3 rbd.c -- Export ceph rados objects as a Linux block device
6 based on drivers/block/osdblk.c:
8 Copyright 2009 Red Hat, Inc.
10 This program is free software; you can redistribute it and/or modify
11 it under the terms of the GNU General Public License as published by
12 the Free Software Foundation.
14 This program is distributed in the hope that it will be useful,
15 but WITHOUT ANY WARRANTY; without even the implied warranty of
16 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
17 GNU General Public License for more details.
19 You should have received a copy of the GNU General Public License
20 along with this program; see the file COPYING. If not, write to
21 the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA.
25 For usage instructions, please refer to:
27 Documentation/ABI/testing/sysfs-bus-rbd
31 #include <linux/ceph/libceph.h>
32 #include <linux/ceph/osd_client.h>
33 #include <linux/ceph/mon_client.h>
34 #include <linux/ceph/decode.h>
35 #include <linux/parser.h>
36 #include <linux/bsearch.h>
38 #include <linux/kernel.h>
39 #include <linux/device.h>
40 #include <linux/module.h>
42 #include <linux/blkdev.h>
43 #include <linux/slab.h>
45 #include "rbd_types.h"
47 #define RBD_DEBUG /* Activate rbd_assert() calls */
50 * The basic unit of block I/O is a sector. It is interpreted in a
51 * number of contexts in Linux (blk, bio, genhd), but the default is
52 * universally 512 bytes. These symbols are just slightly more
53 * meaningful than the bare numbers they represent.
55 #define SECTOR_SHIFT 9
56 #define SECTOR_SIZE (1ULL << SECTOR_SHIFT)
59 * Increment the given counter and return its updated value.
60 * If the counter is already 0 it will not be incremented.
61 * If the counter is already at its maximum value returns
62 * -EINVAL without updating it.
64 static int atomic_inc_return_safe(atomic_t *v)
68 counter = (unsigned int)__atomic_add_unless(v, 1, 0);
69 if (counter <= (unsigned int)INT_MAX)
77 /* Decrement the counter. Return the resulting value, or -EINVAL */
78 static int atomic_dec_return_safe(atomic_t *v)
82 counter = atomic_dec_return(v);
91 #define RBD_DRV_NAME "rbd"
92 #define RBD_DRV_NAME_LONG "rbd (rados block device)"
94 #define RBD_MINORS_PER_MAJOR 256 /* max minors per blkdev */
96 #define RBD_SNAP_DEV_NAME_PREFIX "snap_"
97 #define RBD_MAX_SNAP_NAME_LEN \
98 (NAME_MAX - (sizeof (RBD_SNAP_DEV_NAME_PREFIX) - 1))
100 #define RBD_MAX_SNAP_COUNT 510 /* allows max snapc to fit in 4KB */
102 #define RBD_SNAP_HEAD_NAME "-"
104 #define BAD_SNAP_INDEX U32_MAX /* invalid index into snap array */
106 /* This allows a single page to hold an image name sent by OSD */
107 #define RBD_IMAGE_NAME_LEN_MAX (PAGE_SIZE - sizeof (__le32) - 1)
108 #define RBD_IMAGE_ID_LEN_MAX 64
110 #define RBD_OBJ_PREFIX_LEN_MAX 64
114 #define RBD_FEATURE_LAYERING (1<<0)
115 #define RBD_FEATURE_STRIPINGV2 (1<<1)
116 #define RBD_FEATURES_ALL \
117 (RBD_FEATURE_LAYERING | RBD_FEATURE_STRIPINGV2)
119 /* Features supported by this (client software) implementation. */
121 #define RBD_FEATURES_SUPPORTED (RBD_FEATURES_ALL)
124 * An RBD device name will be "rbd#", where the "rbd" comes from
125 * RBD_DRV_NAME above, and # is a unique integer identifier.
126 * MAX_INT_FORMAT_WIDTH is used in ensuring DEV_NAME_LEN is big
127 * enough to hold all possible device names.
129 #define DEV_NAME_LEN 32
130 #define MAX_INT_FORMAT_WIDTH ((5 * sizeof (int)) / 2 + 1)
133 * block device image metadata (in-memory version)
135 struct rbd_image_header {
136 /* These six fields never change for a given rbd image */
143 u64 features; /* Might be changeable someday? */
145 /* The remaining fields need to be updated occasionally */
147 struct ceph_snap_context *snapc;
148 char *snap_names; /* format 1 only */
149 u64 *snap_sizes; /* format 1 only */
153 * An rbd image specification.
155 * The tuple (pool_id, image_id, snap_id) is sufficient to uniquely
156 * identify an image. Each rbd_dev structure includes a pointer to
157 * an rbd_spec structure that encapsulates this identity.
159 * Each of the id's in an rbd_spec has an associated name. For a
160 * user-mapped image, the names are supplied and the id's associated
161 * with them are looked up. For a layered image, a parent image is
162 * defined by the tuple, and the names are looked up.
164 * An rbd_dev structure contains a parent_spec pointer which is
165 * non-null if the image it represents is a child in a layered
166 * image. This pointer will refer to the rbd_spec structure used
167 * by the parent rbd_dev for its own identity (i.e., the structure
168 * is shared between the parent and child).
170 * Since these structures are populated once, during the discovery
171 * phase of image construction, they are effectively immutable so
172 * we make no effort to synchronize access to them.
174 * Note that code herein does not assume the image name is known (it
175 * could be a null pointer).
179 const char *pool_name;
181 const char *image_id;
182 const char *image_name;
185 const char *snap_name;
191 * an instance of the client. multiple devices may share an rbd client.
194 struct ceph_client *client;
196 struct list_head node;
199 struct rbd_img_request;
200 typedef void (*rbd_img_callback_t)(struct rbd_img_request *);
202 #define BAD_WHICH U32_MAX /* Good which or bad which, which? */
204 struct rbd_obj_request;
205 typedef void (*rbd_obj_callback_t)(struct rbd_obj_request *);
207 enum obj_request_type {
208 OBJ_REQUEST_NODATA, OBJ_REQUEST_BIO, OBJ_REQUEST_PAGES
212 OBJ_REQ_DONE, /* completion flag: not done = 0, done = 1 */
213 OBJ_REQ_IMG_DATA, /* object usage: standalone = 0, image = 1 */
214 OBJ_REQ_KNOWN, /* EXISTS flag valid: no = 0, yes = 1 */
215 OBJ_REQ_EXISTS, /* target exists: no = 0, yes = 1 */
218 struct rbd_obj_request {
219 const char *object_name;
220 u64 offset; /* object start byte */
221 u64 length; /* bytes from offset */
225 * An object request associated with an image will have its
226 * img_data flag set; a standalone object request will not.
228 * A standalone object request will have which == BAD_WHICH
229 * and a null obj_request pointer.
231 * An object request initiated in support of a layered image
232 * object (to check for its existence before a write) will
233 * have which == BAD_WHICH and a non-null obj_request pointer.
235 * Finally, an object request for rbd image data will have
236 * which != BAD_WHICH, and will have a non-null img_request
237 * pointer. The value of which will be in the range
238 * 0..(img_request->obj_request_count-1).
241 struct rbd_obj_request *obj_request; /* STAT op */
243 struct rbd_img_request *img_request;
245 /* links for img_request->obj_requests list */
246 struct list_head links;
249 u32 which; /* posn image request list */
251 enum obj_request_type type;
253 struct bio *bio_list;
259 struct page **copyup_pages;
260 u32 copyup_page_count;
262 struct ceph_osd_request *osd_req;
264 u64 xferred; /* bytes transferred */
267 rbd_obj_callback_t callback;
268 struct completion completion;
274 IMG_REQ_WRITE, /* I/O direction: read = 0, write = 1 */
275 IMG_REQ_CHILD, /* initiator: block = 0, child image = 1 */
276 IMG_REQ_LAYERED, /* ENOENT handling: normal = 0, layered = 1 */
279 struct rbd_img_request {
280 struct rbd_device *rbd_dev;
281 u64 offset; /* starting image byte offset */
282 u64 length; /* byte count from offset */
285 u64 snap_id; /* for reads */
286 struct ceph_snap_context *snapc; /* for writes */
289 struct request *rq; /* block request */
290 struct rbd_obj_request *obj_request; /* obj req initiator */
292 struct page **copyup_pages;
293 u32 copyup_page_count;
294 spinlock_t completion_lock;/* protects next_completion */
296 rbd_img_callback_t callback;
297 u64 xferred;/* aggregate bytes transferred */
298 int result; /* first nonzero obj_request result */
300 u32 obj_request_count;
301 struct list_head obj_requests; /* rbd_obj_request structs */
306 #define for_each_obj_request(ireq, oreq) \
307 list_for_each_entry(oreq, &(ireq)->obj_requests, links)
308 #define for_each_obj_request_from(ireq, oreq) \
309 list_for_each_entry_from(oreq, &(ireq)->obj_requests, links)
310 #define for_each_obj_request_safe(ireq, oreq, n) \
311 list_for_each_entry_safe_reverse(oreq, n, &(ireq)->obj_requests, links)
323 int dev_id; /* blkdev unique id */
325 int major; /* blkdev assigned major */
326 struct gendisk *disk; /* blkdev's gendisk and rq */
328 u32 image_format; /* Either 1 or 2 */
329 struct rbd_client *rbd_client;
331 char name[DEV_NAME_LEN]; /* blkdev name, e.g. rbd3 */
333 spinlock_t lock; /* queue, flags, open_count */
335 struct rbd_image_header header;
336 unsigned long flags; /* possibly lock protected */
337 struct rbd_spec *spec;
341 struct ceph_file_layout layout;
343 struct ceph_osd_event *watch_event;
344 struct rbd_obj_request *watch_request;
346 struct rbd_spec *parent_spec;
349 struct rbd_device *parent;
351 /* protects updating the header */
352 struct rw_semaphore header_rwsem;
354 struct rbd_mapping mapping;
356 struct list_head node;
360 unsigned long open_count; /* protected by lock */
364 * Flag bits for rbd_dev->flags. If atomicity is required,
365 * rbd_dev->lock is used to protect access.
367 * Currently, only the "removing" flag (which is coupled with the
368 * "open_count" field) requires atomic access.
371 RBD_DEV_FLAG_EXISTS, /* mapped snapshot has not been deleted */
372 RBD_DEV_FLAG_REMOVING, /* this mapping is being removed */
375 static DEFINE_MUTEX(ctl_mutex); /* Serialize open/close/setup/teardown */
377 static LIST_HEAD(rbd_dev_list); /* devices */
378 static DEFINE_SPINLOCK(rbd_dev_list_lock);
380 static LIST_HEAD(rbd_client_list); /* clients */
381 static DEFINE_SPINLOCK(rbd_client_list_lock);
383 /* Slab caches for frequently-allocated structures */
385 static struct kmem_cache *rbd_img_request_cache;
386 static struct kmem_cache *rbd_obj_request_cache;
387 static struct kmem_cache *rbd_segment_name_cache;
389 static int rbd_img_request_submit(struct rbd_img_request *img_request);
391 static void rbd_dev_device_release(struct device *dev);
393 static ssize_t rbd_add(struct bus_type *bus, const char *buf,
395 static ssize_t rbd_remove(struct bus_type *bus, const char *buf,
397 static int rbd_dev_image_probe(struct rbd_device *rbd_dev, bool mapping);
398 static void rbd_spec_put(struct rbd_spec *spec);
400 static struct bus_attribute rbd_bus_attrs[] = {
401 __ATTR(add, S_IWUSR, NULL, rbd_add),
402 __ATTR(remove, S_IWUSR, NULL, rbd_remove),
406 static struct bus_type rbd_bus_type = {
408 .bus_attrs = rbd_bus_attrs,
411 static void rbd_root_dev_release(struct device *dev)
415 static struct device rbd_root_dev = {
417 .release = rbd_root_dev_release,
420 static __printf(2, 3)
421 void rbd_warn(struct rbd_device *rbd_dev, const char *fmt, ...)
423 struct va_format vaf;
431 printk(KERN_WARNING "%s: %pV\n", RBD_DRV_NAME, &vaf);
432 else if (rbd_dev->disk)
433 printk(KERN_WARNING "%s: %s: %pV\n",
434 RBD_DRV_NAME, rbd_dev->disk->disk_name, &vaf);
435 else if (rbd_dev->spec && rbd_dev->spec->image_name)
436 printk(KERN_WARNING "%s: image %s: %pV\n",
437 RBD_DRV_NAME, rbd_dev->spec->image_name, &vaf);
438 else if (rbd_dev->spec && rbd_dev->spec->image_id)
439 printk(KERN_WARNING "%s: id %s: %pV\n",
440 RBD_DRV_NAME, rbd_dev->spec->image_id, &vaf);
442 printk(KERN_WARNING "%s: rbd_dev %p: %pV\n",
443 RBD_DRV_NAME, rbd_dev, &vaf);
448 #define rbd_assert(expr) \
449 if (unlikely(!(expr))) { \
450 printk(KERN_ERR "\nAssertion failure in %s() " \
452 "\trbd_assert(%s);\n\n", \
453 __func__, __LINE__, #expr); \
456 #else /* !RBD_DEBUG */
457 # define rbd_assert(expr) ((void) 0)
458 #endif /* !RBD_DEBUG */
460 static int rbd_img_obj_request_submit(struct rbd_obj_request *obj_request);
461 static void rbd_img_parent_read(struct rbd_obj_request *obj_request);
462 static void rbd_dev_remove_parent(struct rbd_device *rbd_dev);
464 static int rbd_dev_refresh(struct rbd_device *rbd_dev);
465 static int rbd_dev_v2_header_onetime(struct rbd_device *rbd_dev);
466 static int rbd_dev_v2_header_info(struct rbd_device *rbd_dev);
467 static const char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev,
469 static int _rbd_dev_v2_snap_size(struct rbd_device *rbd_dev, u64 snap_id,
470 u8 *order, u64 *snap_size);
471 static int _rbd_dev_v2_snap_features(struct rbd_device *rbd_dev, u64 snap_id,
473 static u64 rbd_snap_id_by_name(struct rbd_device *rbd_dev, const char *name);
475 static int rbd_open(struct block_device *bdev, fmode_t mode)
477 struct rbd_device *rbd_dev = bdev->bd_disk->private_data;
478 bool removing = false;
480 if ((mode & FMODE_WRITE) && rbd_dev->mapping.read_only)
483 spin_lock_irq(&rbd_dev->lock);
484 if (test_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags))
487 rbd_dev->open_count++;
488 spin_unlock_irq(&rbd_dev->lock);
492 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
493 (void) get_device(&rbd_dev->dev);
494 set_device_ro(bdev, rbd_dev->mapping.read_only);
495 mutex_unlock(&ctl_mutex);
500 static int rbd_release(struct gendisk *disk, fmode_t mode)
502 struct rbd_device *rbd_dev = disk->private_data;
503 unsigned long open_count_before;
505 spin_lock_irq(&rbd_dev->lock);
506 open_count_before = rbd_dev->open_count--;
507 spin_unlock_irq(&rbd_dev->lock);
508 rbd_assert(open_count_before > 0);
510 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
511 put_device(&rbd_dev->dev);
512 mutex_unlock(&ctl_mutex);
517 static const struct block_device_operations rbd_bd_ops = {
518 .owner = THIS_MODULE,
520 .release = rbd_release,
524 * Initialize an rbd client instance. Success or not, this function
525 * consumes ceph_opts.
527 static struct rbd_client *rbd_client_create(struct ceph_options *ceph_opts)
529 struct rbd_client *rbdc;
532 dout("%s:\n", __func__);
533 rbdc = kmalloc(sizeof(struct rbd_client), GFP_KERNEL);
537 kref_init(&rbdc->kref);
538 INIT_LIST_HEAD(&rbdc->node);
540 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
542 rbdc->client = ceph_create_client(ceph_opts, rbdc, 0, 0);
543 if (IS_ERR(rbdc->client))
545 ceph_opts = NULL; /* Now rbdc->client is responsible for ceph_opts */
547 ret = ceph_open_session(rbdc->client);
551 spin_lock(&rbd_client_list_lock);
552 list_add_tail(&rbdc->node, &rbd_client_list);
553 spin_unlock(&rbd_client_list_lock);
555 mutex_unlock(&ctl_mutex);
556 dout("%s: rbdc %p\n", __func__, rbdc);
561 ceph_destroy_client(rbdc->client);
563 mutex_unlock(&ctl_mutex);
567 ceph_destroy_options(ceph_opts);
568 dout("%s: error %d\n", __func__, ret);
573 static struct rbd_client *__rbd_get_client(struct rbd_client *rbdc)
575 kref_get(&rbdc->kref);
581 * Find a ceph client with specific addr and configuration. If
582 * found, bump its reference count.
584 static struct rbd_client *rbd_client_find(struct ceph_options *ceph_opts)
586 struct rbd_client *client_node;
589 if (ceph_opts->flags & CEPH_OPT_NOSHARE)
592 spin_lock(&rbd_client_list_lock);
593 list_for_each_entry(client_node, &rbd_client_list, node) {
594 if (!ceph_compare_options(ceph_opts, client_node->client)) {
595 __rbd_get_client(client_node);
601 spin_unlock(&rbd_client_list_lock);
603 return found ? client_node : NULL;
613 /* string args above */
616 /* Boolean args above */
620 static match_table_t rbd_opts_tokens = {
622 /* string args above */
623 {Opt_read_only, "read_only"},
624 {Opt_read_only, "ro"}, /* Alternate spelling */
625 {Opt_read_write, "read_write"},
626 {Opt_read_write, "rw"}, /* Alternate spelling */
627 /* Boolean args above */
635 #define RBD_READ_ONLY_DEFAULT false
637 static int parse_rbd_opts_token(char *c, void *private)
639 struct rbd_options *rbd_opts = private;
640 substring_t argstr[MAX_OPT_ARGS];
641 int token, intval, ret;
643 token = match_token(c, rbd_opts_tokens, argstr);
647 if (token < Opt_last_int) {
648 ret = match_int(&argstr[0], &intval);
650 pr_err("bad mount option arg (not int) "
654 dout("got int token %d val %d\n", token, intval);
655 } else if (token > Opt_last_int && token < Opt_last_string) {
656 dout("got string token %d val %s\n", token,
658 } else if (token > Opt_last_string && token < Opt_last_bool) {
659 dout("got Boolean token %d\n", token);
661 dout("got token %d\n", token);
666 rbd_opts->read_only = true;
669 rbd_opts->read_only = false;
679 * Get a ceph client with specific addr and configuration, if one does
680 * not exist create it. Either way, ceph_opts is consumed by this
683 static struct rbd_client *rbd_get_client(struct ceph_options *ceph_opts)
685 struct rbd_client *rbdc;
687 rbdc = rbd_client_find(ceph_opts);
688 if (rbdc) /* using an existing client */
689 ceph_destroy_options(ceph_opts);
691 rbdc = rbd_client_create(ceph_opts);
697 * Destroy ceph client
699 * Caller must hold rbd_client_list_lock.
701 static void rbd_client_release(struct kref *kref)
703 struct rbd_client *rbdc = container_of(kref, struct rbd_client, kref);
705 dout("%s: rbdc %p\n", __func__, rbdc);
706 spin_lock(&rbd_client_list_lock);
707 list_del(&rbdc->node);
708 spin_unlock(&rbd_client_list_lock);
710 ceph_destroy_client(rbdc->client);
715 * Drop reference to ceph client node. If it's not referenced anymore, release
718 static void rbd_put_client(struct rbd_client *rbdc)
721 kref_put(&rbdc->kref, rbd_client_release);
724 static bool rbd_image_format_valid(u32 image_format)
726 return image_format == 1 || image_format == 2;
729 static bool rbd_dev_ondisk_valid(struct rbd_image_header_ondisk *ondisk)
734 /* The header has to start with the magic rbd header text */
735 if (memcmp(&ondisk->text, RBD_HEADER_TEXT, sizeof (RBD_HEADER_TEXT)))
738 /* The bio layer requires at least sector-sized I/O */
740 if (ondisk->options.order < SECTOR_SHIFT)
743 /* If we use u64 in a few spots we may be able to loosen this */
745 if (ondisk->options.order > 8 * sizeof (int) - 1)
749 * The size of a snapshot header has to fit in a size_t, and
750 * that limits the number of snapshots.
752 snap_count = le32_to_cpu(ondisk->snap_count);
753 size = SIZE_MAX - sizeof (struct ceph_snap_context);
754 if (snap_count > size / sizeof (__le64))
758 * Not only that, but the size of the entire the snapshot
759 * header must also be representable in a size_t.
761 size -= snap_count * sizeof (__le64);
762 if ((u64) size < le64_to_cpu(ondisk->snap_names_len))
769 * Fill an rbd image header with information from the given format 1
772 static int rbd_header_from_disk(struct rbd_device *rbd_dev,
773 struct rbd_image_header_ondisk *ondisk)
775 struct rbd_image_header *header = &rbd_dev->header;
776 bool first_time = header->object_prefix == NULL;
777 struct ceph_snap_context *snapc;
778 char *object_prefix = NULL;
779 char *snap_names = NULL;
780 u64 *snap_sizes = NULL;
786 /* Allocate this now to avoid having to handle failure below */
791 len = strnlen(ondisk->object_prefix,
792 sizeof (ondisk->object_prefix));
793 object_prefix = kmalloc(len + 1, GFP_KERNEL);
796 memcpy(object_prefix, ondisk->object_prefix, len);
797 object_prefix[len] = '\0';
800 /* Allocate the snapshot context and fill it in */
802 snap_count = le32_to_cpu(ondisk->snap_count);
803 snapc = ceph_create_snap_context(snap_count, GFP_KERNEL);
806 snapc->seq = le64_to_cpu(ondisk->snap_seq);
808 struct rbd_image_snap_ondisk *snaps;
809 u64 snap_names_len = le64_to_cpu(ondisk->snap_names_len);
811 /* We'll keep a copy of the snapshot names... */
813 if (snap_names_len > (u64)SIZE_MAX)
815 snap_names = kmalloc(snap_names_len, GFP_KERNEL);
819 /* ...as well as the array of their sizes. */
821 size = snap_count * sizeof (*header->snap_sizes);
822 snap_sizes = kmalloc(size, GFP_KERNEL);
827 * Copy the names, and fill in each snapshot's id
830 * Note that rbd_dev_v1_header_info() guarantees the
831 * ondisk buffer we're working with has
832 * snap_names_len bytes beyond the end of the
833 * snapshot id array, this memcpy() is safe.
835 memcpy(snap_names, &ondisk->snaps[snap_count], snap_names_len);
836 snaps = ondisk->snaps;
837 for (i = 0; i < snap_count; i++) {
838 snapc->snaps[i] = le64_to_cpu(snaps[i].id);
839 snap_sizes[i] = le64_to_cpu(snaps[i].image_size);
843 /* We won't fail any more, fill in the header */
845 down_write(&rbd_dev->header_rwsem);
847 header->object_prefix = object_prefix;
848 header->obj_order = ondisk->options.order;
849 header->crypt_type = ondisk->options.crypt_type;
850 header->comp_type = ondisk->options.comp_type;
851 /* The rest aren't used for format 1 images */
852 header->stripe_unit = 0;
853 header->stripe_count = 0;
854 header->features = 0;
856 ceph_put_snap_context(header->snapc);
857 kfree(header->snap_names);
858 kfree(header->snap_sizes);
861 /* The remaining fields always get updated (when we refresh) */
863 header->image_size = le64_to_cpu(ondisk->image_size);
864 header->snapc = snapc;
865 header->snap_names = snap_names;
866 header->snap_sizes = snap_sizes;
868 /* Make sure mapping size is consistent with header info */
870 if (rbd_dev->spec->snap_id == CEPH_NOSNAP || first_time)
871 if (rbd_dev->mapping.size != header->image_size)
872 rbd_dev->mapping.size = header->image_size;
874 up_write(&rbd_dev->header_rwsem);
882 ceph_put_snap_context(snapc);
883 kfree(object_prefix);
888 static const char *_rbd_dev_v1_snap_name(struct rbd_device *rbd_dev, u32 which)
890 const char *snap_name;
892 rbd_assert(which < rbd_dev->header.snapc->num_snaps);
894 /* Skip over names until we find the one we are looking for */
896 snap_name = rbd_dev->header.snap_names;
898 snap_name += strlen(snap_name) + 1;
900 return kstrdup(snap_name, GFP_KERNEL);
904 * Snapshot id comparison function for use with qsort()/bsearch().
905 * Note that result is for snapshots in *descending* order.
907 static int snapid_compare_reverse(const void *s1, const void *s2)
909 u64 snap_id1 = *(u64 *)s1;
910 u64 snap_id2 = *(u64 *)s2;
912 if (snap_id1 < snap_id2)
914 return snap_id1 == snap_id2 ? 0 : -1;
918 * Search a snapshot context to see if the given snapshot id is
921 * Returns the position of the snapshot id in the array if it's found,
922 * or BAD_SNAP_INDEX otherwise.
924 * Note: The snapshot array is in kept sorted (by the osd) in
925 * reverse order, highest snapshot id first.
927 static u32 rbd_dev_snap_index(struct rbd_device *rbd_dev, u64 snap_id)
929 struct ceph_snap_context *snapc = rbd_dev->header.snapc;
932 found = bsearch(&snap_id, &snapc->snaps, snapc->num_snaps,
933 sizeof (snap_id), snapid_compare_reverse);
935 return found ? (u32)(found - &snapc->snaps[0]) : BAD_SNAP_INDEX;
938 static const char *rbd_dev_v1_snap_name(struct rbd_device *rbd_dev,
943 which = rbd_dev_snap_index(rbd_dev, snap_id);
944 if (which == BAD_SNAP_INDEX)
947 return _rbd_dev_v1_snap_name(rbd_dev, which);
950 static const char *rbd_snap_name(struct rbd_device *rbd_dev, u64 snap_id)
952 if (snap_id == CEPH_NOSNAP)
953 return RBD_SNAP_HEAD_NAME;
955 rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
956 if (rbd_dev->image_format == 1)
957 return rbd_dev_v1_snap_name(rbd_dev, snap_id);
959 return rbd_dev_v2_snap_name(rbd_dev, snap_id);
962 static int rbd_snap_size(struct rbd_device *rbd_dev, u64 snap_id,
965 rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
966 if (snap_id == CEPH_NOSNAP) {
967 *snap_size = rbd_dev->header.image_size;
968 } else if (rbd_dev->image_format == 1) {
971 which = rbd_dev_snap_index(rbd_dev, snap_id);
972 if (which == BAD_SNAP_INDEX)
975 *snap_size = rbd_dev->header.snap_sizes[which];
980 ret = _rbd_dev_v2_snap_size(rbd_dev, snap_id, NULL, &size);
989 static int rbd_snap_features(struct rbd_device *rbd_dev, u64 snap_id,
992 rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
993 if (snap_id == CEPH_NOSNAP) {
994 *snap_features = rbd_dev->header.features;
995 } else if (rbd_dev->image_format == 1) {
996 *snap_features = 0; /* No features for format 1 */
1001 ret = _rbd_dev_v2_snap_features(rbd_dev, snap_id, &features);
1005 *snap_features = features;
1010 static int rbd_dev_mapping_set(struct rbd_device *rbd_dev)
1012 u64 snap_id = rbd_dev->spec->snap_id;
1017 ret = rbd_snap_size(rbd_dev, snap_id, &size);
1020 ret = rbd_snap_features(rbd_dev, snap_id, &features);
1024 rbd_dev->mapping.size = size;
1025 rbd_dev->mapping.features = features;
1030 static void rbd_dev_mapping_clear(struct rbd_device *rbd_dev)
1032 rbd_dev->mapping.size = 0;
1033 rbd_dev->mapping.features = 0;
1036 static const char *rbd_segment_name(struct rbd_device *rbd_dev, u64 offset)
1043 name = kmem_cache_alloc(rbd_segment_name_cache, GFP_NOIO);
1046 segment = offset >> rbd_dev->header.obj_order;
1047 name_format = "%s.%012llx";
1048 if (rbd_dev->image_format == 2)
1049 name_format = "%s.%016llx";
1050 ret = snprintf(name, MAX_OBJ_NAME_SIZE + 1, name_format,
1051 rbd_dev->header.object_prefix, segment);
1052 if (ret < 0 || ret > MAX_OBJ_NAME_SIZE) {
1053 pr_err("error formatting segment name for #%llu (%d)\n",
1062 static void rbd_segment_name_free(const char *name)
1064 /* The explicit cast here is needed to drop the const qualifier */
1066 kmem_cache_free(rbd_segment_name_cache, (void *)name);
1069 static u64 rbd_segment_offset(struct rbd_device *rbd_dev, u64 offset)
1071 u64 segment_size = (u64) 1 << rbd_dev->header.obj_order;
1073 return offset & (segment_size - 1);
1076 static u64 rbd_segment_length(struct rbd_device *rbd_dev,
1077 u64 offset, u64 length)
1079 u64 segment_size = (u64) 1 << rbd_dev->header.obj_order;
1081 offset &= segment_size - 1;
1083 rbd_assert(length <= U64_MAX - offset);
1084 if (offset + length > segment_size)
1085 length = segment_size - offset;
1091 * returns the size of an object in the image
1093 static u64 rbd_obj_bytes(struct rbd_image_header *header)
1095 return 1 << header->obj_order;
1102 static void bio_chain_put(struct bio *chain)
1108 chain = chain->bi_next;
1114 * zeros a bio chain, starting at specific offset
1116 static void zero_bio_chain(struct bio *chain, int start_ofs)
1119 unsigned long flags;
1125 bio_for_each_segment(bv, chain, i) {
1126 if (pos + bv->bv_len > start_ofs) {
1127 int remainder = max(start_ofs - pos, 0);
1128 buf = bvec_kmap_irq(bv, &flags);
1129 memset(buf + remainder, 0,
1130 bv->bv_len - remainder);
1131 bvec_kunmap_irq(buf, &flags);
1136 chain = chain->bi_next;
1141 * similar to zero_bio_chain(), zeros data defined by a page array,
1142 * starting at the given byte offset from the start of the array and
1143 * continuing up to the given end offset. The pages array is
1144 * assumed to be big enough to hold all bytes up to the end.
1146 static void zero_pages(struct page **pages, u64 offset, u64 end)
1148 struct page **page = &pages[offset >> PAGE_SHIFT];
1150 rbd_assert(end > offset);
1151 rbd_assert(end - offset <= (u64)SIZE_MAX);
1152 while (offset < end) {
1155 unsigned long flags;
1158 page_offset = (size_t)(offset & ~PAGE_MASK);
1159 length = min(PAGE_SIZE - page_offset, (size_t)(end - offset));
1160 local_irq_save(flags);
1161 kaddr = kmap_atomic(*page);
1162 memset(kaddr + page_offset, 0, length);
1163 kunmap_atomic(kaddr);
1164 local_irq_restore(flags);
1172 * Clone a portion of a bio, starting at the given byte offset
1173 * and continuing for the number of bytes indicated.
1175 static struct bio *bio_clone_range(struct bio *bio_src,
1176 unsigned int offset,
1184 unsigned short end_idx;
1185 unsigned short vcnt;
1188 /* Handle the easy case for the caller */
1190 if (!offset && len == bio_src->bi_size)
1191 return bio_clone(bio_src, gfpmask);
1193 if (WARN_ON_ONCE(!len))
1195 if (WARN_ON_ONCE(len > bio_src->bi_size))
1197 if (WARN_ON_ONCE(offset > bio_src->bi_size - len))
1200 /* Find first affected segment... */
1203 __bio_for_each_segment(bv, bio_src, idx, 0) {
1204 if (resid < bv->bv_len)
1206 resid -= bv->bv_len;
1210 /* ...and the last affected segment */
1213 __bio_for_each_segment(bv, bio_src, end_idx, idx) {
1214 if (resid <= bv->bv_len)
1216 resid -= bv->bv_len;
1218 vcnt = end_idx - idx + 1;
1220 /* Build the clone */
1222 bio = bio_alloc(gfpmask, (unsigned int) vcnt);
1224 return NULL; /* ENOMEM */
1226 bio->bi_bdev = bio_src->bi_bdev;
1227 bio->bi_sector = bio_src->bi_sector + (offset >> SECTOR_SHIFT);
1228 bio->bi_rw = bio_src->bi_rw;
1229 bio->bi_flags |= 1 << BIO_CLONED;
1232 * Copy over our part of the bio_vec, then update the first
1233 * and last (or only) entries.
1235 memcpy(&bio->bi_io_vec[0], &bio_src->bi_io_vec[idx],
1236 vcnt * sizeof (struct bio_vec));
1237 bio->bi_io_vec[0].bv_offset += voff;
1239 bio->bi_io_vec[0].bv_len -= voff;
1240 bio->bi_io_vec[vcnt - 1].bv_len = resid;
1242 bio->bi_io_vec[0].bv_len = len;
1245 bio->bi_vcnt = vcnt;
1253 * Clone a portion of a bio chain, starting at the given byte offset
1254 * into the first bio in the source chain and continuing for the
1255 * number of bytes indicated. The result is another bio chain of
1256 * exactly the given length, or a null pointer on error.
1258 * The bio_src and offset parameters are both in-out. On entry they
1259 * refer to the first source bio and the offset into that bio where
1260 * the start of data to be cloned is located.
1262 * On return, bio_src is updated to refer to the bio in the source
1263 * chain that contains first un-cloned byte, and *offset will
1264 * contain the offset of that byte within that bio.
1266 static struct bio *bio_chain_clone_range(struct bio **bio_src,
1267 unsigned int *offset,
1271 struct bio *bi = *bio_src;
1272 unsigned int off = *offset;
1273 struct bio *chain = NULL;
1276 /* Build up a chain of clone bios up to the limit */
1278 if (!bi || off >= bi->bi_size || !len)
1279 return NULL; /* Nothing to clone */
1283 unsigned int bi_size;
1287 rbd_warn(NULL, "bio_chain exhausted with %u left", len);
1288 goto out_err; /* EINVAL; ran out of bio's */
1290 bi_size = min_t(unsigned int, bi->bi_size - off, len);
1291 bio = bio_clone_range(bi, off, bi_size, gfpmask);
1293 goto out_err; /* ENOMEM */
1296 end = &bio->bi_next;
1299 if (off == bi->bi_size) {
1310 bio_chain_put(chain);
1316 * The default/initial value for all object request flags is 0. For
1317 * each flag, once its value is set to 1 it is never reset to 0
1320 static void obj_request_img_data_set(struct rbd_obj_request *obj_request)
1322 if (test_and_set_bit(OBJ_REQ_IMG_DATA, &obj_request->flags)) {
1323 struct rbd_device *rbd_dev;
1325 rbd_dev = obj_request->img_request->rbd_dev;
1326 rbd_warn(rbd_dev, "obj_request %p already marked img_data\n",
1331 static bool obj_request_img_data_test(struct rbd_obj_request *obj_request)
1334 return test_bit(OBJ_REQ_IMG_DATA, &obj_request->flags) != 0;
1337 static void obj_request_done_set(struct rbd_obj_request *obj_request)
1339 if (test_and_set_bit(OBJ_REQ_DONE, &obj_request->flags)) {
1340 struct rbd_device *rbd_dev = NULL;
1342 if (obj_request_img_data_test(obj_request))
1343 rbd_dev = obj_request->img_request->rbd_dev;
1344 rbd_warn(rbd_dev, "obj_request %p already marked done\n",
1349 static bool obj_request_done_test(struct rbd_obj_request *obj_request)
1352 return test_bit(OBJ_REQ_DONE, &obj_request->flags) != 0;
1356 * This sets the KNOWN flag after (possibly) setting the EXISTS
1357 * flag. The latter is set based on the "exists" value provided.
1359 * Note that for our purposes once an object exists it never goes
1360 * away again. It's possible that the response from two existence
1361 * checks are separated by the creation of the target object, and
1362 * the first ("doesn't exist") response arrives *after* the second
1363 * ("does exist"). In that case we ignore the second one.
1365 static void obj_request_existence_set(struct rbd_obj_request *obj_request,
1369 set_bit(OBJ_REQ_EXISTS, &obj_request->flags);
1370 set_bit(OBJ_REQ_KNOWN, &obj_request->flags);
1374 static bool obj_request_known_test(struct rbd_obj_request *obj_request)
1377 return test_bit(OBJ_REQ_KNOWN, &obj_request->flags) != 0;
1380 static bool obj_request_exists_test(struct rbd_obj_request *obj_request)
1383 return test_bit(OBJ_REQ_EXISTS, &obj_request->flags) != 0;
1386 static void rbd_obj_request_get(struct rbd_obj_request *obj_request)
1388 dout("%s: obj %p (was %d)\n", __func__, obj_request,
1389 atomic_read(&obj_request->kref.refcount));
1390 kref_get(&obj_request->kref);
1393 static void rbd_obj_request_destroy(struct kref *kref);
1394 static void rbd_obj_request_put(struct rbd_obj_request *obj_request)
1396 rbd_assert(obj_request != NULL);
1397 dout("%s: obj %p (was %d)\n", __func__, obj_request,
1398 atomic_read(&obj_request->kref.refcount));
1399 kref_put(&obj_request->kref, rbd_obj_request_destroy);
1402 static bool img_request_child_test(struct rbd_img_request *img_request);
1403 static void rbd_parent_request_destroy(struct kref *kref);
1404 static void rbd_img_request_destroy(struct kref *kref);
1405 static void rbd_img_request_put(struct rbd_img_request *img_request)
1407 rbd_assert(img_request != NULL);
1408 dout("%s: img %p (was %d)\n", __func__, img_request,
1409 atomic_read(&img_request->kref.refcount));
1410 if (img_request_child_test(img_request))
1411 kref_put(&img_request->kref, rbd_parent_request_destroy);
1413 kref_put(&img_request->kref, rbd_img_request_destroy);
1416 static inline void rbd_img_obj_request_add(struct rbd_img_request *img_request,
1417 struct rbd_obj_request *obj_request)
1419 rbd_assert(obj_request->img_request == NULL);
1421 /* Image request now owns object's original reference */
1422 obj_request->img_request = img_request;
1423 obj_request->which = img_request->obj_request_count;
1424 rbd_assert(!obj_request_img_data_test(obj_request));
1425 obj_request_img_data_set(obj_request);
1426 rbd_assert(obj_request->which != BAD_WHICH);
1427 img_request->obj_request_count++;
1428 list_add_tail(&obj_request->links, &img_request->obj_requests);
1429 dout("%s: img %p obj %p w=%u\n", __func__, img_request, obj_request,
1430 obj_request->which);
1433 static inline void rbd_img_obj_request_del(struct rbd_img_request *img_request,
1434 struct rbd_obj_request *obj_request)
1436 rbd_assert(obj_request->which != BAD_WHICH);
1438 dout("%s: img %p obj %p w=%u\n", __func__, img_request, obj_request,
1439 obj_request->which);
1440 list_del(&obj_request->links);
1441 rbd_assert(img_request->obj_request_count > 0);
1442 img_request->obj_request_count--;
1443 rbd_assert(obj_request->which == img_request->obj_request_count);
1444 obj_request->which = BAD_WHICH;
1445 rbd_assert(obj_request_img_data_test(obj_request));
1446 rbd_assert(obj_request->img_request == img_request);
1447 obj_request->img_request = NULL;
1448 obj_request->callback = NULL;
1449 rbd_obj_request_put(obj_request);
1452 static bool obj_request_type_valid(enum obj_request_type type)
1455 case OBJ_REQUEST_NODATA:
1456 case OBJ_REQUEST_BIO:
1457 case OBJ_REQUEST_PAGES:
1464 static int rbd_obj_request_submit(struct ceph_osd_client *osdc,
1465 struct rbd_obj_request *obj_request)
1467 dout("%s: osdc %p obj %p\n", __func__, osdc, obj_request);
1469 return ceph_osdc_start_request(osdc, obj_request->osd_req, false);
1472 static void rbd_img_request_complete(struct rbd_img_request *img_request)
1475 dout("%s: img %p\n", __func__, img_request);
1478 * If no error occurred, compute the aggregate transfer
1479 * count for the image request. We could instead use
1480 * atomic64_cmpxchg() to update it as each object request
1481 * completes; not clear which way is better off hand.
1483 if (!img_request->result) {
1484 struct rbd_obj_request *obj_request;
1487 for_each_obj_request(img_request, obj_request)
1488 xferred += obj_request->xferred;
1489 img_request->xferred = xferred;
1492 if (img_request->callback)
1493 img_request->callback(img_request);
1495 rbd_img_request_put(img_request);
1498 /* Caller is responsible for rbd_obj_request_destroy(obj_request) */
1500 static int rbd_obj_request_wait(struct rbd_obj_request *obj_request)
1502 dout("%s: obj %p\n", __func__, obj_request);
1504 return wait_for_completion_interruptible(&obj_request->completion);
1508 * The default/initial value for all image request flags is 0. Each
1509 * is conditionally set to 1 at image request initialization time
1510 * and currently never change thereafter.
1512 static void img_request_write_set(struct rbd_img_request *img_request)
1514 set_bit(IMG_REQ_WRITE, &img_request->flags);
1518 static bool img_request_write_test(struct rbd_img_request *img_request)
1521 return test_bit(IMG_REQ_WRITE, &img_request->flags) != 0;
1524 static void img_request_child_set(struct rbd_img_request *img_request)
1526 set_bit(IMG_REQ_CHILD, &img_request->flags);
1530 static void img_request_child_clear(struct rbd_img_request *img_request)
1532 clear_bit(IMG_REQ_CHILD, &img_request->flags);
1536 static bool img_request_child_test(struct rbd_img_request *img_request)
1539 return test_bit(IMG_REQ_CHILD, &img_request->flags) != 0;
1542 static void img_request_layered_set(struct rbd_img_request *img_request)
1544 set_bit(IMG_REQ_LAYERED, &img_request->flags);
1548 static void img_request_layered_clear(struct rbd_img_request *img_request)
1550 clear_bit(IMG_REQ_LAYERED, &img_request->flags);
1554 static bool img_request_layered_test(struct rbd_img_request *img_request)
1557 return test_bit(IMG_REQ_LAYERED, &img_request->flags) != 0;
1561 rbd_img_obj_request_read_callback(struct rbd_obj_request *obj_request)
1563 u64 xferred = obj_request->xferred;
1564 u64 length = obj_request->length;
1566 dout("%s: obj %p img %p result %d %llu/%llu\n", __func__,
1567 obj_request, obj_request->img_request, obj_request->result,
1570 * ENOENT means a hole in the image. We zero-fill the
1571 * entire length of the request. A short read also implies
1572 * zero-fill to the end of the request. Either way we
1573 * update the xferred count to indicate the whole request
1576 rbd_assert(obj_request->type != OBJ_REQUEST_NODATA);
1577 if (obj_request->result == -ENOENT) {
1578 if (obj_request->type == OBJ_REQUEST_BIO)
1579 zero_bio_chain(obj_request->bio_list, 0);
1581 zero_pages(obj_request->pages, 0, length);
1582 obj_request->result = 0;
1583 obj_request->xferred = length;
1584 } else if (xferred < length && !obj_request->result) {
1585 if (obj_request->type == OBJ_REQUEST_BIO)
1586 zero_bio_chain(obj_request->bio_list, xferred);
1588 zero_pages(obj_request->pages, xferred, length);
1589 obj_request->xferred = length;
1591 obj_request_done_set(obj_request);
1594 static void rbd_obj_request_complete(struct rbd_obj_request *obj_request)
1596 dout("%s: obj %p cb %p\n", __func__, obj_request,
1597 obj_request->callback);
1598 if (obj_request->callback)
1599 obj_request->callback(obj_request);
1601 complete_all(&obj_request->completion);
1604 static void rbd_osd_trivial_callback(struct rbd_obj_request *obj_request)
1606 dout("%s: obj %p\n", __func__, obj_request);
1607 obj_request_done_set(obj_request);
1610 static void rbd_osd_read_callback(struct rbd_obj_request *obj_request)
1612 struct rbd_img_request *img_request = NULL;
1613 struct rbd_device *rbd_dev = NULL;
1614 bool layered = false;
1616 if (obj_request_img_data_test(obj_request)) {
1617 img_request = obj_request->img_request;
1618 layered = img_request && img_request_layered_test(img_request);
1619 rbd_dev = img_request->rbd_dev;
1622 dout("%s: obj %p img %p result %d %llu/%llu\n", __func__,
1623 obj_request, img_request, obj_request->result,
1624 obj_request->xferred, obj_request->length);
1625 if (layered && obj_request->result == -ENOENT &&
1626 obj_request->img_offset < rbd_dev->parent_overlap)
1627 rbd_img_parent_read(obj_request);
1628 else if (img_request)
1629 rbd_img_obj_request_read_callback(obj_request);
1631 obj_request_done_set(obj_request);
1634 static void rbd_osd_write_callback(struct rbd_obj_request *obj_request)
1636 dout("%s: obj %p result %d %llu\n", __func__, obj_request,
1637 obj_request->result, obj_request->length);
1639 * There is no such thing as a successful short write. Set
1640 * it to our originally-requested length.
1642 obj_request->xferred = obj_request->length;
1643 obj_request_done_set(obj_request);
1647 * For a simple stat call there's nothing to do. We'll do more if
1648 * this is part of a write sequence for a layered image.
1650 static void rbd_osd_stat_callback(struct rbd_obj_request *obj_request)
1652 dout("%s: obj %p\n", __func__, obj_request);
1653 obj_request_done_set(obj_request);
1656 static void rbd_osd_req_callback(struct ceph_osd_request *osd_req,
1657 struct ceph_msg *msg)
1659 struct rbd_obj_request *obj_request = osd_req->r_priv;
1662 dout("%s: osd_req %p msg %p\n", __func__, osd_req, msg);
1663 rbd_assert(osd_req == obj_request->osd_req);
1664 if (obj_request_img_data_test(obj_request)) {
1665 rbd_assert(obj_request->img_request);
1666 rbd_assert(obj_request->which != BAD_WHICH);
1668 rbd_assert(obj_request->which == BAD_WHICH);
1671 if (osd_req->r_result < 0)
1672 obj_request->result = osd_req->r_result;
1674 BUG_ON(osd_req->r_num_ops > 2);
1677 * We support a 64-bit length, but ultimately it has to be
1678 * passed to blk_end_request(), which takes an unsigned int.
1680 obj_request->xferred = osd_req->r_reply_op_len[0];
1681 rbd_assert(obj_request->xferred < (u64)UINT_MAX);
1682 opcode = osd_req->r_ops[0].op;
1684 case CEPH_OSD_OP_READ:
1685 rbd_osd_read_callback(obj_request);
1687 case CEPH_OSD_OP_WRITE:
1688 rbd_osd_write_callback(obj_request);
1690 case CEPH_OSD_OP_STAT:
1691 rbd_osd_stat_callback(obj_request);
1693 case CEPH_OSD_OP_CALL:
1694 case CEPH_OSD_OP_NOTIFY_ACK:
1695 case CEPH_OSD_OP_WATCH:
1696 rbd_osd_trivial_callback(obj_request);
1699 rbd_warn(NULL, "%s: unsupported op %hu\n",
1700 obj_request->object_name, (unsigned short) opcode);
1704 if (obj_request_done_test(obj_request))
1705 rbd_obj_request_complete(obj_request);
1708 static void rbd_osd_req_format_read(struct rbd_obj_request *obj_request)
1710 struct rbd_img_request *img_request = obj_request->img_request;
1711 struct ceph_osd_request *osd_req = obj_request->osd_req;
1714 rbd_assert(osd_req != NULL);
1716 snap_id = img_request ? img_request->snap_id : CEPH_NOSNAP;
1717 ceph_osdc_build_request(osd_req, obj_request->offset,
1718 NULL, snap_id, NULL);
1721 static void rbd_osd_req_format_write(struct rbd_obj_request *obj_request)
1723 struct rbd_img_request *img_request = obj_request->img_request;
1724 struct ceph_osd_request *osd_req = obj_request->osd_req;
1725 struct ceph_snap_context *snapc;
1726 struct timespec mtime = CURRENT_TIME;
1728 rbd_assert(osd_req != NULL);
1730 snapc = img_request ? img_request->snapc : NULL;
1731 ceph_osdc_build_request(osd_req, obj_request->offset,
1732 snapc, CEPH_NOSNAP, &mtime);
1735 static struct ceph_osd_request *rbd_osd_req_create(
1736 struct rbd_device *rbd_dev,
1738 struct rbd_obj_request *obj_request)
1740 struct ceph_snap_context *snapc = NULL;
1741 struct ceph_osd_client *osdc;
1742 struct ceph_osd_request *osd_req;
1744 if (obj_request_img_data_test(obj_request)) {
1745 struct rbd_img_request *img_request = obj_request->img_request;
1747 rbd_assert(write_request ==
1748 img_request_write_test(img_request));
1750 snapc = img_request->snapc;
1753 /* Allocate and initialize the request, for the single op */
1755 osdc = &rbd_dev->rbd_client->client->osdc;
1756 osd_req = ceph_osdc_alloc_request(osdc, snapc, 1, false, GFP_ATOMIC);
1758 return NULL; /* ENOMEM */
1761 osd_req->r_flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK;
1763 osd_req->r_flags = CEPH_OSD_FLAG_READ;
1765 osd_req->r_callback = rbd_osd_req_callback;
1766 osd_req->r_priv = obj_request;
1768 osd_req->r_oid_len = strlen(obj_request->object_name);
1769 rbd_assert(osd_req->r_oid_len < sizeof (osd_req->r_oid));
1770 memcpy(osd_req->r_oid, obj_request->object_name, osd_req->r_oid_len);
1772 osd_req->r_file_layout = rbd_dev->layout; /* struct */
1778 * Create a copyup osd request based on the information in the
1779 * object request supplied. A copyup request has two osd ops,
1780 * a copyup method call, and a "normal" write request.
1782 static struct ceph_osd_request *
1783 rbd_osd_req_create_copyup(struct rbd_obj_request *obj_request)
1785 struct rbd_img_request *img_request;
1786 struct ceph_snap_context *snapc;
1787 struct rbd_device *rbd_dev;
1788 struct ceph_osd_client *osdc;
1789 struct ceph_osd_request *osd_req;
1791 rbd_assert(obj_request_img_data_test(obj_request));
1792 img_request = obj_request->img_request;
1793 rbd_assert(img_request);
1794 rbd_assert(img_request_write_test(img_request));
1796 /* Allocate and initialize the request, for the two ops */
1798 snapc = img_request->snapc;
1799 rbd_dev = img_request->rbd_dev;
1800 osdc = &rbd_dev->rbd_client->client->osdc;
1801 osd_req = ceph_osdc_alloc_request(osdc, snapc, 2, false, GFP_ATOMIC);
1803 return NULL; /* ENOMEM */
1805 osd_req->r_flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK;
1806 osd_req->r_callback = rbd_osd_req_callback;
1807 osd_req->r_priv = obj_request;
1809 osd_req->r_oid_len = strlen(obj_request->object_name);
1810 rbd_assert(osd_req->r_oid_len < sizeof (osd_req->r_oid));
1811 memcpy(osd_req->r_oid, obj_request->object_name, osd_req->r_oid_len);
1813 osd_req->r_file_layout = rbd_dev->layout; /* struct */
1819 static void rbd_osd_req_destroy(struct ceph_osd_request *osd_req)
1821 ceph_osdc_put_request(osd_req);
1824 /* object_name is assumed to be a non-null pointer and NUL-terminated */
1826 static struct rbd_obj_request *rbd_obj_request_create(const char *object_name,
1827 u64 offset, u64 length,
1828 enum obj_request_type type)
1830 struct rbd_obj_request *obj_request;
1834 rbd_assert(obj_request_type_valid(type));
1836 size = strlen(object_name) + 1;
1837 name = kmalloc(size, GFP_KERNEL);
1841 obj_request = kmem_cache_zalloc(rbd_obj_request_cache, GFP_KERNEL);
1847 obj_request->object_name = memcpy(name, object_name, size);
1848 obj_request->offset = offset;
1849 obj_request->length = length;
1850 obj_request->flags = 0;
1851 obj_request->which = BAD_WHICH;
1852 obj_request->type = type;
1853 INIT_LIST_HEAD(&obj_request->links);
1854 init_completion(&obj_request->completion);
1855 kref_init(&obj_request->kref);
1857 dout("%s: \"%s\" %llu/%llu %d -> obj %p\n", __func__, object_name,
1858 offset, length, (int)type, obj_request);
1863 static void rbd_obj_request_destroy(struct kref *kref)
1865 struct rbd_obj_request *obj_request;
1867 obj_request = container_of(kref, struct rbd_obj_request, kref);
1869 dout("%s: obj %p\n", __func__, obj_request);
1871 rbd_assert(obj_request->img_request == NULL);
1872 rbd_assert(obj_request->which == BAD_WHICH);
1874 if (obj_request->osd_req)
1875 rbd_osd_req_destroy(obj_request->osd_req);
1877 rbd_assert(obj_request_type_valid(obj_request->type));
1878 switch (obj_request->type) {
1879 case OBJ_REQUEST_NODATA:
1880 break; /* Nothing to do */
1881 case OBJ_REQUEST_BIO:
1882 if (obj_request->bio_list)
1883 bio_chain_put(obj_request->bio_list);
1885 case OBJ_REQUEST_PAGES:
1886 if (obj_request->pages)
1887 ceph_release_page_vector(obj_request->pages,
1888 obj_request->page_count);
1892 kfree(obj_request->object_name);
1893 obj_request->object_name = NULL;
1894 kmem_cache_free(rbd_obj_request_cache, obj_request);
1897 /* It's OK to call this for a device with no parent */
1899 static void rbd_spec_put(struct rbd_spec *spec);
1900 static void rbd_dev_unparent(struct rbd_device *rbd_dev)
1902 rbd_dev_remove_parent(rbd_dev);
1903 rbd_spec_put(rbd_dev->parent_spec);
1904 rbd_dev->parent_spec = NULL;
1905 rbd_dev->parent_overlap = 0;
1909 * Parent image reference counting is used to determine when an
1910 * image's parent fields can be safely torn down--after there are no
1911 * more in-flight requests to the parent image. When the last
1912 * reference is dropped, cleaning them up is safe.
1914 static void rbd_dev_parent_put(struct rbd_device *rbd_dev)
1918 if (!rbd_dev->parent_spec)
1921 counter = atomic_dec_return_safe(&rbd_dev->parent_ref);
1925 /* Last reference; clean up parent data structures */
1928 rbd_dev_unparent(rbd_dev);
1930 rbd_warn(rbd_dev, "parent reference underflow\n");
1934 * If an image has a non-zero parent overlap, get a reference to its
1937 * We must get the reference before checking for the overlap to
1938 * coordinate properly with zeroing the parent overlap in
1939 * rbd_dev_v2_parent_info() when an image gets flattened. We
1940 * drop it again if there is no overlap.
1942 * Returns true if the rbd device has a parent with a non-zero
1943 * overlap and a reference for it was successfully taken, or
1946 static bool rbd_dev_parent_get(struct rbd_device *rbd_dev)
1950 if (!rbd_dev->parent_spec)
1953 counter = atomic_inc_return_safe(&rbd_dev->parent_ref);
1954 if (counter > 0 && rbd_dev->parent_overlap)
1957 /* Image was flattened, but parent is not yet torn down */
1960 rbd_warn(rbd_dev, "parent reference overflow\n");
1966 * Caller is responsible for filling in the list of object requests
1967 * that comprises the image request, and the Linux request pointer
1968 * (if there is one).
1970 static struct rbd_img_request *rbd_img_request_create(
1971 struct rbd_device *rbd_dev,
1972 u64 offset, u64 length,
1975 struct rbd_img_request *img_request;
1977 img_request = kmem_cache_alloc(rbd_img_request_cache, GFP_ATOMIC);
1981 if (write_request) {
1982 down_read(&rbd_dev->header_rwsem);
1983 ceph_get_snap_context(rbd_dev->header.snapc);
1984 up_read(&rbd_dev->header_rwsem);
1987 img_request->rq = NULL;
1988 img_request->rbd_dev = rbd_dev;
1989 img_request->offset = offset;
1990 img_request->length = length;
1991 img_request->flags = 0;
1992 if (write_request) {
1993 img_request_write_set(img_request);
1994 img_request->snapc = rbd_dev->header.snapc;
1996 img_request->snap_id = rbd_dev->spec->snap_id;
1998 if (rbd_dev_parent_get(rbd_dev))
1999 img_request_layered_set(img_request);
2000 spin_lock_init(&img_request->completion_lock);
2001 img_request->next_completion = 0;
2002 img_request->callback = NULL;
2003 img_request->result = 0;
2004 img_request->obj_request_count = 0;
2005 INIT_LIST_HEAD(&img_request->obj_requests);
2006 kref_init(&img_request->kref);
2008 dout("%s: rbd_dev %p %s %llu/%llu -> img %p\n", __func__, rbd_dev,
2009 write_request ? "write" : "read", offset, length,
2015 static void rbd_img_request_destroy(struct kref *kref)
2017 struct rbd_img_request *img_request;
2018 struct rbd_obj_request *obj_request;
2019 struct rbd_obj_request *next_obj_request;
2021 img_request = container_of(kref, struct rbd_img_request, kref);
2023 dout("%s: img %p\n", __func__, img_request);
2025 for_each_obj_request_safe(img_request, obj_request, next_obj_request)
2026 rbd_img_obj_request_del(img_request, obj_request);
2027 rbd_assert(img_request->obj_request_count == 0);
2029 if (img_request_layered_test(img_request)) {
2030 img_request_layered_clear(img_request);
2031 rbd_dev_parent_put(img_request->rbd_dev);
2034 if (img_request_write_test(img_request))
2035 ceph_put_snap_context(img_request->snapc);
2037 kmem_cache_free(rbd_img_request_cache, img_request);
2040 static struct rbd_img_request *rbd_parent_request_create(
2041 struct rbd_obj_request *obj_request,
2042 u64 img_offset, u64 length)
2044 struct rbd_img_request *parent_request;
2045 struct rbd_device *rbd_dev;
2047 rbd_assert(obj_request->img_request);
2048 rbd_dev = obj_request->img_request->rbd_dev;
2050 parent_request = rbd_img_request_create(rbd_dev->parent,
2051 img_offset, length, false);
2052 if (!parent_request)
2055 img_request_child_set(parent_request);
2056 rbd_obj_request_get(obj_request);
2057 parent_request->obj_request = obj_request;
2059 return parent_request;
2062 static void rbd_parent_request_destroy(struct kref *kref)
2064 struct rbd_img_request *parent_request;
2065 struct rbd_obj_request *orig_request;
2067 parent_request = container_of(kref, struct rbd_img_request, kref);
2068 orig_request = parent_request->obj_request;
2070 parent_request->obj_request = NULL;
2071 rbd_obj_request_put(orig_request);
2072 img_request_child_clear(parent_request);
2074 rbd_img_request_destroy(kref);
2077 static bool rbd_img_obj_end_request(struct rbd_obj_request *obj_request)
2079 struct rbd_img_request *img_request;
2080 unsigned int xferred;
2084 rbd_assert(obj_request_img_data_test(obj_request));
2085 img_request = obj_request->img_request;
2087 rbd_assert(obj_request->xferred <= (u64)UINT_MAX);
2088 xferred = (unsigned int)obj_request->xferred;
2089 result = obj_request->result;
2091 struct rbd_device *rbd_dev = img_request->rbd_dev;
2093 rbd_warn(rbd_dev, "%s %llx at %llx (%llx)\n",
2094 img_request_write_test(img_request) ? "write" : "read",
2095 obj_request->length, obj_request->img_offset,
2096 obj_request->offset);
2097 rbd_warn(rbd_dev, " result %d xferred %x\n",
2099 if (!img_request->result)
2100 img_request->result = result;
2103 /* Image object requests don't own their page array */
2105 if (obj_request->type == OBJ_REQUEST_PAGES) {
2106 obj_request->pages = NULL;
2107 obj_request->page_count = 0;
2110 if (img_request_child_test(img_request)) {
2111 rbd_assert(img_request->obj_request != NULL);
2112 more = obj_request->which < img_request->obj_request_count - 1;
2114 rbd_assert(img_request->rq != NULL);
2115 more = blk_end_request(img_request->rq, result, xferred);
2121 static void rbd_img_obj_callback(struct rbd_obj_request *obj_request)
2123 struct rbd_img_request *img_request;
2124 u32 which = obj_request->which;
2127 rbd_assert(obj_request_img_data_test(obj_request));
2128 img_request = obj_request->img_request;
2130 dout("%s: img %p obj %p\n", __func__, img_request, obj_request);
2131 rbd_assert(img_request != NULL);
2132 rbd_assert(img_request->obj_request_count > 0);
2133 rbd_assert(which != BAD_WHICH);
2134 rbd_assert(which < img_request->obj_request_count);
2135 rbd_assert(which >= img_request->next_completion);
2137 spin_lock_irq(&img_request->completion_lock);
2138 if (which != img_request->next_completion)
2141 for_each_obj_request_from(img_request, obj_request) {
2143 rbd_assert(which < img_request->obj_request_count);
2145 if (!obj_request_done_test(obj_request))
2147 more = rbd_img_obj_end_request(obj_request);
2151 rbd_assert(more ^ (which == img_request->obj_request_count));
2152 img_request->next_completion = which;
2154 spin_unlock_irq(&img_request->completion_lock);
2157 rbd_img_request_complete(img_request);
2161 * Split up an image request into one or more object requests, each
2162 * to a different object. The "type" parameter indicates whether
2163 * "data_desc" is the pointer to the head of a list of bio
2164 * structures, or the base of a page array. In either case this
2165 * function assumes data_desc describes memory sufficient to hold
2166 * all data described by the image request.
2168 static int rbd_img_request_fill(struct rbd_img_request *img_request,
2169 enum obj_request_type type,
2172 struct rbd_device *rbd_dev = img_request->rbd_dev;
2173 struct rbd_obj_request *obj_request = NULL;
2174 struct rbd_obj_request *next_obj_request;
2175 bool write_request = img_request_write_test(img_request);
2176 struct bio *bio_list;
2177 unsigned int bio_offset = 0;
2178 struct page **pages;
2183 dout("%s: img %p type %d data_desc %p\n", __func__, img_request,
2184 (int)type, data_desc);
2186 opcode = write_request ? CEPH_OSD_OP_WRITE : CEPH_OSD_OP_READ;
2187 img_offset = img_request->offset;
2188 resid = img_request->length;
2189 rbd_assert(resid > 0);
2191 if (type == OBJ_REQUEST_BIO) {
2192 bio_list = data_desc;
2193 rbd_assert(img_offset == bio_list->bi_sector << SECTOR_SHIFT);
2195 rbd_assert(type == OBJ_REQUEST_PAGES);
2200 struct ceph_osd_request *osd_req;
2201 const char *object_name;
2205 object_name = rbd_segment_name(rbd_dev, img_offset);
2208 offset = rbd_segment_offset(rbd_dev, img_offset);
2209 length = rbd_segment_length(rbd_dev, img_offset, resid);
2210 obj_request = rbd_obj_request_create(object_name,
2211 offset, length, type);
2212 /* object request has its own copy of the object name */
2213 rbd_segment_name_free(object_name);
2217 if (type == OBJ_REQUEST_BIO) {
2218 unsigned int clone_size;
2220 rbd_assert(length <= (u64)UINT_MAX);
2221 clone_size = (unsigned int)length;
2222 obj_request->bio_list =
2223 bio_chain_clone_range(&bio_list,
2227 if (!obj_request->bio_list)
2230 unsigned int page_count;
2232 obj_request->pages = pages;
2233 page_count = (u32)calc_pages_for(offset, length);
2234 obj_request->page_count = page_count;
2235 if ((offset + length) & ~PAGE_MASK)
2236 page_count--; /* more on last page */
2237 pages += page_count;
2240 osd_req = rbd_osd_req_create(rbd_dev, write_request,
2244 obj_request->osd_req = osd_req;
2245 obj_request->callback = rbd_img_obj_callback;
2247 osd_req_op_extent_init(osd_req, 0, opcode, offset, length,
2249 if (type == OBJ_REQUEST_BIO)
2250 osd_req_op_extent_osd_data_bio(osd_req, 0,
2251 obj_request->bio_list, length);
2253 osd_req_op_extent_osd_data_pages(osd_req, 0,
2254 obj_request->pages, length,
2255 offset & ~PAGE_MASK, false, false);
2258 * set obj_request->img_request before formatting
2259 * the osd_request so that it gets the right snapc
2261 rbd_img_obj_request_add(img_request, obj_request);
2263 rbd_osd_req_format_write(obj_request);
2265 rbd_osd_req_format_read(obj_request);
2267 obj_request->img_offset = img_offset;
2269 img_offset += length;
2276 rbd_obj_request_put(obj_request);
2278 for_each_obj_request_safe(img_request, obj_request, next_obj_request)
2279 rbd_obj_request_put(obj_request);
2285 rbd_img_obj_copyup_callback(struct rbd_obj_request *obj_request)
2287 struct rbd_img_request *img_request;
2288 struct rbd_device *rbd_dev;
2289 struct page **pages;
2292 rbd_assert(obj_request->type == OBJ_REQUEST_BIO);
2293 rbd_assert(obj_request_img_data_test(obj_request));
2294 img_request = obj_request->img_request;
2295 rbd_assert(img_request);
2297 rbd_dev = img_request->rbd_dev;
2298 rbd_assert(rbd_dev);
2300 pages = obj_request->copyup_pages;
2301 rbd_assert(pages != NULL);
2302 obj_request->copyup_pages = NULL;
2303 page_count = obj_request->copyup_page_count;
2304 rbd_assert(page_count);
2305 obj_request->copyup_page_count = 0;
2306 ceph_release_page_vector(pages, page_count);
2309 * We want the transfer count to reflect the size of the
2310 * original write request. There is no such thing as a
2311 * successful short write, so if the request was successful
2312 * we can just set it to the originally-requested length.
2314 if (!obj_request->result)
2315 obj_request->xferred = obj_request->length;
2317 /* Finish up with the normal image object callback */
2319 rbd_img_obj_callback(obj_request);
2323 rbd_img_obj_parent_read_full_callback(struct rbd_img_request *img_request)
2325 struct rbd_obj_request *orig_request;
2326 struct ceph_osd_request *osd_req;
2327 struct ceph_osd_client *osdc;
2328 struct rbd_device *rbd_dev;
2329 struct page **pages;
2336 rbd_assert(img_request_child_test(img_request));
2338 /* First get what we need from the image request */
2340 pages = img_request->copyup_pages;
2341 rbd_assert(pages != NULL);
2342 img_request->copyup_pages = NULL;
2343 page_count = img_request->copyup_page_count;
2344 rbd_assert(page_count);
2345 img_request->copyup_page_count = 0;
2347 orig_request = img_request->obj_request;
2348 rbd_assert(orig_request != NULL);
2349 rbd_assert(obj_request_type_valid(orig_request->type));
2350 img_result = img_request->result;
2351 parent_length = img_request->length;
2352 rbd_assert(parent_length == img_request->xferred);
2353 rbd_img_request_put(img_request);
2355 rbd_assert(orig_request->img_request);
2356 rbd_dev = orig_request->img_request->rbd_dev;
2357 rbd_assert(rbd_dev);
2360 * If the overlap has become 0 (most likely because the
2361 * image has been flattened) we need to free the pages
2362 * and re-submit the original write request.
2364 if (!rbd_dev->parent_overlap) {
2365 struct ceph_osd_client *osdc;
2367 ceph_release_page_vector(pages, page_count);
2368 osdc = &rbd_dev->rbd_client->client->osdc;
2369 img_result = rbd_obj_request_submit(osdc, orig_request);
2378 * The original osd request is of no use to use any more.
2379 * We need a new one that can hold the two ops in a copyup
2380 * request. Allocate the new copyup osd request for the
2381 * original request, and release the old one.
2383 img_result = -ENOMEM;
2384 osd_req = rbd_osd_req_create_copyup(orig_request);
2387 rbd_osd_req_destroy(orig_request->osd_req);
2388 orig_request->osd_req = osd_req;
2389 orig_request->copyup_pages = pages;
2390 orig_request->copyup_page_count = page_count;
2392 /* Initialize the copyup op */
2394 osd_req_op_cls_init(osd_req, 0, CEPH_OSD_OP_CALL, "rbd", "copyup");
2395 osd_req_op_cls_request_data_pages(osd_req, 0, pages, parent_length, 0,
2398 /* Then the original write request op */
2400 offset = orig_request->offset;
2401 length = orig_request->length;
2402 osd_req_op_extent_init(osd_req, 1, CEPH_OSD_OP_WRITE,
2403 offset, length, 0, 0);
2404 if (orig_request->type == OBJ_REQUEST_BIO)
2405 osd_req_op_extent_osd_data_bio(osd_req, 1,
2406 orig_request->bio_list, length);
2408 osd_req_op_extent_osd_data_pages(osd_req, 1,
2409 orig_request->pages, length,
2410 offset & ~PAGE_MASK, false, false);
2412 rbd_osd_req_format_write(orig_request);
2414 /* All set, send it off. */
2416 orig_request->callback = rbd_img_obj_copyup_callback;
2417 osdc = &rbd_dev->rbd_client->client->osdc;
2418 img_result = rbd_obj_request_submit(osdc, orig_request);
2422 /* Record the error code and complete the request */
2424 orig_request->result = img_result;
2425 orig_request->xferred = 0;
2426 obj_request_done_set(orig_request);
2427 rbd_obj_request_complete(orig_request);
2431 * Read from the parent image the range of data that covers the
2432 * entire target of the given object request. This is used for
2433 * satisfying a layered image write request when the target of an
2434 * object request from the image request does not exist.
2436 * A page array big enough to hold the returned data is allocated
2437 * and supplied to rbd_img_request_fill() as the "data descriptor."
2438 * When the read completes, this page array will be transferred to
2439 * the original object request for the copyup operation.
2441 * If an error occurs, record it as the result of the original
2442 * object request and mark it done so it gets completed.
2444 static int rbd_img_obj_parent_read_full(struct rbd_obj_request *obj_request)
2446 struct rbd_img_request *img_request = NULL;
2447 struct rbd_img_request *parent_request = NULL;
2448 struct rbd_device *rbd_dev;
2451 struct page **pages = NULL;
2455 rbd_assert(obj_request_img_data_test(obj_request));
2456 rbd_assert(obj_request_type_valid(obj_request->type));
2458 img_request = obj_request->img_request;
2459 rbd_assert(img_request != NULL);
2460 rbd_dev = img_request->rbd_dev;
2461 rbd_assert(rbd_dev->parent != NULL);
2464 * Determine the byte range covered by the object in the
2465 * child image to which the original request was to be sent.
2467 img_offset = obj_request->img_offset - obj_request->offset;
2468 length = (u64)1 << rbd_dev->header.obj_order;
2471 * There is no defined parent data beyond the parent
2472 * overlap, so limit what we read at that boundary if
2475 if (img_offset + length > rbd_dev->parent_overlap) {
2476 rbd_assert(img_offset < rbd_dev->parent_overlap);
2477 length = rbd_dev->parent_overlap - img_offset;
2481 * Allocate a page array big enough to receive the data read
2484 page_count = (u32)calc_pages_for(0, length);
2485 pages = ceph_alloc_page_vector(page_count, GFP_KERNEL);
2486 if (IS_ERR(pages)) {
2487 result = PTR_ERR(pages);
2493 parent_request = rbd_parent_request_create(obj_request,
2494 img_offset, length);
2495 if (!parent_request)
2498 result = rbd_img_request_fill(parent_request, OBJ_REQUEST_PAGES, pages);
2501 parent_request->copyup_pages = pages;
2502 parent_request->copyup_page_count = page_count;
2504 parent_request->callback = rbd_img_obj_parent_read_full_callback;
2505 result = rbd_img_request_submit(parent_request);
2509 parent_request->copyup_pages = NULL;
2510 parent_request->copyup_page_count = 0;
2511 parent_request->obj_request = NULL;
2512 rbd_obj_request_put(obj_request);
2515 ceph_release_page_vector(pages, page_count);
2517 rbd_img_request_put(parent_request);
2518 obj_request->result = result;
2519 obj_request->xferred = 0;
2520 obj_request_done_set(obj_request);
2525 static void rbd_img_obj_exists_callback(struct rbd_obj_request *obj_request)
2527 struct rbd_obj_request *orig_request;
2528 struct rbd_device *rbd_dev;
2531 rbd_assert(!obj_request_img_data_test(obj_request));
2534 * All we need from the object request is the original
2535 * request and the result of the STAT op. Grab those, then
2536 * we're done with the request.
2538 orig_request = obj_request->obj_request;
2539 obj_request->obj_request = NULL;
2540 rbd_assert(orig_request);
2541 rbd_assert(orig_request->img_request);
2543 result = obj_request->result;
2544 obj_request->result = 0;
2546 dout("%s: obj %p for obj %p result %d %llu/%llu\n", __func__,
2547 obj_request, orig_request, result,
2548 obj_request->xferred, obj_request->length);
2549 rbd_obj_request_put(obj_request);
2552 * If the overlap has become 0 (most likely because the
2553 * image has been flattened) we need to free the pages
2554 * and re-submit the original write request.
2556 rbd_dev = orig_request->img_request->rbd_dev;
2557 if (!rbd_dev->parent_overlap) {
2558 struct ceph_osd_client *osdc;
2560 rbd_obj_request_put(orig_request);
2561 osdc = &rbd_dev->rbd_client->client->osdc;
2562 result = rbd_obj_request_submit(osdc, orig_request);
2568 * Our only purpose here is to determine whether the object
2569 * exists, and we don't want to treat the non-existence as
2570 * an error. If something else comes back, transfer the
2571 * error to the original request and complete it now.
2574 obj_request_existence_set(orig_request, true);
2575 } else if (result == -ENOENT) {
2576 obj_request_existence_set(orig_request, false);
2577 } else if (result) {
2578 orig_request->result = result;
2583 * Resubmit the original request now that we have recorded
2584 * whether the target object exists.
2586 orig_request->result = rbd_img_obj_request_submit(orig_request);
2588 if (orig_request->result)
2589 rbd_obj_request_complete(orig_request);
2590 rbd_obj_request_put(orig_request);
2593 static int rbd_img_obj_exists_submit(struct rbd_obj_request *obj_request)
2595 struct rbd_obj_request *stat_request;
2596 struct rbd_device *rbd_dev;
2597 struct ceph_osd_client *osdc;
2598 struct page **pages = NULL;
2604 * The response data for a STAT call consists of:
2611 size = sizeof (__le64) + sizeof (__le32) + sizeof (__le32);
2612 page_count = (u32)calc_pages_for(0, size);
2613 pages = ceph_alloc_page_vector(page_count, GFP_KERNEL);
2615 return PTR_ERR(pages);
2618 stat_request = rbd_obj_request_create(obj_request->object_name, 0, 0,
2623 rbd_obj_request_get(obj_request);
2624 stat_request->obj_request = obj_request;
2625 stat_request->pages = pages;
2626 stat_request->page_count = page_count;
2628 rbd_assert(obj_request->img_request);
2629 rbd_dev = obj_request->img_request->rbd_dev;
2630 stat_request->osd_req = rbd_osd_req_create(rbd_dev, false,
2632 if (!stat_request->osd_req)
2634 stat_request->callback = rbd_img_obj_exists_callback;
2636 osd_req_op_init(stat_request->osd_req, 0, CEPH_OSD_OP_STAT);
2637 osd_req_op_raw_data_in_pages(stat_request->osd_req, 0, pages, size, 0,
2639 rbd_osd_req_format_read(stat_request);
2641 osdc = &rbd_dev->rbd_client->client->osdc;
2642 ret = rbd_obj_request_submit(osdc, stat_request);
2645 rbd_obj_request_put(obj_request);
2650 static int rbd_img_obj_request_submit(struct rbd_obj_request *obj_request)
2652 struct rbd_img_request *img_request;
2653 struct rbd_device *rbd_dev;
2656 rbd_assert(obj_request_img_data_test(obj_request));
2658 img_request = obj_request->img_request;
2659 rbd_assert(img_request);
2660 rbd_dev = img_request->rbd_dev;
2663 * Only writes to layered images need special handling.
2664 * Reads and non-layered writes are simple object requests.
2665 * Layered writes that start beyond the end of the overlap
2666 * with the parent have no parent data, so they too are
2667 * simple object requests. Finally, if the target object is
2668 * known to already exist, its parent data has already been
2669 * copied, so a write to the object can also be handled as a
2670 * simple object request.
2672 if (!img_request_write_test(img_request) ||
2673 !img_request_layered_test(img_request) ||
2674 rbd_dev->parent_overlap <= obj_request->img_offset ||
2675 ((known = obj_request_known_test(obj_request)) &&
2676 obj_request_exists_test(obj_request))) {
2678 struct rbd_device *rbd_dev;
2679 struct ceph_osd_client *osdc;
2681 rbd_dev = obj_request->img_request->rbd_dev;
2682 osdc = &rbd_dev->rbd_client->client->osdc;
2684 return rbd_obj_request_submit(osdc, obj_request);
2688 * It's a layered write. The target object might exist but
2689 * we may not know that yet. If we know it doesn't exist,
2690 * start by reading the data for the full target object from
2691 * the parent so we can use it for a copyup to the target.
2694 return rbd_img_obj_parent_read_full(obj_request);
2696 /* We don't know whether the target exists. Go find out. */
2698 return rbd_img_obj_exists_submit(obj_request);
2701 static int rbd_img_request_submit(struct rbd_img_request *img_request)
2703 struct rbd_obj_request *obj_request;
2704 struct rbd_obj_request *next_obj_request;
2706 dout("%s: img %p\n", __func__, img_request);
2707 for_each_obj_request_safe(img_request, obj_request, next_obj_request) {
2710 ret = rbd_img_obj_request_submit(obj_request);
2718 static void rbd_img_parent_read_callback(struct rbd_img_request *img_request)
2720 struct rbd_obj_request *obj_request;
2721 struct rbd_device *rbd_dev;
2726 rbd_assert(img_request_child_test(img_request));
2728 /* First get what we need from the image request and release it */
2730 obj_request = img_request->obj_request;
2731 img_xferred = img_request->xferred;
2732 img_result = img_request->result;
2733 rbd_img_request_put(img_request);
2736 * If the overlap has become 0 (most likely because the
2737 * image has been flattened) we need to re-submit the
2740 rbd_assert(obj_request);
2741 rbd_assert(obj_request->img_request);
2742 rbd_dev = obj_request->img_request->rbd_dev;
2743 if (!rbd_dev->parent_overlap) {
2744 struct ceph_osd_client *osdc;
2746 osdc = &rbd_dev->rbd_client->client->osdc;
2747 img_result = rbd_obj_request_submit(osdc, obj_request);
2752 obj_request->result = img_result;
2753 if (obj_request->result)
2757 * We need to zero anything beyond the parent overlap
2758 * boundary. Since rbd_img_obj_request_read_callback()
2759 * will zero anything beyond the end of a short read, an
2760 * easy way to do this is to pretend the data from the
2761 * parent came up short--ending at the overlap boundary.
2763 rbd_assert(obj_request->img_offset < U64_MAX - obj_request->length);
2764 obj_end = obj_request->img_offset + obj_request->length;
2765 if (obj_end > rbd_dev->parent_overlap) {
2768 if (obj_request->img_offset < rbd_dev->parent_overlap)
2769 xferred = rbd_dev->parent_overlap -
2770 obj_request->img_offset;
2772 obj_request->xferred = min(img_xferred, xferred);
2774 obj_request->xferred = img_xferred;
2777 rbd_img_obj_request_read_callback(obj_request);
2778 rbd_obj_request_complete(obj_request);
2781 static void rbd_img_parent_read(struct rbd_obj_request *obj_request)
2783 struct rbd_img_request *img_request;
2786 rbd_assert(obj_request_img_data_test(obj_request));
2787 rbd_assert(obj_request->img_request != NULL);
2788 rbd_assert(obj_request->result == (s32) -ENOENT);
2789 rbd_assert(obj_request_type_valid(obj_request->type));
2791 /* rbd_read_finish(obj_request, obj_request->length); */
2792 img_request = rbd_parent_request_create(obj_request,
2793 obj_request->img_offset,
2794 obj_request->length);
2799 if (obj_request->type == OBJ_REQUEST_BIO)
2800 result = rbd_img_request_fill(img_request, OBJ_REQUEST_BIO,
2801 obj_request->bio_list);
2803 result = rbd_img_request_fill(img_request, OBJ_REQUEST_PAGES,
2804 obj_request->pages);
2808 img_request->callback = rbd_img_parent_read_callback;
2809 result = rbd_img_request_submit(img_request);
2816 rbd_img_request_put(img_request);
2817 obj_request->result = result;
2818 obj_request->xferred = 0;
2819 obj_request_done_set(obj_request);
2822 static int rbd_obj_notify_ack(struct rbd_device *rbd_dev, u64 notify_id)
2824 struct rbd_obj_request *obj_request;
2825 struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
2828 obj_request = rbd_obj_request_create(rbd_dev->header_name, 0, 0,
2829 OBJ_REQUEST_NODATA);
2834 obj_request->osd_req = rbd_osd_req_create(rbd_dev, false, obj_request);
2835 if (!obj_request->osd_req)
2837 obj_request->callback = rbd_obj_request_put;
2839 osd_req_op_watch_init(obj_request->osd_req, 0, CEPH_OSD_OP_NOTIFY_ACK,
2841 rbd_osd_req_format_read(obj_request);
2843 ret = rbd_obj_request_submit(osdc, obj_request);
2846 rbd_obj_request_put(obj_request);
2851 static void rbd_watch_cb(u64 ver, u64 notify_id, u8 opcode, void *data)
2853 struct rbd_device *rbd_dev = (struct rbd_device *)data;
2859 dout("%s: \"%s\" notify_id %llu opcode %u\n", __func__,
2860 rbd_dev->header_name, (unsigned long long)notify_id,
2861 (unsigned int)opcode);
2862 ret = rbd_dev_refresh(rbd_dev);
2864 rbd_warn(rbd_dev, ": header refresh error (%d)\n", ret);
2866 rbd_obj_notify_ack(rbd_dev, notify_id);
2870 * Request sync osd watch/unwatch. The value of "start" determines
2871 * whether a watch request is being initiated or torn down.
2873 static int rbd_dev_header_watch_sync(struct rbd_device *rbd_dev, bool start)
2875 struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
2876 struct rbd_obj_request *obj_request;
2879 rbd_assert(start ^ !!rbd_dev->watch_event);
2880 rbd_assert(start ^ !!rbd_dev->watch_request);
2883 ret = ceph_osdc_create_event(osdc, rbd_watch_cb, rbd_dev,
2884 &rbd_dev->watch_event);
2887 rbd_assert(rbd_dev->watch_event != NULL);
2891 obj_request = rbd_obj_request_create(rbd_dev->header_name, 0, 0,
2892 OBJ_REQUEST_NODATA);
2896 obj_request->osd_req = rbd_osd_req_create(rbd_dev, true, obj_request);
2897 if (!obj_request->osd_req)
2901 ceph_osdc_set_request_linger(osdc, obj_request->osd_req);
2903 ceph_osdc_unregister_linger_request(osdc,
2904 rbd_dev->watch_request->osd_req);
2906 osd_req_op_watch_init(obj_request->osd_req, 0, CEPH_OSD_OP_WATCH,
2907 rbd_dev->watch_event->cookie, 0, start ? 1 : 0);
2908 rbd_osd_req_format_write(obj_request);
2910 ret = rbd_obj_request_submit(osdc, obj_request);
2913 ret = rbd_obj_request_wait(obj_request);
2916 ret = obj_request->result;
2921 * A watch request is set to linger, so the underlying osd
2922 * request won't go away until we unregister it. We retain
2923 * a pointer to the object request during that time (in
2924 * rbd_dev->watch_request), so we'll keep a reference to
2925 * it. We'll drop that reference (below) after we've
2929 rbd_dev->watch_request = obj_request;
2934 /* We have successfully torn down the watch request */
2936 rbd_obj_request_put(rbd_dev->watch_request);
2937 rbd_dev->watch_request = NULL;
2939 /* Cancel the event if we're tearing down, or on error */
2940 ceph_osdc_cancel_event(rbd_dev->watch_event);
2941 rbd_dev->watch_event = NULL;
2943 rbd_obj_request_put(obj_request);
2949 * Synchronous osd object method call. Returns the number of bytes
2950 * returned in the outbound buffer, or a negative error code.
2952 static int rbd_obj_method_sync(struct rbd_device *rbd_dev,
2953 const char *object_name,
2954 const char *class_name,
2955 const char *method_name,
2956 const void *outbound,
2957 size_t outbound_size,
2959 size_t inbound_size)
2961 struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
2962 struct rbd_obj_request *obj_request;
2963 struct page **pages;
2968 * Method calls are ultimately read operations. The result
2969 * should placed into the inbound buffer provided. They
2970 * also supply outbound data--parameters for the object
2971 * method. Currently if this is present it will be a
2974 page_count = (u32)calc_pages_for(0, inbound_size);
2975 pages = ceph_alloc_page_vector(page_count, GFP_KERNEL);
2977 return PTR_ERR(pages);
2980 obj_request = rbd_obj_request_create(object_name, 0, inbound_size,
2985 obj_request->pages = pages;
2986 obj_request->page_count = page_count;
2988 obj_request->osd_req = rbd_osd_req_create(rbd_dev, false, obj_request);
2989 if (!obj_request->osd_req)
2992 osd_req_op_cls_init(obj_request->osd_req, 0, CEPH_OSD_OP_CALL,
2993 class_name, method_name);
2994 if (outbound_size) {
2995 struct ceph_pagelist *pagelist;
2997 pagelist = kmalloc(sizeof (*pagelist), GFP_NOFS);
3001 ceph_pagelist_init(pagelist);
3002 ceph_pagelist_append(pagelist, outbound, outbound_size);
3003 osd_req_op_cls_request_data_pagelist(obj_request->osd_req, 0,
3006 osd_req_op_cls_response_data_pages(obj_request->osd_req, 0,
3007 obj_request->pages, inbound_size,
3009 rbd_osd_req_format_read(obj_request);
3011 ret = rbd_obj_request_submit(osdc, obj_request);
3014 ret = rbd_obj_request_wait(obj_request);
3018 ret = obj_request->result;
3022 rbd_assert(obj_request->xferred < (u64)INT_MAX);
3023 ret = (int)obj_request->xferred;
3024 ceph_copy_from_page_vector(pages, inbound, 0, obj_request->xferred);
3027 rbd_obj_request_put(obj_request);
3029 ceph_release_page_vector(pages, page_count);
3034 static void rbd_request_fn(struct request_queue *q)
3035 __releases(q->queue_lock) __acquires(q->queue_lock)
3037 struct rbd_device *rbd_dev = q->queuedata;
3038 bool read_only = rbd_dev->mapping.read_only;
3042 while ((rq = blk_fetch_request(q))) {
3043 bool write_request = rq_data_dir(rq) == WRITE;
3044 struct rbd_img_request *img_request;
3048 /* Ignore any non-FS requests that filter through. */
3050 if (rq->cmd_type != REQ_TYPE_FS) {
3051 dout("%s: non-fs request type %d\n", __func__,
3052 (int) rq->cmd_type);
3053 __blk_end_request_all(rq, 0);
3057 /* Ignore/skip any zero-length requests */
3059 offset = (u64) blk_rq_pos(rq) << SECTOR_SHIFT;
3060 length = (u64) blk_rq_bytes(rq);
3063 dout("%s: zero-length request\n", __func__);
3064 __blk_end_request_all(rq, 0);
3068 spin_unlock_irq(q->queue_lock);
3070 /* Disallow writes to a read-only device */
3072 if (write_request) {
3076 rbd_assert(rbd_dev->spec->snap_id == CEPH_NOSNAP);
3080 * Quit early if the mapped snapshot no longer
3081 * exists. It's still possible the snapshot will
3082 * have disappeared by the time our request arrives
3083 * at the osd, but there's no sense in sending it if
3086 if (!test_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags)) {
3087 dout("request for non-existent snapshot");
3088 rbd_assert(rbd_dev->spec->snap_id != CEPH_NOSNAP);
3094 if (offset && length > U64_MAX - offset + 1) {
3095 rbd_warn(rbd_dev, "bad request range (%llu~%llu)\n",
3097 goto end_request; /* Shouldn't happen */
3101 if (offset + length > rbd_dev->mapping.size) {
3102 rbd_warn(rbd_dev, "beyond EOD (%llu~%llu > %llu)\n",
3103 offset, length, rbd_dev->mapping.size);
3108 img_request = rbd_img_request_create(rbd_dev, offset, length,
3113 img_request->rq = rq;
3115 result = rbd_img_request_fill(img_request, OBJ_REQUEST_BIO,
3118 result = rbd_img_request_submit(img_request);
3120 rbd_img_request_put(img_request);
3122 spin_lock_irq(q->queue_lock);
3124 rbd_warn(rbd_dev, "%s %llx at %llx result %d\n",
3125 write_request ? "write" : "read",
3126 length, offset, result);
3128 __blk_end_request_all(rq, result);
3134 * a queue callback. Makes sure that we don't create a bio that spans across
3135 * multiple osd objects. One exception would be with a single page bios,
3136 * which we handle later at bio_chain_clone_range()
3138 static int rbd_merge_bvec(struct request_queue *q, struct bvec_merge_data *bmd,
3139 struct bio_vec *bvec)
3141 struct rbd_device *rbd_dev = q->queuedata;
3142 sector_t sector_offset;
3143 sector_t sectors_per_obj;
3144 sector_t obj_sector_offset;
3148 * Find how far into its rbd object the partition-relative
3149 * bio start sector is to offset relative to the enclosing
3152 sector_offset = get_start_sect(bmd->bi_bdev) + bmd->bi_sector;
3153 sectors_per_obj = 1 << (rbd_dev->header.obj_order - SECTOR_SHIFT);
3154 obj_sector_offset = sector_offset & (sectors_per_obj - 1);
3157 * Compute the number of bytes from that offset to the end
3158 * of the object. Account for what's already used by the bio.
3160 ret = (int) (sectors_per_obj - obj_sector_offset) << SECTOR_SHIFT;
3161 if (ret > bmd->bi_size)
3162 ret -= bmd->bi_size;
3167 * Don't send back more than was asked for. And if the bio
3168 * was empty, let the whole thing through because: "Note
3169 * that a block device *must* allow a single page to be
3170 * added to an empty bio."
3172 rbd_assert(bvec->bv_len <= PAGE_SIZE);
3173 if (ret > (int) bvec->bv_len || !bmd->bi_size)
3174 ret = (int) bvec->bv_len;
3179 static void rbd_free_disk(struct rbd_device *rbd_dev)
3181 struct gendisk *disk = rbd_dev->disk;
3186 rbd_dev->disk = NULL;
3187 if (disk->flags & GENHD_FL_UP) {
3190 blk_cleanup_queue(disk->queue);
3195 static int rbd_obj_read_sync(struct rbd_device *rbd_dev,
3196 const char *object_name,
3197 u64 offset, u64 length, void *buf)
3200 struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
3201 struct rbd_obj_request *obj_request;
3202 struct page **pages = NULL;
3207 page_count = (u32) calc_pages_for(offset, length);
3208 pages = ceph_alloc_page_vector(page_count, GFP_KERNEL);
3210 ret = PTR_ERR(pages);
3213 obj_request = rbd_obj_request_create(object_name, offset, length,
3218 obj_request->pages = pages;
3219 obj_request->page_count = page_count;
3221 obj_request->osd_req = rbd_osd_req_create(rbd_dev, false, obj_request);
3222 if (!obj_request->osd_req)
3225 osd_req_op_extent_init(obj_request->osd_req, 0, CEPH_OSD_OP_READ,
3226 offset, length, 0, 0);
3227 osd_req_op_extent_osd_data_pages(obj_request->osd_req, 0,
3229 obj_request->length,
3230 obj_request->offset & ~PAGE_MASK,
3232 rbd_osd_req_format_read(obj_request);
3234 ret = rbd_obj_request_submit(osdc, obj_request);
3237 ret = rbd_obj_request_wait(obj_request);
3241 ret = obj_request->result;
3245 rbd_assert(obj_request->xferred <= (u64) SIZE_MAX);
3246 size = (size_t) obj_request->xferred;
3247 ceph_copy_from_page_vector(pages, buf, 0, size);
3248 rbd_assert(size <= (size_t)INT_MAX);
3252 rbd_obj_request_put(obj_request);
3254 ceph_release_page_vector(pages, page_count);
3260 * Read the complete header for the given rbd device. On successful
3261 * return, the rbd_dev->header field will contain up-to-date
3262 * information about the image.
3264 static int rbd_dev_v1_header_info(struct rbd_device *rbd_dev)
3266 struct rbd_image_header_ondisk *ondisk = NULL;
3273 * The complete header will include an array of its 64-bit
3274 * snapshot ids, followed by the names of those snapshots as
3275 * a contiguous block of NUL-terminated strings. Note that
3276 * the number of snapshots could change by the time we read
3277 * it in, in which case we re-read it.
3284 size = sizeof (*ondisk);
3285 size += snap_count * sizeof (struct rbd_image_snap_ondisk);
3287 ondisk = kmalloc(size, GFP_KERNEL);
3291 ret = rbd_obj_read_sync(rbd_dev, rbd_dev->header_name,
3295 if ((size_t)ret < size) {
3297 rbd_warn(rbd_dev, "short header read (want %zd got %d)",
3301 if (!rbd_dev_ondisk_valid(ondisk)) {
3303 rbd_warn(rbd_dev, "invalid header");
3307 names_size = le64_to_cpu(ondisk->snap_names_len);
3308 want_count = snap_count;
3309 snap_count = le32_to_cpu(ondisk->snap_count);
3310 } while (snap_count != want_count);
3312 ret = rbd_header_from_disk(rbd_dev, ondisk);
3320 * Clear the rbd device's EXISTS flag if the snapshot it's mapped to
3321 * has disappeared from the (just updated) snapshot context.
3323 static void rbd_exists_validate(struct rbd_device *rbd_dev)
3327 if (!test_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags))
3330 snap_id = rbd_dev->spec->snap_id;
3331 if (snap_id == CEPH_NOSNAP)
3334 if (rbd_dev_snap_index(rbd_dev, snap_id) == BAD_SNAP_INDEX)
3335 clear_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags);
3338 static int rbd_dev_refresh(struct rbd_device *rbd_dev)
3343 rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
3344 mapping_size = rbd_dev->mapping.size;
3345 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
3346 if (rbd_dev->image_format == 1)
3347 ret = rbd_dev_v1_header_info(rbd_dev);
3349 ret = rbd_dev_v2_header_info(rbd_dev);
3351 /* If it's a mapped snapshot, validate its EXISTS flag */
3353 rbd_exists_validate(rbd_dev);
3354 mutex_unlock(&ctl_mutex);
3355 if (mapping_size != rbd_dev->mapping.size) {
3358 size = (sector_t)rbd_dev->mapping.size / SECTOR_SIZE;
3359 dout("setting size to %llu sectors", (unsigned long long)size);
3360 set_capacity(rbd_dev->disk, size);
3361 revalidate_disk(rbd_dev->disk);
3367 static int rbd_init_disk(struct rbd_device *rbd_dev)
3369 struct gendisk *disk;
3370 struct request_queue *q;
3373 /* create gendisk info */
3374 disk = alloc_disk(RBD_MINORS_PER_MAJOR);
3378 snprintf(disk->disk_name, sizeof(disk->disk_name), RBD_DRV_NAME "%d",
3380 disk->major = rbd_dev->major;
3381 disk->first_minor = 0;
3382 disk->fops = &rbd_bd_ops;
3383 disk->private_data = rbd_dev;
3385 q = blk_init_queue(rbd_request_fn, &rbd_dev->lock);
3389 /* We use the default size, but let's be explicit about it. */
3390 blk_queue_physical_block_size(q, SECTOR_SIZE);
3392 /* set io sizes to object size */
3393 segment_size = rbd_obj_bytes(&rbd_dev->header);
3394 blk_queue_max_hw_sectors(q, segment_size / SECTOR_SIZE);
3395 blk_queue_max_segment_size(q, segment_size);
3396 blk_queue_io_min(q, segment_size);
3397 blk_queue_io_opt(q, segment_size);
3399 blk_queue_merge_bvec(q, rbd_merge_bvec);
3402 q->queuedata = rbd_dev;
3404 rbd_dev->disk = disk;
3417 static struct rbd_device *dev_to_rbd_dev(struct device *dev)
3419 return container_of(dev, struct rbd_device, dev);
3422 static ssize_t rbd_size_show(struct device *dev,
3423 struct device_attribute *attr, char *buf)
3425 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3427 return sprintf(buf, "%llu\n",
3428 (unsigned long long)rbd_dev->mapping.size);
3432 * Note this shows the features for whatever's mapped, which is not
3433 * necessarily the base image.
3435 static ssize_t rbd_features_show(struct device *dev,
3436 struct device_attribute *attr, char *buf)
3438 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3440 return sprintf(buf, "0x%016llx\n",
3441 (unsigned long long)rbd_dev->mapping.features);
3444 static ssize_t rbd_major_show(struct device *dev,
3445 struct device_attribute *attr, char *buf)
3447 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3450 return sprintf(buf, "%d\n", rbd_dev->major);
3452 return sprintf(buf, "(none)\n");
3456 static ssize_t rbd_client_id_show(struct device *dev,
3457 struct device_attribute *attr, char *buf)
3459 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3461 return sprintf(buf, "client%lld\n",
3462 ceph_client_id(rbd_dev->rbd_client->client));
3465 static ssize_t rbd_pool_show(struct device *dev,
3466 struct device_attribute *attr, char *buf)
3468 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3470 return sprintf(buf, "%s\n", rbd_dev->spec->pool_name);
3473 static ssize_t rbd_pool_id_show(struct device *dev,
3474 struct device_attribute *attr, char *buf)
3476 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3478 return sprintf(buf, "%llu\n",
3479 (unsigned long long) rbd_dev->spec->pool_id);
3482 static ssize_t rbd_name_show(struct device *dev,
3483 struct device_attribute *attr, char *buf)
3485 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3487 if (rbd_dev->spec->image_name)
3488 return sprintf(buf, "%s\n", rbd_dev->spec->image_name);
3490 return sprintf(buf, "(unknown)\n");
3493 static ssize_t rbd_image_id_show(struct device *dev,
3494 struct device_attribute *attr, char *buf)
3496 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3498 return sprintf(buf, "%s\n", rbd_dev->spec->image_id);
3502 * Shows the name of the currently-mapped snapshot (or
3503 * RBD_SNAP_HEAD_NAME for the base image).
3505 static ssize_t rbd_snap_show(struct device *dev,
3506 struct device_attribute *attr,
3509 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3511 return sprintf(buf, "%s\n", rbd_dev->spec->snap_name);
3515 * For an rbd v2 image, shows the pool id, image id, and snapshot id
3516 * for the parent image. If there is no parent, simply shows
3517 * "(no parent image)".
3519 static ssize_t rbd_parent_show(struct device *dev,
3520 struct device_attribute *attr,
3523 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3524 struct rbd_spec *spec = rbd_dev->parent_spec;
3529 return sprintf(buf, "(no parent image)\n");
3531 count = sprintf(bufp, "pool_id %llu\npool_name %s\n",
3532 (unsigned long long) spec->pool_id, spec->pool_name);
3537 count = sprintf(bufp, "image_id %s\nimage_name %s\n", spec->image_id,
3538 spec->image_name ? spec->image_name : "(unknown)");
3543 count = sprintf(bufp, "snap_id %llu\nsnap_name %s\n",
3544 (unsigned long long) spec->snap_id, spec->snap_name);
3549 count = sprintf(bufp, "overlap %llu\n", rbd_dev->parent_overlap);
3554 return (ssize_t) (bufp - buf);
3557 static ssize_t rbd_image_refresh(struct device *dev,
3558 struct device_attribute *attr,
3562 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3565 ret = rbd_dev_refresh(rbd_dev);
3567 rbd_warn(rbd_dev, ": manual header refresh error (%d)\n", ret);
3569 return ret < 0 ? ret : size;
3572 static DEVICE_ATTR(size, S_IRUGO, rbd_size_show, NULL);
3573 static DEVICE_ATTR(features, S_IRUGO, rbd_features_show, NULL);
3574 static DEVICE_ATTR(major, S_IRUGO, rbd_major_show, NULL);
3575 static DEVICE_ATTR(client_id, S_IRUGO, rbd_client_id_show, NULL);
3576 static DEVICE_ATTR(pool, S_IRUGO, rbd_pool_show, NULL);
3577 static DEVICE_ATTR(pool_id, S_IRUGO, rbd_pool_id_show, NULL);
3578 static DEVICE_ATTR(name, S_IRUGO, rbd_name_show, NULL);
3579 static DEVICE_ATTR(image_id, S_IRUGO, rbd_image_id_show, NULL);
3580 static DEVICE_ATTR(refresh, S_IWUSR, NULL, rbd_image_refresh);
3581 static DEVICE_ATTR(current_snap, S_IRUGO, rbd_snap_show, NULL);
3582 static DEVICE_ATTR(parent, S_IRUGO, rbd_parent_show, NULL);
3584 static struct attribute *rbd_attrs[] = {
3585 &dev_attr_size.attr,
3586 &dev_attr_features.attr,
3587 &dev_attr_major.attr,
3588 &dev_attr_client_id.attr,
3589 &dev_attr_pool.attr,
3590 &dev_attr_pool_id.attr,
3591 &dev_attr_name.attr,
3592 &dev_attr_image_id.attr,
3593 &dev_attr_current_snap.attr,
3594 &dev_attr_parent.attr,
3595 &dev_attr_refresh.attr,
3599 static struct attribute_group rbd_attr_group = {
3603 static const struct attribute_group *rbd_attr_groups[] = {
3608 static void rbd_sysfs_dev_release(struct device *dev)
3612 static struct device_type rbd_device_type = {
3614 .groups = rbd_attr_groups,
3615 .release = rbd_sysfs_dev_release,
3618 static struct rbd_spec *rbd_spec_get(struct rbd_spec *spec)
3620 kref_get(&spec->kref);
3625 static void rbd_spec_free(struct kref *kref);
3626 static void rbd_spec_put(struct rbd_spec *spec)
3629 kref_put(&spec->kref, rbd_spec_free);
3632 static struct rbd_spec *rbd_spec_alloc(void)
3634 struct rbd_spec *spec;
3636 spec = kzalloc(sizeof (*spec), GFP_KERNEL);
3639 kref_init(&spec->kref);
3644 static void rbd_spec_free(struct kref *kref)
3646 struct rbd_spec *spec = container_of(kref, struct rbd_spec, kref);
3648 kfree(spec->pool_name);
3649 kfree(spec->image_id);
3650 kfree(spec->image_name);
3651 kfree(spec->snap_name);
3655 static struct rbd_device *rbd_dev_create(struct rbd_client *rbdc,
3656 struct rbd_spec *spec)
3658 struct rbd_device *rbd_dev;
3660 rbd_dev = kzalloc(sizeof (*rbd_dev), GFP_KERNEL);
3664 spin_lock_init(&rbd_dev->lock);
3666 atomic_set(&rbd_dev->parent_ref, 0);
3667 INIT_LIST_HEAD(&rbd_dev->node);
3668 init_rwsem(&rbd_dev->header_rwsem);
3670 rbd_dev->spec = spec;
3671 rbd_dev->rbd_client = rbdc;
3673 /* Initialize the layout used for all rbd requests */
3675 rbd_dev->layout.fl_stripe_unit = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER);
3676 rbd_dev->layout.fl_stripe_count = cpu_to_le32(1);
3677 rbd_dev->layout.fl_object_size = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER);
3678 rbd_dev->layout.fl_pg_pool = cpu_to_le32((u32) spec->pool_id);
3683 static void rbd_dev_destroy(struct rbd_device *rbd_dev)
3685 rbd_put_client(rbd_dev->rbd_client);
3686 rbd_spec_put(rbd_dev->spec);
3691 * Get the size and object order for an image snapshot, or if
3692 * snap_id is CEPH_NOSNAP, gets this information for the base
3695 static int _rbd_dev_v2_snap_size(struct rbd_device *rbd_dev, u64 snap_id,
3696 u8 *order, u64 *snap_size)
3698 __le64 snapid = cpu_to_le64(snap_id);
3703 } __attribute__ ((packed)) size_buf = { 0 };
3705 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
3707 &snapid, sizeof (snapid),
3708 &size_buf, sizeof (size_buf));
3709 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
3712 if (ret < sizeof (size_buf))
3716 *order = size_buf.order;
3717 *snap_size = le64_to_cpu(size_buf.size);
3719 dout(" snap_id 0x%016llx order = %u, snap_size = %llu\n",
3720 (unsigned long long)snap_id, (unsigned int)*order,
3721 (unsigned long long)*snap_size);
3726 static int rbd_dev_v2_image_size(struct rbd_device *rbd_dev)
3728 return _rbd_dev_v2_snap_size(rbd_dev, CEPH_NOSNAP,
3729 &rbd_dev->header.obj_order,
3730 &rbd_dev->header.image_size);
3733 static int rbd_dev_v2_object_prefix(struct rbd_device *rbd_dev)
3739 reply_buf = kzalloc(RBD_OBJ_PREFIX_LEN_MAX, GFP_KERNEL);
3743 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
3744 "rbd", "get_object_prefix", NULL, 0,
3745 reply_buf, RBD_OBJ_PREFIX_LEN_MAX);
3746 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
3751 rbd_dev->header.object_prefix = ceph_extract_encoded_string(&p,
3752 p + ret, NULL, GFP_NOIO);
3755 if (IS_ERR(rbd_dev->header.object_prefix)) {
3756 ret = PTR_ERR(rbd_dev->header.object_prefix);
3757 rbd_dev->header.object_prefix = NULL;
3759 dout(" object_prefix = %s\n", rbd_dev->header.object_prefix);
3767 static int _rbd_dev_v2_snap_features(struct rbd_device *rbd_dev, u64 snap_id,
3770 __le64 snapid = cpu_to_le64(snap_id);
3774 } __attribute__ ((packed)) features_buf = { 0 };
3778 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
3779 "rbd", "get_features",
3780 &snapid, sizeof (snapid),
3781 &features_buf, sizeof (features_buf));
3782 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
3785 if (ret < sizeof (features_buf))
3788 incompat = le64_to_cpu(features_buf.incompat);
3789 if (incompat & ~RBD_FEATURES_SUPPORTED)
3792 *snap_features = le64_to_cpu(features_buf.features);
3794 dout(" snap_id 0x%016llx features = 0x%016llx incompat = 0x%016llx\n",
3795 (unsigned long long)snap_id,
3796 (unsigned long long)*snap_features,
3797 (unsigned long long)le64_to_cpu(features_buf.incompat));
3802 static int rbd_dev_v2_features(struct rbd_device *rbd_dev)
3804 return _rbd_dev_v2_snap_features(rbd_dev, CEPH_NOSNAP,
3805 &rbd_dev->header.features);
3808 static int rbd_dev_v2_parent_info(struct rbd_device *rbd_dev)
3810 struct rbd_spec *parent_spec;
3812 void *reply_buf = NULL;
3821 parent_spec = rbd_spec_alloc();
3825 size = sizeof (__le64) + /* pool_id */
3826 sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX + /* image_id */
3827 sizeof (__le64) + /* snap_id */
3828 sizeof (__le64); /* overlap */
3829 reply_buf = kmalloc(size, GFP_KERNEL);
3835 snapid = cpu_to_le64(CEPH_NOSNAP);
3836 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
3837 "rbd", "get_parent",
3838 &snapid, sizeof (snapid),
3840 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
3845 end = reply_buf + ret;
3847 ceph_decode_64_safe(&p, end, pool_id, out_err);
3848 if (pool_id == CEPH_NOPOOL) {
3850 * Either the parent never existed, or we have
3851 * record of it but the image got flattened so it no
3852 * longer has a parent. When the parent of a
3853 * layered image disappears we immediately set the
3854 * overlap to 0. The effect of this is that all new
3855 * requests will be treated as if the image had no
3858 if (rbd_dev->parent_overlap) {
3859 rbd_dev->parent_overlap = 0;
3861 rbd_dev_parent_put(rbd_dev);
3862 pr_info("%s: clone image has been flattened\n",
3863 rbd_dev->disk->disk_name);
3866 goto out; /* No parent? No problem. */
3869 /* The ceph file layout needs to fit pool id in 32 bits */
3872 if (pool_id > (u64)U32_MAX) {
3873 rbd_warn(NULL, "parent pool id too large (%llu > %u)\n",
3874 (unsigned long long)pool_id, U32_MAX);
3877 parent_spec->pool_id = pool_id;
3879 image_id = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL);
3880 if (IS_ERR(image_id)) {
3881 ret = PTR_ERR(image_id);
3884 parent_spec->image_id = image_id;
3885 ceph_decode_64_safe(&p, end, parent_spec->snap_id, out_err);
3886 ceph_decode_64_safe(&p, end, overlap, out_err);
3889 rbd_spec_put(rbd_dev->parent_spec);
3890 rbd_dev->parent_spec = parent_spec;
3891 parent_spec = NULL; /* rbd_dev now owns this */
3892 rbd_dev->parent_overlap = overlap;
3894 rbd_warn(rbd_dev, "ignoring parent of clone with overlap 0\n");
3900 rbd_spec_put(parent_spec);
3905 static int rbd_dev_v2_striping_info(struct rbd_device *rbd_dev)
3909 __le64 stripe_count;
3910 } __attribute__ ((packed)) striping_info_buf = { 0 };
3911 size_t size = sizeof (striping_info_buf);
3918 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
3919 "rbd", "get_stripe_unit_count", NULL, 0,
3920 (char *)&striping_info_buf, size);
3921 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
3928 * We don't actually support the "fancy striping" feature
3929 * (STRIPINGV2) yet, but if the striping sizes are the
3930 * defaults the behavior is the same as before. So find
3931 * out, and only fail if the image has non-default values.
3934 obj_size = (u64)1 << rbd_dev->header.obj_order;
3935 p = &striping_info_buf;
3936 stripe_unit = ceph_decode_64(&p);
3937 if (stripe_unit != obj_size) {
3938 rbd_warn(rbd_dev, "unsupported stripe unit "
3939 "(got %llu want %llu)",
3940 stripe_unit, obj_size);
3943 stripe_count = ceph_decode_64(&p);
3944 if (stripe_count != 1) {
3945 rbd_warn(rbd_dev, "unsupported stripe count "
3946 "(got %llu want 1)", stripe_count);
3949 rbd_dev->header.stripe_unit = stripe_unit;
3950 rbd_dev->header.stripe_count = stripe_count;
3955 static char *rbd_dev_image_name(struct rbd_device *rbd_dev)
3957 size_t image_id_size;
3962 void *reply_buf = NULL;
3964 char *image_name = NULL;
3967 rbd_assert(!rbd_dev->spec->image_name);
3969 len = strlen(rbd_dev->spec->image_id);
3970 image_id_size = sizeof (__le32) + len;
3971 image_id = kmalloc(image_id_size, GFP_KERNEL);
3976 end = image_id + image_id_size;
3977 ceph_encode_string(&p, end, rbd_dev->spec->image_id, (u32)len);
3979 size = sizeof (__le32) + RBD_IMAGE_NAME_LEN_MAX;
3980 reply_buf = kmalloc(size, GFP_KERNEL);
3984 ret = rbd_obj_method_sync(rbd_dev, RBD_DIRECTORY,
3985 "rbd", "dir_get_name",
3986 image_id, image_id_size,
3991 end = reply_buf + ret;
3993 image_name = ceph_extract_encoded_string(&p, end, &len, GFP_KERNEL);
3994 if (IS_ERR(image_name))
3997 dout("%s: name is %s len is %zd\n", __func__, image_name, len);
4005 static u64 rbd_v1_snap_id_by_name(struct rbd_device *rbd_dev, const char *name)
4007 struct ceph_snap_context *snapc = rbd_dev->header.snapc;
4008 const char *snap_name;
4011 /* Skip over names until we find the one we are looking for */
4013 snap_name = rbd_dev->header.snap_names;
4014 while (which < snapc->num_snaps) {
4015 if (!strcmp(name, snap_name))
4016 return snapc->snaps[which];
4017 snap_name += strlen(snap_name) + 1;
4023 static u64 rbd_v2_snap_id_by_name(struct rbd_device *rbd_dev, const char *name)
4025 struct ceph_snap_context *snapc = rbd_dev->header.snapc;
4030 for (which = 0; !found && which < snapc->num_snaps; which++) {
4031 const char *snap_name;
4033 snap_id = snapc->snaps[which];
4034 snap_name = rbd_dev_v2_snap_name(rbd_dev, snap_id);
4035 if (IS_ERR(snap_name))
4037 found = !strcmp(name, snap_name);
4040 return found ? snap_id : CEPH_NOSNAP;
4044 * Assumes name is never RBD_SNAP_HEAD_NAME; returns CEPH_NOSNAP if
4045 * no snapshot by that name is found, or if an error occurs.
4047 static u64 rbd_snap_id_by_name(struct rbd_device *rbd_dev, const char *name)
4049 if (rbd_dev->image_format == 1)
4050 return rbd_v1_snap_id_by_name(rbd_dev, name);
4052 return rbd_v2_snap_id_by_name(rbd_dev, name);
4056 * When an rbd image has a parent image, it is identified by the
4057 * pool, image, and snapshot ids (not names). This function fills
4058 * in the names for those ids. (It's OK if we can't figure out the
4059 * name for an image id, but the pool and snapshot ids should always
4060 * exist and have names.) All names in an rbd spec are dynamically
4063 * When an image being mapped (not a parent) is probed, we have the
4064 * pool name and pool id, image name and image id, and the snapshot
4065 * name. The only thing we're missing is the snapshot id.
4067 static int rbd_dev_spec_update(struct rbd_device *rbd_dev)
4069 struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
4070 struct rbd_spec *spec = rbd_dev->spec;
4071 const char *pool_name;
4072 const char *image_name;
4073 const char *snap_name;
4077 * An image being mapped will have the pool name (etc.), but
4078 * we need to look up the snapshot id.
4080 if (spec->pool_name) {
4081 if (strcmp(spec->snap_name, RBD_SNAP_HEAD_NAME)) {
4084 snap_id = rbd_snap_id_by_name(rbd_dev, spec->snap_name);
4085 if (snap_id == CEPH_NOSNAP)
4087 spec->snap_id = snap_id;
4089 spec->snap_id = CEPH_NOSNAP;
4095 /* Get the pool name; we have to make our own copy of this */
4097 pool_name = ceph_pg_pool_name_by_id(osdc->osdmap, spec->pool_id);
4099 rbd_warn(rbd_dev, "no pool with id %llu", spec->pool_id);
4102 pool_name = kstrdup(pool_name, GFP_KERNEL);
4106 /* Fetch the image name; tolerate failure here */
4108 image_name = rbd_dev_image_name(rbd_dev);
4110 rbd_warn(rbd_dev, "unable to get image name");
4112 /* Look up the snapshot name, and make a copy */
4114 snap_name = rbd_snap_name(rbd_dev, spec->snap_id);
4120 spec->pool_name = pool_name;
4121 spec->image_name = image_name;
4122 spec->snap_name = snap_name;
4132 static int rbd_dev_v2_snap_context(struct rbd_device *rbd_dev)
4141 struct ceph_snap_context *snapc;
4145 * We'll need room for the seq value (maximum snapshot id),
4146 * snapshot count, and array of that many snapshot ids.
4147 * For now we have a fixed upper limit on the number we're
4148 * prepared to receive.
4150 size = sizeof (__le64) + sizeof (__le32) +
4151 RBD_MAX_SNAP_COUNT * sizeof (__le64);
4152 reply_buf = kzalloc(size, GFP_KERNEL);
4156 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
4157 "rbd", "get_snapcontext", NULL, 0,
4159 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
4164 end = reply_buf + ret;
4166 ceph_decode_64_safe(&p, end, seq, out);
4167 ceph_decode_32_safe(&p, end, snap_count, out);
4170 * Make sure the reported number of snapshot ids wouldn't go
4171 * beyond the end of our buffer. But before checking that,
4172 * make sure the computed size of the snapshot context we
4173 * allocate is representable in a size_t.
4175 if (snap_count > (SIZE_MAX - sizeof (struct ceph_snap_context))
4180 if (!ceph_has_room(&p, end, snap_count * sizeof (__le64)))
4184 snapc = ceph_create_snap_context(snap_count, GFP_KERNEL);
4190 for (i = 0; i < snap_count; i++)
4191 snapc->snaps[i] = ceph_decode_64(&p);
4193 ceph_put_snap_context(rbd_dev->header.snapc);
4194 rbd_dev->header.snapc = snapc;
4196 dout(" snap context seq = %llu, snap_count = %u\n",
4197 (unsigned long long)seq, (unsigned int)snap_count);
4204 static const char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev,
4215 size = sizeof (__le32) + RBD_MAX_SNAP_NAME_LEN;
4216 reply_buf = kmalloc(size, GFP_KERNEL);
4218 return ERR_PTR(-ENOMEM);
4220 snapid = cpu_to_le64(snap_id);
4221 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
4222 "rbd", "get_snapshot_name",
4223 &snapid, sizeof (snapid),
4225 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
4227 snap_name = ERR_PTR(ret);
4232 end = reply_buf + ret;
4233 snap_name = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL);
4234 if (IS_ERR(snap_name))
4237 dout(" snap_id 0x%016llx snap_name = %s\n",
4238 (unsigned long long)snap_id, snap_name);
4245 static int rbd_dev_v2_header_info(struct rbd_device *rbd_dev)
4247 bool first_time = rbd_dev->header.object_prefix == NULL;
4250 down_write(&rbd_dev->header_rwsem);
4252 ret = rbd_dev_v2_image_size(rbd_dev);
4257 ret = rbd_dev_v2_header_onetime(rbd_dev);
4263 * If the image supports layering, get the parent info. We
4264 * need to probe the first time regardless. Thereafter we
4265 * only need to if there's a parent, to see if it has
4266 * disappeared due to the mapped image getting flattened.
4268 if (rbd_dev->header.features & RBD_FEATURE_LAYERING &&
4269 (first_time || rbd_dev->parent_spec)) {
4272 ret = rbd_dev_v2_parent_info(rbd_dev);
4277 * Print a warning if this is the initial probe and
4278 * the image has a parent. Don't print it if the
4279 * image now being probed is itself a parent. We
4280 * can tell at this point because we won't know its
4281 * pool name yet (just its pool id).
4283 warn = rbd_dev->parent_spec && rbd_dev->spec->pool_name;
4284 if (first_time && warn)
4285 rbd_warn(rbd_dev, "WARNING: kernel layering "
4286 "is EXPERIMENTAL!");
4289 if (rbd_dev->spec->snap_id == CEPH_NOSNAP)
4290 if (rbd_dev->mapping.size != rbd_dev->header.image_size)
4291 rbd_dev->mapping.size = rbd_dev->header.image_size;
4293 ret = rbd_dev_v2_snap_context(rbd_dev);
4294 dout("rbd_dev_v2_snap_context returned %d\n", ret);
4296 up_write(&rbd_dev->header_rwsem);
4301 static int rbd_bus_add_dev(struct rbd_device *rbd_dev)
4306 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
4308 dev = &rbd_dev->dev;
4309 dev->bus = &rbd_bus_type;
4310 dev->type = &rbd_device_type;
4311 dev->parent = &rbd_root_dev;
4312 dev->release = rbd_dev_device_release;
4313 dev_set_name(dev, "%d", rbd_dev->dev_id);
4314 ret = device_register(dev);
4316 mutex_unlock(&ctl_mutex);
4321 static void rbd_bus_del_dev(struct rbd_device *rbd_dev)
4323 device_unregister(&rbd_dev->dev);
4326 static atomic64_t rbd_dev_id_max = ATOMIC64_INIT(0);
4329 * Get a unique rbd identifier for the given new rbd_dev, and add
4330 * the rbd_dev to the global list. The minimum rbd id is 1.
4332 static void rbd_dev_id_get(struct rbd_device *rbd_dev)
4334 rbd_dev->dev_id = atomic64_inc_return(&rbd_dev_id_max);
4336 spin_lock(&rbd_dev_list_lock);
4337 list_add_tail(&rbd_dev->node, &rbd_dev_list);
4338 spin_unlock(&rbd_dev_list_lock);
4339 dout("rbd_dev %p given dev id %llu\n", rbd_dev,
4340 (unsigned long long) rbd_dev->dev_id);
4344 * Remove an rbd_dev from the global list, and record that its
4345 * identifier is no longer in use.
4347 static void rbd_dev_id_put(struct rbd_device *rbd_dev)
4349 struct list_head *tmp;
4350 int rbd_id = rbd_dev->dev_id;
4353 rbd_assert(rbd_id > 0);
4355 dout("rbd_dev %p released dev id %llu\n", rbd_dev,
4356 (unsigned long long) rbd_dev->dev_id);
4357 spin_lock(&rbd_dev_list_lock);
4358 list_del_init(&rbd_dev->node);
4361 * If the id being "put" is not the current maximum, there
4362 * is nothing special we need to do.
4364 if (rbd_id != atomic64_read(&rbd_dev_id_max)) {
4365 spin_unlock(&rbd_dev_list_lock);
4370 * We need to update the current maximum id. Search the
4371 * list to find out what it is. We're more likely to find
4372 * the maximum at the end, so search the list backward.
4375 list_for_each_prev(tmp, &rbd_dev_list) {
4376 struct rbd_device *rbd_dev;
4378 rbd_dev = list_entry(tmp, struct rbd_device, node);
4379 if (rbd_dev->dev_id > max_id)
4380 max_id = rbd_dev->dev_id;
4382 spin_unlock(&rbd_dev_list_lock);
4385 * The max id could have been updated by rbd_dev_id_get(), in
4386 * which case it now accurately reflects the new maximum.
4387 * Be careful not to overwrite the maximum value in that
4390 atomic64_cmpxchg(&rbd_dev_id_max, rbd_id, max_id);
4391 dout(" max dev id has been reset\n");
4395 * Skips over white space at *buf, and updates *buf to point to the
4396 * first found non-space character (if any). Returns the length of
4397 * the token (string of non-white space characters) found. Note
4398 * that *buf must be terminated with '\0'.
4400 static inline size_t next_token(const char **buf)
4403 * These are the characters that produce nonzero for
4404 * isspace() in the "C" and "POSIX" locales.
4406 const char *spaces = " \f\n\r\t\v";
4408 *buf += strspn(*buf, spaces); /* Find start of token */
4410 return strcspn(*buf, spaces); /* Return token length */
4414 * Finds the next token in *buf, and if the provided token buffer is
4415 * big enough, copies the found token into it. The result, if
4416 * copied, is guaranteed to be terminated with '\0'. Note that *buf
4417 * must be terminated with '\0' on entry.
4419 * Returns the length of the token found (not including the '\0').
4420 * Return value will be 0 if no token is found, and it will be >=
4421 * token_size if the token would not fit.
4423 * The *buf pointer will be updated to point beyond the end of the
4424 * found token. Note that this occurs even if the token buffer is
4425 * too small to hold it.
4427 static inline size_t copy_token(const char **buf,
4433 len = next_token(buf);
4434 if (len < token_size) {
4435 memcpy(token, *buf, len);
4436 *(token + len) = '\0';
4444 * Finds the next token in *buf, dynamically allocates a buffer big
4445 * enough to hold a copy of it, and copies the token into the new
4446 * buffer. The copy is guaranteed to be terminated with '\0'. Note
4447 * that a duplicate buffer is created even for a zero-length token.
4449 * Returns a pointer to the newly-allocated duplicate, or a null
4450 * pointer if memory for the duplicate was not available. If
4451 * the lenp argument is a non-null pointer, the length of the token
4452 * (not including the '\0') is returned in *lenp.
4454 * If successful, the *buf pointer will be updated to point beyond
4455 * the end of the found token.
4457 * Note: uses GFP_KERNEL for allocation.
4459 static inline char *dup_token(const char **buf, size_t *lenp)
4464 len = next_token(buf);
4465 dup = kmemdup(*buf, len + 1, GFP_KERNEL);
4468 *(dup + len) = '\0';
4478 * Parse the options provided for an "rbd add" (i.e., rbd image
4479 * mapping) request. These arrive via a write to /sys/bus/rbd/add,
4480 * and the data written is passed here via a NUL-terminated buffer.
4481 * Returns 0 if successful or an error code otherwise.
4483 * The information extracted from these options is recorded in
4484 * the other parameters which return dynamically-allocated
4487 * The address of a pointer that will refer to a ceph options
4488 * structure. Caller must release the returned pointer using
4489 * ceph_destroy_options() when it is no longer needed.
4491 * Address of an rbd options pointer. Fully initialized by
4492 * this function; caller must release with kfree().
4494 * Address of an rbd image specification pointer. Fully
4495 * initialized by this function based on parsed options.
4496 * Caller must release with rbd_spec_put().
4498 * The options passed take this form:
4499 * <mon_addrs> <options> <pool_name> <image_name> [<snap_id>]
4502 * A comma-separated list of one or more monitor addresses.
4503 * A monitor address is an ip address, optionally followed
4504 * by a port number (separated by a colon).
4505 * I.e.: ip1[:port1][,ip2[:port2]...]
4507 * A comma-separated list of ceph and/or rbd options.
4509 * The name of the rados pool containing the rbd image.
4511 * The name of the image in that pool to map.
4513 * An optional snapshot id. If provided, the mapping will
4514 * present data from the image at the time that snapshot was
4515 * created. The image head is used if no snapshot id is
4516 * provided. Snapshot mappings are always read-only.
4518 static int rbd_add_parse_args(const char *buf,
4519 struct ceph_options **ceph_opts,
4520 struct rbd_options **opts,
4521 struct rbd_spec **rbd_spec)
4525 const char *mon_addrs;
4527 size_t mon_addrs_size;
4528 struct rbd_spec *spec = NULL;
4529 struct rbd_options *rbd_opts = NULL;
4530 struct ceph_options *copts;
4533 /* The first four tokens are required */
4535 len = next_token(&buf);
4537 rbd_warn(NULL, "no monitor address(es) provided");
4541 mon_addrs_size = len + 1;
4545 options = dup_token(&buf, NULL);
4549 rbd_warn(NULL, "no options provided");
4553 spec = rbd_spec_alloc();
4557 spec->pool_name = dup_token(&buf, NULL);
4558 if (!spec->pool_name)
4560 if (!*spec->pool_name) {
4561 rbd_warn(NULL, "no pool name provided");
4565 spec->image_name = dup_token(&buf, NULL);
4566 if (!spec->image_name)
4568 if (!*spec->image_name) {
4569 rbd_warn(NULL, "no image name provided");
4574 * Snapshot name is optional; default is to use "-"
4575 * (indicating the head/no snapshot).
4577 len = next_token(&buf);
4579 buf = RBD_SNAP_HEAD_NAME; /* No snapshot supplied */
4580 len = sizeof (RBD_SNAP_HEAD_NAME) - 1;
4581 } else if (len > RBD_MAX_SNAP_NAME_LEN) {
4582 ret = -ENAMETOOLONG;
4585 snap_name = kmemdup(buf, len + 1, GFP_KERNEL);
4588 *(snap_name + len) = '\0';
4589 spec->snap_name = snap_name;
4591 /* Initialize all rbd options to the defaults */
4593 rbd_opts = kzalloc(sizeof (*rbd_opts), GFP_KERNEL);
4597 rbd_opts->read_only = RBD_READ_ONLY_DEFAULT;
4599 copts = ceph_parse_options(options, mon_addrs,
4600 mon_addrs + mon_addrs_size - 1,
4601 parse_rbd_opts_token, rbd_opts);
4602 if (IS_ERR(copts)) {
4603 ret = PTR_ERR(copts);
4624 * An rbd format 2 image has a unique identifier, distinct from the
4625 * name given to it by the user. Internally, that identifier is
4626 * what's used to specify the names of objects related to the image.
4628 * A special "rbd id" object is used to map an rbd image name to its
4629 * id. If that object doesn't exist, then there is no v2 rbd image
4630 * with the supplied name.
4632 * This function will record the given rbd_dev's image_id field if
4633 * it can be determined, and in that case will return 0. If any
4634 * errors occur a negative errno will be returned and the rbd_dev's
4635 * image_id field will be unchanged (and should be NULL).
4637 static int rbd_dev_image_id(struct rbd_device *rbd_dev)
4646 * When probing a parent image, the image id is already
4647 * known (and the image name likely is not). There's no
4648 * need to fetch the image id again in this case. We
4649 * do still need to set the image format though.
4651 if (rbd_dev->spec->image_id) {
4652 rbd_dev->image_format = *rbd_dev->spec->image_id ? 2 : 1;
4658 * First, see if the format 2 image id file exists, and if
4659 * so, get the image's persistent id from it.
4661 size = sizeof (RBD_ID_PREFIX) + strlen(rbd_dev->spec->image_name);
4662 object_name = kmalloc(size, GFP_NOIO);
4665 sprintf(object_name, "%s%s", RBD_ID_PREFIX, rbd_dev->spec->image_name);
4666 dout("rbd id object name is %s\n", object_name);
4668 /* Response will be an encoded string, which includes a length */
4670 size = sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX;
4671 response = kzalloc(size, GFP_NOIO);
4677 /* If it doesn't exist we'll assume it's a format 1 image */
4679 ret = rbd_obj_method_sync(rbd_dev, object_name,
4680 "rbd", "get_id", NULL, 0,
4681 response, RBD_IMAGE_ID_LEN_MAX);
4682 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
4683 if (ret == -ENOENT) {
4684 image_id = kstrdup("", GFP_KERNEL);
4685 ret = image_id ? 0 : -ENOMEM;
4687 rbd_dev->image_format = 1;
4688 } else if (ret > sizeof (__le32)) {
4691 image_id = ceph_extract_encoded_string(&p, p + ret,
4693 ret = IS_ERR(image_id) ? PTR_ERR(image_id) : 0;
4695 rbd_dev->image_format = 2;
4701 rbd_dev->spec->image_id = image_id;
4702 dout("image_id is %s\n", image_id);
4712 * Undo whatever state changes are made by v1 or v2 header info
4715 static void rbd_dev_unprobe(struct rbd_device *rbd_dev)
4717 struct rbd_image_header *header;
4719 /* Drop parent reference unless it's already been done (or none) */
4721 if (rbd_dev->parent_overlap)
4722 rbd_dev_parent_put(rbd_dev);
4724 /* Free dynamic fields from the header, then zero it out */
4726 header = &rbd_dev->header;
4727 ceph_put_snap_context(header->snapc);
4728 kfree(header->snap_sizes);
4729 kfree(header->snap_names);
4730 kfree(header->object_prefix);
4731 memset(header, 0, sizeof (*header));
4734 static int rbd_dev_v2_header_onetime(struct rbd_device *rbd_dev)
4738 ret = rbd_dev_v2_object_prefix(rbd_dev);
4743 * Get the and check features for the image. Currently the
4744 * features are assumed to never change.
4746 ret = rbd_dev_v2_features(rbd_dev);
4750 /* If the image supports fancy striping, get its parameters */
4752 if (rbd_dev->header.features & RBD_FEATURE_STRIPINGV2) {
4753 ret = rbd_dev_v2_striping_info(rbd_dev);
4757 /* No support for crypto and compression type format 2 images */
4761 rbd_dev->header.features = 0;
4762 kfree(rbd_dev->header.object_prefix);
4763 rbd_dev->header.object_prefix = NULL;
4768 static int rbd_dev_probe_parent(struct rbd_device *rbd_dev)
4770 struct rbd_device *parent = NULL;
4771 struct rbd_spec *parent_spec;
4772 struct rbd_client *rbdc;
4775 if (!rbd_dev->parent_spec)
4778 * We need to pass a reference to the client and the parent
4779 * spec when creating the parent rbd_dev. Images related by
4780 * parent/child relationships always share both.
4782 parent_spec = rbd_spec_get(rbd_dev->parent_spec);
4783 rbdc = __rbd_get_client(rbd_dev->rbd_client);
4786 parent = rbd_dev_create(rbdc, parent_spec);
4790 ret = rbd_dev_image_probe(parent, false);
4793 rbd_dev->parent = parent;
4794 atomic_set(&rbd_dev->parent_ref, 1);
4799 rbd_dev_unparent(rbd_dev);
4800 kfree(rbd_dev->header_name);
4801 rbd_dev_destroy(parent);
4803 rbd_put_client(rbdc);
4804 rbd_spec_put(parent_spec);
4810 static int rbd_dev_device_setup(struct rbd_device *rbd_dev)
4814 /* generate unique id: find highest unique id, add one */
4815 rbd_dev_id_get(rbd_dev);
4817 /* Fill in the device name, now that we have its id. */
4818 BUILD_BUG_ON(DEV_NAME_LEN
4819 < sizeof (RBD_DRV_NAME) + MAX_INT_FORMAT_WIDTH);
4820 sprintf(rbd_dev->name, "%s%d", RBD_DRV_NAME, rbd_dev->dev_id);
4822 /* Get our block major device number. */
4824 ret = register_blkdev(0, rbd_dev->name);
4827 rbd_dev->major = ret;
4829 /* Set up the blkdev mapping. */
4831 ret = rbd_init_disk(rbd_dev);
4833 goto err_out_blkdev;
4835 ret = rbd_dev_mapping_set(rbd_dev);
4838 set_capacity(rbd_dev->disk, rbd_dev->mapping.size / SECTOR_SIZE);
4840 ret = rbd_bus_add_dev(rbd_dev);
4842 goto err_out_mapping;
4844 /* Everything's ready. Announce the disk to the world. */
4846 set_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags);
4847 add_disk(rbd_dev->disk);
4849 pr_info("%s: added with size 0x%llx\n", rbd_dev->disk->disk_name,
4850 (unsigned long long) rbd_dev->mapping.size);
4855 rbd_dev_mapping_clear(rbd_dev);
4857 rbd_free_disk(rbd_dev);
4859 unregister_blkdev(rbd_dev->major, rbd_dev->name);
4861 rbd_dev_id_put(rbd_dev);
4862 rbd_dev_mapping_clear(rbd_dev);
4867 static int rbd_dev_header_name(struct rbd_device *rbd_dev)
4869 struct rbd_spec *spec = rbd_dev->spec;
4872 /* Record the header object name for this rbd image. */
4874 rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
4876 if (rbd_dev->image_format == 1)
4877 size = strlen(spec->image_name) + sizeof (RBD_SUFFIX);
4879 size = sizeof (RBD_HEADER_PREFIX) + strlen(spec->image_id);
4881 rbd_dev->header_name = kmalloc(size, GFP_KERNEL);
4882 if (!rbd_dev->header_name)
4885 if (rbd_dev->image_format == 1)
4886 sprintf(rbd_dev->header_name, "%s%s",
4887 spec->image_name, RBD_SUFFIX);
4889 sprintf(rbd_dev->header_name, "%s%s",
4890 RBD_HEADER_PREFIX, spec->image_id);
4894 static void rbd_dev_image_release(struct rbd_device *rbd_dev)
4896 rbd_dev_unprobe(rbd_dev);
4897 kfree(rbd_dev->header_name);
4898 rbd_dev->header_name = NULL;
4899 rbd_dev->image_format = 0;
4900 kfree(rbd_dev->spec->image_id);
4901 rbd_dev->spec->image_id = NULL;
4903 rbd_dev_destroy(rbd_dev);
4907 * Probe for the existence of the header object for the given rbd
4908 * device. If this image is the one being mapped (i.e., not a
4909 * parent), initiate a watch on its header object before using that
4910 * object to get detailed information about the rbd image.
4912 static int rbd_dev_image_probe(struct rbd_device *rbd_dev, bool mapping)
4918 * Get the id from the image id object. Unless there's an
4919 * error, rbd_dev->spec->image_id will be filled in with
4920 * a dynamically-allocated string, and rbd_dev->image_format
4921 * will be set to either 1 or 2.
4923 ret = rbd_dev_image_id(rbd_dev);
4926 rbd_assert(rbd_dev->spec->image_id);
4927 rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
4929 ret = rbd_dev_header_name(rbd_dev);
4931 goto err_out_format;
4934 ret = rbd_dev_header_watch_sync(rbd_dev, true);
4936 goto out_header_name;
4939 if (rbd_dev->image_format == 1)
4940 ret = rbd_dev_v1_header_info(rbd_dev);
4942 ret = rbd_dev_v2_header_info(rbd_dev);
4946 ret = rbd_dev_spec_update(rbd_dev);
4950 ret = rbd_dev_probe_parent(rbd_dev);
4954 dout("discovered format %u image, header name is %s\n",
4955 rbd_dev->image_format, rbd_dev->header_name);
4959 rbd_dev_unprobe(rbd_dev);
4962 tmp = rbd_dev_header_watch_sync(rbd_dev, false);
4964 rbd_warn(rbd_dev, "unable to tear down "
4965 "watch request (%d)\n", tmp);
4968 kfree(rbd_dev->header_name);
4969 rbd_dev->header_name = NULL;
4971 rbd_dev->image_format = 0;
4972 kfree(rbd_dev->spec->image_id);
4973 rbd_dev->spec->image_id = NULL;
4975 dout("probe failed, returning %d\n", ret);
4980 static ssize_t rbd_add(struct bus_type *bus,
4984 struct rbd_device *rbd_dev = NULL;
4985 struct ceph_options *ceph_opts = NULL;
4986 struct rbd_options *rbd_opts = NULL;
4987 struct rbd_spec *spec = NULL;
4988 struct rbd_client *rbdc;
4989 struct ceph_osd_client *osdc;
4993 if (!try_module_get(THIS_MODULE))
4996 /* parse add command */
4997 rc = rbd_add_parse_args(buf, &ceph_opts, &rbd_opts, &spec);
4999 goto err_out_module;
5000 read_only = rbd_opts->read_only;
5002 rbd_opts = NULL; /* done with this */
5004 rbdc = rbd_get_client(ceph_opts);
5011 osdc = &rbdc->client->osdc;
5012 rc = ceph_pg_poolid_by_name(osdc->osdmap, spec->pool_name);
5014 goto err_out_client;
5015 spec->pool_id = (u64)rc;
5017 /* The ceph file layout needs to fit pool id in 32 bits */
5019 if (spec->pool_id > (u64)U32_MAX) {
5020 rbd_warn(NULL, "pool id too large (%llu > %u)\n",
5021 (unsigned long long)spec->pool_id, U32_MAX);
5023 goto err_out_client;
5026 rbd_dev = rbd_dev_create(rbdc, spec);
5028 goto err_out_client;
5029 rbdc = NULL; /* rbd_dev now owns this */
5030 spec = NULL; /* rbd_dev now owns this */
5032 rc = rbd_dev_image_probe(rbd_dev, true);
5034 goto err_out_rbd_dev;
5036 /* If we are mapping a snapshot it must be marked read-only */
5038 if (rbd_dev->spec->snap_id != CEPH_NOSNAP)
5040 rbd_dev->mapping.read_only = read_only;
5042 rc = rbd_dev_device_setup(rbd_dev);
5044 rbd_dev_image_release(rbd_dev);
5045 goto err_out_module;
5051 rbd_dev_destroy(rbd_dev);
5053 rbd_put_client(rbdc);
5057 module_put(THIS_MODULE);
5059 dout("Error adding device %s\n", buf);
5064 static struct rbd_device *__rbd_get_dev(unsigned long dev_id)
5066 struct list_head *tmp;
5067 struct rbd_device *rbd_dev;
5069 spin_lock(&rbd_dev_list_lock);
5070 list_for_each(tmp, &rbd_dev_list) {
5071 rbd_dev = list_entry(tmp, struct rbd_device, node);
5072 if (rbd_dev->dev_id == dev_id) {
5073 spin_unlock(&rbd_dev_list_lock);
5077 spin_unlock(&rbd_dev_list_lock);
5081 static void rbd_dev_device_release(struct device *dev)
5083 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
5085 rbd_free_disk(rbd_dev);
5086 clear_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags);
5087 rbd_dev_mapping_clear(rbd_dev);
5088 unregister_blkdev(rbd_dev->major, rbd_dev->name);
5090 rbd_dev_id_put(rbd_dev);
5091 rbd_dev_mapping_clear(rbd_dev);
5094 static void rbd_dev_remove_parent(struct rbd_device *rbd_dev)
5096 while (rbd_dev->parent) {
5097 struct rbd_device *first = rbd_dev;
5098 struct rbd_device *second = first->parent;
5099 struct rbd_device *third;
5102 * Follow to the parent with no grandparent and
5105 while (second && (third = second->parent)) {
5110 rbd_dev_image_release(second);
5111 first->parent = NULL;
5112 first->parent_overlap = 0;
5114 rbd_assert(first->parent_spec);
5115 rbd_spec_put(first->parent_spec);
5116 first->parent_spec = NULL;
5120 static ssize_t rbd_remove(struct bus_type *bus,
5124 struct rbd_device *rbd_dev = NULL;
5129 ret = strict_strtoul(buf, 10, &ul);
5133 /* convert to int; abort if we lost anything in the conversion */
5134 target_id = (int) ul;
5135 if (target_id != ul)
5138 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
5140 rbd_dev = __rbd_get_dev(target_id);
5146 spin_lock_irq(&rbd_dev->lock);
5147 if (rbd_dev->open_count)
5150 set_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags);
5151 spin_unlock_irq(&rbd_dev->lock);
5154 rbd_bus_del_dev(rbd_dev);
5155 ret = rbd_dev_header_watch_sync(rbd_dev, false);
5157 rbd_warn(rbd_dev, "failed to cancel watch event (%d)\n", ret);
5158 rbd_dev_image_release(rbd_dev);
5159 module_put(THIS_MODULE);
5162 mutex_unlock(&ctl_mutex);
5168 * create control files in sysfs
5171 static int rbd_sysfs_init(void)
5175 ret = device_register(&rbd_root_dev);
5179 ret = bus_register(&rbd_bus_type);
5181 device_unregister(&rbd_root_dev);
5186 static void rbd_sysfs_cleanup(void)
5188 bus_unregister(&rbd_bus_type);
5189 device_unregister(&rbd_root_dev);
5192 static int rbd_slab_init(void)
5194 rbd_assert(!rbd_img_request_cache);
5195 rbd_img_request_cache = kmem_cache_create("rbd_img_request",
5196 sizeof (struct rbd_img_request),
5197 __alignof__(struct rbd_img_request),
5199 if (!rbd_img_request_cache)
5202 rbd_assert(!rbd_obj_request_cache);
5203 rbd_obj_request_cache = kmem_cache_create("rbd_obj_request",
5204 sizeof (struct rbd_obj_request),
5205 __alignof__(struct rbd_obj_request),
5207 if (!rbd_obj_request_cache)
5210 rbd_assert(!rbd_segment_name_cache);
5211 rbd_segment_name_cache = kmem_cache_create("rbd_segment_name",
5212 MAX_OBJ_NAME_SIZE + 1, 1, 0, NULL);
5213 if (rbd_segment_name_cache)
5216 if (rbd_obj_request_cache) {
5217 kmem_cache_destroy(rbd_obj_request_cache);
5218 rbd_obj_request_cache = NULL;
5221 kmem_cache_destroy(rbd_img_request_cache);
5222 rbd_img_request_cache = NULL;
5227 static void rbd_slab_exit(void)
5229 rbd_assert(rbd_segment_name_cache);
5230 kmem_cache_destroy(rbd_segment_name_cache);
5231 rbd_segment_name_cache = NULL;
5233 rbd_assert(rbd_obj_request_cache);
5234 kmem_cache_destroy(rbd_obj_request_cache);
5235 rbd_obj_request_cache = NULL;
5237 rbd_assert(rbd_img_request_cache);
5238 kmem_cache_destroy(rbd_img_request_cache);
5239 rbd_img_request_cache = NULL;
5242 static int __init rbd_init(void)
5246 if (!libceph_compatible(NULL)) {
5247 rbd_warn(NULL, "libceph incompatibility (quitting)");
5251 rc = rbd_slab_init();
5254 rc = rbd_sysfs_init();
5258 pr_info("loaded " RBD_DRV_NAME_LONG "\n");
5263 static void __exit rbd_exit(void)
5265 rbd_sysfs_cleanup();
5269 module_init(rbd_init);
5270 module_exit(rbd_exit);
5272 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
5273 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
5274 MODULE_DESCRIPTION("rados block device");
5276 /* following authorship retained from original osdblk.c */
5277 MODULE_AUTHOR("Jeff Garzik <jeff@garzik.org>");
5279 MODULE_LICENSE("GPL");