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 (void) get_device(&rbd_dev->dev);
493 set_device_ro(bdev, rbd_dev->mapping.read_only);
498 static void rbd_release(struct gendisk *disk, fmode_t mode)
500 struct rbd_device *rbd_dev = disk->private_data;
501 unsigned long open_count_before;
503 spin_lock_irq(&rbd_dev->lock);
504 open_count_before = rbd_dev->open_count--;
505 spin_unlock_irq(&rbd_dev->lock);
506 rbd_assert(open_count_before > 0);
508 put_device(&rbd_dev->dev);
511 static const struct block_device_operations rbd_bd_ops = {
512 .owner = THIS_MODULE,
514 .release = rbd_release,
518 * Initialize an rbd client instance. Success or not, this function
519 * consumes ceph_opts. Caller holds ctl_mutex.
521 static struct rbd_client *rbd_client_create(struct ceph_options *ceph_opts)
523 struct rbd_client *rbdc;
526 dout("%s:\n", __func__);
527 rbdc = kmalloc(sizeof(struct rbd_client), GFP_KERNEL);
531 kref_init(&rbdc->kref);
532 INIT_LIST_HEAD(&rbdc->node);
534 rbdc->client = ceph_create_client(ceph_opts, rbdc, 0, 0);
535 if (IS_ERR(rbdc->client))
537 ceph_opts = NULL; /* Now rbdc->client is responsible for ceph_opts */
539 ret = ceph_open_session(rbdc->client);
543 spin_lock(&rbd_client_list_lock);
544 list_add_tail(&rbdc->node, &rbd_client_list);
545 spin_unlock(&rbd_client_list_lock);
547 dout("%s: rbdc %p\n", __func__, rbdc);
551 ceph_destroy_client(rbdc->client);
556 ceph_destroy_options(ceph_opts);
557 dout("%s: error %d\n", __func__, ret);
562 static struct rbd_client *__rbd_get_client(struct rbd_client *rbdc)
564 kref_get(&rbdc->kref);
570 * Find a ceph client with specific addr and configuration. If
571 * found, bump its reference count.
573 static struct rbd_client *rbd_client_find(struct ceph_options *ceph_opts)
575 struct rbd_client *client_node;
578 if (ceph_opts->flags & CEPH_OPT_NOSHARE)
581 spin_lock(&rbd_client_list_lock);
582 list_for_each_entry(client_node, &rbd_client_list, node) {
583 if (!ceph_compare_options(ceph_opts, client_node->client)) {
584 __rbd_get_client(client_node);
590 spin_unlock(&rbd_client_list_lock);
592 return found ? client_node : NULL;
602 /* string args above */
605 /* Boolean args above */
609 static match_table_t rbd_opts_tokens = {
611 /* string args above */
612 {Opt_read_only, "read_only"},
613 {Opt_read_only, "ro"}, /* Alternate spelling */
614 {Opt_read_write, "read_write"},
615 {Opt_read_write, "rw"}, /* Alternate spelling */
616 /* Boolean args above */
624 #define RBD_READ_ONLY_DEFAULT false
626 static int parse_rbd_opts_token(char *c, void *private)
628 struct rbd_options *rbd_opts = private;
629 substring_t argstr[MAX_OPT_ARGS];
630 int token, intval, ret;
632 token = match_token(c, rbd_opts_tokens, argstr);
636 if (token < Opt_last_int) {
637 ret = match_int(&argstr[0], &intval);
639 pr_err("bad mount option arg (not int) "
643 dout("got int token %d val %d\n", token, intval);
644 } else if (token > Opt_last_int && token < Opt_last_string) {
645 dout("got string token %d val %s\n", token,
647 } else if (token > Opt_last_string && token < Opt_last_bool) {
648 dout("got Boolean token %d\n", token);
650 dout("got token %d\n", token);
655 rbd_opts->read_only = true;
658 rbd_opts->read_only = false;
668 * Get a ceph client with specific addr and configuration, if one does
669 * not exist create it. Either way, ceph_opts is consumed by this
672 static struct rbd_client *rbd_get_client(struct ceph_options *ceph_opts)
674 struct rbd_client *rbdc;
676 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
677 rbdc = rbd_client_find(ceph_opts);
678 if (rbdc) /* using an existing client */
679 ceph_destroy_options(ceph_opts);
681 rbdc = rbd_client_create(ceph_opts);
682 mutex_unlock(&ctl_mutex);
688 * Destroy ceph client
690 * Caller must hold rbd_client_list_lock.
692 static void rbd_client_release(struct kref *kref)
694 struct rbd_client *rbdc = container_of(kref, struct rbd_client, kref);
696 dout("%s: rbdc %p\n", __func__, rbdc);
697 spin_lock(&rbd_client_list_lock);
698 list_del(&rbdc->node);
699 spin_unlock(&rbd_client_list_lock);
701 ceph_destroy_client(rbdc->client);
706 * Drop reference to ceph client node. If it's not referenced anymore, release
709 static void rbd_put_client(struct rbd_client *rbdc)
712 kref_put(&rbdc->kref, rbd_client_release);
715 static bool rbd_image_format_valid(u32 image_format)
717 return image_format == 1 || image_format == 2;
720 static bool rbd_dev_ondisk_valid(struct rbd_image_header_ondisk *ondisk)
725 /* The header has to start with the magic rbd header text */
726 if (memcmp(&ondisk->text, RBD_HEADER_TEXT, sizeof (RBD_HEADER_TEXT)))
729 /* The bio layer requires at least sector-sized I/O */
731 if (ondisk->options.order < SECTOR_SHIFT)
734 /* If we use u64 in a few spots we may be able to loosen this */
736 if (ondisk->options.order > 8 * sizeof (int) - 1)
740 * The size of a snapshot header has to fit in a size_t, and
741 * that limits the number of snapshots.
743 snap_count = le32_to_cpu(ondisk->snap_count);
744 size = SIZE_MAX - sizeof (struct ceph_snap_context);
745 if (snap_count > size / sizeof (__le64))
749 * Not only that, but the size of the entire the snapshot
750 * header must also be representable in a size_t.
752 size -= snap_count * sizeof (__le64);
753 if ((u64) size < le64_to_cpu(ondisk->snap_names_len))
760 * Fill an rbd image header with information from the given format 1
763 static int rbd_header_from_disk(struct rbd_device *rbd_dev,
764 struct rbd_image_header_ondisk *ondisk)
766 struct rbd_image_header *header = &rbd_dev->header;
767 bool first_time = header->object_prefix == NULL;
768 struct ceph_snap_context *snapc;
769 char *object_prefix = NULL;
770 char *snap_names = NULL;
771 u64 *snap_sizes = NULL;
777 /* Allocate this now to avoid having to handle failure below */
782 len = strnlen(ondisk->object_prefix,
783 sizeof (ondisk->object_prefix));
784 object_prefix = kmalloc(len + 1, GFP_KERNEL);
787 memcpy(object_prefix, ondisk->object_prefix, len);
788 object_prefix[len] = '\0';
791 /* Allocate the snapshot context and fill it in */
793 snap_count = le32_to_cpu(ondisk->snap_count);
794 snapc = ceph_create_snap_context(snap_count, GFP_KERNEL);
797 snapc->seq = le64_to_cpu(ondisk->snap_seq);
799 struct rbd_image_snap_ondisk *snaps;
800 u64 snap_names_len = le64_to_cpu(ondisk->snap_names_len);
802 /* We'll keep a copy of the snapshot names... */
804 if (snap_names_len > (u64)SIZE_MAX)
806 snap_names = kmalloc(snap_names_len, GFP_KERNEL);
810 /* ...as well as the array of their sizes. */
812 size = snap_count * sizeof (*header->snap_sizes);
813 snap_sizes = kmalloc(size, GFP_KERNEL);
818 * Copy the names, and fill in each snapshot's id
821 * Note that rbd_dev_v1_header_info() guarantees the
822 * ondisk buffer we're working with has
823 * snap_names_len bytes beyond the end of the
824 * snapshot id array, this memcpy() is safe.
826 memcpy(snap_names, &ondisk->snaps[snap_count], snap_names_len);
827 snaps = ondisk->snaps;
828 for (i = 0; i < snap_count; i++) {
829 snapc->snaps[i] = le64_to_cpu(snaps[i].id);
830 snap_sizes[i] = le64_to_cpu(snaps[i].image_size);
834 /* We won't fail any more, fill in the header */
836 down_write(&rbd_dev->header_rwsem);
838 header->object_prefix = object_prefix;
839 header->obj_order = ondisk->options.order;
840 header->crypt_type = ondisk->options.crypt_type;
841 header->comp_type = ondisk->options.comp_type;
842 /* The rest aren't used for format 1 images */
843 header->stripe_unit = 0;
844 header->stripe_count = 0;
845 header->features = 0;
847 ceph_put_snap_context(header->snapc);
848 kfree(header->snap_names);
849 kfree(header->snap_sizes);
852 /* The remaining fields always get updated (when we refresh) */
854 header->image_size = le64_to_cpu(ondisk->image_size);
855 header->snapc = snapc;
856 header->snap_names = snap_names;
857 header->snap_sizes = snap_sizes;
859 /* Make sure mapping size is consistent with header info */
861 if (rbd_dev->spec->snap_id == CEPH_NOSNAP || first_time)
862 if (rbd_dev->mapping.size != header->image_size)
863 rbd_dev->mapping.size = header->image_size;
865 up_write(&rbd_dev->header_rwsem);
873 ceph_put_snap_context(snapc);
874 kfree(object_prefix);
879 static const char *_rbd_dev_v1_snap_name(struct rbd_device *rbd_dev, u32 which)
881 const char *snap_name;
883 rbd_assert(which < rbd_dev->header.snapc->num_snaps);
885 /* Skip over names until we find the one we are looking for */
887 snap_name = rbd_dev->header.snap_names;
889 snap_name += strlen(snap_name) + 1;
891 return kstrdup(snap_name, GFP_KERNEL);
895 * Snapshot id comparison function for use with qsort()/bsearch().
896 * Note that result is for snapshots in *descending* order.
898 static int snapid_compare_reverse(const void *s1, const void *s2)
900 u64 snap_id1 = *(u64 *)s1;
901 u64 snap_id2 = *(u64 *)s2;
903 if (snap_id1 < snap_id2)
905 return snap_id1 == snap_id2 ? 0 : -1;
909 * Search a snapshot context to see if the given snapshot id is
912 * Returns the position of the snapshot id in the array if it's found,
913 * or BAD_SNAP_INDEX otherwise.
915 * Note: The snapshot array is in kept sorted (by the osd) in
916 * reverse order, highest snapshot id first.
918 static u32 rbd_dev_snap_index(struct rbd_device *rbd_dev, u64 snap_id)
920 struct ceph_snap_context *snapc = rbd_dev->header.snapc;
923 found = bsearch(&snap_id, &snapc->snaps, snapc->num_snaps,
924 sizeof (snap_id), snapid_compare_reverse);
926 return found ? (u32)(found - &snapc->snaps[0]) : BAD_SNAP_INDEX;
929 static const char *rbd_dev_v1_snap_name(struct rbd_device *rbd_dev,
934 which = rbd_dev_snap_index(rbd_dev, snap_id);
935 if (which == BAD_SNAP_INDEX)
938 return _rbd_dev_v1_snap_name(rbd_dev, which);
941 static const char *rbd_snap_name(struct rbd_device *rbd_dev, u64 snap_id)
943 if (snap_id == CEPH_NOSNAP)
944 return RBD_SNAP_HEAD_NAME;
946 rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
947 if (rbd_dev->image_format == 1)
948 return rbd_dev_v1_snap_name(rbd_dev, snap_id);
950 return rbd_dev_v2_snap_name(rbd_dev, snap_id);
953 static int rbd_snap_size(struct rbd_device *rbd_dev, u64 snap_id,
956 rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
957 if (snap_id == CEPH_NOSNAP) {
958 *snap_size = rbd_dev->header.image_size;
959 } else if (rbd_dev->image_format == 1) {
962 which = rbd_dev_snap_index(rbd_dev, snap_id);
963 if (which == BAD_SNAP_INDEX)
966 *snap_size = rbd_dev->header.snap_sizes[which];
971 ret = _rbd_dev_v2_snap_size(rbd_dev, snap_id, NULL, &size);
980 static int rbd_snap_features(struct rbd_device *rbd_dev, u64 snap_id,
983 rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
984 if (snap_id == CEPH_NOSNAP) {
985 *snap_features = rbd_dev->header.features;
986 } else if (rbd_dev->image_format == 1) {
987 *snap_features = 0; /* No features for format 1 */
992 ret = _rbd_dev_v2_snap_features(rbd_dev, snap_id, &features);
996 *snap_features = features;
1001 static int rbd_dev_mapping_set(struct rbd_device *rbd_dev)
1003 u64 snap_id = rbd_dev->spec->snap_id;
1008 ret = rbd_snap_size(rbd_dev, snap_id, &size);
1011 ret = rbd_snap_features(rbd_dev, snap_id, &features);
1015 rbd_dev->mapping.size = size;
1016 rbd_dev->mapping.features = features;
1021 static void rbd_dev_mapping_clear(struct rbd_device *rbd_dev)
1023 rbd_dev->mapping.size = 0;
1024 rbd_dev->mapping.features = 0;
1027 static const char *rbd_segment_name(struct rbd_device *rbd_dev, u64 offset)
1034 name = kmem_cache_alloc(rbd_segment_name_cache, GFP_NOIO);
1037 segment = offset >> rbd_dev->header.obj_order;
1038 name_format = "%s.%012llx";
1039 if (rbd_dev->image_format == 2)
1040 name_format = "%s.%016llx";
1041 ret = snprintf(name, MAX_OBJ_NAME_SIZE + 1, name_format,
1042 rbd_dev->header.object_prefix, segment);
1043 if (ret < 0 || ret > MAX_OBJ_NAME_SIZE) {
1044 pr_err("error formatting segment name for #%llu (%d)\n",
1053 static void rbd_segment_name_free(const char *name)
1055 /* The explicit cast here is needed to drop the const qualifier */
1057 kmem_cache_free(rbd_segment_name_cache, (void *)name);
1060 static u64 rbd_segment_offset(struct rbd_device *rbd_dev, u64 offset)
1062 u64 segment_size = (u64) 1 << rbd_dev->header.obj_order;
1064 return offset & (segment_size - 1);
1067 static u64 rbd_segment_length(struct rbd_device *rbd_dev,
1068 u64 offset, u64 length)
1070 u64 segment_size = (u64) 1 << rbd_dev->header.obj_order;
1072 offset &= segment_size - 1;
1074 rbd_assert(length <= U64_MAX - offset);
1075 if (offset + length > segment_size)
1076 length = segment_size - offset;
1082 * returns the size of an object in the image
1084 static u64 rbd_obj_bytes(struct rbd_image_header *header)
1086 return 1 << header->obj_order;
1093 static void bio_chain_put(struct bio *chain)
1099 chain = chain->bi_next;
1105 * zeros a bio chain, starting at specific offset
1107 static void zero_bio_chain(struct bio *chain, int start_ofs)
1110 unsigned long flags;
1116 bio_for_each_segment(bv, chain, i) {
1117 if (pos + bv->bv_len > start_ofs) {
1118 int remainder = max(start_ofs - pos, 0);
1119 buf = bvec_kmap_irq(bv, &flags);
1120 memset(buf + remainder, 0,
1121 bv->bv_len - remainder);
1122 flush_dcache_page(bv->bv_page);
1123 bvec_kunmap_irq(buf, &flags);
1128 chain = chain->bi_next;
1133 * similar to zero_bio_chain(), zeros data defined by a page array,
1134 * starting at the given byte offset from the start of the array and
1135 * continuing up to the given end offset. The pages array is
1136 * assumed to be big enough to hold all bytes up to the end.
1138 static void zero_pages(struct page **pages, u64 offset, u64 end)
1140 struct page **page = &pages[offset >> PAGE_SHIFT];
1142 rbd_assert(end > offset);
1143 rbd_assert(end - offset <= (u64)SIZE_MAX);
1144 while (offset < end) {
1147 unsigned long flags;
1150 page_offset = offset & ~PAGE_MASK;
1151 length = min_t(size_t, PAGE_SIZE - page_offset, end - offset);
1152 local_irq_save(flags);
1153 kaddr = kmap_atomic(*page);
1154 memset(kaddr + page_offset, 0, length);
1155 flush_dcache_page(*page);
1156 kunmap_atomic(kaddr);
1157 local_irq_restore(flags);
1165 * Clone a portion of a bio, starting at the given byte offset
1166 * and continuing for the number of bytes indicated.
1168 static struct bio *bio_clone_range(struct bio *bio_src,
1169 unsigned int offset,
1177 unsigned short end_idx;
1178 unsigned short vcnt;
1181 /* Handle the easy case for the caller */
1183 if (!offset && len == bio_src->bi_size)
1184 return bio_clone(bio_src, gfpmask);
1186 if (WARN_ON_ONCE(!len))
1188 if (WARN_ON_ONCE(len > bio_src->bi_size))
1190 if (WARN_ON_ONCE(offset > bio_src->bi_size - len))
1193 /* Find first affected segment... */
1196 bio_for_each_segment(bv, bio_src, idx) {
1197 if (resid < bv->bv_len)
1199 resid -= bv->bv_len;
1203 /* ...and the last affected segment */
1206 __bio_for_each_segment(bv, bio_src, end_idx, idx) {
1207 if (resid <= bv->bv_len)
1209 resid -= bv->bv_len;
1211 vcnt = end_idx - idx + 1;
1213 /* Build the clone */
1215 bio = bio_alloc(gfpmask, (unsigned int) vcnt);
1217 return NULL; /* ENOMEM */
1219 bio->bi_bdev = bio_src->bi_bdev;
1220 bio->bi_sector = bio_src->bi_sector + (offset >> SECTOR_SHIFT);
1221 bio->bi_rw = bio_src->bi_rw;
1222 bio->bi_flags |= 1 << BIO_CLONED;
1225 * Copy over our part of the bio_vec, then update the first
1226 * and last (or only) entries.
1228 memcpy(&bio->bi_io_vec[0], &bio_src->bi_io_vec[idx],
1229 vcnt * sizeof (struct bio_vec));
1230 bio->bi_io_vec[0].bv_offset += voff;
1232 bio->bi_io_vec[0].bv_len -= voff;
1233 bio->bi_io_vec[vcnt - 1].bv_len = resid;
1235 bio->bi_io_vec[0].bv_len = len;
1238 bio->bi_vcnt = vcnt;
1246 * Clone a portion of a bio chain, starting at the given byte offset
1247 * into the first bio in the source chain and continuing for the
1248 * number of bytes indicated. The result is another bio chain of
1249 * exactly the given length, or a null pointer on error.
1251 * The bio_src and offset parameters are both in-out. On entry they
1252 * refer to the first source bio and the offset into that bio where
1253 * the start of data to be cloned is located.
1255 * On return, bio_src is updated to refer to the bio in the source
1256 * chain that contains first un-cloned byte, and *offset will
1257 * contain the offset of that byte within that bio.
1259 static struct bio *bio_chain_clone_range(struct bio **bio_src,
1260 unsigned int *offset,
1264 struct bio *bi = *bio_src;
1265 unsigned int off = *offset;
1266 struct bio *chain = NULL;
1269 /* Build up a chain of clone bios up to the limit */
1271 if (!bi || off >= bi->bi_size || !len)
1272 return NULL; /* Nothing to clone */
1276 unsigned int bi_size;
1280 rbd_warn(NULL, "bio_chain exhausted with %u left", len);
1281 goto out_err; /* EINVAL; ran out of bio's */
1283 bi_size = min_t(unsigned int, bi->bi_size - off, len);
1284 bio = bio_clone_range(bi, off, bi_size, gfpmask);
1286 goto out_err; /* ENOMEM */
1289 end = &bio->bi_next;
1292 if (off == bi->bi_size) {
1303 bio_chain_put(chain);
1309 * The default/initial value for all object request flags is 0. For
1310 * each flag, once its value is set to 1 it is never reset to 0
1313 static void obj_request_img_data_set(struct rbd_obj_request *obj_request)
1315 if (test_and_set_bit(OBJ_REQ_IMG_DATA, &obj_request->flags)) {
1316 struct rbd_device *rbd_dev;
1318 rbd_dev = obj_request->img_request->rbd_dev;
1319 rbd_warn(rbd_dev, "obj_request %p already marked img_data\n",
1324 static bool obj_request_img_data_test(struct rbd_obj_request *obj_request)
1327 return test_bit(OBJ_REQ_IMG_DATA, &obj_request->flags) != 0;
1330 static void obj_request_done_set(struct rbd_obj_request *obj_request)
1332 if (test_and_set_bit(OBJ_REQ_DONE, &obj_request->flags)) {
1333 struct rbd_device *rbd_dev = NULL;
1335 if (obj_request_img_data_test(obj_request))
1336 rbd_dev = obj_request->img_request->rbd_dev;
1337 rbd_warn(rbd_dev, "obj_request %p already marked done\n",
1342 static bool obj_request_done_test(struct rbd_obj_request *obj_request)
1345 return test_bit(OBJ_REQ_DONE, &obj_request->flags) != 0;
1349 * This sets the KNOWN flag after (possibly) setting the EXISTS
1350 * flag. The latter is set based on the "exists" value provided.
1352 * Note that for our purposes once an object exists it never goes
1353 * away again. It's possible that the response from two existence
1354 * checks are separated by the creation of the target object, and
1355 * the first ("doesn't exist") response arrives *after* the second
1356 * ("does exist"). In that case we ignore the second one.
1358 static void obj_request_existence_set(struct rbd_obj_request *obj_request,
1362 set_bit(OBJ_REQ_EXISTS, &obj_request->flags);
1363 set_bit(OBJ_REQ_KNOWN, &obj_request->flags);
1367 static bool obj_request_known_test(struct rbd_obj_request *obj_request)
1370 return test_bit(OBJ_REQ_KNOWN, &obj_request->flags) != 0;
1373 static bool obj_request_exists_test(struct rbd_obj_request *obj_request)
1376 return test_bit(OBJ_REQ_EXISTS, &obj_request->flags) != 0;
1379 static void rbd_obj_request_get(struct rbd_obj_request *obj_request)
1381 dout("%s: obj %p (was %d)\n", __func__, obj_request,
1382 atomic_read(&obj_request->kref.refcount));
1383 kref_get(&obj_request->kref);
1386 static void rbd_obj_request_destroy(struct kref *kref);
1387 static void rbd_obj_request_put(struct rbd_obj_request *obj_request)
1389 rbd_assert(obj_request != NULL);
1390 dout("%s: obj %p (was %d)\n", __func__, obj_request,
1391 atomic_read(&obj_request->kref.refcount));
1392 kref_put(&obj_request->kref, rbd_obj_request_destroy);
1395 static bool img_request_child_test(struct rbd_img_request *img_request);
1396 static void rbd_parent_request_destroy(struct kref *kref);
1397 static void rbd_img_request_destroy(struct kref *kref);
1398 static void rbd_img_request_put(struct rbd_img_request *img_request)
1400 rbd_assert(img_request != NULL);
1401 dout("%s: img %p (was %d)\n", __func__, img_request,
1402 atomic_read(&img_request->kref.refcount));
1403 if (img_request_child_test(img_request))
1404 kref_put(&img_request->kref, rbd_parent_request_destroy);
1406 kref_put(&img_request->kref, rbd_img_request_destroy);
1409 static inline void rbd_img_obj_request_add(struct rbd_img_request *img_request,
1410 struct rbd_obj_request *obj_request)
1412 rbd_assert(obj_request->img_request == NULL);
1414 /* Image request now owns object's original reference */
1415 obj_request->img_request = img_request;
1416 obj_request->which = img_request->obj_request_count;
1417 rbd_assert(!obj_request_img_data_test(obj_request));
1418 obj_request_img_data_set(obj_request);
1419 rbd_assert(obj_request->which != BAD_WHICH);
1420 img_request->obj_request_count++;
1421 list_add_tail(&obj_request->links, &img_request->obj_requests);
1422 dout("%s: img %p obj %p w=%u\n", __func__, img_request, obj_request,
1423 obj_request->which);
1426 static inline void rbd_img_obj_request_del(struct rbd_img_request *img_request,
1427 struct rbd_obj_request *obj_request)
1429 rbd_assert(obj_request->which != BAD_WHICH);
1431 dout("%s: img %p obj %p w=%u\n", __func__, img_request, obj_request,
1432 obj_request->which);
1433 list_del(&obj_request->links);
1434 rbd_assert(img_request->obj_request_count > 0);
1435 img_request->obj_request_count--;
1436 rbd_assert(obj_request->which == img_request->obj_request_count);
1437 obj_request->which = BAD_WHICH;
1438 rbd_assert(obj_request_img_data_test(obj_request));
1439 rbd_assert(obj_request->img_request == img_request);
1440 obj_request->img_request = NULL;
1441 obj_request->callback = NULL;
1442 rbd_obj_request_put(obj_request);
1445 static bool obj_request_type_valid(enum obj_request_type type)
1448 case OBJ_REQUEST_NODATA:
1449 case OBJ_REQUEST_BIO:
1450 case OBJ_REQUEST_PAGES:
1457 static int rbd_obj_request_submit(struct ceph_osd_client *osdc,
1458 struct rbd_obj_request *obj_request)
1460 dout("%s: osdc %p obj %p\n", __func__, osdc, obj_request);
1462 return ceph_osdc_start_request(osdc, obj_request->osd_req, false);
1465 static void rbd_img_request_complete(struct rbd_img_request *img_request)
1468 dout("%s: img %p\n", __func__, img_request);
1471 * If no error occurred, compute the aggregate transfer
1472 * count for the image request. We could instead use
1473 * atomic64_cmpxchg() to update it as each object request
1474 * completes; not clear which way is better off hand.
1476 if (!img_request->result) {
1477 struct rbd_obj_request *obj_request;
1480 for_each_obj_request(img_request, obj_request)
1481 xferred += obj_request->xferred;
1482 img_request->xferred = xferred;
1485 if (img_request->callback)
1486 img_request->callback(img_request);
1488 rbd_img_request_put(img_request);
1491 /* Caller is responsible for rbd_obj_request_destroy(obj_request) */
1493 static int rbd_obj_request_wait(struct rbd_obj_request *obj_request)
1495 dout("%s: obj %p\n", __func__, obj_request);
1497 return wait_for_completion_interruptible(&obj_request->completion);
1501 * The default/initial value for all image request flags is 0. Each
1502 * is conditionally set to 1 at image request initialization time
1503 * and currently never change thereafter.
1505 static void img_request_write_set(struct rbd_img_request *img_request)
1507 set_bit(IMG_REQ_WRITE, &img_request->flags);
1511 static bool img_request_write_test(struct rbd_img_request *img_request)
1514 return test_bit(IMG_REQ_WRITE, &img_request->flags) != 0;
1517 static void img_request_child_set(struct rbd_img_request *img_request)
1519 set_bit(IMG_REQ_CHILD, &img_request->flags);
1523 static void img_request_child_clear(struct rbd_img_request *img_request)
1525 clear_bit(IMG_REQ_CHILD, &img_request->flags);
1529 static bool img_request_child_test(struct rbd_img_request *img_request)
1532 return test_bit(IMG_REQ_CHILD, &img_request->flags) != 0;
1535 static void img_request_layered_set(struct rbd_img_request *img_request)
1537 set_bit(IMG_REQ_LAYERED, &img_request->flags);
1541 static void img_request_layered_clear(struct rbd_img_request *img_request)
1543 clear_bit(IMG_REQ_LAYERED, &img_request->flags);
1547 static bool img_request_layered_test(struct rbd_img_request *img_request)
1550 return test_bit(IMG_REQ_LAYERED, &img_request->flags) != 0;
1554 rbd_img_obj_request_read_callback(struct rbd_obj_request *obj_request)
1556 u64 xferred = obj_request->xferred;
1557 u64 length = obj_request->length;
1559 dout("%s: obj %p img %p result %d %llu/%llu\n", __func__,
1560 obj_request, obj_request->img_request, obj_request->result,
1563 * ENOENT means a hole in the image. We zero-fill the
1564 * entire length of the request. A short read also implies
1565 * zero-fill to the end of the request. Either way we
1566 * update the xferred count to indicate the whole request
1569 rbd_assert(obj_request->type != OBJ_REQUEST_NODATA);
1570 if (obj_request->result == -ENOENT) {
1571 if (obj_request->type == OBJ_REQUEST_BIO)
1572 zero_bio_chain(obj_request->bio_list, 0);
1574 zero_pages(obj_request->pages, 0, length);
1575 obj_request->result = 0;
1576 obj_request->xferred = length;
1577 } else if (xferred < length && !obj_request->result) {
1578 if (obj_request->type == OBJ_REQUEST_BIO)
1579 zero_bio_chain(obj_request->bio_list, xferred);
1581 zero_pages(obj_request->pages, xferred, length);
1582 obj_request->xferred = length;
1584 obj_request_done_set(obj_request);
1587 static void rbd_obj_request_complete(struct rbd_obj_request *obj_request)
1589 dout("%s: obj %p cb %p\n", __func__, obj_request,
1590 obj_request->callback);
1591 if (obj_request->callback)
1592 obj_request->callback(obj_request);
1594 complete_all(&obj_request->completion);
1597 static void rbd_osd_trivial_callback(struct rbd_obj_request *obj_request)
1599 dout("%s: obj %p\n", __func__, obj_request);
1600 obj_request_done_set(obj_request);
1603 static void rbd_osd_read_callback(struct rbd_obj_request *obj_request)
1605 struct rbd_img_request *img_request = NULL;
1606 struct rbd_device *rbd_dev = NULL;
1607 bool layered = false;
1609 if (obj_request_img_data_test(obj_request)) {
1610 img_request = obj_request->img_request;
1611 layered = img_request && img_request_layered_test(img_request);
1612 rbd_dev = img_request->rbd_dev;
1615 dout("%s: obj %p img %p result %d %llu/%llu\n", __func__,
1616 obj_request, img_request, obj_request->result,
1617 obj_request->xferred, obj_request->length);
1618 if (layered && obj_request->result == -ENOENT &&
1619 obj_request->img_offset < rbd_dev->parent_overlap)
1620 rbd_img_parent_read(obj_request);
1621 else if (img_request)
1622 rbd_img_obj_request_read_callback(obj_request);
1624 obj_request_done_set(obj_request);
1627 static void rbd_osd_write_callback(struct rbd_obj_request *obj_request)
1629 dout("%s: obj %p result %d %llu\n", __func__, obj_request,
1630 obj_request->result, obj_request->length);
1632 * There is no such thing as a successful short write. Set
1633 * it to our originally-requested length.
1635 obj_request->xferred = obj_request->length;
1636 obj_request_done_set(obj_request);
1640 * For a simple stat call there's nothing to do. We'll do more if
1641 * this is part of a write sequence for a layered image.
1643 static void rbd_osd_stat_callback(struct rbd_obj_request *obj_request)
1645 dout("%s: obj %p\n", __func__, obj_request);
1646 obj_request_done_set(obj_request);
1649 static void rbd_osd_req_callback(struct ceph_osd_request *osd_req,
1650 struct ceph_msg *msg)
1652 struct rbd_obj_request *obj_request = osd_req->r_priv;
1655 dout("%s: osd_req %p msg %p\n", __func__, osd_req, msg);
1656 rbd_assert(osd_req == obj_request->osd_req);
1657 if (obj_request_img_data_test(obj_request)) {
1658 rbd_assert(obj_request->img_request);
1659 rbd_assert(obj_request->which != BAD_WHICH);
1661 rbd_assert(obj_request->which == BAD_WHICH);
1664 if (osd_req->r_result < 0)
1665 obj_request->result = osd_req->r_result;
1667 BUG_ON(osd_req->r_num_ops > 2);
1670 * We support a 64-bit length, but ultimately it has to be
1671 * passed to blk_end_request(), which takes an unsigned int.
1673 obj_request->xferred = osd_req->r_reply_op_len[0];
1674 rbd_assert(obj_request->xferred < (u64)UINT_MAX);
1675 opcode = osd_req->r_ops[0].op;
1677 case CEPH_OSD_OP_READ:
1678 rbd_osd_read_callback(obj_request);
1680 case CEPH_OSD_OP_WRITE:
1681 rbd_osd_write_callback(obj_request);
1683 case CEPH_OSD_OP_STAT:
1684 rbd_osd_stat_callback(obj_request);
1686 case CEPH_OSD_OP_CALL:
1687 case CEPH_OSD_OP_NOTIFY_ACK:
1688 case CEPH_OSD_OP_WATCH:
1689 rbd_osd_trivial_callback(obj_request);
1692 rbd_warn(NULL, "%s: unsupported op %hu\n",
1693 obj_request->object_name, (unsigned short) opcode);
1697 if (obj_request_done_test(obj_request))
1698 rbd_obj_request_complete(obj_request);
1701 static void rbd_osd_req_format_read(struct rbd_obj_request *obj_request)
1703 struct rbd_img_request *img_request = obj_request->img_request;
1704 struct ceph_osd_request *osd_req = obj_request->osd_req;
1707 rbd_assert(osd_req != NULL);
1709 snap_id = img_request ? img_request->snap_id : CEPH_NOSNAP;
1710 ceph_osdc_build_request(osd_req, obj_request->offset,
1711 NULL, snap_id, NULL);
1714 static void rbd_osd_req_format_write(struct rbd_obj_request *obj_request)
1716 struct rbd_img_request *img_request = obj_request->img_request;
1717 struct ceph_osd_request *osd_req = obj_request->osd_req;
1718 struct ceph_snap_context *snapc;
1719 struct timespec mtime = CURRENT_TIME;
1721 rbd_assert(osd_req != NULL);
1723 snapc = img_request ? img_request->snapc : NULL;
1724 ceph_osdc_build_request(osd_req, obj_request->offset,
1725 snapc, CEPH_NOSNAP, &mtime);
1728 static struct ceph_osd_request *rbd_osd_req_create(
1729 struct rbd_device *rbd_dev,
1731 struct rbd_obj_request *obj_request)
1733 struct ceph_snap_context *snapc = NULL;
1734 struct ceph_osd_client *osdc;
1735 struct ceph_osd_request *osd_req;
1737 if (obj_request_img_data_test(obj_request)) {
1738 struct rbd_img_request *img_request = obj_request->img_request;
1740 rbd_assert(write_request ==
1741 img_request_write_test(img_request));
1743 snapc = img_request->snapc;
1746 /* Allocate and initialize the request, for the single op */
1748 osdc = &rbd_dev->rbd_client->client->osdc;
1749 osd_req = ceph_osdc_alloc_request(osdc, snapc, 1, false, GFP_ATOMIC);
1751 return NULL; /* ENOMEM */
1754 osd_req->r_flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK;
1756 osd_req->r_flags = CEPH_OSD_FLAG_READ;
1758 osd_req->r_callback = rbd_osd_req_callback;
1759 osd_req->r_priv = obj_request;
1761 osd_req->r_oid_len = strlen(obj_request->object_name);
1762 rbd_assert(osd_req->r_oid_len < sizeof (osd_req->r_oid));
1763 memcpy(osd_req->r_oid, obj_request->object_name, osd_req->r_oid_len);
1765 osd_req->r_file_layout = rbd_dev->layout; /* struct */
1771 * Create a copyup osd request based on the information in the
1772 * object request supplied. A copyup request has two osd ops,
1773 * a copyup method call, and a "normal" write request.
1775 static struct ceph_osd_request *
1776 rbd_osd_req_create_copyup(struct rbd_obj_request *obj_request)
1778 struct rbd_img_request *img_request;
1779 struct ceph_snap_context *snapc;
1780 struct rbd_device *rbd_dev;
1781 struct ceph_osd_client *osdc;
1782 struct ceph_osd_request *osd_req;
1784 rbd_assert(obj_request_img_data_test(obj_request));
1785 img_request = obj_request->img_request;
1786 rbd_assert(img_request);
1787 rbd_assert(img_request_write_test(img_request));
1789 /* Allocate and initialize the request, for the two ops */
1791 snapc = img_request->snapc;
1792 rbd_dev = img_request->rbd_dev;
1793 osdc = &rbd_dev->rbd_client->client->osdc;
1794 osd_req = ceph_osdc_alloc_request(osdc, snapc, 2, false, GFP_ATOMIC);
1796 return NULL; /* ENOMEM */
1798 osd_req->r_flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK;
1799 osd_req->r_callback = rbd_osd_req_callback;
1800 osd_req->r_priv = obj_request;
1802 osd_req->r_oid_len = strlen(obj_request->object_name);
1803 rbd_assert(osd_req->r_oid_len < sizeof (osd_req->r_oid));
1804 memcpy(osd_req->r_oid, obj_request->object_name, osd_req->r_oid_len);
1806 osd_req->r_file_layout = rbd_dev->layout; /* struct */
1812 static void rbd_osd_req_destroy(struct ceph_osd_request *osd_req)
1814 ceph_osdc_put_request(osd_req);
1817 /* object_name is assumed to be a non-null pointer and NUL-terminated */
1819 static struct rbd_obj_request *rbd_obj_request_create(const char *object_name,
1820 u64 offset, u64 length,
1821 enum obj_request_type type)
1823 struct rbd_obj_request *obj_request;
1827 rbd_assert(obj_request_type_valid(type));
1829 size = strlen(object_name) + 1;
1830 name = kmalloc(size, GFP_KERNEL);
1834 obj_request = kmem_cache_zalloc(rbd_obj_request_cache, GFP_KERNEL);
1840 obj_request->object_name = memcpy(name, object_name, size);
1841 obj_request->offset = offset;
1842 obj_request->length = length;
1843 obj_request->flags = 0;
1844 obj_request->which = BAD_WHICH;
1845 obj_request->type = type;
1846 INIT_LIST_HEAD(&obj_request->links);
1847 init_completion(&obj_request->completion);
1848 kref_init(&obj_request->kref);
1850 dout("%s: \"%s\" %llu/%llu %d -> obj %p\n", __func__, object_name,
1851 offset, length, (int)type, obj_request);
1856 static void rbd_obj_request_destroy(struct kref *kref)
1858 struct rbd_obj_request *obj_request;
1860 obj_request = container_of(kref, struct rbd_obj_request, kref);
1862 dout("%s: obj %p\n", __func__, obj_request);
1864 rbd_assert(obj_request->img_request == NULL);
1865 rbd_assert(obj_request->which == BAD_WHICH);
1867 if (obj_request->osd_req)
1868 rbd_osd_req_destroy(obj_request->osd_req);
1870 rbd_assert(obj_request_type_valid(obj_request->type));
1871 switch (obj_request->type) {
1872 case OBJ_REQUEST_NODATA:
1873 break; /* Nothing to do */
1874 case OBJ_REQUEST_BIO:
1875 if (obj_request->bio_list)
1876 bio_chain_put(obj_request->bio_list);
1878 case OBJ_REQUEST_PAGES:
1879 if (obj_request->pages)
1880 ceph_release_page_vector(obj_request->pages,
1881 obj_request->page_count);
1885 kfree(obj_request->object_name);
1886 obj_request->object_name = NULL;
1887 kmem_cache_free(rbd_obj_request_cache, obj_request);
1890 /* It's OK to call this for a device with no parent */
1892 static void rbd_spec_put(struct rbd_spec *spec);
1893 static void rbd_dev_unparent(struct rbd_device *rbd_dev)
1895 rbd_dev_remove_parent(rbd_dev);
1896 rbd_spec_put(rbd_dev->parent_spec);
1897 rbd_dev->parent_spec = NULL;
1898 rbd_dev->parent_overlap = 0;
1902 * Parent image reference counting is used to determine when an
1903 * image's parent fields can be safely torn down--after there are no
1904 * more in-flight requests to the parent image. When the last
1905 * reference is dropped, cleaning them up is safe.
1907 static void rbd_dev_parent_put(struct rbd_device *rbd_dev)
1911 if (!rbd_dev->parent_spec)
1914 counter = atomic_dec_return_safe(&rbd_dev->parent_ref);
1918 /* Last reference; clean up parent data structures */
1921 rbd_dev_unparent(rbd_dev);
1923 rbd_warn(rbd_dev, "parent reference underflow\n");
1927 * If an image has a non-zero parent overlap, get a reference to its
1930 * We must get the reference before checking for the overlap to
1931 * coordinate properly with zeroing the parent overlap in
1932 * rbd_dev_v2_parent_info() when an image gets flattened. We
1933 * drop it again if there is no overlap.
1935 * Returns true if the rbd device has a parent with a non-zero
1936 * overlap and a reference for it was successfully taken, or
1939 static bool rbd_dev_parent_get(struct rbd_device *rbd_dev)
1943 if (!rbd_dev->parent_spec)
1946 counter = atomic_inc_return_safe(&rbd_dev->parent_ref);
1947 if (counter > 0 && rbd_dev->parent_overlap)
1950 /* Image was flattened, but parent is not yet torn down */
1953 rbd_warn(rbd_dev, "parent reference overflow\n");
1959 * Caller is responsible for filling in the list of object requests
1960 * that comprises the image request, and the Linux request pointer
1961 * (if there is one).
1963 static struct rbd_img_request *rbd_img_request_create(
1964 struct rbd_device *rbd_dev,
1965 u64 offset, u64 length,
1968 struct rbd_img_request *img_request;
1970 img_request = kmem_cache_alloc(rbd_img_request_cache, GFP_ATOMIC);
1974 if (write_request) {
1975 down_read(&rbd_dev->header_rwsem);
1976 ceph_get_snap_context(rbd_dev->header.snapc);
1977 up_read(&rbd_dev->header_rwsem);
1980 img_request->rq = NULL;
1981 img_request->rbd_dev = rbd_dev;
1982 img_request->offset = offset;
1983 img_request->length = length;
1984 img_request->flags = 0;
1985 if (write_request) {
1986 img_request_write_set(img_request);
1987 img_request->snapc = rbd_dev->header.snapc;
1989 img_request->snap_id = rbd_dev->spec->snap_id;
1991 if (rbd_dev_parent_get(rbd_dev))
1992 img_request_layered_set(img_request);
1993 spin_lock_init(&img_request->completion_lock);
1994 img_request->next_completion = 0;
1995 img_request->callback = NULL;
1996 img_request->result = 0;
1997 img_request->obj_request_count = 0;
1998 INIT_LIST_HEAD(&img_request->obj_requests);
1999 kref_init(&img_request->kref);
2001 dout("%s: rbd_dev %p %s %llu/%llu -> img %p\n", __func__, rbd_dev,
2002 write_request ? "write" : "read", offset, length,
2008 static void rbd_img_request_destroy(struct kref *kref)
2010 struct rbd_img_request *img_request;
2011 struct rbd_obj_request *obj_request;
2012 struct rbd_obj_request *next_obj_request;
2014 img_request = container_of(kref, struct rbd_img_request, kref);
2016 dout("%s: img %p\n", __func__, img_request);
2018 for_each_obj_request_safe(img_request, obj_request, next_obj_request)
2019 rbd_img_obj_request_del(img_request, obj_request);
2020 rbd_assert(img_request->obj_request_count == 0);
2022 if (img_request_layered_test(img_request)) {
2023 img_request_layered_clear(img_request);
2024 rbd_dev_parent_put(img_request->rbd_dev);
2027 if (img_request_write_test(img_request))
2028 ceph_put_snap_context(img_request->snapc);
2030 kmem_cache_free(rbd_img_request_cache, img_request);
2033 static struct rbd_img_request *rbd_parent_request_create(
2034 struct rbd_obj_request *obj_request,
2035 u64 img_offset, u64 length)
2037 struct rbd_img_request *parent_request;
2038 struct rbd_device *rbd_dev;
2040 rbd_assert(obj_request->img_request);
2041 rbd_dev = obj_request->img_request->rbd_dev;
2043 parent_request = rbd_img_request_create(rbd_dev->parent,
2044 img_offset, length, false);
2045 if (!parent_request)
2048 img_request_child_set(parent_request);
2049 rbd_obj_request_get(obj_request);
2050 parent_request->obj_request = obj_request;
2052 return parent_request;
2055 static void rbd_parent_request_destroy(struct kref *kref)
2057 struct rbd_img_request *parent_request;
2058 struct rbd_obj_request *orig_request;
2060 parent_request = container_of(kref, struct rbd_img_request, kref);
2061 orig_request = parent_request->obj_request;
2063 parent_request->obj_request = NULL;
2064 rbd_obj_request_put(orig_request);
2065 img_request_child_clear(parent_request);
2067 rbd_img_request_destroy(kref);
2070 static bool rbd_img_obj_end_request(struct rbd_obj_request *obj_request)
2072 struct rbd_img_request *img_request;
2073 unsigned int xferred;
2077 rbd_assert(obj_request_img_data_test(obj_request));
2078 img_request = obj_request->img_request;
2080 rbd_assert(obj_request->xferred <= (u64)UINT_MAX);
2081 xferred = (unsigned int)obj_request->xferred;
2082 result = obj_request->result;
2084 struct rbd_device *rbd_dev = img_request->rbd_dev;
2086 rbd_warn(rbd_dev, "%s %llx at %llx (%llx)\n",
2087 img_request_write_test(img_request) ? "write" : "read",
2088 obj_request->length, obj_request->img_offset,
2089 obj_request->offset);
2090 rbd_warn(rbd_dev, " result %d xferred %x\n",
2092 if (!img_request->result)
2093 img_request->result = result;
2096 /* Image object requests don't own their page array */
2098 if (obj_request->type == OBJ_REQUEST_PAGES) {
2099 obj_request->pages = NULL;
2100 obj_request->page_count = 0;
2103 if (img_request_child_test(img_request)) {
2104 rbd_assert(img_request->obj_request != NULL);
2105 more = obj_request->which < img_request->obj_request_count - 1;
2107 rbd_assert(img_request->rq != NULL);
2108 more = blk_end_request(img_request->rq, result, xferred);
2114 static void rbd_img_obj_callback(struct rbd_obj_request *obj_request)
2116 struct rbd_img_request *img_request;
2117 u32 which = obj_request->which;
2120 rbd_assert(obj_request_img_data_test(obj_request));
2121 img_request = obj_request->img_request;
2123 dout("%s: img %p obj %p\n", __func__, img_request, obj_request);
2124 rbd_assert(img_request != NULL);
2125 rbd_assert(img_request->obj_request_count > 0);
2126 rbd_assert(which != BAD_WHICH);
2127 rbd_assert(which < img_request->obj_request_count);
2128 rbd_assert(which >= img_request->next_completion);
2130 spin_lock_irq(&img_request->completion_lock);
2131 if (which != img_request->next_completion)
2134 for_each_obj_request_from(img_request, obj_request) {
2136 rbd_assert(which < img_request->obj_request_count);
2138 if (!obj_request_done_test(obj_request))
2140 more = rbd_img_obj_end_request(obj_request);
2144 rbd_assert(more ^ (which == img_request->obj_request_count));
2145 img_request->next_completion = which;
2147 spin_unlock_irq(&img_request->completion_lock);
2150 rbd_img_request_complete(img_request);
2154 * Split up an image request into one or more object requests, each
2155 * to a different object. The "type" parameter indicates whether
2156 * "data_desc" is the pointer to the head of a list of bio
2157 * structures, or the base of a page array. In either case this
2158 * function assumes data_desc describes memory sufficient to hold
2159 * all data described by the image request.
2161 static int rbd_img_request_fill(struct rbd_img_request *img_request,
2162 enum obj_request_type type,
2165 struct rbd_device *rbd_dev = img_request->rbd_dev;
2166 struct rbd_obj_request *obj_request = NULL;
2167 struct rbd_obj_request *next_obj_request;
2168 bool write_request = img_request_write_test(img_request);
2169 struct bio *bio_list;
2170 unsigned int bio_offset = 0;
2171 struct page **pages;
2176 dout("%s: img %p type %d data_desc %p\n", __func__, img_request,
2177 (int)type, data_desc);
2179 opcode = write_request ? CEPH_OSD_OP_WRITE : CEPH_OSD_OP_READ;
2180 img_offset = img_request->offset;
2181 resid = img_request->length;
2182 rbd_assert(resid > 0);
2184 if (type == OBJ_REQUEST_BIO) {
2185 bio_list = data_desc;
2186 rbd_assert(img_offset == bio_list->bi_sector << SECTOR_SHIFT);
2188 rbd_assert(type == OBJ_REQUEST_PAGES);
2193 struct ceph_osd_request *osd_req;
2194 const char *object_name;
2198 object_name = rbd_segment_name(rbd_dev, img_offset);
2201 offset = rbd_segment_offset(rbd_dev, img_offset);
2202 length = rbd_segment_length(rbd_dev, img_offset, resid);
2203 obj_request = rbd_obj_request_create(object_name,
2204 offset, length, type);
2205 /* object request has its own copy of the object name */
2206 rbd_segment_name_free(object_name);
2210 if (type == OBJ_REQUEST_BIO) {
2211 unsigned int clone_size;
2213 rbd_assert(length <= (u64)UINT_MAX);
2214 clone_size = (unsigned int)length;
2215 obj_request->bio_list =
2216 bio_chain_clone_range(&bio_list,
2220 if (!obj_request->bio_list)
2223 unsigned int page_count;
2225 obj_request->pages = pages;
2226 page_count = (u32)calc_pages_for(offset, length);
2227 obj_request->page_count = page_count;
2228 if ((offset + length) & ~PAGE_MASK)
2229 page_count--; /* more on last page */
2230 pages += page_count;
2233 osd_req = rbd_osd_req_create(rbd_dev, write_request,
2237 obj_request->osd_req = osd_req;
2238 obj_request->callback = rbd_img_obj_callback;
2240 osd_req_op_extent_init(osd_req, 0, opcode, offset, length,
2242 if (type == OBJ_REQUEST_BIO)
2243 osd_req_op_extent_osd_data_bio(osd_req, 0,
2244 obj_request->bio_list, length);
2246 osd_req_op_extent_osd_data_pages(osd_req, 0,
2247 obj_request->pages, length,
2248 offset & ~PAGE_MASK, false, false);
2251 * set obj_request->img_request before formatting
2252 * the osd_request so that it gets the right snapc
2254 rbd_img_obj_request_add(img_request, obj_request);
2256 rbd_osd_req_format_write(obj_request);
2258 rbd_osd_req_format_read(obj_request);
2260 obj_request->img_offset = img_offset;
2262 img_offset += length;
2269 rbd_obj_request_put(obj_request);
2271 for_each_obj_request_safe(img_request, obj_request, next_obj_request)
2272 rbd_obj_request_put(obj_request);
2278 rbd_img_obj_copyup_callback(struct rbd_obj_request *obj_request)
2280 struct rbd_img_request *img_request;
2281 struct rbd_device *rbd_dev;
2282 struct page **pages;
2285 rbd_assert(obj_request->type == OBJ_REQUEST_BIO);
2286 rbd_assert(obj_request_img_data_test(obj_request));
2287 img_request = obj_request->img_request;
2288 rbd_assert(img_request);
2290 rbd_dev = img_request->rbd_dev;
2291 rbd_assert(rbd_dev);
2293 pages = obj_request->copyup_pages;
2294 rbd_assert(pages != NULL);
2295 obj_request->copyup_pages = NULL;
2296 page_count = obj_request->copyup_page_count;
2297 rbd_assert(page_count);
2298 obj_request->copyup_page_count = 0;
2299 ceph_release_page_vector(pages, page_count);
2302 * We want the transfer count to reflect the size of the
2303 * original write request. There is no such thing as a
2304 * successful short write, so if the request was successful
2305 * we can just set it to the originally-requested length.
2307 if (!obj_request->result)
2308 obj_request->xferred = obj_request->length;
2310 /* Finish up with the normal image object callback */
2312 rbd_img_obj_callback(obj_request);
2316 rbd_img_obj_parent_read_full_callback(struct rbd_img_request *img_request)
2318 struct rbd_obj_request *orig_request;
2319 struct ceph_osd_request *osd_req;
2320 struct ceph_osd_client *osdc;
2321 struct rbd_device *rbd_dev;
2322 struct page **pages;
2329 rbd_assert(img_request_child_test(img_request));
2331 /* First get what we need from the image request */
2333 pages = img_request->copyup_pages;
2334 rbd_assert(pages != NULL);
2335 img_request->copyup_pages = NULL;
2336 page_count = img_request->copyup_page_count;
2337 rbd_assert(page_count);
2338 img_request->copyup_page_count = 0;
2340 orig_request = img_request->obj_request;
2341 rbd_assert(orig_request != NULL);
2342 rbd_assert(obj_request_type_valid(orig_request->type));
2343 img_result = img_request->result;
2344 parent_length = img_request->length;
2345 rbd_assert(parent_length == img_request->xferred);
2346 rbd_img_request_put(img_request);
2348 rbd_assert(orig_request->img_request);
2349 rbd_dev = orig_request->img_request->rbd_dev;
2350 rbd_assert(rbd_dev);
2353 * If the overlap has become 0 (most likely because the
2354 * image has been flattened) we need to free the pages
2355 * and re-submit the original write request.
2357 if (!rbd_dev->parent_overlap) {
2358 struct ceph_osd_client *osdc;
2360 ceph_release_page_vector(pages, page_count);
2361 osdc = &rbd_dev->rbd_client->client->osdc;
2362 img_result = rbd_obj_request_submit(osdc, orig_request);
2371 * The original osd request is of no use to use any more.
2372 * We need a new one that can hold the two ops in a copyup
2373 * request. Allocate the new copyup osd request for the
2374 * original request, and release the old one.
2376 img_result = -ENOMEM;
2377 osd_req = rbd_osd_req_create_copyup(orig_request);
2380 rbd_osd_req_destroy(orig_request->osd_req);
2381 orig_request->osd_req = osd_req;
2382 orig_request->copyup_pages = pages;
2383 orig_request->copyup_page_count = page_count;
2385 /* Initialize the copyup op */
2387 osd_req_op_cls_init(osd_req, 0, CEPH_OSD_OP_CALL, "rbd", "copyup");
2388 osd_req_op_cls_request_data_pages(osd_req, 0, pages, parent_length, 0,
2391 /* Then the original write request op */
2393 offset = orig_request->offset;
2394 length = orig_request->length;
2395 osd_req_op_extent_init(osd_req, 1, CEPH_OSD_OP_WRITE,
2396 offset, length, 0, 0);
2397 if (orig_request->type == OBJ_REQUEST_BIO)
2398 osd_req_op_extent_osd_data_bio(osd_req, 1,
2399 orig_request->bio_list, length);
2401 osd_req_op_extent_osd_data_pages(osd_req, 1,
2402 orig_request->pages, length,
2403 offset & ~PAGE_MASK, false, false);
2405 rbd_osd_req_format_write(orig_request);
2407 /* All set, send it off. */
2409 orig_request->callback = rbd_img_obj_copyup_callback;
2410 osdc = &rbd_dev->rbd_client->client->osdc;
2411 img_result = rbd_obj_request_submit(osdc, orig_request);
2415 /* Record the error code and complete the request */
2417 orig_request->result = img_result;
2418 orig_request->xferred = 0;
2419 obj_request_done_set(orig_request);
2420 rbd_obj_request_complete(orig_request);
2424 * Read from the parent image the range of data that covers the
2425 * entire target of the given object request. This is used for
2426 * satisfying a layered image write request when the target of an
2427 * object request from the image request does not exist.
2429 * A page array big enough to hold the returned data is allocated
2430 * and supplied to rbd_img_request_fill() as the "data descriptor."
2431 * When the read completes, this page array will be transferred to
2432 * the original object request for the copyup operation.
2434 * If an error occurs, record it as the result of the original
2435 * object request and mark it done so it gets completed.
2437 static int rbd_img_obj_parent_read_full(struct rbd_obj_request *obj_request)
2439 struct rbd_img_request *img_request = NULL;
2440 struct rbd_img_request *parent_request = NULL;
2441 struct rbd_device *rbd_dev;
2444 struct page **pages = NULL;
2448 rbd_assert(obj_request_img_data_test(obj_request));
2449 rbd_assert(obj_request_type_valid(obj_request->type));
2451 img_request = obj_request->img_request;
2452 rbd_assert(img_request != NULL);
2453 rbd_dev = img_request->rbd_dev;
2454 rbd_assert(rbd_dev->parent != NULL);
2457 * Determine the byte range covered by the object in the
2458 * child image to which the original request was to be sent.
2460 img_offset = obj_request->img_offset - obj_request->offset;
2461 length = (u64)1 << rbd_dev->header.obj_order;
2464 * There is no defined parent data beyond the parent
2465 * overlap, so limit what we read at that boundary if
2468 if (img_offset + length > rbd_dev->parent_overlap) {
2469 rbd_assert(img_offset < rbd_dev->parent_overlap);
2470 length = rbd_dev->parent_overlap - img_offset;
2474 * Allocate a page array big enough to receive the data read
2477 page_count = (u32)calc_pages_for(0, length);
2478 pages = ceph_alloc_page_vector(page_count, GFP_KERNEL);
2479 if (IS_ERR(pages)) {
2480 result = PTR_ERR(pages);
2486 parent_request = rbd_parent_request_create(obj_request,
2487 img_offset, length);
2488 if (!parent_request)
2491 result = rbd_img_request_fill(parent_request, OBJ_REQUEST_PAGES, pages);
2494 parent_request->copyup_pages = pages;
2495 parent_request->copyup_page_count = page_count;
2497 parent_request->callback = rbd_img_obj_parent_read_full_callback;
2498 result = rbd_img_request_submit(parent_request);
2502 parent_request->copyup_pages = NULL;
2503 parent_request->copyup_page_count = 0;
2504 parent_request->obj_request = NULL;
2505 rbd_obj_request_put(obj_request);
2508 ceph_release_page_vector(pages, page_count);
2510 rbd_img_request_put(parent_request);
2511 obj_request->result = result;
2512 obj_request->xferred = 0;
2513 obj_request_done_set(obj_request);
2518 static void rbd_img_obj_exists_callback(struct rbd_obj_request *obj_request)
2520 struct rbd_obj_request *orig_request;
2521 struct rbd_device *rbd_dev;
2524 rbd_assert(!obj_request_img_data_test(obj_request));
2527 * All we need from the object request is the original
2528 * request and the result of the STAT op. Grab those, then
2529 * we're done with the request.
2531 orig_request = obj_request->obj_request;
2532 obj_request->obj_request = NULL;
2533 rbd_obj_request_put(orig_request);
2534 rbd_assert(orig_request);
2535 rbd_assert(orig_request->img_request);
2537 result = obj_request->result;
2538 obj_request->result = 0;
2540 dout("%s: obj %p for obj %p result %d %llu/%llu\n", __func__,
2541 obj_request, orig_request, result,
2542 obj_request->xferred, obj_request->length);
2543 rbd_obj_request_put(obj_request);
2546 * If the overlap has become 0 (most likely because the
2547 * image has been flattened) we need to free the pages
2548 * and re-submit the original write request.
2550 rbd_dev = orig_request->img_request->rbd_dev;
2551 if (!rbd_dev->parent_overlap) {
2552 struct ceph_osd_client *osdc;
2554 osdc = &rbd_dev->rbd_client->client->osdc;
2555 result = rbd_obj_request_submit(osdc, orig_request);
2561 * Our only purpose here is to determine whether the object
2562 * exists, and we don't want to treat the non-existence as
2563 * an error. If something else comes back, transfer the
2564 * error to the original request and complete it now.
2567 obj_request_existence_set(orig_request, true);
2568 } else if (result == -ENOENT) {
2569 obj_request_existence_set(orig_request, false);
2570 } else if (result) {
2571 orig_request->result = result;
2576 * Resubmit the original request now that we have recorded
2577 * whether the target object exists.
2579 orig_request->result = rbd_img_obj_request_submit(orig_request);
2581 if (orig_request->result)
2582 rbd_obj_request_complete(orig_request);
2585 static int rbd_img_obj_exists_submit(struct rbd_obj_request *obj_request)
2587 struct rbd_obj_request *stat_request;
2588 struct rbd_device *rbd_dev;
2589 struct ceph_osd_client *osdc;
2590 struct page **pages = NULL;
2596 * The response data for a STAT call consists of:
2603 size = sizeof (__le64) + sizeof (__le32) + sizeof (__le32);
2604 page_count = (u32)calc_pages_for(0, size);
2605 pages = ceph_alloc_page_vector(page_count, GFP_KERNEL);
2607 return PTR_ERR(pages);
2610 stat_request = rbd_obj_request_create(obj_request->object_name, 0, 0,
2615 rbd_obj_request_get(obj_request);
2616 stat_request->obj_request = obj_request;
2617 stat_request->pages = pages;
2618 stat_request->page_count = page_count;
2620 rbd_assert(obj_request->img_request);
2621 rbd_dev = obj_request->img_request->rbd_dev;
2622 stat_request->osd_req = rbd_osd_req_create(rbd_dev, false,
2624 if (!stat_request->osd_req)
2626 stat_request->callback = rbd_img_obj_exists_callback;
2628 osd_req_op_init(stat_request->osd_req, 0, CEPH_OSD_OP_STAT);
2629 osd_req_op_raw_data_in_pages(stat_request->osd_req, 0, pages, size, 0,
2631 rbd_osd_req_format_read(stat_request);
2633 osdc = &rbd_dev->rbd_client->client->osdc;
2634 ret = rbd_obj_request_submit(osdc, stat_request);
2637 rbd_obj_request_put(obj_request);
2642 static int rbd_img_obj_request_submit(struct rbd_obj_request *obj_request)
2644 struct rbd_img_request *img_request;
2645 struct rbd_device *rbd_dev;
2648 rbd_assert(obj_request_img_data_test(obj_request));
2650 img_request = obj_request->img_request;
2651 rbd_assert(img_request);
2652 rbd_dev = img_request->rbd_dev;
2655 * Only writes to layered images need special handling.
2656 * Reads and non-layered writes are simple object requests.
2657 * Layered writes that start beyond the end of the overlap
2658 * with the parent have no parent data, so they too are
2659 * simple object requests. Finally, if the target object is
2660 * known to already exist, its parent data has already been
2661 * copied, so a write to the object can also be handled as a
2662 * simple object request.
2664 if (!img_request_write_test(img_request) ||
2665 !img_request_layered_test(img_request) ||
2666 rbd_dev->parent_overlap <= obj_request->img_offset ||
2667 ((known = obj_request_known_test(obj_request)) &&
2668 obj_request_exists_test(obj_request))) {
2670 struct rbd_device *rbd_dev;
2671 struct ceph_osd_client *osdc;
2673 rbd_dev = obj_request->img_request->rbd_dev;
2674 osdc = &rbd_dev->rbd_client->client->osdc;
2676 return rbd_obj_request_submit(osdc, obj_request);
2680 * It's a layered write. The target object might exist but
2681 * we may not know that yet. If we know it doesn't exist,
2682 * start by reading the data for the full target object from
2683 * the parent so we can use it for a copyup to the target.
2686 return rbd_img_obj_parent_read_full(obj_request);
2688 /* We don't know whether the target exists. Go find out. */
2690 return rbd_img_obj_exists_submit(obj_request);
2693 static int rbd_img_request_submit(struct rbd_img_request *img_request)
2695 struct rbd_obj_request *obj_request;
2696 struct rbd_obj_request *next_obj_request;
2698 dout("%s: img %p\n", __func__, img_request);
2699 for_each_obj_request_safe(img_request, obj_request, next_obj_request) {
2702 ret = rbd_img_obj_request_submit(obj_request);
2710 static void rbd_img_parent_read_callback(struct rbd_img_request *img_request)
2712 struct rbd_obj_request *obj_request;
2713 struct rbd_device *rbd_dev;
2718 rbd_assert(img_request_child_test(img_request));
2720 /* First get what we need from the image request and release it */
2722 obj_request = img_request->obj_request;
2723 img_xferred = img_request->xferred;
2724 img_result = img_request->result;
2725 rbd_img_request_put(img_request);
2728 * If the overlap has become 0 (most likely because the
2729 * image has been flattened) we need to re-submit the
2732 rbd_assert(obj_request);
2733 rbd_assert(obj_request->img_request);
2734 rbd_dev = obj_request->img_request->rbd_dev;
2735 if (!rbd_dev->parent_overlap) {
2736 struct ceph_osd_client *osdc;
2738 osdc = &rbd_dev->rbd_client->client->osdc;
2739 img_result = rbd_obj_request_submit(osdc, obj_request);
2744 obj_request->result = img_result;
2745 if (obj_request->result)
2749 * We need to zero anything beyond the parent overlap
2750 * boundary. Since rbd_img_obj_request_read_callback()
2751 * will zero anything beyond the end of a short read, an
2752 * easy way to do this is to pretend the data from the
2753 * parent came up short--ending at the overlap boundary.
2755 rbd_assert(obj_request->img_offset < U64_MAX - obj_request->length);
2756 obj_end = obj_request->img_offset + obj_request->length;
2757 if (obj_end > rbd_dev->parent_overlap) {
2760 if (obj_request->img_offset < rbd_dev->parent_overlap)
2761 xferred = rbd_dev->parent_overlap -
2762 obj_request->img_offset;
2764 obj_request->xferred = min(img_xferred, xferred);
2766 obj_request->xferred = img_xferred;
2769 rbd_img_obj_request_read_callback(obj_request);
2770 rbd_obj_request_complete(obj_request);
2773 static void rbd_img_parent_read(struct rbd_obj_request *obj_request)
2775 struct rbd_img_request *img_request;
2778 rbd_assert(obj_request_img_data_test(obj_request));
2779 rbd_assert(obj_request->img_request != NULL);
2780 rbd_assert(obj_request->result == (s32) -ENOENT);
2781 rbd_assert(obj_request_type_valid(obj_request->type));
2783 /* rbd_read_finish(obj_request, obj_request->length); */
2784 img_request = rbd_parent_request_create(obj_request,
2785 obj_request->img_offset,
2786 obj_request->length);
2791 if (obj_request->type == OBJ_REQUEST_BIO)
2792 result = rbd_img_request_fill(img_request, OBJ_REQUEST_BIO,
2793 obj_request->bio_list);
2795 result = rbd_img_request_fill(img_request, OBJ_REQUEST_PAGES,
2796 obj_request->pages);
2800 img_request->callback = rbd_img_parent_read_callback;
2801 result = rbd_img_request_submit(img_request);
2808 rbd_img_request_put(img_request);
2809 obj_request->result = result;
2810 obj_request->xferred = 0;
2811 obj_request_done_set(obj_request);
2814 static int rbd_obj_notify_ack(struct rbd_device *rbd_dev, u64 notify_id)
2816 struct rbd_obj_request *obj_request;
2817 struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
2820 obj_request = rbd_obj_request_create(rbd_dev->header_name, 0, 0,
2821 OBJ_REQUEST_NODATA);
2826 obj_request->osd_req = rbd_osd_req_create(rbd_dev, false, obj_request);
2827 if (!obj_request->osd_req)
2829 obj_request->callback = rbd_obj_request_put;
2831 osd_req_op_watch_init(obj_request->osd_req, 0, CEPH_OSD_OP_NOTIFY_ACK,
2833 rbd_osd_req_format_read(obj_request);
2835 ret = rbd_obj_request_submit(osdc, obj_request);
2838 rbd_obj_request_put(obj_request);
2843 static void rbd_watch_cb(u64 ver, u64 notify_id, u8 opcode, void *data)
2845 struct rbd_device *rbd_dev = (struct rbd_device *)data;
2851 dout("%s: \"%s\" notify_id %llu opcode %u\n", __func__,
2852 rbd_dev->header_name, (unsigned long long)notify_id,
2853 (unsigned int)opcode);
2854 ret = rbd_dev_refresh(rbd_dev);
2856 rbd_warn(rbd_dev, "header refresh error (%d)\n", ret);
2858 rbd_obj_notify_ack(rbd_dev, notify_id);
2862 * Request sync osd watch/unwatch. The value of "start" determines
2863 * whether a watch request is being initiated or torn down.
2865 static int rbd_dev_header_watch_sync(struct rbd_device *rbd_dev, bool start)
2867 struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
2868 struct rbd_obj_request *obj_request;
2871 rbd_assert(start ^ !!rbd_dev->watch_event);
2872 rbd_assert(start ^ !!rbd_dev->watch_request);
2875 ret = ceph_osdc_create_event(osdc, rbd_watch_cb, rbd_dev,
2876 &rbd_dev->watch_event);
2879 rbd_assert(rbd_dev->watch_event != NULL);
2883 obj_request = rbd_obj_request_create(rbd_dev->header_name, 0, 0,
2884 OBJ_REQUEST_NODATA);
2888 obj_request->osd_req = rbd_osd_req_create(rbd_dev, true, obj_request);
2889 if (!obj_request->osd_req)
2893 ceph_osdc_set_request_linger(osdc, obj_request->osd_req);
2895 ceph_osdc_unregister_linger_request(osdc,
2896 rbd_dev->watch_request->osd_req);
2898 osd_req_op_watch_init(obj_request->osd_req, 0, CEPH_OSD_OP_WATCH,
2899 rbd_dev->watch_event->cookie, 0, start ? 1 : 0);
2900 rbd_osd_req_format_write(obj_request);
2902 ret = rbd_obj_request_submit(osdc, obj_request);
2905 ret = rbd_obj_request_wait(obj_request);
2908 ret = obj_request->result;
2913 * A watch request is set to linger, so the underlying osd
2914 * request won't go away until we unregister it. We retain
2915 * a pointer to the object request during that time (in
2916 * rbd_dev->watch_request), so we'll keep a reference to
2917 * it. We'll drop that reference (below) after we've
2921 rbd_dev->watch_request = obj_request;
2926 /* We have successfully torn down the watch request */
2928 rbd_obj_request_put(rbd_dev->watch_request);
2929 rbd_dev->watch_request = NULL;
2931 /* Cancel the event if we're tearing down, or on error */
2932 ceph_osdc_cancel_event(rbd_dev->watch_event);
2933 rbd_dev->watch_event = NULL;
2935 rbd_obj_request_put(obj_request);
2941 * Synchronous osd object method call. Returns the number of bytes
2942 * returned in the outbound buffer, or a negative error code.
2944 static int rbd_obj_method_sync(struct rbd_device *rbd_dev,
2945 const char *object_name,
2946 const char *class_name,
2947 const char *method_name,
2948 const void *outbound,
2949 size_t outbound_size,
2951 size_t inbound_size)
2953 struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
2954 struct rbd_obj_request *obj_request;
2955 struct page **pages;
2960 * Method calls are ultimately read operations. The result
2961 * should placed into the inbound buffer provided. They
2962 * also supply outbound data--parameters for the object
2963 * method. Currently if this is present it will be a
2966 page_count = (u32)calc_pages_for(0, inbound_size);
2967 pages = ceph_alloc_page_vector(page_count, GFP_KERNEL);
2969 return PTR_ERR(pages);
2972 obj_request = rbd_obj_request_create(object_name, 0, inbound_size,
2977 obj_request->pages = pages;
2978 obj_request->page_count = page_count;
2980 obj_request->osd_req = rbd_osd_req_create(rbd_dev, false, obj_request);
2981 if (!obj_request->osd_req)
2984 osd_req_op_cls_init(obj_request->osd_req, 0, CEPH_OSD_OP_CALL,
2985 class_name, method_name);
2986 if (outbound_size) {
2987 struct ceph_pagelist *pagelist;
2989 pagelist = kmalloc(sizeof (*pagelist), GFP_NOFS);
2993 ceph_pagelist_init(pagelist);
2994 ceph_pagelist_append(pagelist, outbound, outbound_size);
2995 osd_req_op_cls_request_data_pagelist(obj_request->osd_req, 0,
2998 osd_req_op_cls_response_data_pages(obj_request->osd_req, 0,
2999 obj_request->pages, inbound_size,
3001 rbd_osd_req_format_read(obj_request);
3003 ret = rbd_obj_request_submit(osdc, obj_request);
3006 ret = rbd_obj_request_wait(obj_request);
3010 ret = obj_request->result;
3014 rbd_assert(obj_request->xferred < (u64)INT_MAX);
3015 ret = (int)obj_request->xferred;
3016 ceph_copy_from_page_vector(pages, inbound, 0, obj_request->xferred);
3019 rbd_obj_request_put(obj_request);
3021 ceph_release_page_vector(pages, page_count);
3026 static void rbd_request_fn(struct request_queue *q)
3027 __releases(q->queue_lock) __acquires(q->queue_lock)
3029 struct rbd_device *rbd_dev = q->queuedata;
3030 bool read_only = rbd_dev->mapping.read_only;
3034 while ((rq = blk_fetch_request(q))) {
3035 bool write_request = rq_data_dir(rq) == WRITE;
3036 struct rbd_img_request *img_request;
3040 /* Ignore any non-FS requests that filter through. */
3042 if (rq->cmd_type != REQ_TYPE_FS) {
3043 dout("%s: non-fs request type %d\n", __func__,
3044 (int) rq->cmd_type);
3045 __blk_end_request_all(rq, 0);
3049 /* Ignore/skip any zero-length requests */
3051 offset = (u64) blk_rq_pos(rq) << SECTOR_SHIFT;
3052 length = (u64) blk_rq_bytes(rq);
3055 dout("%s: zero-length request\n", __func__);
3056 __blk_end_request_all(rq, 0);
3060 spin_unlock_irq(q->queue_lock);
3062 /* Disallow writes to a read-only device */
3064 if (write_request) {
3068 rbd_assert(rbd_dev->spec->snap_id == CEPH_NOSNAP);
3072 * Quit early if the mapped snapshot no longer
3073 * exists. It's still possible the snapshot will
3074 * have disappeared by the time our request arrives
3075 * at the osd, but there's no sense in sending it if
3078 if (!test_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags)) {
3079 dout("request for non-existent snapshot");
3080 rbd_assert(rbd_dev->spec->snap_id != CEPH_NOSNAP);
3086 if (offset && length > U64_MAX - offset + 1) {
3087 rbd_warn(rbd_dev, "bad request range (%llu~%llu)\n",
3089 goto end_request; /* Shouldn't happen */
3093 if (offset + length > rbd_dev->mapping.size) {
3094 rbd_warn(rbd_dev, "beyond EOD (%llu~%llu > %llu)\n",
3095 offset, length, rbd_dev->mapping.size);
3100 img_request = rbd_img_request_create(rbd_dev, offset, length,
3105 img_request->rq = rq;
3107 result = rbd_img_request_fill(img_request, OBJ_REQUEST_BIO,
3110 result = rbd_img_request_submit(img_request);
3112 rbd_img_request_put(img_request);
3114 spin_lock_irq(q->queue_lock);
3116 rbd_warn(rbd_dev, "%s %llx at %llx result %d\n",
3117 write_request ? "write" : "read",
3118 length, offset, result);
3120 __blk_end_request_all(rq, result);
3126 * a queue callback. Makes sure that we don't create a bio that spans across
3127 * multiple osd objects. One exception would be with a single page bios,
3128 * which we handle later at bio_chain_clone_range()
3130 static int rbd_merge_bvec(struct request_queue *q, struct bvec_merge_data *bmd,
3131 struct bio_vec *bvec)
3133 struct rbd_device *rbd_dev = q->queuedata;
3134 sector_t sector_offset;
3135 sector_t sectors_per_obj;
3136 sector_t obj_sector_offset;
3140 * Find how far into its rbd object the partition-relative
3141 * bio start sector is to offset relative to the enclosing
3144 sector_offset = get_start_sect(bmd->bi_bdev) + bmd->bi_sector;
3145 sectors_per_obj = 1 << (rbd_dev->header.obj_order - SECTOR_SHIFT);
3146 obj_sector_offset = sector_offset & (sectors_per_obj - 1);
3149 * Compute the number of bytes from that offset to the end
3150 * of the object. Account for what's already used by the bio.
3152 ret = (int) (sectors_per_obj - obj_sector_offset) << SECTOR_SHIFT;
3153 if (ret > bmd->bi_size)
3154 ret -= bmd->bi_size;
3159 * Don't send back more than was asked for. And if the bio
3160 * was empty, let the whole thing through because: "Note
3161 * that a block device *must* allow a single page to be
3162 * added to an empty bio."
3164 rbd_assert(bvec->bv_len <= PAGE_SIZE);
3165 if (ret > (int) bvec->bv_len || !bmd->bi_size)
3166 ret = (int) bvec->bv_len;
3171 static void rbd_free_disk(struct rbd_device *rbd_dev)
3173 struct gendisk *disk = rbd_dev->disk;
3178 rbd_dev->disk = NULL;
3179 if (disk->flags & GENHD_FL_UP) {
3182 blk_cleanup_queue(disk->queue);
3187 static int rbd_obj_read_sync(struct rbd_device *rbd_dev,
3188 const char *object_name,
3189 u64 offset, u64 length, void *buf)
3192 struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
3193 struct rbd_obj_request *obj_request;
3194 struct page **pages = NULL;
3199 page_count = (u32) calc_pages_for(offset, length);
3200 pages = ceph_alloc_page_vector(page_count, GFP_KERNEL);
3202 ret = PTR_ERR(pages);
3205 obj_request = rbd_obj_request_create(object_name, offset, length,
3210 obj_request->pages = pages;
3211 obj_request->page_count = page_count;
3213 obj_request->osd_req = rbd_osd_req_create(rbd_dev, false, obj_request);
3214 if (!obj_request->osd_req)
3217 osd_req_op_extent_init(obj_request->osd_req, 0, CEPH_OSD_OP_READ,
3218 offset, length, 0, 0);
3219 osd_req_op_extent_osd_data_pages(obj_request->osd_req, 0,
3221 obj_request->length,
3222 obj_request->offset & ~PAGE_MASK,
3224 rbd_osd_req_format_read(obj_request);
3226 ret = rbd_obj_request_submit(osdc, obj_request);
3229 ret = rbd_obj_request_wait(obj_request);
3233 ret = obj_request->result;
3237 rbd_assert(obj_request->xferred <= (u64) SIZE_MAX);
3238 size = (size_t) obj_request->xferred;
3239 ceph_copy_from_page_vector(pages, buf, 0, size);
3240 rbd_assert(size <= (size_t)INT_MAX);
3244 rbd_obj_request_put(obj_request);
3246 ceph_release_page_vector(pages, page_count);
3252 * Read the complete header for the given rbd device. On successful
3253 * return, the rbd_dev->header field will contain up-to-date
3254 * information about the image.
3256 static int rbd_dev_v1_header_info(struct rbd_device *rbd_dev)
3258 struct rbd_image_header_ondisk *ondisk = NULL;
3265 * The complete header will include an array of its 64-bit
3266 * snapshot ids, followed by the names of those snapshots as
3267 * a contiguous block of NUL-terminated strings. Note that
3268 * the number of snapshots could change by the time we read
3269 * it in, in which case we re-read it.
3276 size = sizeof (*ondisk);
3277 size += snap_count * sizeof (struct rbd_image_snap_ondisk);
3279 ondisk = kmalloc(size, GFP_KERNEL);
3283 ret = rbd_obj_read_sync(rbd_dev, rbd_dev->header_name,
3287 if ((size_t)ret < size) {
3289 rbd_warn(rbd_dev, "short header read (want %zd got %d)",
3293 if (!rbd_dev_ondisk_valid(ondisk)) {
3295 rbd_warn(rbd_dev, "invalid header");
3299 names_size = le64_to_cpu(ondisk->snap_names_len);
3300 want_count = snap_count;
3301 snap_count = le32_to_cpu(ondisk->snap_count);
3302 } while (snap_count != want_count);
3304 ret = rbd_header_from_disk(rbd_dev, ondisk);
3312 * Clear the rbd device's EXISTS flag if the snapshot it's mapped to
3313 * has disappeared from the (just updated) snapshot context.
3315 static void rbd_exists_validate(struct rbd_device *rbd_dev)
3319 if (!test_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags))
3322 snap_id = rbd_dev->spec->snap_id;
3323 if (snap_id == CEPH_NOSNAP)
3326 if (rbd_dev_snap_index(rbd_dev, snap_id) == BAD_SNAP_INDEX)
3327 clear_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags);
3330 static int rbd_dev_refresh(struct rbd_device *rbd_dev)
3335 rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
3336 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
3337 mapping_size = rbd_dev->mapping.size;
3338 if (rbd_dev->image_format == 1)
3339 ret = rbd_dev_v1_header_info(rbd_dev);
3341 ret = rbd_dev_v2_header_info(rbd_dev);
3343 /* If it's a mapped snapshot, validate its EXISTS flag */
3345 rbd_exists_validate(rbd_dev);
3346 mutex_unlock(&ctl_mutex);
3347 if (mapping_size != rbd_dev->mapping.size) {
3350 size = (sector_t)rbd_dev->mapping.size / SECTOR_SIZE;
3351 dout("setting size to %llu sectors", (unsigned long long)size);
3352 set_capacity(rbd_dev->disk, size);
3353 revalidate_disk(rbd_dev->disk);
3359 static int rbd_init_disk(struct rbd_device *rbd_dev)
3361 struct gendisk *disk;
3362 struct request_queue *q;
3365 /* create gendisk info */
3366 disk = alloc_disk(RBD_MINORS_PER_MAJOR);
3370 snprintf(disk->disk_name, sizeof(disk->disk_name), RBD_DRV_NAME "%d",
3372 disk->major = rbd_dev->major;
3373 disk->first_minor = 0;
3374 disk->fops = &rbd_bd_ops;
3375 disk->private_data = rbd_dev;
3377 q = blk_init_queue(rbd_request_fn, &rbd_dev->lock);
3381 /* We use the default size, but let's be explicit about it. */
3382 blk_queue_physical_block_size(q, SECTOR_SIZE);
3384 /* set io sizes to object size */
3385 segment_size = rbd_obj_bytes(&rbd_dev->header);
3386 blk_queue_max_hw_sectors(q, segment_size / SECTOR_SIZE);
3387 blk_queue_max_segment_size(q, segment_size);
3388 blk_queue_io_min(q, segment_size);
3389 blk_queue_io_opt(q, segment_size);
3391 blk_queue_merge_bvec(q, rbd_merge_bvec);
3394 q->queuedata = rbd_dev;
3396 rbd_dev->disk = disk;
3409 static struct rbd_device *dev_to_rbd_dev(struct device *dev)
3411 return container_of(dev, struct rbd_device, dev);
3414 static ssize_t rbd_size_show(struct device *dev,
3415 struct device_attribute *attr, char *buf)
3417 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3419 return sprintf(buf, "%llu\n",
3420 (unsigned long long)rbd_dev->mapping.size);
3424 * Note this shows the features for whatever's mapped, which is not
3425 * necessarily the base image.
3427 static ssize_t rbd_features_show(struct device *dev,
3428 struct device_attribute *attr, char *buf)
3430 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3432 return sprintf(buf, "0x%016llx\n",
3433 (unsigned long long)rbd_dev->mapping.features);
3436 static ssize_t rbd_major_show(struct device *dev,
3437 struct device_attribute *attr, char *buf)
3439 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3442 return sprintf(buf, "%d\n", rbd_dev->major);
3444 return sprintf(buf, "(none)\n");
3448 static ssize_t rbd_client_id_show(struct device *dev,
3449 struct device_attribute *attr, char *buf)
3451 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3453 return sprintf(buf, "client%lld\n",
3454 ceph_client_id(rbd_dev->rbd_client->client));
3457 static ssize_t rbd_pool_show(struct device *dev,
3458 struct device_attribute *attr, char *buf)
3460 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3462 return sprintf(buf, "%s\n", rbd_dev->spec->pool_name);
3465 static ssize_t rbd_pool_id_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, "%llu\n",
3471 (unsigned long long) rbd_dev->spec->pool_id);
3474 static ssize_t rbd_name_show(struct device *dev,
3475 struct device_attribute *attr, char *buf)
3477 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3479 if (rbd_dev->spec->image_name)
3480 return sprintf(buf, "%s\n", rbd_dev->spec->image_name);
3482 return sprintf(buf, "(unknown)\n");
3485 static ssize_t rbd_image_id_show(struct device *dev,
3486 struct device_attribute *attr, char *buf)
3488 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3490 return sprintf(buf, "%s\n", rbd_dev->spec->image_id);
3494 * Shows the name of the currently-mapped snapshot (or
3495 * RBD_SNAP_HEAD_NAME for the base image).
3497 static ssize_t rbd_snap_show(struct device *dev,
3498 struct device_attribute *attr,
3501 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3503 return sprintf(buf, "%s\n", rbd_dev->spec->snap_name);
3507 * For an rbd v2 image, shows the pool id, image id, and snapshot id
3508 * for the parent image. If there is no parent, simply shows
3509 * "(no parent image)".
3511 static ssize_t rbd_parent_show(struct device *dev,
3512 struct device_attribute *attr,
3515 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3516 struct rbd_spec *spec = rbd_dev->parent_spec;
3521 return sprintf(buf, "(no parent image)\n");
3523 count = sprintf(bufp, "pool_id %llu\npool_name %s\n",
3524 (unsigned long long) spec->pool_id, spec->pool_name);
3529 count = sprintf(bufp, "image_id %s\nimage_name %s\n", spec->image_id,
3530 spec->image_name ? spec->image_name : "(unknown)");
3535 count = sprintf(bufp, "snap_id %llu\nsnap_name %s\n",
3536 (unsigned long long) spec->snap_id, spec->snap_name);
3541 count = sprintf(bufp, "overlap %llu\n", rbd_dev->parent_overlap);
3546 return (ssize_t) (bufp - buf);
3549 static ssize_t rbd_image_refresh(struct device *dev,
3550 struct device_attribute *attr,
3554 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3557 ret = rbd_dev_refresh(rbd_dev);
3559 rbd_warn(rbd_dev, ": manual header refresh error (%d)\n", ret);
3561 return ret < 0 ? ret : size;
3564 static DEVICE_ATTR(size, S_IRUGO, rbd_size_show, NULL);
3565 static DEVICE_ATTR(features, S_IRUGO, rbd_features_show, NULL);
3566 static DEVICE_ATTR(major, S_IRUGO, rbd_major_show, NULL);
3567 static DEVICE_ATTR(client_id, S_IRUGO, rbd_client_id_show, NULL);
3568 static DEVICE_ATTR(pool, S_IRUGO, rbd_pool_show, NULL);
3569 static DEVICE_ATTR(pool_id, S_IRUGO, rbd_pool_id_show, NULL);
3570 static DEVICE_ATTR(name, S_IRUGO, rbd_name_show, NULL);
3571 static DEVICE_ATTR(image_id, S_IRUGO, rbd_image_id_show, NULL);
3572 static DEVICE_ATTR(refresh, S_IWUSR, NULL, rbd_image_refresh);
3573 static DEVICE_ATTR(current_snap, S_IRUGO, rbd_snap_show, NULL);
3574 static DEVICE_ATTR(parent, S_IRUGO, rbd_parent_show, NULL);
3576 static struct attribute *rbd_attrs[] = {
3577 &dev_attr_size.attr,
3578 &dev_attr_features.attr,
3579 &dev_attr_major.attr,
3580 &dev_attr_client_id.attr,
3581 &dev_attr_pool.attr,
3582 &dev_attr_pool_id.attr,
3583 &dev_attr_name.attr,
3584 &dev_attr_image_id.attr,
3585 &dev_attr_current_snap.attr,
3586 &dev_attr_parent.attr,
3587 &dev_attr_refresh.attr,
3591 static struct attribute_group rbd_attr_group = {
3595 static const struct attribute_group *rbd_attr_groups[] = {
3600 static void rbd_sysfs_dev_release(struct device *dev)
3604 static struct device_type rbd_device_type = {
3606 .groups = rbd_attr_groups,
3607 .release = rbd_sysfs_dev_release,
3610 static struct rbd_spec *rbd_spec_get(struct rbd_spec *spec)
3612 kref_get(&spec->kref);
3617 static void rbd_spec_free(struct kref *kref);
3618 static void rbd_spec_put(struct rbd_spec *spec)
3621 kref_put(&spec->kref, rbd_spec_free);
3624 static struct rbd_spec *rbd_spec_alloc(void)
3626 struct rbd_spec *spec;
3628 spec = kzalloc(sizeof (*spec), GFP_KERNEL);
3631 kref_init(&spec->kref);
3636 static void rbd_spec_free(struct kref *kref)
3638 struct rbd_spec *spec = container_of(kref, struct rbd_spec, kref);
3640 kfree(spec->pool_name);
3641 kfree(spec->image_id);
3642 kfree(spec->image_name);
3643 kfree(spec->snap_name);
3647 static struct rbd_device *rbd_dev_create(struct rbd_client *rbdc,
3648 struct rbd_spec *spec)
3650 struct rbd_device *rbd_dev;
3652 rbd_dev = kzalloc(sizeof (*rbd_dev), GFP_KERNEL);
3656 spin_lock_init(&rbd_dev->lock);
3658 atomic_set(&rbd_dev->parent_ref, 0);
3659 INIT_LIST_HEAD(&rbd_dev->node);
3660 init_rwsem(&rbd_dev->header_rwsem);
3662 rbd_dev->spec = spec;
3663 rbd_dev->rbd_client = rbdc;
3665 /* Initialize the layout used for all rbd requests */
3667 rbd_dev->layout.fl_stripe_unit = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER);
3668 rbd_dev->layout.fl_stripe_count = cpu_to_le32(1);
3669 rbd_dev->layout.fl_object_size = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER);
3670 rbd_dev->layout.fl_pg_pool = cpu_to_le32((u32) spec->pool_id);
3675 static void rbd_dev_destroy(struct rbd_device *rbd_dev)
3677 rbd_put_client(rbd_dev->rbd_client);
3678 rbd_spec_put(rbd_dev->spec);
3683 * Get the size and object order for an image snapshot, or if
3684 * snap_id is CEPH_NOSNAP, gets this information for the base
3687 static int _rbd_dev_v2_snap_size(struct rbd_device *rbd_dev, u64 snap_id,
3688 u8 *order, u64 *snap_size)
3690 __le64 snapid = cpu_to_le64(snap_id);
3695 } __attribute__ ((packed)) size_buf = { 0 };
3697 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
3699 &snapid, sizeof (snapid),
3700 &size_buf, sizeof (size_buf));
3701 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
3704 if (ret < sizeof (size_buf))
3708 *order = size_buf.order;
3709 *snap_size = le64_to_cpu(size_buf.size);
3711 dout(" snap_id 0x%016llx order = %u, snap_size = %llu\n",
3712 (unsigned long long)snap_id, (unsigned int)*order,
3713 (unsigned long long)*snap_size);
3718 static int rbd_dev_v2_image_size(struct rbd_device *rbd_dev)
3720 return _rbd_dev_v2_snap_size(rbd_dev, CEPH_NOSNAP,
3721 &rbd_dev->header.obj_order,
3722 &rbd_dev->header.image_size);
3725 static int rbd_dev_v2_object_prefix(struct rbd_device *rbd_dev)
3731 reply_buf = kzalloc(RBD_OBJ_PREFIX_LEN_MAX, GFP_KERNEL);
3735 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
3736 "rbd", "get_object_prefix", NULL, 0,
3737 reply_buf, RBD_OBJ_PREFIX_LEN_MAX);
3738 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
3743 rbd_dev->header.object_prefix = ceph_extract_encoded_string(&p,
3744 p + ret, NULL, GFP_NOIO);
3747 if (IS_ERR(rbd_dev->header.object_prefix)) {
3748 ret = PTR_ERR(rbd_dev->header.object_prefix);
3749 rbd_dev->header.object_prefix = NULL;
3751 dout(" object_prefix = %s\n", rbd_dev->header.object_prefix);
3759 static int _rbd_dev_v2_snap_features(struct rbd_device *rbd_dev, u64 snap_id,
3762 __le64 snapid = cpu_to_le64(snap_id);
3766 } __attribute__ ((packed)) features_buf = { 0 };
3770 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
3771 "rbd", "get_features",
3772 &snapid, sizeof (snapid),
3773 &features_buf, sizeof (features_buf));
3774 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
3777 if (ret < sizeof (features_buf))
3780 incompat = le64_to_cpu(features_buf.incompat);
3781 if (incompat & ~RBD_FEATURES_SUPPORTED)
3784 *snap_features = le64_to_cpu(features_buf.features);
3786 dout(" snap_id 0x%016llx features = 0x%016llx incompat = 0x%016llx\n",
3787 (unsigned long long)snap_id,
3788 (unsigned long long)*snap_features,
3789 (unsigned long long)le64_to_cpu(features_buf.incompat));
3794 static int rbd_dev_v2_features(struct rbd_device *rbd_dev)
3796 return _rbd_dev_v2_snap_features(rbd_dev, CEPH_NOSNAP,
3797 &rbd_dev->header.features);
3800 static int rbd_dev_v2_parent_info(struct rbd_device *rbd_dev)
3802 struct rbd_spec *parent_spec;
3804 void *reply_buf = NULL;
3814 parent_spec = rbd_spec_alloc();
3818 size = sizeof (__le64) + /* pool_id */
3819 sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX + /* image_id */
3820 sizeof (__le64) + /* snap_id */
3821 sizeof (__le64); /* overlap */
3822 reply_buf = kmalloc(size, GFP_KERNEL);
3828 snapid = cpu_to_le64(CEPH_NOSNAP);
3829 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
3830 "rbd", "get_parent",
3831 &snapid, sizeof (snapid),
3833 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
3838 end = reply_buf + ret;
3840 ceph_decode_64_safe(&p, end, pool_id, out_err);
3841 if (pool_id == CEPH_NOPOOL) {
3843 * Either the parent never existed, or we have
3844 * record of it but the image got flattened so it no
3845 * longer has a parent. When the parent of a
3846 * layered image disappears we immediately set the
3847 * overlap to 0. The effect of this is that all new
3848 * requests will be treated as if the image had no
3851 if (rbd_dev->parent_overlap) {
3852 rbd_dev->parent_overlap = 0;
3854 rbd_dev_parent_put(rbd_dev);
3855 pr_info("%s: clone image has been flattened\n",
3856 rbd_dev->disk->disk_name);
3859 goto out; /* No parent? No problem. */
3862 /* The ceph file layout needs to fit pool id in 32 bits */
3865 if (pool_id > (u64)U32_MAX) {
3866 rbd_warn(NULL, "parent pool id too large (%llu > %u)\n",
3867 (unsigned long long)pool_id, U32_MAX);
3871 image_id = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL);
3872 if (IS_ERR(image_id)) {
3873 ret = PTR_ERR(image_id);
3876 ceph_decode_64_safe(&p, end, snap_id, out_err);
3877 ceph_decode_64_safe(&p, end, overlap, out_err);
3880 * The parent won't change (except when the clone is
3881 * flattened, already handled that). So we only need to
3882 * record the parent spec we have not already done so.
3884 if (!rbd_dev->parent_spec) {
3885 parent_spec->pool_id = pool_id;
3886 parent_spec->image_id = image_id;
3887 parent_spec->snap_id = snap_id;
3888 rbd_dev->parent_spec = parent_spec;
3889 parent_spec = NULL; /* rbd_dev now owns this */
3893 * We always update the parent overlap. If it's zero we
3894 * treat it specially.
3896 rbd_dev->parent_overlap = overlap;
3900 /* A null parent_spec indicates it's the initial probe */
3904 * The overlap has become zero, so the clone
3905 * must have been resized down to 0 at some
3906 * point. Treat this the same as a flatten.
3908 rbd_dev_parent_put(rbd_dev);
3909 pr_info("%s: clone image now standalone\n",
3910 rbd_dev->disk->disk_name);
3913 * For the initial probe, if we find the
3914 * overlap is zero we just pretend there was
3917 rbd_warn(rbd_dev, "ignoring parent of "
3918 "clone with overlap 0\n");
3925 rbd_spec_put(parent_spec);
3930 static int rbd_dev_v2_striping_info(struct rbd_device *rbd_dev)
3934 __le64 stripe_count;
3935 } __attribute__ ((packed)) striping_info_buf = { 0 };
3936 size_t size = sizeof (striping_info_buf);
3943 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
3944 "rbd", "get_stripe_unit_count", NULL, 0,
3945 (char *)&striping_info_buf, size);
3946 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
3953 * We don't actually support the "fancy striping" feature
3954 * (STRIPINGV2) yet, but if the striping sizes are the
3955 * defaults the behavior is the same as before. So find
3956 * out, and only fail if the image has non-default values.
3959 obj_size = (u64)1 << rbd_dev->header.obj_order;
3960 p = &striping_info_buf;
3961 stripe_unit = ceph_decode_64(&p);
3962 if (stripe_unit != obj_size) {
3963 rbd_warn(rbd_dev, "unsupported stripe unit "
3964 "(got %llu want %llu)",
3965 stripe_unit, obj_size);
3968 stripe_count = ceph_decode_64(&p);
3969 if (stripe_count != 1) {
3970 rbd_warn(rbd_dev, "unsupported stripe count "
3971 "(got %llu want 1)", stripe_count);
3974 rbd_dev->header.stripe_unit = stripe_unit;
3975 rbd_dev->header.stripe_count = stripe_count;
3980 static char *rbd_dev_image_name(struct rbd_device *rbd_dev)
3982 size_t image_id_size;
3987 void *reply_buf = NULL;
3989 char *image_name = NULL;
3992 rbd_assert(!rbd_dev->spec->image_name);
3994 len = strlen(rbd_dev->spec->image_id);
3995 image_id_size = sizeof (__le32) + len;
3996 image_id = kmalloc(image_id_size, GFP_KERNEL);
4001 end = image_id + image_id_size;
4002 ceph_encode_string(&p, end, rbd_dev->spec->image_id, (u32)len);
4004 size = sizeof (__le32) + RBD_IMAGE_NAME_LEN_MAX;
4005 reply_buf = kmalloc(size, GFP_KERNEL);
4009 ret = rbd_obj_method_sync(rbd_dev, RBD_DIRECTORY,
4010 "rbd", "dir_get_name",
4011 image_id, image_id_size,
4016 end = reply_buf + ret;
4018 image_name = ceph_extract_encoded_string(&p, end, &len, GFP_KERNEL);
4019 if (IS_ERR(image_name))
4022 dout("%s: name is %s len is %zd\n", __func__, image_name, len);
4030 static u64 rbd_v1_snap_id_by_name(struct rbd_device *rbd_dev, const char *name)
4032 struct ceph_snap_context *snapc = rbd_dev->header.snapc;
4033 const char *snap_name;
4036 /* Skip over names until we find the one we are looking for */
4038 snap_name = rbd_dev->header.snap_names;
4039 while (which < snapc->num_snaps) {
4040 if (!strcmp(name, snap_name))
4041 return snapc->snaps[which];
4042 snap_name += strlen(snap_name) + 1;
4048 static u64 rbd_v2_snap_id_by_name(struct rbd_device *rbd_dev, const char *name)
4050 struct ceph_snap_context *snapc = rbd_dev->header.snapc;
4055 for (which = 0; !found && which < snapc->num_snaps; which++) {
4056 const char *snap_name;
4058 snap_id = snapc->snaps[which];
4059 snap_name = rbd_dev_v2_snap_name(rbd_dev, snap_id);
4060 if (IS_ERR(snap_name))
4062 found = !strcmp(name, snap_name);
4065 return found ? snap_id : CEPH_NOSNAP;
4069 * Assumes name is never RBD_SNAP_HEAD_NAME; returns CEPH_NOSNAP if
4070 * no snapshot by that name is found, or if an error occurs.
4072 static u64 rbd_snap_id_by_name(struct rbd_device *rbd_dev, const char *name)
4074 if (rbd_dev->image_format == 1)
4075 return rbd_v1_snap_id_by_name(rbd_dev, name);
4077 return rbd_v2_snap_id_by_name(rbd_dev, name);
4081 * When an rbd image has a parent image, it is identified by the
4082 * pool, image, and snapshot ids (not names). This function fills
4083 * in the names for those ids. (It's OK if we can't figure out the
4084 * name for an image id, but the pool and snapshot ids should always
4085 * exist and have names.) All names in an rbd spec are dynamically
4088 * When an image being mapped (not a parent) is probed, we have the
4089 * pool name and pool id, image name and image id, and the snapshot
4090 * name. The only thing we're missing is the snapshot id.
4092 static int rbd_dev_spec_update(struct rbd_device *rbd_dev)
4094 struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
4095 struct rbd_spec *spec = rbd_dev->spec;
4096 const char *pool_name;
4097 const char *image_name;
4098 const char *snap_name;
4102 * An image being mapped will have the pool name (etc.), but
4103 * we need to look up the snapshot id.
4105 if (spec->pool_name) {
4106 if (strcmp(spec->snap_name, RBD_SNAP_HEAD_NAME)) {
4109 snap_id = rbd_snap_id_by_name(rbd_dev, spec->snap_name);
4110 if (snap_id == CEPH_NOSNAP)
4112 spec->snap_id = snap_id;
4114 spec->snap_id = CEPH_NOSNAP;
4120 /* Get the pool name; we have to make our own copy of this */
4122 pool_name = ceph_pg_pool_name_by_id(osdc->osdmap, spec->pool_id);
4124 rbd_warn(rbd_dev, "no pool with id %llu", spec->pool_id);
4127 pool_name = kstrdup(pool_name, GFP_KERNEL);
4131 /* Fetch the image name; tolerate failure here */
4133 image_name = rbd_dev_image_name(rbd_dev);
4135 rbd_warn(rbd_dev, "unable to get image name");
4137 /* Look up the snapshot name, and make a copy */
4139 snap_name = rbd_snap_name(rbd_dev, spec->snap_id);
4145 spec->pool_name = pool_name;
4146 spec->image_name = image_name;
4147 spec->snap_name = snap_name;
4157 static int rbd_dev_v2_snap_context(struct rbd_device *rbd_dev)
4166 struct ceph_snap_context *snapc;
4170 * We'll need room for the seq value (maximum snapshot id),
4171 * snapshot count, and array of that many snapshot ids.
4172 * For now we have a fixed upper limit on the number we're
4173 * prepared to receive.
4175 size = sizeof (__le64) + sizeof (__le32) +
4176 RBD_MAX_SNAP_COUNT * sizeof (__le64);
4177 reply_buf = kzalloc(size, GFP_KERNEL);
4181 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
4182 "rbd", "get_snapcontext", NULL, 0,
4184 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
4189 end = reply_buf + ret;
4191 ceph_decode_64_safe(&p, end, seq, out);
4192 ceph_decode_32_safe(&p, end, snap_count, out);
4195 * Make sure the reported number of snapshot ids wouldn't go
4196 * beyond the end of our buffer. But before checking that,
4197 * make sure the computed size of the snapshot context we
4198 * allocate is representable in a size_t.
4200 if (snap_count > (SIZE_MAX - sizeof (struct ceph_snap_context))
4205 if (!ceph_has_room(&p, end, snap_count * sizeof (__le64)))
4209 snapc = ceph_create_snap_context(snap_count, GFP_KERNEL);
4215 for (i = 0; i < snap_count; i++)
4216 snapc->snaps[i] = ceph_decode_64(&p);
4218 ceph_put_snap_context(rbd_dev->header.snapc);
4219 rbd_dev->header.snapc = snapc;
4221 dout(" snap context seq = %llu, snap_count = %u\n",
4222 (unsigned long long)seq, (unsigned int)snap_count);
4229 static const char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev,
4240 size = sizeof (__le32) + RBD_MAX_SNAP_NAME_LEN;
4241 reply_buf = kmalloc(size, GFP_KERNEL);
4243 return ERR_PTR(-ENOMEM);
4245 snapid = cpu_to_le64(snap_id);
4246 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
4247 "rbd", "get_snapshot_name",
4248 &snapid, sizeof (snapid),
4250 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
4252 snap_name = ERR_PTR(ret);
4257 end = reply_buf + ret;
4258 snap_name = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL);
4259 if (IS_ERR(snap_name))
4262 dout(" snap_id 0x%016llx snap_name = %s\n",
4263 (unsigned long long)snap_id, snap_name);
4270 static int rbd_dev_v2_header_info(struct rbd_device *rbd_dev)
4272 bool first_time = rbd_dev->header.object_prefix == NULL;
4275 down_write(&rbd_dev->header_rwsem);
4277 ret = rbd_dev_v2_image_size(rbd_dev);
4282 ret = rbd_dev_v2_header_onetime(rbd_dev);
4288 * If the image supports layering, get the parent info. We
4289 * need to probe the first time regardless. Thereafter we
4290 * only need to if there's a parent, to see if it has
4291 * disappeared due to the mapped image getting flattened.
4293 if (rbd_dev->header.features & RBD_FEATURE_LAYERING &&
4294 (first_time || rbd_dev->parent_spec)) {
4297 ret = rbd_dev_v2_parent_info(rbd_dev);
4302 * Print a warning if this is the initial probe and
4303 * the image has a parent. Don't print it if the
4304 * image now being probed is itself a parent. We
4305 * can tell at this point because we won't know its
4306 * pool name yet (just its pool id).
4308 warn = rbd_dev->parent_spec && rbd_dev->spec->pool_name;
4309 if (first_time && warn)
4310 rbd_warn(rbd_dev, "WARNING: kernel layering "
4311 "is EXPERIMENTAL!");
4314 if (rbd_dev->spec->snap_id == CEPH_NOSNAP)
4315 if (rbd_dev->mapping.size != rbd_dev->header.image_size)
4316 rbd_dev->mapping.size = rbd_dev->header.image_size;
4318 ret = rbd_dev_v2_snap_context(rbd_dev);
4319 dout("rbd_dev_v2_snap_context returned %d\n", ret);
4321 up_write(&rbd_dev->header_rwsem);
4326 static int rbd_bus_add_dev(struct rbd_device *rbd_dev)
4331 dev = &rbd_dev->dev;
4332 dev->bus = &rbd_bus_type;
4333 dev->type = &rbd_device_type;
4334 dev->parent = &rbd_root_dev;
4335 dev->release = rbd_dev_device_release;
4336 dev_set_name(dev, "%d", rbd_dev->dev_id);
4337 ret = device_register(dev);
4342 static void rbd_bus_del_dev(struct rbd_device *rbd_dev)
4344 device_unregister(&rbd_dev->dev);
4347 static atomic64_t rbd_dev_id_max = ATOMIC64_INIT(0);
4350 * Get a unique rbd identifier for the given new rbd_dev, and add
4351 * the rbd_dev to the global list. The minimum rbd id is 1.
4353 static void rbd_dev_id_get(struct rbd_device *rbd_dev)
4355 rbd_dev->dev_id = atomic64_inc_return(&rbd_dev_id_max);
4357 spin_lock(&rbd_dev_list_lock);
4358 list_add_tail(&rbd_dev->node, &rbd_dev_list);
4359 spin_unlock(&rbd_dev_list_lock);
4360 dout("rbd_dev %p given dev id %llu\n", rbd_dev,
4361 (unsigned long long) rbd_dev->dev_id);
4365 * Remove an rbd_dev from the global list, and record that its
4366 * identifier is no longer in use.
4368 static void rbd_dev_id_put(struct rbd_device *rbd_dev)
4370 struct list_head *tmp;
4371 int rbd_id = rbd_dev->dev_id;
4374 rbd_assert(rbd_id > 0);
4376 dout("rbd_dev %p released dev id %llu\n", rbd_dev,
4377 (unsigned long long) rbd_dev->dev_id);
4378 spin_lock(&rbd_dev_list_lock);
4379 list_del_init(&rbd_dev->node);
4382 * If the id being "put" is not the current maximum, there
4383 * is nothing special we need to do.
4385 if (rbd_id != atomic64_read(&rbd_dev_id_max)) {
4386 spin_unlock(&rbd_dev_list_lock);
4391 * We need to update the current maximum id. Search the
4392 * list to find out what it is. We're more likely to find
4393 * the maximum at the end, so search the list backward.
4396 list_for_each_prev(tmp, &rbd_dev_list) {
4397 struct rbd_device *rbd_dev;
4399 rbd_dev = list_entry(tmp, struct rbd_device, node);
4400 if (rbd_dev->dev_id > max_id)
4401 max_id = rbd_dev->dev_id;
4403 spin_unlock(&rbd_dev_list_lock);
4406 * The max id could have been updated by rbd_dev_id_get(), in
4407 * which case it now accurately reflects the new maximum.
4408 * Be careful not to overwrite the maximum value in that
4411 atomic64_cmpxchg(&rbd_dev_id_max, rbd_id, max_id);
4412 dout(" max dev id has been reset\n");
4416 * Skips over white space at *buf, and updates *buf to point to the
4417 * first found non-space character (if any). Returns the length of
4418 * the token (string of non-white space characters) found. Note
4419 * that *buf must be terminated with '\0'.
4421 static inline size_t next_token(const char **buf)
4424 * These are the characters that produce nonzero for
4425 * isspace() in the "C" and "POSIX" locales.
4427 const char *spaces = " \f\n\r\t\v";
4429 *buf += strspn(*buf, spaces); /* Find start of token */
4431 return strcspn(*buf, spaces); /* Return token length */
4435 * Finds the next token in *buf, and if the provided token buffer is
4436 * big enough, copies the found token into it. The result, if
4437 * copied, is guaranteed to be terminated with '\0'. Note that *buf
4438 * must be terminated with '\0' on entry.
4440 * Returns the length of the token found (not including the '\0').
4441 * Return value will be 0 if no token is found, and it will be >=
4442 * token_size if the token would not fit.
4444 * The *buf pointer will be updated to point beyond the end of the
4445 * found token. Note that this occurs even if the token buffer is
4446 * too small to hold it.
4448 static inline size_t copy_token(const char **buf,
4454 len = next_token(buf);
4455 if (len < token_size) {
4456 memcpy(token, *buf, len);
4457 *(token + len) = '\0';
4465 * Finds the next token in *buf, dynamically allocates a buffer big
4466 * enough to hold a copy of it, and copies the token into the new
4467 * buffer. The copy is guaranteed to be terminated with '\0'. Note
4468 * that a duplicate buffer is created even for a zero-length token.
4470 * Returns a pointer to the newly-allocated duplicate, or a null
4471 * pointer if memory for the duplicate was not available. If
4472 * the lenp argument is a non-null pointer, the length of the token
4473 * (not including the '\0') is returned in *lenp.
4475 * If successful, the *buf pointer will be updated to point beyond
4476 * the end of the found token.
4478 * Note: uses GFP_KERNEL for allocation.
4480 static inline char *dup_token(const char **buf, size_t *lenp)
4485 len = next_token(buf);
4486 dup = kmemdup(*buf, len + 1, GFP_KERNEL);
4489 *(dup + len) = '\0';
4499 * Parse the options provided for an "rbd add" (i.e., rbd image
4500 * mapping) request. These arrive via a write to /sys/bus/rbd/add,
4501 * and the data written is passed here via a NUL-terminated buffer.
4502 * Returns 0 if successful or an error code otherwise.
4504 * The information extracted from these options is recorded in
4505 * the other parameters which return dynamically-allocated
4508 * The address of a pointer that will refer to a ceph options
4509 * structure. Caller must release the returned pointer using
4510 * ceph_destroy_options() when it is no longer needed.
4512 * Address of an rbd options pointer. Fully initialized by
4513 * this function; caller must release with kfree().
4515 * Address of an rbd image specification pointer. Fully
4516 * initialized by this function based on parsed options.
4517 * Caller must release with rbd_spec_put().
4519 * The options passed take this form:
4520 * <mon_addrs> <options> <pool_name> <image_name> [<snap_id>]
4523 * A comma-separated list of one or more monitor addresses.
4524 * A monitor address is an ip address, optionally followed
4525 * by a port number (separated by a colon).
4526 * I.e.: ip1[:port1][,ip2[:port2]...]
4528 * A comma-separated list of ceph and/or rbd options.
4530 * The name of the rados pool containing the rbd image.
4532 * The name of the image in that pool to map.
4534 * An optional snapshot id. If provided, the mapping will
4535 * present data from the image at the time that snapshot was
4536 * created. The image head is used if no snapshot id is
4537 * provided. Snapshot mappings are always read-only.
4539 static int rbd_add_parse_args(const char *buf,
4540 struct ceph_options **ceph_opts,
4541 struct rbd_options **opts,
4542 struct rbd_spec **rbd_spec)
4546 const char *mon_addrs;
4548 size_t mon_addrs_size;
4549 struct rbd_spec *spec = NULL;
4550 struct rbd_options *rbd_opts = NULL;
4551 struct ceph_options *copts;
4554 /* The first four tokens are required */
4556 len = next_token(&buf);
4558 rbd_warn(NULL, "no monitor address(es) provided");
4562 mon_addrs_size = len + 1;
4566 options = dup_token(&buf, NULL);
4570 rbd_warn(NULL, "no options provided");
4574 spec = rbd_spec_alloc();
4578 spec->pool_name = dup_token(&buf, NULL);
4579 if (!spec->pool_name)
4581 if (!*spec->pool_name) {
4582 rbd_warn(NULL, "no pool name provided");
4586 spec->image_name = dup_token(&buf, NULL);
4587 if (!spec->image_name)
4589 if (!*spec->image_name) {
4590 rbd_warn(NULL, "no image name provided");
4595 * Snapshot name is optional; default is to use "-"
4596 * (indicating the head/no snapshot).
4598 len = next_token(&buf);
4600 buf = RBD_SNAP_HEAD_NAME; /* No snapshot supplied */
4601 len = sizeof (RBD_SNAP_HEAD_NAME) - 1;
4602 } else if (len > RBD_MAX_SNAP_NAME_LEN) {
4603 ret = -ENAMETOOLONG;
4606 snap_name = kmemdup(buf, len + 1, GFP_KERNEL);
4609 *(snap_name + len) = '\0';
4610 spec->snap_name = snap_name;
4612 /* Initialize all rbd options to the defaults */
4614 rbd_opts = kzalloc(sizeof (*rbd_opts), GFP_KERNEL);
4618 rbd_opts->read_only = RBD_READ_ONLY_DEFAULT;
4620 copts = ceph_parse_options(options, mon_addrs,
4621 mon_addrs + mon_addrs_size - 1,
4622 parse_rbd_opts_token, rbd_opts);
4623 if (IS_ERR(copts)) {
4624 ret = PTR_ERR(copts);
4645 * An rbd format 2 image has a unique identifier, distinct from the
4646 * name given to it by the user. Internally, that identifier is
4647 * what's used to specify the names of objects related to the image.
4649 * A special "rbd id" object is used to map an rbd image name to its
4650 * id. If that object doesn't exist, then there is no v2 rbd image
4651 * with the supplied name.
4653 * This function will record the given rbd_dev's image_id field if
4654 * it can be determined, and in that case will return 0. If any
4655 * errors occur a negative errno will be returned and the rbd_dev's
4656 * image_id field will be unchanged (and should be NULL).
4658 static int rbd_dev_image_id(struct rbd_device *rbd_dev)
4667 * When probing a parent image, the image id is already
4668 * known (and the image name likely is not). There's no
4669 * need to fetch the image id again in this case. We
4670 * do still need to set the image format though.
4672 if (rbd_dev->spec->image_id) {
4673 rbd_dev->image_format = *rbd_dev->spec->image_id ? 2 : 1;
4679 * First, see if the format 2 image id file exists, and if
4680 * so, get the image's persistent id from it.
4682 size = sizeof (RBD_ID_PREFIX) + strlen(rbd_dev->spec->image_name);
4683 object_name = kmalloc(size, GFP_NOIO);
4686 sprintf(object_name, "%s%s", RBD_ID_PREFIX, rbd_dev->spec->image_name);
4687 dout("rbd id object name is %s\n", object_name);
4689 /* Response will be an encoded string, which includes a length */
4691 size = sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX;
4692 response = kzalloc(size, GFP_NOIO);
4698 /* If it doesn't exist we'll assume it's a format 1 image */
4700 ret = rbd_obj_method_sync(rbd_dev, object_name,
4701 "rbd", "get_id", NULL, 0,
4702 response, RBD_IMAGE_ID_LEN_MAX);
4703 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
4704 if (ret == -ENOENT) {
4705 image_id = kstrdup("", GFP_KERNEL);
4706 ret = image_id ? 0 : -ENOMEM;
4708 rbd_dev->image_format = 1;
4709 } else if (ret > sizeof (__le32)) {
4712 image_id = ceph_extract_encoded_string(&p, p + ret,
4714 ret = IS_ERR(image_id) ? PTR_ERR(image_id) : 0;
4716 rbd_dev->image_format = 2;
4722 rbd_dev->spec->image_id = image_id;
4723 dout("image_id is %s\n", image_id);
4733 * Undo whatever state changes are made by v1 or v2 header info
4736 static void rbd_dev_unprobe(struct rbd_device *rbd_dev)
4738 struct rbd_image_header *header;
4740 /* Drop parent reference unless it's already been done (or none) */
4742 if (rbd_dev->parent_overlap)
4743 rbd_dev_parent_put(rbd_dev);
4745 /* Free dynamic fields from the header, then zero it out */
4747 header = &rbd_dev->header;
4748 ceph_put_snap_context(header->snapc);
4749 kfree(header->snap_sizes);
4750 kfree(header->snap_names);
4751 kfree(header->object_prefix);
4752 memset(header, 0, sizeof (*header));
4755 static int rbd_dev_v2_header_onetime(struct rbd_device *rbd_dev)
4759 ret = rbd_dev_v2_object_prefix(rbd_dev);
4764 * Get the and check features for the image. Currently the
4765 * features are assumed to never change.
4767 ret = rbd_dev_v2_features(rbd_dev);
4771 /* If the image supports fancy striping, get its parameters */
4773 if (rbd_dev->header.features & RBD_FEATURE_STRIPINGV2) {
4774 ret = rbd_dev_v2_striping_info(rbd_dev);
4778 /* No support for crypto and compression type format 2 images */
4782 rbd_dev->header.features = 0;
4783 kfree(rbd_dev->header.object_prefix);
4784 rbd_dev->header.object_prefix = NULL;
4789 static int rbd_dev_probe_parent(struct rbd_device *rbd_dev)
4791 struct rbd_device *parent = NULL;
4792 struct rbd_spec *parent_spec;
4793 struct rbd_client *rbdc;
4796 if (!rbd_dev->parent_spec)
4799 * We need to pass a reference to the client and the parent
4800 * spec when creating the parent rbd_dev. Images related by
4801 * parent/child relationships always share both.
4803 parent_spec = rbd_spec_get(rbd_dev->parent_spec);
4804 rbdc = __rbd_get_client(rbd_dev->rbd_client);
4807 parent = rbd_dev_create(rbdc, parent_spec);
4811 ret = rbd_dev_image_probe(parent, false);
4814 rbd_dev->parent = parent;
4815 atomic_set(&rbd_dev->parent_ref, 1);
4820 rbd_dev_unparent(rbd_dev);
4821 kfree(rbd_dev->header_name);
4822 rbd_dev_destroy(parent);
4824 rbd_put_client(rbdc);
4825 rbd_spec_put(parent_spec);
4831 static int rbd_dev_device_setup(struct rbd_device *rbd_dev)
4835 /* generate unique id: find highest unique id, add one */
4836 rbd_dev_id_get(rbd_dev);
4838 /* Fill in the device name, now that we have its id. */
4839 BUILD_BUG_ON(DEV_NAME_LEN
4840 < sizeof (RBD_DRV_NAME) + MAX_INT_FORMAT_WIDTH);
4841 sprintf(rbd_dev->name, "%s%d", RBD_DRV_NAME, rbd_dev->dev_id);
4843 /* Get our block major device number. */
4845 ret = register_blkdev(0, rbd_dev->name);
4848 rbd_dev->major = ret;
4850 /* Set up the blkdev mapping. */
4852 ret = rbd_init_disk(rbd_dev);
4854 goto err_out_blkdev;
4856 ret = rbd_dev_mapping_set(rbd_dev);
4859 set_capacity(rbd_dev->disk, rbd_dev->mapping.size / SECTOR_SIZE);
4861 ret = rbd_bus_add_dev(rbd_dev);
4863 goto err_out_mapping;
4865 /* Everything's ready. Announce the disk to the world. */
4867 set_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags);
4868 add_disk(rbd_dev->disk);
4870 pr_info("%s: added with size 0x%llx\n", rbd_dev->disk->disk_name,
4871 (unsigned long long) rbd_dev->mapping.size);
4876 rbd_dev_mapping_clear(rbd_dev);
4878 rbd_free_disk(rbd_dev);
4880 unregister_blkdev(rbd_dev->major, rbd_dev->name);
4882 rbd_dev_id_put(rbd_dev);
4883 rbd_dev_mapping_clear(rbd_dev);
4888 static int rbd_dev_header_name(struct rbd_device *rbd_dev)
4890 struct rbd_spec *spec = rbd_dev->spec;
4893 /* Record the header object name for this rbd image. */
4895 rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
4897 if (rbd_dev->image_format == 1)
4898 size = strlen(spec->image_name) + sizeof (RBD_SUFFIX);
4900 size = sizeof (RBD_HEADER_PREFIX) + strlen(spec->image_id);
4902 rbd_dev->header_name = kmalloc(size, GFP_KERNEL);
4903 if (!rbd_dev->header_name)
4906 if (rbd_dev->image_format == 1)
4907 sprintf(rbd_dev->header_name, "%s%s",
4908 spec->image_name, RBD_SUFFIX);
4910 sprintf(rbd_dev->header_name, "%s%s",
4911 RBD_HEADER_PREFIX, spec->image_id);
4915 static void rbd_dev_image_release(struct rbd_device *rbd_dev)
4917 rbd_dev_unprobe(rbd_dev);
4918 kfree(rbd_dev->header_name);
4919 rbd_dev->header_name = NULL;
4920 rbd_dev->image_format = 0;
4921 kfree(rbd_dev->spec->image_id);
4922 rbd_dev->spec->image_id = NULL;
4924 rbd_dev_destroy(rbd_dev);
4928 * Probe for the existence of the header object for the given rbd
4929 * device. If this image is the one being mapped (i.e., not a
4930 * parent), initiate a watch on its header object before using that
4931 * object to get detailed information about the rbd image.
4933 static int rbd_dev_image_probe(struct rbd_device *rbd_dev, bool mapping)
4939 * Get the id from the image id object. Unless there's an
4940 * error, rbd_dev->spec->image_id will be filled in with
4941 * a dynamically-allocated string, and rbd_dev->image_format
4942 * will be set to either 1 or 2.
4944 ret = rbd_dev_image_id(rbd_dev);
4947 rbd_assert(rbd_dev->spec->image_id);
4948 rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
4950 ret = rbd_dev_header_name(rbd_dev);
4952 goto err_out_format;
4955 ret = rbd_dev_header_watch_sync(rbd_dev, true);
4957 goto out_header_name;
4960 if (rbd_dev->image_format == 1)
4961 ret = rbd_dev_v1_header_info(rbd_dev);
4963 ret = rbd_dev_v2_header_info(rbd_dev);
4967 ret = rbd_dev_spec_update(rbd_dev);
4971 ret = rbd_dev_probe_parent(rbd_dev);
4975 dout("discovered format %u image, header name is %s\n",
4976 rbd_dev->image_format, rbd_dev->header_name);
4980 rbd_dev_unprobe(rbd_dev);
4983 tmp = rbd_dev_header_watch_sync(rbd_dev, false);
4985 rbd_warn(rbd_dev, "unable to tear down "
4986 "watch request (%d)\n", tmp);
4989 kfree(rbd_dev->header_name);
4990 rbd_dev->header_name = NULL;
4992 rbd_dev->image_format = 0;
4993 kfree(rbd_dev->spec->image_id);
4994 rbd_dev->spec->image_id = NULL;
4996 dout("probe failed, returning %d\n", ret);
5001 static ssize_t rbd_add(struct bus_type *bus,
5005 struct rbd_device *rbd_dev = NULL;
5006 struct ceph_options *ceph_opts = NULL;
5007 struct rbd_options *rbd_opts = NULL;
5008 struct rbd_spec *spec = NULL;
5009 struct rbd_client *rbdc;
5010 struct ceph_osd_client *osdc;
5014 if (!try_module_get(THIS_MODULE))
5017 /* parse add command */
5018 rc = rbd_add_parse_args(buf, &ceph_opts, &rbd_opts, &spec);
5020 goto err_out_module;
5021 read_only = rbd_opts->read_only;
5023 rbd_opts = NULL; /* done with this */
5025 rbdc = rbd_get_client(ceph_opts);
5032 osdc = &rbdc->client->osdc;
5033 rc = ceph_pg_poolid_by_name(osdc->osdmap, spec->pool_name);
5035 goto err_out_client;
5036 spec->pool_id = (u64)rc;
5038 /* The ceph file layout needs to fit pool id in 32 bits */
5040 if (spec->pool_id > (u64)U32_MAX) {
5041 rbd_warn(NULL, "pool id too large (%llu > %u)\n",
5042 (unsigned long long)spec->pool_id, U32_MAX);
5044 goto err_out_client;
5047 rbd_dev = rbd_dev_create(rbdc, spec);
5049 goto err_out_client;
5050 rbdc = NULL; /* rbd_dev now owns this */
5051 spec = NULL; /* rbd_dev now owns this */
5053 rc = rbd_dev_image_probe(rbd_dev, true);
5055 goto err_out_rbd_dev;
5057 /* If we are mapping a snapshot it must be marked read-only */
5059 if (rbd_dev->spec->snap_id != CEPH_NOSNAP)
5061 rbd_dev->mapping.read_only = read_only;
5063 rc = rbd_dev_device_setup(rbd_dev);
5065 rbd_dev_image_release(rbd_dev);
5066 goto err_out_module;
5072 rbd_dev_destroy(rbd_dev);
5074 rbd_put_client(rbdc);
5078 module_put(THIS_MODULE);
5080 dout("Error adding device %s\n", buf);
5085 static void rbd_dev_device_release(struct device *dev)
5087 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
5089 rbd_free_disk(rbd_dev);
5090 clear_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags);
5091 rbd_dev_mapping_clear(rbd_dev);
5092 unregister_blkdev(rbd_dev->major, rbd_dev->name);
5094 rbd_dev_id_put(rbd_dev);
5095 rbd_dev_mapping_clear(rbd_dev);
5098 static void rbd_dev_remove_parent(struct rbd_device *rbd_dev)
5100 while (rbd_dev->parent) {
5101 struct rbd_device *first = rbd_dev;
5102 struct rbd_device *second = first->parent;
5103 struct rbd_device *third;
5106 * Follow to the parent with no grandparent and
5109 while (second && (third = second->parent)) {
5114 rbd_dev_image_release(second);
5115 first->parent = NULL;
5116 first->parent_overlap = 0;
5118 rbd_assert(first->parent_spec);
5119 rbd_spec_put(first->parent_spec);
5120 first->parent_spec = NULL;
5124 static ssize_t rbd_remove(struct bus_type *bus,
5128 struct rbd_device *rbd_dev = NULL;
5129 struct list_head *tmp;
5132 bool already = false;
5135 ret = strict_strtoul(buf, 10, &ul);
5139 /* convert to int; abort if we lost anything in the conversion */
5145 spin_lock(&rbd_dev_list_lock);
5146 list_for_each(tmp, &rbd_dev_list) {
5147 rbd_dev = list_entry(tmp, struct rbd_device, node);
5148 if (rbd_dev->dev_id == dev_id) {
5154 spin_lock_irq(&rbd_dev->lock);
5155 if (rbd_dev->open_count)
5158 already = test_and_set_bit(RBD_DEV_FLAG_REMOVING,
5160 spin_unlock_irq(&rbd_dev->lock);
5162 spin_unlock(&rbd_dev_list_lock);
5163 if (ret < 0 || already)
5166 rbd_bus_del_dev(rbd_dev);
5167 ret = rbd_dev_header_watch_sync(rbd_dev, false);
5169 rbd_warn(rbd_dev, "failed to cancel watch event (%d)\n", ret);
5170 rbd_dev_image_release(rbd_dev);
5171 module_put(THIS_MODULE);
5177 * create control files in sysfs
5180 static int rbd_sysfs_init(void)
5184 ret = device_register(&rbd_root_dev);
5188 ret = bus_register(&rbd_bus_type);
5190 device_unregister(&rbd_root_dev);
5195 static void rbd_sysfs_cleanup(void)
5197 bus_unregister(&rbd_bus_type);
5198 device_unregister(&rbd_root_dev);
5201 static int rbd_slab_init(void)
5203 rbd_assert(!rbd_img_request_cache);
5204 rbd_img_request_cache = kmem_cache_create("rbd_img_request",
5205 sizeof (struct rbd_img_request),
5206 __alignof__(struct rbd_img_request),
5208 if (!rbd_img_request_cache)
5211 rbd_assert(!rbd_obj_request_cache);
5212 rbd_obj_request_cache = kmem_cache_create("rbd_obj_request",
5213 sizeof (struct rbd_obj_request),
5214 __alignof__(struct rbd_obj_request),
5216 if (!rbd_obj_request_cache)
5219 rbd_assert(!rbd_segment_name_cache);
5220 rbd_segment_name_cache = kmem_cache_create("rbd_segment_name",
5221 MAX_OBJ_NAME_SIZE + 1, 1, 0, NULL);
5222 if (rbd_segment_name_cache)
5225 if (rbd_obj_request_cache) {
5226 kmem_cache_destroy(rbd_obj_request_cache);
5227 rbd_obj_request_cache = NULL;
5230 kmem_cache_destroy(rbd_img_request_cache);
5231 rbd_img_request_cache = NULL;
5236 static void rbd_slab_exit(void)
5238 rbd_assert(rbd_segment_name_cache);
5239 kmem_cache_destroy(rbd_segment_name_cache);
5240 rbd_segment_name_cache = NULL;
5242 rbd_assert(rbd_obj_request_cache);
5243 kmem_cache_destroy(rbd_obj_request_cache);
5244 rbd_obj_request_cache = NULL;
5246 rbd_assert(rbd_img_request_cache);
5247 kmem_cache_destroy(rbd_img_request_cache);
5248 rbd_img_request_cache = NULL;
5251 static int __init rbd_init(void)
5255 if (!libceph_compatible(NULL)) {
5256 rbd_warn(NULL, "libceph incompatibility (quitting)");
5260 rc = rbd_slab_init();
5263 rc = rbd_sysfs_init();
5267 pr_info("loaded " RBD_DRV_NAME_LONG "\n");
5272 static void __exit rbd_exit(void)
5274 rbd_sysfs_cleanup();
5278 module_init(rbd_init);
5279 module_exit(rbd_exit);
5281 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
5282 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
5283 MODULE_DESCRIPTION("rados block device");
5285 /* following authorship retained from original osdblk.c */
5286 MODULE_AUTHOR("Jeff Garzik <jeff@garzik.org>");
5288 MODULE_LICENSE("GPL");