(rq
, result
);
2772 * a queue callback. Makes sure that we don't create a bio that spans across
2773 * multiple osd objects. One exception would be with a single page bios,
2774 * which we handle later at bio_chain_clone_range()
2776 static int rbd_merge_bvec(struct request_queue
*q
, struct bvec_merge_data
*bmd
,
2777 struct bio_vec
*bvec
)
2779 struct rbd_device
*rbd_dev
= q
->queuedata
;
2780 sector_t sector_offset
;
2781 sector_t sectors_per_obj
;
2782 sector_t obj_sector_offset
;
2786 * Find how far into its rbd object the partition-relative
2787 * bio start sector is to offset relative to the enclosing
2790 sector_offset
= get_start_sect(bmd
->bi_bdev
) + bmd
->bi_sector
;
2791 sectors_per_obj
= 1 << (rbd_dev
->header
.obj_order
- SECTOR_SHIFT
);
2792 obj_sector_offset
= sector_offset
& (sectors_per_obj
- 1);
2795 * Compute the number of bytes from that offset to the end
2796 * of the object. Account for what's already used by the bio.
2798 ret
= (int) (sectors_per_obj
- obj_sector_offset
) << SECTOR_SHIFT
;
2799 if (ret
> bmd
->bi_size
)
2800 ret
-= bmd
->bi_size
;
2805 * Don't send back more than was asked for. And if the bio
2806 * was empty, let the whole thing through because: "Note
2807 * that a block device *must* allow a single page to be
2808 * added to an empty bio."
2810 rbd_assert(bvec
->bv_len
<= PAGE_SIZE
);
2811 if (ret
> (int) bvec
->bv_len
|| !bmd
->bi_size
)
2812 ret
= (int) bvec
->bv_len
;
2817 static void rbd_free_disk(struct rbd_device
*rbd_dev
)
2819 struct gendisk
*disk
= rbd_dev
->disk
;
2824 rbd_dev
->disk
= NULL
;
2825 if (disk
->flags
& GENHD_FL_UP
) {
2828 blk_cleanup_queue(disk
->queue
);
2833 static int rbd_obj_read_sync(struct rbd_device
*rbd_dev
,
2834 const char *object_name
,
2835 u64 offset
, u64 length
, void *buf
)
2838 struct ceph_osd_client
*osdc
= &rbd_dev
->rbd_client
->client
->osdc
;
2839 struct rbd_obj_request
*obj_request
;
2840 struct page
**pages
= NULL
;
2845 page_count
= (u32
) calc_pages_for(offset
, length
);
2846 pages
= ceph_alloc_page_vector(page_count
, GFP_KERNEL
);
2848 ret
= PTR_ERR(pages
);
2851 obj_request
= rbd_obj_request_create(object_name
, offset
, length
,
2856 obj_request
->pages
= pages
;
2857 obj_request
->page_count
= page_count
;
2859 obj_request
->osd_req
= rbd_osd_req_create(rbd_dev
, false, obj_request
);
2860 if (!obj_request
->osd_req
)
2863 osd_req_op_extent_init(obj_request
->osd_req
, 0, CEPH_OSD_OP_READ
,
2864 offset
, length
, 0, 0);
2865 osd_req_op_extent_osd_data_pages(obj_request
->osd_req
, 0,
2867 obj_request
->length
,
2868 obj_request
->offset
& ~PAGE_MASK
,
2870 rbd_osd_req_format_read(obj_request
);
2872 ret
= rbd_obj_request_submit(osdc
, obj_request
);
2875 ret
= rbd_obj_request_wait(obj_request
);
2879 ret
= obj_request
->result
;
2883 rbd_assert(obj_request
->xferred
<= (u64
) SIZE_MAX
);
2884 size
= (size_t) obj_request
->xferred
;
2885 ceph_copy_from_page_vector(pages
, buf
, 0, size
);
2886 rbd_assert(size
<= (size_t)INT_MAX
);
2890 rbd_obj_request_put(obj_request
);
2892 ceph_release_page_vector(pages
, page_count
);
2898 * Read the complete header for the given rbd device.
2900 * Returns a pointer to a dynamically-allocated buffer containing
2901 * the complete and validated header. Caller can pass the address
2902 * of a variable that will be filled in with the version of the
2903 * header object at the time it was read.
2905 * Returns a pointer-coded errno if a failure occurs.
2907 static struct rbd_image_header_ondisk
*
2908 rbd_dev_v1_header_read(struct rbd_device
*rbd_dev
)
2910 struct rbd_image_header_ondisk
*ondisk
= NULL
;
2917 * The complete header will include an array of its 64-bit
2918 * snapshot ids, followed by the names of those snapshots as
2919 * a contiguous block of NUL-terminated strings. Note that
2920 * the number of snapshots could change by the time we read
2921 * it in, in which case we re-read it.
2928 size
= sizeof (*ondisk
);
2929 size
+= snap_count
* sizeof (struct rbd_image_snap_ondisk
);
2931 ondisk
= kmalloc(size
, GFP_KERNEL
);
2933 return ERR_PTR(-ENOMEM
);
2935 ret
= rbd_obj_read_sync(rbd_dev
, rbd_dev
->header_name
,
2939 if ((size_t)ret
< size
) {
2941 rbd_warn(rbd_dev
, "short header read (want %zd got %d)",
2945 if (!rbd_dev_ondisk_valid(ondisk
)) {
2947 rbd_warn(rbd_dev
, "invalid header");
2951 names_size
= le64_to_cpu(ondisk
->snap_names_len
);
2952 want_count
= snap_count
;
2953 snap_count
= le32_to_cpu(ondisk
->snap_count
);
2954 } while (snap_count
!= want_count
);
2961 return ERR_PTR(ret
);
2965 * reload the ondisk the header
2967 static int rbd_read_header(struct rbd_device
*rbd_dev
,
2968 struct rbd_image_header
*header
)
2970 struct rbd_image_header_ondisk
*ondisk
;
2973 ondisk
= rbd_dev_v1_header_read(rbd_dev
);
2975 return PTR_ERR(ondisk
);
2976 ret
= rbd_header_from_disk(header
, ondisk
);
2982 static void rbd_remove_all_snaps(struct rbd_device
*rbd_dev
)
2984 struct rbd_snap
*snap
;
2985 struct rbd_snap
*next
;
2987 list_for_each_entry_safe(snap
, next
, &rbd_dev
->snaps
, node
) {
2988 list_del(&snap
->node
);
2989 rbd_snap_destroy(snap
);
2993 static void rbd_update_mapping_size(struct rbd_device
*rbd_dev
)
2995 if (rbd_dev
->spec
->snap_id
!= CEPH_NOSNAP
)
2998 if (rbd_dev
->mapping
.size
!= rbd_dev
->header
.image_size
) {
3001 rbd_dev
->mapping
.size
= rbd_dev
->header
.image_size
;
3002 size
= (sector_t
)rbd_dev
->mapping
.size
/ SECTOR_SIZE
;
3003 dout("setting size to %llu sectors", (unsigned long long)size
);
3004 set_capacity(rbd_dev
->disk
, size
);
3009 * only read the first part of the ondisk header, without the snaps info
3011 static int rbd_dev_v1_refresh(struct rbd_device
*rbd_dev
)
3014 struct rbd_image_header h
;
3016 ret
= rbd_read_header(rbd_dev
, &h
);
3020 down_write(&rbd_dev
->header_rwsem
);
3022 /* Update image size, and check for resize of mapped image */
3023 rbd_dev
->header
.image_size
= h
.image_size
;
3024 rbd_update_mapping_size(rbd_dev
);
3026 /* rbd_dev->header.object_prefix shouldn't change */
3027 kfree(rbd_dev
->header
.snap_sizes
);
3028 kfree(rbd_dev
->header
.snap_names
);
3029 /* osd requests may still refer to snapc */
3030 ceph_put_snap_context(rbd_dev
->header
.snapc
);
3032 rbd_dev
->header
.image_size
= h
.image_size
;
3033 rbd_dev
->header
.snapc
= h
.snapc
;
3034 rbd_dev
->header
.snap_names
= h
.snap_names
;
3035 rbd_dev
->header
.snap_sizes
= h
.snap_sizes
;
3036 /* Free the extra copy of the object prefix */
3037 if (strcmp(rbd_dev
->header
.object_prefix
, h
.object_prefix
))
3038 rbd_warn(rbd_dev
, "object prefix changed (ignoring)");
3039 kfree(h
.object_prefix
);
3041 ret
= rbd_dev_snaps_update(rbd_dev
);
3043 up_write(&rbd_dev
->header_rwsem
);
3048 static int rbd_dev_refresh(struct rbd_device
*rbd_dev
)
3053 rbd_assert(rbd_image_format_valid(rbd_dev
->image_format
));
3054 image_size
= rbd_dev
->header
.image_size
;
3055 mutex_lock_nested(&ctl_mutex
, SINGLE_DEPTH_NESTING
);
3056 if (rbd_dev
->image_format
== 1)
3057 ret
= rbd_dev_v1_refresh(rbd_dev
);
3059 ret
= rbd_dev_v2_refresh(rbd_dev
);
3060 mutex_unlock(&ctl_mutex
);
3062 rbd_warn(rbd_dev
, "got notification but failed to "
3063 " update snaps: %d\n", ret
);
3064 if (image_size
!= rbd_dev
->header
.image_size
)
3065 revalidate_disk(rbd_dev
->disk
);
3070 static int rbd_init_disk(struct rbd_device
*rbd_dev
)
3072 struct gendisk
*disk
;
3073 struct request_queue
*q
;
3076 /* create gendisk info */
3077 disk
= alloc_disk(RBD_MINORS_PER_MAJOR
);
3081 snprintf(disk
->disk_name
, sizeof(disk
->disk_name
), RBD_DRV_NAME
"%d",
3083 disk
->major
= rbd_dev
->major
;
3084 disk
->first_minor
= 0;
3085 disk
->fops
= &rbd_bd_ops
;
3086 disk
->private_data
= rbd_dev
;
3088 q
= blk_init_queue(rbd_request_fn
, &rbd_dev
->lock
);
3092 /* We use the default size, but let's be explicit about it. */
3093 blk_queue_physical_block_size(q
, SECTOR_SIZE
);
3095 /* set io sizes to object size */
3096 segment_size
= rbd_obj_bytes(&rbd_dev
->header
);
3097 blk_queue_max_hw_sectors(q
, segment_size
/ SECTOR_SIZE
);
3098 blk_queue_max_segment_size(q
, segment_size
);
3099 blk_queue_io_min(q
, segment_size
);
3100 blk_queue_io_opt(q
, segment_size
);
3102 blk_queue_merge_bvec(q
, rbd_merge_bvec
);
3105 q
->queuedata
= rbd_dev
;
3107 rbd_dev
->disk
= disk
;
3120 static struct rbd_device
*dev_to_rbd_dev(struct device
*dev
)
3122 return container_of(dev
, struct rbd_device
, dev
);
3125 static ssize_t
rbd_size_show(struct device
*dev
,
3126 struct device_attribute
*attr
, char *buf
)
3128 struct rbd_device
*rbd_dev
= dev_to_rbd_dev(dev
);
3130 return sprintf(buf
, "%llu\n",
3131 (unsigned long long)rbd_dev
->mapping
.size
);
3135 * Note this shows the features for whatever's mapped, which is not
3136 * necessarily the base image.
3138 static ssize_t
rbd_features_show(struct device
*dev
,
3139 struct device_attribute
*attr
, char *buf
)
3141 struct rbd_device
*rbd_dev
= dev_to_rbd_dev(dev
);
3143 return sprintf(buf
, "0x%016llx\n",
3144 (unsigned long long)rbd_dev
->mapping
.features
);
3147 static ssize_t
rbd_major_show(struct device
*dev
,
3148 struct device_attribute
*attr
, char *buf
)
3150 struct rbd_device
*rbd_dev
= dev_to_rbd_dev(dev
);
3153 return sprintf(buf
, "%d\n", rbd_dev
->major
);
3155 return sprintf(buf
, "(none)\n");
3159 static ssize_t
rbd_client_id_show(struct device
*dev
,
3160 struct device_attribute
*attr
, char *buf
)
3162 struct rbd_device
*rbd_dev
= dev_to_rbd_dev(dev
);
3164 return sprintf(buf
, "client%lld\n",
3165 ceph_client_id(rbd_dev
->rbd_client
->client
));
3168 static ssize_t
rbd_pool_show(struct device
*dev
,
3169 struct device_attribute
*attr
, char *buf
)
3171 struct rbd_device
*rbd_dev
= dev_to_rbd_dev(dev
);
3173 return sprintf(buf
, "%s\n", rbd_dev
->spec
->pool_name
);
3176 static ssize_t
rbd_pool_id_show(struct device
*dev
,
3177 struct device_attribute
*attr
, char *buf
)
3179 struct rbd_device
*rbd_dev
= dev_to_rbd_dev(dev
);
3181 return sprintf(buf
, "%llu\n",
3182 (unsigned long long) rbd_dev
->spec
->pool_id
);
3185 static ssize_t
rbd_name_show(struct device
*dev
,
3186 struct device_attribute
*attr
, char *buf
)
3188 struct rbd_device
*rbd_dev
= dev_to_rbd_dev(dev
);
3190 if (rbd_dev
->spec
->image_name
)
3191 return sprintf(buf
, "%s\n", rbd_dev
->spec
->image_name
);
3193 return sprintf(buf
, "(unknown)\n");
3196 static ssize_t
rbd_image_id_show(struct device
*dev
,
3197 struct device_attribute
*attr
, char *buf
)
3199 struct rbd_device
*rbd_dev
= dev_to_rbd_dev(dev
);
3201 return sprintf(buf
, "%s\n", rbd_dev
->spec
->image_id
);
3205 * Shows the name of the currently-mapped snapshot (or
3206 * RBD_SNAP_HEAD_NAME for the base image).
3208 static ssize_t
rbd_snap_show(struct device
*dev
,
3209 struct device_attribute
*attr
,
3212 struct rbd_device
*rbd_dev
= dev_to_rbd_dev(dev
);
3214 return sprintf(buf
, "%s\n", rbd_dev
->spec
->snap_name
);
3218 * For an rbd v2 image, shows the pool id, image id, and snapshot id
3219 * for the parent image. If there is no parent, simply shows
3220 * "(no parent image)".
3222 static ssize_t
rbd_parent_show(struct device
*dev
,
3223 struct device_attribute
*attr
,
3226 struct rbd_device
*rbd_dev
= dev_to_rbd_dev(dev
);
3227 struct rbd_spec
*spec
= rbd_dev
->parent_spec
;
3232 return sprintf(buf
, "(no parent image)\n");
3234 count
= sprintf(bufp
, "pool_id %llu\npool_name %s\n",
3235 (unsigned long long) spec
->pool_id
, spec
->pool_name
);
3240 count
= sprintf(bufp
, "image_id %s\nimage_name %s\n", spec
->image_id
,
3241 spec
->image_name
? spec
->image_name
: "(unknown)");
3246 count
= sprintf(bufp
, "snap_id %llu\nsnap_name %s\n",
3247 (unsigned long long) spec
->snap_id
, spec
->snap_name
);
3252 count
= sprintf(bufp
, "overlap %llu\n", rbd_dev
->parent_overlap
);
3257 return (ssize_t
) (bufp
- buf
);
3260 static ssize_t
rbd_image_refresh(struct device
*dev
,
3261 struct device_attribute
*attr
,
3265 struct rbd_device
*rbd_dev
= dev_to_rbd_dev(dev
);
3268 ret
= rbd_dev_refresh(rbd_dev
);
3270 return ret
< 0 ? ret
: size
;
3273 static DEVICE_ATTR(size
, S_IRUGO
, rbd_size_show
, NULL
);
3274 static DEVICE_ATTR(features
, S_IRUGO
, rbd_features_show
, NULL
);
3275 static DEVICE_ATTR(major
, S_IRUGO
, rbd_major_show
, NULL
);
3276 static DEVICE_ATTR(client_id
, S_IRUGO
, rbd_client_id_show
, NULL
);
3277 static DEVICE_ATTR(pool
, S_IRUGO
, rbd_pool_show
, NULL
);
3278 static DEVICE_ATTR(pool_id
, S_IRUGO
, rbd_pool_id_show
, NULL
);
3279 static DEVICE_ATTR(name
, S_IRUGO
, rbd_name_show
, NULL
);
3280 static DEVICE_ATTR(image_id
, S_IRUGO
, rbd_image_id_show
, NULL
);
3281 static DEVICE_ATTR(refresh
, S_IWUSR
, NULL
, rbd_image_refresh
);
3282 static DEVICE_ATTR(current_snap
, S_IRUGO
, rbd_snap_show
, NULL
);
3283 static DEVICE_ATTR(parent
, S_IRUGO
, rbd_parent_show
, NULL
);
3285 static struct attribute
*rbd_attrs
[] = {
3286 &dev_attr_size
.attr
,
3287 &dev_attr_features
.attr
,
3288 &dev_attr_major
.attr
,
3289 &dev_attr_client_id
.attr
,
3290 &dev_attr_pool
.attr
,
3291 &dev_attr_pool_id
.attr
,
3292 &dev_attr_name
.attr
,
3293 &dev_attr_image_id
.attr
,
3294 &dev_attr_current_snap
.attr
,
3295 &dev_attr_parent
.attr
,
3296 &dev_attr_refresh
.attr
,
3300 static struct attribute_group rbd_attr_group
= {
3304 static const struct attribute_group
*rbd_attr_groups
[] = {
3309 static void rbd_sysfs_dev_release(struct device
*dev
)
3313 static struct device_type rbd_device_type
= {
3315 .groups
= rbd_attr_groups
,
3316 .release
= rbd_sysfs_dev_release
,
3319 static struct rbd_spec
*rbd_spec_get(struct rbd_spec
*spec
)
3321 kref_get(&spec
->kref
);
3326 static void rbd_spec_free(struct kref
*kref
);
3327 static void rbd_spec_put(struct rbd_spec
*spec
)
3330 kref_put(&spec
->kref
, rbd_spec_free
);
3333 static struct rbd_spec
*rbd_spec_alloc(void)
3335 struct rbd_spec
*spec
;
3337 spec
= kzalloc(sizeof (*spec
), GFP_KERNEL
);
3340 kref_init(&spec
->kref
);
3345 static void rbd_spec_free(struct kref
*kref
)
3347 struct rbd_spec
*spec
= container_of(kref
, struct rbd_spec
, kref
);
3349 kfree(spec
->pool_name
);
3350 kfree(spec
->image_id
);
3351 kfree(spec
->image_name
);
3352 kfree(spec
->snap_name
);
3356 static struct rbd_device
*rbd_dev_create(struct rbd_client
*rbdc
,
3357 struct rbd_spec
*spec
)
3359 struct rbd_device
*rbd_dev
;
3361 rbd_dev
= kzalloc(sizeof (*rbd_dev
), GFP_KERNEL
);
3365 spin_lock_init(&rbd_dev
->lock
);
3367 INIT_LIST_HEAD(&rbd_dev
->node
);
3368 INIT_LIST_HEAD(&rbd_dev
->snaps
);
3369 init_rwsem(&rbd_dev
->header_rwsem
);
3371 rbd_dev
->spec
= spec
;
3372 rbd_dev
->rbd_client
= rbdc
;
3374 /* Initialize the layout used for all rbd requests */
3376 rbd_dev
->layout
.fl_stripe_unit
= cpu_to_le32(1 << RBD_MAX_OBJ_ORDER
);
3377 rbd_dev
->layout
.fl_stripe_count
= cpu_to_le32(1);
3378 rbd_dev
->layout
.fl_object_size
= cpu_to_le32(1 << RBD_MAX_OBJ_ORDER
);
3379 rbd_dev
->layout
.fl_pg_pool
= cpu_to_le32((u32
) spec
->pool_id
);
3384 static void rbd_dev_destroy(struct rbd_device
*rbd_dev
)
3386 rbd_put_client(rbd_dev
->rbd_client
);
3387 rbd_spec_put(rbd_dev
->spec
);
3391 static void rbd_snap_destroy(struct rbd_snap
*snap
)
3397 static struct rbd_snap
*rbd_snap_create(struct rbd_device
*rbd_dev
,
3398 const char *snap_name
,
3399 u64 snap_id
, u64 snap_size
,
3402 struct rbd_snap
*snap
;
3404 snap
= kzalloc(sizeof (*snap
), GFP_KERNEL
);
3406 return ERR_PTR(-ENOMEM
);
3408 snap
->name
= snap_name
;
3410 snap
->size
= snap_size
;
3411 snap
->features
= snap_features
;
3417 * Returns a dynamically-allocated snapshot name if successful, or a
3418 * pointer-coded error otherwise.
3420 static const char *rbd_dev_v1_snap_info(struct rbd_device
*rbd_dev
, u32 which
,
3421 u64
*snap_size
, u64
*snap_features
)
3423 const char *snap_name
;
3426 rbd_assert(which
< rbd_dev
->header
.snapc
->num_snaps
);
3428 /* Skip over names until we find the one we are looking for */
3430 snap_name
= rbd_dev
->header
.snap_names
;
3431 for (i
= 0; i
< which
; i
++)
3432 snap_name
+= strlen(snap_name
) + 1;
3434 snap_name
= kstrdup(snap_name
, GFP_KERNEL
);
3436 return ERR_PTR(-ENOMEM
);
3438 *snap_size
= rbd_dev
->header
.snap_sizes
[which
];
3439 *snap_features
= 0; /* No features for v1 */
3445 * Get the size and object order for an image snapshot, or if
3446 * snap_id is CEPH_NOSNAP, gets this information for the base
3449 static int _rbd_dev_v2_snap_size(struct rbd_device
*rbd_dev
, u64 snap_id
,
3450 u8
*order
, u64
*snap_size
)
3452 __le64 snapid
= cpu_to_le64(snap_id
);
3457 } __attribute__ ((packed
)) size_buf
= { 0 };
3459 ret
= rbd_obj_method_sync(rbd_dev
, rbd_dev
->header_name
,
3461 &snapid
, sizeof (snapid
),
3462 &size_buf
, sizeof (size_buf
));
3463 dout("%s: rbd_obj_method_sync returned %d\n", __func__
, ret
);
3466 if (ret
< sizeof (size_buf
))
3470 *order
= size_buf
.order
;
3471 *snap_size
= le64_to_cpu(size_buf
.size
);
3473 dout(" snap_id 0x%016llx order = %u, snap_size = %llu\n",
3474 (unsigned long long)snap_id
, (unsigned int)*order
,
3475 (unsigned long long)*snap_size
);
3480 static int rbd_dev_v2_image_size(struct rbd_device
*rbd_dev
)
3482 return _rbd_dev_v2_snap_size(rbd_dev
, CEPH_NOSNAP
,
3483 &rbd_dev
->header
.obj_order
,
3484 &rbd_dev
->header
.image_size
);
3487 static int rbd_dev_v2_object_prefix(struct rbd_device
*rbd_dev
)
3493 reply_buf
= kzalloc(RBD_OBJ_PREFIX_LEN_MAX
, GFP_KERNEL
);
3497 ret
= rbd_obj_method_sync(rbd_dev
, rbd_dev
->header_name
,
3498 "rbd", "get_object_prefix", NULL
, 0,
3499 reply_buf
, RBD_OBJ_PREFIX_LEN_MAX
);
3500 dout("%s: rbd_obj_method_sync returned %d\n", __func__
, ret
);
3505 rbd_dev
->header
.object_prefix
= ceph_extract_encoded_string(&p
,
3506 p
+ ret
, NULL
, GFP_NOIO
);
3509 if (IS_ERR(rbd_dev
->header
.object_prefix
)) {
3510 ret
= PTR_ERR(rbd_dev
->header
.object_prefix
);
3511 rbd_dev
->header
.object_prefix
= NULL
;
3513 dout(" object_prefix = %s\n", rbd_dev
->header
.object_prefix
);
3521 static int _rbd_dev_v2_snap_features(struct rbd_device
*rbd_dev
, u64 snap_id
,
3524 __le64 snapid
= cpu_to_le64(snap_id
);
3528 } __attribute__ ((packed
)) features_buf
= { 0 };
3532 ret
= rbd_obj_method_sync(rbd_dev
, rbd_dev
->header_name
,
3533 "rbd", "get_features",
3534 &snapid
, sizeof (snapid
),
3535 &features_buf
, sizeof (features_buf
));
3536 dout("%s: rbd_obj_method_sync returned %d\n", __func__
, ret
);
3539 if (ret
< sizeof (features_buf
))
3542 incompat
= le64_to_cpu(features_buf
.incompat
);
3543 if (incompat
& ~RBD_FEATURES_SUPPORTED
)
3546 *snap_features
= le64_to_cpu(features_buf
.features
);
3548 dout(" snap_id 0x%016llx features = 0x%016llx incompat = 0x%016llx\n",
3549 (unsigned long long)snap_id
,
3550 (unsigned long long)*snap_features
,
3551 (unsigned long long)le64_to_cpu(features_buf
.incompat
));
3556 static int rbd_dev_v2_features(struct rbd_device
*rbd_dev
)
3558 return _rbd_dev_v2_snap_features(rbd_dev
, CEPH_NOSNAP
,
3559 &rbd_dev
->header
.features
);
3562 static int rbd_dev_v2_parent_info(struct rbd_device
*rbd_dev
)
3564 struct rbd_spec
*parent_spec
;
3566 void *reply_buf
= NULL
;
3574 parent_spec
= rbd_spec_alloc();
3578 size
= sizeof (__le64
) + /* pool_id */
3579 sizeof (__le32
) + RBD_IMAGE_ID_LEN_MAX
+ /* image_id */
3580 sizeof (__le64
) + /* snap_id */
3581 sizeof (__le64
); /* overlap */
3582 reply_buf
= kmalloc(size
, GFP_KERNEL
);
3588 snapid
= cpu_to_le64(CEPH_NOSNAP
);
3589 ret
= rbd_obj_method_sync(rbd_dev
, rbd_dev
->header_name
,
3590 "rbd", "get_parent",
3591 &snapid
, sizeof (snapid
),
3593 dout("%s: rbd_obj_method_sync returned %d\n", __func__
, ret
);
3598 end
= reply_buf
+ ret
;
3600 ceph_decode_64_safe(&p
, end
, parent_spec
->pool_id
, out_err
);
3601 if (parent_spec
->pool_id
== CEPH_NOPOOL
)
3602 goto out
; /* No parent? No problem. */
3604 /* The ceph file layout needs to fit pool id in 32 bits */
3607 if (parent_spec
->pool_id
> (u64
)U32_MAX
) {
3608 rbd_warn(NULL
, "parent pool id too large (%llu > %u)\n",
3609 (unsigned long long)parent_spec
->pool_id
, U32_MAX
);
3613 image_id
= ceph_extract_encoded_string(&p
, end
, NULL
, GFP_KERNEL
);
3614 if (IS_ERR(image_id
)) {
3615 ret
= PTR_ERR(image_id
);
3618 parent_spec
->image_id
= image_id
;
3619 ceph_decode_64_safe(&p
, end
, parent_spec
->snap_id
, out_err
);
3620 ceph_decode_64_safe(&p
, end
, overlap
, out_err
);
3622 rbd_dev
->parent_overlap
= overlap
;
3623 rbd_dev
->parent_spec
= parent_spec
;
3624 parent_spec
= NULL
; /* rbd_dev now owns this */
3629 rbd_spec_put(parent_spec
);
3634 static int rbd_dev_v2_striping_info(struct rbd_device
*rbd_dev
)
3638 __le64 stripe_count
;
3639 } __attribute__ ((packed
)) striping_info_buf
= { 0 };
3640 size_t size
= sizeof (striping_info_buf
);
3647 ret
= rbd_obj_method_sync(rbd_dev
, rbd_dev
->header_name
,
3648 "rbd", "get_stripe_unit_count", NULL
, 0,
3649 (char *)&striping_info_buf
, size
);
3650 dout("%s: rbd_obj_method_sync returned %d\n", __func__
, ret
);
3657 * We don't actually support the "fancy striping" feature
3658 * (STRIPINGV2) yet, but if the striping sizes are the
3659 * defaults the behavior is the same as before. So find
3660 * out, and only fail if the image has non-default values.
3663 obj_size
= (u64
)1 << rbd_dev
->header
.obj_order
;
3664 p
= &striping_info_buf
;
3665 stripe_unit
= ceph_decode_64(&p
);
3666 if (stripe_unit
!= obj_size
) {
3667 rbd_warn(rbd_dev
, "unsupported stripe unit "
3668 "(got %llu want %llu)",
3669 stripe_unit
, obj_size
);
3672 stripe_count
= ceph_decode_64(&p
);
3673 if (stripe_count
!= 1) {
3674 rbd_warn(rbd_dev
, "unsupported stripe count "
3675 "(got %llu want 1)", stripe_count
);
3678 rbd_dev
->header
.stripe_unit
= stripe_unit
;
3679 rbd_dev
->header
.stripe_count
= stripe_count
;
3684 static char *rbd_dev_image_name(struct rbd_device
*rbd_dev
)
3686 size_t image_id_size
;
3691 void *reply_buf
= NULL
;
3693 char *image_name
= NULL
;
3696 rbd_assert(!rbd_dev
->spec
->image_name
);
3698 len
= strlen(rbd_dev
->spec
->image_id
);
3699 image_id_size
= sizeof (__le32
) + len
;
3700 image_id
= kmalloc(image_id_size
, GFP_KERNEL
);
3705 end
= image_id
+ image_id_size
;
3706 ceph_encode_string(&p
, end
, rbd_dev
->spec
->image_id
, (u32
)len
);
3708 size
= sizeof (__le32
) + RBD_IMAGE_NAME_LEN_MAX
;
3709 reply_buf
= kmalloc(size
, GFP_KERNEL
);
3713 ret
= rbd_obj_method_sync(rbd_dev
, RBD_DIRECTORY
,
3714 "rbd", "dir_get_name",
3715 image_id
, image_id_size
,
3720 end
= reply_buf
+ ret
;
3722 image_name
= ceph_extract_encoded_string(&p
, end
, &len
, GFP_KERNEL
);
3723 if (IS_ERR(image_name
))
3726 dout("%s: name is %s len is %zd\n", __func__
, image_name
, len
);
3735 * When an rbd image has a parent image, it is identified by the
3736 * pool, image, and snapshot ids (not names). This function fills
3737 * in the names for those ids. (It's OK if we can't figure out the
3738 * name for an image id, but the pool and snapshot ids should always
3739 * exist and have names.) All names in an rbd spec are dynamically
3742 * When an image being mapped (not a parent) is probed, we have the
3743 * pool name and pool id, image name and image id, and the snapshot
3744 * name. The only thing we're missing is the snapshot id.
3746 * The set of snapshots for an image is not known until they have
3747 * been read by rbd_dev_snaps_update(), so we can't completely fill
3748 * in this information until after that has been called.
3750 static int rbd_dev_spec_update(struct rbd_device
*rbd_dev
)
3752 struct ceph_osd_client
*osdc
= &rbd_dev
->rbd_client
->client
->osdc
;
3753 struct rbd_spec
*spec
= rbd_dev
->spec
;
3754 const char *pool_name
;
3755 const char *image_name
;
3756 const char *snap_name
;
3760 * An image being mapped will have the pool name (etc.), but
3761 * we need to look up the snapshot id.
3763 if (spec
->pool_name
) {
3764 if (strcmp(spec
->snap_name
, RBD_SNAP_HEAD_NAME
)) {
3765 struct rbd_snap
*snap
;
3767 snap
= snap_by_name(rbd_dev
, spec
->snap_name
);
3770 spec
->snap_id
= snap
->id
;
3772 spec
->snap_id
= CEPH_NOSNAP
;
3778 /* Get the pool name; we have to make our own copy of this */
3780 pool_name
= ceph_pg_pool_name_by_id(osdc
->osdmap
, spec
->pool_id
);
3782 rbd_warn(rbd_dev
, "no pool with id %llu", spec
->pool_id
);
3785 pool_name
= kstrdup(pool_name
, GFP_KERNEL
);
3789 /* Fetch the image name; tolerate failure here */
3791 image_name
= rbd_dev_image_name(rbd_dev
);
3793 rbd_warn(rbd_dev
, "unable to get image name");
3795 /* Look up the snapshot name, and make a copy */
3797 snap_name
= rbd_snap_name(rbd_dev
, spec
->snap_id
);
3799 rbd_warn(rbd_dev
, "no snapshot with id %llu", spec
->snap_id
);
3803 snap_name
= kstrdup(snap_name
, GFP_KERNEL
);
3809 spec
->pool_name
= pool_name
;
3810 spec
->image_name
= image_name
;
3811 spec
->snap_name
= snap_name
;
3821 static int rbd_dev_v2_snap_context(struct rbd_device
*rbd_dev
)
3830 struct ceph_snap_context
*snapc
;
3834 * We'll need room for the seq value (maximum snapshot id),
3835 * snapshot count, and array of that many snapshot ids.
3836 * For now we have a fixed upper limit on the number we're
3837 * prepared to receive.
3839 size
= sizeof (__le64
) + sizeof (__le32
) +
3840 RBD_MAX_SNAP_COUNT
* sizeof (__le64
);
3841 reply_buf
= kzalloc(size
, GFP_KERNEL
);
3845 ret
= rbd_obj_method_sync(rbd_dev
, rbd_dev
->header_name
,
3846 "rbd", "get_snapcontext", NULL
, 0,
3848 dout("%s: rbd_obj_method_sync returned %d\n", __func__
, ret
);
3853 end
= reply_buf
+ ret
;
3855 ceph_decode_64_safe(&p
, end
, seq
, out
);
3856 ceph_decode_32_safe(&p
, end
, snap_count
, out
);
3859 * Make sure the reported number of snapshot ids wouldn't go
3860 * beyond the end of our buffer. But before checking that,
3861 * make sure the computed size of the snapshot context we
3862 * allocate is representable in a size_t.
3864 if (snap_count
> (SIZE_MAX
- sizeof (struct ceph_snap_context
))
3869 if (!ceph_has_room(&p
, end
, snap_count
* sizeof (__le64
)))
3873 snapc
= ceph_create_snap_context(snap_count
, GFP_KERNEL
);
3879 for (i
= 0; i
< snap_count
; i
++)
3880 snapc
->snaps
[i
] = ceph_decode_64(&p
);
3882 rbd_dev
->header
.snapc
= snapc
;
3884 dout(" snap context seq = %llu, snap_count = %u\n",
3885 (unsigned long long)seq
, (unsigned int)snap_count
);
3892 static const char *rbd_dev_v2_snap_name(struct rbd_device
*rbd_dev
, u32 which
)
3902 size
= sizeof (__le32
) + RBD_MAX_SNAP_NAME_LEN
;
3903 reply_buf
= kmalloc(size
, GFP_KERNEL
);
3905 return ERR_PTR(-ENOMEM
);
3907 rbd_assert(which
< rbd_dev
->header
.snapc
->num_snaps
);
3908 snap_id
= cpu_to_le64(rbd_dev
->header
.snapc
->snaps
[which
]);
3909 ret
= rbd_obj_method_sync(rbd_dev
, rbd_dev
->header_name
,
3910 "rbd", "get_snapshot_name",
3911 &snap_id
, sizeof (snap_id
),
3913 dout("%s: rbd_obj_method_sync returned %d\n", __func__
, ret
);
3915 snap_name
= ERR_PTR(ret
);
3920 end
= reply_buf
+ ret
;
3921 snap_name
= ceph_extract_encoded_string(&p
, end
, NULL
, GFP_KERNEL
);
3922 if (IS_ERR(snap_name
))
3925 dout(" snap_id 0x%016llx snap_name = %s\n",
3926 (unsigned long long)le64_to_cpu(snap_id
), snap_name
);
3933 static const char *rbd_dev_v2_snap_info(struct rbd_device
*rbd_dev
, u32 which
,
3934 u64
*snap_size
, u64
*snap_features
)
3939 const char *snap_name
;
3942 rbd_assert(which
< rbd_dev
->header
.snapc
->num_snaps
);
3943 snap_id
= rbd_dev
->header
.snapc
->snaps
[which
];
3944 ret
= _rbd_dev_v2_snap_size(rbd_dev
, snap_id
, NULL
, &size
);
3948 ret
= _rbd_dev_v2_snap_features(rbd_dev
, snap_id
, &features
);
3952 snap_name
= rbd_dev_v2_snap_name(rbd_dev
, which
);
3953 if (!IS_ERR(snap_name
)) {
3955 *snap_features
= features
;
3960 return ERR_PTR(ret
);
3963 static const char *rbd_dev_snap_info(struct rbd_device
*rbd_dev
, u32 which
,
3964 u64
*snap_size
, u64
*snap_features
)
3966 if (rbd_dev
->image_format
== 1)
3967 return rbd_dev_v1_snap_info(rbd_dev
, which
,
3968 snap_size
, snap_features
);
3969 if (rbd_dev
->image_format
== 2)
3970 return rbd_dev_v2_snap_info(rbd_dev
, which
,
3971 snap_size
, snap_features
);
3972 return ERR_PTR(-EINVAL
);
3975 static int rbd_dev_v2_refresh(struct rbd_device
*rbd_dev
)
3979 down_write(&rbd_dev
->header_rwsem
);
3981 ret
= rbd_dev_v2_image_size(rbd_dev
);
3984 rbd_update_mapping_size(rbd_dev
);
3986 ret
= rbd_dev_v2_snap_context(rbd_dev
);
3987 dout("rbd_dev_v2_snap_context returned %d\n", ret
);
3990 ret
= rbd_dev_snaps_update(rbd_dev
);
3991 dout("rbd_dev_snaps_update returned %d\n", ret
);
3995 up_write(&rbd_dev
->header_rwsem
);
4001 * Scan the rbd device's current snapshot list and compare it to the
4002 * newly-received snapshot context. Remove any existing snapshots
4003 * not present in the new snapshot context. Add a new snapshot for
4004 * any snaphots in the snapshot context not in the current list.
4005 * And verify there are no changes to snapshots we already know
4008 * Assumes the snapshots in the snapshot context are sorted by
4009 * snapshot id, highest id first. (Snapshots in the rbd_dev's list
4010 * are also maintained in that order.)
4012 * Note that any error occurs while updating the snapshot list
4013 * aborts the update, and the entire list is cleared. The snapshot
4014 * list becomes inconsistent at that point anyway, so it might as
4017 static int rbd_dev_snaps_update(struct rbd_device
*rbd_dev
)
4019 struct ceph_snap_context
*snapc
= rbd_dev
->header
.snapc
;
4020 const u32 snap_count
= snapc
->num_snaps
;
4021 struct list_head
*head
= &rbd_dev
->snaps
;
4022 struct list_head
*links
= head
->next
;
4026 dout("%s: snap count is %u\n", __func__
, (unsigned int)snap_count
);
4027 while (index
< snap_count
|| links
!= head
) {
4029 struct rbd_snap
*snap
;
4030 const char *snap_name
;
4032 u64 snap_features
= 0;
4034 snap_id
= index
< snap_count
? snapc
->snaps
[index
]
4036 snap
= links
!= head
? list_entry(links
, struct rbd_snap
, node
)
4038 rbd_assert(!snap
|| snap
->id
!= CEPH_NOSNAP
);
4040 if (snap_id
== CEPH_NOSNAP
|| (snap
&& snap
->id
> snap_id
)) {
4041 struct list_head
*next
= links
->next
;
4044 * A previously-existing snapshot is not in
4045 * the new snap context.
4047 * If the now-missing snapshot is the one
4048 * the image represents, clear its existence
4049 * flag so we can avoid sending any more
4052 if (rbd_dev
->spec
->snap_id
== snap
->id
)
4053 clear_bit(RBD_DEV_FLAG_EXISTS
, &rbd_dev
->flags
);
4054 dout("removing %ssnap id %llu\n",
4055 rbd_dev
->spec
->snap_id
== snap
->id
?
4057 (unsigned long long)snap
->id
);
4059 list_del(&snap
->node
);
4060 rbd_snap_destroy(snap
);
4062 /* Done with this list entry; advance */
4068 snap_name
= rbd_dev_snap_info(rbd_dev
, index
,
4069 &snap_size
, &snap_features
);
4070 if (IS_ERR(snap_name
)) {
4071 ret
= PTR_ERR(snap_name
);
4072 dout("failed to get snap info, error %d\n", ret
);
4076 dout("entry %u: snap_id = %llu\n", (unsigned int)snap_count
,
4077 (unsigned long long)snap_id
);
4078 if (!snap
|| (snap_id
!= CEPH_NOSNAP
&& snap
->id
< snap_id
)) {
4079 struct rbd_snap
*new_snap
;
4081 /* We haven't seen this snapshot before */
4083 new_snap
= rbd_snap_create(rbd_dev
, snap_name
,
4084 snap_id
, snap_size
, snap_features
);
4085 if (IS_ERR(new_snap
)) {
4086 ret
= PTR_ERR(new_snap
);
4087 dout(" failed to add dev, error %d\n", ret
);
4091 /* New goes before existing, or at end of list */
4093 dout(" added dev%s\n", snap
? "" : " at end\n");
4095 list_add_tail(&new_snap
->node
, &snap
->node
);
4097 list_add_tail(&new_snap
->node
, head
);
4099 /* Already have this one */
4101 dout(" already present\n");
4103 rbd_assert(snap
->size
== snap_size
);
4104 rbd_assert(!strcmp(snap
->name
, snap_name
));
4105 rbd_assert(snap
->features
== snap_features
);
4107 /* Done with this list entry; advance */
4109 links
= links
->next
;
4112 /* Advance to the next entry in the snapshot context */
4116 dout("%s: done\n", __func__
);
4120 rbd_remove_all_snaps(rbd_dev
);
4125 static int rbd_bus_add_dev(struct rbd_device
*rbd_dev
)
4130 mutex_lock_nested(&ctl_mutex
, SINGLE_DEPTH_NESTING
);
4132 dev
= &rbd_dev
->dev
;
4133 dev
->bus
= &rbd_bus_type
;
4134 dev
->type
= &rbd_device_type
;
4135 dev
->parent
= &rbd_root_dev
;
4136 dev
->release
= rbd_dev_device_release
;
4137 dev_set_name(dev
, "%d", rbd_dev
->dev_id
);
4138 ret
= device_register(dev
);
4140 mutex_unlock(&ctl_mutex
);
4145 static void rbd_bus_del_dev(struct rbd_device
*rbd_dev
)
4147 device_unregister(&rbd_dev
->dev
);
4150 static atomic64_t rbd_dev_id_max
= ATOMIC64_INIT(0);
4153 * Get a unique rbd identifier for the given new rbd_dev, and add
4154 * the rbd_dev to the global list. The minimum rbd id is 1.
4156 static void rbd_dev_id_get(struct rbd_device
*rbd_dev
)
4158 rbd_dev
->dev_id
= atomic64_inc_return(&rbd_dev_id_max
);
4160 spin_lock(&rbd_dev_list_lock
);
4161 list_add_tail(&rbd_dev
->node
, &rbd_dev_list
);
4162 spin_unlock(&rbd_dev_list_lock
);
4163 dout("rbd_dev %p given dev id %llu\n", rbd_dev
,
4164 (unsigned long long) rbd_dev
->dev_id
);
4168 * Remove an rbd_dev from the global list, and record that its
4169 * identifier is no longer in use.
4171 static void rbd_dev_id_put(struct rbd_device
*rbd_dev
)
4173 struct list_head
*tmp
;
4174 int rbd_id
= rbd_dev
->dev_id
;
4177 rbd_assert(rbd_id
> 0);
4179 dout("rbd_dev %p released dev id %llu\n", rbd_dev
,
4180 (unsigned long long) rbd_dev
->dev_id
);
4181 spin_lock(&rbd_dev_list_lock
);
4182 list_del_init(&rbd_dev
->node
);
4185 * If the id being "put" is not the current maximum, there
4186 * is nothing special we need to do.
4188 if (rbd_id
!= atomic64_read(&rbd_dev_id_max
)) {
4189 spin_unlock(&rbd_dev_list_lock
);
4194 * We need to update the current maximum id. Search the
4195 * list to find out what it is. We're more likely to find
4196 * the maximum at the end, so search the list backward.
4199 list_for_each_prev(tmp
, &rbd_dev_list
) {
4200 struct rbd_device
*rbd_dev
;
4202 rbd_dev
= list_entry(tmp
, struct rbd_device
, node
);
4203 if (rbd_dev
->dev_id
> max_id
)
4204 max_id
= rbd_dev
->dev_id
;
4206 spin_unlock(&rbd_dev_list_lock
);
4209 * The max id could have been updated by rbd_dev_id_get(), in
4210 * which case it now accurately reflects the new maximum.
4211 * Be careful not to overwrite the maximum value in that
4214 atomic64_cmpxchg(&rbd_dev_id_max
, rbd_id
, max_id
);
4215 dout(" max dev id has been reset\n");
4219 * Skips over white space at *buf, and updates *buf to point to the
4220 * first found non-space character (if any). Returns the length of
4221 * the token (string of non-white space characters) found. Note
4222 * that *buf must be terminated with '\0'.
4224 static inline size_t next_token(const char **buf
)
4227 * These are the characters that produce nonzero for
4228 * isspace() in the "C" and "POSIX" locales.
4230 const char *spaces
= " \f\n\r\t\v";
4232 *buf
+= strspn(*buf
, spaces
); /* Find start of token */
4234 return strcspn(*buf
, spaces
); /* Return token length */
4238 * Finds the next token in *buf, and if the provided token buffer is
4239 * big enough, copies the found token into it. The result, if
4240 * copied, is guaranteed to be terminated with '\0'. Note that *buf
4241 * must be terminated with '\0' on entry.
4243 * Returns the length of the token found (not including the '\0').
4244 * Return value will be 0 if no token is found, and it will be >=
4245 * token_size if the token would not fit.
4247 * The *buf pointer will be updated to point beyond the end of the
4248 * found token. Note that this occurs even if the token buffer is
4249 * too small to hold it.
4251 static inline size_t copy_token(const char **buf
,
4257 len
= next_token(buf
);
4258 if (len
< token_size
) {
4259 memcpy(token
, *buf
, len
);
4260 *(token
+ len
) = '\0';
4268 * Finds the next token in *buf, dynamically allocates a buffer big
4269 * enough to hold a copy of it, and copies the token into the new
4270 * buffer. The copy is guaranteed to be terminated with '\0'. Note
4271 * that a duplicate buffer is created even for a zero-length token.
4273 * Returns a pointer to the newly-allocated duplicate, or a null
4274 * pointer if memory for the duplicate was not available. If
4275 * the lenp argument is a non-null pointer, the length of the token
4276 * (not including the '\0') is returned in *lenp.
4278 * If successful, the *buf pointer will be updated to point beyond
4279 * the end of the found token.
4281 * Note: uses GFP_KERNEL for allocation.
4283 static inline char *dup_token(const char **buf
, size_t *lenp
)
4288 len
= next_token(buf
);
4289 dup
= kmemdup(*buf
, len
+ 1, GFP_KERNEL
);
4292 *(dup
+ len
) = '\0';
4302 * Parse the options provided for an "rbd add" (i.e., rbd image
4303 * mapping) request. These arrive via a write to /sys/bus/rbd/add,
4304 * and the data written is passed here via a NUL-terminated buffer.
4305 * Returns 0 if successful or an error code otherwise.
4307 * The information extracted from these options is recorded in
4308 * the other parameters which return dynamically-allocated
4311 * The address of a pointer that will refer to a ceph options
4312 * structure. Caller must release the returned pointer using
4313 * ceph_destroy_options() when it is no longer needed.
4315 * Address of an rbd options pointer. Fully initialized by
4316 * this function; caller must release with kfree().
4318 * Address of an rbd image specification pointer. Fully
4319 * initialized by this function based on parsed options.
4320 * Caller must release with rbd_spec_put().
4322 * The options passed take this form:
4323 * <mon_addrs> <options> <pool_name> <image_name> [<snap_id>]
4326 * A comma-separated list of one or more monitor addresses.
4327 * A monitor address is an ip address, optionally followed
4328 * by a port number (separated by a colon).
4329 * I.e.: ip1[:port1][,ip2[:port2]...]
4331 * A comma-separated list of ceph and/or rbd options.
4333 * The name of the rados pool containing the rbd image.
4335 * The name of the image in that pool to map.
4337 * An optional snapshot id. If provided, the mapping will
4338 * present data from the image at the time that snapshot was
4339 * created. The image head is used if no snapshot id is
4340 * provided. Snapshot mappings are always read-only.
4342 static int rbd_add_parse_args(const char *buf
,
4343 struct ceph_options
**ceph_opts
,
4344 struct rbd_options
**opts
,
4345 struct rbd_spec
**rbd_spec
)
4349 const char *mon_addrs
;
4351 size_t mon_addrs_size
;
4352 struct rbd_spec
*spec
= NULL
;
4353 struct rbd_options
*rbd_opts
= NULL
;
4354 struct ceph_options
*copts
;
4357 /* The first four tokens are required */
4359 len
= next_token(&buf
);
4361 rbd_warn(NULL
, "no monitor address(es) provided");
4365 mon_addrs_size
= len
+ 1;
4369 options
= dup_token(&buf
, NULL
);
4373 rbd_warn(NULL
, "no options provided");
4377 spec
= rbd_spec_alloc();
4381 spec
->pool_name
= dup_token(&buf
, NULL
);
4382 if (!spec
->pool_name
)
4384 if (!*spec
->pool_name
) {
4385 rbd_warn(NULL
, "no pool name provided");
4389 spec
->image_name
= dup_token(&buf
, NULL
);
4390 if (!spec
->image_name
)
4392 if (!*spec
->image_name
) {
4393 rbd_warn(NULL
, "no image name provided");
4398 * Snapshot name is optional; default is to use "-"
4399 * (indicating the head/no snapshot).
4401 len
= next_token(&buf
);
4403 buf
= RBD_SNAP_HEAD_NAME
; /* No snapshot supplied */
4404 len
= sizeof (RBD_SNAP_HEAD_NAME
) - 1;
4405 } else if (len
> RBD_MAX_SNAP_NAME_LEN
) {
4406 ret
= -ENAMETOOLONG
;
4409 snap_name
= kmemdup(buf
, len
+ 1, GFP_KERNEL
);
4412 *(snap_name
+ len
) = '\0';
4413 spec
->snap_name
= snap_name
;
4415 /* Initialize all rbd options to the defaults */
4417 rbd_opts
= kzalloc(sizeof (*rbd_opts
), GFP_KERNEL
);
4421 rbd_opts
->read_only
= RBD_READ_ONLY_DEFAULT
;
4423 copts
= ceph_parse_options(options
, mon_addrs
,
4424 mon_addrs
+ mon_addrs_size
- 1,
4425 parse_rbd_opts_token
, rbd_opts
);
4426 if (IS_ERR(copts
)) {
4427 ret
= PTR_ERR(copts
);
4448 * An rbd format 2 image has a unique identifier, distinct from the
4449 * name given to it by the user. Internally, that identifier is
4450 * what's used to specify the names of objects related to the image.
4452 * A special "rbd id" object is used to map an rbd image name to its
4453 * id. If that object doesn't exist, then there is no v2 rbd image
4454 * with the supplied name.
4456 * This function will record the given rbd_dev's image_id field if
4457 * it can be determined, and in that case will return 0. If any
4458 * errors occur a negative errno will be returned and the rbd_dev's
4459 * image_id field will be unchanged (and should be NULL).
4461 static int rbd_dev_image_id(struct rbd_device
*rbd_dev
)
4470 * When probing a parent image, the image id is already
4471 * known (and the image name likely is not). There's no
4472 * need to fetch the image id again in this case. We
4473 * do still need to set the image format though.
4475 if (rbd_dev
->spec
->image_id
) {
4476 rbd_dev
->image_format
= *rbd_dev
->spec
->image_id
? 2 : 1;
4482 * First, see if the format 2 image id file exists, and if
4483 * so, get the image's persistent id from it.
4485 size
= sizeof (RBD_ID_PREFIX
) + strlen(rbd_dev
->spec
->image_name
);
4486 object_name
= kmalloc(size
, GFP_NOIO
);
4489 sprintf(object_name
, "%s%s", RBD_ID_PREFIX
, rbd_dev
->spec
->image_name
);
4490 dout("rbd id object name is %s\n", object_name
);
4492 /* Response will be an encoded string, which includes a length */
4494 size
= sizeof (__le32
) + RBD_IMAGE_ID_LEN_MAX
;
4495 response
= kzalloc(size
, GFP_NOIO
);
4501 /* If it doesn't exist we'll assume it's a format 1 image */
4503 ret
= rbd_obj_method_sync(rbd_dev
, object_name
,
4504 "rbd", "get_id", NULL
, 0,
4505 response
, RBD_IMAGE_ID_LEN_MAX
);
4506 dout("%s: rbd_obj_method_sync returned %d\n", __func__
, ret
);
4507 if (ret
== -ENOENT
) {
4508 image_id
= kstrdup("", GFP_KERNEL
);
4509 ret
= image_id
? 0 : -ENOMEM
;
4511 rbd_dev
->image_format
= 1;
4512 } else if (ret
> sizeof (__le32
)) {
4515 image_id
= ceph_extract_encoded_string(&p
, p
+ ret
,
4517 ret
= IS_ERR(image_id
) ? PTR_ERR(image_id
) : 0;
4519 rbd_dev
->image_format
= 2;
4525 rbd_dev
->spec
->image_id
= image_id
;
4526 dout("image_id is %s\n", image_id
);
4535 /* Undo whatever state changes are made by v1 or v2 image probe */
4537 static void rbd_dev_unprobe(struct rbd_device
*rbd_dev
)
4539 struct rbd_image_header
*header
;
4541 rbd_dev_remove_parent(rbd_dev
);
4542 rbd_spec_put(rbd_dev
->parent_spec
);
4543 rbd_dev
->parent_spec
= NULL
;
4544 rbd_dev
->parent_overlap
= 0;
4546 /* Free dynamic fields from the header, then zero it out */
4548 header
= &rbd_dev
->header
;
4549 ceph_put_snap_context(header
->snapc
);
4550 kfree(header
->snap_sizes
);
4551 kfree(header
->snap_names
);
4552 kfree(header
->object_prefix
);
4553 memset(header
, 0, sizeof (*header
));
4556 static int rbd_dev_v1_probe(struct rbd_device
*rbd_dev
)
4560 /* Populate rbd image metadata */
4562 ret
= rbd_read_header(rbd_dev
, &rbd_dev
->header
);
4566 /* Version 1 images have no parent (no layering) */
4568 rbd_dev
->parent_spec
= NULL
;
4569 rbd_dev
->parent_overlap
= 0;
4571 dout("discovered version 1 image, header name is %s\n",
4572 rbd_dev
->header_name
);
4577 kfree(rbd_dev
->header_name
);
4578 rbd_dev
->header_name
= NULL
;
4579 kfree(rbd_dev
->spec
->image_id
);
4580 rbd_dev
->spec
->image_id
= NULL
;
4585 static int rbd_dev_v2_probe(struct rbd_device
*rbd_dev
)
4589 ret
= rbd_dev_v2_image_size(rbd_dev
);
4593 /* Get the object prefix (a.k.a. block_name) for the image */
4595 ret
= rbd_dev_v2_object_prefix(rbd_dev
);
4599 /* Get the and check features for the image */
4601 ret
= rbd_dev_v2_features(rbd_dev
);
4605 /* If the image supports layering, get the parent info */
4607 if (rbd_dev
->header
.features
& RBD_FEATURE_LAYERING
) {
4608 ret
= rbd_dev_v2_parent_info(rbd_dev
);
4613 * Don't print a warning for parent images. We can
4614 * tell this point because we won't know its pool
4615 * name yet (just its pool id).
4617 if (rbd_dev
->spec
->pool_name
)
4618 rbd_warn(rbd_dev
, "WARNING: kernel layering "
4619 "is EXPERIMENTAL!");
4622 /* If the image supports fancy striping, get its parameters */
4624 if (rbd_dev
->header
.features
& RBD_FEATURE_STRIPINGV2
) {
4625 ret
= rbd_dev_v2_striping_info(rbd_dev
);
4630 /* crypto and compression type aren't (yet) supported for v2 images */
4632 rbd_dev
->header
.crypt_type
= 0;
4633 rbd_dev
->header
.comp_type
= 0;
4635 /* Get the snapshot context, plus the header version */
4637 ret
= rbd_dev_v2_snap_context(rbd_dev
);
4641 dout("discovered version 2 image, header name is %s\n",
4642 rbd_dev
->header_name
);
4646 rbd_dev
->parent_overlap
= 0;
4647 rbd_spec_put(rbd_dev
->parent_spec
);
4648 rbd_dev
->parent_spec
= NULL
;
4649 kfree(rbd_dev
->header_name
);
4650 rbd_dev
->header_name
= NULL
;
4651 kfree(rbd_dev
->header
.object_prefix
);
4652 rbd_dev
->header
.object_prefix
= NULL
;
4657 static int rbd_dev_probe_parent(struct rbd_device
*rbd_dev
)
4659 struct rbd_device
*parent
= NULL
;
4660 struct rbd_spec
*parent_spec
;
4661 struct rbd_client
*rbdc
;
4664 if (!rbd_dev
->parent_spec
)
4667 * We need to pass a reference to the client and the parent
4668 * spec when creating the parent rbd_dev. Images related by
4669 * parent/child relationships always share both.
4671 parent_spec
= rbd_spec_get(rbd_dev
->parent_spec
);
4672 rbdc
= __rbd_get_client(rbd_dev
->rbd_client
);
4675 parent
= rbd_dev_create(rbdc
, parent_spec
);
4679 ret
= rbd_dev_image_probe(parent
);
4682 rbd_dev
->parent
= parent
;
4687 rbd_spec_put(rbd_dev
->parent_spec
);
4688 kfree(rbd_dev
->header_name
);
4689 rbd_dev_destroy(parent
);
4691 rbd_put_client(rbdc
);
4692 rbd_spec_put(parent_spec
);
4698 static int rbd_dev_device_setup(struct rbd_device
*rbd_dev
)
4702 ret
= rbd_dev_mapping_set(rbd_dev
);
4706 /* generate unique id: find highest unique id, add one */
4707 rbd_dev_id_get(rbd_dev
);
4709 /* Fill in the device name, now that we have its id. */
4710 BUILD_BUG_ON(DEV_NAME_LEN
4711 < sizeof (RBD_DRV_NAME
) + MAX_INT_FORMAT_WIDTH
);
4712 sprintf(rbd_dev
->name
, "%s%d", RBD_DRV_NAME
, rbd_dev
->dev_id
);
4714 /* Get our block major device number. */
4716 ret
= register_blkdev(0, rbd_dev
->name
);
4719 rbd_dev
->major
= ret
;
4721 /* Set up the blkdev mapping. */
4723 ret
= rbd_init_disk(rbd_dev
);
4725 goto err_out_blkdev
;
4727 ret
= rbd_bus_add_dev(rbd_dev
);
4731 /* Everything's ready. Announce the disk to the world. */
4733 set_capacity(rbd_dev
->disk
, rbd_dev
->mapping
.size
/ SECTOR_SIZE
);
4734 set_bit(RBD_DEV_FLAG_EXISTS
, &rbd_dev
->flags
);
4735 add_disk(rbd_dev
->disk
);
4737 pr_info("%s: added with size 0x%llx\n", rbd_dev
->disk
->disk_name
,
4738 (unsigned long long) rbd_dev
->mapping
.size
);
4743 rbd_free_disk(rbd_dev
);
4745 unregister_blkdev(rbd_dev
->major
, rbd_dev
->name
);
4747 rbd_dev_id_put(rbd_dev
);
4748 rbd_dev_mapping_clear(rbd_dev
);
4753 static int rbd_dev_header_name(struct rbd_device
*rbd_dev
)
4755 struct rbd_spec
*spec
= rbd_dev
->spec
;
4758 /* Record the header object name for this rbd image. */
4760 rbd_assert(rbd_image_format_valid(rbd_dev
->image_format
));
4762 if (rbd_dev
->image_format
== 1)
4763 size
= strlen(spec
->image_name
) + sizeof (RBD_SUFFIX
);
4765 size
= sizeof (RBD_HEADER_PREFIX
) + strlen(spec
->image_id
);
4767 rbd_dev
->header_name
= kmalloc(size
, GFP_KERNEL
);
4768 if (!rbd_dev
->header_name
)
4771 if (rbd_dev
->image_format
== 1)
4772 sprintf(rbd_dev
->header_name
, "%s%s",
4773 spec
->image_name
, RBD_SUFFIX
);
4775 sprintf(rbd_dev
->header_name
, "%s%s",
4776 RBD_HEADER_PREFIX
, spec
->image_id
);
4780 static void rbd_dev_image_release(struct rbd_device
*rbd_dev
)
4784 rbd_remove_all_snaps(rbd_dev
);
4785 rbd_dev_unprobe(rbd_dev
);
4786 ret
= rbd_dev_header_watch_sync(rbd_dev
, 0);
4788 rbd_warn(rbd_dev
, "failed to cancel watch event (%d)\n", ret
);
4789 kfree(rbd_dev
->header_name
);
4790 rbd_dev
->header_name
= NULL
;
4791 rbd_dev
->image_format
= 0;
4792 kfree(rbd_dev
->spec
->image_id
);
4793 rbd_dev
->spec
->image_id
= NULL
;
4795 rbd_dev_destroy(rbd_dev
);
4799 * Probe for the existence of the header object for the given rbd
4800 * device. For format 2 images this includes determining the image
4803 static int rbd_dev_image_probe(struct rbd_device
*rbd_dev
)
4809 * Get the id from the image id object. If it's not a
4810 * format 2 image, we'll get ENOENT back, and we'll assume
4811 * it's a format 1 image.
4813 ret
= rbd_dev_image_id(rbd_dev
);
4816 rbd_assert(rbd_dev
->spec
->image_id
);
4817 rbd_assert(rbd_image_format_valid(rbd_dev
->image_format
));
4819 ret
= rbd_dev_header_name(rbd_dev
);
4821 goto err_out_format
;
4823 ret
= rbd_dev_header_watch_sync(rbd_dev
, 1);
4825 goto out_header_name
;
4827 if (rbd_dev
->image_format
== 1)
4828 ret
= rbd_dev_v1_probe(rbd_dev
);
4830 ret
= rbd_dev_v2_probe(rbd_dev
);
4834 ret
= rbd_dev_snaps_update(rbd_dev
);
4838 ret
= rbd_dev_spec_update(rbd_dev
);
4842 ret
= rbd_dev_probe_parent(rbd_dev
);
4847 rbd_remove_all_snaps(rbd_dev
);
4849 rbd_dev_unprobe(rbd_dev
);
4851 tmp
= rbd_dev_header_watch_sync(rbd_dev
, 0);
4853 rbd_warn(rbd_dev
, "unable to tear down watch request\n");
4855 kfree(rbd_dev
->header_name
);
4856 rbd_dev
->header_name
= NULL
;
4858 rbd_dev
->image_format
= 0;
4859 kfree(rbd_dev
->spec
->image_id
);
4860 rbd_dev
->spec
->image_id
= NULL
;
4862 dout("probe failed, returning %d\n", ret
);
4867 static ssize_t
rbd_add(struct bus_type
*bus
,
4871 struct rbd_device
*rbd_dev
= NULL
;
4872 struct ceph_options
*ceph_opts
= NULL
;
4873 struct rbd_options
*rbd_opts
= NULL
;
4874 struct rbd_spec
*spec
= NULL
;
4875 struct rbd_client
*rbdc
;
4876 struct ceph_osd_client
*osdc
;
4879 if (!try_module_get(THIS_MODULE
))
4882 /* parse add command */
4883 rc
= rbd_add_parse_args(buf
, &ceph_opts
, &rbd_opts
, &spec
);
4885 goto err_out_module
;
4887 rbdc
= rbd_get_client(ceph_opts
);
4892 ceph_opts
= NULL
; /* rbd_dev client now owns this */
4895 osdc
= &rbdc
->client
->osdc
;
4896 rc
= ceph_pg_poolid_by_name(osdc
->osdmap
, spec
->pool_name
);
4898 goto err_out_client
;
4899 spec
->pool_id
= (u64
)rc
;
4901 /* The ceph file layout needs to fit pool id in 32 bits */
4903 if (spec
->pool_id
> (u64
)U32_MAX
) {
4904 rbd_warn(NULL
, "pool id too large (%llu > %u)\n",
4905 (unsigned long long)spec
->pool_id
, U32_MAX
);
4907 goto err_out_client
;
4910 rbd_dev
= rbd_dev_create(rbdc
, spec
);
4912 goto err_out_client
;
4913 rbdc
= NULL
; /* rbd_dev now owns this */
4914 spec
= NULL
; /* rbd_dev now owns this */
4916 rbd_dev
->mapping
.read_only
= rbd_opts
->read_only
;
4918 rbd_opts
= NULL
; /* done with this */
4920 rc
= rbd_dev_image_probe(rbd_dev
);
4922 goto err_out_rbd_dev
;
4924 rc
= rbd_dev_device_setup(rbd_dev
);
4928 rbd_dev_image_release(rbd_dev
);
4930 rbd_dev_destroy(rbd_dev
);
4932 rbd_put_client(rbdc
);
4935 ceph_destroy_options(ceph_opts
);
4939 module_put(THIS_MODULE
);
4941 dout("Error adding device %s\n", buf
);
4946 static struct rbd_device
*__rbd_get_dev(unsigned long dev_id
)
4948 struct list_head
*tmp
;
4949 struct rbd_device
*rbd_dev
;
4951 spin_lock(&rbd_dev_list_lock
);
4952 list_for_each(tmp
, &rbd_dev_list
) {
4953 rbd_dev
= list_entry(tmp
, struct rbd_device
, node
);
4954 if (rbd_dev
->dev_id
== dev_id
) {
4955 spin_unlock(&rbd_dev_list_lock
);
4959 spin_unlock(&rbd_dev_list_lock
);
4963 static void rbd_dev_device_release(struct device
*dev
)
4965 struct rbd_device
*rbd_dev
= dev_to_rbd_dev(dev
);
4967 rbd_free_disk(rbd_dev
);
4968 clear_bit(RBD_DEV_FLAG_EXISTS
, &rbd_dev
->flags
);
4969 rbd_dev_clear_mapping(rbd_dev
);
4970 unregister_blkdev(rbd_dev
->major
, rbd_dev
->name
);
4972 rbd_dev_id_put(rbd_dev
);
4973 rbd_dev_mapping_clear(rbd_dev
);
4976 static void rbd_dev_remove_parent(struct rbd_device
*rbd_dev
)
4978 while (rbd_dev
->parent
) {
4979 struct rbd_device
*first
= rbd_dev
;
4980 struct rbd_device
*second
= first
->parent
;
4981 struct rbd_device
*third
;
4984 * Follow to the parent with no grandparent and
4987 while (second
&& (third
= second
->parent
)) {
4992 rbd_dev_image_release(second
);
4993 first
->parent
= NULL
;
4994 first
->parent_overlap
= 0;
4996 rbd_assert(first
->parent_spec
);
4997 rbd_spec_put(first
->parent_spec
);
4998 first
->parent_spec
= NULL
;
5002 static ssize_t
rbd_remove(struct bus_type
*bus
,
5006 struct rbd_device
*rbd_dev
= NULL
;
5011 ret
= strict_strtoul(buf
, 10, &ul
);
5015 /* convert to int; abort if we lost anything in the conversion */
5016 target_id
= (int) ul
;
5017 if (target_id
!= ul
)
5020 mutex_lock_nested(&ctl_mutex
, SINGLE_DEPTH_NESTING
);
5022 rbd_dev
= __rbd_get_dev(target_id
);
5028 spin_lock_irq(&rbd_dev
->lock
);
5029 if (rbd_dev
->open_count
)
5032 set_bit(RBD_DEV_FLAG_REMOVING
, &rbd_dev
->flags
);
5033 spin_unlock_irq(&rbd_dev
->lock
);
5037 rbd_bus_del_dev(rbd_dev
);
5038 rbd_dev_image_release(rbd_dev
);
5039 module_put(THIS_MODULE
);
5041 mutex_unlock(&ctl_mutex
);
5047 * create control files in sysfs
5050 static int rbd_sysfs_init(void)
5054 ret
= device_register(&rbd_root_dev
);
5058 ret
= bus_register(&rbd_bus_type
);
5060 device_unregister(&rbd_root_dev
);
5065 static void rbd_sysfs_cleanup(void)
5067 bus_unregister(&rbd_bus_type
);
5068 device_unregister(&rbd_root_dev
);
5071 static int __init
rbd_init(void)
5075 if (!libceph_compatible(NULL
)) {
5076 rbd_warn(NULL
, "libceph incompatibility (quitting)");
5080 rc
= rbd_sysfs_init();
5083 pr_info("loaded " RBD_DRV_NAME_LONG
"\n");
5087 static void __exit
rbd_exit(void)
5089 rbd_sysfs_cleanup();
5092 module_init(rbd_init
);
5093 module_exit(rbd_exit
);
5095 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
5096 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
5097 MODULE_DESCRIPTION("rados block device");
5099 /* following authorship retained from original osdblk.c */
5100 MODULE_AUTHOR("Jeff Garzik <jeff@garzik.org>");
5102 MODULE_LICENSE("GPL");