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>
44 #include <linux/idr.h>
45 #include <linux/workqueue.h>
47 #include "rbd_types.h"
49 #define RBD_DEBUG /* Activate rbd_assert() calls */
52 * The basic unit of block I/O is a sector. It is interpreted in a
53 * number of contexts in Linux (blk, bio, genhd), but the default is
54 * universally 512 bytes. These symbols are just slightly more
55 * meaningful than the bare numbers they represent.
57 #define SECTOR_SHIFT 9
58 #define SECTOR_SIZE (1ULL << SECTOR_SHIFT)
61 * Increment the given counter and return its updated value.
62 * If the counter is already 0 it will not be incremented.
63 * If the counter is already at its maximum value returns
64 * -EINVAL without updating it.
66 static int atomic_inc_return_safe(atomic_t
*v
)
70 counter
= (unsigned int)__atomic_add_unless(v
, 1, 0);
71 if (counter
<= (unsigned int)INT_MAX
)
79 /* Decrement the counter. Return the resulting value, or -EINVAL */
80 static int atomic_dec_return_safe(atomic_t
*v
)
84 counter
= atomic_dec_return(v
);
93 #define RBD_DRV_NAME "rbd"
95 #define RBD_MINORS_PER_MAJOR 256
96 #define RBD_SINGLE_MAJOR_PART_SHIFT 4
98 #define RBD_SNAP_DEV_NAME_PREFIX "snap_"
99 #define RBD_MAX_SNAP_NAME_LEN \
100 (NAME_MAX - (sizeof (RBD_SNAP_DEV_NAME_PREFIX) - 1))
102 #define RBD_MAX_SNAP_COUNT 510 /* allows max snapc to fit in 4KB */
104 #define RBD_SNAP_HEAD_NAME "-"
106 #define BAD_SNAP_INDEX U32_MAX /* invalid index into snap array */
108 /* This allows a single page to hold an image name sent by OSD */
109 #define RBD_IMAGE_NAME_LEN_MAX (PAGE_SIZE - sizeof (__le32) - 1)
110 #define RBD_IMAGE_ID_LEN_MAX 64
112 #define RBD_OBJ_PREFIX_LEN_MAX 64
116 #define RBD_FEATURE_LAYERING (1<<0)
117 #define RBD_FEATURE_STRIPINGV2 (1<<1)
118 #define RBD_FEATURES_ALL \
119 (RBD_FEATURE_LAYERING | RBD_FEATURE_STRIPINGV2)
121 /* Features supported by this (client software) implementation. */
123 #define RBD_FEATURES_SUPPORTED (RBD_FEATURES_ALL)
126 * An RBD device name will be "rbd#", where the "rbd" comes from
127 * RBD_DRV_NAME above, and # is a unique integer identifier.
128 * MAX_INT_FORMAT_WIDTH is used in ensuring DEV_NAME_LEN is big
129 * enough to hold all possible device names.
131 #define DEV_NAME_LEN 32
132 #define MAX_INT_FORMAT_WIDTH ((5 * sizeof (int)) / 2 + 1)
135 * block device image metadata (in-memory version)
137 struct rbd_image_header
{
138 /* These six fields never change for a given rbd image */
145 u64 features
; /* Might be changeable someday? */
147 /* The remaining fields need to be updated occasionally */
149 struct ceph_snap_context
*snapc
;
150 char *snap_names
; /* format 1 only */
151 u64
*snap_sizes
; /* format 1 only */
155 * An rbd image specification.
157 * The tuple (pool_id, image_id, snap_id) is sufficient to uniquely
158 * identify an image. Each rbd_dev structure includes a pointer to
159 * an rbd_spec structure that encapsulates this identity.
161 * Each of the id's in an rbd_spec has an associated name. For a
162 * user-mapped image, the names are supplied and the id's associated
163 * with them are looked up. For a layered image, a parent image is
164 * defined by the tuple, and the names are looked up.
166 * An rbd_dev structure contains a parent_spec pointer which is
167 * non-null if the image it represents is a child in a layered
168 * image. This pointer will refer to the rbd_spec structure used
169 * by the parent rbd_dev for its own identity (i.e., the structure
170 * is shared between the parent and child).
172 * Since these structures are populated once, during the discovery
173 * phase of image construction, they are effectively immutable so
174 * we make no effort to synchronize access to them.
176 * Note that code herein does not assume the image name is known (it
177 * could be a null pointer).
181 const char *pool_name
;
183 const char *image_id
;
184 const char *image_name
;
187 const char *snap_name
;
193 * an instance of the client. multiple devices may share an rbd client.
196 struct ceph_client
*client
;
198 struct list_head node
;
201 struct rbd_img_request
;
202 typedef void (*rbd_img_callback_t
)(struct rbd_img_request
*);
204 #define BAD_WHICH U32_MAX /* Good which or bad which, which? */
206 struct rbd_obj_request
;
207 typedef void (*rbd_obj_callback_t
)(struct rbd_obj_request
*);
209 enum obj_request_type
{
210 OBJ_REQUEST_NODATA
, OBJ_REQUEST_BIO
, OBJ_REQUEST_PAGES
213 enum obj_operation_type
{
220 OBJ_REQ_DONE
, /* completion flag: not done = 0, done = 1 */
221 OBJ_REQ_IMG_DATA
, /* object usage: standalone = 0, image = 1 */
222 OBJ_REQ_KNOWN
, /* EXISTS flag valid: no = 0, yes = 1 */
223 OBJ_REQ_EXISTS
, /* target exists: no = 0, yes = 1 */
226 struct rbd_obj_request
{
227 const char *object_name
;
228 u64 offset
; /* object start byte */
229 u64 length
; /* bytes from offset */
233 * An object request associated with an image will have its
234 * img_data flag set; a standalone object request will not.
236 * A standalone object request will have which == BAD_WHICH
237 * and a null obj_request pointer.
239 * An object request initiated in support of a layered image
240 * object (to check for its existence before a write) will
241 * have which == BAD_WHICH and a non-null obj_request pointer.
243 * Finally, an object request for rbd image data will have
244 * which != BAD_WHICH, and will have a non-null img_request
245 * pointer. The value of which will be in the range
246 * 0..(img_request->obj_request_count-1).
249 struct rbd_obj_request
*obj_request
; /* STAT op */
251 struct rbd_img_request
*img_request
;
253 /* links for img_request->obj_requests list */
254 struct list_head links
;
257 u32 which
; /* posn image request list */
259 enum obj_request_type type
;
261 struct bio
*bio_list
;
267 struct page
**copyup_pages
;
268 u32 copyup_page_count
;
270 struct ceph_osd_request
*osd_req
;
272 u64 xferred
; /* bytes transferred */
275 rbd_obj_callback_t callback
;
276 struct completion completion
;
282 IMG_REQ_WRITE
, /* I/O direction: read = 0, write = 1 */
283 IMG_REQ_CHILD
, /* initiator: block = 0, child image = 1 */
284 IMG_REQ_LAYERED
, /* ENOENT handling: normal = 0, layered = 1 */
285 IMG_REQ_DISCARD
, /* discard: normal = 0, discard request = 1 */
288 struct rbd_img_request
{
289 struct rbd_device
*rbd_dev
;
290 u64 offset
; /* starting image byte offset */
291 u64 length
; /* byte count from offset */
294 u64 snap_id
; /* for reads */
295 struct ceph_snap_context
*snapc
; /* for writes */
298 struct request
*rq
; /* block request */
299 struct rbd_obj_request
*obj_request
; /* obj req initiator */
301 struct page
**copyup_pages
;
302 u32 copyup_page_count
;
303 spinlock_t completion_lock
;/* protects next_completion */
305 rbd_img_callback_t callback
;
306 u64 xferred
;/* aggregate bytes transferred */
307 int result
; /* first nonzero obj_request result */
309 u32 obj_request_count
;
310 struct list_head obj_requests
; /* rbd_obj_request structs */
315 #define for_each_obj_request(ireq, oreq) \
316 list_for_each_entry(oreq, &(ireq)->obj_requests, links)
317 #define for_each_obj_request_from(ireq, oreq) \
318 list_for_each_entry_from(oreq, &(ireq)->obj_requests, links)
319 #define for_each_obj_request_safe(ireq, oreq, n) \
320 list_for_each_entry_safe_reverse(oreq, n, &(ireq)->obj_requests, links)
332 int dev_id
; /* blkdev unique id */
334 int major
; /* blkdev assigned major */
336 struct gendisk
*disk
; /* blkdev's gendisk and rq */
338 u32 image_format
; /* Either 1 or 2 */
339 struct rbd_client
*rbd_client
;
341 char name
[DEV_NAME_LEN
]; /* blkdev name, e.g. rbd3 */
343 struct list_head rq_queue
; /* incoming rq queue */
344 spinlock_t lock
; /* queue, flags, open_count */
345 struct work_struct rq_work
;
347 struct rbd_image_header header
;
348 unsigned long flags
; /* possibly lock protected */
349 struct rbd_spec
*spec
;
353 struct ceph_file_layout layout
;
355 struct ceph_osd_event
*watch_event
;
356 struct rbd_obj_request
*watch_request
;
358 struct rbd_spec
*parent_spec
;
361 struct rbd_device
*parent
;
363 /* protects updating the header */
364 struct rw_semaphore header_rwsem
;
366 struct rbd_mapping mapping
;
368 struct list_head node
;
372 unsigned long open_count
; /* protected by lock */
376 * Flag bits for rbd_dev->flags. If atomicity is required,
377 * rbd_dev->lock is used to protect access.
379 * Currently, only the "removing" flag (which is coupled with the
380 * "open_count" field) requires atomic access.
383 RBD_DEV_FLAG_EXISTS
, /* mapped snapshot has not been deleted */
384 RBD_DEV_FLAG_REMOVING
, /* this mapping is being removed */
387 static DEFINE_MUTEX(client_mutex
); /* Serialize client creation */
389 static LIST_HEAD(rbd_dev_list
); /* devices */
390 static DEFINE_SPINLOCK(rbd_dev_list_lock
);
392 static LIST_HEAD(rbd_client_list
); /* clients */
393 static DEFINE_SPINLOCK(rbd_client_list_lock
);
395 /* Slab caches for frequently-allocated structures */
397 static struct kmem_cache
*rbd_img_request_cache
;
398 static struct kmem_cache
*rbd_obj_request_cache
;
399 static struct kmem_cache
*rbd_segment_name_cache
;
401 static int rbd_major
;
402 static DEFINE_IDA(rbd_dev_id_ida
);
404 static struct workqueue_struct
*rbd_wq
;
407 * Default to false for now, as single-major requires >= 0.75 version of
408 * userspace rbd utility.
410 static bool single_major
= false;
411 module_param(single_major
, bool, S_IRUGO
);
412 MODULE_PARM_DESC(single_major
, "Use a single major number for all rbd devices (default: false)");
414 static int rbd_img_request_submit(struct rbd_img_request
*img_request
);
416 static void rbd_dev_device_release(struct device
*dev
);
418 static ssize_t
rbd_add(struct bus_type
*bus
, const char *buf
,
420 static ssize_t
rbd_remove(struct bus_type
*bus
, const char *buf
,
422 static ssize_t
rbd_add_single_major(struct bus_type
*bus
, const char *buf
,
424 static ssize_t
rbd_remove_single_major(struct bus_type
*bus
, const char *buf
,
426 static int rbd_dev_image_probe(struct rbd_device
*rbd_dev
, bool mapping
);
427 static void rbd_spec_put(struct rbd_spec
*spec
);
429 static int rbd_dev_id_to_minor(int dev_id
)
431 return dev_id
<< RBD_SINGLE_MAJOR_PART_SHIFT
;
434 static int minor_to_rbd_dev_id(int minor
)
436 return minor
>> RBD_SINGLE_MAJOR_PART_SHIFT
;
439 static BUS_ATTR(add
, S_IWUSR
, NULL
, rbd_add
);
440 static BUS_ATTR(remove
, S_IWUSR
, NULL
, rbd_remove
);
441 static BUS_ATTR(add_single_major
, S_IWUSR
, NULL
, rbd_add_single_major
);
442 static BUS_ATTR(remove_single_major
, S_IWUSR
, NULL
, rbd_remove_single_major
);
444 static struct attribute
*rbd_bus_attrs
[] = {
446 &bus_attr_remove
.attr
,
447 &bus_attr_add_single_major
.attr
,
448 &bus_attr_remove_single_major
.attr
,
452 static umode_t
rbd_bus_is_visible(struct kobject
*kobj
,
453 struct attribute
*attr
, int index
)
456 (attr
== &bus_attr_add_single_major
.attr
||
457 attr
== &bus_attr_remove_single_major
.attr
))
463 static const struct attribute_group rbd_bus_group
= {
464 .attrs
= rbd_bus_attrs
,
465 .is_visible
= rbd_bus_is_visible
,
467 __ATTRIBUTE_GROUPS(rbd_bus
);
469 static struct bus_type rbd_bus_type
= {
471 .bus_groups
= rbd_bus_groups
,
474 static void rbd_root_dev_release(struct device
*dev
)
478 static struct device rbd_root_dev
= {
480 .release
= rbd_root_dev_release
,
483 static __printf(2, 3)
484 void rbd_warn(struct rbd_device
*rbd_dev
, const char *fmt
, ...)
486 struct va_format vaf
;
494 printk(KERN_WARNING
"%s: %pV\n", RBD_DRV_NAME
, &vaf
);
495 else if (rbd_dev
->disk
)
496 printk(KERN_WARNING
"%s: %s: %pV\n",
497 RBD_DRV_NAME
, rbd_dev
->disk
->disk_name
, &vaf
);
498 else if (rbd_dev
->spec
&& rbd_dev
->spec
->image_name
)
499 printk(KERN_WARNING
"%s: image %s: %pV\n",
500 RBD_DRV_NAME
, rbd_dev
->spec
->image_name
, &vaf
);
501 else if (rbd_dev
->spec
&& rbd_dev
->spec
->image_id
)
502 printk(KERN_WARNING
"%s: id %s: %pV\n",
503 RBD_DRV_NAME
, rbd_dev
->spec
->image_id
, &vaf
);
505 printk(KERN_WARNING
"%s: rbd_dev %p: %pV\n",
506 RBD_DRV_NAME
, rbd_dev
, &vaf
);
511 #define rbd_assert(expr) \
512 if (unlikely(!(expr))) { \
513 printk(KERN_ERR "\nAssertion failure in %s() " \
515 "\trbd_assert(%s);\n\n", \
516 __func__, __LINE__, #expr); \
519 #else /* !RBD_DEBUG */
520 # define rbd_assert(expr) ((void) 0)
521 #endif /* !RBD_DEBUG */
523 static int rbd_img_obj_request_submit(struct rbd_obj_request
*obj_request
);
524 static void rbd_img_parent_read(struct rbd_obj_request
*obj_request
);
525 static void rbd_dev_remove_parent(struct rbd_device
*rbd_dev
);
527 static int rbd_dev_refresh(struct rbd_device
*rbd_dev
);
528 static int rbd_dev_v2_header_onetime(struct rbd_device
*rbd_dev
);
529 static int rbd_dev_header_info(struct rbd_device
*rbd_dev
);
530 static int rbd_dev_v2_parent_info(struct rbd_device
*rbd_dev
);
531 static const char *rbd_dev_v2_snap_name(struct rbd_device
*rbd_dev
,
533 static int _rbd_dev_v2_snap_size(struct rbd_device
*rbd_dev
, u64 snap_id
,
534 u8
*order
, u64
*snap_size
);
535 static int _rbd_dev_v2_snap_features(struct rbd_device
*rbd_dev
, u64 snap_id
,
537 static u64
rbd_snap_id_by_name(struct rbd_device
*rbd_dev
, const char *name
);
539 static int rbd_open(struct block_device
*bdev
, fmode_t mode
)
541 struct rbd_device
*rbd_dev
= bdev
->bd_disk
->private_data
;
542 bool removing
= false;
544 if ((mode
& FMODE_WRITE
) && rbd_dev
->mapping
.read_only
)
547 spin_lock_irq(&rbd_dev
->lock
);
548 if (test_bit(RBD_DEV_FLAG_REMOVING
, &rbd_dev
->flags
))
551 rbd_dev
->open_count
++;
552 spin_unlock_irq(&rbd_dev
->lock
);
556 (void) get_device(&rbd_dev
->dev
);
561 static void rbd_release(struct gendisk
*disk
, fmode_t mode
)
563 struct rbd_device
*rbd_dev
= disk
->private_data
;
564 unsigned long open_count_before
;
566 spin_lock_irq(&rbd_dev
->lock
);
567 open_count_before
= rbd_dev
->open_count
--;
568 spin_unlock_irq(&rbd_dev
->lock
);
569 rbd_assert(open_count_before
> 0);
571 put_device(&rbd_dev
->dev
);
574 static int rbd_ioctl_set_ro(struct rbd_device
*rbd_dev
, unsigned long arg
)
579 bool ro_changed
= false;
581 /* get_user() may sleep, so call it before taking rbd_dev->lock */
582 if (get_user(val
, (int __user
*)(arg
)))
585 ro
= val
? true : false;
586 /* Snapshot doesn't allow to write*/
587 if (rbd_dev
->spec
->snap_id
!= CEPH_NOSNAP
&& !ro
)
590 spin_lock_irq(&rbd_dev
->lock
);
591 /* prevent others open this device */
592 if (rbd_dev
->open_count
> 1) {
597 if (rbd_dev
->mapping
.read_only
!= ro
) {
598 rbd_dev
->mapping
.read_only
= ro
;
603 spin_unlock_irq(&rbd_dev
->lock
);
604 /* set_disk_ro() may sleep, so call it after releasing rbd_dev->lock */
605 if (ret
== 0 && ro_changed
)
606 set_disk_ro(rbd_dev
->disk
, ro
? 1 : 0);
611 static int rbd_ioctl(struct block_device
*bdev
, fmode_t mode
,
612 unsigned int cmd
, unsigned long arg
)
614 struct rbd_device
*rbd_dev
= bdev
->bd_disk
->private_data
;
619 ret
= rbd_ioctl_set_ro(rbd_dev
, arg
);
629 static int rbd_compat_ioctl(struct block_device
*bdev
, fmode_t mode
,
630 unsigned int cmd
, unsigned long arg
)
632 return rbd_ioctl(bdev
, mode
, cmd
, arg
);
634 #endif /* CONFIG_COMPAT */
636 static const struct block_device_operations rbd_bd_ops
= {
637 .owner
= THIS_MODULE
,
639 .release
= rbd_release
,
642 .compat_ioctl
= rbd_compat_ioctl
,
647 * Initialize an rbd client instance. Success or not, this function
648 * consumes ceph_opts. Caller holds client_mutex.
650 static struct rbd_client
*rbd_client_create(struct ceph_options
*ceph_opts
)
652 struct rbd_client
*rbdc
;
655 dout("%s:\n", __func__
);
656 rbdc
= kmalloc(sizeof(struct rbd_client
), GFP_KERNEL
);
660 kref_init(&rbdc
->kref
);
661 INIT_LIST_HEAD(&rbdc
->node
);
663 rbdc
->client
= ceph_create_client(ceph_opts
, rbdc
, 0, 0);
664 if (IS_ERR(rbdc
->client
))
666 ceph_opts
= NULL
; /* Now rbdc->client is responsible for ceph_opts */
668 ret
= ceph_open_session(rbdc
->client
);
672 spin_lock(&rbd_client_list_lock
);
673 list_add_tail(&rbdc
->node
, &rbd_client_list
);
674 spin_unlock(&rbd_client_list_lock
);
676 dout("%s: rbdc %p\n", __func__
, rbdc
);
680 ceph_destroy_client(rbdc
->client
);
685 ceph_destroy_options(ceph_opts
);
686 dout("%s: error %d\n", __func__
, ret
);
691 static struct rbd_client
*__rbd_get_client(struct rbd_client
*rbdc
)
693 kref_get(&rbdc
->kref
);
699 * Find a ceph client with specific addr and configuration. If
700 * found, bump its reference count.
702 static struct rbd_client
*rbd_client_find(struct ceph_options
*ceph_opts
)
704 struct rbd_client
*client_node
;
707 if (ceph_opts
->flags
& CEPH_OPT_NOSHARE
)
710 spin_lock(&rbd_client_list_lock
);
711 list_for_each_entry(client_node
, &rbd_client_list
, node
) {
712 if (!ceph_compare_options(ceph_opts
, client_node
->client
)) {
713 __rbd_get_client(client_node
);
719 spin_unlock(&rbd_client_list_lock
);
721 return found
? client_node
: NULL
;
731 /* string args above */
734 /* Boolean args above */
738 static match_table_t rbd_opts_tokens
= {
740 /* string args above */
741 {Opt_read_only
, "read_only"},
742 {Opt_read_only
, "ro"}, /* Alternate spelling */
743 {Opt_read_write
, "read_write"},
744 {Opt_read_write
, "rw"}, /* Alternate spelling */
745 /* Boolean args above */
753 #define RBD_READ_ONLY_DEFAULT false
755 static int parse_rbd_opts_token(char *c
, void *private)
757 struct rbd_options
*rbd_opts
= private;
758 substring_t argstr
[MAX_OPT_ARGS
];
759 int token
, intval
, ret
;
761 token
= match_token(c
, rbd_opts_tokens
, argstr
);
765 if (token
< Opt_last_int
) {
766 ret
= match_int(&argstr
[0], &intval
);
768 pr_err("bad mount option arg (not int) "
772 dout("got int token %d val %d\n", token
, intval
);
773 } else if (token
> Opt_last_int
&& token
< Opt_last_string
) {
774 dout("got string token %d val %s\n", token
,
776 } else if (token
> Opt_last_string
&& token
< Opt_last_bool
) {
777 dout("got Boolean token %d\n", token
);
779 dout("got token %d\n", token
);
784 rbd_opts
->read_only
= true;
787 rbd_opts
->read_only
= false;
796 static char* obj_op_name(enum obj_operation_type op_type
)
811 * Get a ceph client with specific addr and configuration, if one does
812 * not exist create it. Either way, ceph_opts is consumed by this
815 static struct rbd_client
*rbd_get_client(struct ceph_options
*ceph_opts
)
817 struct rbd_client
*rbdc
;
819 mutex_lock_nested(&client_mutex
, SINGLE_DEPTH_NESTING
);
820 rbdc
= rbd_client_find(ceph_opts
);
821 if (rbdc
) /* using an existing client */
822 ceph_destroy_options(ceph_opts
);
824 rbdc
= rbd_client_create(ceph_opts
);
825 mutex_unlock(&client_mutex
);
831 * Destroy ceph client
833 * Caller must hold rbd_client_list_lock.
835 static void rbd_client_release(struct kref
*kref
)
837 struct rbd_client
*rbdc
= container_of(kref
, struct rbd_client
, kref
);
839 dout("%s: rbdc %p\n", __func__
, rbdc
);
840 spin_lock(&rbd_client_list_lock
);
841 list_del(&rbdc
->node
);
842 spin_unlock(&rbd_client_list_lock
);
844 ceph_destroy_client(rbdc
->client
);
849 * Drop reference to ceph client node. If it's not referenced anymore, release
852 static void rbd_put_client(struct rbd_client
*rbdc
)
855 kref_put(&rbdc
->kref
, rbd_client_release
);
858 static bool rbd_image_format_valid(u32 image_format
)
860 return image_format
== 1 || image_format
== 2;
863 static bool rbd_dev_ondisk_valid(struct rbd_image_header_ondisk
*ondisk
)
868 /* The header has to start with the magic rbd header text */
869 if (memcmp(&ondisk
->text
, RBD_HEADER_TEXT
, sizeof (RBD_HEADER_TEXT
)))
872 /* The bio layer requires at least sector-sized I/O */
874 if (ondisk
->options
.order
< SECTOR_SHIFT
)
877 /* If we use u64 in a few spots we may be able to loosen this */
879 if (ondisk
->options
.order
> 8 * sizeof (int) - 1)
883 * The size of a snapshot header has to fit in a size_t, and
884 * that limits the number of snapshots.
886 snap_count
= le32_to_cpu(ondisk
->snap_count
);
887 size
= SIZE_MAX
- sizeof (struct ceph_snap_context
);
888 if (snap_count
> size
/ sizeof (__le64
))
892 * Not only that, but the size of the entire the snapshot
893 * header must also be representable in a size_t.
895 size
-= snap_count
* sizeof (__le64
);
896 if ((u64
) size
< le64_to_cpu(ondisk
->snap_names_len
))
903 * Fill an rbd image header with information from the given format 1
906 static int rbd_header_from_disk(struct rbd_device
*rbd_dev
,
907 struct rbd_image_header_ondisk
*ondisk
)
909 struct rbd_image_header
*header
= &rbd_dev
->header
;
910 bool first_time
= header
->object_prefix
== NULL
;
911 struct ceph_snap_context
*snapc
;
912 char *object_prefix
= NULL
;
913 char *snap_names
= NULL
;
914 u64
*snap_sizes
= NULL
;
920 /* Allocate this now to avoid having to handle failure below */
925 len
= strnlen(ondisk
->object_prefix
,
926 sizeof (ondisk
->object_prefix
));
927 object_prefix
= kmalloc(len
+ 1, GFP_KERNEL
);
930 memcpy(object_prefix
, ondisk
->object_prefix
, len
);
931 object_prefix
[len
] = '\0';
934 /* Allocate the snapshot context and fill it in */
936 snap_count
= le32_to_cpu(ondisk
->snap_count
);
937 snapc
= ceph_create_snap_context(snap_count
, GFP_KERNEL
);
940 snapc
->seq
= le64_to_cpu(ondisk
->snap_seq
);
942 struct rbd_image_snap_ondisk
*snaps
;
943 u64 snap_names_len
= le64_to_cpu(ondisk
->snap_names_len
);
945 /* We'll keep a copy of the snapshot names... */
947 if (snap_names_len
> (u64
)SIZE_MAX
)
949 snap_names
= kmalloc(snap_names_len
, GFP_KERNEL
);
953 /* ...as well as the array of their sizes. */
955 size
= snap_count
* sizeof (*header
->snap_sizes
);
956 snap_sizes
= kmalloc(size
, GFP_KERNEL
);
961 * Copy the names, and fill in each snapshot's id
964 * Note that rbd_dev_v1_header_info() guarantees the
965 * ondisk buffer we're working with has
966 * snap_names_len bytes beyond the end of the
967 * snapshot id array, this memcpy() is safe.
969 memcpy(snap_names
, &ondisk
->snaps
[snap_count
], snap_names_len
);
970 snaps
= ondisk
->snaps
;
971 for (i
= 0; i
< snap_count
; i
++) {
972 snapc
->snaps
[i
] = le64_to_cpu(snaps
[i
].id
);
973 snap_sizes
[i
] = le64_to_cpu(snaps
[i
].image_size
);
977 /* We won't fail any more, fill in the header */
980 header
->object_prefix
= object_prefix
;
981 header
->obj_order
= ondisk
->options
.order
;
982 header
->crypt_type
= ondisk
->options
.crypt_type
;
983 header
->comp_type
= ondisk
->options
.comp_type
;
984 /* The rest aren't used for format 1 images */
985 header
->stripe_unit
= 0;
986 header
->stripe_count
= 0;
987 header
->features
= 0;
989 ceph_put_snap_context(header
->snapc
);
990 kfree(header
->snap_names
);
991 kfree(header
->snap_sizes
);
994 /* The remaining fields always get updated (when we refresh) */
996 header
->image_size
= le64_to_cpu(ondisk
->image_size
);
997 header
->snapc
= snapc
;
998 header
->snap_names
= snap_names
;
999 header
->snap_sizes
= snap_sizes
;
1007 ceph_put_snap_context(snapc
);
1008 kfree(object_prefix
);
1013 static const char *_rbd_dev_v1_snap_name(struct rbd_device
*rbd_dev
, u32 which
)
1015 const char *snap_name
;
1017 rbd_assert(which
< rbd_dev
->header
.snapc
->num_snaps
);
1019 /* Skip over names until we find the one we are looking for */
1021 snap_name
= rbd_dev
->header
.snap_names
;
1023 snap_name
+= strlen(snap_name
) + 1;
1025 return kstrdup(snap_name
, GFP_KERNEL
);
1029 * Snapshot id comparison function for use with qsort()/bsearch().
1030 * Note that result is for snapshots in *descending* order.
1032 static int snapid_compare_reverse(const void *s1
, const void *s2
)
1034 u64 snap_id1
= *(u64
*)s1
;
1035 u64 snap_id2
= *(u64
*)s2
;
1037 if (snap_id1
< snap_id2
)
1039 return snap_id1
== snap_id2
? 0 : -1;
1043 * Search a snapshot context to see if the given snapshot id is
1046 * Returns the position of the snapshot id in the array if it's found,
1047 * or BAD_SNAP_INDEX otherwise.
1049 * Note: The snapshot array is in kept sorted (by the osd) in
1050 * reverse order, highest snapshot id first.
1052 static u32
rbd_dev_snap_index(struct rbd_device
*rbd_dev
, u64 snap_id
)
1054 struct ceph_snap_context
*snapc
= rbd_dev
->header
.snapc
;
1057 found
= bsearch(&snap_id
, &snapc
->snaps
, snapc
->num_snaps
,
1058 sizeof (snap_id
), snapid_compare_reverse
);
1060 return found
? (u32
)(found
- &snapc
->snaps
[0]) : BAD_SNAP_INDEX
;
1063 static const char *rbd_dev_v1_snap_name(struct rbd_device
*rbd_dev
,
1067 const char *snap_name
;
1069 which
= rbd_dev_snap_index(rbd_dev
, snap_id
);
1070 if (which
== BAD_SNAP_INDEX
)
1071 return ERR_PTR(-ENOENT
);
1073 snap_name
= _rbd_dev_v1_snap_name(rbd_dev
, which
);
1074 return snap_name
? snap_name
: ERR_PTR(-ENOMEM
);
1077 static const char *rbd_snap_name(struct rbd_device
*rbd_dev
, u64 snap_id
)
1079 if (snap_id
== CEPH_NOSNAP
)
1080 return RBD_SNAP_HEAD_NAME
;
1082 rbd_assert(rbd_image_format_valid(rbd_dev
->image_format
));
1083 if (rbd_dev
->image_format
== 1)
1084 return rbd_dev_v1_snap_name(rbd_dev
, snap_id
);
1086 return rbd_dev_v2_snap_name(rbd_dev
, snap_id
);
1089 static int rbd_snap_size(struct rbd_device
*rbd_dev
, u64 snap_id
,
1092 rbd_assert(rbd_image_format_valid(rbd_dev
->image_format
));
1093 if (snap_id
== CEPH_NOSNAP
) {
1094 *snap_size
= rbd_dev
->header
.image_size
;
1095 } else if (rbd_dev
->image_format
== 1) {
1098 which
= rbd_dev_snap_index(rbd_dev
, snap_id
);
1099 if (which
== BAD_SNAP_INDEX
)
1102 *snap_size
= rbd_dev
->header
.snap_sizes
[which
];
1107 ret
= _rbd_dev_v2_snap_size(rbd_dev
, snap_id
, NULL
, &size
);
1116 static int rbd_snap_features(struct rbd_device
*rbd_dev
, u64 snap_id
,
1119 rbd_assert(rbd_image_format_valid(rbd_dev
->image_format
));
1120 if (snap_id
== CEPH_NOSNAP
) {
1121 *snap_features
= rbd_dev
->header
.features
;
1122 } else if (rbd_dev
->image_format
== 1) {
1123 *snap_features
= 0; /* No features for format 1 */
1128 ret
= _rbd_dev_v2_snap_features(rbd_dev
, snap_id
, &features
);
1132 *snap_features
= features
;
1137 static int rbd_dev_mapping_set(struct rbd_device
*rbd_dev
)
1139 u64 snap_id
= rbd_dev
->spec
->snap_id
;
1144 ret
= rbd_snap_size(rbd_dev
, snap_id
, &size
);
1147 ret
= rbd_snap_features(rbd_dev
, snap_id
, &features
);
1151 rbd_dev
->mapping
.size
= size
;
1152 rbd_dev
->mapping
.features
= features
;
1157 static void rbd_dev_mapping_clear(struct rbd_device
*rbd_dev
)
1159 rbd_dev
->mapping
.size
= 0;
1160 rbd_dev
->mapping
.features
= 0;
1163 static void rbd_segment_name_free(const char *name
)
1165 /* The explicit cast here is needed to drop the const qualifier */
1167 kmem_cache_free(rbd_segment_name_cache
, (void *)name
);
1170 static const char *rbd_segment_name(struct rbd_device
*rbd_dev
, u64 offset
)
1177 name
= kmem_cache_alloc(rbd_segment_name_cache
, GFP_NOIO
);
1180 segment
= offset
>> rbd_dev
->header
.obj_order
;
1181 name_format
= "%s.%012llx";
1182 if (rbd_dev
->image_format
== 2)
1183 name_format
= "%s.%016llx";
1184 ret
= snprintf(name
, CEPH_MAX_OID_NAME_LEN
+ 1, name_format
,
1185 rbd_dev
->header
.object_prefix
, segment
);
1186 if (ret
< 0 || ret
> CEPH_MAX_OID_NAME_LEN
) {
1187 pr_err("error formatting segment name for #%llu (%d)\n",
1189 rbd_segment_name_free(name
);
1196 static u64
rbd_segment_offset(struct rbd_device
*rbd_dev
, u64 offset
)
1198 u64 segment_size
= (u64
) 1 << rbd_dev
->header
.obj_order
;
1200 return offset
& (segment_size
- 1);
1203 static u64
rbd_segment_length(struct rbd_device
*rbd_dev
,
1204 u64 offset
, u64 length
)
1206 u64 segment_size
= (u64
) 1 << rbd_dev
->header
.obj_order
;
1208 offset
&= segment_size
- 1;
1210 rbd_assert(length
<= U64_MAX
- offset
);
1211 if (offset
+ length
> segment_size
)
1212 length
= segment_size
- offset
;
1218 * returns the size of an object in the image
1220 static u64
rbd_obj_bytes(struct rbd_image_header
*header
)
1222 return 1 << header
->obj_order
;
1229 static void bio_chain_put(struct bio
*chain
)
1235 chain
= chain
->bi_next
;
1241 * zeros a bio chain, starting at specific offset
1243 static void zero_bio_chain(struct bio
*chain
, int start_ofs
)
1246 struct bvec_iter iter
;
1247 unsigned long flags
;
1252 bio_for_each_segment(bv
, chain
, iter
) {
1253 if (pos
+ bv
.bv_len
> start_ofs
) {
1254 int remainder
= max(start_ofs
- pos
, 0);
1255 buf
= bvec_kmap_irq(&bv
, &flags
);
1256 memset(buf
+ remainder
, 0,
1257 bv
.bv_len
- remainder
);
1258 flush_dcache_page(bv
.bv_page
);
1259 bvec_kunmap_irq(buf
, &flags
);
1264 chain
= chain
->bi_next
;
1269 * similar to zero_bio_chain(), zeros data defined by a page array,
1270 * starting at the given byte offset from the start of the array and
1271 * continuing up to the given end offset. The pages array is
1272 * assumed to be big enough to hold all bytes up to the end.
1274 static void zero_pages(struct page
**pages
, u64 offset
, u64 end
)
1276 struct page
**page
= &pages
[offset
>> PAGE_SHIFT
];
1278 rbd_assert(end
> offset
);
1279 rbd_assert(end
- offset
<= (u64
)SIZE_MAX
);
1280 while (offset
< end
) {
1283 unsigned long flags
;
1286 page_offset
= offset
& ~PAGE_MASK
;
1287 length
= min_t(size_t, PAGE_SIZE
- page_offset
, end
- offset
);
1288 local_irq_save(flags
);
1289 kaddr
= kmap_atomic(*page
);
1290 memset(kaddr
+ page_offset
, 0, length
);
1291 flush_dcache_page(*page
);
1292 kunmap_atomic(kaddr
);
1293 local_irq_restore(flags
);
1301 * Clone a portion of a bio, starting at the given byte offset
1302 * and continuing for the number of bytes indicated.
1304 static struct bio
*bio_clone_range(struct bio
*bio_src
,
1305 unsigned int offset
,
1311 bio
= bio_clone(bio_src
, gfpmask
);
1313 return NULL
; /* ENOMEM */
1315 bio_advance(bio
, offset
);
1316 bio
->bi_iter
.bi_size
= len
;
1322 * Clone a portion of a bio chain, starting at the given byte offset
1323 * into the first bio in the source chain and continuing for the
1324 * number of bytes indicated. The result is another bio chain of
1325 * exactly the given length, or a null pointer on error.
1327 * The bio_src and offset parameters are both in-out. On entry they
1328 * refer to the first source bio and the offset into that bio where
1329 * the start of data to be cloned is located.
1331 * On return, bio_src is updated to refer to the bio in the source
1332 * chain that contains first un-cloned byte, and *offset will
1333 * contain the offset of that byte within that bio.
1335 static struct bio
*bio_chain_clone_range(struct bio
**bio_src
,
1336 unsigned int *offset
,
1340 struct bio
*bi
= *bio_src
;
1341 unsigned int off
= *offset
;
1342 struct bio
*chain
= NULL
;
1345 /* Build up a chain of clone bios up to the limit */
1347 if (!bi
|| off
>= bi
->bi_iter
.bi_size
|| !len
)
1348 return NULL
; /* Nothing to clone */
1352 unsigned int bi_size
;
1356 rbd_warn(NULL
, "bio_chain exhausted with %u left", len
);
1357 goto out_err
; /* EINVAL; ran out of bio's */
1359 bi_size
= min_t(unsigned int, bi
->bi_iter
.bi_size
- off
, len
);
1360 bio
= bio_clone_range(bi
, off
, bi_size
, gfpmask
);
1362 goto out_err
; /* ENOMEM */
1365 end
= &bio
->bi_next
;
1368 if (off
== bi
->bi_iter
.bi_size
) {
1379 bio_chain_put(chain
);
1385 * The default/initial value for all object request flags is 0. For
1386 * each flag, once its value is set to 1 it is never reset to 0
1389 static void obj_request_img_data_set(struct rbd_obj_request
*obj_request
)
1391 if (test_and_set_bit(OBJ_REQ_IMG_DATA
, &obj_request
->flags
)) {
1392 struct rbd_device
*rbd_dev
;
1394 rbd_dev
= obj_request
->img_request
->rbd_dev
;
1395 rbd_warn(rbd_dev
, "obj_request %p already marked img_data",
1400 static bool obj_request_img_data_test(struct rbd_obj_request
*obj_request
)
1403 return test_bit(OBJ_REQ_IMG_DATA
, &obj_request
->flags
) != 0;
1406 static void obj_request_done_set(struct rbd_obj_request
*obj_request
)
1408 if (test_and_set_bit(OBJ_REQ_DONE
, &obj_request
->flags
)) {
1409 struct rbd_device
*rbd_dev
= NULL
;
1411 if (obj_request_img_data_test(obj_request
))
1412 rbd_dev
= obj_request
->img_request
->rbd_dev
;
1413 rbd_warn(rbd_dev
, "obj_request %p already marked done",
1418 static bool obj_request_done_test(struct rbd_obj_request
*obj_request
)
1421 return test_bit(OBJ_REQ_DONE
, &obj_request
->flags
) != 0;
1425 * This sets the KNOWN flag after (possibly) setting the EXISTS
1426 * flag. The latter is set based on the "exists" value provided.
1428 * Note that for our purposes once an object exists it never goes
1429 * away again. It's possible that the response from two existence
1430 * checks are separated by the creation of the target object, and
1431 * the first ("doesn't exist") response arrives *after* the second
1432 * ("does exist"). In that case we ignore the second one.
1434 static void obj_request_existence_set(struct rbd_obj_request
*obj_request
,
1438 set_bit(OBJ_REQ_EXISTS
, &obj_request
->flags
);
1439 set_bit(OBJ_REQ_KNOWN
, &obj_request
->flags
);
1443 static bool obj_request_known_test(struct rbd_obj_request
*obj_request
)
1446 return test_bit(OBJ_REQ_KNOWN
, &obj_request
->flags
) != 0;
1449 static bool obj_request_exists_test(struct rbd_obj_request
*obj_request
)
1452 return test_bit(OBJ_REQ_EXISTS
, &obj_request
->flags
) != 0;
1455 static bool obj_request_overlaps_parent(struct rbd_obj_request
*obj_request
)
1457 struct rbd_device
*rbd_dev
= obj_request
->img_request
->rbd_dev
;
1459 return obj_request
->img_offset
<
1460 round_up(rbd_dev
->parent_overlap
, rbd_obj_bytes(&rbd_dev
->header
));
1463 static void rbd_obj_request_get(struct rbd_obj_request
*obj_request
)
1465 dout("%s: obj %p (was %d)\n", __func__
, obj_request
,
1466 atomic_read(&obj_request
->kref
.refcount
));
1467 kref_get(&obj_request
->kref
);
1470 static void rbd_obj_request_destroy(struct kref
*kref
);
1471 static void rbd_obj_request_put(struct rbd_obj_request
*obj_request
)
1473 rbd_assert(obj_request
!= NULL
);
1474 dout("%s: obj %p (was %d)\n", __func__
, obj_request
,
1475 atomic_read(&obj_request
->kref
.refcount
));
1476 kref_put(&obj_request
->kref
, rbd_obj_request_destroy
);
1479 static void rbd_img_request_get(struct rbd_img_request
*img_request
)
1481 dout("%s: img %p (was %d)\n", __func__
, img_request
,
1482 atomic_read(&img_request
->kref
.refcount
));
1483 kref_get(&img_request
->kref
);
1486 static bool img_request_child_test(struct rbd_img_request
*img_request
);
1487 static void rbd_parent_request_destroy(struct kref
*kref
);
1488 static void rbd_img_request_destroy(struct kref
*kref
);
1489 static void rbd_img_request_put(struct rbd_img_request
*img_request
)
1491 rbd_assert(img_request
!= NULL
);
1492 dout("%s: img %p (was %d)\n", __func__
, img_request
,
1493 atomic_read(&img_request
->kref
.refcount
));
1494 if (img_request_child_test(img_request
))
1495 kref_put(&img_request
->kref
, rbd_parent_request_destroy
);
1497 kref_put(&img_request
->kref
, rbd_img_request_destroy
);
1500 static inline void rbd_img_obj_request_add(struct rbd_img_request
*img_request
,
1501 struct rbd_obj_request
*obj_request
)
1503 rbd_assert(obj_request
->img_request
== NULL
);
1505 /* Image request now owns object's original reference */
1506 obj_request
->img_request
= img_request
;
1507 obj_request
->which
= img_request
->obj_request_count
;
1508 rbd_assert(!obj_request_img_data_test(obj_request
));
1509 obj_request_img_data_set(obj_request
);
1510 rbd_assert(obj_request
->which
!= BAD_WHICH
);
1511 img_request
->obj_request_count
++;
1512 list_add_tail(&obj_request
->links
, &img_request
->obj_requests
);
1513 dout("%s: img %p obj %p w=%u\n", __func__
, img_request
, obj_request
,
1514 obj_request
->which
);
1517 static inline void rbd_img_obj_request_del(struct rbd_img_request
*img_request
,
1518 struct rbd_obj_request
*obj_request
)
1520 rbd_assert(obj_request
->which
!= BAD_WHICH
);
1522 dout("%s: img %p obj %p w=%u\n", __func__
, img_request
, obj_request
,
1523 obj_request
->which
);
1524 list_del(&obj_request
->links
);
1525 rbd_assert(img_request
->obj_request_count
> 0);
1526 img_request
->obj_request_count
--;
1527 rbd_assert(obj_request
->which
== img_request
->obj_request_count
);
1528 obj_request
->which
= BAD_WHICH
;
1529 rbd_assert(obj_request_img_data_test(obj_request
));
1530 rbd_assert(obj_request
->img_request
== img_request
);
1531 obj_request
->img_request
= NULL
;
1532 obj_request
->callback
= NULL
;
1533 rbd_obj_request_put(obj_request
);
1536 static bool obj_request_type_valid(enum obj_request_type type
)
1539 case OBJ_REQUEST_NODATA
:
1540 case OBJ_REQUEST_BIO
:
1541 case OBJ_REQUEST_PAGES
:
1548 static int rbd_obj_request_submit(struct ceph_osd_client
*osdc
,
1549 struct rbd_obj_request
*obj_request
)
1551 dout("%s %p\n", __func__
, obj_request
);
1552 return ceph_osdc_start_request(osdc
, obj_request
->osd_req
, false);
1555 static void rbd_obj_request_end(struct rbd_obj_request
*obj_request
)
1557 dout("%s %p\n", __func__
, obj_request
);
1558 ceph_osdc_cancel_request(obj_request
->osd_req
);
1562 * Wait for an object request to complete. If interrupted, cancel the
1563 * underlying osd request.
1565 static int rbd_obj_request_wait(struct rbd_obj_request
*obj_request
)
1569 dout("%s %p\n", __func__
, obj_request
);
1571 ret
= wait_for_completion_interruptible(&obj_request
->completion
);
1573 dout("%s %p interrupted\n", __func__
, obj_request
);
1574 rbd_obj_request_end(obj_request
);
1578 dout("%s %p done\n", __func__
, obj_request
);
1582 static void rbd_img_request_complete(struct rbd_img_request
*img_request
)
1585 dout("%s: img %p\n", __func__
, img_request
);
1588 * If no error occurred, compute the aggregate transfer
1589 * count for the image request. We could instead use
1590 * atomic64_cmpxchg() to update it as each object request
1591 * completes; not clear which way is better off hand.
1593 if (!img_request
->result
) {
1594 struct rbd_obj_request
*obj_request
;
1597 for_each_obj_request(img_request
, obj_request
)
1598 xferred
+= obj_request
->xferred
;
1599 img_request
->xferred
= xferred
;
1602 if (img_request
->callback
)
1603 img_request
->callback(img_request
);
1605 rbd_img_request_put(img_request
);
1609 * The default/initial value for all image request flags is 0. Each
1610 * is conditionally set to 1 at image request initialization time
1611 * and currently never change thereafter.
1613 static void img_request_write_set(struct rbd_img_request
*img_request
)
1615 set_bit(IMG_REQ_WRITE
, &img_request
->flags
);
1619 static bool img_request_write_test(struct rbd_img_request
*img_request
)
1622 return test_bit(IMG_REQ_WRITE
, &img_request
->flags
) != 0;
1626 * Set the discard flag when the img_request is an discard request
1628 static void img_request_discard_set(struct rbd_img_request
*img_request
)
1630 set_bit(IMG_REQ_DISCARD
, &img_request
->flags
);
1634 static bool img_request_discard_test(struct rbd_img_request
*img_request
)
1637 return test_bit(IMG_REQ_DISCARD
, &img_request
->flags
) != 0;
1640 static void img_request_child_set(struct rbd_img_request
*img_request
)
1642 set_bit(IMG_REQ_CHILD
, &img_request
->flags
);
1646 static void img_request_child_clear(struct rbd_img_request
*img_request
)
1648 clear_bit(IMG_REQ_CHILD
, &img_request
->flags
);
1652 static bool img_request_child_test(struct rbd_img_request
*img_request
)
1655 return test_bit(IMG_REQ_CHILD
, &img_request
->flags
) != 0;
1658 static void img_request_layered_set(struct rbd_img_request
*img_request
)
1660 set_bit(IMG_REQ_LAYERED
, &img_request
->flags
);
1664 static void img_request_layered_clear(struct rbd_img_request
*img_request
)
1666 clear_bit(IMG_REQ_LAYERED
, &img_request
->flags
);
1670 static bool img_request_layered_test(struct rbd_img_request
*img_request
)
1673 return test_bit(IMG_REQ_LAYERED
, &img_request
->flags
) != 0;
1676 static enum obj_operation_type
1677 rbd_img_request_op_type(struct rbd_img_request
*img_request
)
1679 if (img_request_write_test(img_request
))
1680 return OBJ_OP_WRITE
;
1681 else if (img_request_discard_test(img_request
))
1682 return OBJ_OP_DISCARD
;
1688 rbd_img_obj_request_read_callback(struct rbd_obj_request
*obj_request
)
1690 u64 xferred
= obj_request
->xferred
;
1691 u64 length
= obj_request
->length
;
1693 dout("%s: obj %p img %p result %d %llu/%llu\n", __func__
,
1694 obj_request
, obj_request
->img_request
, obj_request
->result
,
1697 * ENOENT means a hole in the image. We zero-fill the entire
1698 * length of the request. A short read also implies zero-fill
1699 * to the end of the request. An error requires the whole
1700 * length of the request to be reported finished with an error
1701 * to the block layer. In each case we update the xferred
1702 * count to indicate the whole request was satisfied.
1704 rbd_assert(obj_request
->type
!= OBJ_REQUEST_NODATA
);
1705 if (obj_request
->result
== -ENOENT
) {
1706 if (obj_request
->type
== OBJ_REQUEST_BIO
)
1707 zero_bio_chain(obj_request
->bio_list
, 0);
1709 zero_pages(obj_request
->pages
, 0, length
);
1710 obj_request
->result
= 0;
1711 } else if (xferred
< length
&& !obj_request
->result
) {
1712 if (obj_request
->type
== OBJ_REQUEST_BIO
)
1713 zero_bio_chain(obj_request
->bio_list
, xferred
);
1715 zero_pages(obj_request
->pages
, xferred
, length
);
1717 obj_request
->xferred
= length
;
1718 obj_request_done_set(obj_request
);
1721 static void rbd_obj_request_complete(struct rbd_obj_request
*obj_request
)
1723 dout("%s: obj %p cb %p\n", __func__
, obj_request
,
1724 obj_request
->callback
);
1725 if (obj_request
->callback
)
1726 obj_request
->callback(obj_request
);
1728 complete_all(&obj_request
->completion
);
1731 static void rbd_osd_trivial_callback(struct rbd_obj_request
*obj_request
)
1733 dout("%s: obj %p\n", __func__
, obj_request
);
1734 obj_request_done_set(obj_request
);
1737 static void rbd_osd_read_callback(struct rbd_obj_request
*obj_request
)
1739 struct rbd_img_request
*img_request
= NULL
;
1740 struct rbd_device
*rbd_dev
= NULL
;
1741 bool layered
= false;
1743 if (obj_request_img_data_test(obj_request
)) {
1744 img_request
= obj_request
->img_request
;
1745 layered
= img_request
&& img_request_layered_test(img_request
);
1746 rbd_dev
= img_request
->rbd_dev
;
1749 dout("%s: obj %p img %p result %d %llu/%llu\n", __func__
,
1750 obj_request
, img_request
, obj_request
->result
,
1751 obj_request
->xferred
, obj_request
->length
);
1752 if (layered
&& obj_request
->result
== -ENOENT
&&
1753 obj_request
->img_offset
< rbd_dev
->parent_overlap
)
1754 rbd_img_parent_read(obj_request
);
1755 else if (img_request
)
1756 rbd_img_obj_request_read_callback(obj_request
);
1758 obj_request_done_set(obj_request
);
1761 static void rbd_osd_write_callback(struct rbd_obj_request
*obj_request
)
1763 dout("%s: obj %p result %d %llu\n", __func__
, obj_request
,
1764 obj_request
->result
, obj_request
->length
);
1766 * There is no such thing as a successful short write. Set
1767 * it to our originally-requested length.
1769 obj_request
->xferred
= obj_request
->length
;
1770 obj_request_done_set(obj_request
);
1773 static void rbd_osd_discard_callback(struct rbd_obj_request
*obj_request
)
1775 dout("%s: obj %p result %d %llu\n", __func__
, obj_request
,
1776 obj_request
->result
, obj_request
->length
);
1778 * There is no such thing as a successful short discard. Set
1779 * it to our originally-requested length.
1781 obj_request
->xferred
= obj_request
->length
;
1782 /* discarding a non-existent object is not a problem */
1783 if (obj_request
->result
== -ENOENT
)
1784 obj_request
->result
= 0;
1785 obj_request_done_set(obj_request
);
1789 * For a simple stat call there's nothing to do. We'll do more if
1790 * this is part of a write sequence for a layered image.
1792 static void rbd_osd_stat_callback(struct rbd_obj_request
*obj_request
)
1794 dout("%s: obj %p\n", __func__
, obj_request
);
1795 obj_request_done_set(obj_request
);
1798 static void rbd_osd_req_callback(struct ceph_osd_request
*osd_req
,
1799 struct ceph_msg
*msg
)
1801 struct rbd_obj_request
*obj_request
= osd_req
->r_priv
;
1804 dout("%s: osd_req %p msg %p\n", __func__
, osd_req
, msg
);
1805 rbd_assert(osd_req
== obj_request
->osd_req
);
1806 if (obj_request_img_data_test(obj_request
)) {
1807 rbd_assert(obj_request
->img_request
);
1808 rbd_assert(obj_request
->which
!= BAD_WHICH
);
1810 rbd_assert(obj_request
->which
== BAD_WHICH
);
1813 if (osd_req
->r_result
< 0)
1814 obj_request
->result
= osd_req
->r_result
;
1816 rbd_assert(osd_req
->r_num_ops
<= CEPH_OSD_MAX_OP
);
1819 * We support a 64-bit length, but ultimately it has to be
1820 * passed to blk_end_request(), which takes an unsigned int.
1822 obj_request
->xferred
= osd_req
->r_reply_op_len
[0];
1823 rbd_assert(obj_request
->xferred
< (u64
)UINT_MAX
);
1825 opcode
= osd_req
->r_ops
[0].op
;
1827 case CEPH_OSD_OP_READ
:
1828 rbd_osd_read_callback(obj_request
);
1830 case CEPH_OSD_OP_SETALLOCHINT
:
1831 rbd_assert(osd_req
->r_ops
[1].op
== CEPH_OSD_OP_WRITE
);
1833 case CEPH_OSD_OP_WRITE
:
1834 rbd_osd_write_callback(obj_request
);
1836 case CEPH_OSD_OP_STAT
:
1837 rbd_osd_stat_callback(obj_request
);
1839 case CEPH_OSD_OP_DELETE
:
1840 case CEPH_OSD_OP_TRUNCATE
:
1841 case CEPH_OSD_OP_ZERO
:
1842 rbd_osd_discard_callback(obj_request
);
1844 case CEPH_OSD_OP_CALL
:
1845 case CEPH_OSD_OP_NOTIFY_ACK
:
1846 case CEPH_OSD_OP_WATCH
:
1847 rbd_osd_trivial_callback(obj_request
);
1850 rbd_warn(NULL
, "%s: unsupported op %hu",
1851 obj_request
->object_name
, (unsigned short) opcode
);
1855 if (obj_request_done_test(obj_request
))
1856 rbd_obj_request_complete(obj_request
);
1859 static void rbd_osd_req_format_read(struct rbd_obj_request
*obj_request
)
1861 struct rbd_img_request
*img_request
= obj_request
->img_request
;
1862 struct ceph_osd_request
*osd_req
= obj_request
->osd_req
;
1865 rbd_assert(osd_req
!= NULL
);
1867 snap_id
= img_request
? img_request
->snap_id
: CEPH_NOSNAP
;
1868 ceph_osdc_build_request(osd_req
, obj_request
->offset
,
1869 NULL
, snap_id
, NULL
);
1872 static void rbd_osd_req_format_write(struct rbd_obj_request
*obj_request
)
1874 struct rbd_img_request
*img_request
= obj_request
->img_request
;
1875 struct ceph_osd_request
*osd_req
= obj_request
->osd_req
;
1876 struct ceph_snap_context
*snapc
;
1877 struct timespec mtime
= CURRENT_TIME
;
1879 rbd_assert(osd_req
!= NULL
);
1881 snapc
= img_request
? img_request
->snapc
: NULL
;
1882 ceph_osdc_build_request(osd_req
, obj_request
->offset
,
1883 snapc
, CEPH_NOSNAP
, &mtime
);
1887 * Create an osd request. A read request has one osd op (read).
1888 * A write request has either one (watch) or two (hint+write) osd ops.
1889 * (All rbd data writes are prefixed with an allocation hint op, but
1890 * technically osd watch is a write request, hence this distinction.)
1892 static struct ceph_osd_request
*rbd_osd_req_create(
1893 struct rbd_device
*rbd_dev
,
1894 enum obj_operation_type op_type
,
1895 unsigned int num_ops
,
1896 struct rbd_obj_request
*obj_request
)
1898 struct ceph_snap_context
*snapc
= NULL
;
1899 struct ceph_osd_client
*osdc
;
1900 struct ceph_osd_request
*osd_req
;
1902 if (obj_request_img_data_test(obj_request
) &&
1903 (op_type
== OBJ_OP_DISCARD
|| op_type
== OBJ_OP_WRITE
)) {
1904 struct rbd_img_request
*img_request
= obj_request
->img_request
;
1905 if (op_type
== OBJ_OP_WRITE
) {
1906 rbd_assert(img_request_write_test(img_request
));
1908 rbd_assert(img_request_discard_test(img_request
));
1910 snapc
= img_request
->snapc
;
1913 rbd_assert(num_ops
== 1 || ((op_type
== OBJ_OP_WRITE
) && num_ops
== 2));
1915 /* Allocate and initialize the request, for the num_ops ops */
1917 osdc
= &rbd_dev
->rbd_client
->client
->osdc
;
1918 osd_req
= ceph_osdc_alloc_request(osdc
, snapc
, num_ops
, false,
1921 return NULL
; /* ENOMEM */
1923 if (op_type
== OBJ_OP_WRITE
|| op_type
== OBJ_OP_DISCARD
)
1924 osd_req
->r_flags
= CEPH_OSD_FLAG_WRITE
| CEPH_OSD_FLAG_ONDISK
;
1926 osd_req
->r_flags
= CEPH_OSD_FLAG_READ
;
1928 osd_req
->r_callback
= rbd_osd_req_callback
;
1929 osd_req
->r_priv
= obj_request
;
1931 osd_req
->r_base_oloc
.pool
= ceph_file_layout_pg_pool(rbd_dev
->layout
);
1932 ceph_oid_set_name(&osd_req
->r_base_oid
, obj_request
->object_name
);
1938 * Create a copyup osd request based on the information in the object
1939 * request supplied. A copyup request has two or three osd ops, a
1940 * copyup method call, potentially a hint op, and a write or truncate
1943 static struct ceph_osd_request
*
1944 rbd_osd_req_create_copyup(struct rbd_obj_request
*obj_request
)
1946 struct rbd_img_request
*img_request
;
1947 struct ceph_snap_context
*snapc
;
1948 struct rbd_device
*rbd_dev
;
1949 struct ceph_osd_client
*osdc
;
1950 struct ceph_osd_request
*osd_req
;
1951 int num_osd_ops
= 3;
1953 rbd_assert(obj_request_img_data_test(obj_request
));
1954 img_request
= obj_request
->img_request
;
1955 rbd_assert(img_request
);
1956 rbd_assert(img_request_write_test(img_request
) ||
1957 img_request_discard_test(img_request
));
1959 if (img_request_discard_test(img_request
))
1962 /* Allocate and initialize the request, for all the ops */
1964 snapc
= img_request
->snapc
;
1965 rbd_dev
= img_request
->rbd_dev
;
1966 osdc
= &rbd_dev
->rbd_client
->client
->osdc
;
1967 osd_req
= ceph_osdc_alloc_request(osdc
, snapc
, num_osd_ops
,
1970 return NULL
; /* ENOMEM */
1972 osd_req
->r_flags
= CEPH_OSD_FLAG_WRITE
| CEPH_OSD_FLAG_ONDISK
;
1973 osd_req
->r_callback
= rbd_osd_req_callback
;
1974 osd_req
->r_priv
= obj_request
;
1976 osd_req
->r_base_oloc
.pool
= ceph_file_layout_pg_pool(rbd_dev
->layout
);
1977 ceph_oid_set_name(&osd_req
->r_base_oid
, obj_request
->object_name
);
1983 static void rbd_osd_req_destroy(struct ceph_osd_request
*osd_req
)
1985 ceph_osdc_put_request(osd_req
);
1988 /* object_name is assumed to be a non-null pointer and NUL-terminated */
1990 static struct rbd_obj_request
*rbd_obj_request_create(const char *object_name
,
1991 u64 offset
, u64 length
,
1992 enum obj_request_type type
)
1994 struct rbd_obj_request
*obj_request
;
1998 rbd_assert(obj_request_type_valid(type
));
2000 size
= strlen(object_name
) + 1;
2001 name
= kmalloc(size
, GFP_KERNEL
);
2005 obj_request
= kmem_cache_zalloc(rbd_obj_request_cache
, GFP_KERNEL
);
2011 obj_request
->object_name
= memcpy(name
, object_name
, size
);
2012 obj_request
->offset
= offset
;
2013 obj_request
->length
= length
;
2014 obj_request
->flags
= 0;
2015 obj_request
->which
= BAD_WHICH
;
2016 obj_request
->type
= type
;
2017 INIT_LIST_HEAD(&obj_request
->links
);
2018 init_completion(&obj_request
->completion
);
2019 kref_init(&obj_request
->kref
);
2021 dout("%s: \"%s\" %llu/%llu %d -> obj %p\n", __func__
, object_name
,
2022 offset
, length
, (int)type
, obj_request
);
2027 static void rbd_obj_request_destroy(struct kref
*kref
)
2029 struct rbd_obj_request
*obj_request
;
2031 obj_request
= container_of(kref
, struct rbd_obj_request
, kref
);
2033 dout("%s: obj %p\n", __func__
, obj_request
);
2035 rbd_assert(obj_request
->img_request
== NULL
);
2036 rbd_assert(obj_request
->which
== BAD_WHICH
);
2038 if (obj_request
->osd_req
)
2039 rbd_osd_req_destroy(obj_request
->osd_req
);
2041 rbd_assert(obj_request_type_valid(obj_request
->type
));
2042 switch (obj_request
->type
) {
2043 case OBJ_REQUEST_NODATA
:
2044 break; /* Nothing to do */
2045 case OBJ_REQUEST_BIO
:
2046 if (obj_request
->bio_list
)
2047 bio_chain_put(obj_request
->bio_list
);
2049 case OBJ_REQUEST_PAGES
:
2050 if (obj_request
->pages
)
2051 ceph_release_page_vector(obj_request
->pages
,
2052 obj_request
->page_count
);
2056 kfree(obj_request
->object_name
);
2057 obj_request
->object_name
= NULL
;
2058 kmem_cache_free(rbd_obj_request_cache
, obj_request
);
2061 /* It's OK to call this for a device with no parent */
2063 static void rbd_spec_put(struct rbd_spec
*spec
);
2064 static void rbd_dev_unparent(struct rbd_device
*rbd_dev
)
2066 rbd_dev_remove_parent(rbd_dev
);
2067 rbd_spec_put(rbd_dev
->parent_spec
);
2068 rbd_dev
->parent_spec
= NULL
;
2069 rbd_dev
->parent_overlap
= 0;
2073 * Parent image reference counting is used to determine when an
2074 * image's parent fields can be safely torn down--after there are no
2075 * more in-flight requests to the parent image. When the last
2076 * reference is dropped, cleaning them up is safe.
2078 static void rbd_dev_parent_put(struct rbd_device
*rbd_dev
)
2082 if (!rbd_dev
->parent_spec
)
2085 counter
= atomic_dec_return_safe(&rbd_dev
->parent_ref
);
2089 /* Last reference; clean up parent data structures */
2092 rbd_dev_unparent(rbd_dev
);
2094 rbd_warn(rbd_dev
, "parent reference underflow");
2098 * If an image has a non-zero parent overlap, get a reference to its
2101 * Returns true if the rbd device has a parent with a non-zero
2102 * overlap and a reference for it was successfully taken, or
2105 static bool rbd_dev_parent_get(struct rbd_device
*rbd_dev
)
2109 if (!rbd_dev
->parent_spec
)
2112 down_read(&rbd_dev
->header_rwsem
);
2113 if (rbd_dev
->parent_overlap
)
2114 counter
= atomic_inc_return_safe(&rbd_dev
->parent_ref
);
2115 up_read(&rbd_dev
->header_rwsem
);
2118 rbd_warn(rbd_dev
, "parent reference overflow");
2124 * Caller is responsible for filling in the list of object requests
2125 * that comprises the image request, and the Linux request pointer
2126 * (if there is one).
2128 static struct rbd_img_request
*rbd_img_request_create(
2129 struct rbd_device
*rbd_dev
,
2130 u64 offset
, u64 length
,
2131 enum obj_operation_type op_type
,
2132 struct ceph_snap_context
*snapc
)
2134 struct rbd_img_request
*img_request
;
2136 img_request
= kmem_cache_alloc(rbd_img_request_cache
, GFP_NOIO
);
2140 img_request
->rq
= NULL
;
2141 img_request
->rbd_dev
= rbd_dev
;
2142 img_request
->offset
= offset
;
2143 img_request
->length
= length
;
2144 img_request
->flags
= 0;
2145 if (op_type
== OBJ_OP_DISCARD
) {
2146 img_request_discard_set(img_request
);
2147 img_request
->snapc
= snapc
;
2148 } else if (op_type
== OBJ_OP_WRITE
) {
2149 img_request_write_set(img_request
);
2150 img_request
->snapc
= snapc
;
2152 img_request
->snap_id
= rbd_dev
->spec
->snap_id
;
2154 if (rbd_dev_parent_get(rbd_dev
))
2155 img_request_layered_set(img_request
);
2156 spin_lock_init(&img_request
->completion_lock
);
2157 img_request
->next_completion
= 0;
2158 img_request
->callback
= NULL
;
2159 img_request
->result
= 0;
2160 img_request
->obj_request_count
= 0;
2161 INIT_LIST_HEAD(&img_request
->obj_requests
);
2162 kref_init(&img_request
->kref
);
2164 dout("%s: rbd_dev %p %s %llu/%llu -> img %p\n", __func__
, rbd_dev
,
2165 obj_op_name(op_type
), offset
, length
, img_request
);
2170 static void rbd_img_request_destroy(struct kref
*kref
)
2172 struct rbd_img_request
*img_request
;
2173 struct rbd_obj_request
*obj_request
;
2174 struct rbd_obj_request
*next_obj_request
;
2176 img_request
= container_of(kref
, struct rbd_img_request
, kref
);
2178 dout("%s: img %p\n", __func__
, img_request
);
2180 for_each_obj_request_safe(img_request
, obj_request
, next_obj_request
)
2181 rbd_img_obj_request_del(img_request
, obj_request
);
2182 rbd_assert(img_request
->obj_request_count
== 0);
2184 if (img_request_layered_test(img_request
)) {
2185 img_request_layered_clear(img_request
);
2186 rbd_dev_parent_put(img_request
->rbd_dev
);
2189 if (img_request_write_test(img_request
) ||
2190 img_request_discard_test(img_request
))
2191 ceph_put_snap_context(img_request
->snapc
);
2193 kmem_cache_free(rbd_img_request_cache
, img_request
);
2196 static struct rbd_img_request
*rbd_parent_request_create(
2197 struct rbd_obj_request
*obj_request
,
2198 u64 img_offset
, u64 length
)
2200 struct rbd_img_request
*parent_request
;
2201 struct rbd_device
*rbd_dev
;
2203 rbd_assert(obj_request
->img_request
);
2204 rbd_dev
= obj_request
->img_request
->rbd_dev
;
2206 parent_request
= rbd_img_request_create(rbd_dev
->parent
, img_offset
,
2207 length
, OBJ_OP_READ
, NULL
);
2208 if (!parent_request
)
2211 img_request_child_set(parent_request
);
2212 rbd_obj_request_get(obj_request
);
2213 parent_request
->obj_request
= obj_request
;
2215 return parent_request
;
2218 static void rbd_parent_request_destroy(struct kref
*kref
)
2220 struct rbd_img_request
*parent_request
;
2221 struct rbd_obj_request
*orig_request
;
2223 parent_request
= container_of(kref
, struct rbd_img_request
, kref
);
2224 orig_request
= parent_request
->obj_request
;
2226 parent_request
->obj_request
= NULL
;
2227 rbd_obj_request_put(orig_request
);
2228 img_request_child_clear(parent_request
);
2230 rbd_img_request_destroy(kref
);
2233 static bool rbd_img_obj_end_request(struct rbd_obj_request
*obj_request
)
2235 struct rbd_img_request
*img_request
;
2236 unsigned int xferred
;
2240 rbd_assert(obj_request_img_data_test(obj_request
));
2241 img_request
= obj_request
->img_request
;
2243 rbd_assert(obj_request
->xferred
<= (u64
)UINT_MAX
);
2244 xferred
= (unsigned int)obj_request
->xferred
;
2245 result
= obj_request
->result
;
2247 struct rbd_device
*rbd_dev
= img_request
->rbd_dev
;
2248 enum obj_operation_type op_type
;
2250 if (img_request_discard_test(img_request
))
2251 op_type
= OBJ_OP_DISCARD
;
2252 else if (img_request_write_test(img_request
))
2253 op_type
= OBJ_OP_WRITE
;
2255 op_type
= OBJ_OP_READ
;
2257 rbd_warn(rbd_dev
, "%s %llx at %llx (%llx)",
2258 obj_op_name(op_type
), obj_request
->length
,
2259 obj_request
->img_offset
, obj_request
->offset
);
2260 rbd_warn(rbd_dev
, " result %d xferred %x",
2262 if (!img_request
->result
)
2263 img_request
->result
= result
;
2266 /* Image object requests don't own their page array */
2268 if (obj_request
->type
== OBJ_REQUEST_PAGES
) {
2269 obj_request
->pages
= NULL
;
2270 obj_request
->page_count
= 0;
2273 if (img_request_child_test(img_request
)) {
2274 rbd_assert(img_request
->obj_request
!= NULL
);
2275 more
= obj_request
->which
< img_request
->obj_request_count
- 1;
2277 rbd_assert(img_request
->rq
!= NULL
);
2278 more
= blk_end_request(img_request
->rq
, result
, xferred
);
2284 static void rbd_img_obj_callback(struct rbd_obj_request
*obj_request
)
2286 struct rbd_img_request
*img_request
;
2287 u32 which
= obj_request
->which
;
2290 rbd_assert(obj_request_img_data_test(obj_request
));
2291 img_request
= obj_request
->img_request
;
2293 dout("%s: img %p obj %p\n", __func__
, img_request
, obj_request
);
2294 rbd_assert(img_request
!= NULL
);
2295 rbd_assert(img_request
->obj_request_count
> 0);
2296 rbd_assert(which
!= BAD_WHICH
);
2297 rbd_assert(which
< img_request
->obj_request_count
);
2299 spin_lock_irq(&img_request
->completion_lock
);
2300 if (which
!= img_request
->next_completion
)
2303 for_each_obj_request_from(img_request
, obj_request
) {
2305 rbd_assert(which
< img_request
->obj_request_count
);
2307 if (!obj_request_done_test(obj_request
))
2309 more
= rbd_img_obj_end_request(obj_request
);
2313 rbd_assert(more
^ (which
== img_request
->obj_request_count
));
2314 img_request
->next_completion
= which
;
2316 spin_unlock_irq(&img_request
->completion_lock
);
2317 rbd_img_request_put(img_request
);
2320 rbd_img_request_complete(img_request
);
2324 * Add individual osd ops to the given ceph_osd_request and prepare
2325 * them for submission. num_ops is the current number of
2326 * osd operations already to the object request.
2328 static void rbd_img_obj_request_fill(struct rbd_obj_request
*obj_request
,
2329 struct ceph_osd_request
*osd_request
,
2330 enum obj_operation_type op_type
,
2331 unsigned int num_ops
)
2333 struct rbd_img_request
*img_request
= obj_request
->img_request
;
2334 struct rbd_device
*rbd_dev
= img_request
->rbd_dev
;
2335 u64 object_size
= rbd_obj_bytes(&rbd_dev
->header
);
2336 u64 offset
= obj_request
->offset
;
2337 u64 length
= obj_request
->length
;
2341 if (op_type
== OBJ_OP_DISCARD
) {
2342 if (!offset
&& length
== object_size
&&
2343 (!img_request_layered_test(img_request
) ||
2344 !obj_request_overlaps_parent(obj_request
))) {
2345 opcode
= CEPH_OSD_OP_DELETE
;
2346 } else if ((offset
+ length
== object_size
)) {
2347 opcode
= CEPH_OSD_OP_TRUNCATE
;
2349 down_read(&rbd_dev
->header_rwsem
);
2350 img_end
= rbd_dev
->header
.image_size
;
2351 up_read(&rbd_dev
->header_rwsem
);
2353 if (obj_request
->img_offset
+ length
== img_end
)
2354 opcode
= CEPH_OSD_OP_TRUNCATE
;
2356 opcode
= CEPH_OSD_OP_ZERO
;
2358 } else if (op_type
== OBJ_OP_WRITE
) {
2359 opcode
= CEPH_OSD_OP_WRITE
;
2360 osd_req_op_alloc_hint_init(osd_request
, num_ops
,
2361 object_size
, object_size
);
2364 opcode
= CEPH_OSD_OP_READ
;
2367 if (opcode
== CEPH_OSD_OP_DELETE
)
2368 osd_req_op_init(osd_request
, num_ops
, opcode
);
2370 osd_req_op_extent_init(osd_request
, num_ops
, opcode
,
2371 offset
, length
, 0, 0);
2373 if (obj_request
->type
== OBJ_REQUEST_BIO
)
2374 osd_req_op_extent_osd_data_bio(osd_request
, num_ops
,
2375 obj_request
->bio_list
, length
);
2376 else if (obj_request
->type
== OBJ_REQUEST_PAGES
)
2377 osd_req_op_extent_osd_data_pages(osd_request
, num_ops
,
2378 obj_request
->pages
, length
,
2379 offset
& ~PAGE_MASK
, false, false);
2381 /* Discards are also writes */
2382 if (op_type
== OBJ_OP_WRITE
|| op_type
== OBJ_OP_DISCARD
)
2383 rbd_osd_req_format_write(obj_request
);
2385 rbd_osd_req_format_read(obj_request
);
2389 * Split up an image request into one or more object requests, each
2390 * to a different object. The "type" parameter indicates whether
2391 * "data_desc" is the pointer to the head of a list of bio
2392 * structures, or the base of a page array. In either case this
2393 * function assumes data_desc describes memory sufficient to hold
2394 * all data described by the image request.
2396 static int rbd_img_request_fill(struct rbd_img_request
*img_request
,
2397 enum obj_request_type type
,
2400 struct rbd_device
*rbd_dev
= img_request
->rbd_dev
;
2401 struct rbd_obj_request
*obj_request
= NULL
;
2402 struct rbd_obj_request
*next_obj_request
;
2403 struct bio
*bio_list
= NULL
;
2404 unsigned int bio_offset
= 0;
2405 struct page
**pages
= NULL
;
2406 enum obj_operation_type op_type
;
2410 dout("%s: img %p type %d data_desc %p\n", __func__
, img_request
,
2411 (int)type
, data_desc
);
2413 img_offset
= img_request
->offset
;
2414 resid
= img_request
->length
;
2415 rbd_assert(resid
> 0);
2416 op_type
= rbd_img_request_op_type(img_request
);
2418 if (type
== OBJ_REQUEST_BIO
) {
2419 bio_list
= data_desc
;
2420 rbd_assert(img_offset
==
2421 bio_list
->bi_iter
.bi_sector
<< SECTOR_SHIFT
);
2422 } else if (type
== OBJ_REQUEST_PAGES
) {
2427 struct ceph_osd_request
*osd_req
;
2428 const char *object_name
;
2432 object_name
= rbd_segment_name(rbd_dev
, img_offset
);
2435 offset
= rbd_segment_offset(rbd_dev
, img_offset
);
2436 length
= rbd_segment_length(rbd_dev
, img_offset
, resid
);
2437 obj_request
= rbd_obj_request_create(object_name
,
2438 offset
, length
, type
);
2439 /* object request has its own copy of the object name */
2440 rbd_segment_name_free(object_name
);
2445 * set obj_request->img_request before creating the
2446 * osd_request so that it gets the right snapc
2448 rbd_img_obj_request_add(img_request
, obj_request
);
2450 if (type
== OBJ_REQUEST_BIO
) {
2451 unsigned int clone_size
;
2453 rbd_assert(length
<= (u64
)UINT_MAX
);
2454 clone_size
= (unsigned int)length
;
2455 obj_request
->bio_list
=
2456 bio_chain_clone_range(&bio_list
,
2460 if (!obj_request
->bio_list
)
2462 } else if (type
== OBJ_REQUEST_PAGES
) {
2463 unsigned int page_count
;
2465 obj_request
->pages
= pages
;
2466 page_count
= (u32
)calc_pages_for(offset
, length
);
2467 obj_request
->page_count
= page_count
;
2468 if ((offset
+ length
) & ~PAGE_MASK
)
2469 page_count
--; /* more on last page */
2470 pages
+= page_count
;
2473 osd_req
= rbd_osd_req_create(rbd_dev
, op_type
,
2474 (op_type
== OBJ_OP_WRITE
) ? 2 : 1,
2479 obj_request
->osd_req
= osd_req
;
2480 obj_request
->callback
= rbd_img_obj_callback
;
2481 obj_request
->img_offset
= img_offset
;
2483 rbd_img_obj_request_fill(obj_request
, osd_req
, op_type
, 0);
2485 rbd_img_request_get(img_request
);
2487 img_offset
+= length
;
2494 for_each_obj_request_safe(img_request
, obj_request
, next_obj_request
)
2495 rbd_img_obj_request_del(img_request
, obj_request
);
2501 rbd_img_obj_copyup_callback(struct rbd_obj_request
*obj_request
)
2503 struct rbd_img_request
*img_request
;
2504 struct rbd_device
*rbd_dev
;
2505 struct page
**pages
;
2508 rbd_assert(obj_request
->type
== OBJ_REQUEST_BIO
||
2509 obj_request
->type
== OBJ_REQUEST_NODATA
);
2510 rbd_assert(obj_request_img_data_test(obj_request
));
2511 img_request
= obj_request
->img_request
;
2512 rbd_assert(img_request
);
2514 rbd_dev
= img_request
->rbd_dev
;
2515 rbd_assert(rbd_dev
);
2517 pages
= obj_request
->copyup_pages
;
2518 rbd_assert(pages
!= NULL
);
2519 obj_request
->copyup_pages
= NULL
;
2520 page_count
= obj_request
->copyup_page_count
;
2521 rbd_assert(page_count
);
2522 obj_request
->copyup_page_count
= 0;
2523 ceph_release_page_vector(pages
, page_count
);
2526 * We want the transfer count to reflect the size of the
2527 * original write request. There is no such thing as a
2528 * successful short write, so if the request was successful
2529 * we can just set it to the originally-requested length.
2531 if (!obj_request
->result
)
2532 obj_request
->xferred
= obj_request
->length
;
2534 /* Finish up with the normal image object callback */
2536 rbd_img_obj_callback(obj_request
);
2540 rbd_img_obj_parent_read_full_callback(struct rbd_img_request
*img_request
)
2542 struct rbd_obj_request
*orig_request
;
2543 struct ceph_osd_request
*osd_req
;
2544 struct ceph_osd_client
*osdc
;
2545 struct rbd_device
*rbd_dev
;
2546 struct page
**pages
;
2547 enum obj_operation_type op_type
;
2552 rbd_assert(img_request_child_test(img_request
));
2554 /* First get what we need from the image request */
2556 pages
= img_request
->copyup_pages
;
2557 rbd_assert(pages
!= NULL
);
2558 img_request
->copyup_pages
= NULL
;
2559 page_count
= img_request
->copyup_page_count
;
2560 rbd_assert(page_count
);
2561 img_request
->copyup_page_count
= 0;
2563 orig_request
= img_request
->obj_request
;
2564 rbd_assert(orig_request
!= NULL
);
2565 rbd_assert(obj_request_type_valid(orig_request
->type
));
2566 img_result
= img_request
->result
;
2567 parent_length
= img_request
->length
;
2568 rbd_assert(parent_length
== img_request
->xferred
);
2569 rbd_img_request_put(img_request
);
2571 rbd_assert(orig_request
->img_request
);
2572 rbd_dev
= orig_request
->img_request
->rbd_dev
;
2573 rbd_assert(rbd_dev
);
2576 * If the overlap has become 0 (most likely because the
2577 * image has been flattened) we need to free the pages
2578 * and re-submit the original write request.
2580 if (!rbd_dev
->parent_overlap
) {
2581 struct ceph_osd_client
*osdc
;
2583 ceph_release_page_vector(pages
, page_count
);
2584 osdc
= &rbd_dev
->rbd_client
->client
->osdc
;
2585 img_result
= rbd_obj_request_submit(osdc
, orig_request
);
2594 * The original osd request is of no use to use any more.
2595 * We need a new one that can hold the three ops in a copyup
2596 * request. Allocate the new copyup osd request for the
2597 * original request, and release the old one.
2599 img_result
= -ENOMEM
;
2600 osd_req
= rbd_osd_req_create_copyup(orig_request
);
2603 rbd_osd_req_destroy(orig_request
->osd_req
);
2604 orig_request
->osd_req
= osd_req
;
2605 orig_request
->copyup_pages
= pages
;
2606 orig_request
->copyup_page_count
= page_count
;
2608 /* Initialize the copyup op */
2610 osd_req_op_cls_init(osd_req
, 0, CEPH_OSD_OP_CALL
, "rbd", "copyup");
2611 osd_req_op_cls_request_data_pages(osd_req
, 0, pages
, parent_length
, 0,
2614 /* Add the other op(s) */
2616 op_type
= rbd_img_request_op_type(orig_request
->img_request
);
2617 rbd_img_obj_request_fill(orig_request
, osd_req
, op_type
, 1);
2619 /* All set, send it off. */
2621 orig_request
->callback
= rbd_img_obj_copyup_callback
;
2622 osdc
= &rbd_dev
->rbd_client
->client
->osdc
;
2623 img_result
= rbd_obj_request_submit(osdc
, orig_request
);
2627 /* Record the error code and complete the request */
2629 orig_request
->result
= img_result
;
2630 orig_request
->xferred
= 0;
2631 obj_request_done_set(orig_request
);
2632 rbd_obj_request_complete(orig_request
);
2636 * Read from the parent image the range of data that covers the
2637 * entire target of the given object request. This is used for
2638 * satisfying a layered image write request when the target of an
2639 * object request from the image request does not exist.
2641 * A page array big enough to hold the returned data is allocated
2642 * and supplied to rbd_img_request_fill() as the "data descriptor."
2643 * When the read completes, this page array will be transferred to
2644 * the original object request for the copyup operation.
2646 * If an error occurs, record it as the result of the original
2647 * object request and mark it done so it gets completed.
2649 static int rbd_img_obj_parent_read_full(struct rbd_obj_request
*obj_request
)
2651 struct rbd_img_request
*img_request
= NULL
;
2652 struct rbd_img_request
*parent_request
= NULL
;
2653 struct rbd_device
*rbd_dev
;
2656 struct page
**pages
= NULL
;
2660 rbd_assert(obj_request_img_data_test(obj_request
));
2661 rbd_assert(obj_request_type_valid(obj_request
->type
));
2663 img_request
= obj_request
->img_request
;
2664 rbd_assert(img_request
!= NULL
);
2665 rbd_dev
= img_request
->rbd_dev
;
2666 rbd_assert(rbd_dev
->parent
!= NULL
);
2669 * Determine the byte range covered by the object in the
2670 * child image to which the original request was to be sent.
2672 img_offset
= obj_request
->img_offset
- obj_request
->offset
;
2673 length
= (u64
)1 << rbd_dev
->header
.obj_order
;
2676 * There is no defined parent data beyond the parent
2677 * overlap, so limit what we read at that boundary if
2680 if (img_offset
+ length
> rbd_dev
->parent_overlap
) {
2681 rbd_assert(img_offset
< rbd_dev
->parent_overlap
);
2682 length
= rbd_dev
->parent_overlap
- img_offset
;
2686 * Allocate a page array big enough to receive the data read
2689 page_count
= (u32
)calc_pages_for(0, length
);
2690 pages
= ceph_alloc_page_vector(page_count
, GFP_KERNEL
);
2691 if (IS_ERR(pages
)) {
2692 result
= PTR_ERR(pages
);
2698 parent_request
= rbd_parent_request_create(obj_request
,
2699 img_offset
, length
);
2700 if (!parent_request
)
2703 result
= rbd_img_request_fill(parent_request
, OBJ_REQUEST_PAGES
, pages
);
2706 parent_request
->copyup_pages
= pages
;
2707 parent_request
->copyup_page_count
= page_count
;
2709 parent_request
->callback
= rbd_img_obj_parent_read_full_callback
;
2710 result
= rbd_img_request_submit(parent_request
);
2714 parent_request
->copyup_pages
= NULL
;
2715 parent_request
->copyup_page_count
= 0;
2716 parent_request
->obj_request
= NULL
;
2717 rbd_obj_request_put(obj_request
);
2720 ceph_release_page_vector(pages
, page_count
);
2722 rbd_img_request_put(parent_request
);
2723 obj_request
->result
= result
;
2724 obj_request
->xferred
= 0;
2725 obj_request_done_set(obj_request
);
2730 static void rbd_img_obj_exists_callback(struct rbd_obj_request
*obj_request
)
2732 struct rbd_obj_request
*orig_request
;
2733 struct rbd_device
*rbd_dev
;
2736 rbd_assert(!obj_request_img_data_test(obj_request
));
2739 * All we need from the object request is the original
2740 * request and the result of the STAT op. Grab those, then
2741 * we're done with the request.
2743 orig_request
= obj_request
->obj_request
;
2744 obj_request
->obj_request
= NULL
;
2745 rbd_obj_request_put(orig_request
);
2746 rbd_assert(orig_request
);
2747 rbd_assert(orig_request
->img_request
);
2749 result
= obj_request
->result
;
2750 obj_request
->result
= 0;
2752 dout("%s: obj %p for obj %p result %d %llu/%llu\n", __func__
,
2753 obj_request
, orig_request
, result
,
2754 obj_request
->xferred
, obj_request
->length
);
2755 rbd_obj_request_put(obj_request
);
2758 * If the overlap has become 0 (most likely because the
2759 * image has been flattened) we need to free the pages
2760 * and re-submit the original write request.
2762 rbd_dev
= orig_request
->img_request
->rbd_dev
;
2763 if (!rbd_dev
->parent_overlap
) {
2764 struct ceph_osd_client
*osdc
;
2766 osdc
= &rbd_dev
->rbd_client
->client
->osdc
;
2767 result
= rbd_obj_request_submit(osdc
, orig_request
);
2773 * Our only purpose here is to determine whether the object
2774 * exists, and we don't want to treat the non-existence as
2775 * an error. If something else comes back, transfer the
2776 * error to the original request and complete it now.
2779 obj_request_existence_set(orig_request
, true);
2780 } else if (result
== -ENOENT
) {
2781 obj_request_existence_set(orig_request
, false);
2782 } else if (result
) {
2783 orig_request
->result
= result
;
2788 * Resubmit the original request now that we have recorded
2789 * whether the target object exists.
2791 orig_request
->result
= rbd_img_obj_request_submit(orig_request
);
2793 if (orig_request
->result
)
2794 rbd_obj_request_complete(orig_request
);
2797 static int rbd_img_obj_exists_submit(struct rbd_obj_request
*obj_request
)
2799 struct rbd_obj_request
*stat_request
;
2800 struct rbd_device
*rbd_dev
;
2801 struct ceph_osd_client
*osdc
;
2802 struct page
**pages
= NULL
;
2808 * The response data for a STAT call consists of:
2815 size
= sizeof (__le64
) + sizeof (__le32
) + sizeof (__le32
);
2816 page_count
= (u32
)calc_pages_for(0, size
);
2817 pages
= ceph_alloc_page_vector(page_count
, GFP_KERNEL
);
2819 return PTR_ERR(pages
);
2822 stat_request
= rbd_obj_request_create(obj_request
->object_name
, 0, 0,
2827 rbd_obj_request_get(obj_request
);
2828 stat_request
->obj_request
= obj_request
;
2829 stat_request
->pages
= pages
;
2830 stat_request
->page_count
= page_count
;
2832 rbd_assert(obj_request
->img_request
);
2833 rbd_dev
= obj_request
->img_request
->rbd_dev
;
2834 stat_request
->osd_req
= rbd_osd_req_create(rbd_dev
, OBJ_OP_READ
, 1,
2836 if (!stat_request
->osd_req
)
2838 stat_request
->callback
= rbd_img_obj_exists_callback
;
2840 osd_req_op_init(stat_request
->osd_req
, 0, CEPH_OSD_OP_STAT
);
2841 osd_req_op_raw_data_in_pages(stat_request
->osd_req
, 0, pages
, size
, 0,
2843 rbd_osd_req_format_read(stat_request
);
2845 osdc
= &rbd_dev
->rbd_client
->client
->osdc
;
2846 ret
= rbd_obj_request_submit(osdc
, stat_request
);
2849 rbd_obj_request_put(obj_request
);
2854 static bool img_obj_request_simple(struct rbd_obj_request
*obj_request
)
2856 struct rbd_img_request
*img_request
;
2857 struct rbd_device
*rbd_dev
;
2859 rbd_assert(obj_request_img_data_test(obj_request
));
2861 img_request
= obj_request
->img_request
;
2862 rbd_assert(img_request
);
2863 rbd_dev
= img_request
->rbd_dev
;
2866 if (!img_request_write_test(img_request
) &&
2867 !img_request_discard_test(img_request
))
2870 /* Non-layered writes */
2871 if (!img_request_layered_test(img_request
))
2875 * Layered writes outside of the parent overlap range don't
2876 * share any data with the parent.
2878 if (!obj_request_overlaps_parent(obj_request
))
2882 * Entire-object layered writes - we will overwrite whatever
2883 * parent data there is anyway.
2885 if (!obj_request
->offset
&&
2886 obj_request
->length
== rbd_obj_bytes(&rbd_dev
->header
))
2890 * If the object is known to already exist, its parent data has
2891 * already been copied.
2893 if (obj_request_known_test(obj_request
) &&
2894 obj_request_exists_test(obj_request
))
2900 static int rbd_img_obj_request_submit(struct rbd_obj_request
*obj_request
)
2902 if (img_obj_request_simple(obj_request
)) {
2903 struct rbd_device
*rbd_dev
;
2904 struct ceph_osd_client
*osdc
;
2906 rbd_dev
= obj_request
->img_request
->rbd_dev
;
2907 osdc
= &rbd_dev
->rbd_client
->client
->osdc
;
2909 return rbd_obj_request_submit(osdc
, obj_request
);
2913 * It's a layered write. The target object might exist but
2914 * we may not know that yet. If we know it doesn't exist,
2915 * start by reading the data for the full target object from
2916 * the parent so we can use it for a copyup to the target.
2918 if (obj_request_known_test(obj_request
))
2919 return rbd_img_obj_parent_read_full(obj_request
);
2921 /* We don't know whether the target exists. Go find out. */
2923 return rbd_img_obj_exists_submit(obj_request
);
2926 static int rbd_img_request_submit(struct rbd_img_request
*img_request
)
2928 struct rbd_obj_request
*obj_request
;
2929 struct rbd_obj_request
*next_obj_request
;
2931 dout("%s: img %p\n", __func__
, img_request
);
2932 for_each_obj_request_safe(img_request
, obj_request
, next_obj_request
) {
2935 ret
= rbd_img_obj_request_submit(obj_request
);
2943 static void rbd_img_parent_read_callback(struct rbd_img_request
*img_request
)
2945 struct rbd_obj_request
*obj_request
;
2946 struct rbd_device
*rbd_dev
;
2951 rbd_assert(img_request_child_test(img_request
));
2953 /* First get what we need from the image request and release it */
2955 obj_request
= img_request
->obj_request
;
2956 img_xferred
= img_request
->xferred
;
2957 img_result
= img_request
->result
;
2958 rbd_img_request_put(img_request
);
2961 * If the overlap has become 0 (most likely because the
2962 * image has been flattened) we need to re-submit the
2965 rbd_assert(obj_request
);
2966 rbd_assert(obj_request
->img_request
);
2967 rbd_dev
= obj_request
->img_request
->rbd_dev
;
2968 if (!rbd_dev
->parent_overlap
) {
2969 struct ceph_osd_client
*osdc
;
2971 osdc
= &rbd_dev
->rbd_client
->client
->osdc
;
2972 img_result
= rbd_obj_request_submit(osdc
, obj_request
);
2977 obj_request
->result
= img_result
;
2978 if (obj_request
->result
)
2982 * We need to zero anything beyond the parent overlap
2983 * boundary. Since rbd_img_obj_request_read_callback()
2984 * will zero anything beyond the end of a short read, an
2985 * easy way to do this is to pretend the data from the
2986 * parent came up short--ending at the overlap boundary.
2988 rbd_assert(obj_request
->img_offset
< U64_MAX
- obj_request
->length
);
2989 obj_end
= obj_request
->img_offset
+ obj_request
->length
;
2990 if (obj_end
> rbd_dev
->parent_overlap
) {
2993 if (obj_request
->img_offset
< rbd_dev
->parent_overlap
)
2994 xferred
= rbd_dev
->parent_overlap
-
2995 obj_request
->img_offset
;
2997 obj_request
->xferred
= min(img_xferred
, xferred
);
2999 obj_request
->xferred
= img_xferred
;
3002 rbd_img_obj_request_read_callback(obj_request
);
3003 rbd_obj_request_complete(obj_request
);
3006 static void rbd_img_parent_read(struct rbd_obj_request
*obj_request
)
3008 struct rbd_img_request
*img_request
;
3011 rbd_assert(obj_request_img_data_test(obj_request
));
3012 rbd_assert(obj_request
->img_request
!= NULL
);
3013 rbd_assert(obj_request
->result
== (s32
) -ENOENT
);
3014 rbd_assert(obj_request_type_valid(obj_request
->type
));
3016 /* rbd_read_finish(obj_request, obj_request->length); */
3017 img_request
= rbd_parent_request_create(obj_request
,
3018 obj_request
->img_offset
,
3019 obj_request
->length
);
3024 if (obj_request
->type
== OBJ_REQUEST_BIO
)
3025 result
= rbd_img_request_fill(img_request
, OBJ_REQUEST_BIO
,
3026 obj_request
->bio_list
);
3028 result
= rbd_img_request_fill(img_request
, OBJ_REQUEST_PAGES
,
3029 obj_request
->pages
);
3033 img_request
->callback
= rbd_img_parent_read_callback
;
3034 result
= rbd_img_request_submit(img_request
);
3041 rbd_img_request_put(img_request
);
3042 obj_request
->result
= result
;
3043 obj_request
->xferred
= 0;
3044 obj_request_done_set(obj_request
);
3047 static int rbd_obj_notify_ack_sync(struct rbd_device
*rbd_dev
, u64 notify_id
)
3049 struct rbd_obj_request
*obj_request
;
3050 struct ceph_osd_client
*osdc
= &rbd_dev
->rbd_client
->client
->osdc
;
3053 obj_request
= rbd_obj_request_create(rbd_dev
->header_name
, 0, 0,
3054 OBJ_REQUEST_NODATA
);
3059 obj_request
->osd_req
= rbd_osd_req_create(rbd_dev
, OBJ_OP_READ
, 1,
3061 if (!obj_request
->osd_req
)
3064 osd_req_op_watch_init(obj_request
->osd_req
, 0, CEPH_OSD_OP_NOTIFY_ACK
,
3066 rbd_osd_req_format_read(obj_request
);
3068 ret
= rbd_obj_request_submit(osdc
, obj_request
);
3071 ret
= rbd_obj_request_wait(obj_request
);
3073 rbd_obj_request_put(obj_request
);
3078 static void rbd_watch_cb(u64 ver
, u64 notify_id
, u8 opcode
, void *data
)
3080 struct rbd_device
*rbd_dev
= (struct rbd_device
*)data
;
3086 dout("%s: \"%s\" notify_id %llu opcode %u\n", __func__
,
3087 rbd_dev
->header_name
, (unsigned long long)notify_id
,
3088 (unsigned int)opcode
);
3091 * Until adequate refresh error handling is in place, there is
3092 * not much we can do here, except warn.
3094 * See http://tracker.ceph.com/issues/5040
3096 ret
= rbd_dev_refresh(rbd_dev
);
3098 rbd_warn(rbd_dev
, "refresh failed: %d", ret
);
3100 ret
= rbd_obj_notify_ack_sync(rbd_dev
, notify_id
);
3102 rbd_warn(rbd_dev
, "notify_ack ret %d", ret
);
3106 * Send a (un)watch request and wait for the ack. Return a request
3107 * with a ref held on success or error.
3109 static struct rbd_obj_request
*rbd_obj_watch_request_helper(
3110 struct rbd_device
*rbd_dev
,
3113 struct ceph_osd_client
*osdc
= &rbd_dev
->rbd_client
->client
->osdc
;
3114 struct rbd_obj_request
*obj_request
;
3117 obj_request
= rbd_obj_request_create(rbd_dev
->header_name
, 0, 0,
3118 OBJ_REQUEST_NODATA
);
3120 return ERR_PTR(-ENOMEM
);
3122 obj_request
->osd_req
= rbd_osd_req_create(rbd_dev
, OBJ_OP_WRITE
, 1,
3124 if (!obj_request
->osd_req
) {
3129 osd_req_op_watch_init(obj_request
->osd_req
, 0, CEPH_OSD_OP_WATCH
,
3130 rbd_dev
->watch_event
->cookie
, 0, watch
);
3131 rbd_osd_req_format_write(obj_request
);
3134 ceph_osdc_set_request_linger(osdc
, obj_request
->osd_req
);
3136 ret
= rbd_obj_request_submit(osdc
, obj_request
);
3140 ret
= rbd_obj_request_wait(obj_request
);
3144 ret
= obj_request
->result
;
3147 rbd_obj_request_end(obj_request
);
3154 rbd_obj_request_put(obj_request
);
3155 return ERR_PTR(ret
);
3159 * Initiate a watch request, synchronously.
3161 static int rbd_dev_header_watch_sync(struct rbd_device
*rbd_dev
)
3163 struct ceph_osd_client
*osdc
= &rbd_dev
->rbd_client
->client
->osdc
;
3164 struct rbd_obj_request
*obj_request
;
3167 rbd_assert(!rbd_dev
->watch_event
);
3168 rbd_assert(!rbd_dev
->watch_request
);
3170 ret
= ceph_osdc_create_event(osdc
, rbd_watch_cb
, rbd_dev
,
3171 &rbd_dev
->watch_event
);
3175 obj_request
= rbd_obj_watch_request_helper(rbd_dev
, true);
3176 if (IS_ERR(obj_request
)) {
3177 ceph_osdc_cancel_event(rbd_dev
->watch_event
);
3178 rbd_dev
->watch_event
= NULL
;
3179 return PTR_ERR(obj_request
);
3183 * A watch request is set to linger, so the underlying osd
3184 * request won't go away until we unregister it. We retain
3185 * a pointer to the object request during that time (in
3186 * rbd_dev->watch_request), so we'll keep a reference to it.
3187 * We'll drop that reference after we've unregistered it in
3188 * rbd_dev_header_unwatch_sync().
3190 rbd_dev
->watch_request
= obj_request
;
3196 * Tear down a watch request, synchronously.
3198 static void rbd_dev_header_unwatch_sync(struct rbd_device
*rbd_dev
)
3200 struct rbd_obj_request
*obj_request
;
3202 rbd_assert(rbd_dev
->watch_event
);
3203 rbd_assert(rbd_dev
->watch_request
);
3205 rbd_obj_request_end(rbd_dev
->watch_request
);
3206 rbd_obj_request_put(rbd_dev
->watch_request
);
3207 rbd_dev
->watch_request
= NULL
;
3209 obj_request
= rbd_obj_watch_request_helper(rbd_dev
, false);
3210 if (!IS_ERR(obj_request
))
3211 rbd_obj_request_put(obj_request
);
3213 rbd_warn(rbd_dev
, "unable to tear down watch request (%ld)",
3214 PTR_ERR(obj_request
));
3216 ceph_osdc_cancel_event(rbd_dev
->watch_event
);
3217 rbd_dev
->watch_event
= NULL
;
3221 * Synchronous osd object method call. Returns the number of bytes
3222 * returned in the outbound buffer, or a negative error code.
3224 static int rbd_obj_method_sync(struct rbd_device
*rbd_dev
,
3225 const char *object_name
,
3226 const char *class_name
,
3227 const char *method_name
,
3228 const void *outbound
,
3229 size_t outbound_size
,
3231 size_t inbound_size
)
3233 struct ceph_osd_client
*osdc
= &rbd_dev
->rbd_client
->client
->osdc
;
3234 struct rbd_obj_request
*obj_request
;
3235 struct page
**pages
;
3240 * Method calls are ultimately read operations. The result
3241 * should placed into the inbound buffer provided. They
3242 * also supply outbound data--parameters for the object
3243 * method. Currently if this is present it will be a
3246 page_count
= (u32
)calc_pages_for(0, inbound_size
);
3247 pages
= ceph_alloc_page_vector(page_count
, GFP_KERNEL
);
3249 return PTR_ERR(pages
);
3252 obj_request
= rbd_obj_request_create(object_name
, 0, inbound_size
,
3257 obj_request
->pages
= pages
;
3258 obj_request
->page_count
= page_count
;
3260 obj_request
->osd_req
= rbd_osd_req_create(rbd_dev
, OBJ_OP_READ
, 1,
3262 if (!obj_request
->osd_req
)
3265 osd_req_op_cls_init(obj_request
->osd_req
, 0, CEPH_OSD_OP_CALL
,
3266 class_name
, method_name
);
3267 if (outbound_size
) {
3268 struct ceph_pagelist
*pagelist
;
3270 pagelist
= kmalloc(sizeof (*pagelist
), GFP_NOFS
);
3274 ceph_pagelist_init(pagelist
);
3275 ceph_pagelist_append(pagelist
, outbound
, outbound_size
);
3276 osd_req_op_cls_request_data_pagelist(obj_request
->osd_req
, 0,
3279 osd_req_op_cls_response_data_pages(obj_request
->osd_req
, 0,
3280 obj_request
->pages
, inbound_size
,
3282 rbd_osd_req_format_read(obj_request
);
3284 ret
= rbd_obj_request_submit(osdc
, obj_request
);
3287 ret
= rbd_obj_request_wait(obj_request
);
3291 ret
= obj_request
->result
;
3295 rbd_assert(obj_request
->xferred
< (u64
)INT_MAX
);
3296 ret
= (int)obj_request
->xferred
;
3297 ceph_copy_from_page_vector(pages
, inbound
, 0, obj_request
->xferred
);
3300 rbd_obj_request_put(obj_request
);
3302 ceph_release_page_vector(pages
, page_count
);
3307 static void rbd_handle_request(struct rbd_device
*rbd_dev
, struct request
*rq
)
3309 struct rbd_img_request
*img_request
;
3310 struct ceph_snap_context
*snapc
= NULL
;
3311 u64 offset
= (u64
)blk_rq_pos(rq
) << SECTOR_SHIFT
;
3312 u64 length
= blk_rq_bytes(rq
);
3313 enum obj_operation_type op_type
;
3317 if (rq
->cmd_flags
& REQ_DISCARD
)
3318 op_type
= OBJ_OP_DISCARD
;
3319 else if (rq
->cmd_flags
& REQ_WRITE
)
3320 op_type
= OBJ_OP_WRITE
;
3322 op_type
= OBJ_OP_READ
;
3324 /* Ignore/skip any zero-length requests */
3327 dout("%s: zero-length request\n", __func__
);
3332 /* Only reads are allowed to a read-only device */
3334 if (op_type
!= OBJ_OP_READ
) {
3335 if (rbd_dev
->mapping
.read_only
) {
3339 rbd_assert(rbd_dev
->spec
->snap_id
== CEPH_NOSNAP
);
3343 * Quit early if the mapped snapshot no longer exists. It's
3344 * still possible the snapshot will have disappeared by the
3345 * time our request arrives at the osd, but there's no sense in
3346 * sending it if we already know.
3348 if (!test_bit(RBD_DEV_FLAG_EXISTS
, &rbd_dev
->flags
)) {
3349 dout("request for non-existent snapshot");
3350 rbd_assert(rbd_dev
->spec
->snap_id
!= CEPH_NOSNAP
);
3355 if (offset
&& length
> U64_MAX
- offset
+ 1) {
3356 rbd_warn(rbd_dev
, "bad request range (%llu~%llu)", offset
,
3359 goto err_rq
; /* Shouldn't happen */
3362 down_read(&rbd_dev
->header_rwsem
);
3363 mapping_size
= rbd_dev
->mapping
.size
;
3364 if (op_type
!= OBJ_OP_READ
) {
3365 snapc
= rbd_dev
->header
.snapc
;
3366 ceph_get_snap_context(snapc
);
3368 up_read(&rbd_dev
->header_rwsem
);
3370 if (offset
+ length
> mapping_size
) {
3371 rbd_warn(rbd_dev
, "beyond EOD (%llu~%llu > %llu)", offset
,
3372 length
, mapping_size
);
3377 img_request
= rbd_img_request_create(rbd_dev
, offset
, length
, op_type
,
3383 img_request
->rq
= rq
;
3385 if (op_type
== OBJ_OP_DISCARD
)
3386 result
= rbd_img_request_fill(img_request
, OBJ_REQUEST_NODATA
,
3389 result
= rbd_img_request_fill(img_request
, OBJ_REQUEST_BIO
,
3392 goto err_img_request
;
3394 result
= rbd_img_request_submit(img_request
);
3396 goto err_img_request
;
3401 rbd_img_request_put(img_request
);
3404 rbd_warn(rbd_dev
, "%s %llx at %llx result %d",
3405 obj_op_name(op_type
), length
, offset
, result
);
3406 ceph_put_snap_context(snapc
);
3407 blk_end_request_all(rq
, result
);
3410 static void rbd_request_workfn(struct work_struct
*work
)
3412 struct rbd_device
*rbd_dev
=
3413 container_of(work
, struct rbd_device
, rq_work
);
3414 struct request
*rq
, *next
;
3415 LIST_HEAD(requests
);
3417 spin_lock_irq(&rbd_dev
->lock
); /* rq->q->queue_lock */
3418 list_splice_init(&rbd_dev
->rq_queue
, &requests
);
3419 spin_unlock_irq(&rbd_dev
->lock
);
3421 list_for_each_entry_safe(rq
, next
, &requests
, queuelist
) {
3422 list_del_init(&rq
->queuelist
);
3423 rbd_handle_request(rbd_dev
, rq
);
3428 * Called with q->queue_lock held and interrupts disabled, possibly on
3429 * the way to schedule(). Do not sleep here!
3431 static void rbd_request_fn(struct request_queue
*q
)
3433 struct rbd_device
*rbd_dev
= q
->queuedata
;
3437 rbd_assert(rbd_dev
);
3439 while ((rq
= blk_fetch_request(q
))) {
3440 /* Ignore any non-FS requests that filter through. */
3441 if (rq
->cmd_type
!= REQ_TYPE_FS
) {
3442 dout("%s: non-fs request type %d\n", __func__
,
3443 (int) rq
->cmd_type
);
3444 __blk_end_request_all(rq
, 0);
3448 list_add_tail(&rq
->queuelist
, &rbd_dev
->rq_queue
);
3453 queue_work(rbd_wq
, &rbd_dev
->rq_work
);
3457 * a queue callback. Makes sure that we don't create a bio that spans across
3458 * multiple osd objects. One exception would be with a single page bios,
3459 * which we handle later at bio_chain_clone_range()
3461 static int rbd_merge_bvec(struct request_queue
*q
, struct bvec_merge_data
*bmd
,
3462 struct bio_vec
*bvec
)
3464 struct rbd_device
*rbd_dev
= q
->queuedata
;
3465 sector_t sector_offset
;
3466 sector_t sectors_per_obj
;
3467 sector_t obj_sector_offset
;
3471 * Find how far into its rbd object the partition-relative
3472 * bio start sector is to offset relative to the enclosing
3475 sector_offset
= get_start_sect(bmd
->bi_bdev
) + bmd
->bi_sector
;
3476 sectors_per_obj
= 1 << (rbd_dev
->header
.obj_order
- SECTOR_SHIFT
);
3477 obj_sector_offset
= sector_offset
& (sectors_per_obj
- 1);
3480 * Compute the number of bytes from that offset to the end
3481 * of the object. Account for what's already used by the bio.
3483 ret
= (int) (sectors_per_obj
- obj_sector_offset
) << SECTOR_SHIFT
;
3484 if (ret
> bmd
->bi_size
)
3485 ret
-= bmd
->bi_size
;
3490 * Don't send back more than was asked for. And if the bio
3491 * was empty, let the whole thing through because: "Note
3492 * that a block device *must* allow a single page to be
3493 * added to an empty bio."
3495 rbd_assert(bvec
->bv_len
<= PAGE_SIZE
);
3496 if (ret
> (int) bvec
->bv_len
|| !bmd
->bi_size
)
3497 ret
= (int) bvec
->bv_len
;
3502 static void rbd_free_disk(struct rbd_device
*rbd_dev
)
3504 struct gendisk
*disk
= rbd_dev
->disk
;
3509 rbd_dev
->disk
= NULL
;
3510 if (disk
->flags
& GENHD_FL_UP
) {
3513 blk_cleanup_queue(disk
->queue
);
3518 static int rbd_obj_read_sync(struct rbd_device
*rbd_dev
,
3519 const char *object_name
,
3520 u64 offset
, u64 length
, void *buf
)
3523 struct ceph_osd_client
*osdc
= &rbd_dev
->rbd_client
->client
->osdc
;
3524 struct rbd_obj_request
*obj_request
;
3525 struct page
**pages
= NULL
;
3530 page_count
= (u32
) calc_pages_for(offset
, length
);
3531 pages
= ceph_alloc_page_vector(page_count
, GFP_KERNEL
);
3533 return PTR_ERR(pages
);
3536 obj_request
= rbd_obj_request_create(object_name
, offset
, length
,
3541 obj_request
->pages
= pages
;
3542 obj_request
->page_count
= page_count
;
3544 obj_request
->osd_req
= rbd_osd_req_create(rbd_dev
, OBJ_OP_READ
, 1,
3546 if (!obj_request
->osd_req
)
3549 osd_req_op_extent_init(obj_request
->osd_req
, 0, CEPH_OSD_OP_READ
,
3550 offset
, length
, 0, 0);
3551 osd_req_op_extent_osd_data_pages(obj_request
->osd_req
, 0,
3553 obj_request
->length
,
3554 obj_request
->offset
& ~PAGE_MASK
,
3556 rbd_osd_req_format_read(obj_request
);
3558 ret
= rbd_obj_request_submit(osdc
, obj_request
);
3561 ret
= rbd_obj_request_wait(obj_request
);
3565 ret
= obj_request
->result
;
3569 rbd_assert(obj_request
->xferred
<= (u64
) SIZE_MAX
);
3570 size
= (size_t) obj_request
->xferred
;
3571 ceph_copy_from_page_vector(pages
, buf
, 0, size
);
3572 rbd_assert(size
<= (size_t)INT_MAX
);
3576 rbd_obj_request_put(obj_request
);
3578 ceph_release_page_vector(pages
, page_count
);
3584 * Read the complete header for the given rbd device. On successful
3585 * return, the rbd_dev->header field will contain up-to-date
3586 * information about the image.
3588 static int rbd_dev_v1_header_info(struct rbd_device
*rbd_dev
)
3590 struct rbd_image_header_ondisk
*ondisk
= NULL
;
3597 * The complete header will include an array of its 64-bit
3598 * snapshot ids, followed by the names of those snapshots as
3599 * a contiguous block of NUL-terminated strings. Note that
3600 * the number of snapshots could change by the time we read
3601 * it in, in which case we re-read it.
3608 size
= sizeof (*ondisk
);
3609 size
+= snap_count
* sizeof (struct rbd_image_snap_ondisk
);
3611 ondisk
= kmalloc(size
, GFP_KERNEL
);
3615 ret
= rbd_obj_read_sync(rbd_dev
, rbd_dev
->header_name
,
3619 if ((size_t)ret
< size
) {
3621 rbd_warn(rbd_dev
, "short header read (want %zd got %d)",
3625 if (!rbd_dev_ondisk_valid(ondisk
)) {
3627 rbd_warn(rbd_dev
, "invalid header");
3631 names_size
= le64_to_cpu(ondisk
->snap_names_len
);
3632 want_count
= snap_count
;
3633 snap_count
= le32_to_cpu(ondisk
->snap_count
);
3634 } while (snap_count
!= want_count
);
3636 ret
= rbd_header_from_disk(rbd_dev
, ondisk
);
3644 * Clear the rbd device's EXISTS flag if the snapshot it's mapped to
3645 * has disappeared from the (just updated) snapshot context.
3647 static void rbd_exists_validate(struct rbd_device
*rbd_dev
)
3651 if (!test_bit(RBD_DEV_FLAG_EXISTS
, &rbd_dev
->flags
))
3654 snap_id
= rbd_dev
->spec
->snap_id
;
3655 if (snap_id
== CEPH_NOSNAP
)
3658 if (rbd_dev_snap_index(rbd_dev
, snap_id
) == BAD_SNAP_INDEX
)
3659 clear_bit(RBD_DEV_FLAG_EXISTS
, &rbd_dev
->flags
);
3662 static void rbd_dev_update_size(struct rbd_device
*rbd_dev
)
3668 * Don't hold the lock while doing disk operations,
3669 * or lock ordering will conflict with the bdev mutex via:
3670 * rbd_add() -> blkdev_get() -> rbd_open()
3672 spin_lock_irq(&rbd_dev
->lock
);
3673 removing
= test_bit(RBD_DEV_FLAG_REMOVING
, &rbd_dev
->flags
);
3674 spin_unlock_irq(&rbd_dev
->lock
);
3676 * If the device is being removed, rbd_dev->disk has
3677 * been destroyed, so don't try to update its size
3680 size
= (sector_t
)rbd_dev
->mapping
.size
/ SECTOR_SIZE
;
3681 dout("setting size to %llu sectors", (unsigned long long)size
);
3682 set_capacity(rbd_dev
->disk
, size
);
3683 revalidate_disk(rbd_dev
->disk
);
3687 static int rbd_dev_refresh(struct rbd_device
*rbd_dev
)
3692 down_write(&rbd_dev
->header_rwsem
);
3693 mapping_size
= rbd_dev
->mapping
.size
;
3695 ret
= rbd_dev_header_info(rbd_dev
);
3700 * If there is a parent, see if it has disappeared due to the
3701 * mapped image getting flattened.
3703 if (rbd_dev
->parent
) {
3704 ret
= rbd_dev_v2_parent_info(rbd_dev
);
3709 if (rbd_dev
->spec
->snap_id
== CEPH_NOSNAP
) {
3710 if (rbd_dev
->mapping
.size
!= rbd_dev
->header
.image_size
)
3711 rbd_dev
->mapping
.size
= rbd_dev
->header
.image_size
;
3713 /* validate mapped snapshot's EXISTS flag */
3714 rbd_exists_validate(rbd_dev
);
3717 up_write(&rbd_dev
->header_rwsem
);
3719 if (mapping_size
!= rbd_dev
->mapping
.size
)
3720 rbd_dev_update_size(rbd_dev
);
3725 static int rbd_init_disk(struct rbd_device
*rbd_dev
)
3727 struct gendisk
*disk
;
3728 struct request_queue
*q
;
3731 /* create gendisk info */
3732 disk
= alloc_disk(single_major
?
3733 (1 << RBD_SINGLE_MAJOR_PART_SHIFT
) :
3734 RBD_MINORS_PER_MAJOR
);
3738 snprintf(disk
->disk_name
, sizeof(disk
->disk_name
), RBD_DRV_NAME
"%d",
3740 disk
->major
= rbd_dev
->major
;
3741 disk
->first_minor
= rbd_dev
->minor
;
3743 disk
->flags
|= GENHD_FL_EXT_DEVT
;
3744 disk
->fops
= &rbd_bd_ops
;
3745 disk
->private_data
= rbd_dev
;
3747 q
= blk_init_queue(rbd_request_fn
, &rbd_dev
->lock
);
3751 /* We use the default size, but let's be explicit about it. */
3752 blk_queue_physical_block_size(q
, SECTOR_SIZE
);
3754 /* set io sizes to object size */
3755 segment_size
= rbd_obj_bytes(&rbd_dev
->header
);
3756 blk_queue_max_hw_sectors(q
, segment_size
/ SECTOR_SIZE
);
3757 blk_queue_max_segment_size(q
, segment_size
);
3758 blk_queue_io_min(q
, segment_size
);
3759 blk_queue_io_opt(q
, segment_size
);
3761 /* enable the discard support */
3762 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD
, q
);
3763 q
->limits
.discard_granularity
= segment_size
;
3764 q
->limits
.discard_alignment
= segment_size
;
3765 q
->limits
.max_discard_sectors
= segment_size
/ SECTOR_SIZE
;
3766 q
->limits
.discard_zeroes_data
= 1;
3768 blk_queue_merge_bvec(q
, rbd_merge_bvec
);
3771 q
->queuedata
= rbd_dev
;
3773 rbd_dev
->disk
= disk
;
3786 static struct rbd_device
*dev_to_rbd_dev(struct device
*dev
)
3788 return container_of(dev
, struct rbd_device
, dev
);
3791 static ssize_t
rbd_size_show(struct device
*dev
,
3792 struct device_attribute
*attr
, char *buf
)
3794 struct rbd_device
*rbd_dev
= dev_to_rbd_dev(dev
);
3796 return sprintf(buf
, "%llu\n",
3797 (unsigned long long)rbd_dev
->mapping
.size
);
3801 * Note this shows the features for whatever's mapped, which is not
3802 * necessarily the base image.
3804 static ssize_t
rbd_features_show(struct device
*dev
,
3805 struct device_attribute
*attr
, char *buf
)
3807 struct rbd_device
*rbd_dev
= dev_to_rbd_dev(dev
);
3809 return sprintf(buf
, "0x%016llx\n",
3810 (unsigned long long)rbd_dev
->mapping
.features
);
3813 static ssize_t
rbd_major_show(struct device
*dev
,
3814 struct device_attribute
*attr
, char *buf
)
3816 struct rbd_device
*rbd_dev
= dev_to_rbd_dev(dev
);
3819 return sprintf(buf
, "%d\n", rbd_dev
->major
);
3821 return sprintf(buf
, "(none)\n");
3824 static ssize_t
rbd_minor_show(struct device
*dev
,
3825 struct device_attribute
*attr
, char *buf
)
3827 struct rbd_device
*rbd_dev
= dev_to_rbd_dev(dev
);
3829 return sprintf(buf
, "%d\n", rbd_dev
->minor
);
3832 static ssize_t
rbd_client_id_show(struct device
*dev
,
3833 struct device_attribute
*attr
, char *buf
)
3835 struct rbd_device
*rbd_dev
= dev_to_rbd_dev(dev
);
3837 return sprintf(buf
, "client%lld\n",
3838 ceph_client_id(rbd_dev
->rbd_client
->client
));
3841 static ssize_t
rbd_pool_show(struct device
*dev
,
3842 struct device_attribute
*attr
, char *buf
)
3844 struct rbd_device
*rbd_dev
= dev_to_rbd_dev(dev
);
3846 return sprintf(buf
, "%s\n", rbd_dev
->spec
->pool_name
);
3849 static ssize_t
rbd_pool_id_show(struct device
*dev
,
3850 struct device_attribute
*attr
, char *buf
)
3852 struct rbd_device
*rbd_dev
= dev_to_rbd_dev(dev
);
3854 return sprintf(buf
, "%llu\n",
3855 (unsigned long long) rbd_dev
->spec
->pool_id
);
3858 static ssize_t
rbd_name_show(struct device
*dev
,
3859 struct device_attribute
*attr
, char *buf
)
3861 struct rbd_device
*rbd_dev
= dev_to_rbd_dev(dev
);
3863 if (rbd_dev
->spec
->image_name
)
3864 return sprintf(buf
, "%s\n", rbd_dev
->spec
->image_name
);
3866 return sprintf(buf
, "(unknown)\n");
3869 static ssize_t
rbd_image_id_show(struct device
*dev
,
3870 struct device_attribute
*attr
, char *buf
)
3872 struct rbd_device
*rbd_dev
= dev_to_rbd_dev(dev
);
3874 return sprintf(buf
, "%s\n", rbd_dev
->spec
->image_id
);
3878 * Shows the name of the currently-mapped snapshot (or
3879 * RBD_SNAP_HEAD_NAME for the base image).
3881 static ssize_t
rbd_snap_show(struct device
*dev
,
3882 struct device_attribute
*attr
,
3885 struct rbd_device
*rbd_dev
= dev_to_rbd_dev(dev
);
3887 return sprintf(buf
, "%s\n", rbd_dev
->spec
->snap_name
);
3891 * For a v2 image, shows the chain of parent images, separated by empty
3892 * lines. For v1 images or if there is no parent, shows "(no parent
3895 static ssize_t
rbd_parent_show(struct device
*dev
,
3896 struct device_attribute
*attr
,
3899 struct rbd_device
*rbd_dev
= dev_to_rbd_dev(dev
);
3902 if (!rbd_dev
->parent
)
3903 return sprintf(buf
, "(no parent image)\n");
3905 for ( ; rbd_dev
->parent
; rbd_dev
= rbd_dev
->parent
) {
3906 struct rbd_spec
*spec
= rbd_dev
->parent_spec
;
3908 count
+= sprintf(&buf
[count
], "%s"
3909 "pool_id %llu\npool_name %s\n"
3910 "image_id %s\nimage_name %s\n"
3911 "snap_id %llu\nsnap_name %s\n"
3913 !count
? "" : "\n", /* first? */
3914 spec
->pool_id
, spec
->pool_name
,
3915 spec
->image_id
, spec
->image_name
?: "(unknown)",
3916 spec
->snap_id
, spec
->snap_name
,
3917 rbd_dev
->parent_overlap
);
3923 static ssize_t
rbd_image_refresh(struct device
*dev
,
3924 struct device_attribute
*attr
,
3928 struct rbd_device
*rbd_dev
= dev_to_rbd_dev(dev
);
3931 ret
= rbd_dev_refresh(rbd_dev
);
3938 static DEVICE_ATTR(size
, S_IRUGO
, rbd_size_show
, NULL
);
3939 static DEVICE_ATTR(features
, S_IRUGO
, rbd_features_show
, NULL
);
3940 static DEVICE_ATTR(major
, S_IRUGO
, rbd_major_show
, NULL
);
3941 static DEVICE_ATTR(minor
, S_IRUGO
, rbd_minor_show
, NULL
);
3942 static DEVICE_ATTR(client_id
, S_IRUGO
, rbd_client_id_show
, NULL
);
3943 static DEVICE_ATTR(pool
, S_IRUGO
, rbd_pool_show
, NULL
);
3944 static DEVICE_ATTR(pool_id
, S_IRUGO
, rbd_pool_id_show
, NULL
);
3945 static DEVICE_ATTR(name
, S_IRUGO
, rbd_name_show
, NULL
);
3946 static DEVICE_ATTR(image_id
, S_IRUGO
, rbd_image_id_show
, NULL
);
3947 static DEVICE_ATTR(refresh
, S_IWUSR
, NULL
, rbd_image_refresh
);
3948 static DEVICE_ATTR(current_snap
, S_IRUGO
, rbd_snap_show
, NULL
);
3949 static DEVICE_ATTR(parent
, S_IRUGO
, rbd_parent_show
, NULL
);
3951 static struct attribute
*rbd_attrs
[] = {
3952 &dev_attr_size
.attr
,
3953 &dev_attr_features
.attr
,
3954 &dev_attr_major
.attr
,
3955 &dev_attr_minor
.attr
,
3956 &dev_attr_client_id
.attr
,
3957 &dev_attr_pool
.attr
,
3958 &dev_attr_pool_id
.attr
,
3959 &dev_attr_name
.attr
,
3960 &dev_attr_image_id
.attr
,
3961 &dev_attr_current_snap
.attr
,
3962 &dev_attr_parent
.attr
,
3963 &dev_attr_refresh
.attr
,
3967 static struct attribute_group rbd_attr_group
= {
3971 static const struct attribute_group
*rbd_attr_groups
[] = {
3976 static void rbd_sysfs_dev_release(struct device
*dev
)
3980 static struct device_type rbd_device_type
= {
3982 .groups
= rbd_attr_groups
,
3983 .release
= rbd_sysfs_dev_release
,
3986 static struct rbd_spec
*rbd_spec_get(struct rbd_spec
*spec
)
3988 kref_get(&spec
->kref
);
3993 static void rbd_spec_free(struct kref
*kref
);
3994 static void rbd_spec_put(struct rbd_spec
*spec
)
3997 kref_put(&spec
->kref
, rbd_spec_free
);
4000 static struct rbd_spec
*rbd_spec_alloc(void)
4002 struct rbd_spec
*spec
;
4004 spec
= kzalloc(sizeof (*spec
), GFP_KERNEL
);
4008 spec
->pool_id
= CEPH_NOPOOL
;
4009 spec
->snap_id
= CEPH_NOSNAP
;
4010 kref_init(&spec
->kref
);
4015 static void rbd_spec_free(struct kref
*kref
)
4017 struct rbd_spec
*spec
= container_of(kref
, struct rbd_spec
, kref
);
4019 kfree(spec
->pool_name
);
4020 kfree(spec
->image_id
);
4021 kfree(spec
->image_name
);
4022 kfree(spec
->snap_name
);
4026 static struct rbd_device
*rbd_dev_create(struct rbd_client
*rbdc
,
4027 struct rbd_spec
*spec
)
4029 struct rbd_device
*rbd_dev
;
4031 rbd_dev
= kzalloc(sizeof (*rbd_dev
), GFP_KERNEL
);
4035 spin_lock_init(&rbd_dev
->lock
);
4036 INIT_LIST_HEAD(&rbd_dev
->rq_queue
);
4037 INIT_WORK(&rbd_dev
->rq_work
, rbd_request_workfn
);
4039 atomic_set(&rbd_dev
->parent_ref
, 0);
4040 INIT_LIST_HEAD(&rbd_dev
->node
);
4041 init_rwsem(&rbd_dev
->header_rwsem
);
4043 rbd_dev
->spec
= spec
;
4044 rbd_dev
->rbd_client
= rbdc
;
4046 /* Initialize the layout used for all rbd requests */
4048 rbd_dev
->layout
.fl_stripe_unit
= cpu_to_le32(1 << RBD_MAX_OBJ_ORDER
);
4049 rbd_dev
->layout
.fl_stripe_count
= cpu_to_le32(1);
4050 rbd_dev
->layout
.fl_object_size
= cpu_to_le32(1 << RBD_MAX_OBJ_ORDER
);
4051 rbd_dev
->layout
.fl_pg_pool
= cpu_to_le32((u32
) spec
->pool_id
);
4056 static void rbd_dev_destroy(struct rbd_device
*rbd_dev
)
4058 rbd_put_client(rbd_dev
->rbd_client
);
4059 rbd_spec_put(rbd_dev
->spec
);
4064 * Get the size and object order for an image snapshot, or if
4065 * snap_id is CEPH_NOSNAP, gets this information for the base
4068 static int _rbd_dev_v2_snap_size(struct rbd_device
*rbd_dev
, u64 snap_id
,
4069 u8
*order
, u64
*snap_size
)
4071 __le64 snapid
= cpu_to_le64(snap_id
);
4076 } __attribute__ ((packed
)) size_buf
= { 0 };
4078 ret
= rbd_obj_method_sync(rbd_dev
, rbd_dev
->header_name
,
4080 &snapid
, sizeof (snapid
),
4081 &size_buf
, sizeof (size_buf
));
4082 dout("%s: rbd_obj_method_sync returned %d\n", __func__
, ret
);
4085 if (ret
< sizeof (size_buf
))
4089 *order
= size_buf
.order
;
4090 dout(" order %u", (unsigned int)*order
);
4092 *snap_size
= le64_to_cpu(size_buf
.size
);
4094 dout(" snap_id 0x%016llx snap_size = %llu\n",
4095 (unsigned long long)snap_id
,
4096 (unsigned long long)*snap_size
);
4101 static int rbd_dev_v2_image_size(struct rbd_device
*rbd_dev
)
4103 return _rbd_dev_v2_snap_size(rbd_dev
, CEPH_NOSNAP
,
4104 &rbd_dev
->header
.obj_order
,
4105 &rbd_dev
->header
.image_size
);
4108 static int rbd_dev_v2_object_prefix(struct rbd_device
*rbd_dev
)
4114 reply_buf
= kzalloc(RBD_OBJ_PREFIX_LEN_MAX
, GFP_KERNEL
);
4118 ret
= rbd_obj_method_sync(rbd_dev
, rbd_dev
->header_name
,
4119 "rbd", "get_object_prefix", NULL
, 0,
4120 reply_buf
, RBD_OBJ_PREFIX_LEN_MAX
);
4121 dout("%s: rbd_obj_method_sync returned %d\n", __func__
, ret
);
4126 rbd_dev
->header
.object_prefix
= ceph_extract_encoded_string(&p
,
4127 p
+ ret
, NULL
, GFP_NOIO
);
4130 if (IS_ERR(rbd_dev
->header
.object_prefix
)) {
4131 ret
= PTR_ERR(rbd_dev
->header
.object_prefix
);
4132 rbd_dev
->header
.object_prefix
= NULL
;
4134 dout(" object_prefix = %s\n", rbd_dev
->header
.object_prefix
);
4142 static int _rbd_dev_v2_snap_features(struct rbd_device
*rbd_dev
, u64 snap_id
,
4145 __le64 snapid
= cpu_to_le64(snap_id
);
4149 } __attribute__ ((packed
)) features_buf
= { 0 };
4153 ret
= rbd_obj_method_sync(rbd_dev
, rbd_dev
->header_name
,
4154 "rbd", "get_features",
4155 &snapid
, sizeof (snapid
),
4156 &features_buf
, sizeof (features_buf
));
4157 dout("%s: rbd_obj_method_sync returned %d\n", __func__
, ret
);
4160 if (ret
< sizeof (features_buf
))
4163 incompat
= le64_to_cpu(features_buf
.incompat
);
4164 if (incompat
& ~RBD_FEATURES_SUPPORTED
)
4167 *snap_features
= le64_to_cpu(features_buf
.features
);
4169 dout(" snap_id 0x%016llx features = 0x%016llx incompat = 0x%016llx\n",
4170 (unsigned long long)snap_id
,
4171 (unsigned long long)*snap_features
,
4172 (unsigned long long)le64_to_cpu(features_buf
.incompat
));
4177 static int rbd_dev_v2_features(struct rbd_device
*rbd_dev
)
4179 return _rbd_dev_v2_snap_features(rbd_dev
, CEPH_NOSNAP
,
4180 &rbd_dev
->header
.features
);
4183 static int rbd_dev_v2_parent_info(struct rbd_device
*rbd_dev
)
4185 struct rbd_spec
*parent_spec
;
4187 void *reply_buf
= NULL
;
4197 parent_spec
= rbd_spec_alloc();
4201 size
= sizeof (__le64
) + /* pool_id */
4202 sizeof (__le32
) + RBD_IMAGE_ID_LEN_MAX
+ /* image_id */
4203 sizeof (__le64
) + /* snap_id */
4204 sizeof (__le64
); /* overlap */
4205 reply_buf
= kmalloc(size
, GFP_KERNEL
);
4211 snapid
= cpu_to_le64(rbd_dev
->spec
->snap_id
);
4212 ret
= rbd_obj_method_sync(rbd_dev
, rbd_dev
->header_name
,
4213 "rbd", "get_parent",
4214 &snapid
, sizeof (snapid
),
4216 dout("%s: rbd_obj_method_sync returned %d\n", __func__
, ret
);
4221 end
= reply_buf
+ ret
;
4223 ceph_decode_64_safe(&p
, end
, pool_id
, out_err
);
4224 if (pool_id
== CEPH_NOPOOL
) {
4226 * Either the parent never existed, or we have
4227 * record of it but the image got flattened so it no
4228 * longer has a parent. When the parent of a
4229 * layered image disappears we immediately set the
4230 * overlap to 0. The effect of this is that all new
4231 * requests will be treated as if the image had no
4234 if (rbd_dev
->parent_overlap
) {
4235 rbd_dev
->parent_overlap
= 0;
4236 rbd_dev_parent_put(rbd_dev
);
4237 pr_info("%s: clone image has been flattened\n",
4238 rbd_dev
->disk
->disk_name
);
4241 goto out
; /* No parent? No problem. */
4244 /* The ceph file layout needs to fit pool id in 32 bits */
4247 if (pool_id
> (u64
)U32_MAX
) {
4248 rbd_warn(NULL
, "parent pool id too large (%llu > %u)",
4249 (unsigned long long)pool_id
, U32_MAX
);
4253 image_id
= ceph_extract_encoded_string(&p
, end
, NULL
, GFP_KERNEL
);
4254 if (IS_ERR(image_id
)) {
4255 ret
= PTR_ERR(image_id
);
4258 ceph_decode_64_safe(&p
, end
, snap_id
, out_err
);
4259 ceph_decode_64_safe(&p
, end
, overlap
, out_err
);
4262 * The parent won't change (except when the clone is
4263 * flattened, already handled that). So we only need to
4264 * record the parent spec we have not already done so.
4266 if (!rbd_dev
->parent_spec
) {
4267 parent_spec
->pool_id
= pool_id
;
4268 parent_spec
->image_id
= image_id
;
4269 parent_spec
->snap_id
= snap_id
;
4270 rbd_dev
->parent_spec
= parent_spec
;
4271 parent_spec
= NULL
; /* rbd_dev now owns this */
4277 * We always update the parent overlap. If it's zero we
4278 * treat it specially.
4280 rbd_dev
->parent_overlap
= overlap
;
4283 /* A null parent_spec indicates it's the initial probe */
4287 * The overlap has become zero, so the clone
4288 * must have been resized down to 0 at some
4289 * point. Treat this the same as a flatten.
4291 rbd_dev_parent_put(rbd_dev
);
4292 pr_info("%s: clone image now standalone\n",
4293 rbd_dev
->disk
->disk_name
);
4296 * For the initial probe, if we find the
4297 * overlap is zero we just pretend there was
4300 rbd_warn(rbd_dev
, "ignoring parent with overlap 0");
4307 rbd_spec_put(parent_spec
);
4312 static int rbd_dev_v2_striping_info(struct rbd_device
*rbd_dev
)
4316 __le64 stripe_count
;
4317 } __attribute__ ((packed
)) striping_info_buf
= { 0 };
4318 size_t size
= sizeof (striping_info_buf
);
4325 ret
= rbd_obj_method_sync(rbd_dev
, rbd_dev
->header_name
,
4326 "rbd", "get_stripe_unit_count", NULL
, 0,
4327 (char *)&striping_info_buf
, size
);
4328 dout("%s: rbd_obj_method_sync returned %d\n", __func__
, ret
);
4335 * We don't actually support the "fancy striping" feature
4336 * (STRIPINGV2) yet, but if the striping sizes are the
4337 * defaults the behavior is the same as before. So find
4338 * out, and only fail if the image has non-default values.
4341 obj_size
= (u64
)1 << rbd_dev
->header
.obj_order
;
4342 p
= &striping_info_buf
;
4343 stripe_unit
= ceph_decode_64(&p
);
4344 if (stripe_unit
!= obj_size
) {
4345 rbd_warn(rbd_dev
, "unsupported stripe unit "
4346 "(got %llu want %llu)",
4347 stripe_unit
, obj_size
);
4350 stripe_count
= ceph_decode_64(&p
);
4351 if (stripe_count
!= 1) {
4352 rbd_warn(rbd_dev
, "unsupported stripe count "
4353 "(got %llu want 1)", stripe_count
);
4356 rbd_dev
->header
.stripe_unit
= stripe_unit
;
4357 rbd_dev
->header
.stripe_count
= stripe_count
;
4362 static char *rbd_dev_image_name(struct rbd_device
*rbd_dev
)
4364 size_t image_id_size
;
4369 void *reply_buf
= NULL
;
4371 char *image_name
= NULL
;
4374 rbd_assert(!rbd_dev
->spec
->image_name
);
4376 len
= strlen(rbd_dev
->spec
->image_id
);
4377 image_id_size
= sizeof (__le32
) + len
;
4378 image_id
= kmalloc(image_id_size
, GFP_KERNEL
);
4383 end
= image_id
+ image_id_size
;
4384 ceph_encode_string(&p
, end
, rbd_dev
->spec
->image_id
, (u32
)len
);
4386 size
= sizeof (__le32
) + RBD_IMAGE_NAME_LEN_MAX
;
4387 reply_buf
= kmalloc(size
, GFP_KERNEL
);
4391 ret
= rbd_obj_method_sync(rbd_dev
, RBD_DIRECTORY
,
4392 "rbd", "dir_get_name",
4393 image_id
, image_id_size
,
4398 end
= reply_buf
+ ret
;
4400 image_name
= ceph_extract_encoded_string(&p
, end
, &len
, GFP_KERNEL
);
4401 if (IS_ERR(image_name
))
4404 dout("%s: name is %s len is %zd\n", __func__
, image_name
, len
);
4412 static u64
rbd_v1_snap_id_by_name(struct rbd_device
*rbd_dev
, const char *name
)
4414 struct ceph_snap_context
*snapc
= rbd_dev
->header
.snapc
;
4415 const char *snap_name
;
4418 /* Skip over names until we find the one we are looking for */
4420 snap_name
= rbd_dev
->header
.snap_names
;
4421 while (which
< snapc
->num_snaps
) {
4422 if (!strcmp(name
, snap_name
))
4423 return snapc
->snaps
[which
];
4424 snap_name
+= strlen(snap_name
) + 1;
4430 static u64
rbd_v2_snap_id_by_name(struct rbd_device
*rbd_dev
, const char *name
)
4432 struct ceph_snap_context
*snapc
= rbd_dev
->header
.snapc
;
4437 for (which
= 0; !found
&& which
< snapc
->num_snaps
; which
++) {
4438 const char *snap_name
;
4440 snap_id
= snapc
->snaps
[which
];
4441 snap_name
= rbd_dev_v2_snap_name(rbd_dev
, snap_id
);
4442 if (IS_ERR(snap_name
)) {
4443 /* ignore no-longer existing snapshots */
4444 if (PTR_ERR(snap_name
) == -ENOENT
)
4449 found
= !strcmp(name
, snap_name
);
4452 return found
? snap_id
: CEPH_NOSNAP
;
4456 * Assumes name is never RBD_SNAP_HEAD_NAME; returns CEPH_NOSNAP if
4457 * no snapshot by that name is found, or if an error occurs.
4459 static u64
rbd_snap_id_by_name(struct rbd_device
*rbd_dev
, const char *name
)
4461 if (rbd_dev
->image_format
== 1)
4462 return rbd_v1_snap_id_by_name(rbd_dev
, name
);
4464 return rbd_v2_snap_id_by_name(rbd_dev
, name
);
4468 * An image being mapped will have everything but the snap id.
4470 static int rbd_spec_fill_snap_id(struct rbd_device
*rbd_dev
)
4472 struct rbd_spec
*spec
= rbd_dev
->spec
;
4474 rbd_assert(spec
->pool_id
!= CEPH_NOPOOL
&& spec
->pool_name
);
4475 rbd_assert(spec
->image_id
&& spec
->image_name
);
4476 rbd_assert(spec
->snap_name
);
4478 if (strcmp(spec
->snap_name
, RBD_SNAP_HEAD_NAME
)) {
4481 snap_id
= rbd_snap_id_by_name(rbd_dev
, spec
->snap_name
);
4482 if (snap_id
== CEPH_NOSNAP
)
4485 spec
->snap_id
= snap_id
;
4487 spec
->snap_id
= CEPH_NOSNAP
;
4494 * A parent image will have all ids but none of the names.
4496 * All names in an rbd spec are dynamically allocated. It's OK if we
4497 * can't figure out the name for an image id.
4499 static int rbd_spec_fill_names(struct rbd_device
*rbd_dev
)
4501 struct ceph_osd_client
*osdc
= &rbd_dev
->rbd_client
->client
->osdc
;
4502 struct rbd_spec
*spec
= rbd_dev
->spec
;
4503 const char *pool_name
;
4504 const char *image_name
;
4505 const char *snap_name
;
4508 rbd_assert(spec
->pool_id
!= CEPH_NOPOOL
);
4509 rbd_assert(spec
->image_id
);
4510 rbd_assert(spec
->snap_id
!= CEPH_NOSNAP
);
4512 /* Get the pool name; we have to make our own copy of this */
4514 pool_name
= ceph_pg_pool_name_by_id(osdc
->osdmap
, spec
->pool_id
);
4516 rbd_warn(rbd_dev
, "no pool with id %llu", spec
->pool_id
);
4519 pool_name
= kstrdup(pool_name
, GFP_KERNEL
);
4523 /* Fetch the image name; tolerate failure here */
4525 image_name
= rbd_dev_image_name(rbd_dev
);
4527 rbd_warn(rbd_dev
, "unable to get image name");
4529 /* Fetch the snapshot name */
4531 snap_name
= rbd_snap_name(rbd_dev
, spec
->snap_id
);
4532 if (IS_ERR(snap_name
)) {
4533 ret
= PTR_ERR(snap_name
);
4537 spec
->pool_name
= pool_name
;
4538 spec
->image_name
= image_name
;
4539 spec
->snap_name
= snap_name
;
4549 static int rbd_dev_v2_snap_context(struct rbd_device
*rbd_dev
)
4558 struct ceph_snap_context
*snapc
;
4562 * We'll need room for the seq value (maximum snapshot id),
4563 * snapshot count, and array of that many snapshot ids.
4564 * For now we have a fixed upper limit on the number we're
4565 * prepared to receive.
4567 size
= sizeof (__le64
) + sizeof (__le32
) +
4568 RBD_MAX_SNAP_COUNT
* sizeof (__le64
);
4569 reply_buf
= kzalloc(size
, GFP_KERNEL
);
4573 ret
= rbd_obj_method_sync(rbd_dev
, rbd_dev
->header_name
,
4574 "rbd", "get_snapcontext", NULL
, 0,
4576 dout("%s: rbd_obj_method_sync returned %d\n", __func__
, ret
);
4581 end
= reply_buf
+ ret
;
4583 ceph_decode_64_safe(&p
, end
, seq
, out
);
4584 ceph_decode_32_safe(&p
, end
, snap_count
, out
);
4587 * Make sure the reported number of snapshot ids wouldn't go
4588 * beyond the end of our buffer. But before checking that,
4589 * make sure the computed size of the snapshot context we
4590 * allocate is representable in a size_t.
4592 if (snap_count
> (SIZE_MAX
- sizeof (struct ceph_snap_context
))
4597 if (!ceph_has_room(&p
, end
, snap_count
* sizeof (__le64
)))
4601 snapc
= ceph_create_snap_context(snap_count
, GFP_KERNEL
);
4607 for (i
= 0; i
< snap_count
; i
++)
4608 snapc
->snaps
[i
] = ceph_decode_64(&p
);
4610 ceph_put_snap_context(rbd_dev
->header
.snapc
);
4611 rbd_dev
->header
.snapc
= snapc
;
4613 dout(" snap context seq = %llu, snap_count = %u\n",
4614 (unsigned long long)seq
, (unsigned int)snap_count
);
4621 static const char *rbd_dev_v2_snap_name(struct rbd_device
*rbd_dev
,
4632 size
= sizeof (__le32
) + RBD_MAX_SNAP_NAME_LEN
;
4633 reply_buf
= kmalloc(size
, GFP_KERNEL
);
4635 return ERR_PTR(-ENOMEM
);
4637 snapid
= cpu_to_le64(snap_id
);
4638 ret
= rbd_obj_method_sync(rbd_dev
, rbd_dev
->header_name
,
4639 "rbd", "get_snapshot_name",
4640 &snapid
, sizeof (snapid
),
4642 dout("%s: rbd_obj_method_sync returned %d\n", __func__
, ret
);
4644 snap_name
= ERR_PTR(ret
);
4649 end
= reply_buf
+ ret
;
4650 snap_name
= ceph_extract_encoded_string(&p
, end
, NULL
, GFP_KERNEL
);
4651 if (IS_ERR(snap_name
))
4654 dout(" snap_id 0x%016llx snap_name = %s\n",
4655 (unsigned long long)snap_id
, snap_name
);
4662 static int rbd_dev_v2_header_info(struct rbd_device
*rbd_dev
)
4664 bool first_time
= rbd_dev
->header
.object_prefix
== NULL
;
4667 ret
= rbd_dev_v2_image_size(rbd_dev
);
4672 ret
= rbd_dev_v2_header_onetime(rbd_dev
);
4677 ret
= rbd_dev_v2_snap_context(rbd_dev
);
4678 dout("rbd_dev_v2_snap_context returned %d\n", ret
);
4683 static int rbd_dev_header_info(struct rbd_device
*rbd_dev
)
4685 rbd_assert(rbd_image_format_valid(rbd_dev
->image_format
));
4687 if (rbd_dev
->image_format
== 1)
4688 return rbd_dev_v1_header_info(rbd_dev
);
4690 return rbd_dev_v2_header_info(rbd_dev
);
4693 static int rbd_bus_add_dev(struct rbd_device
*rbd_dev
)
4698 dev
= &rbd_dev
->dev
;
4699 dev
->bus
= &rbd_bus_type
;
4700 dev
->type
= &rbd_device_type
;
4701 dev
->parent
= &rbd_root_dev
;
4702 dev
->release
= rbd_dev_device_release
;
4703 dev_set_name(dev
, "%d", rbd_dev
->dev_id
);
4704 ret
= device_register(dev
);
4709 static void rbd_bus_del_dev(struct rbd_device
*rbd_dev
)
4711 device_unregister(&rbd_dev
->dev
);
4715 * Get a unique rbd identifier for the given new rbd_dev, and add
4716 * the rbd_dev to the global list.
4718 static int rbd_dev_id_get(struct rbd_device
*rbd_dev
)
4722 new_dev_id
= ida_simple_get(&rbd_dev_id_ida
,
4723 0, minor_to_rbd_dev_id(1 << MINORBITS
),
4728 rbd_dev
->dev_id
= new_dev_id
;
4730 spin_lock(&rbd_dev_list_lock
);
4731 list_add_tail(&rbd_dev
->node
, &rbd_dev_list
);
4732 spin_unlock(&rbd_dev_list_lock
);
4734 dout("rbd_dev %p given dev id %d\n", rbd_dev
, rbd_dev
->dev_id
);
4740 * Remove an rbd_dev from the global list, and record that its
4741 * identifier is no longer in use.
4743 static void rbd_dev_id_put(struct rbd_device
*rbd_dev
)
4745 spin_lock(&rbd_dev_list_lock
);
4746 list_del_init(&rbd_dev
->node
);
4747 spin_unlock(&rbd_dev_list_lock
);
4749 ida_simple_remove(&rbd_dev_id_ida
, rbd_dev
->dev_id
);
4751 dout("rbd_dev %p released dev id %d\n", rbd_dev
, rbd_dev
->dev_id
);
4755 * Skips over white space at *buf, and updates *buf to point to the
4756 * first found non-space character (if any). Returns the length of
4757 * the token (string of non-white space characters) found. Note
4758 * that *buf must be terminated with '\0'.
4760 static inline size_t next_token(const char **buf
)
4763 * These are the characters that produce nonzero for
4764 * isspace() in the "C" and "POSIX" locales.
4766 const char *spaces
= " \f\n\r\t\v";
4768 *buf
+= strspn(*buf
, spaces
); /* Find start of token */
4770 return strcspn(*buf
, spaces
); /* Return token length */
4774 * Finds the next token in *buf, and if the provided token buffer is
4775 * big enough, copies the found token into it. The result, if
4776 * copied, is guaranteed to be terminated with '\0'. Note that *buf
4777 * must be terminated with '\0' on entry.
4779 * Returns the length of the token found (not including the '\0').
4780 * Return value will be 0 if no token is found, and it will be >=
4781 * token_size if the token would not fit.
4783 * The *buf pointer will be updated to point beyond the end of the
4784 * found token. Note that this occurs even if the token buffer is
4785 * too small to hold it.
4787 static inline size_t copy_token(const char **buf
,
4793 len
= next_token(buf
);
4794 if (len
< token_size
) {
4795 memcpy(token
, *buf
, len
);
4796 *(token
+ len
) = '\0';
4804 * Finds the next token in *buf, dynamically allocates a buffer big
4805 * enough to hold a copy of it, and copies the token into the new
4806 * buffer. The copy is guaranteed to be terminated with '\0'. Note
4807 * that a duplicate buffer is created even for a zero-length token.
4809 * Returns a pointer to the newly-allocated duplicate, or a null
4810 * pointer if memory for the duplicate was not available. If
4811 * the lenp argument is a non-null pointer, the length of the token
4812 * (not including the '\0') is returned in *lenp.
4814 * If successful, the *buf pointer will be updated to point beyond
4815 * the end of the found token.
4817 * Note: uses GFP_KERNEL for allocation.
4819 static inline char *dup_token(const char **buf
, size_t *lenp
)
4824 len
= next_token(buf
);
4825 dup
= kmemdup(*buf
, len
+ 1, GFP_KERNEL
);
4828 *(dup
+ len
) = '\0';
4838 * Parse the options provided for an "rbd add" (i.e., rbd image
4839 * mapping) request. These arrive via a write to /sys/bus/rbd/add,
4840 * and the data written is passed here via a NUL-terminated buffer.
4841 * Returns 0 if successful or an error code otherwise.
4843 * The information extracted from these options is recorded in
4844 * the other parameters which return dynamically-allocated
4847 * The address of a pointer that will refer to a ceph options
4848 * structure. Caller must release the returned pointer using
4849 * ceph_destroy_options() when it is no longer needed.
4851 * Address of an rbd options pointer. Fully initialized by
4852 * this function; caller must release with kfree().
4854 * Address of an rbd image specification pointer. Fully
4855 * initialized by this function based on parsed options.
4856 * Caller must release with rbd_spec_put().
4858 * The options passed take this form:
4859 * <mon_addrs> <options> <pool_name> <image_name> [<snap_id>]
4862 * A comma-separated list of one or more monitor addresses.
4863 * A monitor address is an ip address, optionally followed
4864 * by a port number (separated by a colon).
4865 * I.e.: ip1[:port1][,ip2[:port2]...]
4867 * A comma-separated list of ceph and/or rbd options.
4869 * The name of the rados pool containing the rbd image.
4871 * The name of the image in that pool to map.
4873 * An optional snapshot id. If provided, the mapping will
4874 * present data from the image at the time that snapshot was
4875 * created. The image head is used if no snapshot id is
4876 * provided. Snapshot mappings are always read-only.
4878 static int rbd_add_parse_args(const char *buf
,
4879 struct ceph_options
**ceph_opts
,
4880 struct rbd_options
**opts
,
4881 struct rbd_spec
**rbd_spec
)
4885 const char *mon_addrs
;
4887 size_t mon_addrs_size
;
4888 struct rbd_spec
*spec
= NULL
;
4889 struct rbd_options
*rbd_opts
= NULL
;
4890 struct ceph_options
*copts
;
4893 /* The first four tokens are required */
4895 len
= next_token(&buf
);
4897 rbd_warn(NULL
, "no monitor address(es) provided");
4901 mon_addrs_size
= len
+ 1;
4905 options
= dup_token(&buf
, NULL
);
4909 rbd_warn(NULL
, "no options provided");
4913 spec
= rbd_spec_alloc();
4917 spec
->pool_name
= dup_token(&buf
, NULL
);
4918 if (!spec
->pool_name
)
4920 if (!*spec
->pool_name
) {
4921 rbd_warn(NULL
, "no pool name provided");
4925 spec
->image_name
= dup_token(&buf
, NULL
);
4926 if (!spec
->image_name
)
4928 if (!*spec
->image_name
) {
4929 rbd_warn(NULL
, "no image name provided");
4934 * Snapshot name is optional; default is to use "-"
4935 * (indicating the head/no snapshot).
4937 len
= next_token(&buf
);
4939 buf
= RBD_SNAP_HEAD_NAME
; /* No snapshot supplied */
4940 len
= sizeof (RBD_SNAP_HEAD_NAME
) - 1;
4941 } else if (len
> RBD_MAX_SNAP_NAME_LEN
) {
4942 ret
= -ENAMETOOLONG
;
4945 snap_name
= kmemdup(buf
, len
+ 1, GFP_KERNEL
);
4948 *(snap_name
+ len
) = '\0';
4949 spec
->snap_name
= snap_name
;
4951 /* Initialize all rbd options to the defaults */
4953 rbd_opts
= kzalloc(sizeof (*rbd_opts
), GFP_KERNEL
);
4957 rbd_opts
->read_only
= RBD_READ_ONLY_DEFAULT
;
4959 copts
= ceph_parse_options(options
, mon_addrs
,
4960 mon_addrs
+ mon_addrs_size
- 1,
4961 parse_rbd_opts_token
, rbd_opts
);
4962 if (IS_ERR(copts
)) {
4963 ret
= PTR_ERR(copts
);
4984 * Return pool id (>= 0) or a negative error code.
4986 static int rbd_add_get_pool_id(struct rbd_client
*rbdc
, const char *pool_name
)
4989 unsigned long timeout
= rbdc
->client
->options
->mount_timeout
* HZ
;
4994 ret
= ceph_pg_poolid_by_name(rbdc
->client
->osdc
.osdmap
, pool_name
);
4995 if (ret
== -ENOENT
&& tries
++ < 1) {
4996 ret
= ceph_monc_do_get_version(&rbdc
->client
->monc
, "osdmap",
5001 if (rbdc
->client
->osdc
.osdmap
->epoch
< newest_epoch
) {
5002 ceph_monc_request_next_osdmap(&rbdc
->client
->monc
);
5003 (void) ceph_monc_wait_osdmap(&rbdc
->client
->monc
,
5004 newest_epoch
, timeout
);
5007 /* the osdmap we have is new enough */
5016 * An rbd format 2 image has a unique identifier, distinct from the
5017 * name given to it by the user. Internally, that identifier is
5018 * what's used to specify the names of objects related to the image.
5020 * A special "rbd id" object is used to map an rbd image name to its
5021 * id. If that object doesn't exist, then there is no v2 rbd image
5022 * with the supplied name.
5024 * This function will record the given rbd_dev's image_id field if
5025 * it can be determined, and in that case will return 0. If any
5026 * errors occur a negative errno will be returned and the rbd_dev's
5027 * image_id field will be unchanged (and should be NULL).
5029 static int rbd_dev_image_id(struct rbd_device
*rbd_dev
)
5038 * When probing a parent image, the image id is already
5039 * known (and the image name likely is not). There's no
5040 * need to fetch the image id again in this case. We
5041 * do still need to set the image format though.
5043 if (rbd_dev
->spec
->image_id
) {
5044 rbd_dev
->image_format
= *rbd_dev
->spec
->image_id
? 2 : 1;
5050 * First, see if the format 2 image id file exists, and if
5051 * so, get the image's persistent id from it.
5053 size
= sizeof (RBD_ID_PREFIX
) + strlen(rbd_dev
->spec
->image_name
);
5054 object_name
= kmalloc(size
, GFP_NOIO
);
5057 sprintf(object_name
, "%s%s", RBD_ID_PREFIX
, rbd_dev
->spec
->image_name
);
5058 dout("rbd id object name is %s\n", object_name
);
5060 /* Response will be an encoded string, which includes a length */
5062 size
= sizeof (__le32
) + RBD_IMAGE_ID_LEN_MAX
;
5063 response
= kzalloc(size
, GFP_NOIO
);
5069 /* If it doesn't exist we'll assume it's a format 1 image */
5071 ret
= rbd_obj_method_sync(rbd_dev
, object_name
,
5072 "rbd", "get_id", NULL
, 0,
5073 response
, RBD_IMAGE_ID_LEN_MAX
);
5074 dout("%s: rbd_obj_method_sync returned %d\n", __func__
, ret
);
5075 if (ret
== -ENOENT
) {
5076 image_id
= kstrdup("", GFP_KERNEL
);
5077 ret
= image_id
? 0 : -ENOMEM
;
5079 rbd_dev
->image_format
= 1;
5080 } else if (ret
>= 0) {
5083 image_id
= ceph_extract_encoded_string(&p
, p
+ ret
,
5085 ret
= PTR_ERR_OR_ZERO(image_id
);
5087 rbd_dev
->image_format
= 2;
5091 rbd_dev
->spec
->image_id
= image_id
;
5092 dout("image_id is %s\n", image_id
);
5102 * Undo whatever state changes are made by v1 or v2 header info
5105 static void rbd_dev_unprobe(struct rbd_device
*rbd_dev
)
5107 struct rbd_image_header
*header
;
5109 rbd_dev_parent_put(rbd_dev
);
5111 /* Free dynamic fields from the header, then zero it out */
5113 header
= &rbd_dev
->header
;
5114 ceph_put_snap_context(header
->snapc
);
5115 kfree(header
->snap_sizes
);
5116 kfree(header
->snap_names
);
5117 kfree(header
->object_prefix
);
5118 memset(header
, 0, sizeof (*header
));
5121 static int rbd_dev_v2_header_onetime(struct rbd_device
*rbd_dev
)
5125 ret
= rbd_dev_v2_object_prefix(rbd_dev
);
5130 * Get the and check features for the image. Currently the
5131 * features are assumed to never change.
5133 ret
= rbd_dev_v2_features(rbd_dev
);
5137 /* If the image supports fancy striping, get its parameters */
5139 if (rbd_dev
->header
.features
& RBD_FEATURE_STRIPINGV2
) {
5140 ret
= rbd_dev_v2_striping_info(rbd_dev
);
5144 /* No support for crypto and compression type format 2 images */
5148 rbd_dev
->header
.features
= 0;
5149 kfree(rbd_dev
->header
.object_prefix
);
5150 rbd_dev
->header
.object_prefix
= NULL
;
5155 static int rbd_dev_probe_parent(struct rbd_device
*rbd_dev
)
5157 struct rbd_device
*parent
= NULL
;
5158 struct rbd_spec
*parent_spec
;
5159 struct rbd_client
*rbdc
;
5162 if (!rbd_dev
->parent_spec
)
5165 * We need to pass a reference to the client and the parent
5166 * spec when creating the parent rbd_dev. Images related by
5167 * parent/child relationships always share both.
5169 parent_spec
= rbd_spec_get(rbd_dev
->parent_spec
);
5170 rbdc
= __rbd_get_client(rbd_dev
->rbd_client
);
5173 parent
= rbd_dev_create(rbdc
, parent_spec
);
5177 ret
= rbd_dev_image_probe(parent
, false);
5180 rbd_dev
->parent
= parent
;
5181 atomic_set(&rbd_dev
->parent_ref
, 1);
5186 rbd_dev_unparent(rbd_dev
);
5187 kfree(rbd_dev
->header_name
);
5188 rbd_dev_destroy(parent
);
5190 rbd_put_client(rbdc
);
5191 rbd_spec_put(parent_spec
);
5197 static int rbd_dev_device_setup(struct rbd_device
*rbd_dev
)
5201 /* Get an id and fill in device name. */
5203 ret
= rbd_dev_id_get(rbd_dev
);
5207 BUILD_BUG_ON(DEV_NAME_LEN
5208 < sizeof (RBD_DRV_NAME
) + MAX_INT_FORMAT_WIDTH
);
5209 sprintf(rbd_dev
->name
, "%s%d", RBD_DRV_NAME
, rbd_dev
->dev_id
);
5211 /* Record our major and minor device numbers. */
5213 if (!single_major
) {
5214 ret
= register_blkdev(0, rbd_dev
->name
);
5218 rbd_dev
->major
= ret
;
5221 rbd_dev
->major
= rbd_major
;
5222 rbd_dev
->minor
= rbd_dev_id_to_minor(rbd_dev
->dev_id
);
5225 /* Set up the blkdev mapping. */
5227 ret
= rbd_init_disk(rbd_dev
);
5229 goto err_out_blkdev
;
5231 ret
= rbd_dev_mapping_set(rbd_dev
);
5235 set_capacity(rbd_dev
->disk
, rbd_dev
->mapping
.size
/ SECTOR_SIZE
);
5236 set_disk_ro(rbd_dev
->disk
, rbd_dev
->mapping
.read_only
);
5238 ret
= rbd_bus_add_dev(rbd_dev
);
5240 goto err_out_mapping
;
5242 /* Everything's ready. Announce the disk to the world. */
5244 set_bit(RBD_DEV_FLAG_EXISTS
, &rbd_dev
->flags
);
5245 add_disk(rbd_dev
->disk
);
5247 pr_info("%s: added with size 0x%llx\n", rbd_dev
->disk
->disk_name
,
5248 (unsigned long long) rbd_dev
->mapping
.size
);
5253 rbd_dev_mapping_clear(rbd_dev
);
5255 rbd_free_disk(rbd_dev
);
5258 unregister_blkdev(rbd_dev
->major
, rbd_dev
->name
);
5260 rbd_dev_id_put(rbd_dev
);
5261 rbd_dev_mapping_clear(rbd_dev
);
5266 static int rbd_dev_header_name(struct rbd_device
*rbd_dev
)
5268 struct rbd_spec
*spec
= rbd_dev
->spec
;
5271 /* Record the header object name for this rbd image. */
5273 rbd_assert(rbd_image_format_valid(rbd_dev
->image_format
));
5275 if (rbd_dev
->image_format
== 1)
5276 size
= strlen(spec
->image_name
) + sizeof (RBD_SUFFIX
);
5278 size
= sizeof (RBD_HEADER_PREFIX
) + strlen(spec
->image_id
);
5280 rbd_dev
->header_name
= kmalloc(size
, GFP_KERNEL
);
5281 if (!rbd_dev
->header_name
)
5284 if (rbd_dev
->image_format
== 1)
5285 sprintf(rbd_dev
->header_name
, "%s%s",
5286 spec
->image_name
, RBD_SUFFIX
);
5288 sprintf(rbd_dev
->header_name
, "%s%s",
5289 RBD_HEADER_PREFIX
, spec
->image_id
);
5293 static void rbd_dev_image_release(struct rbd_device
*rbd_dev
)
5295 rbd_dev_unprobe(rbd_dev
);
5296 kfree(rbd_dev
->header_name
);
5297 rbd_dev
->header_name
= NULL
;
5298 rbd_dev
->image_format
= 0;
5299 kfree(rbd_dev
->spec
->image_id
);
5300 rbd_dev
->spec
->image_id
= NULL
;
5302 rbd_dev_destroy(rbd_dev
);
5306 * Probe for the existence of the header object for the given rbd
5307 * device. If this image is the one being mapped (i.e., not a
5308 * parent), initiate a watch on its header object before using that
5309 * object to get detailed information about the rbd image.
5311 static int rbd_dev_image_probe(struct rbd_device
*rbd_dev
, bool mapping
)
5316 * Get the id from the image id object. Unless there's an
5317 * error, rbd_dev->spec->image_id will be filled in with
5318 * a dynamically-allocated string, and rbd_dev->image_format
5319 * will be set to either 1 or 2.
5321 ret
= rbd_dev_image_id(rbd_dev
);
5325 ret
= rbd_dev_header_name(rbd_dev
);
5327 goto err_out_format
;
5330 ret
= rbd_dev_header_watch_sync(rbd_dev
);
5332 goto out_header_name
;
5335 ret
= rbd_dev_header_info(rbd_dev
);
5340 * If this image is the one being mapped, we have pool name and
5341 * id, image name and id, and snap name - need to fill snap id.
5342 * Otherwise this is a parent image, identified by pool, image
5343 * and snap ids - need to fill in names for those ids.
5346 ret
= rbd_spec_fill_snap_id(rbd_dev
);
5348 ret
= rbd_spec_fill_names(rbd_dev
);
5352 if (rbd_dev
->header
.features
& RBD_FEATURE_LAYERING
) {
5353 ret
= rbd_dev_v2_parent_info(rbd_dev
);
5358 * Need to warn users if this image is the one being
5359 * mapped and has a parent.
5361 if (mapping
&& rbd_dev
->parent_spec
)
5363 "WARNING: kernel layering is EXPERIMENTAL!");
5366 ret
= rbd_dev_probe_parent(rbd_dev
);
5370 dout("discovered format %u image, header name is %s\n",
5371 rbd_dev
->image_format
, rbd_dev
->header_name
);
5375 rbd_dev_unprobe(rbd_dev
);
5378 rbd_dev_header_unwatch_sync(rbd_dev
);
5380 kfree(rbd_dev
->header_name
);
5381 rbd_dev
->header_name
= NULL
;
5383 rbd_dev
->image_format
= 0;
5384 kfree(rbd_dev
->spec
->image_id
);
5385 rbd_dev
->spec
->image_id
= NULL
;
5389 static ssize_t
do_rbd_add(struct bus_type
*bus
,
5393 struct rbd_device
*rbd_dev
= NULL
;
5394 struct ceph_options
*ceph_opts
= NULL
;
5395 struct rbd_options
*rbd_opts
= NULL
;
5396 struct rbd_spec
*spec
= NULL
;
5397 struct rbd_client
*rbdc
;
5401 if (!try_module_get(THIS_MODULE
))
5404 /* parse add command */
5405 rc
= rbd_add_parse_args(buf
, &ceph_opts
, &rbd_opts
, &spec
);
5407 goto err_out_module
;
5408 read_only
= rbd_opts
->read_only
;
5410 rbd_opts
= NULL
; /* done with this */
5412 rbdc
= rbd_get_client(ceph_opts
);
5419 rc
= rbd_add_get_pool_id(rbdc
, spec
->pool_name
);
5421 goto err_out_client
;
5422 spec
->pool_id
= (u64
)rc
;
5424 /* The ceph file layout needs to fit pool id in 32 bits */
5426 if (spec
->pool_id
> (u64
)U32_MAX
) {
5427 rbd_warn(NULL
, "pool id too large (%llu > %u)",
5428 (unsigned long long)spec
->pool_id
, U32_MAX
);
5430 goto err_out_client
;
5433 rbd_dev
= rbd_dev_create(rbdc
, spec
);
5435 goto err_out_client
;
5436 rbdc
= NULL
; /* rbd_dev now owns this */
5437 spec
= NULL
; /* rbd_dev now owns this */
5439 rc
= rbd_dev_image_probe(rbd_dev
, true);
5441 goto err_out_rbd_dev
;
5443 /* If we are mapping a snapshot it must be marked read-only */
5445 if (rbd_dev
->spec
->snap_id
!= CEPH_NOSNAP
)
5447 rbd_dev
->mapping
.read_only
= read_only
;
5449 rc
= rbd_dev_device_setup(rbd_dev
);
5452 * rbd_dev_header_unwatch_sync() can't be moved into
5453 * rbd_dev_image_release() without refactoring, see
5454 * commit 1f3ef78861ac.
5456 rbd_dev_header_unwatch_sync(rbd_dev
);
5457 rbd_dev_image_release(rbd_dev
);
5458 goto err_out_module
;
5464 rbd_dev_destroy(rbd_dev
);
5466 rbd_put_client(rbdc
);
5470 module_put(THIS_MODULE
);
5472 dout("Error adding device %s\n", buf
);
5477 static ssize_t
rbd_add(struct bus_type
*bus
,
5484 return do_rbd_add(bus
, buf
, count
);
5487 static ssize_t
rbd_add_single_major(struct bus_type
*bus
,
5491 return do_rbd_add(bus
, buf
, count
);
5494 static void rbd_dev_device_release(struct device
*dev
)
5496 struct rbd_device
*rbd_dev
= dev_to_rbd_dev(dev
);
5498 rbd_free_disk(rbd_dev
);
5499 clear_bit(RBD_DEV_FLAG_EXISTS
, &rbd_dev
->flags
);
5500 rbd_dev_mapping_clear(rbd_dev
);
5502 unregister_blkdev(rbd_dev
->major
, rbd_dev
->name
);
5503 rbd_dev_id_put(rbd_dev
);
5504 rbd_dev_mapping_clear(rbd_dev
);
5507 static void rbd_dev_remove_parent(struct rbd_device
*rbd_dev
)
5509 while (rbd_dev
->parent
) {
5510 struct rbd_device
*first
= rbd_dev
;
5511 struct rbd_device
*second
= first
->parent
;
5512 struct rbd_device
*third
;
5515 * Follow to the parent with no grandparent and
5518 while (second
&& (third
= second
->parent
)) {
5523 rbd_dev_image_release(second
);
5524 first
->parent
= NULL
;
5525 first
->parent_overlap
= 0;
5527 rbd_assert(first
->parent_spec
);
5528 rbd_spec_put(first
->parent_spec
);
5529 first
->parent_spec
= NULL
;
5533 static ssize_t
do_rbd_remove(struct bus_type
*bus
,
5537 struct rbd_device
*rbd_dev
= NULL
;
5538 struct list_head
*tmp
;
5541 bool already
= false;
5544 ret
= kstrtoul(buf
, 10, &ul
);
5548 /* convert to int; abort if we lost anything in the conversion */
5554 spin_lock(&rbd_dev_list_lock
);
5555 list_for_each(tmp
, &rbd_dev_list
) {
5556 rbd_dev
= list_entry(tmp
, struct rbd_device
, node
);
5557 if (rbd_dev
->dev_id
== dev_id
) {
5563 spin_lock_irq(&rbd_dev
->lock
);
5564 if (rbd_dev
->open_count
)
5567 already
= test_and_set_bit(RBD_DEV_FLAG_REMOVING
,
5569 spin_unlock_irq(&rbd_dev
->lock
);
5571 spin_unlock(&rbd_dev_list_lock
);
5572 if (ret
< 0 || already
)
5575 rbd_dev_header_unwatch_sync(rbd_dev
);
5577 * flush remaining watch callbacks - these must be complete
5578 * before the osd_client is shutdown
5580 dout("%s: flushing notifies", __func__
);
5581 ceph_osdc_flush_notifies(&rbd_dev
->rbd_client
->client
->osdc
);
5584 * Don't free anything from rbd_dev->disk until after all
5585 * notifies are completely processed. Otherwise
5586 * rbd_bus_del_dev() will race with rbd_watch_cb(), resulting
5587 * in a potential use after free of rbd_dev->disk or rbd_dev.
5589 rbd_bus_del_dev(rbd_dev
);
5590 rbd_dev_image_release(rbd_dev
);
5591 module_put(THIS_MODULE
);
5596 static ssize_t
rbd_remove(struct bus_type
*bus
,
5603 return do_rbd_remove(bus
, buf
, count
);
5606 static ssize_t
rbd_remove_single_major(struct bus_type
*bus
,
5610 return do_rbd_remove(bus
, buf
, count
);
5614 * create control files in sysfs
5617 static int rbd_sysfs_init(void)
5621 ret
= device_register(&rbd_root_dev
);
5625 ret
= bus_register(&rbd_bus_type
);
5627 device_unregister(&rbd_root_dev
);
5632 static void rbd_sysfs_cleanup(void)
5634 bus_unregister(&rbd_bus_type
);
5635 device_unregister(&rbd_root_dev
);
5638 static int rbd_slab_init(void)
5640 rbd_assert(!rbd_img_request_cache
);
5641 rbd_img_request_cache
= kmem_cache_create("rbd_img_request",
5642 sizeof (struct rbd_img_request
),
5643 __alignof__(struct rbd_img_request
),
5645 if (!rbd_img_request_cache
)
5648 rbd_assert(!rbd_obj_request_cache
);
5649 rbd_obj_request_cache
= kmem_cache_create("rbd_obj_request",
5650 sizeof (struct rbd_obj_request
),
5651 __alignof__(struct rbd_obj_request
),
5653 if (!rbd_obj_request_cache
)
5656 rbd_assert(!rbd_segment_name_cache
);
5657 rbd_segment_name_cache
= kmem_cache_create("rbd_segment_name",
5658 CEPH_MAX_OID_NAME_LEN
+ 1, 1, 0, NULL
);
5659 if (rbd_segment_name_cache
)
5662 if (rbd_obj_request_cache
) {
5663 kmem_cache_destroy(rbd_obj_request_cache
);
5664 rbd_obj_request_cache
= NULL
;
5667 kmem_cache_destroy(rbd_img_request_cache
);
5668 rbd_img_request_cache
= NULL
;
5673 static void rbd_slab_exit(void)
5675 rbd_assert(rbd_segment_name_cache
);
5676 kmem_cache_destroy(rbd_segment_name_cache
);
5677 rbd_segment_name_cache
= NULL
;
5679 rbd_assert(rbd_obj_request_cache
);
5680 kmem_cache_destroy(rbd_obj_request_cache
);
5681 rbd_obj_request_cache
= NULL
;
5683 rbd_assert(rbd_img_request_cache
);
5684 kmem_cache_destroy(rbd_img_request_cache
);
5685 rbd_img_request_cache
= NULL
;
5688 static int __init
rbd_init(void)
5692 if (!libceph_compatible(NULL
)) {
5693 rbd_warn(NULL
, "libceph incompatibility (quitting)");
5697 rc
= rbd_slab_init();
5702 * The number of active work items is limited by the number of
5703 * rbd devices, so leave @max_active at default.
5705 rbd_wq
= alloc_workqueue(RBD_DRV_NAME
, WQ_MEM_RECLAIM
, 0);
5712 rbd_major
= register_blkdev(0, RBD_DRV_NAME
);
5713 if (rbd_major
< 0) {
5719 rc
= rbd_sysfs_init();
5721 goto err_out_blkdev
;
5724 pr_info("loaded (major %d)\n", rbd_major
);
5726 pr_info("loaded\n");
5732 unregister_blkdev(rbd_major
, RBD_DRV_NAME
);
5734 destroy_workqueue(rbd_wq
);
5740 static void __exit
rbd_exit(void)
5742 ida_destroy(&rbd_dev_id_ida
);
5743 rbd_sysfs_cleanup();
5745 unregister_blkdev(rbd_major
, RBD_DRV_NAME
);
5746 destroy_workqueue(rbd_wq
);
5750 module_init(rbd_init
);
5751 module_exit(rbd_exit
);
5753 MODULE_AUTHOR("Alex Elder <elder@inktank.com>");
5754 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
5755 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
5756 /* following authorship retained from original osdblk.c */
5757 MODULE_AUTHOR("Jeff Garzik <jeff@garzik.org>");
5759 MODULE_DESCRIPTION("RADOS Block Device (RBD) driver");
5760 MODULE_LICENSE("GPL");