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>
36 #include <linux/bsearch.h>
38 #include <linux/kernel.h>
39 #include <linux/device.h>
40 #include <linux/module.h>
42 #include <linux/blkdev.h>
43 #include <linux/slab.h>
45 #include "rbd_types.h"
47 #define RBD_DEBUG /* Activate rbd_assert() calls */
50 * The basic unit of block I/O is a sector. It is interpreted in a
51 * number of contexts in Linux (blk, bio, genhd), but the default is
52 * universally 512 bytes. These symbols are just slightly more
53 * meaningful than the bare numbers they represent.
55 #define SECTOR_SHIFT 9
56 #define SECTOR_SIZE (1ULL << SECTOR_SHIFT)
59 * Increment the given counter and return its updated value.
60 * If the counter is already 0 it will not be incremented.
61 * If the counter is already at its maximum value returns
62 * -EINVAL without updating it.
64 static int atomic_inc_return_safe(atomic_t
*v
)
68 counter
= (unsigned int)__atomic_add_unless(v
, 1, 0);
69 if (counter
<= (unsigned int)INT_MAX
)
77 /* Decrement the counter. Return the resulting value, or -EINVAL */
78 static int atomic_dec_return_safe(atomic_t
*v
)
82 counter
= atomic_dec_return(v
);
91 #define RBD_DRV_NAME "rbd"
92 #define RBD_DRV_NAME_LONG "rbd (rados block device)"
94 #define RBD_MINORS_PER_MAJOR 256 /* max minors per blkdev */
96 #define RBD_SNAP_DEV_NAME_PREFIX "snap_"
97 #define RBD_MAX_SNAP_NAME_LEN \
98 (NAME_MAX - (sizeof (RBD_SNAP_DEV_NAME_PREFIX) - 1))
100 #define RBD_MAX_SNAP_COUNT 510 /* allows max snapc to fit in 4KB */
102 #define RBD_SNAP_HEAD_NAME "-"
104 #define BAD_SNAP_INDEX U32_MAX /* invalid index into snap array */
106 /* This allows a single page to hold an image name sent by OSD */
107 #define RBD_IMAGE_NAME_LEN_MAX (PAGE_SIZE - sizeof (__le32) - 1)
108 #define RBD_IMAGE_ID_LEN_MAX 64
110 #define RBD_OBJ_PREFIX_LEN_MAX 64
114 #define RBD_FEATURE_LAYERING (1<<0)
115 #define RBD_FEATURE_STRIPINGV2 (1<<1)
116 #define RBD_FEATURES_ALL \
117 (RBD_FEATURE_LAYERING | RBD_FEATURE_STRIPINGV2)
119 /* Features supported by this (client software) implementation. */
121 #define RBD_FEATURES_SUPPORTED (RBD_FEATURES_ALL)
124 * An RBD device name will be "rbd#", where the "rbd" comes from
125 * RBD_DRV_NAME above, and # is a unique integer identifier.
126 * MAX_INT_FORMAT_WIDTH is used in ensuring DEV_NAME_LEN is big
127 * enough to hold all possible device names.
129 #define DEV_NAME_LEN 32
130 #define MAX_INT_FORMAT_WIDTH ((5 * sizeof (int)) / 2 + 1)
133 * block device image metadata (in-memory version)
135 struct rbd_image_header
{
136 /* These six fields never change for a given rbd image */
143 u64 features
; /* Might be changeable someday? */
145 /* The remaining fields need to be updated occasionally */
147 struct ceph_snap_context
*snapc
;
148 char *snap_names
; /* format 1 only */
149 u64
*snap_sizes
; /* format 1 only */
153 * An rbd image specification.
155 * The tuple (pool_id, image_id, snap_id) is sufficient to uniquely
156 * identify an image. Each rbd_dev structure includes a pointer to
157 * an rbd_spec structure that encapsulates this identity.
159 * Each of the id's in an rbd_spec has an associated name. For a
160 * user-mapped image, the names are supplied and the id's associated
161 * with them are looked up. For a layered image, a parent image is
162 * defined by the tuple, and the names are looked up.
164 * An rbd_dev structure contains a parent_spec pointer which is
165 * non-null if the image it represents is a child in a layered
166 * image. This pointer will refer to the rbd_spec structure used
167 * by the parent rbd_dev for its own identity (i.e., the structure
168 * is shared between the parent and child).
170 * Since these structures are populated once, during the discovery
171 * phase of image construction, they are effectively immutable so
172 * we make no effort to synchronize access to them.
174 * Note that code herein does not assume the image name is known (it
175 * could be a null pointer).
179 const char *pool_name
;
181 const char *image_id
;
182 const char *image_name
;
185 const char *snap_name
;
191 * an instance of the client. multiple devices may share an rbd client.
194 struct ceph_client
*client
;
196 struct list_head node
;
199 struct rbd_img_request
;
200 typedef void (*rbd_img_callback_t
)(struct rbd_img_request
*);
202 #define BAD_WHICH U32_MAX /* Good which or bad which, which? */
204 struct rbd_obj_request
;
205 typedef void (*rbd_obj_callback_t
)(struct rbd_obj_request
*);
207 enum obj_request_type
{
208 OBJ_REQUEST_NODATA
, OBJ_REQUEST_BIO
, OBJ_REQUEST_PAGES
212 OBJ_REQ_DONE
, /* completion flag: not done = 0, done = 1 */
213 OBJ_REQ_IMG_DATA
, /* object usage: standalone = 0, image = 1 */
214 OBJ_REQ_KNOWN
, /* EXISTS flag valid: no = 0, yes = 1 */
215 OBJ_REQ_EXISTS
, /* target exists: no = 0, yes = 1 */
218 struct rbd_obj_request
{
219 const char *object_name
;
220 u64 offset
; /* object start byte */
221 u64 length
; /* bytes from offset */
225 * An object request associated with an image will have its
226 * img_data flag set; a standalone object request will not.
228 * A standalone object request will have which == BAD_WHICH
229 * and a null obj_request pointer.
231 * An object request initiated in support of a layered image
232 * object (to check for its existence before a write) will
233 * have which == BAD_WHICH and a non-null obj_request pointer.
235 * Finally, an object request for rbd image data will have
236 * which != BAD_WHICH, and will have a non-null img_request
237 * pointer. The value of which will be in the range
238 * 0..(img_request->obj_request_count-1).
241 struct rbd_obj_request
*obj_request
; /* STAT op */
243 struct rbd_img_request
*img_request
;
245 /* links for img_request->obj_requests list */
246 struct list_head links
;
249 u32 which
; /* posn image request list */
251 enum obj_request_type type
;
253 struct bio
*bio_list
;
259 struct page
**copyup_pages
;
260 u32 copyup_page_count
;
262 struct ceph_osd_request
*osd_req
;
264 u64 xferred
; /* bytes transferred */
267 rbd_obj_callback_t callback
;
268 struct completion completion
;
274 IMG_REQ_WRITE
, /* I/O direction: read = 0, write = 1 */
275 IMG_REQ_CHILD
, /* initiator: block = 0, child image = 1 */
276 IMG_REQ_LAYERED
, /* ENOENT handling: normal = 0, layered = 1 */
279 struct rbd_img_request
{
280 struct rbd_device
*rbd_dev
;
281 u64 offset
; /* starting image byte offset */
282 u64 length
; /* byte count from offset */
285 u64 snap_id
; /* for reads */
286 struct ceph_snap_context
*snapc
; /* for writes */
289 struct request
*rq
; /* block request */
290 struct rbd_obj_request
*obj_request
; /* obj req initiator */
292 struct page
**copyup_pages
;
293 u32 copyup_page_count
;
294 spinlock_t completion_lock
;/* protects next_completion */
296 rbd_img_callback_t callback
;
297 u64 xferred
;/* aggregate bytes transferred */
298 int result
; /* first nonzero obj_request result */
300 u32 obj_request_count
;
301 struct list_head obj_requests
; /* rbd_obj_request structs */
306 #define for_each_obj_request(ireq, oreq) \
307 list_for_each_entry(oreq, &(ireq)->obj_requests, links)
308 #define for_each_obj_request_from(ireq, oreq) \
309 list_for_each_entry_from(oreq, &(ireq)->obj_requests, links)
310 #define for_each_obj_request_safe(ireq, oreq, n) \
311 list_for_each_entry_safe_reverse(oreq, n, &(ireq)->obj_requests, links)
323 int dev_id
; /* blkdev unique id */
325 int major
; /* blkdev assigned major */
326 struct gendisk
*disk
; /* blkdev's gendisk and rq */
328 u32 image_format
; /* Either 1 or 2 */
329 struct rbd_client
*rbd_client
;
331 char name
[DEV_NAME_LEN
]; /* blkdev name, e.g. rbd3 */
333 spinlock_t lock
; /* queue, flags, open_count */
335 struct rbd_image_header header
;
336 unsigned long flags
; /* possibly lock protected */
337 struct rbd_spec
*spec
;
341 struct ceph_file_layout layout
;
343 struct ceph_osd_event
*watch_event
;
344 struct rbd_obj_request
*watch_request
;
346 struct rbd_spec
*parent_spec
;
349 struct rbd_device
*parent
;
351 /* protects updating the header */
352 struct rw_semaphore header_rwsem
;
354 struct rbd_mapping mapping
;
356 struct list_head node
;
360 unsigned long open_count
; /* protected by lock */
364 * Flag bits for rbd_dev->flags. If atomicity is required,
365 * rbd_dev->lock is used to protect access.
367 * Currently, only the "removing" flag (which is coupled with the
368 * "open_count" field) requires atomic access.
371 RBD_DEV_FLAG_EXISTS
, /* mapped snapshot has not been deleted */
372 RBD_DEV_FLAG_REMOVING
, /* this mapping is being removed */
375 static DEFINE_MUTEX(client_mutex
); /* Serialize client creation */
377 static LIST_HEAD(rbd_dev_list
); /* devices */
378 static DEFINE_SPINLOCK(rbd_dev_list_lock
);
380 static LIST_HEAD(rbd_client_list
); /* clients */
381 static DEFINE_SPINLOCK(rbd_client_list_lock
);
383 /* Slab caches for frequently-allocated structures */
385 static struct kmem_cache
*rbd_img_request_cache
;
386 static struct kmem_cache
*rbd_obj_request_cache
;
387 static struct kmem_cache
*rbd_segment_name_cache
;
389 static int rbd_img_request_submit(struct rbd_img_request
*img_request
);
391 static void rbd_dev_device_release(struct device
*dev
);
393 static ssize_t
rbd_add(struct bus_type
*bus
, const char *buf
,
395 static ssize_t
rbd_remove(struct bus_type
*bus
, const char *buf
,
397 static int rbd_dev_image_probe(struct rbd_device
*rbd_dev
, bool mapping
);
398 static void rbd_spec_put(struct rbd_spec
*spec
);
400 static BUS_ATTR(add
, S_IWUSR
, NULL
, rbd_add
);
401 static BUS_ATTR(remove
, S_IWUSR
, NULL
, rbd_remove
);
403 static struct attribute
*rbd_bus_attrs
[] = {
405 &bus_attr_remove
.attr
,
408 ATTRIBUTE_GROUPS(rbd_bus
);
410 static struct bus_type rbd_bus_type
= {
412 .bus_groups
= rbd_bus_groups
,
415 static void rbd_root_dev_release(struct device
*dev
)
419 static struct device rbd_root_dev
= {
421 .release
= rbd_root_dev_release
,
424 static __printf(2, 3)
425 void rbd_warn(struct rbd_device
*rbd_dev
, const char *fmt
, ...)
427 struct va_format vaf
;
435 printk(KERN_WARNING
"%s: %pV\n", RBD_DRV_NAME
, &vaf
);
436 else if (rbd_dev
->disk
)
437 printk(KERN_WARNING
"%s: %s: %pV\n",
438 RBD_DRV_NAME
, rbd_dev
->disk
->disk_name
, &vaf
);
439 else if (rbd_dev
->spec
&& rbd_dev
->spec
->image_name
)
440 printk(KERN_WARNING
"%s: image %s: %pV\n",
441 RBD_DRV_NAME
, rbd_dev
->spec
->image_name
, &vaf
);
442 else if (rbd_dev
->spec
&& rbd_dev
->spec
->image_id
)
443 printk(KERN_WARNING
"%s: id %s: %pV\n",
444 RBD_DRV_NAME
, rbd_dev
->spec
->image_id
, &vaf
);
446 printk(KERN_WARNING
"%s: rbd_dev %p: %pV\n",
447 RBD_DRV_NAME
, rbd_dev
, &vaf
);
452 #define rbd_assert(expr) \
453 if (unlikely(!(expr))) { \
454 printk(KERN_ERR "\nAssertion failure in %s() " \
456 "\trbd_assert(%s);\n\n", \
457 __func__, __LINE__, #expr); \
460 #else /* !RBD_DEBUG */
461 # define rbd_assert(expr) ((void) 0)
462 #endif /* !RBD_DEBUG */
464 static int rbd_img_obj_request_submit(struct rbd_obj_request
*obj_request
);
465 static void rbd_img_parent_read(struct rbd_obj_request
*obj_request
);
466 static void rbd_dev_remove_parent(struct rbd_device
*rbd_dev
);
468 static int rbd_dev_refresh(struct rbd_device
*rbd_dev
);
469 static int rbd_dev_v2_header_onetime(struct rbd_device
*rbd_dev
);
470 static int rbd_dev_v2_header_info(struct rbd_device
*rbd_dev
);
471 static const char *rbd_dev_v2_snap_name(struct rbd_device
*rbd_dev
,
473 static int _rbd_dev_v2_snap_size(struct rbd_device
*rbd_dev
, u64 snap_id
,
474 u8
*order
, u64
*snap_size
);
475 static int _rbd_dev_v2_snap_features(struct rbd_device
*rbd_dev
, u64 snap_id
,
477 static u64
rbd_snap_id_by_name(struct rbd_device
*rbd_dev
, const char *name
);
479 static int rbd_open(struct block_device
*bdev
, fmode_t mode
)
481 struct rbd_device
*rbd_dev
= bdev
->bd_disk
->private_data
;
482 bool removing
= false;
484 if ((mode
& FMODE_WRITE
) && rbd_dev
->mapping
.read_only
)
487 spin_lock_irq(&rbd_dev
->lock
);
488 if (test_bit(RBD_DEV_FLAG_REMOVING
, &rbd_dev
->flags
))
491 rbd_dev
->open_count
++;
492 spin_unlock_irq(&rbd_dev
->lock
);
496 (void) get_device(&rbd_dev
->dev
);
497 set_device_ro(bdev
, rbd_dev
->mapping
.read_only
);
502 static void rbd_release(struct gendisk
*disk
, fmode_t mode
)
504 struct rbd_device
*rbd_dev
= disk
->private_data
;
505 unsigned long open_count_before
;
507 spin_lock_irq(&rbd_dev
->lock
);
508 open_count_before
= rbd_dev
->open_count
--;
509 spin_unlock_irq(&rbd_dev
->lock
);
510 rbd_assert(open_count_before
> 0);
512 put_device(&rbd_dev
->dev
);
515 static const struct block_device_operations rbd_bd_ops
= {
516 .owner
= THIS_MODULE
,
518 .release
= rbd_release
,
522 * Initialize an rbd client instance. Success or not, this function
523 * consumes ceph_opts. Caller holds client_mutex.
525 static struct rbd_client
*rbd_client_create(struct ceph_options
*ceph_opts
)
527 struct rbd_client
*rbdc
;
530 dout("%s:\n", __func__
);
531 rbdc
= kmalloc(sizeof(struct rbd_client
), GFP_KERNEL
);
535 kref_init(&rbdc
->kref
);
536 INIT_LIST_HEAD(&rbdc
->node
);
538 rbdc
->client
= ceph_create_client(ceph_opts
, rbdc
, 0, 0);
539 if (IS_ERR(rbdc
->client
))
541 ceph_opts
= NULL
; /* Now rbdc->client is responsible for ceph_opts */
543 ret
= ceph_open_session(rbdc
->client
);
547 spin_lock(&rbd_client_list_lock
);
548 list_add_tail(&rbdc
->node
, &rbd_client_list
);
549 spin_unlock(&rbd_client_list_lock
);
551 dout("%s: rbdc %p\n", __func__
, rbdc
);
555 ceph_destroy_client(rbdc
->client
);
560 ceph_destroy_options(ceph_opts
);
561 dout("%s: error %d\n", __func__
, ret
);
566 static struct rbd_client
*__rbd_get_client(struct rbd_client
*rbdc
)
568 kref_get(&rbdc
->kref
);
574 * Find a ceph client with specific addr and configuration. If
575 * found, bump its reference count.
577 static struct rbd_client
*rbd_client_find(struct ceph_options
*ceph_opts
)
579 struct rbd_client
*client_node
;
582 if (ceph_opts
->flags
& CEPH_OPT_NOSHARE
)
585 spin_lock(&rbd_client_list_lock
);
586 list_for_each_entry(client_node
, &rbd_client_list
, node
) {
587 if (!ceph_compare_options(ceph_opts
, client_node
->client
)) {
588 __rbd_get_client(client_node
);
594 spin_unlock(&rbd_client_list_lock
);
596 return found
? client_node
: NULL
;
606 /* string args above */
609 /* Boolean args above */
613 static match_table_t rbd_opts_tokens
= {
615 /* string args above */
616 {Opt_read_only
, "read_only"},
617 {Opt_read_only
, "ro"}, /* Alternate spelling */
618 {Opt_read_write
, "read_write"},
619 {Opt_read_write
, "rw"}, /* Alternate spelling */
620 /* Boolean args above */
628 #define RBD_READ_ONLY_DEFAULT false
630 static int parse_rbd_opts_token(char *c
, void *private)
632 struct rbd_options
*rbd_opts
= private;
633 substring_t argstr
[MAX_OPT_ARGS
];
634 int token
, intval
, ret
;
636 token
= match_token(c
, rbd_opts_tokens
, argstr
);
640 if (token
< Opt_last_int
) {
641 ret
= match_int(&argstr
[0], &intval
);
643 pr_err("bad mount option arg (not int) "
647 dout("got int token %d val %d\n", token
, intval
);
648 } else if (token
> Opt_last_int
&& token
< Opt_last_string
) {
649 dout("got string token %d val %s\n", token
,
651 } else if (token
> Opt_last_string
&& token
< Opt_last_bool
) {
652 dout("got Boolean token %d\n", token
);
654 dout("got token %d\n", token
);
659 rbd_opts
->read_only
= true;
662 rbd_opts
->read_only
= false;
672 * Get a ceph client with specific addr and configuration, if one does
673 * not exist create it. Either way, ceph_opts is consumed by this
676 static struct rbd_client
*rbd_get_client(struct ceph_options
*ceph_opts
)
678 struct rbd_client
*rbdc
;
680 mutex_lock_nested(&client_mutex
, SINGLE_DEPTH_NESTING
);
681 rbdc
= rbd_client_find(ceph_opts
);
682 if (rbdc
) /* using an existing client */
683 ceph_destroy_options(ceph_opts
);
685 rbdc
= rbd_client_create(ceph_opts
);
686 mutex_unlock(&client_mutex
);
692 * Destroy ceph client
694 * Caller must hold rbd_client_list_lock.
696 static void rbd_client_release(struct kref
*kref
)
698 struct rbd_client
*rbdc
= container_of(kref
, struct rbd_client
, kref
);
700 dout("%s: rbdc %p\n", __func__
, rbdc
);
701 spin_lock(&rbd_client_list_lock
);
702 list_del(&rbdc
->node
);
703 spin_unlock(&rbd_client_list_lock
);
705 ceph_destroy_client(rbdc
->client
);
710 * Drop reference to ceph client node. If it's not referenced anymore, release
713 static void rbd_put_client(struct rbd_client
*rbdc
)
716 kref_put(&rbdc
->kref
, rbd_client_release
);
719 static bool rbd_image_format_valid(u32 image_format
)
721 return image_format
== 1 || image_format
== 2;
724 static bool rbd_dev_ondisk_valid(struct rbd_image_header_ondisk
*ondisk
)
729 /* The header has to start with the magic rbd header text */
730 if (memcmp(&ondisk
->text
, RBD_HEADER_TEXT
, sizeof (RBD_HEADER_TEXT
)))
733 /* The bio layer requires at least sector-sized I/O */
735 if (ondisk
->options
.order
< SECTOR_SHIFT
)
738 /* If we use u64 in a few spots we may be able to loosen this */
740 if (ondisk
->options
.order
> 8 * sizeof (int) - 1)
744 * The size of a snapshot header has to fit in a size_t, and
745 * that limits the number of snapshots.
747 snap_count
= le32_to_cpu(ondisk
->snap_count
);
748 size
= SIZE_MAX
- sizeof (struct ceph_snap_context
);
749 if (snap_count
> size
/ sizeof (__le64
))
753 * Not only that, but the size of the entire the snapshot
754 * header must also be representable in a size_t.
756 size
-= snap_count
* sizeof (__le64
);
757 if ((u64
) size
< le64_to_cpu(ondisk
->snap_names_len
))
764 * Fill an rbd image header with information from the given format 1
767 static int rbd_header_from_disk(struct rbd_device
*rbd_dev
,
768 struct rbd_image_header_ondisk
*ondisk
)
770 struct rbd_image_header
*header
= &rbd_dev
->header
;
771 bool first_time
= header
->object_prefix
== NULL
;
772 struct ceph_snap_context
*snapc
;
773 char *object_prefix
= NULL
;
774 char *snap_names
= NULL
;
775 u64
*snap_sizes
= NULL
;
781 /* Allocate this now to avoid having to handle failure below */
786 len
= strnlen(ondisk
->object_prefix
,
787 sizeof (ondisk
->object_prefix
));
788 object_prefix
= kmalloc(len
+ 1, GFP_KERNEL
);
791 memcpy(object_prefix
, ondisk
->object_prefix
, len
);
792 object_prefix
[len
] = '\0';
795 /* Allocate the snapshot context and fill it in */
797 snap_count
= le32_to_cpu(ondisk
->snap_count
);
798 snapc
= ceph_create_snap_context(snap_count
, GFP_KERNEL
);
801 snapc
->seq
= le64_to_cpu(ondisk
->snap_seq
);
803 struct rbd_image_snap_ondisk
*snaps
;
804 u64 snap_names_len
= le64_to_cpu(ondisk
->snap_names_len
);
806 /* We'll keep a copy of the snapshot names... */
808 if (snap_names_len
> (u64
)SIZE_MAX
)
810 snap_names
= kmalloc(snap_names_len
, GFP_KERNEL
);
814 /* ...as well as the array of their sizes. */
816 size
= snap_count
* sizeof (*header
->snap_sizes
);
817 snap_sizes
= kmalloc(size
, GFP_KERNEL
);
822 * Copy the names, and fill in each snapshot's id
825 * Note that rbd_dev_v1_header_info() guarantees the
826 * ondisk buffer we're working with has
827 * snap_names_len bytes beyond the end of the
828 * snapshot id array, this memcpy() is safe.
830 memcpy(snap_names
, &ondisk
->snaps
[snap_count
], snap_names_len
);
831 snaps
= ondisk
->snaps
;
832 for (i
= 0; i
< snap_count
; i
++) {
833 snapc
->snaps
[i
] = le64_to_cpu(snaps
[i
].id
);
834 snap_sizes
[i
] = le64_to_cpu(snaps
[i
].image_size
);
838 /* We won't fail any more, fill in the header */
841 header
->object_prefix
= object_prefix
;
842 header
->obj_order
= ondisk
->options
.order
;
843 header
->crypt_type
= ondisk
->options
.crypt_type
;
844 header
->comp_type
= ondisk
->options
.comp_type
;
845 /* The rest aren't used for format 1 images */
846 header
->stripe_unit
= 0;
847 header
->stripe_count
= 0;
848 header
->features
= 0;
850 ceph_put_snap_context(header
->snapc
);
851 kfree(header
->snap_names
);
852 kfree(header
->snap_sizes
);
855 /* The remaining fields always get updated (when we refresh) */
857 header
->image_size
= le64_to_cpu(ondisk
->image_size
);
858 header
->snapc
= snapc
;
859 header
->snap_names
= snap_names
;
860 header
->snap_sizes
= snap_sizes
;
862 /* Make sure mapping size is consistent with header info */
864 if (rbd_dev
->spec
->snap_id
== CEPH_NOSNAP
|| first_time
)
865 if (rbd_dev
->mapping
.size
!= header
->image_size
)
866 rbd_dev
->mapping
.size
= header
->image_size
;
874 ceph_put_snap_context(snapc
);
875 kfree(object_prefix
);
880 static const char *_rbd_dev_v1_snap_name(struct rbd_device
*rbd_dev
, u32 which
)
882 const char *snap_name
;
884 rbd_assert(which
< rbd_dev
->header
.snapc
->num_snaps
);
886 /* Skip over names until we find the one we are looking for */
888 snap_name
= rbd_dev
->header
.snap_names
;
890 snap_name
+= strlen(snap_name
) + 1;
892 return kstrdup(snap_name
, GFP_KERNEL
);
896 * Snapshot id comparison function for use with qsort()/bsearch().
897 * Note that result is for snapshots in *descending* order.
899 static int snapid_compare_reverse(const void *s1
, const void *s2
)
901 u64 snap_id1
= *(u64
*)s1
;
902 u64 snap_id2
= *(u64
*)s2
;
904 if (snap_id1
< snap_id2
)
906 return snap_id1
== snap_id2
? 0 : -1;
910 * Search a snapshot context to see if the given snapshot id is
913 * Returns the position of the snapshot id in the array if it's found,
914 * or BAD_SNAP_INDEX otherwise.
916 * Note: The snapshot array is in kept sorted (by the osd) in
917 * reverse order, highest snapshot id first.
919 static u32
rbd_dev_snap_index(struct rbd_device
*rbd_dev
, u64 snap_id
)
921 struct ceph_snap_context
*snapc
= rbd_dev
->header
.snapc
;
924 found
= bsearch(&snap_id
, &snapc
->snaps
, snapc
->num_snaps
,
925 sizeof (snap_id
), snapid_compare_reverse
);
927 return found
? (u32
)(found
- &snapc
->snaps
[0]) : BAD_SNAP_INDEX
;
930 static const char *rbd_dev_v1_snap_name(struct rbd_device
*rbd_dev
,
934 const char *snap_name
;
936 which
= rbd_dev_snap_index(rbd_dev
, snap_id
);
937 if (which
== BAD_SNAP_INDEX
)
938 return ERR_PTR(-ENOENT
);
940 snap_name
= _rbd_dev_v1_snap_name(rbd_dev
, which
);
941 return snap_name
? snap_name
: ERR_PTR(-ENOMEM
);
944 static const char *rbd_snap_name(struct rbd_device
*rbd_dev
, u64 snap_id
)
946 if (snap_id
== CEPH_NOSNAP
)
947 return RBD_SNAP_HEAD_NAME
;
949 rbd_assert(rbd_image_format_valid(rbd_dev
->image_format
));
950 if (rbd_dev
->image_format
== 1)
951 return rbd_dev_v1_snap_name(rbd_dev
, snap_id
);
953 return rbd_dev_v2_snap_name(rbd_dev
, snap_id
);
956 static int rbd_snap_size(struct rbd_device
*rbd_dev
, u64 snap_id
,
959 rbd_assert(rbd_image_format_valid(rbd_dev
->image_format
));
960 if (snap_id
== CEPH_NOSNAP
) {
961 *snap_size
= rbd_dev
->header
.image_size
;
962 } else if (rbd_dev
->image_format
== 1) {
965 which
= rbd_dev_snap_index(rbd_dev
, snap_id
);
966 if (which
== BAD_SNAP_INDEX
)
969 *snap_size
= rbd_dev
->header
.snap_sizes
[which
];
974 ret
= _rbd_dev_v2_snap_size(rbd_dev
, snap_id
, NULL
, &size
);
983 static int rbd_snap_features(struct rbd_device
*rbd_dev
, u64 snap_id
,
986 rbd_assert(rbd_image_format_valid(rbd_dev
->image_format
));
987 if (snap_id
== CEPH_NOSNAP
) {
988 *snap_features
= rbd_dev
->header
.features
;
989 } else if (rbd_dev
->image_format
== 1) {
990 *snap_features
= 0; /* No features for format 1 */
995 ret
= _rbd_dev_v2_snap_features(rbd_dev
, snap_id
, &features
);
999 *snap_features
= features
;
1004 static int rbd_dev_mapping_set(struct rbd_device
*rbd_dev
)
1006 u64 snap_id
= rbd_dev
->spec
->snap_id
;
1011 ret
= rbd_snap_size(rbd_dev
, snap_id
, &size
);
1014 ret
= rbd_snap_features(rbd_dev
, snap_id
, &features
);
1018 rbd_dev
->mapping
.size
= size
;
1019 rbd_dev
->mapping
.features
= features
;
1024 static void rbd_dev_mapping_clear(struct rbd_device
*rbd_dev
)
1026 rbd_dev
->mapping
.size
= 0;
1027 rbd_dev
->mapping
.features
= 0;
1030 static const char *rbd_segment_name(struct rbd_device
*rbd_dev
, u64 offset
)
1037 name
= kmem_cache_alloc(rbd_segment_name_cache
, GFP_NOIO
);
1040 segment
= offset
>> rbd_dev
->header
.obj_order
;
1041 name_format
= "%s.%012llx";
1042 if (rbd_dev
->image_format
== 2)
1043 name_format
= "%s.%016llx";
1044 ret
= snprintf(name
, MAX_OBJ_NAME_SIZE
+ 1, name_format
,
1045 rbd_dev
->header
.object_prefix
, segment
);
1046 if (ret
< 0 || ret
> MAX_OBJ_NAME_SIZE
) {
1047 pr_err("error formatting segment name for #%llu (%d)\n",
1056 static void rbd_segment_name_free(const char *name
)
1058 /* The explicit cast here is needed to drop the const qualifier */
1060 kmem_cache_free(rbd_segment_name_cache
, (void *)name
);
1063 static u64
rbd_segment_offset(struct rbd_device
*rbd_dev
, u64 offset
)
1065 u64 segment_size
= (u64
) 1 << rbd_dev
->header
.obj_order
;
1067 return offset
& (segment_size
- 1);
1070 static u64
rbd_segment_length(struct rbd_device
*rbd_dev
,
1071 u64 offset
, u64 length
)
1073 u64 segment_size
= (u64
) 1 << rbd_dev
->header
.obj_order
;
1075 offset
&= segment_size
- 1;
1077 rbd_assert(length
<= U64_MAX
- offset
);
1078 if (offset
+ length
> segment_size
)
1079 length
= segment_size
- offset
;
1085 * returns the size of an object in the image
1087 static u64
rbd_obj_bytes(struct rbd_image_header
*header
)
1089 return 1 << header
->obj_order
;
1096 static void bio_chain_put(struct bio
*chain
)
1102 chain
= chain
->bi_next
;
1108 * zeros a bio chain, starting at specific offset
1110 static void zero_bio_chain(struct bio
*chain
, int start_ofs
)
1113 unsigned long flags
;
1119 bio_for_each_segment(bv
, chain
, i
) {
1120 if (pos
+ bv
->bv_len
> start_ofs
) {
1121 int remainder
= max(start_ofs
- pos
, 0);
1122 buf
= bvec_kmap_irq(bv
, &flags
);
1123 memset(buf
+ remainder
, 0,
1124 bv
->bv_len
- remainder
);
1125 flush_dcache_page(bv
->bv_page
);
1126 bvec_kunmap_irq(buf
, &flags
);
1131 chain
= chain
->bi_next
;
1136 * similar to zero_bio_chain(), zeros data defined by a page array,
1137 * starting at the given byte offset from the start of the array and
1138 * continuing up to the given end offset. The pages array is
1139 * assumed to be big enough to hold all bytes up to the end.
1141 static void zero_pages(struct page
**pages
, u64 offset
, u64 end
)
1143 struct page
**page
= &pages
[offset
>> PAGE_SHIFT
];
1145 rbd_assert(end
> offset
);
1146 rbd_assert(end
- offset
<= (u64
)SIZE_MAX
);
1147 while (offset
< end
) {
1150 unsigned long flags
;
1153 page_offset
= offset
& ~PAGE_MASK
;
1154 length
= min_t(size_t, PAGE_SIZE
- page_offset
, end
- offset
);
1155 local_irq_save(flags
);
1156 kaddr
= kmap_atomic(*page
);
1157 memset(kaddr
+ page_offset
, 0, length
);
1158 flush_dcache_page(*page
);
1159 kunmap_atomic(kaddr
);
1160 local_irq_restore(flags
);
1168 * Clone a portion of a bio, starting at the given byte offset
1169 * and continuing for the number of bytes indicated.
1171 static struct bio
*bio_clone_range(struct bio
*bio_src
,
1172 unsigned int offset
,
1180 unsigned short end_idx
;
1181 unsigned short vcnt
;
1184 /* Handle the easy case for the caller */
1186 if (!offset
&& len
== bio_src
->bi_size
)
1187 return bio_clone(bio_src
, gfpmask
);
1189 if (WARN_ON_ONCE(!len
))
1191 if (WARN_ON_ONCE(len
> bio_src
->bi_size
))
1193 if (WARN_ON_ONCE(offset
> bio_src
->bi_size
- len
))
1196 /* Find first affected segment... */
1199 bio_for_each_segment(bv
, bio_src
, idx
) {
1200 if (resid
< bv
->bv_len
)
1202 resid
-= bv
->bv_len
;
1206 /* ...and the last affected segment */
1209 __bio_for_each_segment(bv
, bio_src
, end_idx
, idx
) {
1210 if (resid
<= bv
->bv_len
)
1212 resid
-= bv
->bv_len
;
1214 vcnt
= end_idx
- idx
+ 1;
1216 /* Build the clone */
1218 bio
= bio_alloc(gfpmask
, (unsigned int) vcnt
);
1220 return NULL
; /* ENOMEM */
1222 bio
->bi_bdev
= bio_src
->bi_bdev
;
1223 bio
->bi_sector
= bio_src
->bi_sector
+ (offset
>> SECTOR_SHIFT
);
1224 bio
->bi_rw
= bio_src
->bi_rw
;
1225 bio
->bi_flags
|= 1 << BIO_CLONED
;
1228 * Copy over our part of the bio_vec, then update the first
1229 * and last (or only) entries.
1231 memcpy(&bio
->bi_io_vec
[0], &bio_src
->bi_io_vec
[idx
],
1232 vcnt
* sizeof (struct bio_vec
));
1233 bio
->bi_io_vec
[0].bv_offset
+= voff
;
1235 bio
->bi_io_vec
[0].bv_len
-= voff
;
1236 bio
->bi_io_vec
[vcnt
- 1].bv_len
= resid
;
1238 bio
->bi_io_vec
[0].bv_len
= len
;
1241 bio
->bi_vcnt
= vcnt
;
1249 * Clone a portion of a bio chain, starting at the given byte offset
1250 * into the first bio in the source chain and continuing for the
1251 * number of bytes indicated. The result is another bio chain of
1252 * exactly the given length, or a null pointer on error.
1254 * The bio_src and offset parameters are both in-out. On entry they
1255 * refer to the first source bio and the offset into that bio where
1256 * the start of data to be cloned is located.
1258 * On return, bio_src is updated to refer to the bio in the source
1259 * chain that contains first un-cloned byte, and *offset will
1260 * contain the offset of that byte within that bio.
1262 static struct bio
*bio_chain_clone_range(struct bio
**bio_src
,
1263 unsigned int *offset
,
1267 struct bio
*bi
= *bio_src
;
1268 unsigned int off
= *offset
;
1269 struct bio
*chain
= NULL
;
1272 /* Build up a chain of clone bios up to the limit */
1274 if (!bi
|| off
>= bi
->bi_size
|| !len
)
1275 return NULL
; /* Nothing to clone */
1279 unsigned int bi_size
;
1283 rbd_warn(NULL
, "bio_chain exhausted with %u left", len
);
1284 goto out_err
; /* EINVAL; ran out of bio's */
1286 bi_size
= min_t(unsigned int, bi
->bi_size
- off
, len
);
1287 bio
= bio_clone_range(bi
, off
, bi_size
, gfpmask
);
1289 goto out_err
; /* ENOMEM */
1292 end
= &bio
->bi_next
;
1295 if (off
== bi
->bi_size
) {
1306 bio_chain_put(chain
);
1312 * The default/initial value for all object request flags is 0. For
1313 * each flag, once its value is set to 1 it is never reset to 0
1316 static void obj_request_img_data_set(struct rbd_obj_request
*obj_request
)
1318 if (test_and_set_bit(OBJ_REQ_IMG_DATA
, &obj_request
->flags
)) {
1319 struct rbd_device
*rbd_dev
;
1321 rbd_dev
= obj_request
->img_request
->rbd_dev
;
1322 rbd_warn(rbd_dev
, "obj_request %p already marked img_data\n",
1327 static bool obj_request_img_data_test(struct rbd_obj_request
*obj_request
)
1330 return test_bit(OBJ_REQ_IMG_DATA
, &obj_request
->flags
) != 0;
1333 static void obj_request_done_set(struct rbd_obj_request
*obj_request
)
1335 if (test_and_set_bit(OBJ_REQ_DONE
, &obj_request
->flags
)) {
1336 struct rbd_device
*rbd_dev
= NULL
;
1338 if (obj_request_img_data_test(obj_request
))
1339 rbd_dev
= obj_request
->img_request
->rbd_dev
;
1340 rbd_warn(rbd_dev
, "obj_request %p already marked done\n",
1345 static bool obj_request_done_test(struct rbd_obj_request
*obj_request
)
1348 return test_bit(OBJ_REQ_DONE
, &obj_request
->flags
) != 0;
1352 * This sets the KNOWN flag after (possibly) setting the EXISTS
1353 * flag. The latter is set based on the "exists" value provided.
1355 * Note that for our purposes once an object exists it never goes
1356 * away again. It's possible that the response from two existence
1357 * checks are separated by the creation of the target object, and
1358 * the first ("doesn't exist") response arrives *after* the second
1359 * ("does exist"). In that case we ignore the second one.
1361 static void obj_request_existence_set(struct rbd_obj_request
*obj_request
,
1365 set_bit(OBJ_REQ_EXISTS
, &obj_request
->flags
);
1366 set_bit(OBJ_REQ_KNOWN
, &obj_request
->flags
);
1370 static bool obj_request_known_test(struct rbd_obj_request
*obj_request
)
1373 return test_bit(OBJ_REQ_KNOWN
, &obj_request
->flags
) != 0;
1376 static bool obj_request_exists_test(struct rbd_obj_request
*obj_request
)
1379 return test_bit(OBJ_REQ_EXISTS
, &obj_request
->flags
) != 0;
1382 static bool obj_request_overlaps_parent(struct rbd_obj_request
*obj_request
)
1384 struct rbd_device
*rbd_dev
= obj_request
->img_request
->rbd_dev
;
1386 return obj_request
->img_offset
<
1387 round_up(rbd_dev
->parent_overlap
, rbd_obj_bytes(&rbd_dev
->header
));
1390 static void rbd_obj_request_get(struct rbd_obj_request
*obj_request
)
1392 dout("%s: obj %p (was %d)\n", __func__
, obj_request
,
1393 atomic_read(&obj_request
->kref
.refcount
));
1394 kref_get(&obj_request
->kref
);
1397 static void rbd_obj_request_destroy(struct kref
*kref
);
1398 static void rbd_obj_request_put(struct rbd_obj_request
*obj_request
)
1400 rbd_assert(obj_request
!= NULL
);
1401 dout("%s: obj %p (was %d)\n", __func__
, obj_request
,
1402 atomic_read(&obj_request
->kref
.refcount
));
1403 kref_put(&obj_request
->kref
, rbd_obj_request_destroy
);
1406 static void rbd_img_request_get(struct rbd_img_request
*img_request
)
1408 dout("%s: img %p (was %d)\n", __func__
, img_request
,
1409 atomic_read(&img_request
->kref
.refcount
));
1410 kref_get(&img_request
->kref
);
1413 static bool img_request_child_test(struct rbd_img_request
*img_request
);
1414 static void rbd_parent_request_destroy(struct kref
*kref
);
1415 static void rbd_img_request_destroy(struct kref
*kref
);
1416 static void rbd_img_request_put(struct rbd_img_request
*img_request
)
1418 rbd_assert(img_request
!= NULL
);
1419 dout("%s: img %p (was %d)\n", __func__
, img_request
,
1420 atomic_read(&img_request
->kref
.refcount
));
1421 if (img_request_child_test(img_request
))
1422 kref_put(&img_request
->kref
, rbd_parent_request_destroy
);
1424 kref_put(&img_request
->kref
, rbd_img_request_destroy
);
1427 static inline void rbd_img_obj_request_add(struct rbd_img_request
*img_request
,
1428 struct rbd_obj_request
*obj_request
)
1430 rbd_assert(obj_request
->img_request
== NULL
);
1432 /* Image request now owns object's original reference */
1433 obj_request
->img_request
= img_request
;
1434 obj_request
->which
= img_request
->obj_request_count
;
1435 rbd_assert(!obj_request_img_data_test(obj_request
));
1436 obj_request_img_data_set(obj_request
);
1437 rbd_assert(obj_request
->which
!= BAD_WHICH
);
1438 img_request
->obj_request_count
++;
1439 list_add_tail(&obj_request
->links
, &img_request
->obj_requests
);
1440 dout("%s: img %p obj %p w=%u\n", __func__
, img_request
, obj_request
,
1441 obj_request
->which
);
1444 static inline void rbd_img_obj_request_del(struct rbd_img_request
*img_request
,
1445 struct rbd_obj_request
*obj_request
)
1447 rbd_assert(obj_request
->which
!= BAD_WHICH
);
1449 dout("%s: img %p obj %p w=%u\n", __func__
, img_request
, obj_request
,
1450 obj_request
->which
);
1451 list_del(&obj_request
->links
);
1452 rbd_assert(img_request
->obj_request_count
> 0);
1453 img_request
->obj_request_count
--;
1454 rbd_assert(obj_request
->which
== img_request
->obj_request_count
);
1455 obj_request
->which
= BAD_WHICH
;
1456 rbd_assert(obj_request_img_data_test(obj_request
));
1457 rbd_assert(obj_request
->img_request
== img_request
);
1458 obj_request
->img_request
= NULL
;
1459 obj_request
->callback
= NULL
;
1460 rbd_obj_request_put(obj_request
);
1463 static bool obj_request_type_valid(enum obj_request_type type
)
1466 case OBJ_REQUEST_NODATA
:
1467 case OBJ_REQUEST_BIO
:
1468 case OBJ_REQUEST_PAGES
:
1475 static int rbd_obj_request_submit(struct ceph_osd_client
*osdc
,
1476 struct rbd_obj_request
*obj_request
)
1478 dout("%s: osdc %p obj %p\n", __func__
, osdc
, obj_request
);
1480 return ceph_osdc_start_request(osdc
, obj_request
->osd_req
, false);
1483 static void rbd_img_request_complete(struct rbd_img_request
*img_request
)
1486 dout("%s: img %p\n", __func__
, img_request
);
1489 * If no error occurred, compute the aggregate transfer
1490 * count for the image request. We could instead use
1491 * atomic64_cmpxchg() to update it as each object request
1492 * completes; not clear which way is better off hand.
1494 if (!img_request
->result
) {
1495 struct rbd_obj_request
*obj_request
;
1498 for_each_obj_request(img_request
, obj_request
)
1499 xferred
+= obj_request
->xferred
;
1500 img_request
->xferred
= xferred
;
1503 if (img_request
->callback
)
1504 img_request
->callback(img_request
);
1506 rbd_img_request_put(img_request
);
1509 /* Caller is responsible for rbd_obj_request_destroy(obj_request) */
1511 static int rbd_obj_request_wait(struct rbd_obj_request
*obj_request
)
1513 dout("%s: obj %p\n", __func__
, obj_request
);
1515 return wait_for_completion_interruptible(&obj_request
->completion
);
1519 * The default/initial value for all image request flags is 0. Each
1520 * is conditionally set to 1 at image request initialization time
1521 * and currently never change thereafter.
1523 static void img_request_write_set(struct rbd_img_request
*img_request
)
1525 set_bit(IMG_REQ_WRITE
, &img_request
->flags
);
1529 static bool img_request_write_test(struct rbd_img_request
*img_request
)
1532 return test_bit(IMG_REQ_WRITE
, &img_request
->flags
) != 0;
1535 static void img_request_child_set(struct rbd_img_request
*img_request
)
1537 set_bit(IMG_REQ_CHILD
, &img_request
->flags
);
1541 static void img_request_child_clear(struct rbd_img_request
*img_request
)
1543 clear_bit(IMG_REQ_CHILD
, &img_request
->flags
);
1547 static bool img_request_child_test(struct rbd_img_request
*img_request
)
1550 return test_bit(IMG_REQ_CHILD
, &img_request
->flags
) != 0;
1553 static void img_request_layered_set(struct rbd_img_request
*img_request
)
1555 set_bit(IMG_REQ_LAYERED
, &img_request
->flags
);
1559 static void img_request_layered_clear(struct rbd_img_request
*img_request
)
1561 clear_bit(IMG_REQ_LAYERED
, &img_request
->flags
);
1565 static bool img_request_layered_test(struct rbd_img_request
*img_request
)
1568 return test_bit(IMG_REQ_LAYERED
, &img_request
->flags
) != 0;
1572 rbd_img_obj_request_read_callback(struct rbd_obj_request
*obj_request
)
1574 u64 xferred
= obj_request
->xferred
;
1575 u64 length
= obj_request
->length
;
1577 dout("%s: obj %p img %p result %d %llu/%llu\n", __func__
,
1578 obj_request
, obj_request
->img_request
, obj_request
->result
,
1581 * ENOENT means a hole in the image. We zero-fill the entire
1582 * length of the request. A short read also implies zero-fill
1583 * to the end of the request. An error requires the whole
1584 * length of the request to be reported finished with an error
1585 * to the block layer. In each case we update the xferred
1586 * count to indicate the whole request was satisfied.
1588 rbd_assert(obj_request
->type
!= OBJ_REQUEST_NODATA
);
1589 if (obj_request
->result
== -ENOENT
) {
1590 if (obj_request
->type
== OBJ_REQUEST_BIO
)
1591 zero_bio_chain(obj_request
->bio_list
, 0);
1593 zero_pages(obj_request
->pages
, 0, length
);
1594 obj_request
->result
= 0;
1595 } else if (xferred
< length
&& !obj_request
->result
) {
1596 if (obj_request
->type
== OBJ_REQUEST_BIO
)
1597 zero_bio_chain(obj_request
->bio_list
, xferred
);
1599 zero_pages(obj_request
->pages
, xferred
, length
);
1601 obj_request
->xferred
= length
;
1602 obj_request_done_set(obj_request
);
1605 static void rbd_obj_request_complete(struct rbd_obj_request
*obj_request
)
1607 dout("%s: obj %p cb %p\n", __func__
, obj_request
,
1608 obj_request
->callback
);
1609 if (obj_request
->callback
)
1610 obj_request
->callback(obj_request
);
1612 complete_all(&obj_request
->completion
);
1615 static void rbd_osd_trivial_callback(struct rbd_obj_request
*obj_request
)
1617 dout("%s: obj %p\n", __func__
, obj_request
);
1618 obj_request_done_set(obj_request
);
1621 static void rbd_osd_read_callback(struct rbd_obj_request
*obj_request
)
1623 struct rbd_img_request
*img_request
= NULL
;
1624 struct rbd_device
*rbd_dev
= NULL
;
1625 bool layered
= false;
1627 if (obj_request_img_data_test(obj_request
)) {
1628 img_request
= obj_request
->img_request
;
1629 layered
= img_request
&& img_request_layered_test(img_request
);
1630 rbd_dev
= img_request
->rbd_dev
;
1633 dout("%s: obj %p img %p result %d %llu/%llu\n", __func__
,
1634 obj_request
, img_request
, obj_request
->result
,
1635 obj_request
->xferred
, obj_request
->length
);
1636 if (layered
&& obj_request
->result
== -ENOENT
&&
1637 obj_request
->img_offset
< rbd_dev
->parent_overlap
)
1638 rbd_img_parent_read(obj_request
);
1639 else if (img_request
)
1640 rbd_img_obj_request_read_callback(obj_request
);
1642 obj_request_done_set(obj_request
);
1645 static void rbd_osd_write_callback(struct rbd_obj_request
*obj_request
)
1647 dout("%s: obj %p result %d %llu\n", __func__
, obj_request
,
1648 obj_request
->result
, obj_request
->length
);
1650 * There is no such thing as a successful short write. Set
1651 * it to our originally-requested length.
1653 obj_request
->xferred
= obj_request
->length
;
1654 obj_request_done_set(obj_request
);
1658 * For a simple stat call there's nothing to do. We'll do more if
1659 * this is part of a write sequence for a layered image.
1661 static void rbd_osd_stat_callback(struct rbd_obj_request
*obj_request
)
1663 dout("%s: obj %p\n", __func__
, obj_request
);
1664 obj_request_done_set(obj_request
);
1667 static void rbd_osd_req_callback(struct ceph_osd_request
*osd_req
,
1668 struct ceph_msg
*msg
)
1670 struct rbd_obj_request
*obj_request
= osd_req
->r_priv
;
1673 dout("%s: osd_req %p msg %p\n", __func__
, osd_req
, msg
);
1674 rbd_assert(osd_req
== obj_request
->osd_req
);
1675 if (obj_request_img_data_test(obj_request
)) {
1676 rbd_assert(obj_request
->img_request
);
1677 rbd_assert(obj_request
->which
!= BAD_WHICH
);
1679 rbd_assert(obj_request
->which
== BAD_WHICH
);
1682 if (osd_req
->r_result
< 0)
1683 obj_request
->result
= osd_req
->r_result
;
1685 BUG_ON(osd_req
->r_num_ops
> 2);
1688 * We support a 64-bit length, but ultimately it has to be
1689 * passed to blk_end_request(), which takes an unsigned int.
1691 obj_request
->xferred
= osd_req
->r_reply_op_len
[0];
1692 rbd_assert(obj_request
->xferred
< (u64
)UINT_MAX
);
1693 opcode
= osd_req
->r_ops
[0].op
;
1695 case CEPH_OSD_OP_READ
:
1696 rbd_osd_read_callback(obj_request
);
1698 case CEPH_OSD_OP_WRITE
:
1699 rbd_osd_write_callback(obj_request
);
1701 case CEPH_OSD_OP_STAT
:
1702 rbd_osd_stat_callback(obj_request
);
1704 case CEPH_OSD_OP_CALL
:
1705 case CEPH_OSD_OP_NOTIFY_ACK
:
1706 case CEPH_OSD_OP_WATCH
:
1707 rbd_osd_trivial_callback(obj_request
);
1710 rbd_warn(NULL
, "%s: unsupported op %hu\n",
1711 obj_request
->object_name
, (unsigned short) opcode
);
1715 if (obj_request_done_test(obj_request
))
1716 rbd_obj_request_complete(obj_request
);
1719 static void rbd_osd_req_format_read(struct rbd_obj_request
*obj_request
)
1721 struct rbd_img_request
*img_request
= obj_request
->img_request
;
1722 struct ceph_osd_request
*osd_req
= obj_request
->osd_req
;
1725 rbd_assert(osd_req
!= NULL
);
1727 snap_id
= img_request
? img_request
->snap_id
: CEPH_NOSNAP
;
1728 ceph_osdc_build_request(osd_req
, obj_request
->offset
,
1729 NULL
, snap_id
, NULL
);
1732 static void rbd_osd_req_format_write(struct rbd_obj_request
*obj_request
)
1734 struct rbd_img_request
*img_request
= obj_request
->img_request
;
1735 struct ceph_osd_request
*osd_req
= obj_request
->osd_req
;
1736 struct ceph_snap_context
*snapc
;
1737 struct timespec mtime
= CURRENT_TIME
;
1739 rbd_assert(osd_req
!= NULL
);
1741 snapc
= img_request
? img_request
->snapc
: NULL
;
1742 ceph_osdc_build_request(osd_req
, obj_request
->offset
,
1743 snapc
, CEPH_NOSNAP
, &mtime
);
1746 static struct ceph_osd_request
*rbd_osd_req_create(
1747 struct rbd_device
*rbd_dev
,
1749 struct rbd_obj_request
*obj_request
)
1751 struct ceph_snap_context
*snapc
= NULL
;
1752 struct ceph_osd_client
*osdc
;
1753 struct ceph_osd_request
*osd_req
;
1755 if (obj_request_img_data_test(obj_request
)) {
1756 struct rbd_img_request
*img_request
= obj_request
->img_request
;
1758 rbd_assert(write_request
==
1759 img_request_write_test(img_request
));
1761 snapc
= img_request
->snapc
;
1764 /* Allocate and initialize the request, for the single op */
1766 osdc
= &rbd_dev
->rbd_client
->client
->osdc
;
1767 osd_req
= ceph_osdc_alloc_request(osdc
, snapc
, 1, false, GFP_ATOMIC
);
1769 return NULL
; /* ENOMEM */
1772 osd_req
->r_flags
= CEPH_OSD_FLAG_WRITE
| CEPH_OSD_FLAG_ONDISK
;
1774 osd_req
->r_flags
= CEPH_OSD_FLAG_READ
;
1776 osd_req
->r_callback
= rbd_osd_req_callback
;
1777 osd_req
->r_priv
= obj_request
;
1779 osd_req
->r_oid_len
= strlen(obj_request
->object_name
);
1780 rbd_assert(osd_req
->r_oid_len
< sizeof (osd_req
->r_oid
));
1781 memcpy(osd_req
->r_oid
, obj_request
->object_name
, osd_req
->r_oid_len
);
1783 osd_req
->r_file_layout
= rbd_dev
->layout
; /* struct */
1789 * Create a copyup osd request based on the information in the
1790 * object request supplied. A copyup request has two osd ops,
1791 * a copyup method call, and a "normal" write request.
1793 static struct ceph_osd_request
*
1794 rbd_osd_req_create_copyup(struct rbd_obj_request
*obj_request
)
1796 struct rbd_img_request
*img_request
;
1797 struct ceph_snap_context
*snapc
;
1798 struct rbd_device
*rbd_dev
;
1799 struct ceph_osd_client
*osdc
;
1800 struct ceph_osd_request
*osd_req
;
1802 rbd_assert(obj_request_img_data_test(obj_request
));
1803 img_request
= obj_request
->img_request
;
1804 rbd_assert(img_request
);
1805 rbd_assert(img_request_write_test(img_request
));
1807 /* Allocate and initialize the request, for the two ops */
1809 snapc
= img_request
->snapc
;
1810 rbd_dev
= img_request
->rbd_dev
;
1811 osdc
= &rbd_dev
->rbd_client
->client
->osdc
;
1812 osd_req
= ceph_osdc_alloc_request(osdc
, snapc
, 2, false, GFP_ATOMIC
);
1814 return NULL
; /* ENOMEM */
1816 osd_req
->r_flags
= CEPH_OSD_FLAG_WRITE
| CEPH_OSD_FLAG_ONDISK
;
1817 osd_req
->r_callback
= rbd_osd_req_callback
;
1818 osd_req
->r_priv
= obj_request
;
1820 osd_req
->r_oid_len
= strlen(obj_request
->object_name
);
1821 rbd_assert(osd_req
->r_oid_len
< sizeof (osd_req
->r_oid
));
1822 memcpy(osd_req
->r_oid
, obj_request
->object_name
, osd_req
->r_oid_len
);
1824 osd_req
->r_file_layout
= rbd_dev
->layout
; /* struct */
1830 static void rbd_osd_req_destroy(struct ceph_osd_request
*osd_req
)
1832 ceph_osdc_put_request(osd_req
);
1835 /* object_name is assumed to be a non-null pointer and NUL-terminated */
1837 static struct rbd_obj_request
*rbd_obj_request_create(const char *object_name
,
1838 u64 offset
, u64 length
,
1839 enum obj_request_type type
)
1841 struct rbd_obj_request
*obj_request
;
1845 rbd_assert(obj_request_type_valid(type
));
1847 size
= strlen(object_name
) + 1;
1848 name
= kmalloc(size
, GFP_KERNEL
);
1852 obj_request
= kmem_cache_zalloc(rbd_obj_request_cache
, GFP_KERNEL
);
1858 obj_request
->object_name
= memcpy(name
, object_name
, size
);
1859 obj_request
->offset
= offset
;
1860 obj_request
->length
= length
;
1861 obj_request
->flags
= 0;
1862 obj_request
->which
= BAD_WHICH
;
1863 obj_request
->type
= type
;
1864 INIT_LIST_HEAD(&obj_request
->links
);
1865 init_completion(&obj_request
->completion
);
1866 kref_init(&obj_request
->kref
);
1868 dout("%s: \"%s\" %llu/%llu %d -> obj %p\n", __func__
, object_name
,
1869 offset
, length
, (int)type
, obj_request
);
1874 static void rbd_obj_request_destroy(struct kref
*kref
)
1876 struct rbd_obj_request
*obj_request
;
1878 obj_request
= container_of(kref
, struct rbd_obj_request
, kref
);
1880 dout("%s: obj %p\n", __func__
, obj_request
);
1882 rbd_assert(obj_request
->img_request
== NULL
);
1883 rbd_assert(obj_request
->which
== BAD_WHICH
);
1885 if (obj_request
->osd_req
)
1886 rbd_osd_req_destroy(obj_request
->osd_req
);
1888 rbd_assert(obj_request_type_valid(obj_request
->type
));
1889 switch (obj_request
->type
) {
1890 case OBJ_REQUEST_NODATA
:
1891 break; /* Nothing to do */
1892 case OBJ_REQUEST_BIO
:
1893 if (obj_request
->bio_list
)
1894 bio_chain_put(obj_request
->bio_list
);
1896 case OBJ_REQUEST_PAGES
:
1897 if (obj_request
->pages
)
1898 ceph_release_page_vector(obj_request
->pages
,
1899 obj_request
->page_count
);
1903 kfree(obj_request
->object_name
);
1904 obj_request
->object_name
= NULL
;
1905 kmem_cache_free(rbd_obj_request_cache
, obj_request
);
1908 /* It's OK to call this for a device with no parent */
1910 static void rbd_spec_put(struct rbd_spec
*spec
);
1911 static void rbd_dev_unparent(struct rbd_device
*rbd_dev
)
1913 rbd_dev_remove_parent(rbd_dev
);
1914 rbd_spec_put(rbd_dev
->parent_spec
);
1915 rbd_dev
->parent_spec
= NULL
;
1916 rbd_dev
->parent_overlap
= 0;
1920 * Parent image reference counting is used to determine when an
1921 * image's parent fields can be safely torn down--after there are no
1922 * more in-flight requests to the parent image. When the last
1923 * reference is dropped, cleaning them up is safe.
1925 static void rbd_dev_parent_put(struct rbd_device
*rbd_dev
)
1929 if (!rbd_dev
->parent_spec
)
1932 counter
= atomic_dec_return_safe(&rbd_dev
->parent_ref
);
1936 /* Last reference; clean up parent data structures */
1939 rbd_dev_unparent(rbd_dev
);
1941 rbd_warn(rbd_dev
, "parent reference underflow\n");
1945 * If an image has a non-zero parent overlap, get a reference to its
1948 * We must get the reference before checking for the overlap to
1949 * coordinate properly with zeroing the parent overlap in
1950 * rbd_dev_v2_parent_info() when an image gets flattened. We
1951 * drop it again if there is no overlap.
1953 * Returns true if the rbd device has a parent with a non-zero
1954 * overlap and a reference for it was successfully taken, or
1957 static bool rbd_dev_parent_get(struct rbd_device
*rbd_dev
)
1961 if (!rbd_dev
->parent_spec
)
1964 counter
= atomic_inc_return_safe(&rbd_dev
->parent_ref
);
1965 if (counter
> 0 && rbd_dev
->parent_overlap
)
1968 /* Image was flattened, but parent is not yet torn down */
1971 rbd_warn(rbd_dev
, "parent reference overflow\n");
1977 * Caller is responsible for filling in the list of object requests
1978 * that comprises the image request, and the Linux request pointer
1979 * (if there is one).
1981 static struct rbd_img_request
*rbd_img_request_create(
1982 struct rbd_device
*rbd_dev
,
1983 u64 offset
, u64 length
,
1986 struct rbd_img_request
*img_request
;
1988 img_request
= kmem_cache_alloc(rbd_img_request_cache
, GFP_ATOMIC
);
1992 if (write_request
) {
1993 down_read(&rbd_dev
->header_rwsem
);
1994 ceph_get_snap_context(rbd_dev
->header
.snapc
);
1995 up_read(&rbd_dev
->header_rwsem
);
1998 img_request
->rq
= NULL
;
1999 img_request
->rbd_dev
= rbd_dev
;
2000 img_request
->offset
= offset
;
2001 img_request
->length
= length
;
2002 img_request
->flags
= 0;
2003 if (write_request
) {
2004 img_request_write_set(img_request
);
2005 img_request
->snapc
= rbd_dev
->header
.snapc
;
2007 img_request
->snap_id
= rbd_dev
->spec
->snap_id
;
2009 if (rbd_dev_parent_get(rbd_dev
))
2010 img_request_layered_set(img_request
);
2011 spin_lock_init(&img_request
->completion_lock
);
2012 img_request
->next_completion
= 0;
2013 img_request
->callback
= NULL
;
2014 img_request
->result
= 0;
2015 img_request
->obj_request_count
= 0;
2016 INIT_LIST_HEAD(&img_request
->obj_requests
);
2017 kref_init(&img_request
->kref
);
2019 dout("%s: rbd_dev %p %s %llu/%llu -> img %p\n", __func__
, rbd_dev
,
2020 write_request
? "write" : "read", offset
, length
,
2026 static void rbd_img_request_destroy(struct kref
*kref
)
2028 struct rbd_img_request
*img_request
;
2029 struct rbd_obj_request
*obj_request
;
2030 struct rbd_obj_request
*next_obj_request
;
2032 img_request
= container_of(kref
, struct rbd_img_request
, kref
);
2034 dout("%s: img %p\n", __func__
, img_request
);
2036 for_each_obj_request_safe(img_request
, obj_request
, next_obj_request
)
2037 rbd_img_obj_request_del(img_request
, obj_request
);
2038 rbd_assert(img_request
->obj_request_count
== 0);
2040 if (img_request_layered_test(img_request
)) {
2041 img_request_layered_clear(img_request
);
2042 rbd_dev_parent_put(img_request
->rbd_dev
);
2045 if (img_request_write_test(img_request
))
2046 ceph_put_snap_context(img_request
->snapc
);
2048 kmem_cache_free(rbd_img_request_cache
, img_request
);
2051 static struct rbd_img_request
*rbd_parent_request_create(
2052 struct rbd_obj_request
*obj_request
,
2053 u64 img_offset
, u64 length
)
2055 struct rbd_img_request
*parent_request
;
2056 struct rbd_device
*rbd_dev
;
2058 rbd_assert(obj_request
->img_request
);
2059 rbd_dev
= obj_request
->img_request
->rbd_dev
;
2061 parent_request
= rbd_img_request_create(rbd_dev
->parent
,
2062 img_offset
, length
, false);
2063 if (!parent_request
)
2066 img_request_child_set(parent_request
);
2067 rbd_obj_request_get(obj_request
);
2068 parent_request
->obj_request
= obj_request
;
2070 return parent_request
;
2073 static void rbd_parent_request_destroy(struct kref
*kref
)
2075 struct rbd_img_request
*parent_request
;
2076 struct rbd_obj_request
*orig_request
;
2078 parent_request
= container_of(kref
, struct rbd_img_request
, kref
);
2079 orig_request
= parent_request
->obj_request
;
2081 parent_request
->obj_request
= NULL
;
2082 rbd_obj_request_put(orig_request
);
2083 img_request_child_clear(parent_request
);
2085 rbd_img_request_destroy(kref
);
2088 static bool rbd_img_obj_end_request(struct rbd_obj_request
*obj_request
)
2090 struct rbd_img_request
*img_request
;
2091 unsigned int xferred
;
2095 rbd_assert(obj_request_img_data_test(obj_request
));
2096 img_request
= obj_request
->img_request
;
2098 rbd_assert(obj_request
->xferred
<= (u64
)UINT_MAX
);
2099 xferred
= (unsigned int)obj_request
->xferred
;
2100 result
= obj_request
->result
;
2102 struct rbd_device
*rbd_dev
= img_request
->rbd_dev
;
2104 rbd_warn(rbd_dev
, "%s %llx at %llx (%llx)\n",
2105 img_request_write_test(img_request
) ? "write" : "read",
2106 obj_request
->length
, obj_request
->img_offset
,
2107 obj_request
->offset
);
2108 rbd_warn(rbd_dev
, " result %d xferred %x\n",
2110 if (!img_request
->result
)
2111 img_request
->result
= result
;
2114 /* Image object requests don't own their page array */
2116 if (obj_request
->type
== OBJ_REQUEST_PAGES
) {
2117 obj_request
->pages
= NULL
;
2118 obj_request
->page_count
= 0;
2121 if (img_request_child_test(img_request
)) {
2122 rbd_assert(img_request
->obj_request
!= NULL
);
2123 more
= obj_request
->which
< img_request
->obj_request_count
- 1;
2125 rbd_assert(img_request
->rq
!= NULL
);
2126 more
= blk_end_request(img_request
->rq
, result
, xferred
);
2132 static void rbd_img_obj_callback(struct rbd_obj_request
*obj_request
)
2134 struct rbd_img_request
*img_request
;
2135 u32 which
= obj_request
->which
;
2138 rbd_assert(obj_request_img_data_test(obj_request
));
2139 img_request
= obj_request
->img_request
;
2141 dout("%s: img %p obj %p\n", __func__
, img_request
, obj_request
);
2142 rbd_assert(img_request
!= NULL
);
2143 rbd_assert(img_request
->obj_request_count
> 0);
2144 rbd_assert(which
!= BAD_WHICH
);
2145 rbd_assert(which
< img_request
->obj_request_count
);
2146 rbd_assert(which
>= img_request
->next_completion
);
2148 spin_lock_irq(&img_request
->completion_lock
);
2149 if (which
!= img_request
->next_completion
)
2152 for_each_obj_request_from(img_request
, obj_request
) {
2154 rbd_assert(which
< img_request
->obj_request_count
);
2156 if (!obj_request_done_test(obj_request
))
2158 more
= rbd_img_obj_end_request(obj_request
);
2162 rbd_assert(more
^ (which
== img_request
->obj_request_count
));
2163 img_request
->next_completion
= which
;
2165 spin_unlock_irq(&img_request
->completion_lock
);
2166 rbd_img_request_put(img_request
);
2169 rbd_img_request_complete(img_request
);
2173 * Split up an image request into one or more object requests, each
2174 * to a different object. The "type" parameter indicates whether
2175 * "data_desc" is the pointer to the head of a list of bio
2176 * structures, or the base of a page array. In either case this
2177 * function assumes data_desc describes memory sufficient to hold
2178 * all data described by the image request.
2180 static int rbd_img_request_fill(struct rbd_img_request
*img_request
,
2181 enum obj_request_type type
,
2184 struct rbd_device
*rbd_dev
= img_request
->rbd_dev
;
2185 struct rbd_obj_request
*obj_request
= NULL
;
2186 struct rbd_obj_request
*next_obj_request
;
2187 bool write_request
= img_request_write_test(img_request
);
2188 struct bio
*bio_list
= NULL
;
2189 unsigned int bio_offset
= 0;
2190 struct page
**pages
= NULL
;
2195 dout("%s: img %p type %d data_desc %p\n", __func__
, img_request
,
2196 (int)type
, data_desc
);
2198 opcode
= write_request
? CEPH_OSD_OP_WRITE
: CEPH_OSD_OP_READ
;
2199 img_offset
= img_request
->offset
;
2200 resid
= img_request
->length
;
2201 rbd_assert(resid
> 0);
2203 if (type
== OBJ_REQUEST_BIO
) {
2204 bio_list
= data_desc
;
2205 rbd_assert(img_offset
== bio_list
->bi_sector
<< SECTOR_SHIFT
);
2207 rbd_assert(type
== OBJ_REQUEST_PAGES
);
2212 struct ceph_osd_request
*osd_req
;
2213 const char *object_name
;
2217 object_name
= rbd_segment_name(rbd_dev
, img_offset
);
2220 offset
= rbd_segment_offset(rbd_dev
, img_offset
);
2221 length
= rbd_segment_length(rbd_dev
, img_offset
, resid
);
2222 obj_request
= rbd_obj_request_create(object_name
,
2223 offset
, length
, type
);
2224 /* object request has its own copy of the object name */
2225 rbd_segment_name_free(object_name
);
2229 * set obj_request->img_request before creating the
2230 * osd_request so that it gets the right snapc
2232 rbd_img_obj_request_add(img_request
, obj_request
);
2234 if (type
== OBJ_REQUEST_BIO
) {
2235 unsigned int clone_size
;
2237 rbd_assert(length
<= (u64
)UINT_MAX
);
2238 clone_size
= (unsigned int)length
;
2239 obj_request
->bio_list
=
2240 bio_chain_clone_range(&bio_list
,
2244 if (!obj_request
->bio_list
)
2247 unsigned int page_count
;
2249 obj_request
->pages
= pages
;
2250 page_count
= (u32
)calc_pages_for(offset
, length
);
2251 obj_request
->page_count
= page_count
;
2252 if ((offset
+ length
) & ~PAGE_MASK
)
2253 page_count
--; /* more on last page */
2254 pages
+= page_count
;
2257 osd_req
= rbd_osd_req_create(rbd_dev
, write_request
,
2261 obj_request
->osd_req
= osd_req
;
2262 obj_request
->callback
= rbd_img_obj_callback
;
2263 rbd_img_request_get(img_request
);
2265 osd_req_op_extent_init(osd_req
, 0, opcode
, offset
, length
,
2267 if (type
== OBJ_REQUEST_BIO
)
2268 osd_req_op_extent_osd_data_bio(osd_req
, 0,
2269 obj_request
->bio_list
, length
);
2271 osd_req_op_extent_osd_data_pages(osd_req
, 0,
2272 obj_request
->pages
, length
,
2273 offset
& ~PAGE_MASK
, false, false);
2276 rbd_osd_req_format_write(obj_request
);
2278 rbd_osd_req_format_read(obj_request
);
2280 obj_request
->img_offset
= img_offset
;
2282 img_offset
+= length
;
2289 rbd_obj_request_put(obj_request
);
2291 for_each_obj_request_safe(img_request
, obj_request
, next_obj_request
)
2292 rbd_img_obj_request_del(img_request
, obj_request
);
2298 rbd_img_obj_copyup_callback(struct rbd_obj_request
*obj_request
)
2300 struct rbd_img_request
*img_request
;
2301 struct rbd_device
*rbd_dev
;
2302 struct page
**pages
;
2305 rbd_assert(obj_request
->type
== OBJ_REQUEST_BIO
);
2306 rbd_assert(obj_request_img_data_test(obj_request
));
2307 img_request
= obj_request
->img_request
;
2308 rbd_assert(img_request
);
2310 rbd_dev
= img_request
->rbd_dev
;
2311 rbd_assert(rbd_dev
);
2313 pages
= obj_request
->copyup_pages
;
2314 rbd_assert(pages
!= NULL
);
2315 obj_request
->copyup_pages
= NULL
;
2316 page_count
= obj_request
->copyup_page_count
;
2317 rbd_assert(page_count
);
2318 obj_request
->copyup_page_count
= 0;
2319 ceph_release_page_vector(pages
, page_count
);
2322 * We want the transfer count to reflect the size of the
2323 * original write request. There is no such thing as a
2324 * successful short write, so if the request was successful
2325 * we can just set it to the originally-requested length.
2327 if (!obj_request
->result
)
2328 obj_request
->xferred
= obj_request
->length
;
2330 /* Finish up with the normal image object callback */
2332 rbd_img_obj_callback(obj_request
);
2336 rbd_img_obj_parent_read_full_callback(struct rbd_img_request
*img_request
)
2338 struct rbd_obj_request
*orig_request
;
2339 struct ceph_osd_request
*osd_req
;
2340 struct ceph_osd_client
*osdc
;
2341 struct rbd_device
*rbd_dev
;
2342 struct page
**pages
;
2349 rbd_assert(img_request_child_test(img_request
));
2351 /* First get what we need from the image request */
2353 pages
= img_request
->copyup_pages
;
2354 rbd_assert(pages
!= NULL
);
2355 img_request
->copyup_pages
= NULL
;
2356 page_count
= img_request
->copyup_page_count
;
2357 rbd_assert(page_count
);
2358 img_request
->copyup_page_count
= 0;
2360 orig_request
= img_request
->obj_request
;
2361 rbd_assert(orig_request
!= NULL
);
2362 rbd_assert(obj_request_type_valid(orig_request
->type
));
2363 img_result
= img_request
->result
;
2364 parent_length
= img_request
->length
;
2365 rbd_assert(parent_length
== img_request
->xferred
);
2366 rbd_img_request_put(img_request
);
2368 rbd_assert(orig_request
->img_request
);
2369 rbd_dev
= orig_request
->img_request
->rbd_dev
;
2370 rbd_assert(rbd_dev
);
2373 * If the overlap has become 0 (most likely because the
2374 * image has been flattened) we need to free the pages
2375 * and re-submit the original write request.
2377 if (!rbd_dev
->parent_overlap
) {
2378 struct ceph_osd_client
*osdc
;
2380 ceph_release_page_vector(pages
, page_count
);
2381 osdc
= &rbd_dev
->rbd_client
->client
->osdc
;
2382 img_result
= rbd_obj_request_submit(osdc
, orig_request
);
2391 * The original osd request is of no use to use any more.
2392 * We need a new one that can hold the two ops in a copyup
2393 * request. Allocate the new copyup osd request for the
2394 * original request, and release the old one.
2396 img_result
= -ENOMEM
;
2397 osd_req
= rbd_osd_req_create_copyup(orig_request
);
2400 rbd_osd_req_destroy(orig_request
->osd_req
);
2401 orig_request
->osd_req
= osd_req
;
2402 orig_request
->copyup_pages
= pages
;
2403 orig_request
->copyup_page_count
= page_count
;
2405 /* Initialize the copyup op */
2407 osd_req_op_cls_init(osd_req
, 0, CEPH_OSD_OP_CALL
, "rbd", "copyup");
2408 osd_req_op_cls_request_data_pages(osd_req
, 0, pages
, parent_length
, 0,
2411 /* Then the original write request op */
2413 offset
= orig_request
->offset
;
2414 length
= orig_request
->length
;
2415 osd_req_op_extent_init(osd_req
, 1, CEPH_OSD_OP_WRITE
,
2416 offset
, length
, 0, 0);
2417 if (orig_request
->type
== OBJ_REQUEST_BIO
)
2418 osd_req_op_extent_osd_data_bio(osd_req
, 1,
2419 orig_request
->bio_list
, length
);
2421 osd_req_op_extent_osd_data_pages(osd_req
, 1,
2422 orig_request
->pages
, length
,
2423 offset
& ~PAGE_MASK
, false, false);
2425 rbd_osd_req_format_write(orig_request
);
2427 /* All set, send it off. */
2429 orig_request
->callback
= rbd_img_obj_copyup_callback
;
2430 osdc
= &rbd_dev
->rbd_client
->client
->osdc
;
2431 img_result
= rbd_obj_request_submit(osdc
, orig_request
);
2435 /* Record the error code and complete the request */
2437 orig_request
->result
= img_result
;
2438 orig_request
->xferred
= 0;
2439 obj_request_done_set(orig_request
);
2440 rbd_obj_request_complete(orig_request
);
2444 * Read from the parent image the range of data that covers the
2445 * entire target of the given object request. This is used for
2446 * satisfying a layered image write request when the target of an
2447 * object request from the image request does not exist.
2449 * A page array big enough to hold the returned data is allocated
2450 * and supplied to rbd_img_request_fill() as the "data descriptor."
2451 * When the read completes, this page array will be transferred to
2452 * the original object request for the copyup operation.
2454 * If an error occurs, record it as the result of the original
2455 * object request and mark it done so it gets completed.
2457 static int rbd_img_obj_parent_read_full(struct rbd_obj_request
*obj_request
)
2459 struct rbd_img_request
*img_request
= NULL
;
2460 struct rbd_img_request
*parent_request
= NULL
;
2461 struct rbd_device
*rbd_dev
;
2464 struct page
**pages
= NULL
;
2468 rbd_assert(obj_request_img_data_test(obj_request
));
2469 rbd_assert(obj_request_type_valid(obj_request
->type
));
2471 img_request
= obj_request
->img_request
;
2472 rbd_assert(img_request
!= NULL
);
2473 rbd_dev
= img_request
->rbd_dev
;
2474 rbd_assert(rbd_dev
->parent
!= NULL
);
2477 * Determine the byte range covered by the object in the
2478 * child image to which the original request was to be sent.
2480 img_offset
= obj_request
->img_offset
- obj_request
->offset
;
2481 length
= (u64
)1 << rbd_dev
->header
.obj_order
;
2484 * There is no defined parent data beyond the parent
2485 * overlap, so limit what we read at that boundary if
2488 if (img_offset
+ length
> rbd_dev
->parent_overlap
) {
2489 rbd_assert(img_offset
< rbd_dev
->parent_overlap
);
2490 length
= rbd_dev
->parent_overlap
- img_offset
;
2494 * Allocate a page array big enough to receive the data read
2497 page_count
= (u32
)calc_pages_for(0, length
);
2498 pages
= ceph_alloc_page_vector(page_count
, GFP_KERNEL
);
2499 if (IS_ERR(pages
)) {
2500 result
= PTR_ERR(pages
);
2506 parent_request
= rbd_parent_request_create(obj_request
,
2507 img_offset
, length
);
2508 if (!parent_request
)
2511 result
= rbd_img_request_fill(parent_request
, OBJ_REQUEST_PAGES
, pages
);
2514 parent_request
->copyup_pages
= pages
;
2515 parent_request
->copyup_page_count
= page_count
;
2517 parent_request
->callback
= rbd_img_obj_parent_read_full_callback
;
2518 result
= rbd_img_request_submit(parent_request
);
2522 parent_request
->copyup_pages
= NULL
;
2523 parent_request
->copyup_page_count
= 0;
2524 parent_request
->obj_request
= NULL
;
2525 rbd_obj_request_put(obj_request
);
2528 ceph_release_page_vector(pages
, page_count
);
2530 rbd_img_request_put(parent_request
);
2531 obj_request
->result
= result
;
2532 obj_request
->xferred
= 0;
2533 obj_request_done_set(obj_request
);
2538 static void rbd_img_obj_exists_callback(struct rbd_obj_request
*obj_request
)
2540 struct rbd_obj_request
*orig_request
;
2541 struct rbd_device
*rbd_dev
;
2544 rbd_assert(!obj_request_img_data_test(obj_request
));
2547 * All we need from the object request is the original
2548 * request and the result of the STAT op. Grab those, then
2549 * we're done with the request.
2551 orig_request
= obj_request
->obj_request
;
2552 obj_request
->obj_request
= NULL
;
2553 rbd_obj_request_put(orig_request
);
2554 rbd_assert(orig_request
);
2555 rbd_assert(orig_request
->img_request
);
2557 result
= obj_request
->result
;
2558 obj_request
->result
= 0;
2560 dout("%s: obj %p for obj %p result %d %llu/%llu\n", __func__
,
2561 obj_request
, orig_request
, result
,
2562 obj_request
->xferred
, obj_request
->length
);
2563 rbd_obj_request_put(obj_request
);
2566 * If the overlap has become 0 (most likely because the
2567 * image has been flattened) we need to free the pages
2568 * and re-submit the original write request.
2570 rbd_dev
= orig_request
->img_request
->rbd_dev
;
2571 if (!rbd_dev
->parent_overlap
) {
2572 struct ceph_osd_client
*osdc
;
2574 osdc
= &rbd_dev
->rbd_client
->client
->osdc
;
2575 result
= rbd_obj_request_submit(osdc
, orig_request
);
2581 * Our only purpose here is to determine whether the object
2582 * exists, and we don't want to treat the non-existence as
2583 * an error. If something else comes back, transfer the
2584 * error to the original request and complete it now.
2587 obj_request_existence_set(orig_request
, true);
2588 } else if (result
== -ENOENT
) {
2589 obj_request_existence_set(orig_request
, false);
2590 } else if (result
) {
2591 orig_request
->result
= result
;
2596 * Resubmit the original request now that we have recorded
2597 * whether the target object exists.
2599 orig_request
->result
= rbd_img_obj_request_submit(orig_request
);
2601 if (orig_request
->result
)
2602 rbd_obj_request_complete(orig_request
);
2605 static int rbd_img_obj_exists_submit(struct rbd_obj_request
*obj_request
)
2607 struct rbd_obj_request
*stat_request
;
2608 struct rbd_device
*rbd_dev
;
2609 struct ceph_osd_client
*osdc
;
2610 struct page
**pages
= NULL
;
2616 * The response data for a STAT call consists of:
2623 size
= sizeof (__le64
) + sizeof (__le32
) + sizeof (__le32
);
2624 page_count
= (u32
)calc_pages_for(0, size
);
2625 pages
= ceph_alloc_page_vector(page_count
, GFP_KERNEL
);
2627 return PTR_ERR(pages
);
2630 stat_request
= rbd_obj_request_create(obj_request
->object_name
, 0, 0,
2635 rbd_obj_request_get(obj_request
);
2636 stat_request
->obj_request
= obj_request
;
2637 stat_request
->pages
= pages
;
2638 stat_request
->page_count
= page_count
;
2640 rbd_assert(obj_request
->img_request
);
2641 rbd_dev
= obj_request
->img_request
->rbd_dev
;
2642 stat_request
->osd_req
= rbd_osd_req_create(rbd_dev
, false,
2644 if (!stat_request
->osd_req
)
2646 stat_request
->callback
= rbd_img_obj_exists_callback
;
2648 osd_req_op_init(stat_request
->osd_req
, 0, CEPH_OSD_OP_STAT
);
2649 osd_req_op_raw_data_in_pages(stat_request
->osd_req
, 0, pages
, size
, 0,
2651 rbd_osd_req_format_read(stat_request
);
2653 osdc
= &rbd_dev
->rbd_client
->client
->osdc
;
2654 ret
= rbd_obj_request_submit(osdc
, stat_request
);
2657 rbd_obj_request_put(obj_request
);
2662 static int rbd_img_obj_request_submit(struct rbd_obj_request
*obj_request
)
2664 struct rbd_img_request
*img_request
;
2665 struct rbd_device
*rbd_dev
;
2668 rbd_assert(obj_request_img_data_test(obj_request
));
2670 img_request
= obj_request
->img_request
;
2671 rbd_assert(img_request
);
2672 rbd_dev
= img_request
->rbd_dev
;
2675 * Only writes to layered images need special handling.
2676 * Reads and non-layered writes are simple object requests.
2677 * Layered writes that start beyond the end of the overlap
2678 * with the parent have no parent data, so they too are
2679 * simple object requests. Finally, if the target object is
2680 * known to already exist, its parent data has already been
2681 * copied, so a write to the object can also be handled as a
2682 * simple object request.
2684 if (!img_request_write_test(img_request
) ||
2685 !img_request_layered_test(img_request
) ||
2686 !obj_request_overlaps_parent(obj_request
) ||
2687 ((known
= obj_request_known_test(obj_request
)) &&
2688 obj_request_exists_test(obj_request
))) {
2690 struct rbd_device
*rbd_dev
;
2691 struct ceph_osd_client
*osdc
;
2693 rbd_dev
= obj_request
->img_request
->rbd_dev
;
2694 osdc
= &rbd_dev
->rbd_client
->client
->osdc
;
2696 return rbd_obj_request_submit(osdc
, obj_request
);
2700 * It's a layered write. The target object might exist but
2701 * we may not know that yet. If we know it doesn't exist,
2702 * start by reading the data for the full target object from
2703 * the parent so we can use it for a copyup to the target.
2706 return rbd_img_obj_parent_read_full(obj_request
);
2708 /* We don't know whether the target exists. Go find out. */
2710 return rbd_img_obj_exists_submit(obj_request
);
2713 static int rbd_img_request_submit(struct rbd_img_request
*img_request
)
2715 struct rbd_obj_request
*obj_request
;
2716 struct rbd_obj_request
*next_obj_request
;
2718 dout("%s: img %p\n", __func__
, img_request
);
2719 for_each_obj_request_safe(img_request
, obj_request
, next_obj_request
) {
2722 ret
= rbd_img_obj_request_submit(obj_request
);
2730 static void rbd_img_parent_read_callback(struct rbd_img_request
*img_request
)
2732 struct rbd_obj_request
*obj_request
;
2733 struct rbd_device
*rbd_dev
;
2738 rbd_assert(img_request_child_test(img_request
));
2740 /* First get what we need from the image request and release it */
2742 obj_request
= img_request
->obj_request
;
2743 img_xferred
= img_request
->xferred
;
2744 img_result
= img_request
->result
;
2745 rbd_img_request_put(img_request
);
2748 * If the overlap has become 0 (most likely because the
2749 * image has been flattened) we need to re-submit the
2752 rbd_assert(obj_request
);
2753 rbd_assert(obj_request
->img_request
);
2754 rbd_dev
= obj_request
->img_request
->rbd_dev
;
2755 if (!rbd_dev
->parent_overlap
) {
2756 struct ceph_osd_client
*osdc
;
2758 osdc
= &rbd_dev
->rbd_client
->client
->osdc
;
2759 img_result
= rbd_obj_request_submit(osdc
, obj_request
);
2764 obj_request
->result
= img_result
;
2765 if (obj_request
->result
)
2769 * We need to zero anything beyond the parent overlap
2770 * boundary. Since rbd_img_obj_request_read_callback()
2771 * will zero anything beyond the end of a short read, an
2772 * easy way to do this is to pretend the data from the
2773 * parent came up short--ending at the overlap boundary.
2775 rbd_assert(obj_request
->img_offset
< U64_MAX
- obj_request
->length
);
2776 obj_end
= obj_request
->img_offset
+ obj_request
->length
;
2777 if (obj_end
> rbd_dev
->parent_overlap
) {
2780 if (obj_request
->img_offset
< rbd_dev
->parent_overlap
)
2781 xferred
= rbd_dev
->parent_overlap
-
2782 obj_request
->img_offset
;
2784 obj_request
->xferred
= min(img_xferred
, xferred
);
2786 obj_request
->xferred
= img_xferred
;
2789 rbd_img_obj_request_read_callback(obj_request
);
2790 rbd_obj_request_complete(obj_request
);
2793 static void rbd_img_parent_read(struct rbd_obj_request
*obj_request
)
2795 struct rbd_img_request
*img_request
;
2798 rbd_assert(obj_request_img_data_test(obj_request
));
2799 rbd_assert(obj_request
->img_request
!= NULL
);
2800 rbd_assert(obj_request
->result
== (s32
) -ENOENT
);
2801 rbd_assert(obj_request_type_valid(obj_request
->type
));
2803 /* rbd_read_finish(obj_request, obj_request->length); */
2804 img_request
= rbd_parent_request_create(obj_request
,
2805 obj_request
->img_offset
,
2806 obj_request
->length
);
2811 if (obj_request
->type
== OBJ_REQUEST_BIO
)
2812 result
= rbd_img_request_fill(img_request
, OBJ_REQUEST_BIO
,
2813 obj_request
->bio_list
);
2815 result
= rbd_img_request_fill(img_request
, OBJ_REQUEST_PAGES
,
2816 obj_request
->pages
);
2820 img_request
->callback
= rbd_img_parent_read_callback
;
2821 result
= rbd_img_request_submit(img_request
);
2828 rbd_img_request_put(img_request
);
2829 obj_request
->result
= result
;
2830 obj_request
->xferred
= 0;
2831 obj_request_done_set(obj_request
);
2834 static int rbd_obj_notify_ack_sync(struct rbd_device
*rbd_dev
, u64 notify_id
)
2836 struct rbd_obj_request
*obj_request
;
2837 struct ceph_osd_client
*osdc
= &rbd_dev
->rbd_client
->client
->osdc
;
2840 obj_request
= rbd_obj_request_create(rbd_dev
->header_name
, 0, 0,
2841 OBJ_REQUEST_NODATA
);
2846 obj_request
->osd_req
= rbd_osd_req_create(rbd_dev
, false, obj_request
);
2847 if (!obj_request
->osd_req
)
2850 osd_req_op_watch_init(obj_request
->osd_req
, 0, CEPH_OSD_OP_NOTIFY_ACK
,
2852 rbd_osd_req_format_read(obj_request
);
2854 ret
= rbd_obj_request_submit(osdc
, obj_request
);
2857 ret
= rbd_obj_request_wait(obj_request
);
2859 rbd_obj_request_put(obj_request
);
2864 static void rbd_watch_cb(u64 ver
, u64 notify_id
, u8 opcode
, void *data
)
2866 struct rbd_device
*rbd_dev
= (struct rbd_device
*)data
;
2872 dout("%s: \"%s\" notify_id %llu opcode %u\n", __func__
,
2873 rbd_dev
->header_name
, (unsigned long long)notify_id
,
2874 (unsigned int)opcode
);
2875 ret
= rbd_dev_refresh(rbd_dev
);
2877 rbd_warn(rbd_dev
, "header refresh error (%d)\n", ret
);
2879 rbd_obj_notify_ack_sync(rbd_dev
, notify_id
);
2883 * Request sync osd watch/unwatch. The value of "start" determines
2884 * whether a watch request is being initiated or torn down.
2886 static int rbd_dev_header_watch_sync(struct rbd_device
*rbd_dev
, bool start
)
2888 struct ceph_osd_client
*osdc
= &rbd_dev
->rbd_client
->client
->osdc
;
2889 struct rbd_obj_request
*obj_request
;
2892 rbd_assert(start
^ !!rbd_dev
->watch_event
);
2893 rbd_assert(start
^ !!rbd_dev
->watch_request
);
2896 ret
= ceph_osdc_create_event(osdc
, rbd_watch_cb
, rbd_dev
,
2897 &rbd_dev
->watch_event
);
2900 rbd_assert(rbd_dev
->watch_event
!= NULL
);
2904 obj_request
= rbd_obj_request_create(rbd_dev
->header_name
, 0, 0,
2905 OBJ_REQUEST_NODATA
);
2909 obj_request
->osd_req
= rbd_osd_req_create(rbd_dev
, true, obj_request
);
2910 if (!obj_request
->osd_req
)
2914 ceph_osdc_set_request_linger(osdc
, obj_request
->osd_req
);
2916 ceph_osdc_unregister_linger_request(osdc
,
2917 rbd_dev
->watch_request
->osd_req
);
2919 osd_req_op_watch_init(obj_request
->osd_req
, 0, CEPH_OSD_OP_WATCH
,
2920 rbd_dev
->watch_event
->cookie
, 0, start
? 1 : 0);
2921 rbd_osd_req_format_write(obj_request
);
2923 ret
= rbd_obj_request_submit(osdc
, obj_request
);
2926 ret
= rbd_obj_request_wait(obj_request
);
2929 ret
= obj_request
->result
;
2934 * A watch request is set to linger, so the underlying osd
2935 * request won't go away until we unregister it. We retain
2936 * a pointer to the object request during that time (in
2937 * rbd_dev->watch_request), so we'll keep a reference to
2938 * it. We'll drop that reference (below) after we've
2942 rbd_dev
->watch_request
= obj_request
;
2947 /* We have successfully torn down the watch request */
2949 rbd_obj_request_put(rbd_dev
->watch_request
);
2950 rbd_dev
->watch_request
= NULL
;
2952 /* Cancel the event if we're tearing down, or on error */
2953 ceph_osdc_cancel_event(rbd_dev
->watch_event
);
2954 rbd_dev
->watch_event
= NULL
;
2956 rbd_obj_request_put(obj_request
);
2962 * Synchronous osd object method call. Returns the number of bytes
2963 * returned in the outbound buffer, or a negative error code.
2965 static int rbd_obj_method_sync(struct rbd_device
*rbd_dev
,
2966 const char *object_name
,
2967 const char *class_name
,
2968 const char *method_name
,
2969 const void *outbound
,
2970 size_t outbound_size
,
2972 size_t inbound_size
)
2974 struct ceph_osd_client
*osdc
= &rbd_dev
->rbd_client
->client
->osdc
;
2975 struct rbd_obj_request
*obj_request
;
2976 struct page
**pages
;
2981 * Method calls are ultimately read operations. The result
2982 * should placed into the inbound buffer provided. They
2983 * also supply outbound data--parameters for the object
2984 * method. Currently if this is present it will be a
2987 page_count
= (u32
)calc_pages_for(0, inbound_size
);
2988 pages
= ceph_alloc_page_vector(page_count
, GFP_KERNEL
);
2990 return PTR_ERR(pages
);
2993 obj_request
= rbd_obj_request_create(object_name
, 0, inbound_size
,
2998 obj_request
->pages
= pages
;
2999 obj_request
->page_count
= page_count
;
3001 obj_request
->osd_req
= rbd_osd_req_create(rbd_dev
, false, obj_request
);
3002 if (!obj_request
->osd_req
)
3005 osd_req_op_cls_init(obj_request
->osd_req
, 0, CEPH_OSD_OP_CALL
,
3006 class_name
, method_name
);
3007 if (outbound_size
) {
3008 struct ceph_pagelist
*pagelist
;
3010 pagelist
= kmalloc(sizeof (*pagelist
), GFP_NOFS
);
3014 ceph_pagelist_init(pagelist
);
3015 ceph_pagelist_append(pagelist
, outbound
, outbound_size
);
3016 osd_req_op_cls_request_data_pagelist(obj_request
->osd_req
, 0,
3019 osd_req_op_cls_response_data_pages(obj_request
->osd_req
, 0,
3020 obj_request
->pages
, inbound_size
,
3022 rbd_osd_req_format_read(obj_request
);
3024 ret
= rbd_obj_request_submit(osdc
, obj_request
);
3027 ret
= rbd_obj_request_wait(obj_request
);
3031 ret
= obj_request
->result
;
3035 rbd_assert(obj_request
->xferred
< (u64
)INT_MAX
);
3036 ret
= (int)obj_request
->xferred
;
3037 ceph_copy_from_page_vector(pages
, inbound
, 0, obj_request
->xferred
);
3040 rbd_obj_request_put(obj_request
);
3042 ceph_release_page_vector(pages
, page_count
);
3047 static void rbd_request_fn(struct request_queue
*q
)
3048 __releases(q
->queue_lock
) __acquires(q
->queue_lock
)
3050 struct rbd_device
*rbd_dev
= q
->queuedata
;
3051 bool read_only
= rbd_dev
->mapping
.read_only
;
3055 while ((rq
= blk_fetch_request(q
))) {
3056 bool write_request
= rq_data_dir(rq
) == WRITE
;
3057 struct rbd_img_request
*img_request
;
3061 /* Ignore any non-FS requests that filter through. */
3063 if (rq
->cmd_type
!= REQ_TYPE_FS
) {
3064 dout("%s: non-fs request type %d\n", __func__
,
3065 (int) rq
->cmd_type
);
3066 __blk_end_request_all(rq
, 0);
3070 /* Ignore/skip any zero-length requests */
3072 offset
= (u64
) blk_rq_pos(rq
) << SECTOR_SHIFT
;
3073 length
= (u64
) blk_rq_bytes(rq
);
3076 dout("%s: zero-length request\n", __func__
);
3077 __blk_end_request_all(rq
, 0);
3081 spin_unlock_irq(q
->queue_lock
);
3083 /* Disallow writes to a read-only device */
3085 if (write_request
) {
3089 rbd_assert(rbd_dev
->spec
->snap_id
== CEPH_NOSNAP
);
3093 * Quit early if the mapped snapshot no longer
3094 * exists. It's still possible the snapshot will
3095 * have disappeared by the time our request arrives
3096 * at the osd, but there's no sense in sending it if
3099 if (!test_bit(RBD_DEV_FLAG_EXISTS
, &rbd_dev
->flags
)) {
3100 dout("request for non-existent snapshot");
3101 rbd_assert(rbd_dev
->spec
->snap_id
!= CEPH_NOSNAP
);
3107 if (offset
&& length
> U64_MAX
- offset
+ 1) {
3108 rbd_warn(rbd_dev
, "bad request range (%llu~%llu)\n",
3110 goto end_request
; /* Shouldn't happen */
3114 if (offset
+ length
> rbd_dev
->mapping
.size
) {
3115 rbd_warn(rbd_dev
, "beyond EOD (%llu~%llu > %llu)\n",
3116 offset
, length
, rbd_dev
->mapping
.size
);
3121 img_request
= rbd_img_request_create(rbd_dev
, offset
, length
,
3126 img_request
->rq
= rq
;
3128 result
= rbd_img_request_fill(img_request
, OBJ_REQUEST_BIO
,
3131 result
= rbd_img_request_submit(img_request
);
3133 rbd_img_request_put(img_request
);
3135 spin_lock_irq(q
->queue_lock
);
3137 rbd_warn(rbd_dev
, "%s %llx at %llx result %d\n",
3138 write_request
? "write" : "read",
3139 length
, offset
, result
);
3141 __blk_end_request_all(rq
, result
);
3147 * a queue callback. Makes sure that we don't create a bio that spans across
3148 * multiple osd objects. One exception would be with a single page bios,
3149 * which we handle later at bio_chain_clone_range()
3151 static int rbd_merge_bvec(struct request_queue
*q
, struct bvec_merge_data
*bmd
,
3152 struct bio_vec
*bvec
)
3154 struct rbd_device
*rbd_dev
= q
->queuedata
;
3155 sector_t sector_offset
;
3156 sector_t sectors_per_obj
;
3157 sector_t obj_sector_offset
;
3161 * Find how far into its rbd object the partition-relative
3162 * bio start sector is to offset relative to the enclosing
3165 sector_offset
= get_start_sect(bmd
->bi_bdev
) + bmd
->bi_sector
;
3166 sectors_per_obj
= 1 << (rbd_dev
->header
.obj_order
- SECTOR_SHIFT
);
3167 obj_sector_offset
= sector_offset
& (sectors_per_obj
- 1);
3170 * Compute the number of bytes from that offset to the end
3171 * of the object. Account for what's already used by the bio.
3173 ret
= (int) (sectors_per_obj
- obj_sector_offset
) << SECTOR_SHIFT
;
3174 if (ret
> bmd
->bi_size
)
3175 ret
-= bmd
->bi_size
;
3180 * Don't send back more than was asked for. And if the bio
3181 * was empty, let the whole thing through because: "Note
3182 * that a block device *must* allow a single page to be
3183 * added to an empty bio."
3185 rbd_assert(bvec
->bv_len
<= PAGE_SIZE
);
3186 if (ret
> (int) bvec
->bv_len
|| !bmd
->bi_size
)
3187 ret
= (int) bvec
->bv_len
;
3192 static void rbd_free_disk(struct rbd_device
*rbd_dev
)
3194 struct gendisk
*disk
= rbd_dev
->disk
;
3199 rbd_dev
->disk
= NULL
;
3200 if (disk
->flags
& GENHD_FL_UP
) {
3203 blk_cleanup_queue(disk
->queue
);
3208 static int rbd_obj_read_sync(struct rbd_device
*rbd_dev
,
3209 const char *object_name
,
3210 u64 offset
, u64 length
, void *buf
)
3213 struct ceph_osd_client
*osdc
= &rbd_dev
->rbd_client
->client
->osdc
;
3214 struct rbd_obj_request
*obj_request
;
3215 struct page
**pages
= NULL
;
3220 page_count
= (u32
) calc_pages_for(offset
, length
);
3221 pages
= ceph_alloc_page_vector(page_count
, GFP_KERNEL
);
3223 ret
= PTR_ERR(pages
);
3226 obj_request
= rbd_obj_request_create(object_name
, offset
, length
,
3231 obj_request
->pages
= pages
;
3232 obj_request
->page_count
= page_count
;
3234 obj_request
->osd_req
= rbd_osd_req_create(rbd_dev
, false, obj_request
);
3235 if (!obj_request
->osd_req
)
3238 osd_req_op_extent_init(obj_request
->osd_req
, 0, CEPH_OSD_OP_READ
,
3239 offset
, length
, 0, 0);
3240 osd_req_op_extent_osd_data_pages(obj_request
->osd_req
, 0,
3242 obj_request
->length
,
3243 obj_request
->offset
& ~PAGE_MASK
,
3245 rbd_osd_req_format_read(obj_request
);
3247 ret
= rbd_obj_request_submit(osdc
, obj_request
);
3250 ret
= rbd_obj_request_wait(obj_request
);
3254 ret
= obj_request
->result
;
3258 rbd_assert(obj_request
->xferred
<= (u64
) SIZE_MAX
);
3259 size
= (size_t) obj_request
->xferred
;
3260 ceph_copy_from_page_vector(pages
, buf
, 0, size
);
3261 rbd_assert(size
<= (size_t)INT_MAX
);
3265 rbd_obj_request_put(obj_request
);
3267 ceph_release_page_vector(pages
, page_count
);
3273 * Read the complete header for the given rbd device. On successful
3274 * return, the rbd_dev->header field will contain up-to-date
3275 * information about the image.
3277 static int rbd_dev_v1_header_info(struct rbd_device
*rbd_dev
)
3279 struct rbd_image_header_ondisk
*ondisk
= NULL
;
3286 * The complete header will include an array of its 64-bit
3287 * snapshot ids, followed by the names of those snapshots as
3288 * a contiguous block of NUL-terminated strings. Note that
3289 * the number of snapshots could change by the time we read
3290 * it in, in which case we re-read it.
3297 size
= sizeof (*ondisk
);
3298 size
+= snap_count
* sizeof (struct rbd_image_snap_ondisk
);
3300 ondisk
= kmalloc(size
, GFP_KERNEL
);
3304 ret
= rbd_obj_read_sync(rbd_dev
, rbd_dev
->header_name
,
3308 if ((size_t)ret
< size
) {
3310 rbd_warn(rbd_dev
, "short header read (want %zd got %d)",
3314 if (!rbd_dev_ondisk_valid(ondisk
)) {
3316 rbd_warn(rbd_dev
, "invalid header");
3320 names_size
= le64_to_cpu(ondisk
->snap_names_len
);
3321 want_count
= snap_count
;
3322 snap_count
= le32_to_cpu(ondisk
->snap_count
);
3323 } while (snap_count
!= want_count
);
3325 ret
= rbd_header_from_disk(rbd_dev
, ondisk
);
3333 * Clear the rbd device's EXISTS flag if the snapshot it's mapped to
3334 * has disappeared from the (just updated) snapshot context.
3336 static void rbd_exists_validate(struct rbd_device
*rbd_dev
)
3340 if (!test_bit(RBD_DEV_FLAG_EXISTS
, &rbd_dev
->flags
))
3343 snap_id
= rbd_dev
->spec
->snap_id
;
3344 if (snap_id
== CEPH_NOSNAP
)
3347 if (rbd_dev_snap_index(rbd_dev
, snap_id
) == BAD_SNAP_INDEX
)
3348 clear_bit(RBD_DEV_FLAG_EXISTS
, &rbd_dev
->flags
);
3351 static void rbd_dev_update_size(struct rbd_device
*rbd_dev
)
3357 * Don't hold the lock while doing disk operations,
3358 * or lock ordering will conflict with the bdev mutex via:
3359 * rbd_add() -> blkdev_get() -> rbd_open()
3361 spin_lock_irq(&rbd_dev
->lock
);
3362 removing
= test_bit(RBD_DEV_FLAG_REMOVING
, &rbd_dev
->flags
);
3363 spin_unlock_irq(&rbd_dev
->lock
);
3365 * If the device is being removed, rbd_dev->disk has
3366 * been destroyed, so don't try to update its size
3369 size
= (sector_t
)rbd_dev
->mapping
.size
/ SECTOR_SIZE
;
3370 dout("setting size to %llu sectors", (unsigned long long)size
);
3371 set_capacity(rbd_dev
->disk
, size
);
3372 revalidate_disk(rbd_dev
->disk
);
3376 static int rbd_dev_refresh(struct rbd_device
*rbd_dev
)
3381 rbd_assert(rbd_image_format_valid(rbd_dev
->image_format
));
3382 down_write(&rbd_dev
->header_rwsem
);
3383 mapping_size
= rbd_dev
->mapping
.size
;
3384 if (rbd_dev
->image_format
== 1)
3385 ret
= rbd_dev_v1_header_info(rbd_dev
);
3387 ret
= rbd_dev_v2_header_info(rbd_dev
);
3389 /* If it's a mapped snapshot, validate its EXISTS flag */
3391 rbd_exists_validate(rbd_dev
);
3392 up_write(&rbd_dev
->header_rwsem
);
3394 if (mapping_size
!= rbd_dev
->mapping
.size
) {
3395 rbd_dev_update_size(rbd_dev
);
3401 static int rbd_init_disk(struct rbd_device
*rbd_dev
)
3403 struct gendisk
*disk
;
3404 struct request_queue
*q
;
3407 /* create gendisk info */
3408 disk
= alloc_disk(RBD_MINORS_PER_MAJOR
);
3412 snprintf(disk
->disk_name
, sizeof(disk
->disk_name
), RBD_DRV_NAME
"%d",
3414 disk
->major
= rbd_dev
->major
;
3415 disk
->first_minor
= 0;
3416 disk
->fops
= &rbd_bd_ops
;
3417 disk
->private_data
= rbd_dev
;
3419 q
= blk_init_queue(rbd_request_fn
, &rbd_dev
->lock
);
3423 /* We use the default size, but let's be explicit about it. */
3424 blk_queue_physical_block_size(q
, SECTOR_SIZE
);
3426 /* set io sizes to object size */
3427 segment_size
= rbd_obj_bytes(&rbd_dev
->header
);
3428 blk_queue_max_hw_sectors(q
, segment_size
/ SECTOR_SIZE
);
3429 blk_queue_max_segment_size(q
, segment_size
);
3430 blk_queue_io_min(q
, segment_size
);
3431 blk_queue_io_opt(q
, segment_size
);
3433 blk_queue_merge_bvec(q
, rbd_merge_bvec
);
3436 q
->queuedata
= rbd_dev
;
3438 rbd_dev
->disk
= disk
;
3451 static struct rbd_device
*dev_to_rbd_dev(struct device
*dev
)
3453 return container_of(dev
, struct rbd_device
, dev
);
3456 static ssize_t
rbd_size_show(struct device
*dev
,
3457 struct device_attribute
*attr
, char *buf
)
3459 struct rbd_device
*rbd_dev
= dev_to_rbd_dev(dev
);
3461 return sprintf(buf
, "%llu\n",
3462 (unsigned long long)rbd_dev
->mapping
.size
);
3466 * Note this shows the features for whatever's mapped, which is not
3467 * necessarily the base image.
3469 static ssize_t
rbd_features_show(struct device
*dev
,
3470 struct device_attribute
*attr
, char *buf
)
3472 struct rbd_device
*rbd_dev
= dev_to_rbd_dev(dev
);
3474 return sprintf(buf
, "0x%016llx\n",
3475 (unsigned long long)rbd_dev
->mapping
.features
);
3478 static ssize_t
rbd_major_show(struct device
*dev
,
3479 struct device_attribute
*attr
, char *buf
)
3481 struct rbd_device
*rbd_dev
= dev_to_rbd_dev(dev
);
3484 return sprintf(buf
, "%d\n", rbd_dev
->major
);
3486 return sprintf(buf
, "(none)\n");
3490 static ssize_t
rbd_client_id_show(struct device
*dev
,
3491 struct device_attribute
*attr
, char *buf
)
3493 struct rbd_device
*rbd_dev
= dev_to_rbd_dev(dev
);
3495 return sprintf(buf
, "client%lld\n",
3496 ceph_client_id(rbd_dev
->rbd_client
->client
));
3499 static ssize_t
rbd_pool_show(struct device
*dev
,
3500 struct device_attribute
*attr
, char *buf
)
3502 struct rbd_device
*rbd_dev
= dev_to_rbd_dev(dev
);
3504 return sprintf(buf
, "%s\n", rbd_dev
->spec
->pool_name
);
3507 static ssize_t
rbd_pool_id_show(struct device
*dev
,
3508 struct device_attribute
*attr
, char *buf
)
3510 struct rbd_device
*rbd_dev
= dev_to_rbd_dev(dev
);
3512 return sprintf(buf
, "%llu\n",
3513 (unsigned long long) rbd_dev
->spec
->pool_id
);
3516 static ssize_t
rbd_name_show(struct device
*dev
,
3517 struct device_attribute
*attr
, char *buf
)
3519 struct rbd_device
*rbd_dev
= dev_to_rbd_dev(dev
);
3521 if (rbd_dev
->spec
->image_name
)
3522 return sprintf(buf
, "%s\n", rbd_dev
->spec
->image_name
);
3524 return sprintf(buf
, "(unknown)\n");
3527 static ssize_t
rbd_image_id_show(struct device
*dev
,
3528 struct device_attribute
*attr
, char *buf
)
3530 struct rbd_device
*rbd_dev
= dev_to_rbd_dev(dev
);
3532 return sprintf(buf
, "%s\n", rbd_dev
->spec
->image_id
);
3536 * Shows the name of the currently-mapped snapshot (or
3537 * RBD_SNAP_HEAD_NAME for the base image).
3539 static ssize_t
rbd_snap_show(struct device
*dev
,
3540 struct device_attribute
*attr
,
3543 struct rbd_device
*rbd_dev
= dev_to_rbd_dev(dev
);
3545 return sprintf(buf
, "%s\n", rbd_dev
->spec
->snap_name
);
3549 * For an rbd v2 image, shows the pool id, image id, and snapshot id
3550 * for the parent image. If there is no parent, simply shows
3551 * "(no parent image)".
3553 static ssize_t
rbd_parent_show(struct device
*dev
,
3554 struct device_attribute
*attr
,
3557 struct rbd_device
*rbd_dev
= dev_to_rbd_dev(dev
);
3558 struct rbd_spec
*spec
= rbd_dev
->parent_spec
;
3563 return sprintf(buf
, "(no parent image)\n");
3565 count
= sprintf(bufp
, "pool_id %llu\npool_name %s\n",
3566 (unsigned long long) spec
->pool_id
, spec
->pool_name
);
3571 count
= sprintf(bufp
, "image_id %s\nimage_name %s\n", spec
->image_id
,
3572 spec
->image_name
? spec
->image_name
: "(unknown)");
3577 count
= sprintf(bufp
, "snap_id %llu\nsnap_name %s\n",
3578 (unsigned long long) spec
->snap_id
, spec
->snap_name
);
3583 count
= sprintf(bufp
, "overlap %llu\n", rbd_dev
->parent_overlap
);
3588 return (ssize_t
) (bufp
- buf
);
3591 static ssize_t
rbd_image_refresh(struct device
*dev
,
3592 struct device_attribute
*attr
,
3596 struct rbd_device
*rbd_dev
= dev_to_rbd_dev(dev
);
3599 ret
= rbd_dev_refresh(rbd_dev
);
3601 rbd_warn(rbd_dev
, ": manual header refresh error (%d)\n", ret
);
3603 return ret
< 0 ? ret
: size
;
3606 static DEVICE_ATTR(size
, S_IRUGO
, rbd_size_show
, NULL
);
3607 static DEVICE_ATTR(features
, S_IRUGO
, rbd_features_show
, NULL
);
3608 static DEVICE_ATTR(major
, S_IRUGO
, rbd_major_show
, NULL
);
3609 static DEVICE_ATTR(client_id
, S_IRUGO
, rbd_client_id_show
, NULL
);
3610 static DEVICE_ATTR(pool
, S_IRUGO
, rbd_pool_show
, NULL
);
3611 static DEVICE_ATTR(pool_id
, S_IRUGO
, rbd_pool_id_show
, NULL
);
3612 static DEVICE_ATTR(name
, S_IRUGO
, rbd_name_show
, NULL
);
3613 static DEVICE_ATTR(image_id
, S_IRUGO
, rbd_image_id_show
, NULL
);
3614 static DEVICE_ATTR(refresh
, S_IWUSR
, NULL
, rbd_image_refresh
);
3615 static DEVICE_ATTR(current_snap
, S_IRUGO
, rbd_snap_show
, NULL
);
3616 static DEVICE_ATTR(parent
, S_IRUGO
, rbd_parent_show
, NULL
);
3618 static struct attribute
*rbd_attrs
[] = {
3619 &dev_attr_size
.attr
,
3620 &dev_attr_features
.attr
,
3621 &dev_attr_major
.attr
,
3622 &dev_attr_client_id
.attr
,
3623 &dev_attr_pool
.attr
,
3624 &dev_attr_pool_id
.attr
,
3625 &dev_attr_name
.attr
,
3626 &dev_attr_image_id
.attr
,
3627 &dev_attr_current_snap
.attr
,
3628 &dev_attr_parent
.attr
,
3629 &dev_attr_refresh
.attr
,
3633 static struct attribute_group rbd_attr_group
= {
3637 static const struct attribute_group
*rbd_attr_groups
[] = {
3642 static void rbd_sysfs_dev_release(struct device
*dev
)
3646 static struct device_type rbd_device_type
= {
3648 .groups
= rbd_attr_groups
,
3649 .release
= rbd_sysfs_dev_release
,
3652 static struct rbd_spec
*rbd_spec_get(struct rbd_spec
*spec
)
3654 kref_get(&spec
->kref
);
3659 static void rbd_spec_free(struct kref
*kref
);
3660 static void rbd_spec_put(struct rbd_spec
*spec
)
3663 kref_put(&spec
->kref
, rbd_spec_free
);
3666 static struct rbd_spec
*rbd_spec_alloc(void)
3668 struct rbd_spec
*spec
;
3670 spec
= kzalloc(sizeof (*spec
), GFP_KERNEL
);
3673 kref_init(&spec
->kref
);
3678 static void rbd_spec_free(struct kref
*kref
)
3680 struct rbd_spec
*spec
= container_of(kref
, struct rbd_spec
, kref
);
3682 kfree(spec
->pool_name
);
3683 kfree(spec
->image_id
);
3684 kfree(spec
->image_name
);
3685 kfree(spec
->snap_name
);
3689 static struct rbd_device
*rbd_dev_create(struct rbd_client
*rbdc
,
3690 struct rbd_spec
*spec
)
3692 struct rbd_device
*rbd_dev
;
3694 rbd_dev
= kzalloc(sizeof (*rbd_dev
), GFP_KERNEL
);
3698 spin_lock_init(&rbd_dev
->lock
);
3700 atomic_set(&rbd_dev
->parent_ref
, 0);
3701 INIT_LIST_HEAD(&rbd_dev
->node
);
3702 init_rwsem(&rbd_dev
->header_rwsem
);
3704 rbd_dev
->spec
= spec
;
3705 rbd_dev
->rbd_client
= rbdc
;
3707 /* Initialize the layout used for all rbd requests */
3709 rbd_dev
->layout
.fl_stripe_unit
= cpu_to_le32(1 << RBD_MAX_OBJ_ORDER
);
3710 rbd_dev
->layout
.fl_stripe_count
= cpu_to_le32(1);
3711 rbd_dev
->layout
.fl_object_size
= cpu_to_le32(1 << RBD_MAX_OBJ_ORDER
);
3712 rbd_dev
->layout
.fl_pg_pool
= cpu_to_le32((u32
) spec
->pool_id
);
3717 static void rbd_dev_destroy(struct rbd_device
*rbd_dev
)
3719 rbd_put_client(rbd_dev
->rbd_client
);
3720 rbd_spec_put(rbd_dev
->spec
);
3725 * Get the size and object order for an image snapshot, or if
3726 * snap_id is CEPH_NOSNAP, gets this information for the base
3729 static int _rbd_dev_v2_snap_size(struct rbd_device
*rbd_dev
, u64 snap_id
,
3730 u8
*order
, u64
*snap_size
)
3732 __le64 snapid
= cpu_to_le64(snap_id
);
3737 } __attribute__ ((packed
)) size_buf
= { 0 };
3739 ret
= rbd_obj_method_sync(rbd_dev
, rbd_dev
->header_name
,
3741 &snapid
, sizeof (snapid
),
3742 &size_buf
, sizeof (size_buf
));
3743 dout("%s: rbd_obj_method_sync returned %d\n", __func__
, ret
);
3746 if (ret
< sizeof (size_buf
))
3750 *order
= size_buf
.order
;
3751 dout(" order %u", (unsigned int)*order
);
3753 *snap_size
= le64_to_cpu(size_buf
.size
);
3755 dout(" snap_id 0x%016llx snap_size = %llu\n",
3756 (unsigned long long)snap_id
,
3757 (unsigned long long)*snap_size
);
3762 static int rbd_dev_v2_image_size(struct rbd_device
*rbd_dev
)
3764 return _rbd_dev_v2_snap_size(rbd_dev
, CEPH_NOSNAP
,
3765 &rbd_dev
->header
.obj_order
,
3766 &rbd_dev
->header
.image_size
);
3769 static int rbd_dev_v2_object_prefix(struct rbd_device
*rbd_dev
)
3775 reply_buf
= kzalloc(RBD_OBJ_PREFIX_LEN_MAX
, GFP_KERNEL
);
3779 ret
= rbd_obj_method_sync(rbd_dev
, rbd_dev
->header_name
,
3780 "rbd", "get_object_prefix", NULL
, 0,
3781 reply_buf
, RBD_OBJ_PREFIX_LEN_MAX
);
3782 dout("%s: rbd_obj_method_sync returned %d\n", __func__
, ret
);
3787 rbd_dev
->header
.object_prefix
= ceph_extract_encoded_string(&p
,
3788 p
+ ret
, NULL
, GFP_NOIO
);
3791 if (IS_ERR(rbd_dev
->header
.object_prefix
)) {
3792 ret
= PTR_ERR(rbd_dev
->header
.object_prefix
);
3793 rbd_dev
->header
.object_prefix
= NULL
;
3795 dout(" object_prefix = %s\n", rbd_dev
->header
.object_prefix
);
3803 static int _rbd_dev_v2_snap_features(struct rbd_device
*rbd_dev
, u64 snap_id
,
3806 __le64 snapid
= cpu_to_le64(snap_id
);
3810 } __attribute__ ((packed
)) features_buf
= { 0 };
3814 ret
= rbd_obj_method_sync(rbd_dev
, rbd_dev
->header_name
,
3815 "rbd", "get_features",
3816 &snapid
, sizeof (snapid
),
3817 &features_buf
, sizeof (features_buf
));
3818 dout("%s: rbd_obj_method_sync returned %d\n", __func__
, ret
);
3821 if (ret
< sizeof (features_buf
))
3824 incompat
= le64_to_cpu(features_buf
.incompat
);
3825 if (incompat
& ~RBD_FEATURES_SUPPORTED
)
3828 *snap_features
= le64_to_cpu(features_buf
.features
);
3830 dout(" snap_id 0x%016llx features = 0x%016llx incompat = 0x%016llx\n",
3831 (unsigned long long)snap_id
,
3832 (unsigned long long)*snap_features
,
3833 (unsigned long long)le64_to_cpu(features_buf
.incompat
));
3838 static int rbd_dev_v2_features(struct rbd_device
*rbd_dev
)
3840 return _rbd_dev_v2_snap_features(rbd_dev
, CEPH_NOSNAP
,
3841 &rbd_dev
->header
.features
);
3844 static int rbd_dev_v2_parent_info(struct rbd_device
*rbd_dev
)
3846 struct rbd_spec
*parent_spec
;
3848 void *reply_buf
= NULL
;
3858 parent_spec
= rbd_spec_alloc();
3862 size
= sizeof (__le64
) + /* pool_id */
3863 sizeof (__le32
) + RBD_IMAGE_ID_LEN_MAX
+ /* image_id */
3864 sizeof (__le64
) + /* snap_id */
3865 sizeof (__le64
); /* overlap */
3866 reply_buf
= kmalloc(size
, GFP_KERNEL
);
3872 snapid
= cpu_to_le64(CEPH_NOSNAP
);
3873 ret
= rbd_obj_method_sync(rbd_dev
, rbd_dev
->header_name
,
3874 "rbd", "get_parent",
3875 &snapid
, sizeof (snapid
),
3877 dout("%s: rbd_obj_method_sync returned %d\n", __func__
, ret
);
3882 end
= reply_buf
+ ret
;
3884 ceph_decode_64_safe(&p
, end
, pool_id
, out_err
);
3885 if (pool_id
== CEPH_NOPOOL
) {
3887 * Either the parent never existed, or we have
3888 * record of it but the image got flattened so it no
3889 * longer has a parent. When the parent of a
3890 * layered image disappears we immediately set the
3891 * overlap to 0. The effect of this is that all new
3892 * requests will be treated as if the image had no
3895 if (rbd_dev
->parent_overlap
) {
3896 rbd_dev
->parent_overlap
= 0;
3898 rbd_dev_parent_put(rbd_dev
);
3899 pr_info("%s: clone image has been flattened\n",
3900 rbd_dev
->disk
->disk_name
);
3903 goto out
; /* No parent? No problem. */
3906 /* The ceph file layout needs to fit pool id in 32 bits */
3909 if (pool_id
> (u64
)U32_MAX
) {
3910 rbd_warn(NULL
, "parent pool id too large (%llu > %u)\n",
3911 (unsigned long long)pool_id
, U32_MAX
);
3915 image_id
= ceph_extract_encoded_string(&p
, end
, NULL
, GFP_KERNEL
);
3916 if (IS_ERR(image_id
)) {
3917 ret
= PTR_ERR(image_id
);
3920 ceph_decode_64_safe(&p
, end
, snap_id
, out_err
);
3921 ceph_decode_64_safe(&p
, end
, overlap
, out_err
);
3924 * The parent won't change (except when the clone is
3925 * flattened, already handled that). So we only need to
3926 * record the parent spec we have not already done so.
3928 if (!rbd_dev
->parent_spec
) {
3929 parent_spec
->pool_id
= pool_id
;
3930 parent_spec
->image_id
= image_id
;
3931 parent_spec
->snap_id
= snap_id
;
3932 rbd_dev
->parent_spec
= parent_spec
;
3933 parent_spec
= NULL
; /* rbd_dev now owns this */
3937 * We always update the parent overlap. If it's zero we
3938 * treat it specially.
3940 rbd_dev
->parent_overlap
= overlap
;
3944 /* A null parent_spec indicates it's the initial probe */
3948 * The overlap has become zero, so the clone
3949 * must have been resized down to 0 at some
3950 * point. Treat this the same as a flatten.
3952 rbd_dev_parent_put(rbd_dev
);
3953 pr_info("%s: clone image now standalone\n",
3954 rbd_dev
->disk
->disk_name
);
3957 * For the initial probe, if we find the
3958 * overlap is zero we just pretend there was
3961 rbd_warn(rbd_dev
, "ignoring parent of "
3962 "clone with overlap 0\n");
3969 rbd_spec_put(parent_spec
);
3974 static int rbd_dev_v2_striping_info(struct rbd_device
*rbd_dev
)
3978 __le64 stripe_count
;
3979 } __attribute__ ((packed
)) striping_info_buf
= { 0 };
3980 size_t size
= sizeof (striping_info_buf
);
3987 ret
= rbd_obj_method_sync(rbd_dev
, rbd_dev
->header_name
,
3988 "rbd", "get_stripe_unit_count", NULL
, 0,
3989 (char *)&striping_info_buf
, size
);
3990 dout("%s: rbd_obj_method_sync returned %d\n", __func__
, ret
);
3997 * We don't actually support the "fancy striping" feature
3998 * (STRIPINGV2) yet, but if the striping sizes are the
3999 * defaults the behavior is the same as before. So find
4000 * out, and only fail if the image has non-default values.
4003 obj_size
= (u64
)1 << rbd_dev
->header
.obj_order
;
4004 p
= &striping_info_buf
;
4005 stripe_unit
= ceph_decode_64(&p
);
4006 if (stripe_unit
!= obj_size
) {
4007 rbd_warn(rbd_dev
, "unsupported stripe unit "
4008 "(got %llu want %llu)",
4009 stripe_unit
, obj_size
);
4012 stripe_count
= ceph_decode_64(&p
);
4013 if (stripe_count
!= 1) {
4014 rbd_warn(rbd_dev
, "unsupported stripe count "
4015 "(got %llu want 1)", stripe_count
);
4018 rbd_dev
->header
.stripe_unit
= stripe_unit
;
4019 rbd_dev
->header
.stripe_count
= stripe_count
;
4024 static char *rbd_dev_image_name(struct rbd_device
*rbd_dev
)
4026 size_t image_id_size
;
4031 void *reply_buf
= NULL
;
4033 char *image_name
= NULL
;
4036 rbd_assert(!rbd_dev
->spec
->image_name
);
4038 len
= strlen(rbd_dev
->spec
->image_id
);
4039 image_id_size
= sizeof (__le32
) + len
;
4040 image_id
= kmalloc(image_id_size
, GFP_KERNEL
);
4045 end
= image_id
+ image_id_size
;
4046 ceph_encode_string(&p
, end
, rbd_dev
->spec
->image_id
, (u32
)len
);
4048 size
= sizeof (__le32
) + RBD_IMAGE_NAME_LEN_MAX
;
4049 reply_buf
= kmalloc(size
, GFP_KERNEL
);
4053 ret
= rbd_obj_method_sync(rbd_dev
, RBD_DIRECTORY
,
4054 "rbd", "dir_get_name",
4055 image_id
, image_id_size
,
4060 end
= reply_buf
+ ret
;
4062 image_name
= ceph_extract_encoded_string(&p
, end
, &len
, GFP_KERNEL
);
4063 if (IS_ERR(image_name
))
4066 dout("%s: name is %s len is %zd\n", __func__
, image_name
, len
);
4074 static u64
rbd_v1_snap_id_by_name(struct rbd_device
*rbd_dev
, const char *name
)
4076 struct ceph_snap_context
*snapc
= rbd_dev
->header
.snapc
;
4077 const char *snap_name
;
4080 /* Skip over names until we find the one we are looking for */
4082 snap_name
= rbd_dev
->header
.snap_names
;
4083 while (which
< snapc
->num_snaps
) {
4084 if (!strcmp(name
, snap_name
))
4085 return snapc
->snaps
[which
];
4086 snap_name
+= strlen(snap_name
) + 1;
4092 static u64
rbd_v2_snap_id_by_name(struct rbd_device
*rbd_dev
, const char *name
)
4094 struct ceph_snap_context
*snapc
= rbd_dev
->header
.snapc
;
4099 for (which
= 0; !found
&& which
< snapc
->num_snaps
; which
++) {
4100 const char *snap_name
;
4102 snap_id
= snapc
->snaps
[which
];
4103 snap_name
= rbd_dev_v2_snap_name(rbd_dev
, snap_id
);
4104 if (IS_ERR(snap_name
)) {
4105 /* ignore no-longer existing snapshots */
4106 if (PTR_ERR(snap_name
) == -ENOENT
)
4111 found
= !strcmp(name
, snap_name
);
4114 return found
? snap_id
: CEPH_NOSNAP
;
4118 * Assumes name is never RBD_SNAP_HEAD_NAME; returns CEPH_NOSNAP if
4119 * no snapshot by that name is found, or if an error occurs.
4121 static u64
rbd_snap_id_by_name(struct rbd_device
*rbd_dev
, const char *name
)
4123 if (rbd_dev
->image_format
== 1)
4124 return rbd_v1_snap_id_by_name(rbd_dev
, name
);
4126 return rbd_v2_snap_id_by_name(rbd_dev
, name
);
4130 * When an rbd image has a parent image, it is identified by the
4131 * pool, image, and snapshot ids (not names). This function fills
4132 * in the names for those ids. (It's OK if we can't figure out the
4133 * name for an image id, but the pool and snapshot ids should always
4134 * exist and have names.) All names in an rbd spec are dynamically
4137 * When an image being mapped (not a parent) is probed, we have the
4138 * pool name and pool id, image name and image id, and the snapshot
4139 * name. The only thing we're missing is the snapshot id.
4141 static int rbd_dev_spec_update(struct rbd_device
*rbd_dev
)
4143 struct ceph_osd_client
*osdc
= &rbd_dev
->rbd_client
->client
->osdc
;
4144 struct rbd_spec
*spec
= rbd_dev
->spec
;
4145 const char *pool_name
;
4146 const char *image_name
;
4147 const char *snap_name
;
4151 * An image being mapped will have the pool name (etc.), but
4152 * we need to look up the snapshot id.
4154 if (spec
->pool_name
) {
4155 if (strcmp(spec
->snap_name
, RBD_SNAP_HEAD_NAME
)) {
4158 snap_id
= rbd_snap_id_by_name(rbd_dev
, spec
->snap_name
);
4159 if (snap_id
== CEPH_NOSNAP
)
4161 spec
->snap_id
= snap_id
;
4163 spec
->snap_id
= CEPH_NOSNAP
;
4169 /* Get the pool name; we have to make our own copy of this */
4171 pool_name
= ceph_pg_pool_name_by_id(osdc
->osdmap
, spec
->pool_id
);
4173 rbd_warn(rbd_dev
, "no pool with id %llu", spec
->pool_id
);
4176 pool_name
= kstrdup(pool_name
, GFP_KERNEL
);
4180 /* Fetch the image name; tolerate failure here */
4182 image_name
= rbd_dev_image_name(rbd_dev
);
4184 rbd_warn(rbd_dev
, "unable to get image name");
4186 /* Look up the snapshot name, and make a copy */
4188 snap_name
= rbd_snap_name(rbd_dev
, spec
->snap_id
);
4189 if (IS_ERR(snap_name
)) {
4190 ret
= PTR_ERR(snap_name
);
4194 spec
->pool_name
= pool_name
;
4195 spec
->image_name
= image_name
;
4196 spec
->snap_name
= snap_name
;
4206 static int rbd_dev_v2_snap_context(struct rbd_device
*rbd_dev
)
4215 struct ceph_snap_context
*snapc
;
4219 * We'll need room for the seq value (maximum snapshot id),
4220 * snapshot count, and array of that many snapshot ids.
4221 * For now we have a fixed upper limit on the number we're
4222 * prepared to receive.
4224 size
= sizeof (__le64
) + sizeof (__le32
) +
4225 RBD_MAX_SNAP_COUNT
* sizeof (__le64
);
4226 reply_buf
= kzalloc(size
, GFP_KERNEL
);
4230 ret
= rbd_obj_method_sync(rbd_dev
, rbd_dev
->header_name
,
4231 "rbd", "get_snapcontext", NULL
, 0,
4233 dout("%s: rbd_obj_method_sync returned %d\n", __func__
, ret
);
4238 end
= reply_buf
+ ret
;
4240 ceph_decode_64_safe(&p
, end
, seq
, out
);
4241 ceph_decode_32_safe(&p
, end
, snap_count
, out
);
4244 * Make sure the reported number of snapshot ids wouldn't go
4245 * beyond the end of our buffer. But before checking that,
4246 * make sure the computed size of the snapshot context we
4247 * allocate is representable in a size_t.
4249 if (snap_count
> (SIZE_MAX
- sizeof (struct ceph_snap_context
))
4254 if (!ceph_has_room(&p
, end
, snap_count
* sizeof (__le64
)))
4258 snapc
= ceph_create_snap_context(snap_count
, GFP_KERNEL
);
4264 for (i
= 0; i
< snap_count
; i
++)
4265 snapc
->snaps
[i
] = ceph_decode_64(&p
);
4267 ceph_put_snap_context(rbd_dev
->header
.snapc
);
4268 rbd_dev
->header
.snapc
= snapc
;
4270 dout(" snap context seq = %llu, snap_count = %u\n",
4271 (unsigned long long)seq
, (unsigned int)snap_count
);
4278 static const char *rbd_dev_v2_snap_name(struct rbd_device
*rbd_dev
,
4289 size
= sizeof (__le32
) + RBD_MAX_SNAP_NAME_LEN
;
4290 reply_buf
= kmalloc(size
, GFP_KERNEL
);
4292 return ERR_PTR(-ENOMEM
);
4294 snapid
= cpu_to_le64(snap_id
);
4295 ret
= rbd_obj_method_sync(rbd_dev
, rbd_dev
->header_name
,
4296 "rbd", "get_snapshot_name",
4297 &snapid
, sizeof (snapid
),
4299 dout("%s: rbd_obj_method_sync returned %d\n", __func__
, ret
);
4301 snap_name
= ERR_PTR(ret
);
4306 end
= reply_buf
+ ret
;
4307 snap_name
= ceph_extract_encoded_string(&p
, end
, NULL
, GFP_KERNEL
);
4308 if (IS_ERR(snap_name
))
4311 dout(" snap_id 0x%016llx snap_name = %s\n",
4312 (unsigned long long)snap_id
, snap_name
);
4319 static int rbd_dev_v2_header_info(struct rbd_device
*rbd_dev
)
4321 bool first_time
= rbd_dev
->header
.object_prefix
== NULL
;
4324 ret
= rbd_dev_v2_image_size(rbd_dev
);
4329 ret
= rbd_dev_v2_header_onetime(rbd_dev
);
4335 * If the image supports layering, get the parent info. We
4336 * need to probe the first time regardless. Thereafter we
4337 * only need to if there's a parent, to see if it has
4338 * disappeared due to the mapped image getting flattened.
4340 if (rbd_dev
->header
.features
& RBD_FEATURE_LAYERING
&&
4341 (first_time
|| rbd_dev
->parent_spec
)) {
4344 ret
= rbd_dev_v2_parent_info(rbd_dev
);
4349 * Print a warning if this is the initial probe and
4350 * the image has a parent. Don't print it if the
4351 * image now being probed is itself a parent. We
4352 * can tell at this point because we won't know its
4353 * pool name yet (just its pool id).
4355 warn
= rbd_dev
->parent_spec
&& rbd_dev
->spec
->pool_name
;
4356 if (first_time
&& warn
)
4357 rbd_warn(rbd_dev
, "WARNING: kernel layering "
4358 "is EXPERIMENTAL!");
4361 if (rbd_dev
->spec
->snap_id
== CEPH_NOSNAP
)
4362 if (rbd_dev
->mapping
.size
!= rbd_dev
->header
.image_size
)
4363 rbd_dev
->mapping
.size
= rbd_dev
->header
.image_size
;
4365 ret
= rbd_dev_v2_snap_context(rbd_dev
);
4366 dout("rbd_dev_v2_snap_context returned %d\n", ret
);
4371 static int rbd_bus_add_dev(struct rbd_device
*rbd_dev
)
4376 dev
= &rbd_dev
->dev
;
4377 dev
->bus
= &rbd_bus_type
;
4378 dev
->type
= &rbd_device_type
;
4379 dev
->parent
= &rbd_root_dev
;
4380 dev
->release
= rbd_dev_device_release
;
4381 dev_set_name(dev
, "%d", rbd_dev
->dev_id
);
4382 ret
= device_register(dev
);
4387 static void rbd_bus_del_dev(struct rbd_device
*rbd_dev
)
4389 device_unregister(&rbd_dev
->dev
);
4392 static atomic64_t rbd_dev_id_max
= ATOMIC64_INIT(0);
4395 * Get a unique rbd identifier for the given new rbd_dev, and add
4396 * the rbd_dev to the global list. The minimum rbd id is 1.
4398 static void rbd_dev_id_get(struct rbd_device
*rbd_dev
)
4400 rbd_dev
->dev_id
= atomic64_inc_return(&rbd_dev_id_max
);
4402 spin_lock(&rbd_dev_list_lock
);
4403 list_add_tail(&rbd_dev
->node
, &rbd_dev_list
);
4404 spin_unlock(&rbd_dev_list_lock
);
4405 dout("rbd_dev %p given dev id %llu\n", rbd_dev
,
4406 (unsigned long long) rbd_dev
->dev_id
);
4410 * Remove an rbd_dev from the global list, and record that its
4411 * identifier is no longer in use.
4413 static void rbd_dev_id_put(struct rbd_device
*rbd_dev
)
4415 struct list_head
*tmp
;
4416 int rbd_id
= rbd_dev
->dev_id
;
4419 rbd_assert(rbd_id
> 0);
4421 dout("rbd_dev %p released dev id %llu\n", rbd_dev
,
4422 (unsigned long long) rbd_dev
->dev_id
);
4423 spin_lock(&rbd_dev_list_lock
);
4424 list_del_init(&rbd_dev
->node
);
4427 * If the id being "put" is not the current maximum, there
4428 * is nothing special we need to do.
4430 if (rbd_id
!= atomic64_read(&rbd_dev_id_max
)) {
4431 spin_unlock(&rbd_dev_list_lock
);
4436 * We need to update the current maximum id. Search the
4437 * list to find out what it is. We're more likely to find
4438 * the maximum at the end, so search the list backward.
4441 list_for_each_prev(tmp
, &rbd_dev_list
) {
4442 struct rbd_device
*rbd_dev
;
4444 rbd_dev
= list_entry(tmp
, struct rbd_device
, node
);
4445 if (rbd_dev
->dev_id
> max_id
)
4446 max_id
= rbd_dev
->dev_id
;
4448 spin_unlock(&rbd_dev_list_lock
);
4451 * The max id could have been updated by rbd_dev_id_get(), in
4452 * which case it now accurately reflects the new maximum.
4453 * Be careful not to overwrite the maximum value in that
4456 atomic64_cmpxchg(&rbd_dev_id_max
, rbd_id
, max_id
);
4457 dout(" max dev id has been reset\n");
4461 * Skips over white space at *buf, and updates *buf to point to the
4462 * first found non-space character (if any). Returns the length of
4463 * the token (string of non-white space characters) found. Note
4464 * that *buf must be terminated with '\0'.
4466 static inline size_t next_token(const char **buf
)
4469 * These are the characters that produce nonzero for
4470 * isspace() in the "C" and "POSIX" locales.
4472 const char *spaces
= " \f\n\r\t\v";
4474 *buf
+= strspn(*buf
, spaces
); /* Find start of token */
4476 return strcspn(*buf
, spaces
); /* Return token length */
4480 * Finds the next token in *buf, and if the provided token buffer is
4481 * big enough, copies the found token into it. The result, if
4482 * copied, is guaranteed to be terminated with '\0'. Note that *buf
4483 * must be terminated with '\0' on entry.
4485 * Returns the length of the token found (not including the '\0').
4486 * Return value will be 0 if no token is found, and it will be >=
4487 * token_size if the token would not fit.
4489 * The *buf pointer will be updated to point beyond the end of the
4490 * found token. Note that this occurs even if the token buffer is
4491 * too small to hold it.
4493 static inline size_t copy_token(const char **buf
,
4499 len
= next_token(buf
);
4500 if (len
< token_size
) {
4501 memcpy(token
, *buf
, len
);
4502 *(token
+ len
) = '\0';
4510 * Finds the next token in *buf, dynamically allocates a buffer big
4511 * enough to hold a copy of it, and copies the token into the new
4512 * buffer. The copy is guaranteed to be terminated with '\0'. Note
4513 * that a duplicate buffer is created even for a zero-length token.
4515 * Returns a pointer to the newly-allocated duplicate, or a null
4516 * pointer if memory for the duplicate was not available. If
4517 * the lenp argument is a non-null pointer, the length of the token
4518 * (not including the '\0') is returned in *lenp.
4520 * If successful, the *buf pointer will be updated to point beyond
4521 * the end of the found token.
4523 * Note: uses GFP_KERNEL for allocation.
4525 static inline char *dup_token(const char **buf
, size_t *lenp
)
4530 len
= next_token(buf
);
4531 dup
= kmemdup(*buf
, len
+ 1, GFP_KERNEL
);
4534 *(dup
+ len
) = '\0';
4544 * Parse the options provided for an "rbd add" (i.e., rbd image
4545 * mapping) request. These arrive via a write to /sys/bus/rbd/add,
4546 * and the data written is passed here via a NUL-terminated buffer.
4547 * Returns 0 if successful or an error code otherwise.
4549 * The information extracted from these options is recorded in
4550 * the other parameters which return dynamically-allocated
4553 * The address of a pointer that will refer to a ceph options
4554 * structure. Caller must release the returned pointer using
4555 * ceph_destroy_options() when it is no longer needed.
4557 * Address of an rbd options pointer. Fully initialized by
4558 * this function; caller must release with kfree().
4560 * Address of an rbd image specification pointer. Fully
4561 * initialized by this function based on parsed options.
4562 * Caller must release with rbd_spec_put().
4564 * The options passed take this form:
4565 * <mon_addrs> <options> <pool_name> <image_name> [<snap_id>]
4568 * A comma-separated list of one or more monitor addresses.
4569 * A monitor address is an ip address, optionally followed
4570 * by a port number (separated by a colon).
4571 * I.e.: ip1[:port1][,ip2[:port2]...]
4573 * A comma-separated list of ceph and/or rbd options.
4575 * The name of the rados pool containing the rbd image.
4577 * The name of the image in that pool to map.
4579 * An optional snapshot id. If provided, the mapping will
4580 * present data from the image at the time that snapshot was
4581 * created. The image head is used if no snapshot id is
4582 * provided. Snapshot mappings are always read-only.
4584 static int rbd_add_parse_args(const char *buf
,
4585 struct ceph_options
**ceph_opts
,
4586 struct rbd_options
**opts
,
4587 struct rbd_spec
**rbd_spec
)
4591 const char *mon_addrs
;
4593 size_t mon_addrs_size
;
4594 struct rbd_spec
*spec
= NULL
;
4595 struct rbd_options
*rbd_opts
= NULL
;
4596 struct ceph_options
*copts
;
4599 /* The first four tokens are required */
4601 len
= next_token(&buf
);
4603 rbd_warn(NULL
, "no monitor address(es) provided");
4607 mon_addrs_size
= len
+ 1;
4611 options
= dup_token(&buf
, NULL
);
4615 rbd_warn(NULL
, "no options provided");
4619 spec
= rbd_spec_alloc();
4623 spec
->pool_name
= dup_token(&buf
, NULL
);
4624 if (!spec
->pool_name
)
4626 if (!*spec
->pool_name
) {
4627 rbd_warn(NULL
, "no pool name provided");
4631 spec
->image_name
= dup_token(&buf
, NULL
);
4632 if (!spec
->image_name
)
4634 if (!*spec
->image_name
) {
4635 rbd_warn(NULL
, "no image name provided");
4640 * Snapshot name is optional; default is to use "-"
4641 * (indicating the head/no snapshot).
4643 len
= next_token(&buf
);
4645 buf
= RBD_SNAP_HEAD_NAME
; /* No snapshot supplied */
4646 len
= sizeof (RBD_SNAP_HEAD_NAME
) - 1;
4647 } else if (len
> RBD_MAX_SNAP_NAME_LEN
) {
4648 ret
= -ENAMETOOLONG
;
4651 snap_name
= kmemdup(buf
, len
+ 1, GFP_KERNEL
);
4654 *(snap_name
+ len
) = '\0';
4655 spec
->snap_name
= snap_name
;
4657 /* Initialize all rbd options to the defaults */
4659 rbd_opts
= kzalloc(sizeof (*rbd_opts
), GFP_KERNEL
);
4663 rbd_opts
->read_only
= RBD_READ_ONLY_DEFAULT
;
4665 copts
= ceph_parse_options(options
, mon_addrs
,
4666 mon_addrs
+ mon_addrs_size
- 1,
4667 parse_rbd_opts_token
, rbd_opts
);
4668 if (IS_ERR(copts
)) {
4669 ret
= PTR_ERR(copts
);
4690 * An rbd format 2 image has a unique identifier, distinct from the
4691 * name given to it by the user. Internally, that identifier is
4692 * what's used to specify the names of objects related to the image.
4694 * A special "rbd id" object is used to map an rbd image name to its
4695 * id. If that object doesn't exist, then there is no v2 rbd image
4696 * with the supplied name.
4698 * This function will record the given rbd_dev's image_id field if
4699 * it can be determined, and in that case will return 0. If any
4700 * errors occur a negative errno will be returned and the rbd_dev's
4701 * image_id field will be unchanged (and should be NULL).
4703 static int rbd_dev_image_id(struct rbd_device
*rbd_dev
)
4712 * When probing a parent image, the image id is already
4713 * known (and the image name likely is not). There's no
4714 * need to fetch the image id again in this case. We
4715 * do still need to set the image format though.
4717 if (rbd_dev
->spec
->image_id
) {
4718 rbd_dev
->image_format
= *rbd_dev
->spec
->image_id
? 2 : 1;
4724 * First, see if the format 2 image id file exists, and if
4725 * so, get the image's persistent id from it.
4727 size
= sizeof (RBD_ID_PREFIX
) + strlen(rbd_dev
->spec
->image_name
);
4728 object_name
= kmalloc(size
, GFP_NOIO
);
4731 sprintf(object_name
, "%s%s", RBD_ID_PREFIX
, rbd_dev
->spec
->image_name
);
4732 dout("rbd id object name is %s\n", object_name
);
4734 /* Response will be an encoded string, which includes a length */
4736 size
= sizeof (__le32
) + RBD_IMAGE_ID_LEN_MAX
;
4737 response
= kzalloc(size
, GFP_NOIO
);
4743 /* If it doesn't exist we'll assume it's a format 1 image */
4745 ret
= rbd_obj_method_sync(rbd_dev
, object_name
,
4746 "rbd", "get_id", NULL
, 0,
4747 response
, RBD_IMAGE_ID_LEN_MAX
);
4748 dout("%s: rbd_obj_method_sync returned %d\n", __func__
, ret
);
4749 if (ret
== -ENOENT
) {
4750 image_id
= kstrdup("", GFP_KERNEL
);
4751 ret
= image_id
? 0 : -ENOMEM
;
4753 rbd_dev
->image_format
= 1;
4754 } else if (ret
> sizeof (__le32
)) {
4757 image_id
= ceph_extract_encoded_string(&p
, p
+ ret
,
4759 ret
= IS_ERR(image_id
) ? PTR_ERR(image_id
) : 0;
4761 rbd_dev
->image_format
= 2;
4767 rbd_dev
->spec
->image_id
= image_id
;
4768 dout("image_id is %s\n", image_id
);
4778 * Undo whatever state changes are made by v1 or v2 header info
4781 static void rbd_dev_unprobe(struct rbd_device
*rbd_dev
)
4783 struct rbd_image_header
*header
;
4785 /* Drop parent reference unless it's already been done (or none) */
4787 if (rbd_dev
->parent_overlap
)
4788 rbd_dev_parent_put(rbd_dev
);
4790 /* Free dynamic fields from the header, then zero it out */
4792 header
= &rbd_dev
->header
;
4793 ceph_put_snap_context(header
->snapc
);
4794 kfree(header
->snap_sizes
);
4795 kfree(header
->snap_names
);
4796 kfree(header
->object_prefix
);
4797 memset(header
, 0, sizeof (*header
));
4800 static int rbd_dev_v2_header_onetime(struct rbd_device
*rbd_dev
)
4804 ret
= rbd_dev_v2_object_prefix(rbd_dev
);
4809 * Get the and check features for the image. Currently the
4810 * features are assumed to never change.
4812 ret
= rbd_dev_v2_features(rbd_dev
);
4816 /* If the image supports fancy striping, get its parameters */
4818 if (rbd_dev
->header
.features
& RBD_FEATURE_STRIPINGV2
) {
4819 ret
= rbd_dev_v2_striping_info(rbd_dev
);
4823 /* No support for crypto and compression type format 2 images */
4827 rbd_dev
->header
.features
= 0;
4828 kfree(rbd_dev
->header
.object_prefix
);
4829 rbd_dev
->header
.object_prefix
= NULL
;
4834 static int rbd_dev_probe_parent(struct rbd_device
*rbd_dev
)
4836 struct rbd_device
*parent
= NULL
;
4837 struct rbd_spec
*parent_spec
;
4838 struct rbd_client
*rbdc
;
4841 if (!rbd_dev
->parent_spec
)
4844 * We need to pass a reference to the client and the parent
4845 * spec when creating the parent rbd_dev. Images related by
4846 * parent/child relationships always share both.
4848 parent_spec
= rbd_spec_get(rbd_dev
->parent_spec
);
4849 rbdc
= __rbd_get_client(rbd_dev
->rbd_client
);
4852 parent
= rbd_dev_create(rbdc
, parent_spec
);
4856 ret
= rbd_dev_image_probe(parent
, false);
4859 rbd_dev
->parent
= parent
;
4860 atomic_set(&rbd_dev
->parent_ref
, 1);
4865 rbd_dev_unparent(rbd_dev
);
4866 kfree(rbd_dev
->header_name
);
4867 rbd_dev_destroy(parent
);
4869 rbd_put_client(rbdc
);
4870 rbd_spec_put(parent_spec
);
4876 static int rbd_dev_device_setup(struct rbd_device
*rbd_dev
)
4880 /* generate unique id: find highest unique id, add one */
4881 rbd_dev_id_get(rbd_dev
);
4883 /* Fill in the device name, now that we have its id. */
4884 BUILD_BUG_ON(DEV_NAME_LEN
4885 < sizeof (RBD_DRV_NAME
) + MAX_INT_FORMAT_WIDTH
);
4886 sprintf(rbd_dev
->name
, "%s%d", RBD_DRV_NAME
, rbd_dev
->dev_id
);
4888 /* Get our block major device number. */
4890 ret
= register_blkdev(0, rbd_dev
->name
);
4893 rbd_dev
->major
= ret
;
4895 /* Set up the blkdev mapping. */
4897 ret
= rbd_init_disk(rbd_dev
);
4899 goto err_out_blkdev
;
4901 ret
= rbd_dev_mapping_set(rbd_dev
);
4904 set_capacity(rbd_dev
->disk
, rbd_dev
->mapping
.size
/ SECTOR_SIZE
);
4906 ret
= rbd_bus_add_dev(rbd_dev
);
4908 goto err_out_mapping
;
4910 /* Everything's ready. Announce the disk to the world. */
4912 set_bit(RBD_DEV_FLAG_EXISTS
, &rbd_dev
->flags
);
4913 add_disk(rbd_dev
->disk
);
4915 pr_info("%s: added with size 0x%llx\n", rbd_dev
->disk
->disk_name
,
4916 (unsigned long long) rbd_dev
->mapping
.size
);
4921 rbd_dev_mapping_clear(rbd_dev
);
4923 rbd_free_disk(rbd_dev
);
4925 unregister_blkdev(rbd_dev
->major
, rbd_dev
->name
);
4927 rbd_dev_id_put(rbd_dev
);
4928 rbd_dev_mapping_clear(rbd_dev
);
4933 static int rbd_dev_header_name(struct rbd_device
*rbd_dev
)
4935 struct rbd_spec
*spec
= rbd_dev
->spec
;
4938 /* Record the header object name for this rbd image. */
4940 rbd_assert(rbd_image_format_valid(rbd_dev
->image_format
));
4942 if (rbd_dev
->image_format
== 1)
4943 size
= strlen(spec
->image_name
) + sizeof (RBD_SUFFIX
);
4945 size
= sizeof (RBD_HEADER_PREFIX
) + strlen(spec
->image_id
);
4947 rbd_dev
->header_name
= kmalloc(size
, GFP_KERNEL
);
4948 if (!rbd_dev
->header_name
)
4951 if (rbd_dev
->image_format
== 1)
4952 sprintf(rbd_dev
->header_name
, "%s%s",
4953 spec
->image_name
, RBD_SUFFIX
);
4955 sprintf(rbd_dev
->header_name
, "%s%s",
4956 RBD_HEADER_PREFIX
, spec
->image_id
);
4960 static void rbd_dev_image_release(struct rbd_device
*rbd_dev
)
4962 rbd_dev_unprobe(rbd_dev
);
4963 kfree(rbd_dev
->header_name
);
4964 rbd_dev
->header_name
= NULL
;
4965 rbd_dev
->image_format
= 0;
4966 kfree(rbd_dev
->spec
->image_id
);
4967 rbd_dev
->spec
->image_id
= NULL
;
4969 rbd_dev_destroy(rbd_dev
);
4973 * Probe for the existence of the header object for the given rbd
4974 * device. If this image is the one being mapped (i.e., not a
4975 * parent), initiate a watch on its header object before using that
4976 * object to get detailed information about the rbd image.
4978 static int rbd_dev_image_probe(struct rbd_device
*rbd_dev
, bool mapping
)
4984 * Get the id from the image id object. Unless there's an
4985 * error, rbd_dev->spec->image_id will be filled in with
4986 * a dynamically-allocated string, and rbd_dev->image_format
4987 * will be set to either 1 or 2.
4989 ret
= rbd_dev_image_id(rbd_dev
);
4992 rbd_assert(rbd_dev
->spec
->image_id
);
4993 rbd_assert(rbd_image_format_valid(rbd_dev
->image_format
));
4995 ret
= rbd_dev_header_name(rbd_dev
);
4997 goto err_out_format
;
5000 ret
= rbd_dev_header_watch_sync(rbd_dev
, true);
5002 goto out_header_name
;
5005 if (rbd_dev
->image_format
== 1)
5006 ret
= rbd_dev_v1_header_info(rbd_dev
);
5008 ret
= rbd_dev_v2_header_info(rbd_dev
);
5012 ret
= rbd_dev_spec_update(rbd_dev
);
5016 ret
= rbd_dev_probe_parent(rbd_dev
);
5020 dout("discovered format %u image, header name is %s\n",
5021 rbd_dev
->image_format
, rbd_dev
->header_name
);
5025 rbd_dev_unprobe(rbd_dev
);
5028 tmp
= rbd_dev_header_watch_sync(rbd_dev
, false);
5030 rbd_warn(rbd_dev
, "unable to tear down "
5031 "watch request (%d)\n", tmp
);
5034 kfree(rbd_dev
->header_name
);
5035 rbd_dev
->header_name
= NULL
;
5037 rbd_dev
->image_format
= 0;
5038 kfree(rbd_dev
->spec
->image_id
);
5039 rbd_dev
->spec
->image_id
= NULL
;
5041 dout("probe failed, returning %d\n", ret
);
5046 static ssize_t
rbd_add(struct bus_type
*bus
,
5050 struct rbd_device
*rbd_dev
= NULL
;
5051 struct ceph_options
*ceph_opts
= NULL
;
5052 struct rbd_options
*rbd_opts
= NULL
;
5053 struct rbd_spec
*spec
= NULL
;
5054 struct rbd_client
*rbdc
;
5055 struct ceph_osd_client
*osdc
;
5059 if (!try_module_get(THIS_MODULE
))
5062 /* parse add command */
5063 rc
= rbd_add_parse_args(buf
, &ceph_opts
, &rbd_opts
, &spec
);
5065 goto err_out_module
;
5066 read_only
= rbd_opts
->read_only
;
5068 rbd_opts
= NULL
; /* done with this */
5070 rbdc
= rbd_get_client(ceph_opts
);
5077 osdc
= &rbdc
->client
->osdc
;
5078 rc
= ceph_pg_poolid_by_name(osdc
->osdmap
, spec
->pool_name
);
5080 goto err_out_client
;
5081 spec
->pool_id
= (u64
)rc
;
5083 /* The ceph file layout needs to fit pool id in 32 bits */
5085 if (spec
->pool_id
> (u64
)U32_MAX
) {
5086 rbd_warn(NULL
, "pool id too large (%llu > %u)\n",
5087 (unsigned long long)spec
->pool_id
, U32_MAX
);
5089 goto err_out_client
;
5092 rbd_dev
= rbd_dev_create(rbdc
, spec
);
5094 goto err_out_client
;
5095 rbdc
= NULL
; /* rbd_dev now owns this */
5096 spec
= NULL
; /* rbd_dev now owns this */
5098 rc
= rbd_dev_image_probe(rbd_dev
, true);
5100 goto err_out_rbd_dev
;
5102 /* If we are mapping a snapshot it must be marked read-only */
5104 if (rbd_dev
->spec
->snap_id
!= CEPH_NOSNAP
)
5106 rbd_dev
->mapping
.read_only
= read_only
;
5108 rc
= rbd_dev_device_setup(rbd_dev
);
5110 rbd_dev_image_release(rbd_dev
);
5111 goto err_out_module
;
5117 rbd_dev_destroy(rbd_dev
);
5119 rbd_put_client(rbdc
);
5123 module_put(THIS_MODULE
);
5125 dout("Error adding device %s\n", buf
);
5130 static void rbd_dev_device_release(struct device
*dev
)
5132 struct rbd_device
*rbd_dev
= dev_to_rbd_dev(dev
);
5134 rbd_free_disk(rbd_dev
);
5135 clear_bit(RBD_DEV_FLAG_EXISTS
, &rbd_dev
->flags
);
5136 rbd_dev_mapping_clear(rbd_dev
);
5137 unregister_blkdev(rbd_dev
->major
, rbd_dev
->name
);
5139 rbd_dev_id_put(rbd_dev
);
5140 rbd_dev_mapping_clear(rbd_dev
);
5143 static void rbd_dev_remove_parent(struct rbd_device
*rbd_dev
)
5145 while (rbd_dev
->parent
) {
5146 struct rbd_device
*first
= rbd_dev
;
5147 struct rbd_device
*second
= first
->parent
;
5148 struct rbd_device
*third
;
5151 * Follow to the parent with no grandparent and
5154 while (second
&& (third
= second
->parent
)) {
5159 rbd_dev_image_release(second
);
5160 first
->parent
= NULL
;
5161 first
->parent_overlap
= 0;
5163 rbd_assert(first
->parent_spec
);
5164 rbd_spec_put(first
->parent_spec
);
5165 first
->parent_spec
= NULL
;
5169 static ssize_t
rbd_remove(struct bus_type
*bus
,
5173 struct rbd_device
*rbd_dev
= NULL
;
5174 struct list_head
*tmp
;
5177 bool already
= false;
5180 ret
= kstrtoul(buf
, 10, &ul
);
5184 /* convert to int; abort if we lost anything in the conversion */
5190 spin_lock(&rbd_dev_list_lock
);
5191 list_for_each(tmp
, &rbd_dev_list
) {
5192 rbd_dev
= list_entry(tmp
, struct rbd_device
, node
);
5193 if (rbd_dev
->dev_id
== dev_id
) {
5199 spin_lock_irq(&rbd_dev
->lock
);
5200 if (rbd_dev
->open_count
)
5203 already
= test_and_set_bit(RBD_DEV_FLAG_REMOVING
,
5205 spin_unlock_irq(&rbd_dev
->lock
);
5207 spin_unlock(&rbd_dev_list_lock
);
5208 if (ret
< 0 || already
)
5211 ret
= rbd_dev_header_watch_sync(rbd_dev
, false);
5213 rbd_warn(rbd_dev
, "failed to cancel watch event (%d)\n", ret
);
5216 * flush remaining watch callbacks - these must be complete
5217 * before the osd_client is shutdown
5219 dout("%s: flushing notifies", __func__
);
5220 ceph_osdc_flush_notifies(&rbd_dev
->rbd_client
->client
->osdc
);
5222 * Don't free anything from rbd_dev->disk until after all
5223 * notifies are completely processed. Otherwise
5224 * rbd_bus_del_dev() will race with rbd_watch_cb(), resulting
5225 * in a potential use after free of rbd_dev->disk or rbd_dev.
5227 rbd_bus_del_dev(rbd_dev
);
5228 rbd_dev_image_release(rbd_dev
);
5229 module_put(THIS_MODULE
);
5235 * create control files in sysfs
5238 static int rbd_sysfs_init(void)
5242 ret
= device_register(&rbd_root_dev
);
5246 ret
= bus_register(&rbd_bus_type
);
5248 device_unregister(&rbd_root_dev
);
5253 static void rbd_sysfs_cleanup(void)
5255 bus_unregister(&rbd_bus_type
);
5256 device_unregister(&rbd_root_dev
);
5259 static int rbd_slab_init(void)
5261 rbd_assert(!rbd_img_request_cache
);
5262 rbd_img_request_cache
= kmem_cache_create("rbd_img_request",
5263 sizeof (struct rbd_img_request
),
5264 __alignof__(struct rbd_img_request
),
5266 if (!rbd_img_request_cache
)
5269 rbd_assert(!rbd_obj_request_cache
);
5270 rbd_obj_request_cache
= kmem_cache_create("rbd_obj_request",
5271 sizeof (struct rbd_obj_request
),
5272 __alignof__(struct rbd_obj_request
),
5274 if (!rbd_obj_request_cache
)
5277 rbd_assert(!rbd_segment_name_cache
);
5278 rbd_segment_name_cache
= kmem_cache_create("rbd_segment_name",
5279 MAX_OBJ_NAME_SIZE
+ 1, 1, 0, NULL
);
5280 if (rbd_segment_name_cache
)
5283 if (rbd_obj_request_cache
) {
5284 kmem_cache_destroy(rbd_obj_request_cache
);
5285 rbd_obj_request_cache
= NULL
;
5288 kmem_cache_destroy(rbd_img_request_cache
);
5289 rbd_img_request_cache
= NULL
;
5294 static void rbd_slab_exit(void)
5296 rbd_assert(rbd_segment_name_cache
);
5297 kmem_cache_destroy(rbd_segment_name_cache
);
5298 rbd_segment_name_cache
= NULL
;
5300 rbd_assert(rbd_obj_request_cache
);
5301 kmem_cache_destroy(rbd_obj_request_cache
);
5302 rbd_obj_request_cache
= NULL
;
5304 rbd_assert(rbd_img_request_cache
);
5305 kmem_cache_destroy(rbd_img_request_cache
);
5306 rbd_img_request_cache
= NULL
;
5309 static int __init
rbd_init(void)
5313 if (!libceph_compatible(NULL
)) {
5314 rbd_warn(NULL
, "libceph incompatibility (quitting)");
5318 rc
= rbd_slab_init();
5321 rc
= rbd_sysfs_init();
5325 pr_info("loaded " RBD_DRV_NAME_LONG
"\n");
5330 static void __exit
rbd_exit(void)
5332 rbd_sysfs_cleanup();
5336 module_init(rbd_init
);
5337 module_exit(rbd_exit
);
5339 MODULE_AUTHOR("Alex Elder <elder@inktank.com>");
5340 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
5341 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
5342 MODULE_DESCRIPTION("rados block device");
5344 /* following authorship retained from original osdblk.c */
5345 MODULE_AUTHOR("Jeff Garzik <jeff@garzik.org>");
5347 MODULE_LICENSE("GPL");