2 rbd.c -- Export ceph rados objects as a Linux block device
5 based on drivers/block/osdblk.c:
7 Copyright 2009 Red Hat, Inc.
9 This program is free software; you can redistribute it and/or modify
10 it under the terms of the GNU General Public License as published by
11 the Free Software Foundation.
13 This program is distributed in the hope that it will be useful,
14 but WITHOUT ANY WARRANTY; without even the implied warranty of
15 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 GNU General Public License for more details.
18 You should have received a copy of the GNU General Public License
19 along with this program; see the file COPYING. If not, write to
20 the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA.
24 For usage instructions, please refer to:
26 Documentation/ABI/testing/sysfs-bus-rbd
30 #include <linux/ceph/libceph.h>
31 #include <linux/ceph/osd_client.h>
32 #include <linux/ceph/mon_client.h>
33 #include <linux/ceph/decode.h>
34 #include <linux/parser.h>
36 #include <linux/kernel.h>
37 #include <linux/device.h>
38 #include <linux/module.h>
40 #include <linux/blkdev.h>
42 #include "rbd_types.h"
44 #define RBD_DEBUG /* Activate rbd_assert() calls */
47 * The basic unit of block I/O is a sector. It is interpreted in a
48 * number of contexts in Linux (blk, bio, genhd), but the default is
49 * universally 512 bytes. These symbols are just slightly more
50 * meaningful than the bare numbers they represent.
52 #define SECTOR_SHIFT 9
53 #define SECTOR_SIZE (1ULL << SECTOR_SHIFT)
55 /* It might be useful to have these defined elsewhere */
57 #define U8_MAX ((u8) (~0U))
58 #define U16_MAX ((u16) (~0U))
59 #define U32_MAX ((u32) (~0U))
60 #define U64_MAX ((u64) (~0ULL))
62 #define RBD_DRV_NAME "rbd"
63 #define RBD_DRV_NAME_LONG "rbd (rados block device)"
65 #define RBD_MINORS_PER_MAJOR 256 /* max minors per blkdev */
67 #define RBD_SNAP_DEV_NAME_PREFIX "snap_"
68 #define RBD_MAX_SNAP_NAME_LEN \
69 (NAME_MAX - (sizeof (RBD_SNAP_DEV_NAME_PREFIX) - 1))
71 #define RBD_MAX_SNAP_COUNT 510 /* allows max snapc to fit in 4KB */
73 #define RBD_SNAP_HEAD_NAME "-"
75 /* This allows a single page to hold an image name sent by OSD */
76 #define RBD_IMAGE_NAME_LEN_MAX (PAGE_SIZE - sizeof (__le32) - 1)
77 #define RBD_IMAGE_ID_LEN_MAX 64
79 #define RBD_OBJ_PREFIX_LEN_MAX 64
83 #define RBD_FEATURE_LAYERING 1
85 /* Features supported by this (client software) implementation. */
87 #define RBD_FEATURES_ALL (0)
90 * An RBD device name will be "rbd#", where the "rbd" comes from
91 * RBD_DRV_NAME above, and # is a unique integer identifier.
92 * MAX_INT_FORMAT_WIDTH is used in ensuring DEV_NAME_LEN is big
93 * enough to hold all possible device names.
95 #define DEV_NAME_LEN 32
96 #define MAX_INT_FORMAT_WIDTH ((5 * sizeof (int)) / 2 + 1)
99 * block device image metadata (in-memory version)
101 struct rbd_image_header {
102 /* These four fields never change for a given rbd image */
109 /* The remaining fields need to be updated occasionally */
111 struct ceph_snap_context *snapc;
119 * An rbd image specification.
121 * The tuple (pool_id, image_id, snap_id) is sufficient to uniquely
122 * identify an image. Each rbd_dev structure includes a pointer to
123 * an rbd_spec structure that encapsulates this identity.
125 * Each of the id's in an rbd_spec has an associated name. For a
126 * user-mapped image, the names are supplied and the id's associated
127 * with them are looked up. For a layered image, a parent image is
128 * defined by the tuple, and the names are looked up.
130 * An rbd_dev structure contains a parent_spec pointer which is
131 * non-null if the image it represents is a child in a layered
132 * image. This pointer will refer to the rbd_spec structure used
133 * by the parent rbd_dev for its own identity (i.e., the structure
134 * is shared between the parent and child).
136 * Since these structures are populated once, during the discovery
137 * phase of image construction, they are effectively immutable so
138 * we make no effort to synchronize access to them.
140 * Note that code herein does not assume the image name is known (it
141 * could be a null pointer).
157 * an instance of the client. multiple devices may share an rbd client.
160 struct ceph_client *client;
162 struct list_head node;
165 struct rbd_img_request;
166 typedef void (*rbd_img_callback_t)(struct rbd_img_request *);
168 #define BAD_WHICH U32_MAX /* Good which or bad which, which? */
170 struct rbd_obj_request;
171 typedef void (*rbd_obj_callback_t)(struct rbd_obj_request *);
173 enum obj_request_type {
174 OBJ_REQUEST_NODATA, OBJ_REQUEST_BIO, OBJ_REQUEST_PAGES
177 struct rbd_obj_request {
178 const char *object_name;
179 u64 offset; /* object start byte */
180 u64 length; /* bytes from offset */
182 struct rbd_img_request *img_request;
183 struct list_head links; /* img_request->obj_requests */
184 u32 which; /* posn image request list */
186 enum obj_request_type type;
188 struct bio *bio_list;
195 struct ceph_osd_request *osd_req;
197 u64 xferred; /* bytes transferred */
202 rbd_obj_callback_t callback;
203 struct completion completion;
208 struct rbd_img_request {
210 struct rbd_device *rbd_dev;
211 u64 offset; /* starting image byte offset */
212 u64 length; /* byte count from offset */
213 bool write_request; /* false for read */
215 struct ceph_snap_context *snapc; /* for writes */
216 u64 snap_id; /* for reads */
218 spinlock_t completion_lock;/* protects next_completion */
220 rbd_img_callback_t callback;
222 u32 obj_request_count;
223 struct list_head obj_requests; /* rbd_obj_request structs */
228 #define for_each_obj_request(ireq, oreq) \
229 list_for_each_entry(oreq, &(ireq)->obj_requests, links)
230 #define for_each_obj_request_from(ireq, oreq) \
231 list_for_each_entry_from(oreq, &(ireq)->obj_requests, links)
232 #define for_each_obj_request_safe(ireq, oreq, n) \
233 list_for_each_entry_safe_reverse(oreq, n, &(ireq)->obj_requests, links)
239 struct list_head node;
254 int dev_id; /* blkdev unique id */
256 int major; /* blkdev assigned major */
257 struct gendisk *disk; /* blkdev's gendisk and rq */
259 u32 image_format; /* Either 1 or 2 */
260 struct rbd_client *rbd_client;
262 char name[DEV_NAME_LEN]; /* blkdev name, e.g. rbd3 */
264 spinlock_t lock; /* queue, flags, open_count */
266 struct rbd_image_header header;
267 unsigned long flags; /* possibly lock protected */
268 struct rbd_spec *spec;
272 struct ceph_file_layout layout;
274 struct ceph_osd_event *watch_event;
275 struct rbd_obj_request *watch_request;
277 struct rbd_spec *parent_spec;
280 /* protects updating the header */
281 struct rw_semaphore header_rwsem;
283 struct rbd_mapping mapping;
285 struct list_head node;
287 /* list of snapshots */
288 struct list_head snaps;
292 unsigned long open_count; /* protected by lock */
296 * Flag bits for rbd_dev->flags. If atomicity is required,
297 * rbd_dev->lock is used to protect access.
299 * Currently, only the "removing" flag (which is coupled with the
300 * "open_count" field) requires atomic access.
303 RBD_DEV_FLAG_EXISTS, /* mapped snapshot has not been deleted */
304 RBD_DEV_FLAG_REMOVING, /* this mapping is being removed */
307 static DEFINE_MUTEX(ctl_mutex); /* Serialize open/close/setup/teardown */
309 static LIST_HEAD(rbd_dev_list); /* devices */
310 static DEFINE_SPINLOCK(rbd_dev_list_lock);
312 static LIST_HEAD(rbd_client_list); /* clients */
313 static DEFINE_SPINLOCK(rbd_client_list_lock);
315 static int rbd_dev_snaps_update(struct rbd_device *rbd_dev);
316 static int rbd_dev_snaps_register(struct rbd_device *rbd_dev);
318 static void rbd_dev_release(struct device *dev);
319 static void rbd_remove_snap_dev(struct rbd_snap *snap);
321 static ssize_t rbd_add(struct bus_type *bus, const char *buf,
323 static ssize_t rbd_remove(struct bus_type *bus, const char *buf,
326 static struct bus_attribute rbd_bus_attrs[] = {
327 __ATTR(add, S_IWUSR, NULL, rbd_add),
328 __ATTR(remove, S_IWUSR, NULL, rbd_remove),
332 static struct bus_type rbd_bus_type = {
334 .bus_attrs = rbd_bus_attrs,
337 static void rbd_root_dev_release(struct device *dev)
341 static struct device rbd_root_dev = {
343 .release = rbd_root_dev_release,
346 static __printf(2, 3)
347 void rbd_warn(struct rbd_device *rbd_dev, const char *fmt, ...)
349 struct va_format vaf;
357 printk(KERN_WARNING "%s: %pV\n", RBD_DRV_NAME, &vaf);
358 else if (rbd_dev->disk)
359 printk(KERN_WARNING "%s: %s: %pV\n",
360 RBD_DRV_NAME, rbd_dev->disk->disk_name, &vaf);
361 else if (rbd_dev->spec && rbd_dev->spec->image_name)
362 printk(KERN_WARNING "%s: image %s: %pV\n",
363 RBD_DRV_NAME, rbd_dev->spec->image_name, &vaf);
364 else if (rbd_dev->spec && rbd_dev->spec->image_id)
365 printk(KERN_WARNING "%s: id %s: %pV\n",
366 RBD_DRV_NAME, rbd_dev->spec->image_id, &vaf);
368 printk(KERN_WARNING "%s: rbd_dev %p: %pV\n",
369 RBD_DRV_NAME, rbd_dev, &vaf);
374 #define rbd_assert(expr) \
375 if (unlikely(!(expr))) { \
376 printk(KERN_ERR "\nAssertion failure in %s() " \
378 "\trbd_assert(%s);\n\n", \
379 __func__, __LINE__, #expr); \
382 #else /* !RBD_DEBUG */
383 # define rbd_assert(expr) ((void) 0)
384 #endif /* !RBD_DEBUG */
386 static int rbd_dev_refresh(struct rbd_device *rbd_dev, u64 *hver);
387 static int rbd_dev_v2_refresh(struct rbd_device *rbd_dev, u64 *hver);
389 static int rbd_open(struct block_device *bdev, fmode_t mode)
391 struct rbd_device *rbd_dev = bdev->bd_disk->private_data;
392 bool removing = false;
394 if ((mode & FMODE_WRITE) && rbd_dev->mapping.read_only)
397 spin_lock_irq(&rbd_dev->lock);
398 if (test_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags))
401 rbd_dev->open_count++;
402 spin_unlock_irq(&rbd_dev->lock);
406 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
407 (void) get_device(&rbd_dev->dev);
408 set_device_ro(bdev, rbd_dev->mapping.read_only);
409 mutex_unlock(&ctl_mutex);
414 static int rbd_release(struct gendisk *disk, fmode_t mode)
416 struct rbd_device *rbd_dev = disk->private_data;
417 unsigned long open_count_before;
419 spin_lock_irq(&rbd_dev->lock);
420 open_count_before = rbd_dev->open_count--;
421 spin_unlock_irq(&rbd_dev->lock);
422 rbd_assert(open_count_before > 0);
424 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
425 put_device(&rbd_dev->dev);
426 mutex_unlock(&ctl_mutex);
431 static const struct block_device_operations rbd_bd_ops = {
432 .owner = THIS_MODULE,
434 .release = rbd_release,
438 * Initialize an rbd client instance.
441 static struct rbd_client *rbd_client_create(struct ceph_options *ceph_opts)
443 struct rbd_client *rbdc;
446 dout("rbd_client_create\n");
447 rbdc = kmalloc(sizeof(struct rbd_client), GFP_KERNEL);
451 kref_init(&rbdc->kref);
452 INIT_LIST_HEAD(&rbdc->node);
454 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
456 rbdc->client = ceph_create_client(ceph_opts, rbdc, 0, 0);
457 if (IS_ERR(rbdc->client))
459 ceph_opts = NULL; /* Now rbdc->client is responsible for ceph_opts */
461 ret = ceph_open_session(rbdc->client);
465 spin_lock(&rbd_client_list_lock);
466 list_add_tail(&rbdc->node, &rbd_client_list);
467 spin_unlock(&rbd_client_list_lock);
469 mutex_unlock(&ctl_mutex);
471 dout("rbd_client_create created %p\n", rbdc);
475 ceph_destroy_client(rbdc->client);
477 mutex_unlock(&ctl_mutex);
481 ceph_destroy_options(ceph_opts);
486 * Find a ceph client with specific addr and configuration. If
487 * found, bump its reference count.
489 static struct rbd_client *rbd_client_find(struct ceph_options *ceph_opts)
491 struct rbd_client *client_node;
494 if (ceph_opts->flags & CEPH_OPT_NOSHARE)
497 spin_lock(&rbd_client_list_lock);
498 list_for_each_entry(client_node, &rbd_client_list, node) {
499 if (!ceph_compare_options(ceph_opts, client_node->client)) {
500 kref_get(&client_node->kref);
505 spin_unlock(&rbd_client_list_lock);
507 return found ? client_node : NULL;
517 /* string args above */
520 /* Boolean args above */
524 static match_table_t rbd_opts_tokens = {
526 /* string args above */
527 {Opt_read_only, "read_only"},
528 {Opt_read_only, "ro"}, /* Alternate spelling */
529 {Opt_read_write, "read_write"},
530 {Opt_read_write, "rw"}, /* Alternate spelling */
531 /* Boolean args above */
539 #define RBD_READ_ONLY_DEFAULT false
541 static int parse_rbd_opts_token(char *c, void *private)
543 struct rbd_options *rbd_opts = private;
544 substring_t argstr[MAX_OPT_ARGS];
545 int token, intval, ret;
547 token = match_token(c, rbd_opts_tokens, argstr);
551 if (token < Opt_last_int) {
552 ret = match_int(&argstr[0], &intval);
554 pr_err("bad mount option arg (not int) "
558 dout("got int token %d val %d\n", token, intval);
559 } else if (token > Opt_last_int && token < Opt_last_string) {
560 dout("got string token %d val %s\n", token,
562 } else if (token > Opt_last_string && token < Opt_last_bool) {
563 dout("got Boolean token %d\n", token);
565 dout("got token %d\n", token);
570 rbd_opts->read_only = true;
573 rbd_opts->read_only = false;
583 * Get a ceph client with specific addr and configuration, if one does
584 * not exist create it.
586 static struct rbd_client *rbd_get_client(struct ceph_options *ceph_opts)
588 struct rbd_client *rbdc;
590 rbdc = rbd_client_find(ceph_opts);
591 if (rbdc) /* using an existing client */
592 ceph_destroy_options(ceph_opts);
594 rbdc = rbd_client_create(ceph_opts);
600 * Destroy ceph client
602 * Caller must hold rbd_client_list_lock.
604 static void rbd_client_release(struct kref *kref)
606 struct rbd_client *rbdc = container_of(kref, struct rbd_client, kref);
608 dout("rbd_release_client %p\n", rbdc);
609 spin_lock(&rbd_client_list_lock);
610 list_del(&rbdc->node);
611 spin_unlock(&rbd_client_list_lock);
613 ceph_destroy_client(rbdc->client);
618 * Drop reference to ceph client node. If it's not referenced anymore, release
621 static void rbd_put_client(struct rbd_client *rbdc)
624 kref_put(&rbdc->kref, rbd_client_release);
627 static bool rbd_image_format_valid(u32 image_format)
629 return image_format == 1 || image_format == 2;
632 static bool rbd_dev_ondisk_valid(struct rbd_image_header_ondisk *ondisk)
637 /* The header has to start with the magic rbd header text */
638 if (memcmp(&ondisk->text, RBD_HEADER_TEXT, sizeof (RBD_HEADER_TEXT)))
641 /* The bio layer requires at least sector-sized I/O */
643 if (ondisk->options.order < SECTOR_SHIFT)
646 /* If we use u64 in a few spots we may be able to loosen this */
648 if (ondisk->options.order > 8 * sizeof (int) - 1)
652 * The size of a snapshot header has to fit in a size_t, and
653 * that limits the number of snapshots.
655 snap_count = le32_to_cpu(ondisk->snap_count);
656 size = SIZE_MAX - sizeof (struct ceph_snap_context);
657 if (snap_count > size / sizeof (__le64))
661 * Not only that, but the size of the entire the snapshot
662 * header must also be representable in a size_t.
664 size -= snap_count * sizeof (__le64);
665 if ((u64) size < le64_to_cpu(ondisk->snap_names_len))
672 * Create a new header structure, translate header format from the on-disk
675 static int rbd_header_from_disk(struct rbd_image_header *header,
676 struct rbd_image_header_ondisk *ondisk)
683 memset(header, 0, sizeof (*header));
685 snap_count = le32_to_cpu(ondisk->snap_count);
687 len = strnlen(ondisk->object_prefix, sizeof (ondisk->object_prefix));
688 header->object_prefix = kmalloc(len + 1, GFP_KERNEL);
689 if (!header->object_prefix)
691 memcpy(header->object_prefix, ondisk->object_prefix, len);
692 header->object_prefix[len] = '\0';
695 u64 snap_names_len = le64_to_cpu(ondisk->snap_names_len);
697 /* Save a copy of the snapshot names */
699 if (snap_names_len > (u64) SIZE_MAX)
701 header->snap_names = kmalloc(snap_names_len, GFP_KERNEL);
702 if (!header->snap_names)
705 * Note that rbd_dev_v1_header_read() guarantees
706 * the ondisk buffer we're working with has
707 * snap_names_len bytes beyond the end of the
708 * snapshot id array, this memcpy() is safe.
710 memcpy(header->snap_names, &ondisk->snaps[snap_count],
713 /* Record each snapshot's size */
715 size = snap_count * sizeof (*header->snap_sizes);
716 header->snap_sizes = kmalloc(size, GFP_KERNEL);
717 if (!header->snap_sizes)
719 for (i = 0; i < snap_count; i++)
720 header->snap_sizes[i] =
721 le64_to_cpu(ondisk->snaps[i].image_size);
723 WARN_ON(ondisk->snap_names_len);
724 header->snap_names = NULL;
725 header->snap_sizes = NULL;
728 header->features = 0; /* No features support in v1 images */
729 header->obj_order = ondisk->options.order;
730 header->crypt_type = ondisk->options.crypt_type;
731 header->comp_type = ondisk->options.comp_type;
733 /* Allocate and fill in the snapshot context */
735 header->image_size = le64_to_cpu(ondisk->image_size);
736 size = sizeof (struct ceph_snap_context);
737 size += snap_count * sizeof (header->snapc->snaps[0]);
738 header->snapc = kzalloc(size, GFP_KERNEL);
742 atomic_set(&header->snapc->nref, 1);
743 header->snapc->seq = le64_to_cpu(ondisk->snap_seq);
744 header->snapc->num_snaps = snap_count;
745 for (i = 0; i < snap_count; i++)
746 header->snapc->snaps[i] =
747 le64_to_cpu(ondisk->snaps[i].id);
752 kfree(header->snap_sizes);
753 header->snap_sizes = NULL;
754 kfree(header->snap_names);
755 header->snap_names = NULL;
756 kfree(header->object_prefix);
757 header->object_prefix = NULL;
762 static const char *rbd_snap_name(struct rbd_device *rbd_dev, u64 snap_id)
764 struct rbd_snap *snap;
766 if (snap_id == CEPH_NOSNAP)
767 return RBD_SNAP_HEAD_NAME;
769 list_for_each_entry(snap, &rbd_dev->snaps, node)
770 if (snap_id == snap->id)
776 static int snap_by_name(struct rbd_device *rbd_dev, const char *snap_name)
779 struct rbd_snap *snap;
781 list_for_each_entry(snap, &rbd_dev->snaps, node) {
782 if (!strcmp(snap_name, snap->name)) {
783 rbd_dev->spec->snap_id = snap->id;
784 rbd_dev->mapping.size = snap->size;
785 rbd_dev->mapping.features = snap->features;
794 static int rbd_dev_set_mapping(struct rbd_device *rbd_dev)
798 if (!memcmp(rbd_dev->spec->snap_name, RBD_SNAP_HEAD_NAME,
799 sizeof (RBD_SNAP_HEAD_NAME))) {
800 rbd_dev->spec->snap_id = CEPH_NOSNAP;
801 rbd_dev->mapping.size = rbd_dev->header.image_size;
802 rbd_dev->mapping.features = rbd_dev->header.features;
805 ret = snap_by_name(rbd_dev, rbd_dev->spec->snap_name);
808 rbd_dev->mapping.read_only = true;
810 set_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags);
816 static void rbd_header_free(struct rbd_image_header *header)
818 kfree(header->object_prefix);
819 header->object_prefix = NULL;
820 kfree(header->snap_sizes);
821 header->snap_sizes = NULL;
822 kfree(header->snap_names);
823 header->snap_names = NULL;
824 ceph_put_snap_context(header->snapc);
825 header->snapc = NULL;
828 static const char *rbd_segment_name(struct rbd_device *rbd_dev, u64 offset)
834 name = kmalloc(MAX_OBJ_NAME_SIZE + 1, GFP_NOIO);
837 segment = offset >> rbd_dev->header.obj_order;
838 ret = snprintf(name, MAX_OBJ_NAME_SIZE + 1, "%s.%012llx",
839 rbd_dev->header.object_prefix, segment);
840 if (ret < 0 || ret > MAX_OBJ_NAME_SIZE) {
841 pr_err("error formatting segment name for #%llu (%d)\n",
850 static u64 rbd_segment_offset(struct rbd_device *rbd_dev, u64 offset)
852 u64 segment_size = (u64) 1 << rbd_dev->header.obj_order;
854 return offset & (segment_size - 1);
857 static u64 rbd_segment_length(struct rbd_device *rbd_dev,
858 u64 offset, u64 length)
860 u64 segment_size = (u64) 1 << rbd_dev->header.obj_order;
862 offset &= segment_size - 1;
864 rbd_assert(length <= U64_MAX - offset);
865 if (offset + length > segment_size)
866 length = segment_size - offset;
872 * returns the size of an object in the image
874 static u64 rbd_obj_bytes(struct rbd_image_header *header)
876 return 1 << header->obj_order;
883 static void bio_chain_put(struct bio *chain)
889 chain = chain->bi_next;
895 * zeros a bio chain, starting at specific offset
897 static void zero_bio_chain(struct bio *chain, int start_ofs)
906 bio_for_each_segment(bv, chain, i) {
907 if (pos + bv->bv_len > start_ofs) {
908 int remainder = max(start_ofs - pos, 0);
909 buf = bvec_kmap_irq(bv, &flags);
910 memset(buf + remainder, 0,
911 bv->bv_len - remainder);
912 bvec_kunmap_irq(buf, &flags);
917 chain = chain->bi_next;
922 * Clone a portion of a bio, starting at the given byte offset
923 * and continuing for the number of bytes indicated.
925 static struct bio *bio_clone_range(struct bio *bio_src,
934 unsigned short end_idx;
938 /* Handle the easy case for the caller */
940 if (!offset && len == bio_src->bi_size)
941 return bio_clone(bio_src, gfpmask);
943 if (WARN_ON_ONCE(!len))
945 if (WARN_ON_ONCE(len > bio_src->bi_size))
947 if (WARN_ON_ONCE(offset > bio_src->bi_size - len))
950 /* Find first affected segment... */
953 __bio_for_each_segment(bv, bio_src, idx, 0) {
954 if (resid < bv->bv_len)
960 /* ...and the last affected segment */
963 __bio_for_each_segment(bv, bio_src, end_idx, idx) {
964 if (resid <= bv->bv_len)
968 vcnt = end_idx - idx + 1;
970 /* Build the clone */
972 bio = bio_alloc(gfpmask, (unsigned int) vcnt);
974 return NULL; /* ENOMEM */
976 bio->bi_bdev = bio_src->bi_bdev;
977 bio->bi_sector = bio_src->bi_sector + (offset >> SECTOR_SHIFT);
978 bio->bi_rw = bio_src->bi_rw;
979 bio->bi_flags |= 1 << BIO_CLONED;
982 * Copy over our part of the bio_vec, then update the first
983 * and last (or only) entries.
985 memcpy(&bio->bi_io_vec[0], &bio_src->bi_io_vec[idx],
986 vcnt * sizeof (struct bio_vec));
987 bio->bi_io_vec[0].bv_offset += voff;
989 bio->bi_io_vec[0].bv_len -= voff;
990 bio->bi_io_vec[vcnt - 1].bv_len = resid;
992 bio->bi_io_vec[0].bv_len = len;
1003 * Clone a portion of a bio chain, starting at the given byte offset
1004 * into the first bio in the source chain and continuing for the
1005 * number of bytes indicated. The result is another bio chain of
1006 * exactly the given length, or a null pointer on error.
1008 * The bio_src and offset parameters are both in-out. On entry they
1009 * refer to the first source bio and the offset into that bio where
1010 * the start of data to be cloned is located.
1012 * On return, bio_src is updated to refer to the bio in the source
1013 * chain that contains first un-cloned byte, and *offset will
1014 * contain the offset of that byte within that bio.
1016 static struct bio *bio_chain_clone_range(struct bio **bio_src,
1017 unsigned int *offset,
1021 struct bio *bi = *bio_src;
1022 unsigned int off = *offset;
1023 struct bio *chain = NULL;
1026 /* Build up a chain of clone bios up to the limit */
1028 if (!bi || off >= bi->bi_size || !len)
1029 return NULL; /* Nothing to clone */
1033 unsigned int bi_size;
1037 rbd_warn(NULL, "bio_chain exhausted with %u left", len);
1038 goto out_err; /* EINVAL; ran out of bio's */
1040 bi_size = min_t(unsigned int, bi->bi_size - off, len);
1041 bio = bio_clone_range(bi, off, bi_size, gfpmask);
1043 goto out_err; /* ENOMEM */
1046 end = &bio->bi_next;
1049 if (off == bi->bi_size) {
1060 bio_chain_put(chain);
1065 static void rbd_obj_request_get(struct rbd_obj_request *obj_request)
1067 kref_get(&obj_request->kref);
1070 static void rbd_obj_request_destroy(struct kref *kref);
1071 static void rbd_obj_request_put(struct rbd_obj_request *obj_request)
1073 rbd_assert(obj_request != NULL);
1074 kref_put(&obj_request->kref, rbd_obj_request_destroy);
1077 static void rbd_img_request_get(struct rbd_img_request *img_request)
1079 kref_get(&img_request->kref);
1082 static void rbd_img_request_destroy(struct kref *kref);
1083 static void rbd_img_request_put(struct rbd_img_request *img_request)
1085 rbd_assert(img_request != NULL);
1086 kref_put(&img_request->kref, rbd_img_request_destroy);
1089 static inline void rbd_img_obj_request_add(struct rbd_img_request *img_request,
1090 struct rbd_obj_request *obj_request)
1092 rbd_assert(obj_request->img_request == NULL);
1094 rbd_obj_request_get(obj_request);
1095 obj_request->img_request = img_request;
1096 obj_request->which = img_request->obj_request_count;
1097 rbd_assert(obj_request->which != BAD_WHICH);
1098 img_request->obj_request_count++;
1099 list_add_tail(&obj_request->links, &img_request->obj_requests);
1102 static inline void rbd_img_obj_request_del(struct rbd_img_request *img_request,
1103 struct rbd_obj_request *obj_request)
1105 rbd_assert(obj_request->which != BAD_WHICH);
1107 list_del(&obj_request->links);
1108 rbd_assert(img_request->obj_request_count > 0);
1109 img_request->obj_request_count--;
1110 rbd_assert(obj_request->which == img_request->obj_request_count);
1111 obj_request->which = BAD_WHICH;
1112 rbd_assert(obj_request->img_request == img_request);
1113 obj_request->img_request = NULL;
1114 obj_request->callback = NULL;
1115 rbd_obj_request_put(obj_request);
1118 static bool obj_request_type_valid(enum obj_request_type type)
1121 case OBJ_REQUEST_NODATA:
1122 case OBJ_REQUEST_BIO:
1123 case OBJ_REQUEST_PAGES:
1130 struct ceph_osd_req_op *rbd_osd_req_op_create(u16 opcode, ...)
1132 struct ceph_osd_req_op *op;
1136 op = kzalloc(sizeof (*op), GFP_NOIO);
1140 va_start(args, opcode);
1142 case CEPH_OSD_OP_READ:
1143 case CEPH_OSD_OP_WRITE:
1144 /* rbd_osd_req_op_create(READ, offset, length) */
1145 /* rbd_osd_req_op_create(WRITE, offset, length) */
1146 op->extent.offset = va_arg(args, u64);
1147 op->extent.length = va_arg(args, u64);
1148 if (opcode == CEPH_OSD_OP_WRITE)
1149 op->payload_len = op->extent.length;
1151 case CEPH_OSD_OP_STAT:
1153 case CEPH_OSD_OP_CALL:
1154 /* rbd_osd_req_op_create(CALL, class, method, data, datalen) */
1155 op->cls.class_name = va_arg(args, char *);
1156 size = strlen(op->cls.class_name);
1157 rbd_assert(size <= (size_t) U8_MAX);
1158 op->cls.class_len = size;
1159 op->payload_len = size;
1161 op->cls.method_name = va_arg(args, char *);
1162 size = strlen(op->cls.method_name);
1163 rbd_assert(size <= (size_t) U8_MAX);
1164 op->cls.method_len = size;
1165 op->payload_len += size;
1168 op->cls.indata = va_arg(args, void *);
1169 size = va_arg(args, size_t);
1170 rbd_assert(size <= (size_t) U32_MAX);
1171 op->cls.indata_len = (u32) size;
1172 op->payload_len += size;
1174 case CEPH_OSD_OP_NOTIFY_ACK:
1175 case CEPH_OSD_OP_WATCH:
1176 /* rbd_osd_req_op_create(NOTIFY_ACK, cookie, version) */
1177 /* rbd_osd_req_op_create(WATCH, cookie, version, flag) */
1178 op->watch.cookie = va_arg(args, u64);
1179 op->watch.ver = va_arg(args, u64);
1180 op->watch.ver = cpu_to_le64(op->watch.ver);
1181 if (opcode == CEPH_OSD_OP_WATCH && va_arg(args, int))
1182 op->watch.flag = (u8) 1;
1185 rbd_warn(NULL, "unsupported opcode %hu\n", opcode);
1195 static void rbd_osd_req_op_destroy(struct ceph_osd_req_op *op)
1200 static int rbd_obj_request_submit(struct ceph_osd_client *osdc,
1201 struct rbd_obj_request *obj_request)
1203 return ceph_osdc_start_request(osdc, obj_request->osd_req, false);
1206 static void rbd_img_request_complete(struct rbd_img_request *img_request)
1208 if (img_request->callback)
1209 img_request->callback(img_request);
1211 rbd_img_request_put(img_request);
1214 /* Caller is responsible for rbd_obj_request_destroy(obj_request) */
1216 static int rbd_obj_request_wait(struct rbd_obj_request *obj_request)
1218 return wait_for_completion_interruptible(&obj_request->completion);
1221 static void obj_request_done_init(struct rbd_obj_request *obj_request)
1223 atomic_set(&obj_request->done, 0);
1227 static void obj_request_done_set(struct rbd_obj_request *obj_request)
1229 atomic_set(&obj_request->done, 1);
1233 static bool obj_request_done_test(struct rbd_obj_request *obj_request)
1236 return atomic_read(&obj_request->done) != 0;
1239 static void rbd_osd_trivial_callback(struct rbd_obj_request *obj_request,
1240 struct ceph_osd_op *op)
1242 obj_request_done_set(obj_request);
1245 static void rbd_obj_request_complete(struct rbd_obj_request *obj_request)
1247 if (obj_request->callback)
1248 obj_request->callback(obj_request);
1250 complete_all(&obj_request->completion);
1253 static void rbd_osd_read_callback(struct rbd_obj_request *obj_request,
1254 struct ceph_osd_op *op)
1259 * We support a 64-bit length, but ultimately it has to be
1260 * passed to blk_end_request(), which takes an unsigned int.
1262 xferred = le64_to_cpu(op->extent.length);
1263 rbd_assert(xferred < (u64) UINT_MAX);
1264 if (obj_request->result == (s32) -ENOENT) {
1265 zero_bio_chain(obj_request->bio_list, 0);
1266 obj_request->result = 0;
1267 } else if (xferred < obj_request->length && !obj_request->result) {
1268 zero_bio_chain(obj_request->bio_list, xferred);
1269 xferred = obj_request->length;
1271 obj_request->xferred = xferred;
1272 obj_request_done_set(obj_request);
1275 static void rbd_osd_write_callback(struct rbd_obj_request *obj_request,
1276 struct ceph_osd_op *op)
1278 obj_request->xferred = le64_to_cpu(op->extent.length);
1279 obj_request_done_set(obj_request);
1283 * For a simple stat call there's nothing to do. We'll do more if
1284 * this is part of a write sequence for a layered image.
1286 static void rbd_osd_stat_callback(struct rbd_obj_request *obj_request,
1287 struct ceph_osd_op *op)
1289 obj_request_done_set(obj_request);
1292 static void rbd_osd_req_callback(struct ceph_osd_request *osd_req,
1293 struct ceph_msg *msg)
1295 struct rbd_obj_request *obj_request = osd_req->r_priv;
1296 struct ceph_osd_reply_head *reply_head;
1297 struct ceph_osd_op *op;
1301 rbd_assert(osd_req == obj_request->osd_req);
1302 rbd_assert(!!obj_request->img_request ^
1303 (obj_request->which == BAD_WHICH));
1305 obj_request->xferred = le32_to_cpu(msg->hdr.data_len);
1306 reply_head = msg->front.iov_base;
1307 obj_request->result = (s32) le32_to_cpu(reply_head->result);
1308 obj_request->version = le64_to_cpu(osd_req->r_reassert_version.version);
1310 num_ops = le32_to_cpu(reply_head->num_ops);
1311 WARN_ON(num_ops != 1); /* For now */
1313 op = &reply_head->ops[0];
1314 opcode = le16_to_cpu(op->op);
1316 case CEPH_OSD_OP_READ:
1317 rbd_osd_read_callback(obj_request, op);
1319 case CEPH_OSD_OP_WRITE:
1320 rbd_osd_write_callback(obj_request, op);
1322 case CEPH_OSD_OP_STAT:
1323 rbd_osd_stat_callback(obj_request, op);
1325 case CEPH_OSD_OP_CALL:
1326 case CEPH_OSD_OP_NOTIFY_ACK:
1327 case CEPH_OSD_OP_WATCH:
1328 rbd_osd_trivial_callback(obj_request, op);
1331 rbd_warn(NULL, "%s: unsupported op %hu\n",
1332 obj_request->object_name, (unsigned short) opcode);
1336 if (obj_request_done_test(obj_request))
1337 rbd_obj_request_complete(obj_request);
1340 static struct ceph_osd_request *rbd_osd_req_create(
1341 struct rbd_device *rbd_dev,
1343 struct rbd_obj_request *obj_request,
1344 struct ceph_osd_req_op *op)
1346 struct rbd_img_request *img_request = obj_request->img_request;
1347 struct ceph_snap_context *snapc = NULL;
1348 struct ceph_osd_client *osdc;
1349 struct ceph_osd_request *osd_req;
1350 struct timespec now;
1351 struct timespec *mtime;
1352 u64 snap_id = CEPH_NOSNAP;
1353 u64 offset = obj_request->offset;
1354 u64 length = obj_request->length;
1357 rbd_assert(img_request->write_request == write_request);
1358 if (img_request->write_request)
1359 snapc = img_request->snapc;
1361 snap_id = img_request->snap_id;
1364 /* Allocate and initialize the request, for the single op */
1366 osdc = &rbd_dev->rbd_client->client->osdc;
1367 osd_req = ceph_osdc_alloc_request(osdc, snapc, 1, false, GFP_ATOMIC);
1369 return NULL; /* ENOMEM */
1371 rbd_assert(obj_request_type_valid(obj_request->type));
1372 switch (obj_request->type) {
1373 case OBJ_REQUEST_NODATA:
1374 break; /* Nothing to do */
1375 case OBJ_REQUEST_BIO:
1376 rbd_assert(obj_request->bio_list != NULL);
1377 osd_req->r_bio = obj_request->bio_list;
1379 case OBJ_REQUEST_PAGES:
1380 osd_req->r_pages = obj_request->pages;
1381 osd_req->r_num_pages = obj_request->page_count;
1382 osd_req->r_page_alignment = offset & ~PAGE_MASK;
1386 if (write_request) {
1387 osd_req->r_flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK;
1391 osd_req->r_flags = CEPH_OSD_FLAG_READ;
1392 mtime = NULL; /* not needed for reads */
1393 offset = 0; /* These are not used... */
1394 length = 0; /* ...for osd read requests */
1397 osd_req->r_callback = rbd_osd_req_callback;
1398 osd_req->r_priv = obj_request;
1400 osd_req->r_oid_len = strlen(obj_request->object_name);
1401 rbd_assert(osd_req->r_oid_len < sizeof (osd_req->r_oid));
1402 memcpy(osd_req->r_oid, obj_request->object_name, osd_req->r_oid_len);
1404 osd_req->r_file_layout = rbd_dev->layout; /* struct */
1406 /* osd_req will get its own reference to snapc (if non-null) */
1408 ceph_osdc_build_request(osd_req, offset, length, 1, op,
1409 snapc, snap_id, mtime);
1414 static void rbd_osd_req_destroy(struct ceph_osd_request *osd_req)
1416 ceph_osdc_put_request(osd_req);
1419 /* object_name is assumed to be a non-null pointer and NUL-terminated */
1421 static struct rbd_obj_request *rbd_obj_request_create(const char *object_name,
1422 u64 offset, u64 length,
1423 enum obj_request_type type)
1425 struct rbd_obj_request *obj_request;
1429 rbd_assert(obj_request_type_valid(type));
1431 size = strlen(object_name) + 1;
1432 obj_request = kzalloc(sizeof (*obj_request) + size, GFP_KERNEL);
1436 name = (char *)(obj_request + 1);
1437 obj_request->object_name = memcpy(name, object_name, size);
1438 obj_request->offset = offset;
1439 obj_request->length = length;
1440 obj_request->which = BAD_WHICH;
1441 obj_request->type = type;
1442 INIT_LIST_HEAD(&obj_request->links);
1443 obj_request_done_init(obj_request);
1444 init_completion(&obj_request->completion);
1445 kref_init(&obj_request->kref);
1450 static void rbd_obj_request_destroy(struct kref *kref)
1452 struct rbd_obj_request *obj_request;
1454 obj_request = container_of(kref, struct rbd_obj_request, kref);
1456 rbd_assert(obj_request->img_request == NULL);
1457 rbd_assert(obj_request->which == BAD_WHICH);
1459 if (obj_request->osd_req)
1460 rbd_osd_req_destroy(obj_request->osd_req);
1462 rbd_assert(obj_request_type_valid(obj_request->type));
1463 switch (obj_request->type) {
1464 case OBJ_REQUEST_NODATA:
1465 break; /* Nothing to do */
1466 case OBJ_REQUEST_BIO:
1467 if (obj_request->bio_list)
1468 bio_chain_put(obj_request->bio_list);
1470 case OBJ_REQUEST_PAGES:
1471 if (obj_request->pages)
1472 ceph_release_page_vector(obj_request->pages,
1473 obj_request->page_count);
1481 * Caller is responsible for filling in the list of object requests
1482 * that comprises the image request, and the Linux request pointer
1483 * (if there is one).
1485 struct rbd_img_request *rbd_img_request_create(struct rbd_device *rbd_dev,
1486 u64 offset, u64 length,
1489 struct rbd_img_request *img_request;
1490 struct ceph_snap_context *snapc = NULL;
1492 img_request = kmalloc(sizeof (*img_request), GFP_ATOMIC);
1496 if (write_request) {
1497 down_read(&rbd_dev->header_rwsem);
1498 snapc = ceph_get_snap_context(rbd_dev->header.snapc);
1499 up_read(&rbd_dev->header_rwsem);
1500 if (WARN_ON(!snapc)) {
1502 return NULL; /* Shouldn't happen */
1506 img_request->rq = NULL;
1507 img_request->rbd_dev = rbd_dev;
1508 img_request->offset = offset;
1509 img_request->length = length;
1510 img_request->write_request = write_request;
1512 img_request->snapc = snapc;
1514 img_request->snap_id = rbd_dev->spec->snap_id;
1515 spin_lock_init(&img_request->completion_lock);
1516 img_request->next_completion = 0;
1517 img_request->callback = NULL;
1518 img_request->obj_request_count = 0;
1519 INIT_LIST_HEAD(&img_request->obj_requests);
1520 kref_init(&img_request->kref);
1522 rbd_img_request_get(img_request); /* Avoid a warning */
1523 rbd_img_request_put(img_request); /* TEMPORARY */
1528 static void rbd_img_request_destroy(struct kref *kref)
1530 struct rbd_img_request *img_request;
1531 struct rbd_obj_request *obj_request;
1532 struct rbd_obj_request *next_obj_request;
1534 img_request = container_of(kref, struct rbd_img_request, kref);
1536 for_each_obj_request_safe(img_request, obj_request, next_obj_request)
1537 rbd_img_obj_request_del(img_request, obj_request);
1538 rbd_assert(img_request->obj_request_count == 0);
1540 if (img_request->write_request)
1541 ceph_put_snap_context(img_request->snapc);
1546 static int rbd_img_request_fill_bio(struct rbd_img_request *img_request,
1547 struct bio *bio_list)
1549 struct rbd_device *rbd_dev = img_request->rbd_dev;
1550 struct rbd_obj_request *obj_request = NULL;
1551 struct rbd_obj_request *next_obj_request;
1552 unsigned int bio_offset;
1557 opcode = img_request->write_request ? CEPH_OSD_OP_WRITE
1560 image_offset = img_request->offset;
1561 rbd_assert(image_offset == bio_list->bi_sector << SECTOR_SHIFT);
1562 resid = img_request->length;
1564 const char *object_name;
1565 unsigned int clone_size;
1566 struct ceph_osd_req_op *op;
1570 object_name = rbd_segment_name(rbd_dev, image_offset);
1573 offset = rbd_segment_offset(rbd_dev, image_offset);
1574 length = rbd_segment_length(rbd_dev, image_offset, resid);
1575 obj_request = rbd_obj_request_create(object_name,
1578 kfree(object_name); /* object request has its own copy */
1582 rbd_assert(length <= (u64) UINT_MAX);
1583 clone_size = (unsigned int) length;
1584 obj_request->bio_list = bio_chain_clone_range(&bio_list,
1585 &bio_offset, clone_size,
1587 if (!obj_request->bio_list)
1591 * Build up the op to use in building the osd
1592 * request. Note that the contents of the op are
1593 * copied by rbd_osd_req_create().
1595 op = rbd_osd_req_op_create(opcode, offset, length);
1598 obj_request->osd_req = rbd_osd_req_create(rbd_dev,
1599 img_request->write_request,
1601 rbd_osd_req_op_destroy(op);
1602 if (!obj_request->osd_req)
1604 /* status and version are initially zero-filled */
1606 rbd_img_obj_request_add(img_request, obj_request);
1608 image_offset += length;
1615 rbd_obj_request_put(obj_request);
1617 for_each_obj_request_safe(img_request, obj_request, next_obj_request)
1618 rbd_obj_request_put(obj_request);
1623 static void rbd_img_obj_callback(struct rbd_obj_request *obj_request)
1625 struct rbd_img_request *img_request;
1626 u32 which = obj_request->which;
1629 img_request = obj_request->img_request;
1630 rbd_assert(img_request != NULL);
1631 rbd_assert(img_request->rq != NULL);
1632 rbd_assert(which != BAD_WHICH);
1633 rbd_assert(which < img_request->obj_request_count);
1634 rbd_assert(which >= img_request->next_completion);
1636 spin_lock_irq(&img_request->completion_lock);
1637 if (which != img_request->next_completion)
1640 for_each_obj_request_from(img_request, obj_request) {
1641 unsigned int xferred;
1645 rbd_assert(which < img_request->obj_request_count);
1647 if (!obj_request_done_test(obj_request))
1650 rbd_assert(obj_request->xferred <= (u64) UINT_MAX);
1651 xferred = (unsigned int) obj_request->xferred;
1652 result = (int) obj_request->result;
1654 rbd_warn(NULL, "obj_request %s result %d xferred %u\n",
1655 img_request->write_request ? "write" : "read",
1658 more = blk_end_request(img_request->rq, result, xferred);
1661 rbd_assert(more ^ (which == img_request->obj_request_count));
1662 img_request->next_completion = which;
1664 spin_unlock_irq(&img_request->completion_lock);
1667 rbd_img_request_complete(img_request);
1670 static int rbd_img_request_submit(struct rbd_img_request *img_request)
1672 struct rbd_device *rbd_dev = img_request->rbd_dev;
1673 struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
1674 struct rbd_obj_request *obj_request;
1676 for_each_obj_request(img_request, obj_request) {
1679 obj_request->callback = rbd_img_obj_callback;
1680 ret = rbd_obj_request_submit(osdc, obj_request);
1684 * The image request has its own reference to each
1685 * of its object requests, so we can safely drop the
1688 rbd_obj_request_put(obj_request);
1694 static int rbd_obj_notify_ack(struct rbd_device *rbd_dev,
1695 u64 ver, u64 notify_id)
1697 struct rbd_obj_request *obj_request;
1698 struct ceph_osd_req_op *op;
1699 struct ceph_osd_client *osdc;
1702 obj_request = rbd_obj_request_create(rbd_dev->header_name, 0, 0,
1703 OBJ_REQUEST_NODATA);
1708 op = rbd_osd_req_op_create(CEPH_OSD_OP_NOTIFY_ACK, notify_id, ver);
1711 obj_request->osd_req = rbd_osd_req_create(rbd_dev, false,
1713 rbd_osd_req_op_destroy(op);
1714 if (!obj_request->osd_req)
1717 osdc = &rbd_dev->rbd_client->client->osdc;
1718 obj_request->callback = rbd_obj_request_put;
1719 ret = rbd_obj_request_submit(osdc, obj_request);
1722 rbd_obj_request_put(obj_request);
1727 static void rbd_watch_cb(u64 ver, u64 notify_id, u8 opcode, void *data)
1729 struct rbd_device *rbd_dev = (struct rbd_device *)data;
1736 dout("rbd_watch_cb %s notify_id=%llu opcode=%u\n",
1737 rbd_dev->header_name, (unsigned long long) notify_id,
1738 (unsigned int) opcode);
1739 rc = rbd_dev_refresh(rbd_dev, &hver);
1741 rbd_warn(rbd_dev, "got notification but failed to "
1742 " update snaps: %d\n", rc);
1744 rbd_obj_notify_ack(rbd_dev, hver, notify_id);
1748 * Request sync osd watch/unwatch. The value of "start" determines
1749 * whether a watch request is being initiated or torn down.
1751 static int rbd_dev_header_watch_sync(struct rbd_device *rbd_dev, int start)
1753 struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
1754 struct rbd_obj_request *obj_request;
1755 struct ceph_osd_req_op *op;
1758 rbd_assert(start ^ !!rbd_dev->watch_event);
1759 rbd_assert(start ^ !!rbd_dev->watch_request);
1762 ret = ceph_osdc_create_event(osdc, rbd_watch_cb, rbd_dev,
1763 &rbd_dev->watch_event);
1766 rbd_assert(rbd_dev->watch_event != NULL);
1770 obj_request = rbd_obj_request_create(rbd_dev->header_name, 0, 0,
1771 OBJ_REQUEST_NODATA);
1775 op = rbd_osd_req_op_create(CEPH_OSD_OP_WATCH,
1776 rbd_dev->watch_event->cookie,
1777 rbd_dev->header.obj_version, start);
1780 obj_request->osd_req = rbd_osd_req_create(rbd_dev, true,
1782 rbd_osd_req_op_destroy(op);
1783 if (!obj_request->osd_req)
1787 ceph_osdc_set_request_linger(osdc, obj_request->osd_req);
1789 ceph_osdc_unregister_linger_request(osdc,
1790 rbd_dev->watch_request->osd_req);
1791 ret = rbd_obj_request_submit(osdc, obj_request);
1794 ret = rbd_obj_request_wait(obj_request);
1797 ret = obj_request->result;
1802 * A watch request is set to linger, so the underlying osd
1803 * request won't go away until we unregister it. We retain
1804 * a pointer to the object request during that time (in
1805 * rbd_dev->watch_request), so we'll keep a reference to
1806 * it. We'll drop that reference (below) after we've
1810 rbd_dev->watch_request = obj_request;
1815 /* We have successfully torn down the watch request */
1817 rbd_obj_request_put(rbd_dev->watch_request);
1818 rbd_dev->watch_request = NULL;
1820 /* Cancel the event if we're tearing down, or on error */
1821 ceph_osdc_cancel_event(rbd_dev->watch_event);
1822 rbd_dev->watch_event = NULL;
1824 rbd_obj_request_put(obj_request);
1830 * Synchronous osd object method call
1832 static int rbd_obj_method_sync(struct rbd_device *rbd_dev,
1833 const char *object_name,
1834 const char *class_name,
1835 const char *method_name,
1836 const char *outbound,
1837 size_t outbound_size,
1839 size_t inbound_size,
1842 struct rbd_obj_request *obj_request;
1843 struct ceph_osd_client *osdc;
1844 struct ceph_osd_req_op *op;
1845 struct page **pages;
1850 * Method calls are ultimately read operations but they
1851 * don't involve object data (so no offset or length).
1852 * The result should placed into the inbound buffer
1853 * provided. They also supply outbound data--parameters for
1854 * the object method. Currently if this is present it will
1857 page_count = (u32) calc_pages_for(0, inbound_size);
1858 pages = ceph_alloc_page_vector(page_count, GFP_KERNEL);
1860 return PTR_ERR(pages);
1863 obj_request = rbd_obj_request_create(object_name, 0, 0,
1868 obj_request->pages = pages;
1869 obj_request->page_count = page_count;
1871 op = rbd_osd_req_op_create(CEPH_OSD_OP_CALL, class_name,
1872 method_name, outbound, outbound_size);
1875 obj_request->osd_req = rbd_osd_req_create(rbd_dev, false,
1877 rbd_osd_req_op_destroy(op);
1878 if (!obj_request->osd_req)
1881 osdc = &rbd_dev->rbd_client->client->osdc;
1882 ret = rbd_obj_request_submit(osdc, obj_request);
1885 ret = rbd_obj_request_wait(obj_request);
1889 ret = obj_request->result;
1893 (void) ceph_copy_from_page_vector(pages, inbound, 0,
1894 obj_request->xferred);
1896 *version = obj_request->version;
1899 rbd_obj_request_put(obj_request);
1901 ceph_release_page_vector(pages, page_count);
1906 static void rbd_request_fn(struct request_queue *q)
1908 struct rbd_device *rbd_dev = q->queuedata;
1909 bool read_only = rbd_dev->mapping.read_only;
1913 while ((rq = blk_fetch_request(q))) {
1914 bool write_request = rq_data_dir(rq) == WRITE;
1915 struct rbd_img_request *img_request;
1919 /* Ignore any non-FS requests that filter through. */
1921 if (rq->cmd_type != REQ_TYPE_FS) {
1922 __blk_end_request_all(rq, 0);
1926 spin_unlock_irq(q->queue_lock);
1928 /* Disallow writes to a read-only device */
1930 if (write_request) {
1934 rbd_assert(rbd_dev->spec->snap_id == CEPH_NOSNAP);
1938 * Quit early if the mapped snapshot no longer
1939 * exists. It's still possible the snapshot will
1940 * have disappeared by the time our request arrives
1941 * at the osd, but there's no sense in sending it if
1944 if (!test_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags)) {
1945 dout("request for non-existent snapshot");
1946 rbd_assert(rbd_dev->spec->snap_id != CEPH_NOSNAP);
1951 offset = (u64) blk_rq_pos(rq) << SECTOR_SHIFT;
1952 length = (u64) blk_rq_bytes(rq);
1955 if (WARN_ON(offset && length > U64_MAX - offset + 1))
1956 goto end_request; /* Shouldn't happen */
1959 img_request = rbd_img_request_create(rbd_dev, offset, length,
1964 img_request->rq = rq;
1966 result = rbd_img_request_fill_bio(img_request, rq->bio);
1968 result = rbd_img_request_submit(img_request);
1970 rbd_img_request_put(img_request);
1972 spin_lock_irq(q->queue_lock);
1974 rbd_warn(rbd_dev, "obj_request %s result %d\n",
1975 write_request ? "write" : "read", result);
1976 __blk_end_request_all(rq, result);
1982 * a queue callback. Makes sure that we don't create a bio that spans across
1983 * multiple osd objects. One exception would be with a single page bios,
1984 * which we handle later at bio_chain_clone_range()
1986 static int rbd_merge_bvec(struct request_queue *q, struct bvec_merge_data *bmd,
1987 struct bio_vec *bvec)
1989 struct rbd_device *rbd_dev = q->queuedata;
1990 sector_t sector_offset;
1991 sector_t sectors_per_obj;
1992 sector_t obj_sector_offset;
1996 * Find how far into its rbd object the partition-relative
1997 * bio start sector is to offset relative to the enclosing
2000 sector_offset = get_start_sect(bmd->bi_bdev) + bmd->bi_sector;
2001 sectors_per_obj = 1 << (rbd_dev->header.obj_order - SECTOR_SHIFT);
2002 obj_sector_offset = sector_offset & (sectors_per_obj - 1);
2005 * Compute the number of bytes from that offset to the end
2006 * of the object. Account for what's already used by the bio.
2008 ret = (int) (sectors_per_obj - obj_sector_offset) << SECTOR_SHIFT;
2009 if (ret > bmd->bi_size)
2010 ret -= bmd->bi_size;
2015 * Don't send back more than was asked for. And if the bio
2016 * was empty, let the whole thing through because: "Note
2017 * that a block device *must* allow a single page to be
2018 * added to an empty bio."
2020 rbd_assert(bvec->bv_len <= PAGE_SIZE);
2021 if (ret > (int) bvec->bv_len || !bmd->bi_size)
2022 ret = (int) bvec->bv_len;
2027 static void rbd_free_disk(struct rbd_device *rbd_dev)
2029 struct gendisk *disk = rbd_dev->disk;
2034 if (disk->flags & GENHD_FL_UP)
2037 blk_cleanup_queue(disk->queue);
2041 static int rbd_obj_read_sync(struct rbd_device *rbd_dev,
2042 const char *object_name,
2043 u64 offset, u64 length,
2044 char *buf, u64 *version)
2047 struct ceph_osd_req_op *op;
2048 struct rbd_obj_request *obj_request;
2049 struct ceph_osd_client *osdc;
2050 struct page **pages = NULL;
2055 page_count = (u32) calc_pages_for(offset, length);
2056 pages = ceph_alloc_page_vector(page_count, GFP_KERNEL);
2058 ret = PTR_ERR(pages);
2061 obj_request = rbd_obj_request_create(object_name, offset, length,
2066 obj_request->pages = pages;
2067 obj_request->page_count = page_count;
2069 op = rbd_osd_req_op_create(CEPH_OSD_OP_READ, offset, length);
2072 obj_request->osd_req = rbd_osd_req_create(rbd_dev, false,
2074 rbd_osd_req_op_destroy(op);
2075 if (!obj_request->osd_req)
2078 osdc = &rbd_dev->rbd_client->client->osdc;
2079 ret = rbd_obj_request_submit(osdc, obj_request);
2082 ret = rbd_obj_request_wait(obj_request);
2086 ret = obj_request->result;
2090 rbd_assert(obj_request->xferred <= (u64) SIZE_MAX);
2091 size = (size_t) obj_request->xferred;
2092 (void) ceph_copy_from_page_vector(pages, buf, 0, size);
2093 rbd_assert(size <= (size_t) INT_MAX);
2096 *version = obj_request->version;
2099 rbd_obj_request_put(obj_request);
2101 ceph_release_page_vector(pages, page_count);
2107 * Read the complete header for the given rbd device.
2109 * Returns a pointer to a dynamically-allocated buffer containing
2110 * the complete and validated header. Caller can pass the address
2111 * of a variable that will be filled in with the version of the
2112 * header object at the time it was read.
2114 * Returns a pointer-coded errno if a failure occurs.
2116 static struct rbd_image_header_ondisk *
2117 rbd_dev_v1_header_read(struct rbd_device *rbd_dev, u64 *version)
2119 struct rbd_image_header_ondisk *ondisk = NULL;
2126 * The complete header will include an array of its 64-bit
2127 * snapshot ids, followed by the names of those snapshots as
2128 * a contiguous block of NUL-terminated strings. Note that
2129 * the number of snapshots could change by the time we read
2130 * it in, in which case we re-read it.
2137 size = sizeof (*ondisk);
2138 size += snap_count * sizeof (struct rbd_image_snap_ondisk);
2140 ondisk = kmalloc(size, GFP_KERNEL);
2142 return ERR_PTR(-ENOMEM);
2144 ret = rbd_obj_read_sync(rbd_dev, rbd_dev->header_name,
2146 (char *) ondisk, version);
2149 if (WARN_ON((size_t) ret < size)) {
2151 rbd_warn(rbd_dev, "short header read (want %zd got %d)",
2155 if (!rbd_dev_ondisk_valid(ondisk)) {
2157 rbd_warn(rbd_dev, "invalid header");
2161 names_size = le64_to_cpu(ondisk->snap_names_len);
2162 want_count = snap_count;
2163 snap_count = le32_to_cpu(ondisk->snap_count);
2164 } while (snap_count != want_count);
2171 return ERR_PTR(ret);
2175 * reload the ondisk the header
2177 static int rbd_read_header(struct rbd_device *rbd_dev,
2178 struct rbd_image_header *header)
2180 struct rbd_image_header_ondisk *ondisk;
2184 ondisk = rbd_dev_v1_header_read(rbd_dev, &ver);
2186 return PTR_ERR(ondisk);
2187 ret = rbd_header_from_disk(header, ondisk);
2189 header->obj_version = ver;
2195 static void rbd_remove_all_snaps(struct rbd_device *rbd_dev)
2197 struct rbd_snap *snap;
2198 struct rbd_snap *next;
2200 list_for_each_entry_safe(snap, next, &rbd_dev->snaps, node)
2201 rbd_remove_snap_dev(snap);
2204 static void rbd_update_mapping_size(struct rbd_device *rbd_dev)
2208 if (rbd_dev->spec->snap_id != CEPH_NOSNAP)
2211 size = (sector_t) rbd_dev->header.image_size / SECTOR_SIZE;
2212 dout("setting size to %llu sectors", (unsigned long long) size);
2213 rbd_dev->mapping.size = (u64) size;
2214 set_capacity(rbd_dev->disk, size);
2218 * only read the first part of the ondisk header, without the snaps info
2220 static int rbd_dev_v1_refresh(struct rbd_device *rbd_dev, u64 *hver)
2223 struct rbd_image_header h;
2225 ret = rbd_read_header(rbd_dev, &h);
2229 down_write(&rbd_dev->header_rwsem);
2231 /* Update image size, and check for resize of mapped image */
2232 rbd_dev->header.image_size = h.image_size;
2233 rbd_update_mapping_size(rbd_dev);
2235 /* rbd_dev->header.object_prefix shouldn't change */
2236 kfree(rbd_dev->header.snap_sizes);
2237 kfree(rbd_dev->header.snap_names);
2238 /* osd requests may still refer to snapc */
2239 ceph_put_snap_context(rbd_dev->header.snapc);
2242 *hver = h.obj_version;
2243 rbd_dev->header.obj_version = h.obj_version;
2244 rbd_dev->header.image_size = h.image_size;
2245 rbd_dev->header.snapc = h.snapc;
2246 rbd_dev->header.snap_names = h.snap_names;
2247 rbd_dev->header.snap_sizes = h.snap_sizes;
2248 /* Free the extra copy of the object prefix */
2249 WARN_ON(strcmp(rbd_dev->header.object_prefix, h.object_prefix));
2250 kfree(h.object_prefix);
2252 ret = rbd_dev_snaps_update(rbd_dev);
2254 ret = rbd_dev_snaps_register(rbd_dev);
2256 up_write(&rbd_dev->header_rwsem);
2261 static int rbd_dev_refresh(struct rbd_device *rbd_dev, u64 *hver)
2265 rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
2266 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
2267 if (rbd_dev->image_format == 1)
2268 ret = rbd_dev_v1_refresh(rbd_dev, hver);
2270 ret = rbd_dev_v2_refresh(rbd_dev, hver);
2271 mutex_unlock(&ctl_mutex);
2276 static int rbd_init_disk(struct rbd_device *rbd_dev)
2278 struct gendisk *disk;
2279 struct request_queue *q;
2282 /* create gendisk info */
2283 disk = alloc_disk(RBD_MINORS_PER_MAJOR);
2287 snprintf(disk->disk_name, sizeof(disk->disk_name), RBD_DRV_NAME "%d",
2289 disk->major = rbd_dev->major;
2290 disk->first_minor = 0;
2291 disk->fops = &rbd_bd_ops;
2292 disk->private_data = rbd_dev;
2294 q = blk_init_queue(rbd_request_fn, &rbd_dev->lock);
2298 /* We use the default size, but let's be explicit about it. */
2299 blk_queue_physical_block_size(q, SECTOR_SIZE);
2301 /* set io sizes to object size */
2302 segment_size = rbd_obj_bytes(&rbd_dev->header);
2303 blk_queue_max_hw_sectors(q, segment_size / SECTOR_SIZE);
2304 blk_queue_max_segment_size(q, segment_size);
2305 blk_queue_io_min(q, segment_size);
2306 blk_queue_io_opt(q, segment_size);
2308 blk_queue_merge_bvec(q, rbd_merge_bvec);
2311 q->queuedata = rbd_dev;
2313 rbd_dev->disk = disk;
2315 set_capacity(rbd_dev->disk, rbd_dev->mapping.size / SECTOR_SIZE);
2328 static struct rbd_device *dev_to_rbd_dev(struct device *dev)
2330 return container_of(dev, struct rbd_device, dev);
2333 static ssize_t rbd_size_show(struct device *dev,
2334 struct device_attribute *attr, char *buf)
2336 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
2339 down_read(&rbd_dev->header_rwsem);
2340 size = get_capacity(rbd_dev->disk);
2341 up_read(&rbd_dev->header_rwsem);
2343 return sprintf(buf, "%llu\n", (unsigned long long) size * SECTOR_SIZE);
2347 * Note this shows the features for whatever's mapped, which is not
2348 * necessarily the base image.
2350 static ssize_t rbd_features_show(struct device *dev,
2351 struct device_attribute *attr, char *buf)
2353 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
2355 return sprintf(buf, "0x%016llx\n",
2356 (unsigned long long) rbd_dev->mapping.features);
2359 static ssize_t rbd_major_show(struct device *dev,
2360 struct device_attribute *attr, char *buf)
2362 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
2364 return sprintf(buf, "%d\n", rbd_dev->major);
2367 static ssize_t rbd_client_id_show(struct device *dev,
2368 struct device_attribute *attr, char *buf)
2370 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
2372 return sprintf(buf, "client%lld\n",
2373 ceph_client_id(rbd_dev->rbd_client->client));
2376 static ssize_t rbd_pool_show(struct device *dev,
2377 struct device_attribute *attr, char *buf)
2379 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
2381 return sprintf(buf, "%s\n", rbd_dev->spec->pool_name);
2384 static ssize_t rbd_pool_id_show(struct device *dev,
2385 struct device_attribute *attr, char *buf)
2387 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
2389 return sprintf(buf, "%llu\n",
2390 (unsigned long long) rbd_dev->spec->pool_id);
2393 static ssize_t rbd_name_show(struct device *dev,
2394 struct device_attribute *attr, char *buf)
2396 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
2398 if (rbd_dev->spec->image_name)
2399 return sprintf(buf, "%s\n", rbd_dev->spec->image_name);
2401 return sprintf(buf, "(unknown)\n");
2404 static ssize_t rbd_image_id_show(struct device *dev,
2405 struct device_attribute *attr, char *buf)
2407 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
2409 return sprintf(buf, "%s\n", rbd_dev->spec->image_id);
2413 * Shows the name of the currently-mapped snapshot (or
2414 * RBD_SNAP_HEAD_NAME for the base image).
2416 static ssize_t rbd_snap_show(struct device *dev,
2417 struct device_attribute *attr,
2420 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
2422 return sprintf(buf, "%s\n", rbd_dev->spec->snap_name);
2426 * For an rbd v2 image, shows the pool id, image id, and snapshot id
2427 * for the parent image. If there is no parent, simply shows
2428 * "(no parent image)".
2430 static ssize_t rbd_parent_show(struct device *dev,
2431 struct device_attribute *attr,
2434 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
2435 struct rbd_spec *spec = rbd_dev->parent_spec;
2440 return sprintf(buf, "(no parent image)\n");
2442 count = sprintf(bufp, "pool_id %llu\npool_name %s\n",
2443 (unsigned long long) spec->pool_id, spec->pool_name);
2448 count = sprintf(bufp, "image_id %s\nimage_name %s\n", spec->image_id,
2449 spec->image_name ? spec->image_name : "(unknown)");
2454 count = sprintf(bufp, "snap_id %llu\nsnap_name %s\n",
2455 (unsigned long long) spec->snap_id, spec->snap_name);
2460 count = sprintf(bufp, "overlap %llu\n", rbd_dev->parent_overlap);
2465 return (ssize_t) (bufp - buf);
2468 static ssize_t rbd_image_refresh(struct device *dev,
2469 struct device_attribute *attr,
2473 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
2476 ret = rbd_dev_refresh(rbd_dev, NULL);
2478 return ret < 0 ? ret : size;
2481 static DEVICE_ATTR(size, S_IRUGO, rbd_size_show, NULL);
2482 static DEVICE_ATTR(features, S_IRUGO, rbd_features_show, NULL);
2483 static DEVICE_ATTR(major, S_IRUGO, rbd_major_show, NULL);
2484 static DEVICE_ATTR(client_id, S_IRUGO, rbd_client_id_show, NULL);
2485 static DEVICE_ATTR(pool, S_IRUGO, rbd_pool_show, NULL);
2486 static DEVICE_ATTR(pool_id, S_IRUGO, rbd_pool_id_show, NULL);
2487 static DEVICE_ATTR(name, S_IRUGO, rbd_name_show, NULL);
2488 static DEVICE_ATTR(image_id, S_IRUGO, rbd_image_id_show, NULL);
2489 static DEVICE_ATTR(refresh, S_IWUSR, NULL, rbd_image_refresh);
2490 static DEVICE_ATTR(current_snap, S_IRUGO, rbd_snap_show, NULL);
2491 static DEVICE_ATTR(parent, S_IRUGO, rbd_parent_show, NULL);
2493 static struct attribute *rbd_attrs[] = {
2494 &dev_attr_size.attr,
2495 &dev_attr_features.attr,
2496 &dev_attr_major.attr,
2497 &dev_attr_client_id.attr,
2498 &dev_attr_pool.attr,
2499 &dev_attr_pool_id.attr,
2500 &dev_attr_name.attr,
2501 &dev_attr_image_id.attr,
2502 &dev_attr_current_snap.attr,
2503 &dev_attr_parent.attr,
2504 &dev_attr_refresh.attr,
2508 static struct attribute_group rbd_attr_group = {
2512 static const struct attribute_group *rbd_attr_groups[] = {
2517 static void rbd_sysfs_dev_release(struct device *dev)
2521 static struct device_type rbd_device_type = {
2523 .groups = rbd_attr_groups,
2524 .release = rbd_sysfs_dev_release,
2532 static ssize_t rbd_snap_size_show(struct device *dev,
2533 struct device_attribute *attr,
2536 struct rbd_snap *snap = container_of(dev, struct rbd_snap, dev);
2538 return sprintf(buf, "%llu\n", (unsigned long long)snap->size);
2541 static ssize_t rbd_snap_id_show(struct device *dev,
2542 struct device_attribute *attr,
2545 struct rbd_snap *snap = container_of(dev, struct rbd_snap, dev);
2547 return sprintf(buf, "%llu\n", (unsigned long long)snap->id);
2550 static ssize_t rbd_snap_features_show(struct device *dev,
2551 struct device_attribute *attr,
2554 struct rbd_snap *snap = container_of(dev, struct rbd_snap, dev);
2556 return sprintf(buf, "0x%016llx\n",
2557 (unsigned long long) snap->features);
2560 static DEVICE_ATTR(snap_size, S_IRUGO, rbd_snap_size_show, NULL);
2561 static DEVICE_ATTR(snap_id, S_IRUGO, rbd_snap_id_show, NULL);
2562 static DEVICE_ATTR(snap_features, S_IRUGO, rbd_snap_features_show, NULL);
2564 static struct attribute *rbd_snap_attrs[] = {
2565 &dev_attr_snap_size.attr,
2566 &dev_attr_snap_id.attr,
2567 &dev_attr_snap_features.attr,
2571 static struct attribute_group rbd_snap_attr_group = {
2572 .attrs = rbd_snap_attrs,
2575 static void rbd_snap_dev_release(struct device *dev)
2577 struct rbd_snap *snap = container_of(dev, struct rbd_snap, dev);
2582 static const struct attribute_group *rbd_snap_attr_groups[] = {
2583 &rbd_snap_attr_group,
2587 static struct device_type rbd_snap_device_type = {
2588 .groups = rbd_snap_attr_groups,
2589 .release = rbd_snap_dev_release,
2592 static struct rbd_spec *rbd_spec_get(struct rbd_spec *spec)
2594 kref_get(&spec->kref);
2599 static void rbd_spec_free(struct kref *kref);
2600 static void rbd_spec_put(struct rbd_spec *spec)
2603 kref_put(&spec->kref, rbd_spec_free);
2606 static struct rbd_spec *rbd_spec_alloc(void)
2608 struct rbd_spec *spec;
2610 spec = kzalloc(sizeof (*spec), GFP_KERNEL);
2613 kref_init(&spec->kref);
2615 rbd_spec_put(rbd_spec_get(spec)); /* TEMPORARY */
2620 static void rbd_spec_free(struct kref *kref)
2622 struct rbd_spec *spec = container_of(kref, struct rbd_spec, kref);
2624 kfree(spec->pool_name);
2625 kfree(spec->image_id);
2626 kfree(spec->image_name);
2627 kfree(spec->snap_name);
2631 struct rbd_device *rbd_dev_create(struct rbd_client *rbdc,
2632 struct rbd_spec *spec)
2634 struct rbd_device *rbd_dev;
2636 rbd_dev = kzalloc(sizeof (*rbd_dev), GFP_KERNEL);
2640 spin_lock_init(&rbd_dev->lock);
2642 INIT_LIST_HEAD(&rbd_dev->node);
2643 INIT_LIST_HEAD(&rbd_dev->snaps);
2644 init_rwsem(&rbd_dev->header_rwsem);
2646 rbd_dev->spec = spec;
2647 rbd_dev->rbd_client = rbdc;
2649 /* Initialize the layout used for all rbd requests */
2651 rbd_dev->layout.fl_stripe_unit = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER);
2652 rbd_dev->layout.fl_stripe_count = cpu_to_le32(1);
2653 rbd_dev->layout.fl_object_size = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER);
2654 rbd_dev->layout.fl_pg_pool = cpu_to_le32((u32) spec->pool_id);
2659 static void rbd_dev_destroy(struct rbd_device *rbd_dev)
2661 rbd_spec_put(rbd_dev->parent_spec);
2662 kfree(rbd_dev->header_name);
2663 rbd_put_client(rbd_dev->rbd_client);
2664 rbd_spec_put(rbd_dev->spec);
2668 static bool rbd_snap_registered(struct rbd_snap *snap)
2670 bool ret = snap->dev.type == &rbd_snap_device_type;
2671 bool reg = device_is_registered(&snap->dev);
2673 rbd_assert(!ret ^ reg);
2678 static void rbd_remove_snap_dev(struct rbd_snap *snap)
2680 list_del(&snap->node);
2681 if (device_is_registered(&snap->dev))
2682 device_unregister(&snap->dev);
2685 static int rbd_register_snap_dev(struct rbd_snap *snap,
2686 struct device *parent)
2688 struct device *dev = &snap->dev;
2691 dev->type = &rbd_snap_device_type;
2692 dev->parent = parent;
2693 dev->release = rbd_snap_dev_release;
2694 dev_set_name(dev, "%s%s", RBD_SNAP_DEV_NAME_PREFIX, snap->name);
2695 dout("%s: registering device for snapshot %s\n", __func__, snap->name);
2697 ret = device_register(dev);
2702 static struct rbd_snap *__rbd_add_snap_dev(struct rbd_device *rbd_dev,
2703 const char *snap_name,
2704 u64 snap_id, u64 snap_size,
2707 struct rbd_snap *snap;
2710 snap = kzalloc(sizeof (*snap), GFP_KERNEL);
2712 return ERR_PTR(-ENOMEM);
2715 snap->name = kstrdup(snap_name, GFP_KERNEL);
2720 snap->size = snap_size;
2721 snap->features = snap_features;
2729 return ERR_PTR(ret);
2732 static char *rbd_dev_v1_snap_info(struct rbd_device *rbd_dev, u32 which,
2733 u64 *snap_size, u64 *snap_features)
2737 rbd_assert(which < rbd_dev->header.snapc->num_snaps);
2739 *snap_size = rbd_dev->header.snap_sizes[which];
2740 *snap_features = 0; /* No features for v1 */
2742 /* Skip over names until we find the one we are looking for */
2744 snap_name = rbd_dev->header.snap_names;
2746 snap_name += strlen(snap_name) + 1;
2752 * Get the size and object order for an image snapshot, or if
2753 * snap_id is CEPH_NOSNAP, gets this information for the base
2756 static int _rbd_dev_v2_snap_size(struct rbd_device *rbd_dev, u64 snap_id,
2757 u8 *order, u64 *snap_size)
2759 __le64 snapid = cpu_to_le64(snap_id);
2764 } __attribute__ ((packed)) size_buf = { 0 };
2766 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
2768 (char *) &snapid, sizeof (snapid),
2769 (char *) &size_buf, sizeof (size_buf), NULL);
2770 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
2774 *order = size_buf.order;
2775 *snap_size = le64_to_cpu(size_buf.size);
2777 dout(" snap_id 0x%016llx order = %u, snap_size = %llu\n",
2778 (unsigned long long) snap_id, (unsigned int) *order,
2779 (unsigned long long) *snap_size);
2784 static int rbd_dev_v2_image_size(struct rbd_device *rbd_dev)
2786 return _rbd_dev_v2_snap_size(rbd_dev, CEPH_NOSNAP,
2787 &rbd_dev->header.obj_order,
2788 &rbd_dev->header.image_size);
2791 static int rbd_dev_v2_object_prefix(struct rbd_device *rbd_dev)
2797 reply_buf = kzalloc(RBD_OBJ_PREFIX_LEN_MAX, GFP_KERNEL);
2801 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
2802 "rbd", "get_object_prefix",
2804 reply_buf, RBD_OBJ_PREFIX_LEN_MAX, NULL);
2805 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
2810 rbd_dev->header.object_prefix = ceph_extract_encoded_string(&p,
2811 p + RBD_OBJ_PREFIX_LEN_MAX,
2814 if (IS_ERR(rbd_dev->header.object_prefix)) {
2815 ret = PTR_ERR(rbd_dev->header.object_prefix);
2816 rbd_dev->header.object_prefix = NULL;
2818 dout(" object_prefix = %s\n", rbd_dev->header.object_prefix);
2827 static int _rbd_dev_v2_snap_features(struct rbd_device *rbd_dev, u64 snap_id,
2830 __le64 snapid = cpu_to_le64(snap_id);
2834 } features_buf = { 0 };
2838 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
2839 "rbd", "get_features",
2840 (char *) &snapid, sizeof (snapid),
2841 (char *) &features_buf, sizeof (features_buf),
2843 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
2847 incompat = le64_to_cpu(features_buf.incompat);
2848 if (incompat & ~RBD_FEATURES_ALL)
2851 *snap_features = le64_to_cpu(features_buf.features);
2853 dout(" snap_id 0x%016llx features = 0x%016llx incompat = 0x%016llx\n",
2854 (unsigned long long) snap_id,
2855 (unsigned long long) *snap_features,
2856 (unsigned long long) le64_to_cpu(features_buf.incompat));
2861 static int rbd_dev_v2_features(struct rbd_device *rbd_dev)
2863 return _rbd_dev_v2_snap_features(rbd_dev, CEPH_NOSNAP,
2864 &rbd_dev->header.features);
2867 static int rbd_dev_v2_parent_info(struct rbd_device *rbd_dev)
2869 struct rbd_spec *parent_spec;
2871 void *reply_buf = NULL;
2879 parent_spec = rbd_spec_alloc();
2883 size = sizeof (__le64) + /* pool_id */
2884 sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX + /* image_id */
2885 sizeof (__le64) + /* snap_id */
2886 sizeof (__le64); /* overlap */
2887 reply_buf = kmalloc(size, GFP_KERNEL);
2893 snapid = cpu_to_le64(CEPH_NOSNAP);
2894 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
2895 "rbd", "get_parent",
2896 (char *) &snapid, sizeof (snapid),
2897 (char *) reply_buf, size, NULL);
2898 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
2904 end = (char *) reply_buf + size;
2905 ceph_decode_64_safe(&p, end, parent_spec->pool_id, out_err);
2906 if (parent_spec->pool_id == CEPH_NOPOOL)
2907 goto out; /* No parent? No problem. */
2909 /* The ceph file layout needs to fit pool id in 32 bits */
2912 if (WARN_ON(parent_spec->pool_id > (u64) U32_MAX))
2915 image_id = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL);
2916 if (IS_ERR(image_id)) {
2917 ret = PTR_ERR(image_id);
2920 parent_spec->image_id = image_id;
2921 ceph_decode_64_safe(&p, end, parent_spec->snap_id, out_err);
2922 ceph_decode_64_safe(&p, end, overlap, out_err);
2924 rbd_dev->parent_overlap = overlap;
2925 rbd_dev->parent_spec = parent_spec;
2926 parent_spec = NULL; /* rbd_dev now owns this */
2931 rbd_spec_put(parent_spec);
2936 static char *rbd_dev_image_name(struct rbd_device *rbd_dev)
2938 size_t image_id_size;
2943 void *reply_buf = NULL;
2945 char *image_name = NULL;
2948 rbd_assert(!rbd_dev->spec->image_name);
2950 len = strlen(rbd_dev->spec->image_id);
2951 image_id_size = sizeof (__le32) + len;
2952 image_id = kmalloc(image_id_size, GFP_KERNEL);
2957 end = (char *) image_id + image_id_size;
2958 ceph_encode_string(&p, end, rbd_dev->spec->image_id, (u32) len);
2960 size = sizeof (__le32) + RBD_IMAGE_NAME_LEN_MAX;
2961 reply_buf = kmalloc(size, GFP_KERNEL);
2965 ret = rbd_obj_method_sync(rbd_dev, RBD_DIRECTORY,
2966 "rbd", "dir_get_name",
2967 image_id, image_id_size,
2968 (char *) reply_buf, size, NULL);
2972 end = (char *) reply_buf + size;
2973 image_name = ceph_extract_encoded_string(&p, end, &len, GFP_KERNEL);
2974 if (IS_ERR(image_name))
2977 dout("%s: name is %s len is %zd\n", __func__, image_name, len);
2986 * When a parent image gets probed, we only have the pool, image,
2987 * and snapshot ids but not the names of any of them. This call
2988 * is made later to fill in those names. It has to be done after
2989 * rbd_dev_snaps_update() has completed because some of the
2990 * information (in particular, snapshot name) is not available
2993 static int rbd_dev_probe_update_spec(struct rbd_device *rbd_dev)
2995 struct ceph_osd_client *osdc;
2997 void *reply_buf = NULL;
3000 if (rbd_dev->spec->pool_name)
3001 return 0; /* Already have the names */
3003 /* Look up the pool name */
3005 osdc = &rbd_dev->rbd_client->client->osdc;
3006 name = ceph_pg_pool_name_by_id(osdc->osdmap, rbd_dev->spec->pool_id);
3008 rbd_warn(rbd_dev, "there is no pool with id %llu",
3009 rbd_dev->spec->pool_id); /* Really a BUG() */
3013 rbd_dev->spec->pool_name = kstrdup(name, GFP_KERNEL);
3014 if (!rbd_dev->spec->pool_name)
3017 /* Fetch the image name; tolerate failure here */
3019 name = rbd_dev_image_name(rbd_dev);
3021 rbd_dev->spec->image_name = (char *) name;
3023 rbd_warn(rbd_dev, "unable to get image name");
3025 /* Look up the snapshot name. */
3027 name = rbd_snap_name(rbd_dev, rbd_dev->spec->snap_id);
3029 rbd_warn(rbd_dev, "no snapshot with id %llu",
3030 rbd_dev->spec->snap_id); /* Really a BUG() */
3034 rbd_dev->spec->snap_name = kstrdup(name, GFP_KERNEL);
3035 if(!rbd_dev->spec->snap_name)
3041 kfree(rbd_dev->spec->pool_name);
3042 rbd_dev->spec->pool_name = NULL;
3047 static int rbd_dev_v2_snap_context(struct rbd_device *rbd_dev, u64 *ver)
3056 struct ceph_snap_context *snapc;
3060 * We'll need room for the seq value (maximum snapshot id),
3061 * snapshot count, and array of that many snapshot ids.
3062 * For now we have a fixed upper limit on the number we're
3063 * prepared to receive.
3065 size = sizeof (__le64) + sizeof (__le32) +
3066 RBD_MAX_SNAP_COUNT * sizeof (__le64);
3067 reply_buf = kzalloc(size, GFP_KERNEL);
3071 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
3072 "rbd", "get_snapcontext",
3074 reply_buf, size, ver);
3075 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
3081 end = (char *) reply_buf + size;
3082 ceph_decode_64_safe(&p, end, seq, out);
3083 ceph_decode_32_safe(&p, end, snap_count, out);
3086 * Make sure the reported number of snapshot ids wouldn't go
3087 * beyond the end of our buffer. But before checking that,
3088 * make sure the computed size of the snapshot context we
3089 * allocate is representable in a size_t.
3091 if (snap_count > (SIZE_MAX - sizeof (struct ceph_snap_context))
3096 if (!ceph_has_room(&p, end, snap_count * sizeof (__le64)))
3099 size = sizeof (struct ceph_snap_context) +
3100 snap_count * sizeof (snapc->snaps[0]);
3101 snapc = kmalloc(size, GFP_KERNEL);
3107 atomic_set(&snapc->nref, 1);
3109 snapc->num_snaps = snap_count;
3110 for (i = 0; i < snap_count; i++)
3111 snapc->snaps[i] = ceph_decode_64(&p);
3113 rbd_dev->header.snapc = snapc;
3115 dout(" snap context seq = %llu, snap_count = %u\n",
3116 (unsigned long long) seq, (unsigned int) snap_count);
3124 static char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev, u32 which)
3134 size = sizeof (__le32) + RBD_MAX_SNAP_NAME_LEN;
3135 reply_buf = kmalloc(size, GFP_KERNEL);
3137 return ERR_PTR(-ENOMEM);
3139 snap_id = cpu_to_le64(rbd_dev->header.snapc->snaps[which]);
3140 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
3141 "rbd", "get_snapshot_name",
3142 (char *) &snap_id, sizeof (snap_id),
3143 reply_buf, size, NULL);
3144 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
3149 end = (char *) reply_buf + size;
3150 snap_name = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL);
3151 if (IS_ERR(snap_name)) {
3152 ret = PTR_ERR(snap_name);
3155 dout(" snap_id 0x%016llx snap_name = %s\n",
3156 (unsigned long long) le64_to_cpu(snap_id), snap_name);
3164 return ERR_PTR(ret);
3167 static char *rbd_dev_v2_snap_info(struct rbd_device *rbd_dev, u32 which,
3168 u64 *snap_size, u64 *snap_features)
3174 snap_id = rbd_dev->header.snapc->snaps[which];
3175 ret = _rbd_dev_v2_snap_size(rbd_dev, snap_id, &order, snap_size);
3177 return ERR_PTR(ret);
3178 ret = _rbd_dev_v2_snap_features(rbd_dev, snap_id, snap_features);
3180 return ERR_PTR(ret);
3182 return rbd_dev_v2_snap_name(rbd_dev, which);
3185 static char *rbd_dev_snap_info(struct rbd_device *rbd_dev, u32 which,
3186 u64 *snap_size, u64 *snap_features)
3188 if (rbd_dev->image_format == 1)
3189 return rbd_dev_v1_snap_info(rbd_dev, which,
3190 snap_size, snap_features);
3191 if (rbd_dev->image_format == 2)
3192 return rbd_dev_v2_snap_info(rbd_dev, which,
3193 snap_size, snap_features);
3194 return ERR_PTR(-EINVAL);
3197 static int rbd_dev_v2_refresh(struct rbd_device *rbd_dev, u64 *hver)
3202 down_write(&rbd_dev->header_rwsem);
3204 /* Grab old order first, to see if it changes */
3206 obj_order = rbd_dev->header.obj_order,
3207 ret = rbd_dev_v2_image_size(rbd_dev);
3210 if (rbd_dev->header.obj_order != obj_order) {
3214 rbd_update_mapping_size(rbd_dev);
3216 ret = rbd_dev_v2_snap_context(rbd_dev, hver);
3217 dout("rbd_dev_v2_snap_context returned %d\n", ret);
3220 ret = rbd_dev_snaps_update(rbd_dev);
3221 dout("rbd_dev_snaps_update returned %d\n", ret);
3224 ret = rbd_dev_snaps_register(rbd_dev);
3225 dout("rbd_dev_snaps_register returned %d\n", ret);
3227 up_write(&rbd_dev->header_rwsem);
3233 * Scan the rbd device's current snapshot list and compare it to the
3234 * newly-received snapshot context. Remove any existing snapshots
3235 * not present in the new snapshot context. Add a new snapshot for
3236 * any snaphots in the snapshot context not in the current list.
3237 * And verify there are no changes to snapshots we already know
3240 * Assumes the snapshots in the snapshot context are sorted by
3241 * snapshot id, highest id first. (Snapshots in the rbd_dev's list
3242 * are also maintained in that order.)
3244 static int rbd_dev_snaps_update(struct rbd_device *rbd_dev)
3246 struct ceph_snap_context *snapc = rbd_dev->header.snapc;
3247 const u32 snap_count = snapc->num_snaps;
3248 struct list_head *head = &rbd_dev->snaps;
3249 struct list_head *links = head->next;
3252 dout("%s: snap count is %u\n", __func__, (unsigned int) snap_count);
3253 while (index < snap_count || links != head) {
3255 struct rbd_snap *snap;
3258 u64 snap_features = 0;
3260 snap_id = index < snap_count ? snapc->snaps[index]
3262 snap = links != head ? list_entry(links, struct rbd_snap, node)
3264 rbd_assert(!snap || snap->id != CEPH_NOSNAP);
3266 if (snap_id == CEPH_NOSNAP || (snap && snap->id > snap_id)) {
3267 struct list_head *next = links->next;
3270 * A previously-existing snapshot is not in
3271 * the new snap context.
3273 * If the now missing snapshot is the one the
3274 * image is mapped to, clear its exists flag
3275 * so we can avoid sending any more requests
3278 if (rbd_dev->spec->snap_id == snap->id)
3279 clear_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags);
3280 rbd_remove_snap_dev(snap);
3281 dout("%ssnap id %llu has been removed\n",
3282 rbd_dev->spec->snap_id == snap->id ?
3284 (unsigned long long) snap->id);
3286 /* Done with this list entry; advance */
3292 snap_name = rbd_dev_snap_info(rbd_dev, index,
3293 &snap_size, &snap_features);
3294 if (IS_ERR(snap_name))
3295 return PTR_ERR(snap_name);
3297 dout("entry %u: snap_id = %llu\n", (unsigned int) snap_count,
3298 (unsigned long long) snap_id);
3299 if (!snap || (snap_id != CEPH_NOSNAP && snap->id < snap_id)) {
3300 struct rbd_snap *new_snap;
3302 /* We haven't seen this snapshot before */
3304 new_snap = __rbd_add_snap_dev(rbd_dev, snap_name,
3305 snap_id, snap_size, snap_features);
3306 if (IS_ERR(new_snap)) {
3307 int err = PTR_ERR(new_snap);
3309 dout(" failed to add dev, error %d\n", err);
3314 /* New goes before existing, or at end of list */
3316 dout(" added dev%s\n", snap ? "" : " at end\n");
3318 list_add_tail(&new_snap->node, &snap->node);
3320 list_add_tail(&new_snap->node, head);
3322 /* Already have this one */
3324 dout(" already present\n");
3326 rbd_assert(snap->size == snap_size);
3327 rbd_assert(!strcmp(snap->name, snap_name));
3328 rbd_assert(snap->features == snap_features);
3330 /* Done with this list entry; advance */
3332 links = links->next;
3335 /* Advance to the next entry in the snapshot context */
3339 dout("%s: done\n", __func__);
3345 * Scan the list of snapshots and register the devices for any that
3346 * have not already been registered.
3348 static int rbd_dev_snaps_register(struct rbd_device *rbd_dev)
3350 struct rbd_snap *snap;
3353 dout("%s called\n", __func__);
3354 if (WARN_ON(!device_is_registered(&rbd_dev->dev)))
3357 list_for_each_entry(snap, &rbd_dev->snaps, node) {
3358 if (!rbd_snap_registered(snap)) {
3359 ret = rbd_register_snap_dev(snap, &rbd_dev->dev);
3364 dout("%s: returning %d\n", __func__, ret);
3369 static int rbd_bus_add_dev(struct rbd_device *rbd_dev)
3374 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
3376 dev = &rbd_dev->dev;
3377 dev->bus = &rbd_bus_type;
3378 dev->type = &rbd_device_type;
3379 dev->parent = &rbd_root_dev;
3380 dev->release = rbd_dev_release;
3381 dev_set_name(dev, "%d", rbd_dev->dev_id);
3382 ret = device_register(dev);
3384 mutex_unlock(&ctl_mutex);
3389 static void rbd_bus_del_dev(struct rbd_device *rbd_dev)
3391 device_unregister(&rbd_dev->dev);
3394 static atomic64_t rbd_dev_id_max = ATOMIC64_INIT(0);
3397 * Get a unique rbd identifier for the given new rbd_dev, and add
3398 * the rbd_dev to the global list. The minimum rbd id is 1.
3400 static void rbd_dev_id_get(struct rbd_device *rbd_dev)
3402 rbd_dev->dev_id = atomic64_inc_return(&rbd_dev_id_max);
3404 spin_lock(&rbd_dev_list_lock);
3405 list_add_tail(&rbd_dev->node, &rbd_dev_list);
3406 spin_unlock(&rbd_dev_list_lock);
3407 dout("rbd_dev %p given dev id %llu\n", rbd_dev,
3408 (unsigned long long) rbd_dev->dev_id);
3412 * Remove an rbd_dev from the global list, and record that its
3413 * identifier is no longer in use.
3415 static void rbd_dev_id_put(struct rbd_device *rbd_dev)
3417 struct list_head *tmp;
3418 int rbd_id = rbd_dev->dev_id;
3421 rbd_assert(rbd_id > 0);
3423 dout("rbd_dev %p released dev id %llu\n", rbd_dev,
3424 (unsigned long long) rbd_dev->dev_id);
3425 spin_lock(&rbd_dev_list_lock);
3426 list_del_init(&rbd_dev->node);
3429 * If the id being "put" is not the current maximum, there
3430 * is nothing special we need to do.
3432 if (rbd_id != atomic64_read(&rbd_dev_id_max)) {
3433 spin_unlock(&rbd_dev_list_lock);
3438 * We need to update the current maximum id. Search the
3439 * list to find out what it is. We're more likely to find
3440 * the maximum at the end, so search the list backward.
3443 list_for_each_prev(tmp, &rbd_dev_list) {
3444 struct rbd_device *rbd_dev;
3446 rbd_dev = list_entry(tmp, struct rbd_device, node);
3447 if (rbd_dev->dev_id > max_id)
3448 max_id = rbd_dev->dev_id;
3450 spin_unlock(&rbd_dev_list_lock);
3453 * The max id could have been updated by rbd_dev_id_get(), in
3454 * which case it now accurately reflects the new maximum.
3455 * Be careful not to overwrite the maximum value in that
3458 atomic64_cmpxchg(&rbd_dev_id_max, rbd_id, max_id);
3459 dout(" max dev id has been reset\n");
3463 * Skips over white space at *buf, and updates *buf to point to the
3464 * first found non-space character (if any). Returns the length of
3465 * the token (string of non-white space characters) found. Note
3466 * that *buf must be terminated with '\0'.
3468 static inline size_t next_token(const char **buf)
3471 * These are the characters that produce nonzero for
3472 * isspace() in the "C" and "POSIX" locales.
3474 const char *spaces = " \f\n\r\t\v";
3476 *buf += strspn(*buf, spaces); /* Find start of token */
3478 return strcspn(*buf, spaces); /* Return token length */
3482 * Finds the next token in *buf, and if the provided token buffer is
3483 * big enough, copies the found token into it. The result, if
3484 * copied, is guaranteed to be terminated with '\0'. Note that *buf
3485 * must be terminated with '\0' on entry.
3487 * Returns the length of the token found (not including the '\0').
3488 * Return value will be 0 if no token is found, and it will be >=
3489 * token_size if the token would not fit.
3491 * The *buf pointer will be updated to point beyond the end of the
3492 * found token. Note that this occurs even if the token buffer is
3493 * too small to hold it.
3495 static inline size_t copy_token(const char **buf,
3501 len = next_token(buf);
3502 if (len < token_size) {
3503 memcpy(token, *buf, len);
3504 *(token + len) = '\0';
3512 * Finds the next token in *buf, dynamically allocates a buffer big
3513 * enough to hold a copy of it, and copies the token into the new
3514 * buffer. The copy is guaranteed to be terminated with '\0'. Note
3515 * that a duplicate buffer is created even for a zero-length token.
3517 * Returns a pointer to the newly-allocated duplicate, or a null
3518 * pointer if memory for the duplicate was not available. If
3519 * the lenp argument is a non-null pointer, the length of the token
3520 * (not including the '\0') is returned in *lenp.
3522 * If successful, the *buf pointer will be updated to point beyond
3523 * the end of the found token.
3525 * Note: uses GFP_KERNEL for allocation.
3527 static inline char *dup_token(const char **buf, size_t *lenp)
3532 len = next_token(buf);
3533 dup = kmemdup(*buf, len + 1, GFP_KERNEL);
3536 *(dup + len) = '\0';
3546 * Parse the options provided for an "rbd add" (i.e., rbd image
3547 * mapping) request. These arrive via a write to /sys/bus/rbd/add,
3548 * and the data written is passed here via a NUL-terminated buffer.
3549 * Returns 0 if successful or an error code otherwise.
3551 * The information extracted from these options is recorded in
3552 * the other parameters which return dynamically-allocated
3555 * The address of a pointer that will refer to a ceph options
3556 * structure. Caller must release the returned pointer using
3557 * ceph_destroy_options() when it is no longer needed.
3559 * Address of an rbd options pointer. Fully initialized by
3560 * this function; caller must release with kfree().
3562 * Address of an rbd image specification pointer. Fully
3563 * initialized by this function based on parsed options.
3564 * Caller must release with rbd_spec_put().
3566 * The options passed take this form:
3567 * <mon_addrs> <options> <pool_name> <image_name> [<snap_id>]
3570 * A comma-separated list of one or more monitor addresses.
3571 * A monitor address is an ip address, optionally followed
3572 * by a port number (separated by a colon).
3573 * I.e.: ip1[:port1][,ip2[:port2]...]
3575 * A comma-separated list of ceph and/or rbd options.
3577 * The name of the rados pool containing the rbd image.
3579 * The name of the image in that pool to map.
3581 * An optional snapshot id. If provided, the mapping will
3582 * present data from the image at the time that snapshot was
3583 * created. The image head is used if no snapshot id is
3584 * provided. Snapshot mappings are always read-only.
3586 static int rbd_add_parse_args(const char *buf,
3587 struct ceph_options **ceph_opts,
3588 struct rbd_options **opts,
3589 struct rbd_spec **rbd_spec)
3593 const char *mon_addrs;
3594 size_t mon_addrs_size;
3595 struct rbd_spec *spec = NULL;
3596 struct rbd_options *rbd_opts = NULL;
3597 struct ceph_options *copts;
3600 /* The first four tokens are required */
3602 len = next_token(&buf);
3604 rbd_warn(NULL, "no monitor address(es) provided");
3608 mon_addrs_size = len + 1;
3612 options = dup_token(&buf, NULL);
3616 rbd_warn(NULL, "no options provided");
3620 spec = rbd_spec_alloc();
3624 spec->pool_name = dup_token(&buf, NULL);
3625 if (!spec->pool_name)
3627 if (!*spec->pool_name) {
3628 rbd_warn(NULL, "no pool name provided");
3632 spec->image_name = dup_token(&buf, NULL);
3633 if (!spec->image_name)
3635 if (!*spec->image_name) {
3636 rbd_warn(NULL, "no image name provided");
3641 * Snapshot name is optional; default is to use "-"
3642 * (indicating the head/no snapshot).
3644 len = next_token(&buf);
3646 buf = RBD_SNAP_HEAD_NAME; /* No snapshot supplied */
3647 len = sizeof (RBD_SNAP_HEAD_NAME) - 1;
3648 } else if (len > RBD_MAX_SNAP_NAME_LEN) {
3649 ret = -ENAMETOOLONG;
3652 spec->snap_name = kmemdup(buf, len + 1, GFP_KERNEL);
3653 if (!spec->snap_name)
3655 *(spec->snap_name + len) = '\0';
3657 /* Initialize all rbd options to the defaults */
3659 rbd_opts = kzalloc(sizeof (*rbd_opts), GFP_KERNEL);
3663 rbd_opts->read_only = RBD_READ_ONLY_DEFAULT;
3665 copts = ceph_parse_options(options, mon_addrs,
3666 mon_addrs + mon_addrs_size - 1,
3667 parse_rbd_opts_token, rbd_opts);
3668 if (IS_ERR(copts)) {
3669 ret = PTR_ERR(copts);
3690 * An rbd format 2 image has a unique identifier, distinct from the
3691 * name given to it by the user. Internally, that identifier is
3692 * what's used to specify the names of objects related to the image.
3694 * A special "rbd id" object is used to map an rbd image name to its
3695 * id. If that object doesn't exist, then there is no v2 rbd image
3696 * with the supplied name.
3698 * This function will record the given rbd_dev's image_id field if
3699 * it can be determined, and in that case will return 0. If any
3700 * errors occur a negative errno will be returned and the rbd_dev's
3701 * image_id field will be unchanged (and should be NULL).
3703 static int rbd_dev_image_id(struct rbd_device *rbd_dev)
3712 * When probing a parent image, the image id is already
3713 * known (and the image name likely is not). There's no
3714 * need to fetch the image id again in this case.
3716 if (rbd_dev->spec->image_id)
3720 * First, see if the format 2 image id file exists, and if
3721 * so, get the image's persistent id from it.
3723 size = sizeof (RBD_ID_PREFIX) + strlen(rbd_dev->spec->image_name);
3724 object_name = kmalloc(size, GFP_NOIO);
3727 sprintf(object_name, "%s%s", RBD_ID_PREFIX, rbd_dev->spec->image_name);
3728 dout("rbd id object name is %s\n", object_name);
3730 /* Response will be an encoded string, which includes a length */
3732 size = sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX;
3733 response = kzalloc(size, GFP_NOIO);
3739 ret = rbd_obj_method_sync(rbd_dev, object_name,
3742 response, RBD_IMAGE_ID_LEN_MAX, NULL);
3743 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
3748 rbd_dev->spec->image_id = ceph_extract_encoded_string(&p,
3749 p + RBD_IMAGE_ID_LEN_MAX,
3751 if (IS_ERR(rbd_dev->spec->image_id)) {
3752 ret = PTR_ERR(rbd_dev->spec->image_id);
3753 rbd_dev->spec->image_id = NULL;
3755 dout("image_id is %s\n", rbd_dev->spec->image_id);
3764 static int rbd_dev_v1_probe(struct rbd_device *rbd_dev)
3769 /* Version 1 images have no id; empty string is used */
3771 rbd_dev->spec->image_id = kstrdup("", GFP_KERNEL);
3772 if (!rbd_dev->spec->image_id)
3775 /* Record the header object name for this rbd image. */
3777 size = strlen(rbd_dev->spec->image_name) + sizeof (RBD_SUFFIX);
3778 rbd_dev->header_name = kmalloc(size, GFP_KERNEL);
3779 if (!rbd_dev->header_name) {
3783 sprintf(rbd_dev->header_name, "%s%s",
3784 rbd_dev->spec->image_name, RBD_SUFFIX);
3786 /* Populate rbd image metadata */
3788 ret = rbd_read_header(rbd_dev, &rbd_dev->header);
3792 /* Version 1 images have no parent (no layering) */
3794 rbd_dev->parent_spec = NULL;
3795 rbd_dev->parent_overlap = 0;
3797 rbd_dev->image_format = 1;
3799 dout("discovered version 1 image, header name is %s\n",
3800 rbd_dev->header_name);
3805 kfree(rbd_dev->header_name);
3806 rbd_dev->header_name = NULL;
3807 kfree(rbd_dev->spec->image_id);
3808 rbd_dev->spec->image_id = NULL;
3813 static int rbd_dev_v2_probe(struct rbd_device *rbd_dev)
3820 * Image id was filled in by the caller. Record the header
3821 * object name for this rbd image.
3823 size = sizeof (RBD_HEADER_PREFIX) + strlen(rbd_dev->spec->image_id);
3824 rbd_dev->header_name = kmalloc(size, GFP_KERNEL);
3825 if (!rbd_dev->header_name)
3827 sprintf(rbd_dev->header_name, "%s%s",
3828 RBD_HEADER_PREFIX, rbd_dev->spec->image_id);
3830 /* Get the size and object order for the image */
3832 ret = rbd_dev_v2_image_size(rbd_dev);
3836 /* Get the object prefix (a.k.a. block_name) for the image */
3838 ret = rbd_dev_v2_object_prefix(rbd_dev);
3842 /* Get the and check features for the image */
3844 ret = rbd_dev_v2_features(rbd_dev);
3848 /* If the image supports layering, get the parent info */
3850 if (rbd_dev->header.features & RBD_FEATURE_LAYERING) {
3851 ret = rbd_dev_v2_parent_info(rbd_dev);
3856 /* crypto and compression type aren't (yet) supported for v2 images */
3858 rbd_dev->header.crypt_type = 0;
3859 rbd_dev->header.comp_type = 0;
3861 /* Get the snapshot context, plus the header version */
3863 ret = rbd_dev_v2_snap_context(rbd_dev, &ver);
3866 rbd_dev->header.obj_version = ver;
3868 rbd_dev->image_format = 2;
3870 dout("discovered version 2 image, header name is %s\n",
3871 rbd_dev->header_name);
3875 rbd_dev->parent_overlap = 0;
3876 rbd_spec_put(rbd_dev->parent_spec);
3877 rbd_dev->parent_spec = NULL;
3878 kfree(rbd_dev->header_name);
3879 rbd_dev->header_name = NULL;
3880 kfree(rbd_dev->header.object_prefix);
3881 rbd_dev->header.object_prefix = NULL;
3886 static int rbd_dev_probe_finish(struct rbd_device *rbd_dev)
3890 /* no need to lock here, as rbd_dev is not registered yet */
3891 ret = rbd_dev_snaps_update(rbd_dev);
3895 ret = rbd_dev_probe_update_spec(rbd_dev);
3899 ret = rbd_dev_set_mapping(rbd_dev);
3903 /* generate unique id: find highest unique id, add one */
3904 rbd_dev_id_get(rbd_dev);
3906 /* Fill in the device name, now that we have its id. */
3907 BUILD_BUG_ON(DEV_NAME_LEN
3908 < sizeof (RBD_DRV_NAME) + MAX_INT_FORMAT_WIDTH);
3909 sprintf(rbd_dev->name, "%s%d", RBD_DRV_NAME, rbd_dev->dev_id);
3911 /* Get our block major device number. */
3913 ret = register_blkdev(0, rbd_dev->name);
3916 rbd_dev->major = ret;
3918 /* Set up the blkdev mapping. */
3920 ret = rbd_init_disk(rbd_dev);
3922 goto err_out_blkdev;
3924 ret = rbd_bus_add_dev(rbd_dev);
3929 * At this point cleanup in the event of an error is the job
3930 * of the sysfs code (initiated by rbd_bus_del_dev()).
3932 down_write(&rbd_dev->header_rwsem);
3933 ret = rbd_dev_snaps_register(rbd_dev);
3934 up_write(&rbd_dev->header_rwsem);
3938 ret = rbd_dev_header_watch_sync(rbd_dev, 1);
3942 /* Everything's ready. Announce the disk to the world. */
3944 add_disk(rbd_dev->disk);
3946 pr_info("%s: added with size 0x%llx\n", rbd_dev->disk->disk_name,
3947 (unsigned long long) rbd_dev->mapping.size);
3951 /* this will also clean up rest of rbd_dev stuff */
3953 rbd_bus_del_dev(rbd_dev);
3957 rbd_free_disk(rbd_dev);
3959 unregister_blkdev(rbd_dev->major, rbd_dev->name);
3961 rbd_dev_id_put(rbd_dev);
3963 rbd_remove_all_snaps(rbd_dev);
3969 * Probe for the existence of the header object for the given rbd
3970 * device. For format 2 images this includes determining the image
3973 static int rbd_dev_probe(struct rbd_device *rbd_dev)
3978 * Get the id from the image id object. If it's not a
3979 * format 2 image, we'll get ENOENT back, and we'll assume
3980 * it's a format 1 image.
3982 ret = rbd_dev_image_id(rbd_dev);
3984 ret = rbd_dev_v1_probe(rbd_dev);
3986 ret = rbd_dev_v2_probe(rbd_dev);
3988 dout("probe failed, returning %d\n", ret);
3993 ret = rbd_dev_probe_finish(rbd_dev);
3995 rbd_header_free(&rbd_dev->header);
4000 static ssize_t rbd_add(struct bus_type *bus,
4004 struct rbd_device *rbd_dev = NULL;
4005 struct ceph_options *ceph_opts = NULL;
4006 struct rbd_options *rbd_opts = NULL;
4007 struct rbd_spec *spec = NULL;
4008 struct rbd_client *rbdc;
4009 struct ceph_osd_client *osdc;
4012 if (!try_module_get(THIS_MODULE))
4015 /* parse add command */
4016 rc = rbd_add_parse_args(buf, &ceph_opts, &rbd_opts, &spec);
4018 goto err_out_module;
4020 rbdc = rbd_get_client(ceph_opts);
4025 ceph_opts = NULL; /* rbd_dev client now owns this */
4028 osdc = &rbdc->client->osdc;
4029 rc = ceph_pg_poolid_by_name(osdc->osdmap, spec->pool_name);
4031 goto err_out_client;
4032 spec->pool_id = (u64) rc;
4034 /* The ceph file layout needs to fit pool id in 32 bits */
4036 if (WARN_ON(spec->pool_id > (u64) U32_MAX)) {
4038 goto err_out_client;
4041 rbd_dev = rbd_dev_create(rbdc, spec);
4043 goto err_out_client;
4044 rbdc = NULL; /* rbd_dev now owns this */
4045 spec = NULL; /* rbd_dev now owns this */
4047 rbd_dev->mapping.read_only = rbd_opts->read_only;
4049 rbd_opts = NULL; /* done with this */
4051 rc = rbd_dev_probe(rbd_dev);
4053 goto err_out_rbd_dev;
4057 rbd_dev_destroy(rbd_dev);
4059 rbd_put_client(rbdc);
4062 ceph_destroy_options(ceph_opts);
4066 module_put(THIS_MODULE);
4068 dout("Error adding device %s\n", buf);
4070 return (ssize_t) rc;
4073 static struct rbd_device *__rbd_get_dev(unsigned long dev_id)
4075 struct list_head *tmp;
4076 struct rbd_device *rbd_dev;
4078 spin_lock(&rbd_dev_list_lock);
4079 list_for_each(tmp, &rbd_dev_list) {
4080 rbd_dev = list_entry(tmp, struct rbd_device, node);
4081 if (rbd_dev->dev_id == dev_id) {
4082 spin_unlock(&rbd_dev_list_lock);
4086 spin_unlock(&rbd_dev_list_lock);
4090 static void rbd_dev_release(struct device *dev)
4092 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
4094 if (rbd_dev->watch_event)
4095 rbd_dev_header_watch_sync(rbd_dev, 0);
4097 /* clean up and free blkdev */
4098 rbd_free_disk(rbd_dev);
4099 unregister_blkdev(rbd_dev->major, rbd_dev->name);
4101 /* release allocated disk header fields */
4102 rbd_header_free(&rbd_dev->header);
4104 /* done with the id, and with the rbd_dev */
4105 rbd_dev_id_put(rbd_dev);
4106 rbd_assert(rbd_dev->rbd_client != NULL);
4107 rbd_dev_destroy(rbd_dev);
4109 /* release module ref */
4110 module_put(THIS_MODULE);
4113 static ssize_t rbd_remove(struct bus_type *bus,
4117 struct rbd_device *rbd_dev = NULL;
4122 rc = strict_strtoul(buf, 10, &ul);
4126 /* convert to int; abort if we lost anything in the conversion */
4127 target_id = (int) ul;
4128 if (target_id != ul)
4131 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
4133 rbd_dev = __rbd_get_dev(target_id);
4139 spin_lock_irq(&rbd_dev->lock);
4140 if (rbd_dev->open_count)
4143 set_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags);
4144 spin_unlock_irq(&rbd_dev->lock);
4148 rbd_remove_all_snaps(rbd_dev);
4149 rbd_bus_del_dev(rbd_dev);
4152 mutex_unlock(&ctl_mutex);
4158 * create control files in sysfs
4161 static int rbd_sysfs_init(void)
4165 ret = device_register(&rbd_root_dev);
4169 ret = bus_register(&rbd_bus_type);
4171 device_unregister(&rbd_root_dev);
4176 static void rbd_sysfs_cleanup(void)
4178 bus_unregister(&rbd_bus_type);
4179 device_unregister(&rbd_root_dev);
4182 int __init rbd_init(void)
4186 if (!libceph_compatible(NULL)) {
4187 rbd_warn(NULL, "libceph incompatibility (quitting)");
4191 rc = rbd_sysfs_init();
4194 pr_info("loaded " RBD_DRV_NAME_LONG "\n");
4198 void __exit rbd_exit(void)
4200 rbd_sysfs_cleanup();
4203 module_init(rbd_init);
4204 module_exit(rbd_exit);
4206 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
4207 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
4208 MODULE_DESCRIPTION("rados block device");
4210 /* following authorship retained from original osdblk.c */
4211 MODULE_AUTHOR("Jeff Garzik <jeff@garzik.org>");
4213 MODULE_LICENSE("GPL");