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(ctl_mutex
); /* Serialize open/close/setup/teardown */
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 struct bus_attribute rbd_bus_attrs
[] = {
401 __ATTR(add
, S_IWUSR
, NULL
, rbd_add
),
402 __ATTR(remove
, S_IWUSR
, NULL
, rbd_remove
),
406 static struct bus_type rbd_bus_type
= {
408 .bus_attrs
= rbd_bus_attrs
,
411 static void rbd_root_dev_release(struct device
*dev
)
415 static struct device rbd_root_dev
= {
417 .release
= rbd_root_dev_release
,
420 static __printf(2, 3)
421 void rbd_warn(struct rbd_device
*rbd_dev
, const char *fmt
, ...)
423 struct va_format vaf
;
431 printk(KERN_WARNING
"%s: %pV\n", RBD_DRV_NAME
, &vaf
);
432 else if (rbd_dev
->disk
)
433 printk(KERN_WARNING
"%s: %s: %pV\n",
434 RBD_DRV_NAME
, rbd_dev
->disk
->disk_name
, &vaf
);
435 else if (rbd_dev
->spec
&& rbd_dev
->spec
->image_name
)
436 printk(KERN_WARNING
"%s: image %s: %pV\n",
437 RBD_DRV_NAME
, rbd_dev
->spec
->image_name
, &vaf
);
438 else if (rbd_dev
->spec
&& rbd_dev
->spec
->image_id
)
439 printk(KERN_WARNING
"%s: id %s: %pV\n",
440 RBD_DRV_NAME
, rbd_dev
->spec
->image_id
, &vaf
);
442 printk(KERN_WARNING
"%s: rbd_dev %p: %pV\n",
443 RBD_DRV_NAME
, rbd_dev
, &vaf
);
448 #define rbd_assert(expr) \
449 if (unlikely(!(expr))) { \
450 printk(KERN_ERR "\nAssertion failure in %s() " \
452 "\trbd_assert(%s);\n\n", \
453 __func__, __LINE__, #expr); \
456 #else /* !RBD_DEBUG */
457 # define rbd_assert(expr) ((void) 0)
458 #endif /* !RBD_DEBUG */
460 static void rbd_osd_copyup_callback(struct rbd_obj_request
*obj_request
);
461 static int rbd_img_obj_request_submit(struct rbd_obj_request
*obj_request
);
462 static void rbd_img_parent_read(struct rbd_obj_request
*obj_request
);
463 static void rbd_dev_remove_parent(struct rbd_device
*rbd_dev
);
465 static int rbd_dev_refresh(struct rbd_device
*rbd_dev
);
466 static int rbd_dev_v2_header_onetime(struct rbd_device
*rbd_dev
);
467 static int rbd_dev_v2_header_info(struct rbd_device
*rbd_dev
);
468 static const char *rbd_dev_v2_snap_name(struct rbd_device
*rbd_dev
,
470 static int _rbd_dev_v2_snap_size(struct rbd_device
*rbd_dev
, u64 snap_id
,
471 u8
*order
, u64
*snap_size
);
472 static int _rbd_dev_v2_snap_features(struct rbd_device
*rbd_dev
, u64 snap_id
,
474 static u64
rbd_snap_id_by_name(struct rbd_device
*rbd_dev
, const char *name
);
476 static int rbd_open(struct block_device
*bdev
, fmode_t mode
)
478 struct rbd_device
*rbd_dev
= bdev
->bd_disk
->private_data
;
479 bool removing
= false;
481 if ((mode
& FMODE_WRITE
) && rbd_dev
->mapping
.read_only
)
484 spin_lock_irq(&rbd_dev
->lock
);
485 if (test_bit(RBD_DEV_FLAG_REMOVING
, &rbd_dev
->flags
))
488 rbd_dev
->open_count
++;
489 spin_unlock_irq(&rbd_dev
->lock
);
493 mutex_lock_nested(&ctl_mutex
, SINGLE_DEPTH_NESTING
);
494 (void) get_device(&rbd_dev
->dev
);
495 set_device_ro(bdev
, rbd_dev
->mapping
.read_only
);
496 mutex_unlock(&ctl_mutex
);
501 static void rbd_release(struct gendisk
*disk
, fmode_t mode
)
503 struct rbd_device
*rbd_dev
= disk
->private_data
;
504 unsigned long open_count_before
;
506 spin_lock_irq(&rbd_dev
->lock
);
507 open_count_before
= rbd_dev
->open_count
--;
508 spin_unlock_irq(&rbd_dev
->lock
);
509 rbd_assert(open_count_before
> 0);
511 mutex_lock_nested(&ctl_mutex
, SINGLE_DEPTH_NESTING
);
512 put_device(&rbd_dev
->dev
);
513 mutex_unlock(&ctl_mutex
);
516 static const struct block_device_operations rbd_bd_ops
= {
517 .owner
= THIS_MODULE
,
519 .release
= rbd_release
,
523 * Initialize an rbd client instance. Success or not, this function
524 * consumes ceph_opts.
526 static struct rbd_client
*rbd_client_create(struct ceph_options
*ceph_opts
)
528 struct rbd_client
*rbdc
;
531 dout("%s:\n", __func__
);
532 rbdc
= kmalloc(sizeof(struct rbd_client
), GFP_KERNEL
);
536 kref_init(&rbdc
->kref
);
537 INIT_LIST_HEAD(&rbdc
->node
);
539 mutex_lock_nested(&ctl_mutex
, SINGLE_DEPTH_NESTING
);
541 rbdc
->client
= ceph_create_client(ceph_opts
, rbdc
, 0, 0);
542 if (IS_ERR(rbdc
->client
))
544 ceph_opts
= NULL
; /* Now rbdc->client is responsible for ceph_opts */
546 ret
= ceph_open_session(rbdc
->client
);
550 spin_lock(&rbd_client_list_lock
);
551 list_add_tail(&rbdc
->node
, &rbd_client_list
);
552 spin_unlock(&rbd_client_list_lock
);
554 mutex_unlock(&ctl_mutex
);
555 dout("%s: rbdc %p\n", __func__
, rbdc
);
560 ceph_destroy_client(rbdc
->client
);
562 mutex_unlock(&ctl_mutex
);
566 ceph_destroy_options(ceph_opts
);
567 dout("%s: error %d\n", __func__
, ret
);
572 static struct rbd_client
*__rbd_get_client(struct rbd_client
*rbdc
)
574 kref_get(&rbdc
->kref
);
580 * Find a ceph client with specific addr and configuration. If
581 * found, bump its reference count.
583 static struct rbd_client
*rbd_client_find(struct ceph_options
*ceph_opts
)
585 struct rbd_client
*client_node
;
588 if (ceph_opts
->flags
& CEPH_OPT_NOSHARE
)
591 spin_lock(&rbd_client_list_lock
);
592 list_for_each_entry(client_node
, &rbd_client_list
, node
) {
593 if (!ceph_compare_options(ceph_opts
, client_node
->client
)) {
594 __rbd_get_client(client_node
);
600 spin_unlock(&rbd_client_list_lock
);
602 return found
? client_node
: NULL
;
612 /* string args above */
615 /* Boolean args above */
619 static match_table_t rbd_opts_tokens
= {
621 /* string args above */
622 {Opt_read_only
, "read_only"},
623 {Opt_read_only
, "ro"}, /* Alternate spelling */
624 {Opt_read_write
, "read_write"},
625 {Opt_read_write
, "rw"}, /* Alternate spelling */
626 /* Boolean args above */
634 #define RBD_READ_ONLY_DEFAULT false
636 static int parse_rbd_opts_token(char *c
, void *private)
638 struct rbd_options
*rbd_opts
= private;
639 substring_t argstr
[MAX_OPT_ARGS
];
640 int token
, intval
, ret
;
642 token
= match_token(c
, rbd_opts_tokens
, argstr
);
646 if (token
< Opt_last_int
) {
647 ret
= match_int(&argstr
[0], &intval
);
649 pr_err("bad mount option arg (not int) "
653 dout("got int token %d val %d\n", token
, intval
);
654 } else if (token
> Opt_last_int
&& token
< Opt_last_string
) {
655 dout("got string token %d val %s\n", token
,
657 } else if (token
> Opt_last_string
&& token
< Opt_last_bool
) {
658 dout("got Boolean token %d\n", token
);
660 dout("got token %d\n", token
);
665 rbd_opts
->read_only
= true;
668 rbd_opts
->read_only
= false;
678 * Get a ceph client with specific addr and configuration, if one does
679 * not exist create it. Either way, ceph_opts is consumed by this
682 static struct rbd_client
*rbd_get_client(struct ceph_options
*ceph_opts
)
684 struct rbd_client
*rbdc
;
686 rbdc
= rbd_client_find(ceph_opts
);
687 if (rbdc
) /* using an existing client */
688 ceph_destroy_options(ceph_opts
);
690 rbdc
= rbd_client_create(ceph_opts
);
696 * Destroy ceph client
698 * Caller must hold rbd_client_list_lock.
700 static void rbd_client_release(struct kref
*kref
)
702 struct rbd_client
*rbdc
= container_of(kref
, struct rbd_client
, kref
);
704 dout("%s: rbdc %p\n", __func__
, rbdc
);
705 spin_lock(&rbd_client_list_lock
);
706 list_del(&rbdc
->node
);
707 spin_unlock(&rbd_client_list_lock
);
709 ceph_destroy_client(rbdc
->client
);
714 * Drop reference to ceph client node. If it's not referenced anymore, release
717 static void rbd_put_client(struct rbd_client
*rbdc
)
720 kref_put(&rbdc
->kref
, rbd_client_release
);
723 static bool rbd_image_format_valid(u32 image_format
)
725 return image_format
== 1 || image_format
== 2;
728 static bool rbd_dev_ondisk_valid(struct rbd_image_header_ondisk
*ondisk
)
733 /* The header has to start with the magic rbd header text */
734 if (memcmp(&ondisk
->text
, RBD_HEADER_TEXT
, sizeof (RBD_HEADER_TEXT
)))
737 /* The bio layer requires at least sector-sized I/O */
739 if (ondisk
->options
.order
< SECTOR_SHIFT
)
742 /* If we use u64 in a few spots we may be able to loosen this */
744 if (ondisk
->options
.order
> 8 * sizeof (int) - 1)
748 * The size of a snapshot header has to fit in a size_t, and
749 * that limits the number of snapshots.
751 snap_count
= le32_to_cpu(ondisk
->snap_count
);
752 size
= SIZE_MAX
- sizeof (struct ceph_snap_context
);
753 if (snap_count
> size
/ sizeof (__le64
))
757 * Not only that, but the size of the entire the snapshot
758 * header must also be representable in a size_t.
760 size
-= snap_count
* sizeof (__le64
);
761 if ((u64
) size
< le64_to_cpu(ondisk
->snap_names_len
))
768 * Fill an rbd image header with information from the given format 1
771 static int rbd_header_from_disk(struct rbd_device
*rbd_dev
,
772 struct rbd_image_header_ondisk
*ondisk
)
774 struct rbd_image_header
*header
= &rbd_dev
->header
;
775 bool first_time
= header
->object_prefix
== NULL
;
776 struct ceph_snap_context
*snapc
;
777 char *object_prefix
= NULL
;
778 char *snap_names
= NULL
;
779 u64
*snap_sizes
= NULL
;
785 /* Allocate this now to avoid having to handle failure below */
790 len
= strnlen(ondisk
->object_prefix
,
791 sizeof (ondisk
->object_prefix
));
792 object_prefix
= kmalloc(len
+ 1, GFP_KERNEL
);
795 memcpy(object_prefix
, ondisk
->object_prefix
, len
);
796 object_prefix
[len
] = '\0';
799 /* Allocate the snapshot context and fill it in */
801 snap_count
= le32_to_cpu(ondisk
->snap_count
);
802 snapc
= ceph_create_snap_context(snap_count
, GFP_KERNEL
);
805 snapc
->seq
= le64_to_cpu(ondisk
->snap_seq
);
807 struct rbd_image_snap_ondisk
*snaps
;
808 u64 snap_names_len
= le64_to_cpu(ondisk
->snap_names_len
);
810 /* We'll keep a copy of the snapshot names... */
812 if (snap_names_len
> (u64
)SIZE_MAX
)
814 snap_names
= kmalloc(snap_names_len
, GFP_KERNEL
);
818 /* ...as well as the array of their sizes. */
820 size
= snap_count
* sizeof (*header
->snap_sizes
);
821 snap_sizes
= kmalloc(size
, GFP_KERNEL
);
826 * Copy the names, and fill in each snapshot's id
829 * Note that rbd_dev_v1_header_info() guarantees the
830 * ondisk buffer we're working with has
831 * snap_names_len bytes beyond the end of the
832 * snapshot id array, this memcpy() is safe.
834 memcpy(snap_names
, &ondisk
->snaps
[snap_count
], snap_names_len
);
835 snaps
= ondisk
->snaps
;
836 for (i
= 0; i
< snap_count
; i
++) {
837 snapc
->snaps
[i
] = le64_to_cpu(snaps
[i
].id
);
838 snap_sizes
[i
] = le64_to_cpu(snaps
[i
].image_size
);
842 /* We won't fail any more, fill in the header */
844 down_write(&rbd_dev
->header_rwsem
);
846 header
->object_prefix
= object_prefix
;
847 header
->obj_order
= ondisk
->options
.order
;
848 header
->crypt_type
= ondisk
->options
.crypt_type
;
849 header
->comp_type
= ondisk
->options
.comp_type
;
850 /* The rest aren't used for format 1 images */
851 header
->stripe_unit
= 0;
852 header
->stripe_count
= 0;
853 header
->features
= 0;
855 ceph_put_snap_context(header
->snapc
);
856 kfree(header
->snap_names
);
857 kfree(header
->snap_sizes
);
860 /* The remaining fields always get updated (when we refresh) */
862 header
->image_size
= le64_to_cpu(ondisk
->image_size
);
863 header
->snapc
= snapc
;
864 header
->snap_names
= snap_names
;
865 header
->snap_sizes
= snap_sizes
;
867 /* Make sure mapping size is consistent with header info */
869 if (rbd_dev
->spec
->snap_id
== CEPH_NOSNAP
|| first_time
)
870 if (rbd_dev
->mapping
.size
!= header
->image_size
)
871 rbd_dev
->mapping
.size
= header
->image_size
;
873 up_write(&rbd_dev
->header_rwsem
);
881 ceph_put_snap_context(snapc
);
882 kfree(object_prefix
);
887 static const char *_rbd_dev_v1_snap_name(struct rbd_device
*rbd_dev
, u32 which
)
889 const char *snap_name
;
891 rbd_assert(which
< rbd_dev
->header
.snapc
->num_snaps
);
893 /* Skip over names until we find the one we are looking for */
895 snap_name
= rbd_dev
->header
.snap_names
;
897 snap_name
+= strlen(snap_name
) + 1;
899 return kstrdup(snap_name
, GFP_KERNEL
);
903 * Snapshot id comparison function for use with qsort()/bsearch().
904 * Note that result is for snapshots in *descending* order.
906 static int snapid_compare_reverse(const void *s1
, const void *s2
)
908 u64 snap_id1
= *(u64
*)s1
;
909 u64 snap_id2
= *(u64
*)s2
;
911 if (snap_id1
< snap_id2
)
913 return snap_id1
== snap_id2
? 0 : -1;
917 * Search a snapshot context to see if the given snapshot id is
920 * Returns the position of the snapshot id in the array if it's found,
921 * or BAD_SNAP_INDEX otherwise.
923 * Note: The snapshot array is in kept sorted (by the osd) in
924 * reverse order, highest snapshot id first.
926 static u32
rbd_dev_snap_index(struct rbd_device
*rbd_dev
, u64 snap_id
)
928 struct ceph_snap_context
*snapc
= rbd_dev
->header
.snapc
;
931 found
= bsearch(&snap_id
, &snapc
->snaps
, snapc
->num_snaps
,
932 sizeof (snap_id
), snapid_compare_reverse
);
934 return found
? (u32
)(found
- &snapc
->snaps
[0]) : BAD_SNAP_INDEX
;
937 static const char *rbd_dev_v1_snap_name(struct rbd_device
*rbd_dev
,
941 const char *snap_name
;
943 which
= rbd_dev_snap_index(rbd_dev
, snap_id
);
944 if (which
== BAD_SNAP_INDEX
)
945 return ERR_PTR(-ENOENT
);
947 snap_name
= _rbd_dev_v1_snap_name(rbd_dev
, which
);
948 return snap_name
? snap_name
: ERR_PTR(-ENOMEM
);
951 static const char *rbd_snap_name(struct rbd_device
*rbd_dev
, u64 snap_id
)
953 if (snap_id
== CEPH_NOSNAP
)
954 return RBD_SNAP_HEAD_NAME
;
956 rbd_assert(rbd_image_format_valid(rbd_dev
->image_format
));
957 if (rbd_dev
->image_format
== 1)
958 return rbd_dev_v1_snap_name(rbd_dev
, snap_id
);
960 return rbd_dev_v2_snap_name(rbd_dev
, snap_id
);
963 static int rbd_snap_size(struct rbd_device
*rbd_dev
, u64 snap_id
,
966 rbd_assert(rbd_image_format_valid(rbd_dev
->image_format
));
967 if (snap_id
== CEPH_NOSNAP
) {
968 *snap_size
= rbd_dev
->header
.image_size
;
969 } else if (rbd_dev
->image_format
== 1) {
972 which
= rbd_dev_snap_index(rbd_dev
, snap_id
);
973 if (which
== BAD_SNAP_INDEX
)
976 *snap_size
= rbd_dev
->header
.snap_sizes
[which
];
981 ret
= _rbd_dev_v2_snap_size(rbd_dev
, snap_id
, NULL
, &size
);
990 static int rbd_snap_features(struct rbd_device
*rbd_dev
, u64 snap_id
,
993 rbd_assert(rbd_image_format_valid(rbd_dev
->image_format
));
994 if (snap_id
== CEPH_NOSNAP
) {
995 *snap_features
= rbd_dev
->header
.features
;
996 } else if (rbd_dev
->image_format
== 1) {
997 *snap_features
= 0; /* No features for format 1 */
1002 ret
= _rbd_dev_v2_snap_features(rbd_dev
, snap_id
, &features
);
1006 *snap_features
= features
;
1011 static int rbd_dev_mapping_set(struct rbd_device
*rbd_dev
)
1013 u64 snap_id
= rbd_dev
->spec
->snap_id
;
1018 ret
= rbd_snap_size(rbd_dev
, snap_id
, &size
);
1021 ret
= rbd_snap_features(rbd_dev
, snap_id
, &features
);
1025 rbd_dev
->mapping
.size
= size
;
1026 rbd_dev
->mapping
.features
= features
;
1031 static void rbd_dev_mapping_clear(struct rbd_device
*rbd_dev
)
1033 rbd_dev
->mapping
.size
= 0;
1034 rbd_dev
->mapping
.features
= 0;
1037 static const char *rbd_segment_name(struct rbd_device
*rbd_dev
, u64 offset
)
1044 name
= kmem_cache_alloc(rbd_segment_name_cache
, GFP_NOIO
);
1047 segment
= offset
>> rbd_dev
->header
.obj_order
;
1048 name_format
= "%s.%012llx";
1049 if (rbd_dev
->image_format
== 2)
1050 name_format
= "%s.%016llx";
1051 ret
= snprintf(name
, MAX_OBJ_NAME_SIZE
+ 1, name_format
,
1052 rbd_dev
->header
.object_prefix
, segment
);
1053 if (ret
< 0 || ret
> MAX_OBJ_NAME_SIZE
) {
1054 pr_err("error formatting segment name for #%llu (%d)\n",
1063 static void rbd_segment_name_free(const char *name
)
1065 /* The explicit cast here is needed to drop the const qualifier */
1067 kmem_cache_free(rbd_segment_name_cache
, (void *)name
);
1070 static u64
rbd_segment_offset(struct rbd_device
*rbd_dev
, u64 offset
)
1072 u64 segment_size
= (u64
) 1 << rbd_dev
->header
.obj_order
;
1074 return offset
& (segment_size
- 1);
1077 static u64
rbd_segment_length(struct rbd_device
*rbd_dev
,
1078 u64 offset
, u64 length
)
1080 u64 segment_size
= (u64
) 1 << rbd_dev
->header
.obj_order
;
1082 offset
&= segment_size
- 1;
1084 rbd_assert(length
<= U64_MAX
- offset
);
1085 if (offset
+ length
> segment_size
)
1086 length
= segment_size
- offset
;
1092 * returns the size of an object in the image
1094 static u64
rbd_obj_bytes(struct rbd_image_header
*header
)
1096 return 1 << header
->obj_order
;
1103 static void bio_chain_put(struct bio
*chain
)
1109 chain
= chain
->bi_next
;
1115 * zeros a bio chain, starting at specific offset
1117 static void zero_bio_chain(struct bio
*chain
, int start_ofs
)
1120 unsigned long flags
;
1126 bio_for_each_segment(bv
, chain
, i
) {
1127 if (pos
+ bv
->bv_len
> start_ofs
) {
1128 int remainder
= max(start_ofs
- pos
, 0);
1129 buf
= bvec_kmap_irq(bv
, &flags
);
1130 memset(buf
+ remainder
, 0,
1131 bv
->bv_len
- remainder
);
1132 flush_dcache_page(bv
->bv_page
);
1133 bvec_kunmap_irq(buf
, &flags
);
1138 chain
= chain
->bi_next
;
1143 * similar to zero_bio_chain(), zeros data defined by a page array,
1144 * starting at the given byte offset from the start of the array and
1145 * continuing up to the given end offset. The pages array is
1146 * assumed to be big enough to hold all bytes up to the end.
1148 static void zero_pages(struct page
**pages
, u64 offset
, u64 end
)
1150 struct page
**page
= &pages
[offset
>> PAGE_SHIFT
];
1152 rbd_assert(end
> offset
);
1153 rbd_assert(end
- offset
<= (u64
)SIZE_MAX
);
1154 while (offset
< end
) {
1157 unsigned long flags
;
1160 page_offset
= (size_t)(offset
& ~PAGE_MASK
);
1161 length
= min(PAGE_SIZE
- page_offset
, (size_t)(end
- offset
));
1162 local_irq_save(flags
);
1163 kaddr
= kmap_atomic(*page
);
1164 memset(kaddr
+ page_offset
, 0, length
);
1165 flush_dcache_page(*page
);
1166 kunmap_atomic(kaddr
);
1167 local_irq_restore(flags
);
1175 * Clone a portion of a bio, starting at the given byte offset
1176 * and continuing for the number of bytes indicated.
1178 static struct bio
*bio_clone_range(struct bio
*bio_src
,
1179 unsigned int offset
,
1187 unsigned short end_idx
;
1188 unsigned short vcnt
;
1191 /* Handle the easy case for the caller */
1193 if (!offset
&& len
== bio_src
->bi_size
)
1194 return bio_clone(bio_src
, gfpmask
);
1196 if (WARN_ON_ONCE(!len
))
1198 if (WARN_ON_ONCE(len
> bio_src
->bi_size
))
1200 if (WARN_ON_ONCE(offset
> bio_src
->bi_size
- len
))
1203 /* Find first affected segment... */
1206 bio_for_each_segment(bv
, bio_src
, idx
) {
1207 if (resid
< bv
->bv_len
)
1209 resid
-= bv
->bv_len
;
1213 /* ...and the last affected segment */
1216 __bio_for_each_segment(bv
, bio_src
, end_idx
, idx
) {
1217 if (resid
<= bv
->bv_len
)
1219 resid
-= bv
->bv_len
;
1221 vcnt
= end_idx
- idx
+ 1;
1223 /* Build the clone */
1225 bio
= bio_alloc(gfpmask
, (unsigned int) vcnt
);
1227 return NULL
; /* ENOMEM */
1229 bio
->bi_bdev
= bio_src
->bi_bdev
;
1230 bio
->bi_sector
= bio_src
->bi_sector
+ (offset
>> SECTOR_SHIFT
);
1231 bio
->bi_rw
= bio_src
->bi_rw
;
1232 bio
->bi_flags
|= 1 << BIO_CLONED
;
1235 * Copy over our part of the bio_vec, then update the first
1236 * and last (or only) entries.
1238 memcpy(&bio
->bi_io_vec
[0], &bio_src
->bi_io_vec
[idx
],
1239 vcnt
* sizeof (struct bio_vec
));
1240 bio
->bi_io_vec
[0].bv_offset
+= voff
;
1242 bio
->bi_io_vec
[0].bv_len
-= voff
;
1243 bio
->bi_io_vec
[vcnt
- 1].bv_len
= resid
;
1245 bio
->bi_io_vec
[0].bv_len
= len
;
1248 bio
->bi_vcnt
= vcnt
;
1256 * Clone a portion of a bio chain, starting at the given byte offset
1257 * into the first bio in the source chain and continuing for the
1258 * number of bytes indicated. The result is another bio chain of
1259 * exactly the given length, or a null pointer on error.
1261 * The bio_src and offset parameters are both in-out. On entry they
1262 * refer to the first source bio and the offset into that bio where
1263 * the start of data to be cloned is located.
1265 * On return, bio_src is updated to refer to the bio in the source
1266 * chain that contains first un-cloned byte, and *offset will
1267 * contain the offset of that byte within that bio.
1269 static struct bio
*bio_chain_clone_range(struct bio
**bio_src
,
1270 unsigned int *offset
,
1274 struct bio
*bi
= *bio_src
;
1275 unsigned int off
= *offset
;
1276 struct bio
*chain
= NULL
;
1279 /* Build up a chain of clone bios up to the limit */
1281 if (!bi
|| off
>= bi
->bi_size
|| !len
)
1282 return NULL
; /* Nothing to clone */
1286 unsigned int bi_size
;
1290 rbd_warn(NULL
, "bio_chain exhausted with %u left", len
);
1291 goto out_err
; /* EINVAL; ran out of bio's */
1293 bi_size
= min_t(unsigned int, bi
->bi_size
- off
, len
);
1294 bio
= bio_clone_range(bi
, off
, bi_size
, gfpmask
);
1296 goto out_err
; /* ENOMEM */
1299 end
= &bio
->bi_next
;
1302 if (off
== bi
->bi_size
) {
1313 bio_chain_put(chain
);
1319 * The default/initial value for all object request flags is 0. For
1320 * each flag, once its value is set to 1 it is never reset to 0
1323 static void obj_request_img_data_set(struct rbd_obj_request
*obj_request
)
1325 if (test_and_set_bit(OBJ_REQ_IMG_DATA
, &obj_request
->flags
)) {
1326 struct rbd_device
*rbd_dev
;
1328 rbd_dev
= obj_request
->img_request
->rbd_dev
;
1329 rbd_warn(rbd_dev
, "obj_request %p already marked img_data\n",
1334 static bool obj_request_img_data_test(struct rbd_obj_request
*obj_request
)
1337 return test_bit(OBJ_REQ_IMG_DATA
, &obj_request
->flags
) != 0;
1340 static void obj_request_done_set(struct rbd_obj_request
*obj_request
)
1342 if (test_and_set_bit(OBJ_REQ_DONE
, &obj_request
->flags
)) {
1343 struct rbd_device
*rbd_dev
= NULL
;
1345 if (obj_request_img_data_test(obj_request
))
1346 rbd_dev
= obj_request
->img_request
->rbd_dev
;
1347 rbd_warn(rbd_dev
, "obj_request %p already marked done\n",
1352 static bool obj_request_done_test(struct rbd_obj_request
*obj_request
)
1355 return test_bit(OBJ_REQ_DONE
, &obj_request
->flags
) != 0;
1359 * This sets the KNOWN flag after (possibly) setting the EXISTS
1360 * flag. The latter is set based on the "exists" value provided.
1362 * Note that for our purposes once an object exists it never goes
1363 * away again. It's possible that the response from two existence
1364 * checks are separated by the creation of the target object, and
1365 * the first ("doesn't exist") response arrives *after* the second
1366 * ("does exist"). In that case we ignore the second one.
1368 static void obj_request_existence_set(struct rbd_obj_request
*obj_request
,
1372 set_bit(OBJ_REQ_EXISTS
, &obj_request
->flags
);
1373 set_bit(OBJ_REQ_KNOWN
, &obj_request
->flags
);
1377 static bool obj_request_known_test(struct rbd_obj_request
*obj_request
)
1380 return test_bit(OBJ_REQ_KNOWN
, &obj_request
->flags
) != 0;
1383 static bool obj_request_exists_test(struct rbd_obj_request
*obj_request
)
1386 return test_bit(OBJ_REQ_EXISTS
, &obj_request
->flags
) != 0;
1389 static bool obj_request_overlaps_parent(struct rbd_obj_request
*obj_request
)
1391 struct rbd_device
*rbd_dev
= obj_request
->img_request
->rbd_dev
;
1393 return obj_request
->img_offset
<
1394 round_up(rbd_dev
->parent_overlap
, rbd_obj_bytes(&rbd_dev
->header
));
1397 static void rbd_obj_request_get(struct rbd_obj_request
*obj_request
)
1399 dout("%s: obj %p (was %d)\n", __func__
, obj_request
,
1400 atomic_read(&obj_request
->kref
.refcount
));
1401 kref_get(&obj_request
->kref
);
1404 static void rbd_obj_request_destroy(struct kref
*kref
);
1405 static void rbd_obj_request_put(struct rbd_obj_request
*obj_request
)
1407 rbd_assert(obj_request
!= NULL
);
1408 dout("%s: obj %p (was %d)\n", __func__
, obj_request
,
1409 atomic_read(&obj_request
->kref
.refcount
));
1410 kref_put(&obj_request
->kref
, rbd_obj_request_destroy
);
1413 static void rbd_img_request_get(struct rbd_img_request
*img_request
)
1415 dout("%s: img %p (was %d)\n", __func__
, img_request
,
1416 atomic_read(&img_request
->kref
.refcount
));
1417 kref_get(&img_request
->kref
);
1420 static bool img_request_child_test(struct rbd_img_request
*img_request
);
1421 static void rbd_parent_request_destroy(struct kref
*kref
);
1422 static void rbd_img_request_destroy(struct kref
*kref
);
1423 static void rbd_img_request_put(struct rbd_img_request
*img_request
)
1425 rbd_assert(img_request
!= NULL
);
1426 dout("%s: img %p (was %d)\n", __func__
, img_request
,
1427 atomic_read(&img_request
->kref
.refcount
));
1428 if (img_request_child_test(img_request
))
1429 kref_put(&img_request
->kref
, rbd_parent_request_destroy
);
1431 kref_put(&img_request
->kref
, rbd_img_request_destroy
);
1434 static inline void rbd_img_obj_request_add(struct rbd_img_request
*img_request
,
1435 struct rbd_obj_request
*obj_request
)
1437 rbd_assert(obj_request
->img_request
== NULL
);
1439 /* Image request now owns object's original reference */
1440 obj_request
->img_request
= img_request
;
1441 obj_request
->which
= img_request
->obj_request_count
;
1442 rbd_assert(!obj_request_img_data_test(obj_request
));
1443 obj_request_img_data_set(obj_request
);
1444 rbd_assert(obj_request
->which
!= BAD_WHICH
);
1445 img_request
->obj_request_count
++;
1446 list_add_tail(&obj_request
->links
, &img_request
->obj_requests
);
1447 dout("%s: img %p obj %p w=%u\n", __func__
, img_request
, obj_request
,
1448 obj_request
->which
);
1451 static inline void rbd_img_obj_request_del(struct rbd_img_request
*img_request
,
1452 struct rbd_obj_request
*obj_request
)
1454 rbd_assert(obj_request
->which
!= BAD_WHICH
);
1456 dout("%s: img %p obj %p w=%u\n", __func__
, img_request
, obj_request
,
1457 obj_request
->which
);
1458 list_del(&obj_request
->links
);
1459 rbd_assert(img_request
->obj_request_count
> 0);
1460 img_request
->obj_request_count
--;
1461 rbd_assert(obj_request
->which
== img_request
->obj_request_count
);
1462 obj_request
->which
= BAD_WHICH
;
1463 rbd_assert(obj_request_img_data_test(obj_request
));
1464 rbd_assert(obj_request
->img_request
== img_request
);
1465 obj_request
->img_request
= NULL
;
1466 obj_request
->callback
= NULL
;
1467 rbd_obj_request_put(obj_request
);
1470 static bool obj_request_type_valid(enum obj_request_type type
)
1473 case OBJ_REQUEST_NODATA
:
1474 case OBJ_REQUEST_BIO
:
1475 case OBJ_REQUEST_PAGES
:
1482 static int rbd_obj_request_submit(struct ceph_osd_client
*osdc
,
1483 struct rbd_obj_request
*obj_request
)
1485 dout("%s: osdc %p obj %p\n", __func__
, osdc
, obj_request
);
1487 return ceph_osdc_start_request(osdc
, obj_request
->osd_req
, false);
1490 static void rbd_img_request_complete(struct rbd_img_request
*img_request
)
1493 dout("%s: img %p\n", __func__
, img_request
);
1496 * If no error occurred, compute the aggregate transfer
1497 * count for the image request. We could instead use
1498 * atomic64_cmpxchg() to update it as each object request
1499 * completes; not clear which way is better off hand.
1501 if (!img_request
->result
) {
1502 struct rbd_obj_request
*obj_request
;
1505 for_each_obj_request(img_request
, obj_request
)
1506 xferred
+= obj_request
->xferred
;
1507 img_request
->xferred
= xferred
;
1510 if (img_request
->callback
)
1511 img_request
->callback(img_request
);
1513 rbd_img_request_put(img_request
);
1516 /* Caller is responsible for rbd_obj_request_destroy(obj_request) */
1518 static int rbd_obj_request_wait(struct rbd_obj_request
*obj_request
)
1520 dout("%s: obj %p\n", __func__
, obj_request
);
1522 return wait_for_completion_interruptible(&obj_request
->completion
);
1526 * The default/initial value for all image request flags is 0. Each
1527 * is conditionally set to 1 at image request initialization time
1528 * and currently never change thereafter.
1530 static void img_request_write_set(struct rbd_img_request
*img_request
)
1532 set_bit(IMG_REQ_WRITE
, &img_request
->flags
);
1536 static bool img_request_write_test(struct rbd_img_request
*img_request
)
1539 return test_bit(IMG_REQ_WRITE
, &img_request
->flags
) != 0;
1542 static void img_request_child_set(struct rbd_img_request
*img_request
)
1544 set_bit(IMG_REQ_CHILD
, &img_request
->flags
);
1548 static void img_request_child_clear(struct rbd_img_request
*img_request
)
1550 clear_bit(IMG_REQ_CHILD
, &img_request
->flags
);
1554 static bool img_request_child_test(struct rbd_img_request
*img_request
)
1557 return test_bit(IMG_REQ_CHILD
, &img_request
->flags
) != 0;
1560 static void img_request_layered_set(struct rbd_img_request
*img_request
)
1562 set_bit(IMG_REQ_LAYERED
, &img_request
->flags
);
1566 static void img_request_layered_clear(struct rbd_img_request
*img_request
)
1568 clear_bit(IMG_REQ_LAYERED
, &img_request
->flags
);
1572 static bool img_request_layered_test(struct rbd_img_request
*img_request
)
1575 return test_bit(IMG_REQ_LAYERED
, &img_request
->flags
) != 0;
1579 rbd_img_obj_request_read_callback(struct rbd_obj_request
*obj_request
)
1581 u64 xferred
= obj_request
->xferred
;
1582 u64 length
= obj_request
->length
;
1584 dout("%s: obj %p img %p result %d %llu/%llu\n", __func__
,
1585 obj_request
, obj_request
->img_request
, obj_request
->result
,
1588 * ENOENT means a hole in the image. We zero-fill the entire
1589 * length of the request. A short read also implies zero-fill
1590 * to the end of the request. An error requires the whole
1591 * length of the request to be reported finished with an error
1592 * to the block layer. In each case we update the xferred
1593 * count to indicate the whole request was satisfied.
1595 rbd_assert(obj_request
->type
!= OBJ_REQUEST_NODATA
);
1596 if (obj_request
->result
== -ENOENT
) {
1597 if (obj_request
->type
== OBJ_REQUEST_BIO
)
1598 zero_bio_chain(obj_request
->bio_list
, 0);
1600 zero_pages(obj_request
->pages
, 0, length
);
1601 obj_request
->result
= 0;
1602 } else if (xferred
< length
&& !obj_request
->result
) {
1603 if (obj_request
->type
== OBJ_REQUEST_BIO
)
1604 zero_bio_chain(obj_request
->bio_list
, xferred
);
1606 zero_pages(obj_request
->pages
, xferred
, length
);
1608 obj_request
->xferred
= length
;
1609 obj_request_done_set(obj_request
);
1612 static void rbd_obj_request_complete(struct rbd_obj_request
*obj_request
)
1614 dout("%s: obj %p cb %p\n", __func__
, obj_request
,
1615 obj_request
->callback
);
1616 if (obj_request
->callback
)
1617 obj_request
->callback(obj_request
);
1619 complete_all(&obj_request
->completion
);
1622 static void rbd_osd_trivial_callback(struct rbd_obj_request
*obj_request
)
1624 dout("%s: obj %p\n", __func__
, obj_request
);
1625 obj_request_done_set(obj_request
);
1628 static void rbd_osd_read_callback(struct rbd_obj_request
*obj_request
)
1630 struct rbd_img_request
*img_request
= NULL
;
1631 struct rbd_device
*rbd_dev
= NULL
;
1632 bool layered
= false;
1634 if (obj_request_img_data_test(obj_request
)) {
1635 img_request
= obj_request
->img_request
;
1636 layered
= img_request
&& img_request_layered_test(img_request
);
1637 rbd_dev
= img_request
->rbd_dev
;
1640 dout("%s: obj %p img %p result %d %llu/%llu\n", __func__
,
1641 obj_request
, img_request
, obj_request
->result
,
1642 obj_request
->xferred
, obj_request
->length
);
1643 if (layered
&& obj_request
->result
== -ENOENT
&&
1644 obj_request
->img_offset
< rbd_dev
->parent_overlap
)
1645 rbd_img_parent_read(obj_request
);
1646 else if (img_request
)
1647 rbd_img_obj_request_read_callback(obj_request
);
1649 obj_request_done_set(obj_request
);
1652 static void rbd_osd_write_callback(struct rbd_obj_request
*obj_request
)
1654 dout("%s: obj %p result %d %llu\n", __func__
, obj_request
,
1655 obj_request
->result
, obj_request
->length
);
1657 * There is no such thing as a successful short write. Set
1658 * it to our originally-requested length.
1660 obj_request
->xferred
= obj_request
->length
;
1661 obj_request_done_set(obj_request
);
1665 * For a simple stat call there's nothing to do. We'll do more if
1666 * this is part of a write sequence for a layered image.
1668 static void rbd_osd_stat_callback(struct rbd_obj_request
*obj_request
)
1670 dout("%s: obj %p\n", __func__
, obj_request
);
1671 obj_request_done_set(obj_request
);
1674 static void rbd_osd_call_callback(struct rbd_obj_request
*obj_request
)
1676 dout("%s: obj %p\n", __func__
, obj_request
);
1678 if (obj_request_img_data_test(obj_request
))
1679 rbd_osd_copyup_callback(obj_request
);
1681 obj_request_done_set(obj_request
);
1684 static void rbd_osd_req_callback(struct ceph_osd_request
*osd_req
,
1685 struct ceph_msg
*msg
)
1687 struct rbd_obj_request
*obj_request
= osd_req
->r_priv
;
1690 dout("%s: osd_req %p msg %p\n", __func__
, osd_req
, msg
);
1691 rbd_assert(osd_req
== obj_request
->osd_req
);
1692 if (obj_request_img_data_test(obj_request
)) {
1693 rbd_assert(obj_request
->img_request
);
1694 rbd_assert(obj_request
->which
!= BAD_WHICH
);
1696 rbd_assert(obj_request
->which
== BAD_WHICH
);
1699 if (osd_req
->r_result
< 0)
1700 obj_request
->result
= osd_req
->r_result
;
1702 BUG_ON(osd_req
->r_num_ops
> 2);
1705 * We support a 64-bit length, but ultimately it has to be
1706 * passed to blk_end_request(), which takes an unsigned int.
1708 obj_request
->xferred
= osd_req
->r_reply_op_len
[0];
1709 rbd_assert(obj_request
->xferred
< (u64
)UINT_MAX
);
1710 opcode
= osd_req
->r_ops
[0].op
;
1712 case CEPH_OSD_OP_READ
:
1713 rbd_osd_read_callback(obj_request
);
1715 case CEPH_OSD_OP_WRITE
:
1716 rbd_osd_write_callback(obj_request
);
1718 case CEPH_OSD_OP_STAT
:
1719 rbd_osd_stat_callback(obj_request
);
1721 case CEPH_OSD_OP_CALL
:
1722 rbd_osd_call_callback(obj_request
);
1724 case CEPH_OSD_OP_NOTIFY_ACK
:
1725 case CEPH_OSD_OP_WATCH
:
1726 rbd_osd_trivial_callback(obj_request
);
1729 rbd_warn(NULL
, "%s: unsupported op %hu\n",
1730 obj_request
->object_name
, (unsigned short) opcode
);
1734 if (obj_request_done_test(obj_request
))
1735 rbd_obj_request_complete(obj_request
);
1738 static void rbd_osd_req_format_read(struct rbd_obj_request
*obj_request
)
1740 struct rbd_img_request
*img_request
= obj_request
->img_request
;
1741 struct ceph_osd_request
*osd_req
= obj_request
->osd_req
;
1744 rbd_assert(osd_req
!= NULL
);
1746 snap_id
= img_request
? img_request
->snap_id
: CEPH_NOSNAP
;
1747 ceph_osdc_build_request(osd_req
, obj_request
->offset
,
1748 NULL
, snap_id
, NULL
);
1751 static void rbd_osd_req_format_write(struct rbd_obj_request
*obj_request
)
1753 struct rbd_img_request
*img_request
= obj_request
->img_request
;
1754 struct ceph_osd_request
*osd_req
= obj_request
->osd_req
;
1755 struct ceph_snap_context
*snapc
;
1756 struct timespec mtime
= CURRENT_TIME
;
1758 rbd_assert(osd_req
!= NULL
);
1760 snapc
= img_request
? img_request
->snapc
: NULL
;
1761 ceph_osdc_build_request(osd_req
, obj_request
->offset
,
1762 snapc
, CEPH_NOSNAP
, &mtime
);
1765 static struct ceph_osd_request
*rbd_osd_req_create(
1766 struct rbd_device
*rbd_dev
,
1768 struct rbd_obj_request
*obj_request
)
1770 struct ceph_snap_context
*snapc
= NULL
;
1771 struct ceph_osd_client
*osdc
;
1772 struct ceph_osd_request
*osd_req
;
1774 if (obj_request_img_data_test(obj_request
)) {
1775 struct rbd_img_request
*img_request
= obj_request
->img_request
;
1777 rbd_assert(write_request
==
1778 img_request_write_test(img_request
));
1780 snapc
= img_request
->snapc
;
1783 /* Allocate and initialize the request, for the single op */
1785 osdc
= &rbd_dev
->rbd_client
->client
->osdc
;
1786 osd_req
= ceph_osdc_alloc_request(osdc
, snapc
, 1, false, GFP_ATOMIC
);
1788 return NULL
; /* ENOMEM */
1791 osd_req
->r_flags
= CEPH_OSD_FLAG_WRITE
| CEPH_OSD_FLAG_ONDISK
;
1793 osd_req
->r_flags
= CEPH_OSD_FLAG_READ
;
1795 osd_req
->r_callback
= rbd_osd_req_callback
;
1796 osd_req
->r_priv
= obj_request
;
1798 osd_req
->r_oid_len
= strlen(obj_request
->object_name
);
1799 rbd_assert(osd_req
->r_oid_len
< sizeof (osd_req
->r_oid
));
1800 memcpy(osd_req
->r_oid
, obj_request
->object_name
, osd_req
->r_oid_len
);
1802 osd_req
->r_file_layout
= rbd_dev
->layout
; /* struct */
1808 * Create a copyup osd request based on the information in the
1809 * object request supplied. A copyup request has two osd ops,
1810 * a copyup method call, and a "normal" write request.
1812 static struct ceph_osd_request
*
1813 rbd_osd_req_create_copyup(struct rbd_obj_request
*obj_request
)
1815 struct rbd_img_request
*img_request
;
1816 struct ceph_snap_context
*snapc
;
1817 struct rbd_device
*rbd_dev
;
1818 struct ceph_osd_client
*osdc
;
1819 struct ceph_osd_request
*osd_req
;
1821 rbd_assert(obj_request_img_data_test(obj_request
));
1822 img_request
= obj_request
->img_request
;
1823 rbd_assert(img_request
);
1824 rbd_assert(img_request_write_test(img_request
));
1826 /* Allocate and initialize the request, for the two ops */
1828 snapc
= img_request
->snapc
;
1829 rbd_dev
= img_request
->rbd_dev
;
1830 osdc
= &rbd_dev
->rbd_client
->client
->osdc
;
1831 osd_req
= ceph_osdc_alloc_request(osdc
, snapc
, 2, false, GFP_ATOMIC
);
1833 return NULL
; /* ENOMEM */
1835 osd_req
->r_flags
= CEPH_OSD_FLAG_WRITE
| CEPH_OSD_FLAG_ONDISK
;
1836 osd_req
->r_callback
= rbd_osd_req_callback
;
1837 osd_req
->r_priv
= obj_request
;
1839 osd_req
->r_oid_len
= strlen(obj_request
->object_name
);
1840 rbd_assert(osd_req
->r_oid_len
< sizeof (osd_req
->r_oid
));
1841 memcpy(osd_req
->r_oid
, obj_request
->object_name
, osd_req
->r_oid_len
);
1843 osd_req
->r_file_layout
= rbd_dev
->layout
; /* struct */
1849 static void rbd_osd_req_destroy(struct ceph_osd_request
*osd_req
)
1851 ceph_osdc_put_request(osd_req
);
1854 /* object_name is assumed to be a non-null pointer and NUL-terminated */
1856 static struct rbd_obj_request
*rbd_obj_request_create(const char *object_name
,
1857 u64 offset
, u64 length
,
1858 enum obj_request_type type
)
1860 struct rbd_obj_request
*obj_request
;
1864 rbd_assert(obj_request_type_valid(type
));
1866 size
= strlen(object_name
) + 1;
1867 name
= kmalloc(size
, GFP_NOIO
);
1871 obj_request
= kmem_cache_zalloc(rbd_obj_request_cache
, GFP_NOIO
);
1877 obj_request
->object_name
= memcpy(name
, object_name
, size
);
1878 obj_request
->offset
= offset
;
1879 obj_request
->length
= length
;
1880 obj_request
->flags
= 0;
1881 obj_request
->which
= BAD_WHICH
;
1882 obj_request
->type
= type
;
1883 INIT_LIST_HEAD(&obj_request
->links
);
1884 init_completion(&obj_request
->completion
);
1885 kref_init(&obj_request
->kref
);
1887 dout("%s: \"%s\" %llu/%llu %d -> obj %p\n", __func__
, object_name
,
1888 offset
, length
, (int)type
, obj_request
);
1893 static void rbd_obj_request_destroy(struct kref
*kref
)
1895 struct rbd_obj_request
*obj_request
;
1897 obj_request
= container_of(kref
, struct rbd_obj_request
, kref
);
1899 dout("%s: obj %p\n", __func__
, obj_request
);
1901 rbd_assert(obj_request
->img_request
== NULL
);
1902 rbd_assert(obj_request
->which
== BAD_WHICH
);
1904 if (obj_request
->osd_req
)
1905 rbd_osd_req_destroy(obj_request
->osd_req
);
1907 rbd_assert(obj_request_type_valid(obj_request
->type
));
1908 switch (obj_request
->type
) {
1909 case OBJ_REQUEST_NODATA
:
1910 break; /* Nothing to do */
1911 case OBJ_REQUEST_BIO
:
1912 if (obj_request
->bio_list
)
1913 bio_chain_put(obj_request
->bio_list
);
1915 case OBJ_REQUEST_PAGES
:
1916 if (obj_request
->pages
)
1917 ceph_release_page_vector(obj_request
->pages
,
1918 obj_request
->page_count
);
1922 kfree(obj_request
->object_name
);
1923 obj_request
->object_name
= NULL
;
1924 kmem_cache_free(rbd_obj_request_cache
, obj_request
);
1927 /* It's OK to call this for a device with no parent */
1929 static void rbd_spec_put(struct rbd_spec
*spec
);
1930 static void rbd_dev_unparent(struct rbd_device
*rbd_dev
)
1932 rbd_dev_remove_parent(rbd_dev
);
1933 rbd_spec_put(rbd_dev
->parent_spec
);
1934 rbd_dev
->parent_spec
= NULL
;
1935 rbd_dev
->parent_overlap
= 0;
1939 * Parent image reference counting is used to determine when an
1940 * image's parent fields can be safely torn down--after there are no
1941 * more in-flight requests to the parent image. When the last
1942 * reference is dropped, cleaning them up is safe.
1944 static void rbd_dev_parent_put(struct rbd_device
*rbd_dev
)
1948 if (!rbd_dev
->parent_spec
)
1951 counter
= atomic_dec_return_safe(&rbd_dev
->parent_ref
);
1955 /* Last reference; clean up parent data structures */
1958 rbd_dev_unparent(rbd_dev
);
1960 rbd_warn(rbd_dev
, "parent reference underflow\n");
1964 * If an image has a non-zero parent overlap, get a reference to its
1967 * We must get the reference before checking for the overlap to
1968 * coordinate properly with zeroing the parent overlap in
1969 * rbd_dev_v2_parent_info() when an image gets flattened. We
1970 * drop it again if there is no overlap.
1972 * Returns true if the rbd device has a parent with a non-zero
1973 * overlap and a reference for it was successfully taken, or
1976 static bool rbd_dev_parent_get(struct rbd_device
*rbd_dev
)
1980 if (!rbd_dev
->parent_spec
)
1983 counter
= atomic_inc_return_safe(&rbd_dev
->parent_ref
);
1984 if (counter
> 0 && rbd_dev
->parent_overlap
)
1987 /* Image was flattened, but parent is not yet torn down */
1990 rbd_warn(rbd_dev
, "parent reference overflow\n");
1996 * Caller is responsible for filling in the list of object requests
1997 * that comprises the image request, and the Linux request pointer
1998 * (if there is one).
2000 static struct rbd_img_request
*rbd_img_request_create(
2001 struct rbd_device
*rbd_dev
,
2002 u64 offset
, u64 length
,
2005 struct rbd_img_request
*img_request
;
2007 img_request
= kmem_cache_alloc(rbd_img_request_cache
, GFP_ATOMIC
);
2011 if (write_request
) {
2012 down_read(&rbd_dev
->header_rwsem
);
2013 ceph_get_snap_context(rbd_dev
->header
.snapc
);
2014 up_read(&rbd_dev
->header_rwsem
);
2017 img_request
->rq
= NULL
;
2018 img_request
->rbd_dev
= rbd_dev
;
2019 img_request
->offset
= offset
;
2020 img_request
->length
= length
;
2021 img_request
->flags
= 0;
2022 if (write_request
) {
2023 img_request_write_set(img_request
);
2024 img_request
->snapc
= rbd_dev
->header
.snapc
;
2026 img_request
->snap_id
= rbd_dev
->spec
->snap_id
;
2028 if (rbd_dev_parent_get(rbd_dev
))
2029 img_request_layered_set(img_request
);
2030 spin_lock_init(&img_request
->completion_lock
);
2031 img_request
->next_completion
= 0;
2032 img_request
->callback
= NULL
;
2033 img_request
->result
= 0;
2034 img_request
->obj_request_count
= 0;
2035 INIT_LIST_HEAD(&img_request
->obj_requests
);
2036 kref_init(&img_request
->kref
);
2038 dout("%s: rbd_dev %p %s %llu/%llu -> img %p\n", __func__
, rbd_dev
,
2039 write_request
? "write" : "read", offset
, length
,
2045 static void rbd_img_request_destroy(struct kref
*kref
)
2047 struct rbd_img_request
*img_request
;
2048 struct rbd_obj_request
*obj_request
;
2049 struct rbd_obj_request
*next_obj_request
;
2051 img_request
= container_of(kref
, struct rbd_img_request
, kref
);
2053 dout("%s: img %p\n", __func__
, img_request
);
2055 for_each_obj_request_safe(img_request
, obj_request
, next_obj_request
)
2056 rbd_img_obj_request_del(img_request
, obj_request
);
2057 rbd_assert(img_request
->obj_request_count
== 0);
2059 if (img_request_layered_test(img_request
)) {
2060 img_request_layered_clear(img_request
);
2061 rbd_dev_parent_put(img_request
->rbd_dev
);
2064 if (img_request_write_test(img_request
))
2065 ceph_put_snap_context(img_request
->snapc
);
2067 kmem_cache_free(rbd_img_request_cache
, img_request
);
2070 static struct rbd_img_request
*rbd_parent_request_create(
2071 struct rbd_obj_request
*obj_request
,
2072 u64 img_offset
, u64 length
)
2074 struct rbd_img_request
*parent_request
;
2075 struct rbd_device
*rbd_dev
;
2077 rbd_assert(obj_request
->img_request
);
2078 rbd_dev
= obj_request
->img_request
->rbd_dev
;
2080 parent_request
= rbd_img_request_create(rbd_dev
->parent
,
2081 img_offset
, length
, false);
2082 if (!parent_request
)
2085 img_request_child_set(parent_request
);
2086 rbd_obj_request_get(obj_request
);
2087 parent_request
->obj_request
= obj_request
;
2089 return parent_request
;
2092 static void rbd_parent_request_destroy(struct kref
*kref
)
2094 struct rbd_img_request
*parent_request
;
2095 struct rbd_obj_request
*orig_request
;
2097 parent_request
= container_of(kref
, struct rbd_img_request
, kref
);
2098 orig_request
= parent_request
->obj_request
;
2100 parent_request
->obj_request
= NULL
;
2101 rbd_obj_request_put(orig_request
);
2102 img_request_child_clear(parent_request
);
2104 rbd_img_request_destroy(kref
);
2107 static bool rbd_img_obj_end_request(struct rbd_obj_request
*obj_request
)
2109 struct rbd_img_request
*img_request
;
2110 unsigned int xferred
;
2114 rbd_assert(obj_request_img_data_test(obj_request
));
2115 img_request
= obj_request
->img_request
;
2117 rbd_assert(obj_request
->xferred
<= (u64
)UINT_MAX
);
2118 xferred
= (unsigned int)obj_request
->xferred
;
2119 result
= obj_request
->result
;
2121 struct rbd_device
*rbd_dev
= img_request
->rbd_dev
;
2123 rbd_warn(rbd_dev
, "%s %llx at %llx (%llx)\n",
2124 img_request_write_test(img_request
) ? "write" : "read",
2125 obj_request
->length
, obj_request
->img_offset
,
2126 obj_request
->offset
);
2127 rbd_warn(rbd_dev
, " result %d xferred %x\n",
2129 if (!img_request
->result
)
2130 img_request
->result
= result
;
2132 * Need to end I/O on the entire obj_request worth of
2133 * bytes in case of error.
2135 xferred
= obj_request
->length
;
2138 /* Image object requests don't own their page array */
2140 if (obj_request
->type
== OBJ_REQUEST_PAGES
) {
2141 obj_request
->pages
= NULL
;
2142 obj_request
->page_count
= 0;
2145 if (img_request_child_test(img_request
)) {
2146 rbd_assert(img_request
->obj_request
!= NULL
);
2147 more
= obj_request
->which
< img_request
->obj_request_count
- 1;
2149 rbd_assert(img_request
->rq
!= NULL
);
2150 more
= blk_end_request(img_request
->rq
, result
, xferred
);
2156 static void rbd_img_obj_callback(struct rbd_obj_request
*obj_request
)
2158 struct rbd_img_request
*img_request
;
2159 u32 which
= obj_request
->which
;
2162 rbd_assert(obj_request_img_data_test(obj_request
));
2163 img_request
= obj_request
->img_request
;
2165 dout("%s: img %p obj %p\n", __func__
, img_request
, obj_request
);
2166 rbd_assert(img_request
!= NULL
);
2167 rbd_assert(img_request
->obj_request_count
> 0);
2168 rbd_assert(which
!= BAD_WHICH
);
2169 rbd_assert(which
< img_request
->obj_request_count
);
2171 spin_lock_irq(&img_request
->completion_lock
);
2172 if (which
!= img_request
->next_completion
)
2175 for_each_obj_request_from(img_request
, obj_request
) {
2177 rbd_assert(which
< img_request
->obj_request_count
);
2179 if (!obj_request_done_test(obj_request
))
2181 more
= rbd_img_obj_end_request(obj_request
);
2185 rbd_assert(more
^ (which
== img_request
->obj_request_count
));
2186 img_request
->next_completion
= which
;
2188 spin_unlock_irq(&img_request
->completion_lock
);
2189 rbd_img_request_put(img_request
);
2192 rbd_img_request_complete(img_request
);
2196 * Split up an image request into one or more object requests, each
2197 * to a different object. The "type" parameter indicates whether
2198 * "data_desc" is the pointer to the head of a list of bio
2199 * structures, or the base of a page array. In either case this
2200 * function assumes data_desc describes memory sufficient to hold
2201 * all data described by the image request.
2203 static int rbd_img_request_fill(struct rbd_img_request
*img_request
,
2204 enum obj_request_type type
,
2207 struct rbd_device
*rbd_dev
= img_request
->rbd_dev
;
2208 struct rbd_obj_request
*obj_request
= NULL
;
2209 struct rbd_obj_request
*next_obj_request
;
2210 bool write_request
= img_request_write_test(img_request
);
2211 struct bio
*bio_list
= 0;
2212 unsigned int bio_offset
= 0;
2213 struct page
**pages
= 0;
2218 dout("%s: img %p type %d data_desc %p\n", __func__
, img_request
,
2219 (int)type
, data_desc
);
2221 opcode
= write_request
? CEPH_OSD_OP_WRITE
: CEPH_OSD_OP_READ
;
2222 img_offset
= img_request
->offset
;
2223 resid
= img_request
->length
;
2224 rbd_assert(resid
> 0);
2226 if (type
== OBJ_REQUEST_BIO
) {
2227 bio_list
= data_desc
;
2228 rbd_assert(img_offset
== bio_list
->bi_sector
<< SECTOR_SHIFT
);
2230 rbd_assert(type
== OBJ_REQUEST_PAGES
);
2235 struct ceph_osd_request
*osd_req
;
2236 const char *object_name
;
2240 object_name
= rbd_segment_name(rbd_dev
, img_offset
);
2243 offset
= rbd_segment_offset(rbd_dev
, img_offset
);
2244 length
= rbd_segment_length(rbd_dev
, img_offset
, resid
);
2245 obj_request
= rbd_obj_request_create(object_name
,
2246 offset
, length
, type
);
2247 /* object request has its own copy of the object name */
2248 rbd_segment_name_free(object_name
);
2252 * set obj_request->img_request before creating the
2253 * osd_request so that it gets the right snapc
2255 rbd_img_obj_request_add(img_request
, obj_request
);
2257 if (type
== OBJ_REQUEST_BIO
) {
2258 unsigned int clone_size
;
2260 rbd_assert(length
<= (u64
)UINT_MAX
);
2261 clone_size
= (unsigned int)length
;
2262 obj_request
->bio_list
=
2263 bio_chain_clone_range(&bio_list
,
2267 if (!obj_request
->bio_list
)
2270 unsigned int page_count
;
2272 obj_request
->pages
= pages
;
2273 page_count
= (u32
)calc_pages_for(offset
, length
);
2274 obj_request
->page_count
= page_count
;
2275 if ((offset
+ length
) & ~PAGE_MASK
)
2276 page_count
--; /* more on last page */
2277 pages
+= page_count
;
2280 osd_req
= rbd_osd_req_create(rbd_dev
, write_request
,
2284 obj_request
->osd_req
= osd_req
;
2285 obj_request
->callback
= rbd_img_obj_callback
;
2286 rbd_img_request_get(img_request
);
2288 osd_req_op_extent_init(osd_req
, 0, opcode
, offset
, length
,
2290 if (type
== OBJ_REQUEST_BIO
)
2291 osd_req_op_extent_osd_data_bio(osd_req
, 0,
2292 obj_request
->bio_list
, length
);
2294 osd_req_op_extent_osd_data_pages(osd_req
, 0,
2295 obj_request
->pages
, length
,
2296 offset
& ~PAGE_MASK
, false, false);
2299 rbd_osd_req_format_write(obj_request
);
2301 rbd_osd_req_format_read(obj_request
);
2303 obj_request
->img_offset
= img_offset
;
2305 img_offset
+= length
;
2312 rbd_obj_request_put(obj_request
);
2314 for_each_obj_request_safe(img_request
, obj_request
, next_obj_request
)
2315 rbd_img_obj_request_del(img_request
, obj_request
);
2321 rbd_osd_copyup_callback(struct rbd_obj_request
*obj_request
)
2323 struct rbd_img_request
*img_request
;
2324 struct rbd_device
*rbd_dev
;
2325 struct page
**pages
;
2328 dout("%s: obj %p\n", __func__
, obj_request
);
2330 rbd_assert(obj_request
->type
== OBJ_REQUEST_BIO
);
2331 rbd_assert(obj_request_img_data_test(obj_request
));
2332 img_request
= obj_request
->img_request
;
2333 rbd_assert(img_request
);
2335 rbd_dev
= img_request
->rbd_dev
;
2336 rbd_assert(rbd_dev
);
2338 pages
= obj_request
->copyup_pages
;
2339 rbd_assert(pages
!= NULL
);
2340 obj_request
->copyup_pages
= NULL
;
2341 page_count
= obj_request
->copyup_page_count
;
2342 rbd_assert(page_count
);
2343 obj_request
->copyup_page_count
= 0;
2344 ceph_release_page_vector(pages
, page_count
);
2347 * We want the transfer count to reflect the size of the
2348 * original write request. There is no such thing as a
2349 * successful short write, so if the request was successful
2350 * we can just set it to the originally-requested length.
2352 if (!obj_request
->result
)
2353 obj_request
->xferred
= obj_request
->length
;
2355 obj_request_done_set(obj_request
);
2359 rbd_img_obj_parent_read_full_callback(struct rbd_img_request
*img_request
)
2361 struct rbd_obj_request
*orig_request
;
2362 struct ceph_osd_request
*osd_req
;
2363 struct ceph_osd_client
*osdc
;
2364 struct rbd_device
*rbd_dev
;
2365 struct page
**pages
;
2372 rbd_assert(img_request_child_test(img_request
));
2374 /* First get what we need from the image request */
2376 pages
= img_request
->copyup_pages
;
2377 rbd_assert(pages
!= NULL
);
2378 img_request
->copyup_pages
= NULL
;
2379 page_count
= img_request
->copyup_page_count
;
2380 rbd_assert(page_count
);
2381 img_request
->copyup_page_count
= 0;
2383 orig_request
= img_request
->obj_request
;
2384 rbd_assert(orig_request
!= NULL
);
2385 rbd_assert(obj_request_type_valid(orig_request
->type
));
2386 img_result
= img_request
->result
;
2387 parent_length
= img_request
->length
;
2388 rbd_assert(parent_length
== img_request
->xferred
);
2389 rbd_img_request_put(img_request
);
2391 rbd_assert(orig_request
->img_request
);
2392 rbd_dev
= orig_request
->img_request
->rbd_dev
;
2393 rbd_assert(rbd_dev
);
2396 * If the overlap has become 0 (most likely because the
2397 * image has been flattened) we need to free the pages
2398 * and re-submit the original write request.
2400 if (!rbd_dev
->parent_overlap
) {
2401 struct ceph_osd_client
*osdc
;
2403 ceph_release_page_vector(pages
, page_count
);
2404 osdc
= &rbd_dev
->rbd_client
->client
->osdc
;
2405 img_result
= rbd_obj_request_submit(osdc
, orig_request
);
2414 * The original osd request is of no use to use any more.
2415 * We need a new one that can hold the two ops in a copyup
2416 * request. Allocate the new copyup osd request for the
2417 * original request, and release the old one.
2419 img_result
= -ENOMEM
;
2420 osd_req
= rbd_osd_req_create_copyup(orig_request
);
2423 rbd_osd_req_destroy(orig_request
->osd_req
);
2424 orig_request
->osd_req
= osd_req
;
2425 orig_request
->copyup_pages
= pages
;
2426 orig_request
->copyup_page_count
= page_count
;
2428 /* Initialize the copyup op */
2430 osd_req_op_cls_init(osd_req
, 0, CEPH_OSD_OP_CALL
, "rbd", "copyup");
2431 osd_req_op_cls_request_data_pages(osd_req
, 0, pages
, parent_length
, 0,
2434 /* Then the original write request op */
2436 offset
= orig_request
->offset
;
2437 length
= orig_request
->length
;
2438 osd_req_op_extent_init(osd_req
, 1, CEPH_OSD_OP_WRITE
,
2439 offset
, length
, 0, 0);
2440 if (orig_request
->type
== OBJ_REQUEST_BIO
)
2441 osd_req_op_extent_osd_data_bio(osd_req
, 1,
2442 orig_request
->bio_list
, length
);
2444 osd_req_op_extent_osd_data_pages(osd_req
, 1,
2445 orig_request
->pages
, length
,
2446 offset
& ~PAGE_MASK
, false, false);
2448 rbd_osd_req_format_write(orig_request
);
2450 /* All set, send it off. */
2452 osdc
= &rbd_dev
->rbd_client
->client
->osdc
;
2453 img_result
= rbd_obj_request_submit(osdc
, orig_request
);
2457 /* Record the error code and complete the request */
2459 orig_request
->result
= img_result
;
2460 orig_request
->xferred
= 0;
2461 obj_request_done_set(orig_request
);
2462 rbd_obj_request_complete(orig_request
);
2466 * Read from the parent image the range of data that covers the
2467 * entire target of the given object request. This is used for
2468 * satisfying a layered image write request when the target of an
2469 * object request from the image request does not exist.
2471 * A page array big enough to hold the returned data is allocated
2472 * and supplied to rbd_img_request_fill() as the "data descriptor."
2473 * When the read completes, this page array will be transferred to
2474 * the original object request for the copyup operation.
2476 * If an error occurs, record it as the result of the original
2477 * object request and mark it done so it gets completed.
2479 static int rbd_img_obj_parent_read_full(struct rbd_obj_request
*obj_request
)
2481 struct rbd_img_request
*img_request
= NULL
;
2482 struct rbd_img_request
*parent_request
= NULL
;
2483 struct rbd_device
*rbd_dev
;
2486 struct page
**pages
= NULL
;
2490 rbd_assert(obj_request_img_data_test(obj_request
));
2491 rbd_assert(obj_request_type_valid(obj_request
->type
));
2493 img_request
= obj_request
->img_request
;
2494 rbd_assert(img_request
!= NULL
);
2495 rbd_dev
= img_request
->rbd_dev
;
2496 rbd_assert(rbd_dev
->parent
!= NULL
);
2499 * Determine the byte range covered by the object in the
2500 * child image to which the original request was to be sent.
2502 img_offset
= obj_request
->img_offset
- obj_request
->offset
;
2503 length
= (u64
)1 << rbd_dev
->header
.obj_order
;
2506 * There is no defined parent data beyond the parent
2507 * overlap, so limit what we read at that boundary if
2510 if (img_offset
+ length
> rbd_dev
->parent_overlap
) {
2511 rbd_assert(img_offset
< rbd_dev
->parent_overlap
);
2512 length
= rbd_dev
->parent_overlap
- img_offset
;
2516 * Allocate a page array big enough to receive the data read
2519 page_count
= (u32
)calc_pages_for(0, length
);
2520 pages
= ceph_alloc_page_vector(page_count
, GFP_KERNEL
);
2521 if (IS_ERR(pages
)) {
2522 result
= PTR_ERR(pages
);
2528 parent_request
= rbd_parent_request_create(obj_request
,
2529 img_offset
, length
);
2530 if (!parent_request
)
2533 result
= rbd_img_request_fill(parent_request
, OBJ_REQUEST_PAGES
, pages
);
2536 parent_request
->copyup_pages
= pages
;
2537 parent_request
->copyup_page_count
= page_count
;
2539 parent_request
->callback
= rbd_img_obj_parent_read_full_callback
;
2540 result
= rbd_img_request_submit(parent_request
);
2544 parent_request
->copyup_pages
= NULL
;
2545 parent_request
->copyup_page_count
= 0;
2546 parent_request
->obj_request
= NULL
;
2547 rbd_obj_request_put(obj_request
);
2550 ceph_release_page_vector(pages
, page_count
);
2552 rbd_img_request_put(parent_request
);
2553 obj_request
->result
= result
;
2554 obj_request
->xferred
= 0;
2555 obj_request_done_set(obj_request
);
2560 static void rbd_img_obj_exists_callback(struct rbd_obj_request
*obj_request
)
2562 struct rbd_obj_request
*orig_request
;
2563 struct rbd_device
*rbd_dev
;
2566 rbd_assert(!obj_request_img_data_test(obj_request
));
2569 * All we need from the object request is the original
2570 * request and the result of the STAT op. Grab those, then
2571 * we're done with the request.
2573 orig_request
= obj_request
->obj_request
;
2574 obj_request
->obj_request
= NULL
;
2575 rbd_assert(orig_request
);
2576 rbd_assert(orig_request
->img_request
);
2578 result
= obj_request
->result
;
2579 obj_request
->result
= 0;
2581 dout("%s: obj %p for obj %p result %d %llu/%llu\n", __func__
,
2582 obj_request
, orig_request
, result
,
2583 obj_request
->xferred
, obj_request
->length
);
2584 rbd_obj_request_put(obj_request
);
2587 * If the overlap has become 0 (most likely because the
2588 * image has been flattened) we need to free the pages
2589 * and re-submit the original write request.
2591 rbd_dev
= orig_request
->img_request
->rbd_dev
;
2592 if (!rbd_dev
->parent_overlap
) {
2593 struct ceph_osd_client
*osdc
;
2595 rbd_obj_request_put(orig_request
);
2596 osdc
= &rbd_dev
->rbd_client
->client
->osdc
;
2597 result
= rbd_obj_request_submit(osdc
, orig_request
);
2603 * Our only purpose here is to determine whether the object
2604 * exists, and we don't want to treat the non-existence as
2605 * an error. If something else comes back, transfer the
2606 * error to the original request and complete it now.
2609 obj_request_existence_set(orig_request
, true);
2610 } else if (result
== -ENOENT
) {
2611 obj_request_existence_set(orig_request
, false);
2612 } else if (result
) {
2613 orig_request
->result
= result
;
2618 * Resubmit the original request now that we have recorded
2619 * whether the target object exists.
2621 orig_request
->result
= rbd_img_obj_request_submit(orig_request
);
2623 if (orig_request
->result
)
2624 rbd_obj_request_complete(orig_request
);
2625 rbd_obj_request_put(orig_request
);
2628 static int rbd_img_obj_exists_submit(struct rbd_obj_request
*obj_request
)
2630 struct rbd_obj_request
*stat_request
;
2631 struct rbd_device
*rbd_dev
;
2632 struct ceph_osd_client
*osdc
;
2633 struct page
**pages
= NULL
;
2639 * The response data for a STAT call consists of:
2646 size
= sizeof (__le64
) + sizeof (__le32
) + sizeof (__le32
);
2647 page_count
= (u32
)calc_pages_for(0, size
);
2648 pages
= ceph_alloc_page_vector(page_count
, GFP_KERNEL
);
2650 return PTR_ERR(pages
);
2653 stat_request
= rbd_obj_request_create(obj_request
->object_name
, 0, 0,
2658 rbd_obj_request_get(obj_request
);
2659 stat_request
->obj_request
= obj_request
;
2660 stat_request
->pages
= pages
;
2661 stat_request
->page_count
= page_count
;
2663 rbd_assert(obj_request
->img_request
);
2664 rbd_dev
= obj_request
->img_request
->rbd_dev
;
2665 stat_request
->osd_req
= rbd_osd_req_create(rbd_dev
, false,
2667 if (!stat_request
->osd_req
)
2669 stat_request
->callback
= rbd_img_obj_exists_callback
;
2671 osd_req_op_init(stat_request
->osd_req
, 0, CEPH_OSD_OP_STAT
);
2672 osd_req_op_raw_data_in_pages(stat_request
->osd_req
, 0, pages
, size
, 0,
2674 rbd_osd_req_format_read(stat_request
);
2676 osdc
= &rbd_dev
->rbd_client
->client
->osdc
;
2677 ret
= rbd_obj_request_submit(osdc
, stat_request
);
2680 rbd_obj_request_put(obj_request
);
2685 static int rbd_img_obj_request_submit(struct rbd_obj_request
*obj_request
)
2687 struct rbd_img_request
*img_request
;
2688 struct rbd_device
*rbd_dev
;
2691 rbd_assert(obj_request_img_data_test(obj_request
));
2693 img_request
= obj_request
->img_request
;
2694 rbd_assert(img_request
);
2695 rbd_dev
= img_request
->rbd_dev
;
2698 * Only writes to layered images need special handling.
2699 * Reads and non-layered writes are simple object requests.
2700 * Layered writes that start beyond the end of the overlap
2701 * with the parent have no parent data, so they too are
2702 * simple object requests. Finally, if the target object is
2703 * known to already exist, its parent data has already been
2704 * copied, so a write to the object can also be handled as a
2705 * simple object request.
2707 if (!img_request_write_test(img_request
) ||
2708 !img_request_layered_test(img_request
) ||
2709 !obj_request_overlaps_parent(obj_request
) ||
2710 ((known
= obj_request_known_test(obj_request
)) &&
2711 obj_request_exists_test(obj_request
))) {
2713 struct rbd_device
*rbd_dev
;
2714 struct ceph_osd_client
*osdc
;
2716 rbd_dev
= obj_request
->img_request
->rbd_dev
;
2717 osdc
= &rbd_dev
->rbd_client
->client
->osdc
;
2719 return rbd_obj_request_submit(osdc
, obj_request
);
2723 * It's a layered write. The target object might exist but
2724 * we may not know that yet. If we know it doesn't exist,
2725 * start by reading the data for the full target object from
2726 * the parent so we can use it for a copyup to the target.
2729 return rbd_img_obj_parent_read_full(obj_request
);
2731 /* We don't know whether the target exists. Go find out. */
2733 return rbd_img_obj_exists_submit(obj_request
);
2736 static int rbd_img_request_submit(struct rbd_img_request
*img_request
)
2738 struct rbd_obj_request
*obj_request
;
2739 struct rbd_obj_request
*next_obj_request
;
2741 dout("%s: img %p\n", __func__
, img_request
);
2742 for_each_obj_request_safe(img_request
, obj_request
, next_obj_request
) {
2745 ret
= rbd_img_obj_request_submit(obj_request
);
2753 static void rbd_img_parent_read_callback(struct rbd_img_request
*img_request
)
2755 struct rbd_obj_request
*obj_request
;
2756 struct rbd_device
*rbd_dev
;
2761 rbd_assert(img_request_child_test(img_request
));
2763 /* First get what we need from the image request and release it */
2765 obj_request
= img_request
->obj_request
;
2766 img_xferred
= img_request
->xferred
;
2767 img_result
= img_request
->result
;
2768 rbd_img_request_put(img_request
);
2771 * If the overlap has become 0 (most likely because the
2772 * image has been flattened) we need to re-submit the
2775 rbd_assert(obj_request
);
2776 rbd_assert(obj_request
->img_request
);
2777 rbd_dev
= obj_request
->img_request
->rbd_dev
;
2778 if (!rbd_dev
->parent_overlap
) {
2779 struct ceph_osd_client
*osdc
;
2781 osdc
= &rbd_dev
->rbd_client
->client
->osdc
;
2782 img_result
= rbd_obj_request_submit(osdc
, obj_request
);
2787 obj_request
->result
= img_result
;
2788 if (obj_request
->result
)
2792 * We need to zero anything beyond the parent overlap
2793 * boundary. Since rbd_img_obj_request_read_callback()
2794 * will zero anything beyond the end of a short read, an
2795 * easy way to do this is to pretend the data from the
2796 * parent came up short--ending at the overlap boundary.
2798 rbd_assert(obj_request
->img_offset
< U64_MAX
- obj_request
->length
);
2799 obj_end
= obj_request
->img_offset
+ obj_request
->length
;
2800 if (obj_end
> rbd_dev
->parent_overlap
) {
2803 if (obj_request
->img_offset
< rbd_dev
->parent_overlap
)
2804 xferred
= rbd_dev
->parent_overlap
-
2805 obj_request
->img_offset
;
2807 obj_request
->xferred
= min(img_xferred
, xferred
);
2809 obj_request
->xferred
= img_xferred
;
2812 rbd_img_obj_request_read_callback(obj_request
);
2813 rbd_obj_request_complete(obj_request
);
2816 static void rbd_img_parent_read(struct rbd_obj_request
*obj_request
)
2818 struct rbd_img_request
*img_request
;
2821 rbd_assert(obj_request_img_data_test(obj_request
));
2822 rbd_assert(obj_request
->img_request
!= NULL
);
2823 rbd_assert(obj_request
->result
== (s32
) -ENOENT
);
2824 rbd_assert(obj_request_type_valid(obj_request
->type
));
2826 /* rbd_read_finish(obj_request, obj_request->length); */
2827 img_request
= rbd_parent_request_create(obj_request
,
2828 obj_request
->img_offset
,
2829 obj_request
->length
);
2834 if (obj_request
->type
== OBJ_REQUEST_BIO
)
2835 result
= rbd_img_request_fill(img_request
, OBJ_REQUEST_BIO
,
2836 obj_request
->bio_list
);
2838 result
= rbd_img_request_fill(img_request
, OBJ_REQUEST_PAGES
,
2839 obj_request
->pages
);
2843 img_request
->callback
= rbd_img_parent_read_callback
;
2844 result
= rbd_img_request_submit(img_request
);
2851 rbd_img_request_put(img_request
);
2852 obj_request
->result
= result
;
2853 obj_request
->xferred
= 0;
2854 obj_request_done_set(obj_request
);
2857 static int rbd_obj_notify_ack_sync(struct rbd_device
*rbd_dev
, u64 notify_id
)
2859 struct rbd_obj_request
*obj_request
;
2860 struct ceph_osd_client
*osdc
= &rbd_dev
->rbd_client
->client
->osdc
;
2863 obj_request
= rbd_obj_request_create(rbd_dev
->header_name
, 0, 0,
2864 OBJ_REQUEST_NODATA
);
2869 obj_request
->osd_req
= rbd_osd_req_create(rbd_dev
, false, obj_request
);
2870 if (!obj_request
->osd_req
)
2873 osd_req_op_watch_init(obj_request
->osd_req
, 0, CEPH_OSD_OP_NOTIFY_ACK
,
2875 rbd_osd_req_format_read(obj_request
);
2877 ret
= rbd_obj_request_submit(osdc
, obj_request
);
2880 ret
= rbd_obj_request_wait(obj_request
);
2882 rbd_obj_request_put(obj_request
);
2887 static void rbd_watch_cb(u64 ver
, u64 notify_id
, u8 opcode
, void *data
)
2889 struct rbd_device
*rbd_dev
= (struct rbd_device
*)data
;
2895 dout("%s: \"%s\" notify_id %llu opcode %u\n", __func__
,
2896 rbd_dev
->header_name
, (unsigned long long)notify_id
,
2897 (unsigned int)opcode
);
2898 ret
= rbd_dev_refresh(rbd_dev
);
2900 rbd_warn(rbd_dev
, ": header refresh error (%d)\n", ret
);
2902 rbd_obj_notify_ack_sync(rbd_dev
, notify_id
);
2906 * Request sync osd watch/unwatch. The value of "start" determines
2907 * whether a watch request is being initiated or torn down.
2909 static int rbd_dev_header_watch_sync(struct rbd_device
*rbd_dev
, bool start
)
2911 struct ceph_osd_client
*osdc
= &rbd_dev
->rbd_client
->client
->osdc
;
2912 struct rbd_obj_request
*obj_request
;
2915 rbd_assert(start
^ !!rbd_dev
->watch_event
);
2916 rbd_assert(start
^ !!rbd_dev
->watch_request
);
2919 ret
= ceph_osdc_create_event(osdc
, rbd_watch_cb
, rbd_dev
,
2920 &rbd_dev
->watch_event
);
2923 rbd_assert(rbd_dev
->watch_event
!= NULL
);
2927 obj_request
= rbd_obj_request_create(rbd_dev
->header_name
, 0, 0,
2928 OBJ_REQUEST_NODATA
);
2932 obj_request
->osd_req
= rbd_osd_req_create(rbd_dev
, true, obj_request
);
2933 if (!obj_request
->osd_req
)
2937 ceph_osdc_set_request_linger(osdc
, obj_request
->osd_req
);
2939 ceph_osdc_unregister_linger_request(osdc
,
2940 rbd_dev
->watch_request
->osd_req
);
2942 osd_req_op_watch_init(obj_request
->osd_req
, 0, CEPH_OSD_OP_WATCH
,
2943 rbd_dev
->watch_event
->cookie
, 0, start
? 1 : 0);
2944 rbd_osd_req_format_write(obj_request
);
2946 ret
= rbd_obj_request_submit(osdc
, obj_request
);
2949 ret
= rbd_obj_request_wait(obj_request
);
2952 ret
= obj_request
->result
;
2957 * A watch request is set to linger, so the underlying osd
2958 * request won't go away until we unregister it. We retain
2959 * a pointer to the object request during that time (in
2960 * rbd_dev->watch_request), so we'll keep a reference to
2961 * it. We'll drop that reference (below) after we've
2965 rbd_dev
->watch_request
= obj_request
;
2970 /* We have successfully torn down the watch request */
2972 rbd_obj_request_put(rbd_dev
->watch_request
);
2973 rbd_dev
->watch_request
= NULL
;
2975 /* Cancel the event if we're tearing down, or on error */
2976 ceph_osdc_cancel_event(rbd_dev
->watch_event
);
2977 rbd_dev
->watch_event
= NULL
;
2979 rbd_obj_request_put(obj_request
);
2985 * Synchronous osd object method call. Returns the number of bytes
2986 * returned in the outbound buffer, or a negative error code.
2988 static int rbd_obj_method_sync(struct rbd_device
*rbd_dev
,
2989 const char *object_name
,
2990 const char *class_name
,
2991 const char *method_name
,
2992 const void *outbound
,
2993 size_t outbound_size
,
2995 size_t inbound_size
)
2997 struct ceph_osd_client
*osdc
= &rbd_dev
->rbd_client
->client
->osdc
;
2998 struct rbd_obj_request
*obj_request
;
2999 struct page
**pages
;
3004 * Method calls are ultimately read operations. The result
3005 * should placed into the inbound buffer provided. They
3006 * also supply outbound data--parameters for the object
3007 * method. Currently if this is present it will be a
3010 page_count
= (u32
)calc_pages_for(0, inbound_size
);
3011 pages
= ceph_alloc_page_vector(page_count
, GFP_KERNEL
);
3013 return PTR_ERR(pages
);
3016 obj_request
= rbd_obj_request_create(object_name
, 0, inbound_size
,
3021 obj_request
->pages
= pages
;
3022 obj_request
->page_count
= page_count
;
3024 obj_request
->osd_req
= rbd_osd_req_create(rbd_dev
, false, obj_request
);
3025 if (!obj_request
->osd_req
)
3028 osd_req_op_cls_init(obj_request
->osd_req
, 0, CEPH_OSD_OP_CALL
,
3029 class_name
, method_name
);
3030 if (outbound_size
) {
3031 struct ceph_pagelist
*pagelist
;
3033 pagelist
= kmalloc(sizeof (*pagelist
), GFP_NOFS
);
3037 ceph_pagelist_init(pagelist
);
3038 ceph_pagelist_append(pagelist
, outbound
, outbound_size
);
3039 osd_req_op_cls_request_data_pagelist(obj_request
->osd_req
, 0,
3042 osd_req_op_cls_response_data_pages(obj_request
->osd_req
, 0,
3043 obj_request
->pages
, inbound_size
,
3045 rbd_osd_req_format_read(obj_request
);
3047 ret
= rbd_obj_request_submit(osdc
, obj_request
);
3050 ret
= rbd_obj_request_wait(obj_request
);
3054 ret
= obj_request
->result
;
3058 rbd_assert(obj_request
->xferred
< (u64
)INT_MAX
);
3059 ret
= (int)obj_request
->xferred
;
3060 ceph_copy_from_page_vector(pages
, inbound
, 0, obj_request
->xferred
);
3063 rbd_obj_request_put(obj_request
);
3065 ceph_release_page_vector(pages
, page_count
);
3070 static void rbd_request_fn(struct request_queue
*q
)
3071 __releases(q
->queue_lock
) __acquires(q
->queue_lock
)
3073 struct rbd_device
*rbd_dev
= q
->queuedata
;
3074 bool read_only
= rbd_dev
->mapping
.read_only
;
3078 while ((rq
= blk_fetch_request(q
))) {
3079 bool write_request
= rq_data_dir(rq
) == WRITE
;
3080 struct rbd_img_request
*img_request
;
3084 /* Ignore any non-FS requests that filter through. */
3086 if (rq
->cmd_type
!= REQ_TYPE_FS
) {
3087 dout("%s: non-fs request type %d\n", __func__
,
3088 (int) rq
->cmd_type
);
3089 __blk_end_request_all(rq
, 0);
3093 /* Ignore/skip any zero-length requests */
3095 offset
= (u64
) blk_rq_pos(rq
) << SECTOR_SHIFT
;
3096 length
= (u64
) blk_rq_bytes(rq
);
3099 dout("%s: zero-length request\n", __func__
);
3100 __blk_end_request_all(rq
, 0);
3104 spin_unlock_irq(q
->queue_lock
);
3106 /* Disallow writes to a read-only device */
3108 if (write_request
) {
3112 rbd_assert(rbd_dev
->spec
->snap_id
== CEPH_NOSNAP
);
3116 * Quit early if the mapped snapshot no longer
3117 * exists. It's still possible the snapshot will
3118 * have disappeared by the time our request arrives
3119 * at the osd, but there's no sense in sending it if
3122 if (!test_bit(RBD_DEV_FLAG_EXISTS
, &rbd_dev
->flags
)) {
3123 dout("request for non-existent snapshot");
3124 rbd_assert(rbd_dev
->spec
->snap_id
!= CEPH_NOSNAP
);
3130 if (offset
&& length
> U64_MAX
- offset
+ 1) {
3131 rbd_warn(rbd_dev
, "bad request range (%llu~%llu)\n",
3133 goto end_request
; /* Shouldn't happen */
3137 if (offset
+ length
> rbd_dev
->mapping
.size
) {
3138 rbd_warn(rbd_dev
, "beyond EOD (%llu~%llu > %llu)\n",
3139 offset
, length
, rbd_dev
->mapping
.size
);
3144 img_request
= rbd_img_request_create(rbd_dev
, offset
, length
,
3149 img_request
->rq
= rq
;
3151 result
= rbd_img_request_fill(img_request
, OBJ_REQUEST_BIO
,
3154 result
= rbd_img_request_submit(img_request
);
3156 rbd_img_request_put(img_request
);
3158 spin_lock_irq(q
->queue_lock
);
3160 rbd_warn(rbd_dev
, "%s %llx at %llx result %d\n",
3161 write_request
? "write" : "read",
3162 length
, offset
, result
);
3164 __blk_end_request_all(rq
, result
);
3170 * a queue callback. Makes sure that we don't create a bio that spans across
3171 * multiple osd objects. One exception would be with a single page bios,
3172 * which we handle later at bio_chain_clone_range()
3174 static int rbd_merge_bvec(struct request_queue
*q
, struct bvec_merge_data
*bmd
,
3175 struct bio_vec
*bvec
)
3177 struct rbd_device
*rbd_dev
= q
->queuedata
;
3178 sector_t sector_offset
;
3179 sector_t sectors_per_obj
;
3180 sector_t obj_sector_offset
;
3184 * Find how far into its rbd object the partition-relative
3185 * bio start sector is to offset relative to the enclosing
3188 sector_offset
= get_start_sect(bmd
->bi_bdev
) + bmd
->bi_sector
;
3189 sectors_per_obj
= 1 << (rbd_dev
->header
.obj_order
- SECTOR_SHIFT
);
3190 obj_sector_offset
= sector_offset
& (sectors_per_obj
- 1);
3193 * Compute the number of bytes from that offset to the end
3194 * of the object. Account for what's already used by the bio.
3196 ret
= (int) (sectors_per_obj
- obj_sector_offset
) << SECTOR_SHIFT
;
3197 if (ret
> bmd
->bi_size
)
3198 ret
-= bmd
->bi_size
;
3203 * Don't send back more than was asked for. And if the bio
3204 * was empty, let the whole thing through because: "Note
3205 * that a block device *must* allow a single page to be
3206 * added to an empty bio."
3208 rbd_assert(bvec
->bv_len
<= PAGE_SIZE
);
3209 if (ret
> (int) bvec
->bv_len
|| !bmd
->bi_size
)
3210 ret
= (int) bvec
->bv_len
;
3215 static void rbd_free_disk(struct rbd_device
*rbd_dev
)
3217 struct gendisk
*disk
= rbd_dev
->disk
;
3222 rbd_dev
->disk
= NULL
;
3223 if (disk
->flags
& GENHD_FL_UP
) {
3226 blk_cleanup_queue(disk
->queue
);
3231 static int rbd_obj_read_sync(struct rbd_device
*rbd_dev
,
3232 const char *object_name
,
3233 u64 offset
, u64 length
, void *buf
)
3236 struct ceph_osd_client
*osdc
= &rbd_dev
->rbd_client
->client
->osdc
;
3237 struct rbd_obj_request
*obj_request
;
3238 struct page
**pages
= NULL
;
3243 page_count
= (u32
) calc_pages_for(offset
, length
);
3244 pages
= ceph_alloc_page_vector(page_count
, GFP_KERNEL
);
3246 return PTR_ERR(pages
);
3249 obj_request
= rbd_obj_request_create(object_name
, offset
, length
,
3254 obj_request
->pages
= pages
;
3255 obj_request
->page_count
= page_count
;
3257 obj_request
->osd_req
= rbd_osd_req_create(rbd_dev
, false, obj_request
);
3258 if (!obj_request
->osd_req
)
3261 osd_req_op_extent_init(obj_request
->osd_req
, 0, CEPH_OSD_OP_READ
,
3262 offset
, length
, 0, 0);
3263 osd_req_op_extent_osd_data_pages(obj_request
->osd_req
, 0,
3265 obj_request
->length
,
3266 obj_request
->offset
& ~PAGE_MASK
,
3268 rbd_osd_req_format_read(obj_request
);
3270 ret
= rbd_obj_request_submit(osdc
, obj_request
);
3273 ret
= rbd_obj_request_wait(obj_request
);
3277 ret
= obj_request
->result
;
3281 rbd_assert(obj_request
->xferred
<= (u64
) SIZE_MAX
);
3282 size
= (size_t) obj_request
->xferred
;
3283 ceph_copy_from_page_vector(pages
, buf
, 0, size
);
3284 rbd_assert(size
<= (size_t)INT_MAX
);
3288 rbd_obj_request_put(obj_request
);
3290 ceph_release_page_vector(pages
, page_count
);
3296 * Read the complete header for the given rbd device. On successful
3297 * return, the rbd_dev->header field will contain up-to-date
3298 * information about the image.
3300 static int rbd_dev_v1_header_info(struct rbd_device
*rbd_dev
)
3302 struct rbd_image_header_ondisk
*ondisk
= NULL
;
3309 * The complete header will include an array of its 64-bit
3310 * snapshot ids, followed by the names of those snapshots as
3311 * a contiguous block of NUL-terminated strings. Note that
3312 * the number of snapshots could change by the time we read
3313 * it in, in which case we re-read it.
3320 size
= sizeof (*ondisk
);
3321 size
+= snap_count
* sizeof (struct rbd_image_snap_ondisk
);
3323 ondisk
= kmalloc(size
, GFP_KERNEL
);
3327 ret
= rbd_obj_read_sync(rbd_dev
, rbd_dev
->header_name
,
3331 if ((size_t)ret
< size
) {
3333 rbd_warn(rbd_dev
, "short header read (want %zd got %d)",
3337 if (!rbd_dev_ondisk_valid(ondisk
)) {
3339 rbd_warn(rbd_dev
, "invalid header");
3343 names_size
= le64_to_cpu(ondisk
->snap_names_len
);
3344 want_count
= snap_count
;
3345 snap_count
= le32_to_cpu(ondisk
->snap_count
);
3346 } while (snap_count
!= want_count
);
3348 ret
= rbd_header_from_disk(rbd_dev
, ondisk
);
3356 * Clear the rbd device's EXISTS flag if the snapshot it's mapped to
3357 * has disappeared from the (just updated) snapshot context.
3359 static void rbd_exists_validate(struct rbd_device
*rbd_dev
)
3363 if (!test_bit(RBD_DEV_FLAG_EXISTS
, &rbd_dev
->flags
))
3366 snap_id
= rbd_dev
->spec
->snap_id
;
3367 if (snap_id
== CEPH_NOSNAP
)
3370 if (rbd_dev_snap_index(rbd_dev
, snap_id
) == BAD_SNAP_INDEX
)
3371 clear_bit(RBD_DEV_FLAG_EXISTS
, &rbd_dev
->flags
);
3374 static void rbd_dev_update_size(struct rbd_device
*rbd_dev
)
3380 * Don't hold the lock while doing disk operations,
3381 * or lock ordering will conflict with the bdev mutex via:
3382 * rbd_add() -> blkdev_get() -> rbd_open()
3384 spin_lock_irq(&rbd_dev
->lock
);
3385 removing
= test_bit(RBD_DEV_FLAG_REMOVING
, &rbd_dev
->flags
);
3386 spin_unlock_irq(&rbd_dev
->lock
);
3388 * If the device is being removed, rbd_dev->disk has
3389 * been destroyed, so don't try to update its size
3392 size
= (sector_t
)rbd_dev
->mapping
.size
/ SECTOR_SIZE
;
3393 dout("setting size to %llu sectors", (unsigned long long)size
);
3394 set_capacity(rbd_dev
->disk
, size
);
3395 revalidate_disk(rbd_dev
->disk
);
3399 static int rbd_dev_refresh(struct rbd_device
*rbd_dev
)
3404 rbd_assert(rbd_image_format_valid(rbd_dev
->image_format
));
3405 mapping_size
= rbd_dev
->mapping
.size
;
3406 mutex_lock_nested(&ctl_mutex
, SINGLE_DEPTH_NESTING
);
3407 if (rbd_dev
->image_format
== 1)
3408 ret
= rbd_dev_v1_header_info(rbd_dev
);
3410 ret
= rbd_dev_v2_header_info(rbd_dev
);
3412 /* If it's a mapped snapshot, validate its EXISTS flag */
3414 rbd_exists_validate(rbd_dev
);
3415 mutex_unlock(&ctl_mutex
);
3416 if (mapping_size
!= rbd_dev
->mapping
.size
) {
3417 rbd_dev_update_size(rbd_dev
);
3423 static int rbd_init_disk(struct rbd_device
*rbd_dev
)
3425 struct gendisk
*disk
;
3426 struct request_queue
*q
;
3429 /* create gendisk info */
3430 disk
= alloc_disk(RBD_MINORS_PER_MAJOR
);
3434 snprintf(disk
->disk_name
, sizeof(disk
->disk_name
), RBD_DRV_NAME
"%d",
3436 disk
->major
= rbd_dev
->major
;
3437 disk
->first_minor
= 0;
3438 disk
->fops
= &rbd_bd_ops
;
3439 disk
->private_data
= rbd_dev
;
3441 q
= blk_init_queue(rbd_request_fn
, &rbd_dev
->lock
);
3445 /* We use the default size, but let's be explicit about it. */
3446 blk_queue_physical_block_size(q
, SECTOR_SIZE
);
3448 /* set io sizes to object size */
3449 segment_size
= rbd_obj_bytes(&rbd_dev
->header
);
3450 blk_queue_max_hw_sectors(q
, segment_size
/ SECTOR_SIZE
);
3451 blk_queue_max_segment_size(q
, segment_size
);
3452 blk_queue_io_min(q
, segment_size
);
3453 blk_queue_io_opt(q
, segment_size
);
3455 blk_queue_merge_bvec(q
, rbd_merge_bvec
);
3456 if (!ceph_test_opt(rbd_dev
->rbd_client
->client
, NOCRC
))
3457 q
->backing_dev_info
.capabilities
|= BDI_CAP_STABLE_WRITES
;
3461 q
->queuedata
= rbd_dev
;
3463 rbd_dev
->disk
= disk
;
3476 static struct rbd_device
*dev_to_rbd_dev(struct device
*dev
)
3478 return container_of(dev
, struct rbd_device
, dev
);
3481 static ssize_t
rbd_size_show(struct device
*dev
,
3482 struct device_attribute
*attr
, char *buf
)
3484 struct rbd_device
*rbd_dev
= dev_to_rbd_dev(dev
);
3486 return sprintf(buf
, "%llu\n",
3487 (unsigned long long)rbd_dev
->mapping
.size
);
3491 * Note this shows the features for whatever's mapped, which is not
3492 * necessarily the base image.
3494 static ssize_t
rbd_features_show(struct device
*dev
,
3495 struct device_attribute
*attr
, char *buf
)
3497 struct rbd_device
*rbd_dev
= dev_to_rbd_dev(dev
);
3499 return sprintf(buf
, "0x%016llx\n",
3500 (unsigned long long)rbd_dev
->mapping
.features
);
3503 static ssize_t
rbd_major_show(struct device
*dev
,
3504 struct device_attribute
*attr
, char *buf
)
3506 struct rbd_device
*rbd_dev
= dev_to_rbd_dev(dev
);
3509 return sprintf(buf
, "%d\n", rbd_dev
->major
);
3511 return sprintf(buf
, "(none)\n");
3515 static ssize_t
rbd_client_id_show(struct device
*dev
,
3516 struct device_attribute
*attr
, char *buf
)
3518 struct rbd_device
*rbd_dev
= dev_to_rbd_dev(dev
);
3520 return sprintf(buf
, "client%lld\n",
3521 ceph_client_id(rbd_dev
->rbd_client
->client
));
3524 static ssize_t
rbd_pool_show(struct device
*dev
,
3525 struct device_attribute
*attr
, char *buf
)
3527 struct rbd_device
*rbd_dev
= dev_to_rbd_dev(dev
);
3529 return sprintf(buf
, "%s\n", rbd_dev
->spec
->pool_name
);
3532 static ssize_t
rbd_pool_id_show(struct device
*dev
,
3533 struct device_attribute
*attr
, char *buf
)
3535 struct rbd_device
*rbd_dev
= dev_to_rbd_dev(dev
);
3537 return sprintf(buf
, "%llu\n",
3538 (unsigned long long) rbd_dev
->spec
->pool_id
);
3541 static ssize_t
rbd_name_show(struct device
*dev
,
3542 struct device_attribute
*attr
, char *buf
)
3544 struct rbd_device
*rbd_dev
= dev_to_rbd_dev(dev
);
3546 if (rbd_dev
->spec
->image_name
)
3547 return sprintf(buf
, "%s\n", rbd_dev
->spec
->image_name
);
3549 return sprintf(buf
, "(unknown)\n");
3552 static ssize_t
rbd_image_id_show(struct device
*dev
,
3553 struct device_attribute
*attr
, char *buf
)
3555 struct rbd_device
*rbd_dev
= dev_to_rbd_dev(dev
);
3557 return sprintf(buf
, "%s\n", rbd_dev
->spec
->image_id
);
3561 * Shows the name of the currently-mapped snapshot (or
3562 * RBD_SNAP_HEAD_NAME for the base image).
3564 static ssize_t
rbd_snap_show(struct device
*dev
,
3565 struct device_attribute
*attr
,
3568 struct rbd_device
*rbd_dev
= dev_to_rbd_dev(dev
);
3570 return sprintf(buf
, "%s\n", rbd_dev
->spec
->snap_name
);
3574 * For an rbd v2 image, shows the pool id, image id, and snapshot id
3575 * for the parent image. If there is no parent, simply shows
3576 * "(no parent image)".
3578 static ssize_t
rbd_parent_show(struct device
*dev
,
3579 struct device_attribute
*attr
,
3582 struct rbd_device
*rbd_dev
= dev_to_rbd_dev(dev
);
3583 struct rbd_spec
*spec
= rbd_dev
->parent_spec
;
3588 return sprintf(buf
, "(no parent image)\n");
3590 count
= sprintf(bufp
, "pool_id %llu\npool_name %s\n",
3591 (unsigned long long) spec
->pool_id
, spec
->pool_name
);
3596 count
= sprintf(bufp
, "image_id %s\nimage_name %s\n", spec
->image_id
,
3597 spec
->image_name
? spec
->image_name
: "(unknown)");
3602 count
= sprintf(bufp
, "snap_id %llu\nsnap_name %s\n",
3603 (unsigned long long) spec
->snap_id
, spec
->snap_name
);
3608 count
= sprintf(bufp
, "overlap %llu\n", rbd_dev
->parent_overlap
);
3613 return (ssize_t
) (bufp
- buf
);
3616 static ssize_t
rbd_image_refresh(struct device
*dev
,
3617 struct device_attribute
*attr
,
3621 struct rbd_device
*rbd_dev
= dev_to_rbd_dev(dev
);
3624 ret
= rbd_dev_refresh(rbd_dev
);
3626 rbd_warn(rbd_dev
, ": manual header refresh error (%d)\n", ret
);
3628 return ret
< 0 ? ret
: size
;
3631 static DEVICE_ATTR(size
, S_IRUGO
, rbd_size_show
, NULL
);
3632 static DEVICE_ATTR(features
, S_IRUGO
, rbd_features_show
, NULL
);
3633 static DEVICE_ATTR(major
, S_IRUGO
, rbd_major_show
, NULL
);
3634 static DEVICE_ATTR(client_id
, S_IRUGO
, rbd_client_id_show
, NULL
);
3635 static DEVICE_ATTR(pool
, S_IRUGO
, rbd_pool_show
, NULL
);
3636 static DEVICE_ATTR(pool_id
, S_IRUGO
, rbd_pool_id_show
, NULL
);
3637 static DEVICE_ATTR(name
, S_IRUGO
, rbd_name_show
, NULL
);
3638 static DEVICE_ATTR(image_id
, S_IRUGO
, rbd_image_id_show
, NULL
);
3639 static DEVICE_ATTR(refresh
, S_IWUSR
, NULL
, rbd_image_refresh
);
3640 static DEVICE_ATTR(current_snap
, S_IRUGO
, rbd_snap_show
, NULL
);
3641 static DEVICE_ATTR(parent
, S_IRUGO
, rbd_parent_show
, NULL
);
3643 static struct attribute
*rbd_attrs
[] = {
3644 &dev_attr_size
.attr
,
3645 &dev_attr_features
.attr
,
3646 &dev_attr_major
.attr
,
3647 &dev_attr_client_id
.attr
,
3648 &dev_attr_pool
.attr
,
3649 &dev_attr_pool_id
.attr
,
3650 &dev_attr_name
.attr
,
3651 &dev_attr_image_id
.attr
,
3652 &dev_attr_current_snap
.attr
,
3653 &dev_attr_parent
.attr
,
3654 &dev_attr_refresh
.attr
,
3658 static struct attribute_group rbd_attr_group
= {
3662 static const struct attribute_group
*rbd_attr_groups
[] = {
3667 static void rbd_sysfs_dev_release(struct device
*dev
)
3671 static struct device_type rbd_device_type
= {
3673 .groups
= rbd_attr_groups
,
3674 .release
= rbd_sysfs_dev_release
,
3677 static struct rbd_spec
*rbd_spec_get(struct rbd_spec
*spec
)
3679 kref_get(&spec
->kref
);
3684 static void rbd_spec_free(struct kref
*kref
);
3685 static void rbd_spec_put(struct rbd_spec
*spec
)
3688 kref_put(&spec
->kref
, rbd_spec_free
);
3691 static struct rbd_spec
*rbd_spec_alloc(void)
3693 struct rbd_spec
*spec
;
3695 spec
= kzalloc(sizeof (*spec
), GFP_KERNEL
);
3698 kref_init(&spec
->kref
);
3703 static void rbd_spec_free(struct kref
*kref
)
3705 struct rbd_spec
*spec
= container_of(kref
, struct rbd_spec
, kref
);
3707 kfree(spec
->pool_name
);
3708 kfree(spec
->image_id
);
3709 kfree(spec
->image_name
);
3710 kfree(spec
->snap_name
);
3714 static struct rbd_device
*rbd_dev_create(struct rbd_client
*rbdc
,
3715 struct rbd_spec
*spec
)
3717 struct rbd_device
*rbd_dev
;
3719 rbd_dev
= kzalloc(sizeof (*rbd_dev
), GFP_KERNEL
);
3723 spin_lock_init(&rbd_dev
->lock
);
3725 atomic_set(&rbd_dev
->parent_ref
, 0);
3726 INIT_LIST_HEAD(&rbd_dev
->node
);
3727 init_rwsem(&rbd_dev
->header_rwsem
);
3729 rbd_dev
->spec
= spec
;
3730 rbd_dev
->rbd_client
= rbdc
;
3732 /* Initialize the layout used for all rbd requests */
3734 rbd_dev
->layout
.fl_stripe_unit
= cpu_to_le32(1 << RBD_MAX_OBJ_ORDER
);
3735 rbd_dev
->layout
.fl_stripe_count
= cpu_to_le32(1);
3736 rbd_dev
->layout
.fl_object_size
= cpu_to_le32(1 << RBD_MAX_OBJ_ORDER
);
3737 rbd_dev
->layout
.fl_pg_pool
= cpu_to_le32((u32
) spec
->pool_id
);
3742 static void rbd_dev_destroy(struct rbd_device
*rbd_dev
)
3744 rbd_put_client(rbd_dev
->rbd_client
);
3745 rbd_spec_put(rbd_dev
->spec
);
3750 * Get the size and object order for an image snapshot, or if
3751 * snap_id is CEPH_NOSNAP, gets this information for the base
3754 static int _rbd_dev_v2_snap_size(struct rbd_device
*rbd_dev
, u64 snap_id
,
3755 u8
*order
, u64
*snap_size
)
3757 __le64 snapid
= cpu_to_le64(snap_id
);
3762 } __attribute__ ((packed
)) size_buf
= { 0 };
3764 ret
= rbd_obj_method_sync(rbd_dev
, rbd_dev
->header_name
,
3766 &snapid
, sizeof (snapid
),
3767 &size_buf
, sizeof (size_buf
));
3768 dout("%s: rbd_obj_method_sync returned %d\n", __func__
, ret
);
3771 if (ret
< sizeof (size_buf
))
3775 *order
= size_buf
.order
;
3776 dout(" order %u", (unsigned int)*order
);
3778 *snap_size
= le64_to_cpu(size_buf
.size
);
3780 dout(" snap_id 0x%016llx snap_size = %llu\n",
3781 (unsigned long long)snap_id
,
3782 (unsigned long long)*snap_size
);
3787 static int rbd_dev_v2_image_size(struct rbd_device
*rbd_dev
)
3789 return _rbd_dev_v2_snap_size(rbd_dev
, CEPH_NOSNAP
,
3790 &rbd_dev
->header
.obj_order
,
3791 &rbd_dev
->header
.image_size
);
3794 static int rbd_dev_v2_object_prefix(struct rbd_device
*rbd_dev
)
3800 reply_buf
= kzalloc(RBD_OBJ_PREFIX_LEN_MAX
, GFP_KERNEL
);
3804 ret
= rbd_obj_method_sync(rbd_dev
, rbd_dev
->header_name
,
3805 "rbd", "get_object_prefix", NULL
, 0,
3806 reply_buf
, RBD_OBJ_PREFIX_LEN_MAX
);
3807 dout("%s: rbd_obj_method_sync returned %d\n", __func__
, ret
);
3812 rbd_dev
->header
.object_prefix
= ceph_extract_encoded_string(&p
,
3813 p
+ ret
, NULL
, GFP_NOIO
);
3816 if (IS_ERR(rbd_dev
->header
.object_prefix
)) {
3817 ret
= PTR_ERR(rbd_dev
->header
.object_prefix
);
3818 rbd_dev
->header
.object_prefix
= NULL
;
3820 dout(" object_prefix = %s\n", rbd_dev
->header
.object_prefix
);
3828 static int _rbd_dev_v2_snap_features(struct rbd_device
*rbd_dev
, u64 snap_id
,
3831 __le64 snapid
= cpu_to_le64(snap_id
);
3835 } __attribute__ ((packed
)) features_buf
= { 0 };
3839 ret
= rbd_obj_method_sync(rbd_dev
, rbd_dev
->header_name
,
3840 "rbd", "get_features",
3841 &snapid
, sizeof (snapid
),
3842 &features_buf
, sizeof (features_buf
));
3843 dout("%s: rbd_obj_method_sync returned %d\n", __func__
, ret
);
3846 if (ret
< sizeof (features_buf
))
3849 incompat
= le64_to_cpu(features_buf
.incompat
);
3850 if (incompat
& ~RBD_FEATURES_SUPPORTED
)
3853 *snap_features
= le64_to_cpu(features_buf
.features
);
3855 dout(" snap_id 0x%016llx features = 0x%016llx incompat = 0x%016llx\n",
3856 (unsigned long long)snap_id
,
3857 (unsigned long long)*snap_features
,
3858 (unsigned long long)le64_to_cpu(features_buf
.incompat
));
3863 static int rbd_dev_v2_features(struct rbd_device
*rbd_dev
)
3865 return _rbd_dev_v2_snap_features(rbd_dev
, CEPH_NOSNAP
,
3866 &rbd_dev
->header
.features
);
3869 static int rbd_dev_v2_parent_info(struct rbd_device
*rbd_dev
)
3871 struct rbd_spec
*parent_spec
;
3873 void *reply_buf
= NULL
;
3882 parent_spec
= rbd_spec_alloc();
3886 size
= sizeof (__le64
) + /* pool_id */
3887 sizeof (__le32
) + RBD_IMAGE_ID_LEN_MAX
+ /* image_id */
3888 sizeof (__le64
) + /* snap_id */
3889 sizeof (__le64
); /* overlap */
3890 reply_buf
= kmalloc(size
, GFP_KERNEL
);
3896 snapid
= cpu_to_le64(CEPH_NOSNAP
);
3897 ret
= rbd_obj_method_sync(rbd_dev
, rbd_dev
->header_name
,
3898 "rbd", "get_parent",
3899 &snapid
, sizeof (snapid
),
3901 dout("%s: rbd_obj_method_sync returned %d\n", __func__
, ret
);
3906 end
= reply_buf
+ ret
;
3908 ceph_decode_64_safe(&p
, end
, pool_id
, out_err
);
3909 if (pool_id
== CEPH_NOPOOL
) {
3911 * Either the parent never existed, or we have
3912 * record of it but the image got flattened so it no
3913 * longer has a parent. When the parent of a
3914 * layered image disappears we immediately set the
3915 * overlap to 0. The effect of this is that all new
3916 * requests will be treated as if the image had no
3919 if (rbd_dev
->parent_overlap
) {
3920 rbd_dev
->parent_overlap
= 0;
3922 rbd_dev_parent_put(rbd_dev
);
3923 pr_info("%s: clone image has been flattened\n",
3924 rbd_dev
->disk
->disk_name
);
3927 goto out
; /* No parent? No problem. */
3930 /* The ceph file layout needs to fit pool id in 32 bits */
3933 if (pool_id
> (u64
)U32_MAX
) {
3934 rbd_warn(NULL
, "parent pool id too large (%llu > %u)\n",
3935 (unsigned long long)pool_id
, U32_MAX
);
3938 parent_spec
->pool_id
= pool_id
;
3940 image_id
= ceph_extract_encoded_string(&p
, end
, NULL
, GFP_KERNEL
);
3941 if (IS_ERR(image_id
)) {
3942 ret
= PTR_ERR(image_id
);
3945 parent_spec
->image_id
= image_id
;
3946 ceph_decode_64_safe(&p
, end
, parent_spec
->snap_id
, out_err
);
3947 ceph_decode_64_safe(&p
, end
, overlap
, out_err
);
3950 rbd_spec_put(rbd_dev
->parent_spec
);
3951 rbd_dev
->parent_spec
= parent_spec
;
3952 parent_spec
= NULL
; /* rbd_dev now owns this */
3953 rbd_dev
->parent_overlap
= overlap
;
3955 rbd_warn(rbd_dev
, "ignoring parent of clone with overlap 0\n");
3961 rbd_spec_put(parent_spec
);
3966 static int rbd_dev_v2_striping_info(struct rbd_device
*rbd_dev
)
3970 __le64 stripe_count
;
3971 } __attribute__ ((packed
)) striping_info_buf
= { 0 };
3972 size_t size
= sizeof (striping_info_buf
);
3979 ret
= rbd_obj_method_sync(rbd_dev
, rbd_dev
->header_name
,
3980 "rbd", "get_stripe_unit_count", NULL
, 0,
3981 (char *)&striping_info_buf
, size
);
3982 dout("%s: rbd_obj_method_sync returned %d\n", __func__
, ret
);
3989 * We don't actually support the "fancy striping" feature
3990 * (STRIPINGV2) yet, but if the striping sizes are the
3991 * defaults the behavior is the same as before. So find
3992 * out, and only fail if the image has non-default values.
3995 obj_size
= (u64
)1 << rbd_dev
->header
.obj_order
;
3996 p
= &striping_info_buf
;
3997 stripe_unit
= ceph_decode_64(&p
);
3998 if (stripe_unit
!= obj_size
) {
3999 rbd_warn(rbd_dev
, "unsupported stripe unit "
4000 "(got %llu want %llu)",
4001 stripe_unit
, obj_size
);
4004 stripe_count
= ceph_decode_64(&p
);
4005 if (stripe_count
!= 1) {
4006 rbd_warn(rbd_dev
, "unsupported stripe count "
4007 "(got %llu want 1)", stripe_count
);
4010 rbd_dev
->header
.stripe_unit
= stripe_unit
;
4011 rbd_dev
->header
.stripe_count
= stripe_count
;
4016 static char *rbd_dev_image_name(struct rbd_device
*rbd_dev
)
4018 size_t image_id_size
;
4023 void *reply_buf
= NULL
;
4025 char *image_name
= NULL
;
4028 rbd_assert(!rbd_dev
->spec
->image_name
);
4030 len
= strlen(rbd_dev
->spec
->image_id
);
4031 image_id_size
= sizeof (__le32
) + len
;
4032 image_id
= kmalloc(image_id_size
, GFP_KERNEL
);
4037 end
= image_id
+ image_id_size
;
4038 ceph_encode_string(&p
, end
, rbd_dev
->spec
->image_id
, (u32
)len
);
4040 size
= sizeof (__le32
) + RBD_IMAGE_NAME_LEN_MAX
;
4041 reply_buf
= kmalloc(size
, GFP_KERNEL
);
4045 ret
= rbd_obj_method_sync(rbd_dev
, RBD_DIRECTORY
,
4046 "rbd", "dir_get_name",
4047 image_id
, image_id_size
,
4052 end
= reply_buf
+ ret
;
4054 image_name
= ceph_extract_encoded_string(&p
, end
, &len
, GFP_KERNEL
);
4055 if (IS_ERR(image_name
))
4058 dout("%s: name is %s len is %zd\n", __func__
, image_name
, len
);
4066 static u64
rbd_v1_snap_id_by_name(struct rbd_device
*rbd_dev
, const char *name
)
4068 struct ceph_snap_context
*snapc
= rbd_dev
->header
.snapc
;
4069 const char *snap_name
;
4072 /* Skip over names until we find the one we are looking for */
4074 snap_name
= rbd_dev
->header
.snap_names
;
4075 while (which
< snapc
->num_snaps
) {
4076 if (!strcmp(name
, snap_name
))
4077 return snapc
->snaps
[which
];
4078 snap_name
+= strlen(snap_name
) + 1;
4084 static u64
rbd_v2_snap_id_by_name(struct rbd_device
*rbd_dev
, const char *name
)
4086 struct ceph_snap_context
*snapc
= rbd_dev
->header
.snapc
;
4091 for (which
= 0; !found
&& which
< snapc
->num_snaps
; which
++) {
4092 const char *snap_name
;
4094 snap_id
= snapc
->snaps
[which
];
4095 snap_name
= rbd_dev_v2_snap_name(rbd_dev
, snap_id
);
4096 if (IS_ERR(snap_name
)) {
4097 /* ignore no-longer existing snapshots */
4098 if (PTR_ERR(snap_name
) == -ENOENT
)
4103 found
= !strcmp(name
, snap_name
);
4106 return found
? snap_id
: CEPH_NOSNAP
;
4110 * Assumes name is never RBD_SNAP_HEAD_NAME; returns CEPH_NOSNAP if
4111 * no snapshot by that name is found, or if an error occurs.
4113 static u64
rbd_snap_id_by_name(struct rbd_device
*rbd_dev
, const char *name
)
4115 if (rbd_dev
->image_format
== 1)
4116 return rbd_v1_snap_id_by_name(rbd_dev
, name
);
4118 return rbd_v2_snap_id_by_name(rbd_dev
, name
);
4122 * When an rbd image has a parent image, it is identified by the
4123 * pool, image, and snapshot ids (not names). This function fills
4124 * in the names for those ids. (It's OK if we can't figure out the
4125 * name for an image id, but the pool and snapshot ids should always
4126 * exist and have names.) All names in an rbd spec are dynamically
4129 * When an image being mapped (not a parent) is probed, we have the
4130 * pool name and pool id, image name and image id, and the snapshot
4131 * name. The only thing we're missing is the snapshot id.
4133 static int rbd_dev_spec_update(struct rbd_device
*rbd_dev
)
4135 struct ceph_osd_client
*osdc
= &rbd_dev
->rbd_client
->client
->osdc
;
4136 struct rbd_spec
*spec
= rbd_dev
->spec
;
4137 const char *pool_name
;
4138 const char *image_name
;
4139 const char *snap_name
;
4143 * An image being mapped will have the pool name (etc.), but
4144 * we need to look up the snapshot id.
4146 if (spec
->pool_name
) {
4147 if (strcmp(spec
->snap_name
, RBD_SNAP_HEAD_NAME
)) {
4150 snap_id
= rbd_snap_id_by_name(rbd_dev
, spec
->snap_name
);
4151 if (snap_id
== CEPH_NOSNAP
)
4153 spec
->snap_id
= snap_id
;
4155 spec
->snap_id
= CEPH_NOSNAP
;
4161 /* Get the pool name; we have to make our own copy of this */
4163 pool_name
= ceph_pg_pool_name_by_id(osdc
->osdmap
, spec
->pool_id
);
4165 rbd_warn(rbd_dev
, "no pool with id %llu", spec
->pool_id
);
4168 pool_name
= kstrdup(pool_name
, GFP_KERNEL
);
4172 /* Fetch the image name; tolerate failure here */
4174 image_name
= rbd_dev_image_name(rbd_dev
);
4176 rbd_warn(rbd_dev
, "unable to get image name");
4178 /* Look up the snapshot name, and make a copy */
4180 snap_name
= rbd_snap_name(rbd_dev
, spec
->snap_id
);
4181 if (IS_ERR(snap_name
)) {
4182 ret
= PTR_ERR(snap_name
);
4186 spec
->pool_name
= pool_name
;
4187 spec
->image_name
= image_name
;
4188 spec
->snap_name
= snap_name
;
4198 static int rbd_dev_v2_snap_context(struct rbd_device
*rbd_dev
)
4207 struct ceph_snap_context
*snapc
;
4211 * We'll need room for the seq value (maximum snapshot id),
4212 * snapshot count, and array of that many snapshot ids.
4213 * For now we have a fixed upper limit on the number we're
4214 * prepared to receive.
4216 size
= sizeof (__le64
) + sizeof (__le32
) +
4217 RBD_MAX_SNAP_COUNT
* sizeof (__le64
);
4218 reply_buf
= kzalloc(size
, GFP_KERNEL
);
4222 ret
= rbd_obj_method_sync(rbd_dev
, rbd_dev
->header_name
,
4223 "rbd", "get_snapcontext", NULL
, 0,
4225 dout("%s: rbd_obj_method_sync returned %d\n", __func__
, ret
);
4230 end
= reply_buf
+ ret
;
4232 ceph_decode_64_safe(&p
, end
, seq
, out
);
4233 ceph_decode_32_safe(&p
, end
, snap_count
, out
);
4236 * Make sure the reported number of snapshot ids wouldn't go
4237 * beyond the end of our buffer. But before checking that,
4238 * make sure the computed size of the snapshot context we
4239 * allocate is representable in a size_t.
4241 if (snap_count
> (SIZE_MAX
- sizeof (struct ceph_snap_context
))
4246 if (!ceph_has_room(&p
, end
, snap_count
* sizeof (__le64
)))
4250 snapc
= ceph_create_snap_context(snap_count
, GFP_KERNEL
);
4256 for (i
= 0; i
< snap_count
; i
++)
4257 snapc
->snaps
[i
] = ceph_decode_64(&p
);
4259 ceph_put_snap_context(rbd_dev
->header
.snapc
);
4260 rbd_dev
->header
.snapc
= snapc
;
4262 dout(" snap context seq = %llu, snap_count = %u\n",
4263 (unsigned long long)seq
, (unsigned int)snap_count
);
4270 static const char *rbd_dev_v2_snap_name(struct rbd_device
*rbd_dev
,
4281 size
= sizeof (__le32
) + RBD_MAX_SNAP_NAME_LEN
;
4282 reply_buf
= kmalloc(size
, GFP_KERNEL
);
4284 return ERR_PTR(-ENOMEM
);
4286 snapid
= cpu_to_le64(snap_id
);
4287 ret
= rbd_obj_method_sync(rbd_dev
, rbd_dev
->header_name
,
4288 "rbd", "get_snapshot_name",
4289 &snapid
, sizeof (snapid
),
4291 dout("%s: rbd_obj_method_sync returned %d\n", __func__
, ret
);
4293 snap_name
= ERR_PTR(ret
);
4298 end
= reply_buf
+ ret
;
4299 snap_name
= ceph_extract_encoded_string(&p
, end
, NULL
, GFP_KERNEL
);
4300 if (IS_ERR(snap_name
))
4303 dout(" snap_id 0x%016llx snap_name = %s\n",
4304 (unsigned long long)snap_id
, snap_name
);
4311 static int rbd_dev_v2_header_info(struct rbd_device
*rbd_dev
)
4313 bool first_time
= rbd_dev
->header
.object_prefix
== NULL
;
4316 down_write(&rbd_dev
->header_rwsem
);
4318 ret
= rbd_dev_v2_image_size(rbd_dev
);
4323 ret
= rbd_dev_v2_header_onetime(rbd_dev
);
4329 * If the image supports layering, get the parent info. We
4330 * need to probe the first time regardless. Thereafter we
4331 * only need to if there's a parent, to see if it has
4332 * disappeared due to the mapped image getting flattened.
4334 if (rbd_dev
->header
.features
& RBD_FEATURE_LAYERING
&&
4335 (first_time
|| rbd_dev
->parent_spec
)) {
4338 ret
= rbd_dev_v2_parent_info(rbd_dev
);
4343 * Print a warning if this is the initial probe and
4344 * the image has a parent. Don't print it if the
4345 * image now being probed is itself a parent. We
4346 * can tell at this point because we won't know its
4347 * pool name yet (just its pool id).
4349 warn
= rbd_dev
->parent_spec
&& rbd_dev
->spec
->pool_name
;
4350 if (first_time
&& warn
)
4351 rbd_warn(rbd_dev
, "WARNING: kernel layering "
4352 "is EXPERIMENTAL!");
4355 if (rbd_dev
->spec
->snap_id
== CEPH_NOSNAP
)
4356 if (rbd_dev
->mapping
.size
!= rbd_dev
->header
.image_size
)
4357 rbd_dev
->mapping
.size
= rbd_dev
->header
.image_size
;
4359 ret
= rbd_dev_v2_snap_context(rbd_dev
);
4360 dout("rbd_dev_v2_snap_context returned %d\n", ret
);
4362 up_write(&rbd_dev
->header_rwsem
);
4367 static int rbd_bus_add_dev(struct rbd_device
*rbd_dev
)
4372 mutex_lock_nested(&ctl_mutex
, SINGLE_DEPTH_NESTING
);
4374 dev
= &rbd_dev
->dev
;
4375 dev
->bus
= &rbd_bus_type
;
4376 dev
->type
= &rbd_device_type
;
4377 dev
->parent
= &rbd_root_dev
;
4378 dev
->release
= rbd_dev_device_release
;
4379 dev_set_name(dev
, "%d", rbd_dev
->dev_id
);
4380 ret
= device_register(dev
);
4382 mutex_unlock(&ctl_mutex
);
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 rbd_dev_destroy(parent
);
4868 rbd_put_client(rbdc
);
4869 rbd_spec_put(parent_spec
);
4875 static int rbd_dev_device_setup(struct rbd_device
*rbd_dev
)
4879 /* generate unique id: find highest unique id, add one */
4880 rbd_dev_id_get(rbd_dev
);
4882 /* Fill in the device name, now that we have its id. */
4883 BUILD_BUG_ON(DEV_NAME_LEN
4884 < sizeof (RBD_DRV_NAME
) + MAX_INT_FORMAT_WIDTH
);
4885 sprintf(rbd_dev
->name
, "%s%d", RBD_DRV_NAME
, rbd_dev
->dev_id
);
4887 /* Get our block major device number. */
4889 ret
= register_blkdev(0, rbd_dev
->name
);
4892 rbd_dev
->major
= ret
;
4894 /* Set up the blkdev mapping. */
4896 ret
= rbd_init_disk(rbd_dev
);
4898 goto err_out_blkdev
;
4900 ret
= rbd_dev_mapping_set(rbd_dev
);
4903 set_capacity(rbd_dev
->disk
, rbd_dev
->mapping
.size
/ SECTOR_SIZE
);
4905 ret
= rbd_bus_add_dev(rbd_dev
);
4907 goto err_out_mapping
;
4909 /* Everything's ready. Announce the disk to the world. */
4911 set_bit(RBD_DEV_FLAG_EXISTS
, &rbd_dev
->flags
);
4912 add_disk(rbd_dev
->disk
);
4914 pr_info("%s: added with size 0x%llx\n", rbd_dev
->disk
->disk_name
,
4915 (unsigned long long) rbd_dev
->mapping
.size
);
4920 rbd_dev_mapping_clear(rbd_dev
);
4922 rbd_free_disk(rbd_dev
);
4924 unregister_blkdev(rbd_dev
->major
, rbd_dev
->name
);
4926 rbd_dev_id_put(rbd_dev
);
4927 rbd_dev_mapping_clear(rbd_dev
);
4932 static int rbd_dev_header_name(struct rbd_device
*rbd_dev
)
4934 struct rbd_spec
*spec
= rbd_dev
->spec
;
4937 /* Record the header object name for this rbd image. */
4939 rbd_assert(rbd_image_format_valid(rbd_dev
->image_format
));
4941 if (rbd_dev
->image_format
== 1)
4942 size
= strlen(spec
->image_name
) + sizeof (RBD_SUFFIX
);
4944 size
= sizeof (RBD_HEADER_PREFIX
) + strlen(spec
->image_id
);
4946 rbd_dev
->header_name
= kmalloc(size
, GFP_KERNEL
);
4947 if (!rbd_dev
->header_name
)
4950 if (rbd_dev
->image_format
== 1)
4951 sprintf(rbd_dev
->header_name
, "%s%s",
4952 spec
->image_name
, RBD_SUFFIX
);
4954 sprintf(rbd_dev
->header_name
, "%s%s",
4955 RBD_HEADER_PREFIX
, spec
->image_id
);
4959 static void rbd_dev_image_release(struct rbd_device
*rbd_dev
)
4961 rbd_dev_unprobe(rbd_dev
);
4962 kfree(rbd_dev
->header_name
);
4963 rbd_dev
->header_name
= NULL
;
4964 rbd_dev
->image_format
= 0;
4965 kfree(rbd_dev
->spec
->image_id
);
4966 rbd_dev
->spec
->image_id
= NULL
;
4968 rbd_dev_destroy(rbd_dev
);
4972 * Probe for the existence of the header object for the given rbd
4973 * device. If this image is the one being mapped (i.e., not a
4974 * parent), initiate a watch on its header object before using that
4975 * object to get detailed information about the rbd image.
4977 static int rbd_dev_image_probe(struct rbd_device
*rbd_dev
, bool mapping
)
4983 * Get the id from the image id object. Unless there's an
4984 * error, rbd_dev->spec->image_id will be filled in with
4985 * a dynamically-allocated string, and rbd_dev->image_format
4986 * will be set to either 1 or 2.
4988 ret
= rbd_dev_image_id(rbd_dev
);
4991 rbd_assert(rbd_dev
->spec
->image_id
);
4992 rbd_assert(rbd_image_format_valid(rbd_dev
->image_format
));
4994 ret
= rbd_dev_header_name(rbd_dev
);
4996 goto err_out_format
;
4999 ret
= rbd_dev_header_watch_sync(rbd_dev
, true);
5001 goto out_header_name
;
5004 if (rbd_dev
->image_format
== 1)
5005 ret
= rbd_dev_v1_header_info(rbd_dev
);
5007 ret
= rbd_dev_v2_header_info(rbd_dev
);
5011 ret
= rbd_dev_spec_update(rbd_dev
);
5015 ret
= rbd_dev_probe_parent(rbd_dev
);
5019 dout("discovered format %u image, header name is %s\n",
5020 rbd_dev
->image_format
, rbd_dev
->header_name
);
5024 rbd_dev_unprobe(rbd_dev
);
5027 tmp
= rbd_dev_header_watch_sync(rbd_dev
, false);
5029 rbd_warn(rbd_dev
, "unable to tear down "
5030 "watch request (%d)\n", tmp
);
5033 kfree(rbd_dev
->header_name
);
5034 rbd_dev
->header_name
= NULL
;
5036 rbd_dev
->image_format
= 0;
5037 kfree(rbd_dev
->spec
->image_id
);
5038 rbd_dev
->spec
->image_id
= NULL
;
5040 dout("probe failed, returning %d\n", ret
);
5045 static ssize_t
rbd_add(struct bus_type
*bus
,
5049 struct rbd_device
*rbd_dev
= NULL
;
5050 struct ceph_options
*ceph_opts
= NULL
;
5051 struct rbd_options
*rbd_opts
= NULL
;
5052 struct rbd_spec
*spec
= NULL
;
5053 struct rbd_client
*rbdc
;
5054 struct ceph_osd_client
*osdc
;
5058 if (!try_module_get(THIS_MODULE
))
5061 /* parse add command */
5062 rc
= rbd_add_parse_args(buf
, &ceph_opts
, &rbd_opts
, &spec
);
5064 goto err_out_module
;
5065 read_only
= rbd_opts
->read_only
;
5067 rbd_opts
= NULL
; /* done with this */
5069 rbdc
= rbd_get_client(ceph_opts
);
5076 osdc
= &rbdc
->client
->osdc
;
5077 rc
= ceph_pg_poolid_by_name(osdc
->osdmap
, spec
->pool_name
);
5079 goto err_out_client
;
5080 spec
->pool_id
= (u64
)rc
;
5082 /* The ceph file layout needs to fit pool id in 32 bits */
5084 if (spec
->pool_id
> (u64
)U32_MAX
) {
5085 rbd_warn(NULL
, "pool id too large (%llu > %u)\n",
5086 (unsigned long long)spec
->pool_id
, U32_MAX
);
5088 goto err_out_client
;
5091 rbd_dev
= rbd_dev_create(rbdc
, spec
);
5093 goto err_out_client
;
5094 rbdc
= NULL
; /* rbd_dev now owns this */
5095 spec
= NULL
; /* rbd_dev now owns this */
5097 rc
= rbd_dev_image_probe(rbd_dev
, true);
5099 goto err_out_rbd_dev
;
5101 /* If we are mapping a snapshot it must be marked read-only */
5103 if (rbd_dev
->spec
->snap_id
!= CEPH_NOSNAP
)
5105 rbd_dev
->mapping
.read_only
= read_only
;
5107 rc
= rbd_dev_device_setup(rbd_dev
);
5109 rbd_dev_image_release(rbd_dev
);
5110 goto err_out_module
;
5116 rbd_dev_destroy(rbd_dev
);
5118 rbd_put_client(rbdc
);
5122 module_put(THIS_MODULE
);
5124 dout("Error adding device %s\n", buf
);
5129 static void rbd_dev_device_release(struct device
*dev
)
5131 struct rbd_device
*rbd_dev
= dev_to_rbd_dev(dev
);
5133 rbd_free_disk(rbd_dev
);
5134 clear_bit(RBD_DEV_FLAG_EXISTS
, &rbd_dev
->flags
);
5135 rbd_dev_mapping_clear(rbd_dev
);
5136 unregister_blkdev(rbd_dev
->major
, rbd_dev
->name
);
5138 rbd_dev_id_put(rbd_dev
);
5139 rbd_dev_mapping_clear(rbd_dev
);
5142 static void rbd_dev_remove_parent(struct rbd_device
*rbd_dev
)
5144 while (rbd_dev
->parent
) {
5145 struct rbd_device
*first
= rbd_dev
;
5146 struct rbd_device
*second
= first
->parent
;
5147 struct rbd_device
*third
;
5150 * Follow to the parent with no grandparent and
5153 while (second
&& (third
= second
->parent
)) {
5158 rbd_dev_image_release(second
);
5159 first
->parent
= NULL
;
5160 first
->parent_overlap
= 0;
5162 rbd_assert(first
->parent_spec
);
5163 rbd_spec_put(first
->parent_spec
);
5164 first
->parent_spec
= NULL
;
5168 static ssize_t
rbd_remove(struct bus_type
*bus
,
5172 struct rbd_device
*rbd_dev
= NULL
;
5173 struct list_head
*tmp
;
5176 bool already
= false;
5179 ret
= strict_strtoul(buf
, 10, &ul
);
5183 /* convert to int; abort if we lost anything in the conversion */
5188 mutex_lock_nested(&ctl_mutex
, SINGLE_DEPTH_NESTING
);
5191 spin_lock(&rbd_dev_list_lock
);
5192 list_for_each(tmp
, &rbd_dev_list
) {
5193 rbd_dev
= list_entry(tmp
, struct rbd_device
, node
);
5194 if (rbd_dev
->dev_id
== dev_id
) {
5200 spin_lock_irq(&rbd_dev
->lock
);
5201 if (rbd_dev
->open_count
)
5204 already
= test_and_set_bit(RBD_DEV_FLAG_REMOVING
,
5206 spin_unlock_irq(&rbd_dev
->lock
);
5208 spin_unlock(&rbd_dev_list_lock
);
5209 if (ret
< 0 || already
)
5212 ret
= rbd_dev_header_watch_sync(rbd_dev
, false);
5214 rbd_warn(rbd_dev
, "failed to cancel watch event (%d)\n", ret
);
5217 * flush remaining watch callbacks - these must be complete
5218 * before the osd_client is shutdown
5220 dout("%s: flushing notifies", __func__
);
5221 ceph_osdc_flush_notifies(&rbd_dev
->rbd_client
->client
->osdc
);
5223 * Don't free anything from rbd_dev->disk until after all
5224 * notifies are completely processed. Otherwise
5225 * rbd_bus_del_dev() will race with rbd_watch_cb(), resulting
5226 * in a potential use after free of rbd_dev->disk or rbd_dev.
5228 rbd_bus_del_dev(rbd_dev
);
5229 rbd_dev_image_release(rbd_dev
);
5230 module_put(THIS_MODULE
);
5233 mutex_unlock(&ctl_mutex
);
5239 * create control files in sysfs
5242 static int rbd_sysfs_init(void)
5246 ret
= device_register(&rbd_root_dev
);
5250 ret
= bus_register(&rbd_bus_type
);
5252 device_unregister(&rbd_root_dev
);
5257 static void rbd_sysfs_cleanup(void)
5259 bus_unregister(&rbd_bus_type
);
5260 device_unregister(&rbd_root_dev
);
5263 static int rbd_slab_init(void)
5265 rbd_assert(!rbd_img_request_cache
);
5266 rbd_img_request_cache
= kmem_cache_create("rbd_img_request",
5267 sizeof (struct rbd_img_request
),
5268 __alignof__(struct rbd_img_request
),
5270 if (!rbd_img_request_cache
)
5273 rbd_assert(!rbd_obj_request_cache
);
5274 rbd_obj_request_cache
= kmem_cache_create("rbd_obj_request",
5275 sizeof (struct rbd_obj_request
),
5276 __alignof__(struct rbd_obj_request
),
5278 if (!rbd_obj_request_cache
)
5281 rbd_assert(!rbd_segment_name_cache
);
5282 rbd_segment_name_cache
= kmem_cache_create("rbd_segment_name",
5283 MAX_OBJ_NAME_SIZE
+ 1, 1, 0, NULL
);
5284 if (rbd_segment_name_cache
)
5287 if (rbd_obj_request_cache
) {
5288 kmem_cache_destroy(rbd_obj_request_cache
);
5289 rbd_obj_request_cache
= NULL
;
5292 kmem_cache_destroy(rbd_img_request_cache
);
5293 rbd_img_request_cache
= NULL
;
5298 static void rbd_slab_exit(void)
5300 rbd_assert(rbd_segment_name_cache
);
5301 kmem_cache_destroy(rbd_segment_name_cache
);
5302 rbd_segment_name_cache
= NULL
;
5304 rbd_assert(rbd_obj_request_cache
);
5305 kmem_cache_destroy(rbd_obj_request_cache
);
5306 rbd_obj_request_cache
= NULL
;
5308 rbd_assert(rbd_img_request_cache
);
5309 kmem_cache_destroy(rbd_img_request_cache
);
5310 rbd_img_request_cache
= NULL
;
5313 static int __init
rbd_init(void)
5317 if (!libceph_compatible(NULL
)) {
5318 rbd_warn(NULL
, "libceph incompatibility (quitting)");
5322 rc
= rbd_slab_init();
5325 rc
= rbd_sysfs_init();
5329 pr_info("loaded " RBD_DRV_NAME_LONG
"\n");
5334 static void __exit
rbd_exit(void)
5336 rbd_sysfs_cleanup();
5340 module_init(rbd_init
);
5341 module_exit(rbd_exit
);
5343 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
5344 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
5345 MODULE_DESCRIPTION("rados block device");
5347 /* following authorship retained from original osdblk.c */
5348 MODULE_AUTHOR("Jeff Garzik <jeff@garzik.org>");
5350 MODULE_LICENSE("GPL");