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("%s:\n", __func__);
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);
470 dout("%s: rbdc %p\n", __func__, rbdc);
475 ceph_destroy_client(rbdc->client);
477 mutex_unlock(&ctl_mutex);
481 ceph_destroy_options(ceph_opts);
482 dout("%s: error %d\n", __func__, ret);
488 * Find a ceph client with specific addr and configuration. If
489 * found, bump its reference count.
491 static struct rbd_client *rbd_client_find(struct ceph_options *ceph_opts)
493 struct rbd_client *client_node;
496 if (ceph_opts->flags & CEPH_OPT_NOSHARE)
499 spin_lock(&rbd_client_list_lock);
500 list_for_each_entry(client_node, &rbd_client_list, node) {
501 if (!ceph_compare_options(ceph_opts, client_node->client)) {
502 kref_get(&client_node->kref);
507 spin_unlock(&rbd_client_list_lock);
509 return found ? client_node : NULL;
519 /* string args above */
522 /* Boolean args above */
526 static match_table_t rbd_opts_tokens = {
528 /* string args above */
529 {Opt_read_only, "read_only"},
530 {Opt_read_only, "ro"}, /* Alternate spelling */
531 {Opt_read_write, "read_write"},
532 {Opt_read_write, "rw"}, /* Alternate spelling */
533 /* Boolean args above */
541 #define RBD_READ_ONLY_DEFAULT false
543 static int parse_rbd_opts_token(char *c, void *private)
545 struct rbd_options *rbd_opts = private;
546 substring_t argstr[MAX_OPT_ARGS];
547 int token, intval, ret;
549 token = match_token(c, rbd_opts_tokens, argstr);
553 if (token < Opt_last_int) {
554 ret = match_int(&argstr[0], &intval);
556 pr_err("bad mount option arg (not int) "
560 dout("got int token %d val %d\n", token, intval);
561 } else if (token > Opt_last_int && token < Opt_last_string) {
562 dout("got string token %d val %s\n", token,
564 } else if (token > Opt_last_string && token < Opt_last_bool) {
565 dout("got Boolean token %d\n", token);
567 dout("got token %d\n", token);
572 rbd_opts->read_only = true;
575 rbd_opts->read_only = false;
585 * Get a ceph client with specific addr and configuration, if one does
586 * not exist create it.
588 static struct rbd_client *rbd_get_client(struct ceph_options *ceph_opts)
590 struct rbd_client *rbdc;
592 rbdc = rbd_client_find(ceph_opts);
593 if (rbdc) /* using an existing client */
594 ceph_destroy_options(ceph_opts);
596 rbdc = rbd_client_create(ceph_opts);
602 * Destroy ceph client
604 * Caller must hold rbd_client_list_lock.
606 static void rbd_client_release(struct kref *kref)
608 struct rbd_client *rbdc = container_of(kref, struct rbd_client, kref);
610 dout("%s: rbdc %p\n", __func__, rbdc);
611 spin_lock(&rbd_client_list_lock);
612 list_del(&rbdc->node);
613 spin_unlock(&rbd_client_list_lock);
615 ceph_destroy_client(rbdc->client);
620 * Drop reference to ceph client node. If it's not referenced anymore, release
623 static void rbd_put_client(struct rbd_client *rbdc)
626 kref_put(&rbdc->kref, rbd_client_release);
629 static bool rbd_image_format_valid(u32 image_format)
631 return image_format == 1 || image_format == 2;
634 static bool rbd_dev_ondisk_valid(struct rbd_image_header_ondisk *ondisk)
639 /* The header has to start with the magic rbd header text */
640 if (memcmp(&ondisk->text, RBD_HEADER_TEXT, sizeof (RBD_HEADER_TEXT)))
643 /* The bio layer requires at least sector-sized I/O */
645 if (ondisk->options.order < SECTOR_SHIFT)
648 /* If we use u64 in a few spots we may be able to loosen this */
650 if (ondisk->options.order > 8 * sizeof (int) - 1)
654 * The size of a snapshot header has to fit in a size_t, and
655 * that limits the number of snapshots.
657 snap_count = le32_to_cpu(ondisk->snap_count);
658 size = SIZE_MAX - sizeof (struct ceph_snap_context);
659 if (snap_count > size / sizeof (__le64))
663 * Not only that, but the size of the entire the snapshot
664 * header must also be representable in a size_t.
666 size -= snap_count * sizeof (__le64);
667 if ((u64) size < le64_to_cpu(ondisk->snap_names_len))
674 * Create a new header structure, translate header format from the on-disk
677 static int rbd_header_from_disk(struct rbd_image_header *header,
678 struct rbd_image_header_ondisk *ondisk)
685 memset(header, 0, sizeof (*header));
687 snap_count = le32_to_cpu(ondisk->snap_count);
689 len = strnlen(ondisk->object_prefix, sizeof (ondisk->object_prefix));
690 header->object_prefix = kmalloc(len + 1, GFP_KERNEL);
691 if (!header->object_prefix)
693 memcpy(header->object_prefix, ondisk->object_prefix, len);
694 header->object_prefix[len] = '\0';
697 u64 snap_names_len = le64_to_cpu(ondisk->snap_names_len);
699 /* Save a copy of the snapshot names */
701 if (snap_names_len > (u64) SIZE_MAX)
703 header->snap_names = kmalloc(snap_names_len, GFP_KERNEL);
704 if (!header->snap_names)
707 * Note that rbd_dev_v1_header_read() guarantees
708 * the ondisk buffer we're working with has
709 * snap_names_len bytes beyond the end of the
710 * snapshot id array, this memcpy() is safe.
712 memcpy(header->snap_names, &ondisk->snaps[snap_count],
715 /* Record each snapshot's size */
717 size = snap_count * sizeof (*header->snap_sizes);
718 header->snap_sizes = kmalloc(size, GFP_KERNEL);
719 if (!header->snap_sizes)
721 for (i = 0; i < snap_count; i++)
722 header->snap_sizes[i] =
723 le64_to_cpu(ondisk->snaps[i].image_size);
725 WARN_ON(ondisk->snap_names_len);
726 header->snap_names = NULL;
727 header->snap_sizes = NULL;
730 header->features = 0; /* No features support in v1 images */
731 header->obj_order = ondisk->options.order;
732 header->crypt_type = ondisk->options.crypt_type;
733 header->comp_type = ondisk->options.comp_type;
735 /* Allocate and fill in the snapshot context */
737 header->image_size = le64_to_cpu(ondisk->image_size);
738 size = sizeof (struct ceph_snap_context);
739 size += snap_count * sizeof (header->snapc->snaps[0]);
740 header->snapc = kzalloc(size, GFP_KERNEL);
744 atomic_set(&header->snapc->nref, 1);
745 header->snapc->seq = le64_to_cpu(ondisk->snap_seq);
746 header->snapc->num_snaps = snap_count;
747 for (i = 0; i < snap_count; i++)
748 header->snapc->snaps[i] =
749 le64_to_cpu(ondisk->snaps[i].id);
754 kfree(header->snap_sizes);
755 header->snap_sizes = NULL;
756 kfree(header->snap_names);
757 header->snap_names = NULL;
758 kfree(header->object_prefix);
759 header->object_prefix = NULL;
764 static const char *rbd_snap_name(struct rbd_device *rbd_dev, u64 snap_id)
766 struct rbd_snap *snap;
768 if (snap_id == CEPH_NOSNAP)
769 return RBD_SNAP_HEAD_NAME;
771 list_for_each_entry(snap, &rbd_dev->snaps, node)
772 if (snap_id == snap->id)
778 static int snap_by_name(struct rbd_device *rbd_dev, const char *snap_name)
781 struct rbd_snap *snap;
783 list_for_each_entry(snap, &rbd_dev->snaps, node) {
784 if (!strcmp(snap_name, snap->name)) {
785 rbd_dev->spec->snap_id = snap->id;
786 rbd_dev->mapping.size = snap->size;
787 rbd_dev->mapping.features = snap->features;
796 static int rbd_dev_set_mapping(struct rbd_device *rbd_dev)
800 if (!memcmp(rbd_dev->spec->snap_name, RBD_SNAP_HEAD_NAME,
801 sizeof (RBD_SNAP_HEAD_NAME))) {
802 rbd_dev->spec->snap_id = CEPH_NOSNAP;
803 rbd_dev->mapping.size = rbd_dev->header.image_size;
804 rbd_dev->mapping.features = rbd_dev->header.features;
807 ret = snap_by_name(rbd_dev, rbd_dev->spec->snap_name);
810 rbd_dev->mapping.read_only = true;
812 set_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags);
818 static void rbd_header_free(struct rbd_image_header *header)
820 kfree(header->object_prefix);
821 header->object_prefix = NULL;
822 kfree(header->snap_sizes);
823 header->snap_sizes = NULL;
824 kfree(header->snap_names);
825 header->snap_names = NULL;
826 ceph_put_snap_context(header->snapc);
827 header->snapc = NULL;
830 static const char *rbd_segment_name(struct rbd_device *rbd_dev, u64 offset)
836 name = kmalloc(MAX_OBJ_NAME_SIZE + 1, GFP_NOIO);
839 segment = offset >> rbd_dev->header.obj_order;
840 ret = snprintf(name, MAX_OBJ_NAME_SIZE + 1, "%s.%012llx",
841 rbd_dev->header.object_prefix, segment);
842 if (ret < 0 || ret > MAX_OBJ_NAME_SIZE) {
843 pr_err("error formatting segment name for #%llu (%d)\n",
852 static u64 rbd_segment_offset(struct rbd_device *rbd_dev, u64 offset)
854 u64 segment_size = (u64) 1 << rbd_dev->header.obj_order;
856 return offset & (segment_size - 1);
859 static u64 rbd_segment_length(struct rbd_device *rbd_dev,
860 u64 offset, u64 length)
862 u64 segment_size = (u64) 1 << rbd_dev->header.obj_order;
864 offset &= segment_size - 1;
866 rbd_assert(length <= U64_MAX - offset);
867 if (offset + length > segment_size)
868 length = segment_size - offset;
874 * returns the size of an object in the image
876 static u64 rbd_obj_bytes(struct rbd_image_header *header)
878 return 1 << header->obj_order;
885 static void bio_chain_put(struct bio *chain)
891 chain = chain->bi_next;
897 * zeros a bio chain, starting at specific offset
899 static void zero_bio_chain(struct bio *chain, int start_ofs)
908 bio_for_each_segment(bv, chain, i) {
909 if (pos + bv->bv_len > start_ofs) {
910 int remainder = max(start_ofs - pos, 0);
911 buf = bvec_kmap_irq(bv, &flags);
912 memset(buf + remainder, 0,
913 bv->bv_len - remainder);
914 bvec_kunmap_irq(buf, &flags);
919 chain = chain->bi_next;
924 * Clone a portion of a bio, starting at the given byte offset
925 * and continuing for the number of bytes indicated.
927 static struct bio *bio_clone_range(struct bio *bio_src,
936 unsigned short end_idx;
940 /* Handle the easy case for the caller */
942 if (!offset && len == bio_src->bi_size)
943 return bio_clone(bio_src, gfpmask);
945 if (WARN_ON_ONCE(!len))
947 if (WARN_ON_ONCE(len > bio_src->bi_size))
949 if (WARN_ON_ONCE(offset > bio_src->bi_size - len))
952 /* Find first affected segment... */
955 __bio_for_each_segment(bv, bio_src, idx, 0) {
956 if (resid < bv->bv_len)
962 /* ...and the last affected segment */
965 __bio_for_each_segment(bv, bio_src, end_idx, idx) {
966 if (resid <= bv->bv_len)
970 vcnt = end_idx - idx + 1;
972 /* Build the clone */
974 bio = bio_alloc(gfpmask, (unsigned int) vcnt);
976 return NULL; /* ENOMEM */
978 bio->bi_bdev = bio_src->bi_bdev;
979 bio->bi_sector = bio_src->bi_sector + (offset >> SECTOR_SHIFT);
980 bio->bi_rw = bio_src->bi_rw;
981 bio->bi_flags |= 1 << BIO_CLONED;
984 * Copy over our part of the bio_vec, then update the first
985 * and last (or only) entries.
987 memcpy(&bio->bi_io_vec[0], &bio_src->bi_io_vec[idx],
988 vcnt * sizeof (struct bio_vec));
989 bio->bi_io_vec[0].bv_offset += voff;
991 bio->bi_io_vec[0].bv_len -= voff;
992 bio->bi_io_vec[vcnt - 1].bv_len = resid;
994 bio->bi_io_vec[0].bv_len = len;
1005 * Clone a portion of a bio chain, starting at the given byte offset
1006 * into the first bio in the source chain and continuing for the
1007 * number of bytes indicated. The result is another bio chain of
1008 * exactly the given length, or a null pointer on error.
1010 * The bio_src and offset parameters are both in-out. On entry they
1011 * refer to the first source bio and the offset into that bio where
1012 * the start of data to be cloned is located.
1014 * On return, bio_src is updated to refer to the bio in the source
1015 * chain that contains first un-cloned byte, and *offset will
1016 * contain the offset of that byte within that bio.
1018 static struct bio *bio_chain_clone_range(struct bio **bio_src,
1019 unsigned int *offset,
1023 struct bio *bi = *bio_src;
1024 unsigned int off = *offset;
1025 struct bio *chain = NULL;
1028 /* Build up a chain of clone bios up to the limit */
1030 if (!bi || off >= bi->bi_size || !len)
1031 return NULL; /* Nothing to clone */
1035 unsigned int bi_size;
1039 rbd_warn(NULL, "bio_chain exhausted with %u left", len);
1040 goto out_err; /* EINVAL; ran out of bio's */
1042 bi_size = min_t(unsigned int, bi->bi_size - off, len);
1043 bio = bio_clone_range(bi, off, bi_size, gfpmask);
1045 goto out_err; /* ENOMEM */
1048 end = &bio->bi_next;
1051 if (off == bi->bi_size) {
1062 bio_chain_put(chain);
1067 static void rbd_obj_request_get(struct rbd_obj_request *obj_request)
1069 dout("%s: obj %p (was %d)\n", __func__, obj_request,
1070 atomic_read(&obj_request->kref.refcount));
1071 kref_get(&obj_request->kref);
1074 static void rbd_obj_request_destroy(struct kref *kref);
1075 static void rbd_obj_request_put(struct rbd_obj_request *obj_request)
1077 rbd_assert(obj_request != NULL);
1078 dout("%s: obj %p (was %d)\n", __func__, obj_request,
1079 atomic_read(&obj_request->kref.refcount));
1080 kref_put(&obj_request->kref, rbd_obj_request_destroy);
1083 static void rbd_img_request_get(struct rbd_img_request *img_request)
1085 dout("%s: img %p (was %d)\n", __func__, img_request,
1086 atomic_read(&img_request->kref.refcount));
1087 kref_get(&img_request->kref);
1090 static void rbd_img_request_destroy(struct kref *kref);
1091 static void rbd_img_request_put(struct rbd_img_request *img_request)
1093 rbd_assert(img_request != NULL);
1094 dout("%s: img %p (was %d)\n", __func__, img_request,
1095 atomic_read(&img_request->kref.refcount));
1096 kref_put(&img_request->kref, rbd_img_request_destroy);
1099 static inline void rbd_img_obj_request_add(struct rbd_img_request *img_request,
1100 struct rbd_obj_request *obj_request)
1102 rbd_assert(obj_request->img_request == NULL);
1104 rbd_obj_request_get(obj_request);
1105 obj_request->img_request = img_request;
1106 obj_request->which = img_request->obj_request_count;
1107 rbd_assert(obj_request->which != BAD_WHICH);
1108 img_request->obj_request_count++;
1109 list_add_tail(&obj_request->links, &img_request->obj_requests);
1110 dout("%s: img %p obj %p w=%u\n", __func__, img_request, obj_request,
1111 obj_request->which);
1114 static inline void rbd_img_obj_request_del(struct rbd_img_request *img_request,
1115 struct rbd_obj_request *obj_request)
1117 rbd_assert(obj_request->which != BAD_WHICH);
1119 dout("%s: img %p obj %p w=%u\n", __func__, img_request, obj_request,
1120 obj_request->which);
1121 list_del(&obj_request->links);
1122 rbd_assert(img_request->obj_request_count > 0);
1123 img_request->obj_request_count--;
1124 rbd_assert(obj_request->which == img_request->obj_request_count);
1125 obj_request->which = BAD_WHICH;
1126 rbd_assert(obj_request->img_request == img_request);
1127 obj_request->img_request = NULL;
1128 obj_request->callback = NULL;
1129 rbd_obj_request_put(obj_request);
1132 static bool obj_request_type_valid(enum obj_request_type type)
1135 case OBJ_REQUEST_NODATA:
1136 case OBJ_REQUEST_BIO:
1137 case OBJ_REQUEST_PAGES:
1144 static struct ceph_osd_req_op *rbd_osd_req_op_create(u16 opcode, ...)
1146 struct ceph_osd_req_op *op;
1150 op = kzalloc(sizeof (*op), GFP_NOIO);
1154 va_start(args, opcode);
1156 case CEPH_OSD_OP_READ:
1157 case CEPH_OSD_OP_WRITE:
1158 /* rbd_osd_req_op_create(READ, offset, length) */
1159 /* rbd_osd_req_op_create(WRITE, offset, length) */
1160 op->extent.offset = va_arg(args, u64);
1161 op->extent.length = va_arg(args, u64);
1162 if (opcode == CEPH_OSD_OP_WRITE)
1163 op->payload_len = op->extent.length;
1165 case CEPH_OSD_OP_STAT:
1167 case CEPH_OSD_OP_CALL:
1168 /* rbd_osd_req_op_create(CALL, class, method, data, datalen) */
1169 op->cls.class_name = va_arg(args, char *);
1170 size = strlen(op->cls.class_name);
1171 rbd_assert(size <= (size_t) U8_MAX);
1172 op->cls.class_len = size;
1173 op->payload_len = size;
1175 op->cls.method_name = va_arg(args, char *);
1176 size = strlen(op->cls.method_name);
1177 rbd_assert(size <= (size_t) U8_MAX);
1178 op->cls.method_len = size;
1179 op->payload_len += size;
1182 op->cls.indata = va_arg(args, void *);
1183 size = va_arg(args, size_t);
1184 rbd_assert(size <= (size_t) U32_MAX);
1185 op->cls.indata_len = (u32) size;
1186 op->payload_len += size;
1188 case CEPH_OSD_OP_NOTIFY_ACK:
1189 case CEPH_OSD_OP_WATCH:
1190 /* rbd_osd_req_op_create(NOTIFY_ACK, cookie, version) */
1191 /* rbd_osd_req_op_create(WATCH, cookie, version, flag) */
1192 op->watch.cookie = va_arg(args, u64);
1193 op->watch.ver = va_arg(args, u64);
1194 op->watch.ver = cpu_to_le64(op->watch.ver);
1195 if (opcode == CEPH_OSD_OP_WATCH && va_arg(args, int))
1196 op->watch.flag = (u8) 1;
1199 rbd_warn(NULL, "unsupported opcode %hu\n", opcode);
1209 static void rbd_osd_req_op_destroy(struct ceph_osd_req_op *op)
1214 static int rbd_obj_request_submit(struct ceph_osd_client *osdc,
1215 struct rbd_obj_request *obj_request)
1217 dout("%s: osdc %p obj %p\n", __func__, osdc, obj_request);
1219 return ceph_osdc_start_request(osdc, obj_request->osd_req, false);
1222 static void rbd_img_request_complete(struct rbd_img_request *img_request)
1224 dout("%s: img %p\n", __func__, img_request);
1225 if (img_request->callback)
1226 img_request->callback(img_request);
1228 rbd_img_request_put(img_request);
1231 /* Caller is responsible for rbd_obj_request_destroy(obj_request) */
1233 static int rbd_obj_request_wait(struct rbd_obj_request *obj_request)
1235 dout("%s: obj %p\n", __func__, obj_request);
1237 return wait_for_completion_interruptible(&obj_request->completion);
1240 static void obj_request_done_init(struct rbd_obj_request *obj_request)
1242 atomic_set(&obj_request->done, 0);
1246 static void obj_request_done_set(struct rbd_obj_request *obj_request)
1250 done = atomic_inc_return(&obj_request->done);
1252 struct rbd_img_request *img_request = obj_request->img_request;
1253 struct rbd_device *rbd_dev;
1255 rbd_dev = img_request ? img_request->rbd_dev : NULL;
1256 rbd_warn(rbd_dev, "obj_request %p was already done\n",
1261 static bool obj_request_done_test(struct rbd_obj_request *obj_request)
1264 return atomic_read(&obj_request->done) != 0;
1268 rbd_img_obj_request_read_callback(struct rbd_obj_request *obj_request)
1270 dout("%s: obj %p img %p result %d %llu/%llu\n", __func__,
1271 obj_request, obj_request->img_request, obj_request->result,
1272 obj_request->xferred, obj_request->length);
1274 * ENOENT means a hole in the image. We zero-fill the
1275 * entire length of the request. A short read also implies
1276 * zero-fill to the end of the request. Either way we
1277 * update the xferred count to indicate the whole request
1280 BUG_ON(obj_request->type != OBJ_REQUEST_BIO);
1281 if (obj_request->result == -ENOENT) {
1282 zero_bio_chain(obj_request->bio_list, 0);
1283 obj_request->result = 0;
1284 obj_request->xferred = obj_request->length;
1285 } else if (obj_request->xferred < obj_request->length &&
1286 !obj_request->result) {
1287 zero_bio_chain(obj_request->bio_list, obj_request->xferred);
1288 obj_request->xferred = obj_request->length;
1290 obj_request_done_set(obj_request);
1293 static void rbd_obj_request_complete(struct rbd_obj_request *obj_request)
1295 dout("%s: obj %p cb %p\n", __func__, obj_request,
1296 obj_request->callback);
1297 if (obj_request->callback)
1298 obj_request->callback(obj_request);
1300 complete_all(&obj_request->completion);
1303 static void rbd_osd_trivial_callback(struct rbd_obj_request *obj_request)
1305 dout("%s: obj %p\n", __func__, obj_request);
1306 obj_request_done_set(obj_request);
1309 static void rbd_osd_read_callback(struct rbd_obj_request *obj_request)
1311 dout("%s: obj %p result %d %llu/%llu\n", __func__, obj_request,
1312 obj_request->result, obj_request->xferred, obj_request->length);
1313 if (obj_request->img_request)
1314 rbd_img_obj_request_read_callback(obj_request);
1316 obj_request_done_set(obj_request);
1319 static void rbd_osd_write_callback(struct rbd_obj_request *obj_request)
1321 dout("%s: obj %p result %d %llu\n", __func__, obj_request,
1322 obj_request->result, obj_request->length);
1324 * There is no such thing as a successful short write.
1325 * Our xferred value is the number of bytes transferred
1326 * back. Set it to our originally-requested length.
1328 obj_request->xferred = obj_request->length;
1329 obj_request_done_set(obj_request);
1333 * For a simple stat call there's nothing to do. We'll do more if
1334 * this is part of a write sequence for a layered image.
1336 static void rbd_osd_stat_callback(struct rbd_obj_request *obj_request)
1338 dout("%s: obj %p\n", __func__, obj_request);
1339 obj_request_done_set(obj_request);
1342 static void rbd_osd_req_callback(struct ceph_osd_request *osd_req,
1343 struct ceph_msg *msg)
1345 struct rbd_obj_request *obj_request = osd_req->r_priv;
1348 dout("%s: osd_req %p msg %p\n", __func__, osd_req, msg);
1349 rbd_assert(osd_req == obj_request->osd_req);
1350 rbd_assert(!!obj_request->img_request ^
1351 (obj_request->which == BAD_WHICH));
1353 if (osd_req->r_result < 0)
1354 obj_request->result = osd_req->r_result;
1355 obj_request->version = le64_to_cpu(osd_req->r_reassert_version.version);
1357 WARN_ON(osd_req->r_num_ops != 1); /* For now */
1360 * We support a 64-bit length, but ultimately it has to be
1361 * passed to blk_end_request(), which takes an unsigned int.
1363 obj_request->xferred = osd_req->r_reply_op_len[0];
1364 rbd_assert(obj_request->xferred < (u64) UINT_MAX);
1365 opcode = osd_req->r_request_ops[0].op;
1367 case CEPH_OSD_OP_READ:
1368 rbd_osd_read_callback(obj_request);
1370 case CEPH_OSD_OP_WRITE:
1371 rbd_osd_write_callback(obj_request);
1373 case CEPH_OSD_OP_STAT:
1374 rbd_osd_stat_callback(obj_request);
1376 case CEPH_OSD_OP_CALL:
1377 case CEPH_OSD_OP_NOTIFY_ACK:
1378 case CEPH_OSD_OP_WATCH:
1379 rbd_osd_trivial_callback(obj_request);
1382 rbd_warn(NULL, "%s: unsupported op %hu\n",
1383 obj_request->object_name, (unsigned short) opcode);
1387 if (obj_request_done_test(obj_request))
1388 rbd_obj_request_complete(obj_request);
1391 static struct ceph_osd_request *rbd_osd_req_create(
1392 struct rbd_device *rbd_dev,
1394 struct rbd_obj_request *obj_request,
1395 struct ceph_osd_req_op *op)
1397 struct rbd_img_request *img_request = obj_request->img_request;
1398 struct ceph_snap_context *snapc = NULL;
1399 struct ceph_osd_client *osdc;
1400 struct ceph_osd_request *osd_req;
1401 struct timespec now;
1402 struct timespec *mtime;
1403 u64 snap_id = CEPH_NOSNAP;
1404 u64 offset = obj_request->offset;
1405 u64 length = obj_request->length;
1408 rbd_assert(img_request->write_request == write_request);
1409 if (img_request->write_request)
1410 snapc = img_request->snapc;
1412 snap_id = img_request->snap_id;
1415 /* Allocate and initialize the request, for the single op */
1417 osdc = &rbd_dev->rbd_client->client->osdc;
1418 osd_req = ceph_osdc_alloc_request(osdc, snapc, 1, false, GFP_ATOMIC);
1420 return NULL; /* ENOMEM */
1422 rbd_assert(obj_request_type_valid(obj_request->type));
1423 switch (obj_request->type) {
1424 case OBJ_REQUEST_NODATA:
1425 break; /* Nothing to do */
1426 case OBJ_REQUEST_BIO:
1427 rbd_assert(obj_request->bio_list != NULL);
1428 osd_req->r_data.type = CEPH_OSD_DATA_TYPE_BIO;
1429 osd_req->r_data.bio = obj_request->bio_list;
1431 case OBJ_REQUEST_PAGES:
1432 osd_req->r_data.type = CEPH_OSD_DATA_TYPE_PAGES;
1433 osd_req->r_data.pages = obj_request->pages;
1434 osd_req->r_data.num_pages = obj_request->page_count;
1435 osd_req->r_data.alignment = offset & ~PAGE_MASK;
1436 osd_req->r_data.pages_from_pool = false;
1437 osd_req->r_data.own_pages = false;
1441 if (write_request) {
1442 osd_req->r_flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK;
1446 osd_req->r_flags = CEPH_OSD_FLAG_READ;
1447 mtime = NULL; /* not needed for reads */
1448 offset = 0; /* These are not used... */
1449 length = 0; /* ...for osd read requests */
1452 osd_req->r_callback = rbd_osd_req_callback;
1453 osd_req->r_priv = obj_request;
1455 osd_req->r_oid_len = strlen(obj_request->object_name);
1456 rbd_assert(osd_req->r_oid_len < sizeof (osd_req->r_oid));
1457 memcpy(osd_req->r_oid, obj_request->object_name, osd_req->r_oid_len);
1459 osd_req->r_file_layout = rbd_dev->layout; /* struct */
1461 /* osd_req will get its own reference to snapc (if non-null) */
1463 ceph_osdc_build_request(osd_req, offset, length, 1, op,
1464 snapc, snap_id, mtime);
1469 static void rbd_osd_req_destroy(struct ceph_osd_request *osd_req)
1471 ceph_osdc_put_request(osd_req);
1474 /* object_name is assumed to be a non-null pointer and NUL-terminated */
1476 static struct rbd_obj_request *rbd_obj_request_create(const char *object_name,
1477 u64 offset, u64 length,
1478 enum obj_request_type type)
1480 struct rbd_obj_request *obj_request;
1484 rbd_assert(obj_request_type_valid(type));
1486 size = strlen(object_name) + 1;
1487 obj_request = kzalloc(sizeof (*obj_request) + size, GFP_KERNEL);
1491 name = (char *)(obj_request + 1);
1492 obj_request->object_name = memcpy(name, object_name, size);
1493 obj_request->offset = offset;
1494 obj_request->length = length;
1495 obj_request->which = BAD_WHICH;
1496 obj_request->type = type;
1497 INIT_LIST_HEAD(&obj_request->links);
1498 obj_request_done_init(obj_request);
1499 init_completion(&obj_request->completion);
1500 kref_init(&obj_request->kref);
1502 dout("%s: \"%s\" %llu/%llu %d -> obj %p\n", __func__, object_name,
1503 offset, length, (int)type, obj_request);
1508 static void rbd_obj_request_destroy(struct kref *kref)
1510 struct rbd_obj_request *obj_request;
1512 obj_request = container_of(kref, struct rbd_obj_request, kref);
1514 dout("%s: obj %p\n", __func__, obj_request);
1516 rbd_assert(obj_request->img_request == NULL);
1517 rbd_assert(obj_request->which == BAD_WHICH);
1519 if (obj_request->osd_req)
1520 rbd_osd_req_destroy(obj_request->osd_req);
1522 rbd_assert(obj_request_type_valid(obj_request->type));
1523 switch (obj_request->type) {
1524 case OBJ_REQUEST_NODATA:
1525 break; /* Nothing to do */
1526 case OBJ_REQUEST_BIO:
1527 if (obj_request->bio_list)
1528 bio_chain_put(obj_request->bio_list);
1530 case OBJ_REQUEST_PAGES:
1531 if (obj_request->pages)
1532 ceph_release_page_vector(obj_request->pages,
1533 obj_request->page_count);
1541 * Caller is responsible for filling in the list of object requests
1542 * that comprises the image request, and the Linux request pointer
1543 * (if there is one).
1545 static struct rbd_img_request *rbd_img_request_create(
1546 struct rbd_device *rbd_dev,
1547 u64 offset, u64 length,
1550 struct rbd_img_request *img_request;
1551 struct ceph_snap_context *snapc = NULL;
1553 img_request = kmalloc(sizeof (*img_request), GFP_ATOMIC);
1557 if (write_request) {
1558 down_read(&rbd_dev->header_rwsem);
1559 snapc = ceph_get_snap_context(rbd_dev->header.snapc);
1560 up_read(&rbd_dev->header_rwsem);
1561 if (WARN_ON(!snapc)) {
1563 return NULL; /* Shouldn't happen */
1567 img_request->rq = NULL;
1568 img_request->rbd_dev = rbd_dev;
1569 img_request->offset = offset;
1570 img_request->length = length;
1571 img_request->write_request = write_request;
1573 img_request->snapc = snapc;
1575 img_request->snap_id = rbd_dev->spec->snap_id;
1576 spin_lock_init(&img_request->completion_lock);
1577 img_request->next_completion = 0;
1578 img_request->callback = NULL;
1579 img_request->obj_request_count = 0;
1580 INIT_LIST_HEAD(&img_request->obj_requests);
1581 kref_init(&img_request->kref);
1583 rbd_img_request_get(img_request); /* Avoid a warning */
1584 rbd_img_request_put(img_request); /* TEMPORARY */
1586 dout("%s: rbd_dev %p %s %llu/%llu -> img %p\n", __func__, rbd_dev,
1587 write_request ? "write" : "read", offset, length,
1593 static void rbd_img_request_destroy(struct kref *kref)
1595 struct rbd_img_request *img_request;
1596 struct rbd_obj_request *obj_request;
1597 struct rbd_obj_request *next_obj_request;
1599 img_request = container_of(kref, struct rbd_img_request, kref);
1601 dout("%s: img %p\n", __func__, img_request);
1603 for_each_obj_request_safe(img_request, obj_request, next_obj_request)
1604 rbd_img_obj_request_del(img_request, obj_request);
1605 rbd_assert(img_request->obj_request_count == 0);
1607 if (img_request->write_request)
1608 ceph_put_snap_context(img_request->snapc);
1613 static int rbd_img_request_fill_bio(struct rbd_img_request *img_request,
1614 struct bio *bio_list)
1616 struct rbd_device *rbd_dev = img_request->rbd_dev;
1617 struct rbd_obj_request *obj_request = NULL;
1618 struct rbd_obj_request *next_obj_request;
1619 unsigned int bio_offset;
1624 dout("%s: img %p bio %p\n", __func__, img_request, bio_list);
1626 opcode = img_request->write_request ? CEPH_OSD_OP_WRITE
1629 image_offset = img_request->offset;
1630 rbd_assert(image_offset == bio_list->bi_sector << SECTOR_SHIFT);
1631 resid = img_request->length;
1632 rbd_assert(resid > 0);
1634 const char *object_name;
1635 unsigned int clone_size;
1636 struct ceph_osd_req_op *op;
1640 object_name = rbd_segment_name(rbd_dev, image_offset);
1643 offset = rbd_segment_offset(rbd_dev, image_offset);
1644 length = rbd_segment_length(rbd_dev, image_offset, resid);
1645 obj_request = rbd_obj_request_create(object_name,
1648 kfree(object_name); /* object request has its own copy */
1652 rbd_assert(length <= (u64) UINT_MAX);
1653 clone_size = (unsigned int) length;
1654 obj_request->bio_list = bio_chain_clone_range(&bio_list,
1655 &bio_offset, clone_size,
1657 if (!obj_request->bio_list)
1661 * Build up the op to use in building the osd
1662 * request. Note that the contents of the op are
1663 * copied by rbd_osd_req_create().
1665 op = rbd_osd_req_op_create(opcode, offset, length);
1668 obj_request->osd_req = rbd_osd_req_create(rbd_dev,
1669 img_request->write_request,
1671 rbd_osd_req_op_destroy(op);
1672 if (!obj_request->osd_req)
1674 /* status and version are initially zero-filled */
1676 rbd_img_obj_request_add(img_request, obj_request);
1678 image_offset += length;
1685 rbd_obj_request_put(obj_request);
1687 for_each_obj_request_safe(img_request, obj_request, next_obj_request)
1688 rbd_obj_request_put(obj_request);
1693 static void rbd_img_obj_callback(struct rbd_obj_request *obj_request)
1695 struct rbd_img_request *img_request;
1696 u32 which = obj_request->which;
1699 img_request = obj_request->img_request;
1701 dout("%s: img %p obj %p\n", __func__, img_request, obj_request);
1702 rbd_assert(img_request != NULL);
1703 rbd_assert(img_request->rq != NULL);
1704 rbd_assert(img_request->obj_request_count > 0);
1705 rbd_assert(which != BAD_WHICH);
1706 rbd_assert(which < img_request->obj_request_count);
1707 rbd_assert(which >= img_request->next_completion);
1709 spin_lock_irq(&img_request->completion_lock);
1710 if (which != img_request->next_completion)
1713 for_each_obj_request_from(img_request, obj_request) {
1714 unsigned int xferred;
1718 rbd_assert(which < img_request->obj_request_count);
1720 if (!obj_request_done_test(obj_request))
1723 rbd_assert(obj_request->xferred <= (u64) UINT_MAX);
1724 xferred = (unsigned int) obj_request->xferred;
1725 result = (int) obj_request->result;
1727 rbd_warn(NULL, "obj_request %s result %d xferred %u\n",
1728 img_request->write_request ? "write" : "read",
1731 more = blk_end_request(img_request->rq, result, xferred);
1735 rbd_assert(more ^ (which == img_request->obj_request_count));
1736 img_request->next_completion = which;
1738 spin_unlock_irq(&img_request->completion_lock);
1741 rbd_img_request_complete(img_request);
1744 static int rbd_img_request_submit(struct rbd_img_request *img_request)
1746 struct rbd_device *rbd_dev = img_request->rbd_dev;
1747 struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
1748 struct rbd_obj_request *obj_request;
1749 struct rbd_obj_request *next_obj_request;
1751 dout("%s: img %p\n", __func__, img_request);
1752 for_each_obj_request_safe(img_request, obj_request, next_obj_request) {
1755 obj_request->callback = rbd_img_obj_callback;
1756 ret = rbd_obj_request_submit(osdc, obj_request);
1760 * The image request has its own reference to each
1761 * of its object requests, so we can safely drop the
1764 rbd_obj_request_put(obj_request);
1770 static int rbd_obj_notify_ack(struct rbd_device *rbd_dev,
1771 u64 ver, u64 notify_id)
1773 struct rbd_obj_request *obj_request;
1774 struct ceph_osd_req_op *op;
1775 struct ceph_osd_client *osdc;
1778 obj_request = rbd_obj_request_create(rbd_dev->header_name, 0, 0,
1779 OBJ_REQUEST_NODATA);
1784 op = rbd_osd_req_op_create(CEPH_OSD_OP_NOTIFY_ACK, notify_id, ver);
1787 obj_request->osd_req = rbd_osd_req_create(rbd_dev, false,
1789 rbd_osd_req_op_destroy(op);
1790 if (!obj_request->osd_req)
1793 osdc = &rbd_dev->rbd_client->client->osdc;
1794 obj_request->callback = rbd_obj_request_put;
1795 ret = rbd_obj_request_submit(osdc, obj_request);
1798 rbd_obj_request_put(obj_request);
1803 static void rbd_watch_cb(u64 ver, u64 notify_id, u8 opcode, void *data)
1805 struct rbd_device *rbd_dev = (struct rbd_device *)data;
1812 dout("%s: \"%s\" notify_id %llu opcode %u\n", __func__,
1813 rbd_dev->header_name, (unsigned long long) notify_id,
1814 (unsigned int) opcode);
1815 rc = rbd_dev_refresh(rbd_dev, &hver);
1817 rbd_warn(rbd_dev, "got notification but failed to "
1818 " update snaps: %d\n", rc);
1820 rbd_obj_notify_ack(rbd_dev, hver, notify_id);
1824 * Request sync osd watch/unwatch. The value of "start" determines
1825 * whether a watch request is being initiated or torn down.
1827 static int rbd_dev_header_watch_sync(struct rbd_device *rbd_dev, int start)
1829 struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
1830 struct rbd_obj_request *obj_request;
1831 struct ceph_osd_req_op *op;
1834 rbd_assert(start ^ !!rbd_dev->watch_event);
1835 rbd_assert(start ^ !!rbd_dev->watch_request);
1838 ret = ceph_osdc_create_event(osdc, rbd_watch_cb, rbd_dev,
1839 &rbd_dev->watch_event);
1842 rbd_assert(rbd_dev->watch_event != NULL);
1846 obj_request = rbd_obj_request_create(rbd_dev->header_name, 0, 0,
1847 OBJ_REQUEST_NODATA);
1851 op = rbd_osd_req_op_create(CEPH_OSD_OP_WATCH,
1852 rbd_dev->watch_event->cookie,
1853 rbd_dev->header.obj_version, start);
1856 obj_request->osd_req = rbd_osd_req_create(rbd_dev, true,
1858 rbd_osd_req_op_destroy(op);
1859 if (!obj_request->osd_req)
1863 ceph_osdc_set_request_linger(osdc, obj_request->osd_req);
1865 ceph_osdc_unregister_linger_request(osdc,
1866 rbd_dev->watch_request->osd_req);
1867 ret = rbd_obj_request_submit(osdc, obj_request);
1870 ret = rbd_obj_request_wait(obj_request);
1873 ret = obj_request->result;
1878 * A watch request is set to linger, so the underlying osd
1879 * request won't go away until we unregister it. We retain
1880 * a pointer to the object request during that time (in
1881 * rbd_dev->watch_request), so we'll keep a reference to
1882 * it. We'll drop that reference (below) after we've
1886 rbd_dev->watch_request = obj_request;
1891 /* We have successfully torn down the watch request */
1893 rbd_obj_request_put(rbd_dev->watch_request);
1894 rbd_dev->watch_request = NULL;
1896 /* Cancel the event if we're tearing down, or on error */
1897 ceph_osdc_cancel_event(rbd_dev->watch_event);
1898 rbd_dev->watch_event = NULL;
1900 rbd_obj_request_put(obj_request);
1906 * Synchronous osd object method call
1908 static int rbd_obj_method_sync(struct rbd_device *rbd_dev,
1909 const char *object_name,
1910 const char *class_name,
1911 const char *method_name,
1912 const char *outbound,
1913 size_t outbound_size,
1915 size_t inbound_size,
1918 struct rbd_obj_request *obj_request;
1919 struct ceph_osd_client *osdc;
1920 struct ceph_osd_req_op *op;
1921 struct page **pages;
1926 * Method calls are ultimately read operations but they
1927 * don't involve object data (so no offset or length).
1928 * The result should placed into the inbound buffer
1929 * provided. They also supply outbound data--parameters for
1930 * the object method. Currently if this is present it will
1933 page_count = (u32) calc_pages_for(0, inbound_size);
1934 pages = ceph_alloc_page_vector(page_count, GFP_KERNEL);
1936 return PTR_ERR(pages);
1939 obj_request = rbd_obj_request_create(object_name, 0, 0,
1944 obj_request->pages = pages;
1945 obj_request->page_count = page_count;
1947 op = rbd_osd_req_op_create(CEPH_OSD_OP_CALL, class_name,
1948 method_name, outbound, outbound_size);
1951 obj_request->osd_req = rbd_osd_req_create(rbd_dev, false,
1953 rbd_osd_req_op_destroy(op);
1954 if (!obj_request->osd_req)
1957 osdc = &rbd_dev->rbd_client->client->osdc;
1958 ret = rbd_obj_request_submit(osdc, obj_request);
1961 ret = rbd_obj_request_wait(obj_request);
1965 ret = obj_request->result;
1969 ceph_copy_from_page_vector(pages, inbound, 0, obj_request->xferred);
1971 *version = obj_request->version;
1974 rbd_obj_request_put(obj_request);
1976 ceph_release_page_vector(pages, page_count);
1981 static void rbd_request_fn(struct request_queue *q)
1982 __releases(q->queue_lock) __acquires(q->queue_lock)
1984 struct rbd_device *rbd_dev = q->queuedata;
1985 bool read_only = rbd_dev->mapping.read_only;
1989 while ((rq = blk_fetch_request(q))) {
1990 bool write_request = rq_data_dir(rq) == WRITE;
1991 struct rbd_img_request *img_request;
1995 /* Ignore any non-FS requests that filter through. */
1997 if (rq->cmd_type != REQ_TYPE_FS) {
1998 dout("%s: non-fs request type %d\n", __func__,
1999 (int) rq->cmd_type);
2000 __blk_end_request_all(rq, 0);
2004 /* Ignore/skip any zero-length requests */
2006 offset = (u64) blk_rq_pos(rq) << SECTOR_SHIFT;
2007 length = (u64) blk_rq_bytes(rq);
2010 dout("%s: zero-length request\n", __func__);
2011 __blk_end_request_all(rq, 0);
2015 spin_unlock_irq(q->queue_lock);
2017 /* Disallow writes to a read-only device */
2019 if (write_request) {
2023 rbd_assert(rbd_dev->spec->snap_id == CEPH_NOSNAP);
2027 * Quit early if the mapped snapshot no longer
2028 * exists. It's still possible the snapshot will
2029 * have disappeared by the time our request arrives
2030 * at the osd, but there's no sense in sending it if
2033 if (!test_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags)) {
2034 dout("request for non-existent snapshot");
2035 rbd_assert(rbd_dev->spec->snap_id != CEPH_NOSNAP);
2041 if (WARN_ON(offset && length > U64_MAX - offset + 1))
2042 goto end_request; /* Shouldn't happen */
2045 img_request = rbd_img_request_create(rbd_dev, offset, length,
2050 img_request->rq = rq;
2052 result = rbd_img_request_fill_bio(img_request, rq->bio);
2054 result = rbd_img_request_submit(img_request);
2056 rbd_img_request_put(img_request);
2058 spin_lock_irq(q->queue_lock);
2060 rbd_warn(rbd_dev, "obj_request %s result %d\n",
2061 write_request ? "write" : "read", result);
2062 __blk_end_request_all(rq, result);
2068 * a queue callback. Makes sure that we don't create a bio that spans across
2069 * multiple osd objects. One exception would be with a single page bios,
2070 * which we handle later at bio_chain_clone_range()
2072 static int rbd_merge_bvec(struct request_queue *q, struct bvec_merge_data *bmd,
2073 struct bio_vec *bvec)
2075 struct rbd_device *rbd_dev = q->queuedata;
2076 sector_t sector_offset;
2077 sector_t sectors_per_obj;
2078 sector_t obj_sector_offset;
2082 * Find how far into its rbd object the partition-relative
2083 * bio start sector is to offset relative to the enclosing
2086 sector_offset = get_start_sect(bmd->bi_bdev) + bmd->bi_sector;
2087 sectors_per_obj = 1 << (rbd_dev->header.obj_order - SECTOR_SHIFT);
2088 obj_sector_offset = sector_offset & (sectors_per_obj - 1);
2091 * Compute the number of bytes from that offset to the end
2092 * of the object. Account for what's already used by the bio.
2094 ret = (int) (sectors_per_obj - obj_sector_offset) << SECTOR_SHIFT;
2095 if (ret > bmd->bi_size)
2096 ret -= bmd->bi_size;
2101 * Don't send back more than was asked for. And if the bio
2102 * was empty, let the whole thing through because: "Note
2103 * that a block device *must* allow a single page to be
2104 * added to an empty bio."
2106 rbd_assert(bvec->bv_len <= PAGE_SIZE);
2107 if (ret > (int) bvec->bv_len || !bmd->bi_size)
2108 ret = (int) bvec->bv_len;
2113 static void rbd_free_disk(struct rbd_device *rbd_dev)
2115 struct gendisk *disk = rbd_dev->disk;
2120 if (disk->flags & GENHD_FL_UP)
2123 blk_cleanup_queue(disk->queue);
2127 static int rbd_obj_read_sync(struct rbd_device *rbd_dev,
2128 const char *object_name,
2129 u64 offset, u64 length,
2130 char *buf, u64 *version)
2133 struct ceph_osd_req_op *op;
2134 struct rbd_obj_request *obj_request;
2135 struct ceph_osd_client *osdc;
2136 struct page **pages = NULL;
2141 page_count = (u32) calc_pages_for(offset, length);
2142 pages = ceph_alloc_page_vector(page_count, GFP_KERNEL);
2144 ret = PTR_ERR(pages);
2147 obj_request = rbd_obj_request_create(object_name, offset, length,
2152 obj_request->pages = pages;
2153 obj_request->page_count = page_count;
2155 op = rbd_osd_req_op_create(CEPH_OSD_OP_READ, offset, length);
2158 obj_request->osd_req = rbd_osd_req_create(rbd_dev, false,
2160 rbd_osd_req_op_destroy(op);
2161 if (!obj_request->osd_req)
2164 osdc = &rbd_dev->rbd_client->client->osdc;
2165 ret = rbd_obj_request_submit(osdc, obj_request);
2168 ret = rbd_obj_request_wait(obj_request);
2172 ret = obj_request->result;
2176 rbd_assert(obj_request->xferred <= (u64) SIZE_MAX);
2177 size = (size_t) obj_request->xferred;
2178 ceph_copy_from_page_vector(pages, buf, 0, size);
2179 rbd_assert(size <= (size_t) INT_MAX);
2182 *version = obj_request->version;
2185 rbd_obj_request_put(obj_request);
2187 ceph_release_page_vector(pages, page_count);
2193 * Read the complete header for the given rbd device.
2195 * Returns a pointer to a dynamically-allocated buffer containing
2196 * the complete and validated header. Caller can pass the address
2197 * of a variable that will be filled in with the version of the
2198 * header object at the time it was read.
2200 * Returns a pointer-coded errno if a failure occurs.
2202 static struct rbd_image_header_ondisk *
2203 rbd_dev_v1_header_read(struct rbd_device *rbd_dev, u64 *version)
2205 struct rbd_image_header_ondisk *ondisk = NULL;
2212 * The complete header will include an array of its 64-bit
2213 * snapshot ids, followed by the names of those snapshots as
2214 * a contiguous block of NUL-terminated strings. Note that
2215 * the number of snapshots could change by the time we read
2216 * it in, in which case we re-read it.
2223 size = sizeof (*ondisk);
2224 size += snap_count * sizeof (struct rbd_image_snap_ondisk);
2226 ondisk = kmalloc(size, GFP_KERNEL);
2228 return ERR_PTR(-ENOMEM);
2230 ret = rbd_obj_read_sync(rbd_dev, rbd_dev->header_name,
2232 (char *) ondisk, version);
2235 if (WARN_ON((size_t) ret < size)) {
2237 rbd_warn(rbd_dev, "short header read (want %zd got %d)",
2241 if (!rbd_dev_ondisk_valid(ondisk)) {
2243 rbd_warn(rbd_dev, "invalid header");
2247 names_size = le64_to_cpu(ondisk->snap_names_len);
2248 want_count = snap_count;
2249 snap_count = le32_to_cpu(ondisk->snap_count);
2250 } while (snap_count != want_count);
2257 return ERR_PTR(ret);
2261 * reload the ondisk the header
2263 static int rbd_read_header(struct rbd_device *rbd_dev,
2264 struct rbd_image_header *header)
2266 struct rbd_image_header_ondisk *ondisk;
2270 ondisk = rbd_dev_v1_header_read(rbd_dev, &ver);
2272 return PTR_ERR(ondisk);
2273 ret = rbd_header_from_disk(header, ondisk);
2275 header->obj_version = ver;
2281 static void rbd_remove_all_snaps(struct rbd_device *rbd_dev)
2283 struct rbd_snap *snap;
2284 struct rbd_snap *next;
2286 list_for_each_entry_safe(snap, next, &rbd_dev->snaps, node)
2287 rbd_remove_snap_dev(snap);
2290 static void rbd_update_mapping_size(struct rbd_device *rbd_dev)
2294 if (rbd_dev->spec->snap_id != CEPH_NOSNAP)
2297 size = (sector_t) rbd_dev->header.image_size / SECTOR_SIZE;
2298 dout("setting size to %llu sectors", (unsigned long long) size);
2299 rbd_dev->mapping.size = (u64) size;
2300 set_capacity(rbd_dev->disk, size);
2304 * only read the first part of the ondisk header, without the snaps info
2306 static int rbd_dev_v1_refresh(struct rbd_device *rbd_dev, u64 *hver)
2309 struct rbd_image_header h;
2311 ret = rbd_read_header(rbd_dev, &h);
2315 down_write(&rbd_dev->header_rwsem);
2317 /* Update image size, and check for resize of mapped image */
2318 rbd_dev->header.image_size = h.image_size;
2319 rbd_update_mapping_size(rbd_dev);
2321 /* rbd_dev->header.object_prefix shouldn't change */
2322 kfree(rbd_dev->header.snap_sizes);
2323 kfree(rbd_dev->header.snap_names);
2324 /* osd requests may still refer to snapc */
2325 ceph_put_snap_context(rbd_dev->header.snapc);
2328 *hver = h.obj_version;
2329 rbd_dev->header.obj_version = h.obj_version;
2330 rbd_dev->header.image_size = h.image_size;
2331 rbd_dev->header.snapc = h.snapc;
2332 rbd_dev->header.snap_names = h.snap_names;
2333 rbd_dev->header.snap_sizes = h.snap_sizes;
2334 /* Free the extra copy of the object prefix */
2335 WARN_ON(strcmp(rbd_dev->header.object_prefix, h.object_prefix));
2336 kfree(h.object_prefix);
2338 ret = rbd_dev_snaps_update(rbd_dev);
2340 ret = rbd_dev_snaps_register(rbd_dev);
2342 up_write(&rbd_dev->header_rwsem);
2347 static int rbd_dev_refresh(struct rbd_device *rbd_dev, u64 *hver)
2351 rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
2352 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
2353 if (rbd_dev->image_format == 1)
2354 ret = rbd_dev_v1_refresh(rbd_dev, hver);
2356 ret = rbd_dev_v2_refresh(rbd_dev, hver);
2357 mutex_unlock(&ctl_mutex);
2362 static int rbd_init_disk(struct rbd_device *rbd_dev)
2364 struct gendisk *disk;
2365 struct request_queue *q;
2368 /* create gendisk info */
2369 disk = alloc_disk(RBD_MINORS_PER_MAJOR);
2373 snprintf(disk->disk_name, sizeof(disk->disk_name), RBD_DRV_NAME "%d",
2375 disk->major = rbd_dev->major;
2376 disk->first_minor = 0;
2377 disk->fops = &rbd_bd_ops;
2378 disk->private_data = rbd_dev;
2380 q = blk_init_queue(rbd_request_fn, &rbd_dev->lock);
2384 /* We use the default size, but let's be explicit about it. */
2385 blk_queue_physical_block_size(q, SECTOR_SIZE);
2387 /* set io sizes to object size */
2388 segment_size = rbd_obj_bytes(&rbd_dev->header);
2389 blk_queue_max_hw_sectors(q, segment_size / SECTOR_SIZE);
2390 blk_queue_max_segment_size(q, segment_size);
2391 blk_queue_io_min(q, segment_size);
2392 blk_queue_io_opt(q, segment_size);
2394 blk_queue_merge_bvec(q, rbd_merge_bvec);
2397 q->queuedata = rbd_dev;
2399 rbd_dev->disk = disk;
2401 set_capacity(rbd_dev->disk, rbd_dev->mapping.size / SECTOR_SIZE);
2414 static struct rbd_device *dev_to_rbd_dev(struct device *dev)
2416 return container_of(dev, struct rbd_device, dev);
2419 static ssize_t rbd_size_show(struct device *dev,
2420 struct device_attribute *attr, char *buf)
2422 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
2425 down_read(&rbd_dev->header_rwsem);
2426 size = get_capacity(rbd_dev->disk);
2427 up_read(&rbd_dev->header_rwsem);
2429 return sprintf(buf, "%llu\n", (unsigned long long) size * SECTOR_SIZE);
2433 * Note this shows the features for whatever's mapped, which is not
2434 * necessarily the base image.
2436 static ssize_t rbd_features_show(struct device *dev,
2437 struct device_attribute *attr, char *buf)
2439 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
2441 return sprintf(buf, "0x%016llx\n",
2442 (unsigned long long) rbd_dev->mapping.features);
2445 static ssize_t rbd_major_show(struct device *dev,
2446 struct device_attribute *attr, char *buf)
2448 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
2450 return sprintf(buf, "%d\n", rbd_dev->major);
2453 static ssize_t rbd_client_id_show(struct device *dev,
2454 struct device_attribute *attr, char *buf)
2456 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
2458 return sprintf(buf, "client%lld\n",
2459 ceph_client_id(rbd_dev->rbd_client->client));
2462 static ssize_t rbd_pool_show(struct device *dev,
2463 struct device_attribute *attr, char *buf)
2465 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
2467 return sprintf(buf, "%s\n", rbd_dev->spec->pool_name);
2470 static ssize_t rbd_pool_id_show(struct device *dev,
2471 struct device_attribute *attr, char *buf)
2473 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
2475 return sprintf(buf, "%llu\n",
2476 (unsigned long long) rbd_dev->spec->pool_id);
2479 static ssize_t rbd_name_show(struct device *dev,
2480 struct device_attribute *attr, char *buf)
2482 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
2484 if (rbd_dev->spec->image_name)
2485 return sprintf(buf, "%s\n", rbd_dev->spec->image_name);
2487 return sprintf(buf, "(unknown)\n");
2490 static ssize_t rbd_image_id_show(struct device *dev,
2491 struct device_attribute *attr, char *buf)
2493 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
2495 return sprintf(buf, "%s\n", rbd_dev->spec->image_id);
2499 * Shows the name of the currently-mapped snapshot (or
2500 * RBD_SNAP_HEAD_NAME for the base image).
2502 static ssize_t rbd_snap_show(struct device *dev,
2503 struct device_attribute *attr,
2506 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
2508 return sprintf(buf, "%s\n", rbd_dev->spec->snap_name);
2512 * For an rbd v2 image, shows the pool id, image id, and snapshot id
2513 * for the parent image. If there is no parent, simply shows
2514 * "(no parent image)".
2516 static ssize_t rbd_parent_show(struct device *dev,
2517 struct device_attribute *attr,
2520 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
2521 struct rbd_spec *spec = rbd_dev->parent_spec;
2526 return sprintf(buf, "(no parent image)\n");
2528 count = sprintf(bufp, "pool_id %llu\npool_name %s\n",
2529 (unsigned long long) spec->pool_id, spec->pool_name);
2534 count = sprintf(bufp, "image_id %s\nimage_name %s\n", spec->image_id,
2535 spec->image_name ? spec->image_name : "(unknown)");
2540 count = sprintf(bufp, "snap_id %llu\nsnap_name %s\n",
2541 (unsigned long long) spec->snap_id, spec->snap_name);
2546 count = sprintf(bufp, "overlap %llu\n", rbd_dev->parent_overlap);
2551 return (ssize_t) (bufp - buf);
2554 static ssize_t rbd_image_refresh(struct device *dev,
2555 struct device_attribute *attr,
2559 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
2562 ret = rbd_dev_refresh(rbd_dev, NULL);
2564 return ret < 0 ? ret : size;
2567 static DEVICE_ATTR(size, S_IRUGO, rbd_size_show, NULL);
2568 static DEVICE_ATTR(features, S_IRUGO, rbd_features_show, NULL);
2569 static DEVICE_ATTR(major, S_IRUGO, rbd_major_show, NULL);
2570 static DEVICE_ATTR(client_id, S_IRUGO, rbd_client_id_show, NULL);
2571 static DEVICE_ATTR(pool, S_IRUGO, rbd_pool_show, NULL);
2572 static DEVICE_ATTR(pool_id, S_IRUGO, rbd_pool_id_show, NULL);
2573 static DEVICE_ATTR(name, S_IRUGO, rbd_name_show, NULL);
2574 static DEVICE_ATTR(image_id, S_IRUGO, rbd_image_id_show, NULL);
2575 static DEVICE_ATTR(refresh, S_IWUSR, NULL, rbd_image_refresh);
2576 static DEVICE_ATTR(current_snap, S_IRUGO, rbd_snap_show, NULL);
2577 static DEVICE_ATTR(parent, S_IRUGO, rbd_parent_show, NULL);
2579 static struct attribute *rbd_attrs[] = {
2580 &dev_attr_size.attr,
2581 &dev_attr_features.attr,
2582 &dev_attr_major.attr,
2583 &dev_attr_client_id.attr,
2584 &dev_attr_pool.attr,
2585 &dev_attr_pool_id.attr,
2586 &dev_attr_name.attr,
2587 &dev_attr_image_id.attr,
2588 &dev_attr_current_snap.attr,
2589 &dev_attr_parent.attr,
2590 &dev_attr_refresh.attr,
2594 static struct attribute_group rbd_attr_group = {
2598 static const struct attribute_group *rbd_attr_groups[] = {
2603 static void rbd_sysfs_dev_release(struct device *dev)
2607 static struct device_type rbd_device_type = {
2609 .groups = rbd_attr_groups,
2610 .release = rbd_sysfs_dev_release,
2618 static ssize_t rbd_snap_size_show(struct device *dev,
2619 struct device_attribute *attr,
2622 struct rbd_snap *snap = container_of(dev, struct rbd_snap, dev);
2624 return sprintf(buf, "%llu\n", (unsigned long long)snap->size);
2627 static ssize_t rbd_snap_id_show(struct device *dev,
2628 struct device_attribute *attr,
2631 struct rbd_snap *snap = container_of(dev, struct rbd_snap, dev);
2633 return sprintf(buf, "%llu\n", (unsigned long long)snap->id);
2636 static ssize_t rbd_snap_features_show(struct device *dev,
2637 struct device_attribute *attr,
2640 struct rbd_snap *snap = container_of(dev, struct rbd_snap, dev);
2642 return sprintf(buf, "0x%016llx\n",
2643 (unsigned long long) snap->features);
2646 static DEVICE_ATTR(snap_size, S_IRUGO, rbd_snap_size_show, NULL);
2647 static DEVICE_ATTR(snap_id, S_IRUGO, rbd_snap_id_show, NULL);
2648 static DEVICE_ATTR(snap_features, S_IRUGO, rbd_snap_features_show, NULL);
2650 static struct attribute *rbd_snap_attrs[] = {
2651 &dev_attr_snap_size.attr,
2652 &dev_attr_snap_id.attr,
2653 &dev_attr_snap_features.attr,
2657 static struct attribute_group rbd_snap_attr_group = {
2658 .attrs = rbd_snap_attrs,
2661 static void rbd_snap_dev_release(struct device *dev)
2663 struct rbd_snap *snap = container_of(dev, struct rbd_snap, dev);
2668 static const struct attribute_group *rbd_snap_attr_groups[] = {
2669 &rbd_snap_attr_group,
2673 static struct device_type rbd_snap_device_type = {
2674 .groups = rbd_snap_attr_groups,
2675 .release = rbd_snap_dev_release,
2678 static struct rbd_spec *rbd_spec_get(struct rbd_spec *spec)
2680 kref_get(&spec->kref);
2685 static void rbd_spec_free(struct kref *kref);
2686 static void rbd_spec_put(struct rbd_spec *spec)
2689 kref_put(&spec->kref, rbd_spec_free);
2692 static struct rbd_spec *rbd_spec_alloc(void)
2694 struct rbd_spec *spec;
2696 spec = kzalloc(sizeof (*spec), GFP_KERNEL);
2699 kref_init(&spec->kref);
2701 rbd_spec_put(rbd_spec_get(spec)); /* TEMPORARY */
2706 static void rbd_spec_free(struct kref *kref)
2708 struct rbd_spec *spec = container_of(kref, struct rbd_spec, kref);
2710 kfree(spec->pool_name);
2711 kfree(spec->image_id);
2712 kfree(spec->image_name);
2713 kfree(spec->snap_name);
2717 static struct rbd_device *rbd_dev_create(struct rbd_client *rbdc,
2718 struct rbd_spec *spec)
2720 struct rbd_device *rbd_dev;
2722 rbd_dev = kzalloc(sizeof (*rbd_dev), GFP_KERNEL);
2726 spin_lock_init(&rbd_dev->lock);
2728 INIT_LIST_HEAD(&rbd_dev->node);
2729 INIT_LIST_HEAD(&rbd_dev->snaps);
2730 init_rwsem(&rbd_dev->header_rwsem);
2732 rbd_dev->spec = spec;
2733 rbd_dev->rbd_client = rbdc;
2735 /* Initialize the layout used for all rbd requests */
2737 rbd_dev->layout.fl_stripe_unit = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER);
2738 rbd_dev->layout.fl_stripe_count = cpu_to_le32(1);
2739 rbd_dev->layout.fl_object_size = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER);
2740 rbd_dev->layout.fl_pg_pool = cpu_to_le32((u32) spec->pool_id);
2745 static void rbd_dev_destroy(struct rbd_device *rbd_dev)
2747 rbd_spec_put(rbd_dev->parent_spec);
2748 kfree(rbd_dev->header_name);
2749 rbd_put_client(rbd_dev->rbd_client);
2750 rbd_spec_put(rbd_dev->spec);
2754 static bool rbd_snap_registered(struct rbd_snap *snap)
2756 bool ret = snap->dev.type == &rbd_snap_device_type;
2757 bool reg = device_is_registered(&snap->dev);
2759 rbd_assert(!ret ^ reg);
2764 static void rbd_remove_snap_dev(struct rbd_snap *snap)
2766 list_del(&snap->node);
2767 if (device_is_registered(&snap->dev))
2768 device_unregister(&snap->dev);
2771 static int rbd_register_snap_dev(struct rbd_snap *snap,
2772 struct device *parent)
2774 struct device *dev = &snap->dev;
2777 dev->type = &rbd_snap_device_type;
2778 dev->parent = parent;
2779 dev->release = rbd_snap_dev_release;
2780 dev_set_name(dev, "%s%s", RBD_SNAP_DEV_NAME_PREFIX, snap->name);
2781 dout("%s: registering device for snapshot %s\n", __func__, snap->name);
2783 ret = device_register(dev);
2788 static struct rbd_snap *__rbd_add_snap_dev(struct rbd_device *rbd_dev,
2789 const char *snap_name,
2790 u64 snap_id, u64 snap_size,
2793 struct rbd_snap *snap;
2796 snap = kzalloc(sizeof (*snap), GFP_KERNEL);
2798 return ERR_PTR(-ENOMEM);
2801 snap->name = kstrdup(snap_name, GFP_KERNEL);
2806 snap->size = snap_size;
2807 snap->features = snap_features;
2815 return ERR_PTR(ret);
2818 static char *rbd_dev_v1_snap_info(struct rbd_device *rbd_dev, u32 which,
2819 u64 *snap_size, u64 *snap_features)
2823 rbd_assert(which < rbd_dev->header.snapc->num_snaps);
2825 *snap_size = rbd_dev->header.snap_sizes[which];
2826 *snap_features = 0; /* No features for v1 */
2828 /* Skip over names until we find the one we are looking for */
2830 snap_name = rbd_dev->header.snap_names;
2832 snap_name += strlen(snap_name) + 1;
2838 * Get the size and object order for an image snapshot, or if
2839 * snap_id is CEPH_NOSNAP, gets this information for the base
2842 static int _rbd_dev_v2_snap_size(struct rbd_device *rbd_dev, u64 snap_id,
2843 u8 *order, u64 *snap_size)
2845 __le64 snapid = cpu_to_le64(snap_id);
2850 } __attribute__ ((packed)) size_buf = { 0 };
2852 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
2854 (char *) &snapid, sizeof (snapid),
2855 (char *) &size_buf, sizeof (size_buf), NULL);
2856 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
2860 *order = size_buf.order;
2861 *snap_size = le64_to_cpu(size_buf.size);
2863 dout(" snap_id 0x%016llx order = %u, snap_size = %llu\n",
2864 (unsigned long long) snap_id, (unsigned int) *order,
2865 (unsigned long long) *snap_size);
2870 static int rbd_dev_v2_image_size(struct rbd_device *rbd_dev)
2872 return _rbd_dev_v2_snap_size(rbd_dev, CEPH_NOSNAP,
2873 &rbd_dev->header.obj_order,
2874 &rbd_dev->header.image_size);
2877 static int rbd_dev_v2_object_prefix(struct rbd_device *rbd_dev)
2883 reply_buf = kzalloc(RBD_OBJ_PREFIX_LEN_MAX, GFP_KERNEL);
2887 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
2888 "rbd", "get_object_prefix",
2890 reply_buf, RBD_OBJ_PREFIX_LEN_MAX, NULL);
2891 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
2896 rbd_dev->header.object_prefix = ceph_extract_encoded_string(&p,
2897 p + RBD_OBJ_PREFIX_LEN_MAX,
2900 if (IS_ERR(rbd_dev->header.object_prefix)) {
2901 ret = PTR_ERR(rbd_dev->header.object_prefix);
2902 rbd_dev->header.object_prefix = NULL;
2904 dout(" object_prefix = %s\n", rbd_dev->header.object_prefix);
2913 static int _rbd_dev_v2_snap_features(struct rbd_device *rbd_dev, u64 snap_id,
2916 __le64 snapid = cpu_to_le64(snap_id);
2920 } features_buf = { 0 };
2924 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
2925 "rbd", "get_features",
2926 (char *) &snapid, sizeof (snapid),
2927 (char *) &features_buf, sizeof (features_buf),
2929 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
2933 incompat = le64_to_cpu(features_buf.incompat);
2934 if (incompat & ~RBD_FEATURES_ALL)
2937 *snap_features = le64_to_cpu(features_buf.features);
2939 dout(" snap_id 0x%016llx features = 0x%016llx incompat = 0x%016llx\n",
2940 (unsigned long long) snap_id,
2941 (unsigned long long) *snap_features,
2942 (unsigned long long) le64_to_cpu(features_buf.incompat));
2947 static int rbd_dev_v2_features(struct rbd_device *rbd_dev)
2949 return _rbd_dev_v2_snap_features(rbd_dev, CEPH_NOSNAP,
2950 &rbd_dev->header.features);
2953 static int rbd_dev_v2_parent_info(struct rbd_device *rbd_dev)
2955 struct rbd_spec *parent_spec;
2957 void *reply_buf = NULL;
2965 parent_spec = rbd_spec_alloc();
2969 size = sizeof (__le64) + /* pool_id */
2970 sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX + /* image_id */
2971 sizeof (__le64) + /* snap_id */
2972 sizeof (__le64); /* overlap */
2973 reply_buf = kmalloc(size, GFP_KERNEL);
2979 snapid = cpu_to_le64(CEPH_NOSNAP);
2980 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
2981 "rbd", "get_parent",
2982 (char *) &snapid, sizeof (snapid),
2983 (char *) reply_buf, size, NULL);
2984 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
2990 end = (char *) reply_buf + size;
2991 ceph_decode_64_safe(&p, end, parent_spec->pool_id, out_err);
2992 if (parent_spec->pool_id == CEPH_NOPOOL)
2993 goto out; /* No parent? No problem. */
2995 /* The ceph file layout needs to fit pool id in 32 bits */
2998 if (WARN_ON(parent_spec->pool_id > (u64) U32_MAX))
3001 image_id = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL);
3002 if (IS_ERR(image_id)) {
3003 ret = PTR_ERR(image_id);
3006 parent_spec->image_id = image_id;
3007 ceph_decode_64_safe(&p, end, parent_spec->snap_id, out_err);
3008 ceph_decode_64_safe(&p, end, overlap, out_err);
3010 rbd_dev->parent_overlap = overlap;
3011 rbd_dev->parent_spec = parent_spec;
3012 parent_spec = NULL; /* rbd_dev now owns this */
3017 rbd_spec_put(parent_spec);
3022 static char *rbd_dev_image_name(struct rbd_device *rbd_dev)
3024 size_t image_id_size;
3029 void *reply_buf = NULL;
3031 char *image_name = NULL;
3034 rbd_assert(!rbd_dev->spec->image_name);
3036 len = strlen(rbd_dev->spec->image_id);
3037 image_id_size = sizeof (__le32) + len;
3038 image_id = kmalloc(image_id_size, GFP_KERNEL);
3043 end = (char *) image_id + image_id_size;
3044 ceph_encode_string(&p, end, rbd_dev->spec->image_id, (u32) len);
3046 size = sizeof (__le32) + RBD_IMAGE_NAME_LEN_MAX;
3047 reply_buf = kmalloc(size, GFP_KERNEL);
3051 ret = rbd_obj_method_sync(rbd_dev, RBD_DIRECTORY,
3052 "rbd", "dir_get_name",
3053 image_id, image_id_size,
3054 (char *) reply_buf, size, NULL);
3058 end = (char *) reply_buf + size;
3059 image_name = ceph_extract_encoded_string(&p, end, &len, GFP_KERNEL);
3060 if (IS_ERR(image_name))
3063 dout("%s: name is %s len is %zd\n", __func__, image_name, len);
3072 * When a parent image gets probed, we only have the pool, image,
3073 * and snapshot ids but not the names of any of them. This call
3074 * is made later to fill in those names. It has to be done after
3075 * rbd_dev_snaps_update() has completed because some of the
3076 * information (in particular, snapshot name) is not available
3079 static int rbd_dev_probe_update_spec(struct rbd_device *rbd_dev)
3081 struct ceph_osd_client *osdc;
3083 void *reply_buf = NULL;
3086 if (rbd_dev->spec->pool_name)
3087 return 0; /* Already have the names */
3089 /* Look up the pool name */
3091 osdc = &rbd_dev->rbd_client->client->osdc;
3092 name = ceph_pg_pool_name_by_id(osdc->osdmap, rbd_dev->spec->pool_id);
3094 rbd_warn(rbd_dev, "there is no pool with id %llu",
3095 rbd_dev->spec->pool_id); /* Really a BUG() */
3099 rbd_dev->spec->pool_name = kstrdup(name, GFP_KERNEL);
3100 if (!rbd_dev->spec->pool_name)
3103 /* Fetch the image name; tolerate failure here */
3105 name = rbd_dev_image_name(rbd_dev);
3107 rbd_dev->spec->image_name = (char *) name;
3109 rbd_warn(rbd_dev, "unable to get image name");
3111 /* Look up the snapshot name. */
3113 name = rbd_snap_name(rbd_dev, rbd_dev->spec->snap_id);
3115 rbd_warn(rbd_dev, "no snapshot with id %llu",
3116 rbd_dev->spec->snap_id); /* Really a BUG() */
3120 rbd_dev->spec->snap_name = kstrdup(name, GFP_KERNEL);
3121 if(!rbd_dev->spec->snap_name)
3127 kfree(rbd_dev->spec->pool_name);
3128 rbd_dev->spec->pool_name = NULL;
3133 static int rbd_dev_v2_snap_context(struct rbd_device *rbd_dev, u64 *ver)
3142 struct ceph_snap_context *snapc;
3146 * We'll need room for the seq value (maximum snapshot id),
3147 * snapshot count, and array of that many snapshot ids.
3148 * For now we have a fixed upper limit on the number we're
3149 * prepared to receive.
3151 size = sizeof (__le64) + sizeof (__le32) +
3152 RBD_MAX_SNAP_COUNT * sizeof (__le64);
3153 reply_buf = kzalloc(size, GFP_KERNEL);
3157 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
3158 "rbd", "get_snapcontext",
3160 reply_buf, size, ver);
3161 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
3167 end = (char *) reply_buf + size;
3168 ceph_decode_64_safe(&p, end, seq, out);
3169 ceph_decode_32_safe(&p, end, snap_count, out);
3172 * Make sure the reported number of snapshot ids wouldn't go
3173 * beyond the end of our buffer. But before checking that,
3174 * make sure the computed size of the snapshot context we
3175 * allocate is representable in a size_t.
3177 if (snap_count > (SIZE_MAX - sizeof (struct ceph_snap_context))
3182 if (!ceph_has_room(&p, end, snap_count * sizeof (__le64)))
3185 size = sizeof (struct ceph_snap_context) +
3186 snap_count * sizeof (snapc->snaps[0]);
3187 snapc = kmalloc(size, GFP_KERNEL);
3193 atomic_set(&snapc->nref, 1);
3195 snapc->num_snaps = snap_count;
3196 for (i = 0; i < snap_count; i++)
3197 snapc->snaps[i] = ceph_decode_64(&p);
3199 rbd_dev->header.snapc = snapc;
3201 dout(" snap context seq = %llu, snap_count = %u\n",
3202 (unsigned long long) seq, (unsigned int) snap_count);
3210 static char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev, u32 which)
3220 size = sizeof (__le32) + RBD_MAX_SNAP_NAME_LEN;
3221 reply_buf = kmalloc(size, GFP_KERNEL);
3223 return ERR_PTR(-ENOMEM);
3225 snap_id = cpu_to_le64(rbd_dev->header.snapc->snaps[which]);
3226 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
3227 "rbd", "get_snapshot_name",
3228 (char *) &snap_id, sizeof (snap_id),
3229 reply_buf, size, NULL);
3230 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
3235 end = (char *) reply_buf + size;
3236 snap_name = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL);
3237 if (IS_ERR(snap_name)) {
3238 ret = PTR_ERR(snap_name);
3241 dout(" snap_id 0x%016llx snap_name = %s\n",
3242 (unsigned long long) le64_to_cpu(snap_id), snap_name);
3250 return ERR_PTR(ret);
3253 static char *rbd_dev_v2_snap_info(struct rbd_device *rbd_dev, u32 which,
3254 u64 *snap_size, u64 *snap_features)
3260 snap_id = rbd_dev->header.snapc->snaps[which];
3261 ret = _rbd_dev_v2_snap_size(rbd_dev, snap_id, &order, snap_size);
3263 return ERR_PTR(ret);
3264 ret = _rbd_dev_v2_snap_features(rbd_dev, snap_id, snap_features);
3266 return ERR_PTR(ret);
3268 return rbd_dev_v2_snap_name(rbd_dev, which);
3271 static char *rbd_dev_snap_info(struct rbd_device *rbd_dev, u32 which,
3272 u64 *snap_size, u64 *snap_features)
3274 if (rbd_dev->image_format == 1)
3275 return rbd_dev_v1_snap_info(rbd_dev, which,
3276 snap_size, snap_features);
3277 if (rbd_dev->image_format == 2)
3278 return rbd_dev_v2_snap_info(rbd_dev, which,
3279 snap_size, snap_features);
3280 return ERR_PTR(-EINVAL);
3283 static int rbd_dev_v2_refresh(struct rbd_device *rbd_dev, u64 *hver)
3288 down_write(&rbd_dev->header_rwsem);
3290 /* Grab old order first, to see if it changes */
3292 obj_order = rbd_dev->header.obj_order,
3293 ret = rbd_dev_v2_image_size(rbd_dev);
3296 if (rbd_dev->header.obj_order != obj_order) {
3300 rbd_update_mapping_size(rbd_dev);
3302 ret = rbd_dev_v2_snap_context(rbd_dev, hver);
3303 dout("rbd_dev_v2_snap_context returned %d\n", ret);
3306 ret = rbd_dev_snaps_update(rbd_dev);
3307 dout("rbd_dev_snaps_update returned %d\n", ret);
3310 ret = rbd_dev_snaps_register(rbd_dev);
3311 dout("rbd_dev_snaps_register returned %d\n", ret);
3313 up_write(&rbd_dev->header_rwsem);
3319 * Scan the rbd device's current snapshot list and compare it to the
3320 * newly-received snapshot context. Remove any existing snapshots
3321 * not present in the new snapshot context. Add a new snapshot for
3322 * any snaphots in the snapshot context not in the current list.
3323 * And verify there are no changes to snapshots we already know
3326 * Assumes the snapshots in the snapshot context are sorted by
3327 * snapshot id, highest id first. (Snapshots in the rbd_dev's list
3328 * are also maintained in that order.)
3330 static int rbd_dev_snaps_update(struct rbd_device *rbd_dev)
3332 struct ceph_snap_context *snapc = rbd_dev->header.snapc;
3333 const u32 snap_count = snapc->num_snaps;
3334 struct list_head *head = &rbd_dev->snaps;
3335 struct list_head *links = head->next;
3338 dout("%s: snap count is %u\n", __func__, (unsigned int) snap_count);
3339 while (index < snap_count || links != head) {
3341 struct rbd_snap *snap;
3344 u64 snap_features = 0;
3346 snap_id = index < snap_count ? snapc->snaps[index]
3348 snap = links != head ? list_entry(links, struct rbd_snap, node)
3350 rbd_assert(!snap || snap->id != CEPH_NOSNAP);
3352 if (snap_id == CEPH_NOSNAP || (snap && snap->id > snap_id)) {
3353 struct list_head *next = links->next;
3356 * A previously-existing snapshot is not in
3357 * the new snap context.
3359 * If the now missing snapshot is the one the
3360 * image is mapped to, clear its exists flag
3361 * so we can avoid sending any more requests
3364 if (rbd_dev->spec->snap_id == snap->id)
3365 clear_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags);
3366 rbd_remove_snap_dev(snap);
3367 dout("%ssnap id %llu has been removed\n",
3368 rbd_dev->spec->snap_id == snap->id ?
3370 (unsigned long long) snap->id);
3372 /* Done with this list entry; advance */
3378 snap_name = rbd_dev_snap_info(rbd_dev, index,
3379 &snap_size, &snap_features);
3380 if (IS_ERR(snap_name))
3381 return PTR_ERR(snap_name);
3383 dout("entry %u: snap_id = %llu\n", (unsigned int) snap_count,
3384 (unsigned long long) snap_id);
3385 if (!snap || (snap_id != CEPH_NOSNAP && snap->id < snap_id)) {
3386 struct rbd_snap *new_snap;
3388 /* We haven't seen this snapshot before */
3390 new_snap = __rbd_add_snap_dev(rbd_dev, snap_name,
3391 snap_id, snap_size, snap_features);
3392 if (IS_ERR(new_snap)) {
3393 int err = PTR_ERR(new_snap);
3395 dout(" failed to add dev, error %d\n", err);
3400 /* New goes before existing, or at end of list */
3402 dout(" added dev%s\n", snap ? "" : " at end\n");
3404 list_add_tail(&new_snap->node, &snap->node);
3406 list_add_tail(&new_snap->node, head);
3408 /* Already have this one */
3410 dout(" already present\n");
3412 rbd_assert(snap->size == snap_size);
3413 rbd_assert(!strcmp(snap->name, snap_name));
3414 rbd_assert(snap->features == snap_features);
3416 /* Done with this list entry; advance */
3418 links = links->next;
3421 /* Advance to the next entry in the snapshot context */
3425 dout("%s: done\n", __func__);
3431 * Scan the list of snapshots and register the devices for any that
3432 * have not already been registered.
3434 static int rbd_dev_snaps_register(struct rbd_device *rbd_dev)
3436 struct rbd_snap *snap;
3439 dout("%s:\n", __func__);
3440 if (WARN_ON(!device_is_registered(&rbd_dev->dev)))
3443 list_for_each_entry(snap, &rbd_dev->snaps, node) {
3444 if (!rbd_snap_registered(snap)) {
3445 ret = rbd_register_snap_dev(snap, &rbd_dev->dev);
3450 dout("%s: returning %d\n", __func__, ret);
3455 static int rbd_bus_add_dev(struct rbd_device *rbd_dev)
3460 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
3462 dev = &rbd_dev->dev;
3463 dev->bus = &rbd_bus_type;
3464 dev->type = &rbd_device_type;
3465 dev->parent = &rbd_root_dev;
3466 dev->release = rbd_dev_release;
3467 dev_set_name(dev, "%d", rbd_dev->dev_id);
3468 ret = device_register(dev);
3470 mutex_unlock(&ctl_mutex);
3475 static void rbd_bus_del_dev(struct rbd_device *rbd_dev)
3477 device_unregister(&rbd_dev->dev);
3480 static atomic64_t rbd_dev_id_max = ATOMIC64_INIT(0);
3483 * Get a unique rbd identifier for the given new rbd_dev, and add
3484 * the rbd_dev to the global list. The minimum rbd id is 1.
3486 static void rbd_dev_id_get(struct rbd_device *rbd_dev)
3488 rbd_dev->dev_id = atomic64_inc_return(&rbd_dev_id_max);
3490 spin_lock(&rbd_dev_list_lock);
3491 list_add_tail(&rbd_dev->node, &rbd_dev_list);
3492 spin_unlock(&rbd_dev_list_lock);
3493 dout("rbd_dev %p given dev id %llu\n", rbd_dev,
3494 (unsigned long long) rbd_dev->dev_id);
3498 * Remove an rbd_dev from the global list, and record that its
3499 * identifier is no longer in use.
3501 static void rbd_dev_id_put(struct rbd_device *rbd_dev)
3503 struct list_head *tmp;
3504 int rbd_id = rbd_dev->dev_id;
3507 rbd_assert(rbd_id > 0);
3509 dout("rbd_dev %p released dev id %llu\n", rbd_dev,
3510 (unsigned long long) rbd_dev->dev_id);
3511 spin_lock(&rbd_dev_list_lock);
3512 list_del_init(&rbd_dev->node);
3515 * If the id being "put" is not the current maximum, there
3516 * is nothing special we need to do.
3518 if (rbd_id != atomic64_read(&rbd_dev_id_max)) {
3519 spin_unlock(&rbd_dev_list_lock);
3524 * We need to update the current maximum id. Search the
3525 * list to find out what it is. We're more likely to find
3526 * the maximum at the end, so search the list backward.
3529 list_for_each_prev(tmp, &rbd_dev_list) {
3530 struct rbd_device *rbd_dev;
3532 rbd_dev = list_entry(tmp, struct rbd_device, node);
3533 if (rbd_dev->dev_id > max_id)
3534 max_id = rbd_dev->dev_id;
3536 spin_unlock(&rbd_dev_list_lock);
3539 * The max id could have been updated by rbd_dev_id_get(), in
3540 * which case it now accurately reflects the new maximum.
3541 * Be careful not to overwrite the maximum value in that
3544 atomic64_cmpxchg(&rbd_dev_id_max, rbd_id, max_id);
3545 dout(" max dev id has been reset\n");
3549 * Skips over white space at *buf, and updates *buf to point to the
3550 * first found non-space character (if any). Returns the length of
3551 * the token (string of non-white space characters) found. Note
3552 * that *buf must be terminated with '\0'.
3554 static inline size_t next_token(const char **buf)
3557 * These are the characters that produce nonzero for
3558 * isspace() in the "C" and "POSIX" locales.
3560 const char *spaces = " \f\n\r\t\v";
3562 *buf += strspn(*buf, spaces); /* Find start of token */
3564 return strcspn(*buf, spaces); /* Return token length */
3568 * Finds the next token in *buf, and if the provided token buffer is
3569 * big enough, copies the found token into it. The result, if
3570 * copied, is guaranteed to be terminated with '\0'. Note that *buf
3571 * must be terminated with '\0' on entry.
3573 * Returns the length of the token found (not including the '\0').
3574 * Return value will be 0 if no token is found, and it will be >=
3575 * token_size if the token would not fit.
3577 * The *buf pointer will be updated to point beyond the end of the
3578 * found token. Note that this occurs even if the token buffer is
3579 * too small to hold it.
3581 static inline size_t copy_token(const char **buf,
3587 len = next_token(buf);
3588 if (len < token_size) {
3589 memcpy(token, *buf, len);
3590 *(token + len) = '\0';
3598 * Finds the next token in *buf, dynamically allocates a buffer big
3599 * enough to hold a copy of it, and copies the token into the new
3600 * buffer. The copy is guaranteed to be terminated with '\0'. Note
3601 * that a duplicate buffer is created even for a zero-length token.
3603 * Returns a pointer to the newly-allocated duplicate, or a null
3604 * pointer if memory for the duplicate was not available. If
3605 * the lenp argument is a non-null pointer, the length of the token
3606 * (not including the '\0') is returned in *lenp.
3608 * If successful, the *buf pointer will be updated to point beyond
3609 * the end of the found token.
3611 * Note: uses GFP_KERNEL for allocation.
3613 static inline char *dup_token(const char **buf, size_t *lenp)
3618 len = next_token(buf);
3619 dup = kmemdup(*buf, len + 1, GFP_KERNEL);
3622 *(dup + len) = '\0';
3632 * Parse the options provided for an "rbd add" (i.e., rbd image
3633 * mapping) request. These arrive via a write to /sys/bus/rbd/add,
3634 * and the data written is passed here via a NUL-terminated buffer.
3635 * Returns 0 if successful or an error code otherwise.
3637 * The information extracted from these options is recorded in
3638 * the other parameters which return dynamically-allocated
3641 * The address of a pointer that will refer to a ceph options
3642 * structure. Caller must release the returned pointer using
3643 * ceph_destroy_options() when it is no longer needed.
3645 * Address of an rbd options pointer. Fully initialized by
3646 * this function; caller must release with kfree().
3648 * Address of an rbd image specification pointer. Fully
3649 * initialized by this function based on parsed options.
3650 * Caller must release with rbd_spec_put().
3652 * The options passed take this form:
3653 * <mon_addrs> <options> <pool_name> <image_name> [<snap_id>]
3656 * A comma-separated list of one or more monitor addresses.
3657 * A monitor address is an ip address, optionally followed
3658 * by a port number (separated by a colon).
3659 * I.e.: ip1[:port1][,ip2[:port2]...]
3661 * A comma-separated list of ceph and/or rbd options.
3663 * The name of the rados pool containing the rbd image.
3665 * The name of the image in that pool to map.
3667 * An optional snapshot id. If provided, the mapping will
3668 * present data from the image at the time that snapshot was
3669 * created. The image head is used if no snapshot id is
3670 * provided. Snapshot mappings are always read-only.
3672 static int rbd_add_parse_args(const char *buf,
3673 struct ceph_options **ceph_opts,
3674 struct rbd_options **opts,
3675 struct rbd_spec **rbd_spec)
3679 const char *mon_addrs;
3680 size_t mon_addrs_size;
3681 struct rbd_spec *spec = NULL;
3682 struct rbd_options *rbd_opts = NULL;
3683 struct ceph_options *copts;
3686 /* The first four tokens are required */
3688 len = next_token(&buf);
3690 rbd_warn(NULL, "no monitor address(es) provided");
3694 mon_addrs_size = len + 1;
3698 options = dup_token(&buf, NULL);
3702 rbd_warn(NULL, "no options provided");
3706 spec = rbd_spec_alloc();
3710 spec->pool_name = dup_token(&buf, NULL);
3711 if (!spec->pool_name)
3713 if (!*spec->pool_name) {
3714 rbd_warn(NULL, "no pool name provided");
3718 spec->image_name = dup_token(&buf, NULL);
3719 if (!spec->image_name)
3721 if (!*spec->image_name) {
3722 rbd_warn(NULL, "no image name provided");
3727 * Snapshot name is optional; default is to use "-"
3728 * (indicating the head/no snapshot).
3730 len = next_token(&buf);
3732 buf = RBD_SNAP_HEAD_NAME; /* No snapshot supplied */
3733 len = sizeof (RBD_SNAP_HEAD_NAME) - 1;
3734 } else if (len > RBD_MAX_SNAP_NAME_LEN) {
3735 ret = -ENAMETOOLONG;
3738 spec->snap_name = kmemdup(buf, len + 1, GFP_KERNEL);
3739 if (!spec->snap_name)
3741 *(spec->snap_name + len) = '\0';
3743 /* Initialize all rbd options to the defaults */
3745 rbd_opts = kzalloc(sizeof (*rbd_opts), GFP_KERNEL);
3749 rbd_opts->read_only = RBD_READ_ONLY_DEFAULT;
3751 copts = ceph_parse_options(options, mon_addrs,
3752 mon_addrs + mon_addrs_size - 1,
3753 parse_rbd_opts_token, rbd_opts);
3754 if (IS_ERR(copts)) {
3755 ret = PTR_ERR(copts);
3776 * An rbd format 2 image has a unique identifier, distinct from the
3777 * name given to it by the user. Internally, that identifier is
3778 * what's used to specify the names of objects related to the image.
3780 * A special "rbd id" object is used to map an rbd image name to its
3781 * id. If that object doesn't exist, then there is no v2 rbd image
3782 * with the supplied name.
3784 * This function will record the given rbd_dev's image_id field if
3785 * it can be determined, and in that case will return 0. If any
3786 * errors occur a negative errno will be returned and the rbd_dev's
3787 * image_id field will be unchanged (and should be NULL).
3789 static int rbd_dev_image_id(struct rbd_device *rbd_dev)
3798 * When probing a parent image, the image id is already
3799 * known (and the image name likely is not). There's no
3800 * need to fetch the image id again in this case.
3802 if (rbd_dev->spec->image_id)
3806 * First, see if the format 2 image id file exists, and if
3807 * so, get the image's persistent id from it.
3809 size = sizeof (RBD_ID_PREFIX) + strlen(rbd_dev->spec->image_name);
3810 object_name = kmalloc(size, GFP_NOIO);
3813 sprintf(object_name, "%s%s", RBD_ID_PREFIX, rbd_dev->spec->image_name);
3814 dout("rbd id object name is %s\n", object_name);
3816 /* Response will be an encoded string, which includes a length */
3818 size = sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX;
3819 response = kzalloc(size, GFP_NOIO);
3825 ret = rbd_obj_method_sync(rbd_dev, object_name,
3828 response, RBD_IMAGE_ID_LEN_MAX, NULL);
3829 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
3834 rbd_dev->spec->image_id = ceph_extract_encoded_string(&p,
3835 p + RBD_IMAGE_ID_LEN_MAX,
3837 if (IS_ERR(rbd_dev->spec->image_id)) {
3838 ret = PTR_ERR(rbd_dev->spec->image_id);
3839 rbd_dev->spec->image_id = NULL;
3841 dout("image_id is %s\n", rbd_dev->spec->image_id);
3850 static int rbd_dev_v1_probe(struct rbd_device *rbd_dev)
3855 /* Version 1 images have no id; empty string is used */
3857 rbd_dev->spec->image_id = kstrdup("", GFP_KERNEL);
3858 if (!rbd_dev->spec->image_id)
3861 /* Record the header object name for this rbd image. */
3863 size = strlen(rbd_dev->spec->image_name) + sizeof (RBD_SUFFIX);
3864 rbd_dev->header_name = kmalloc(size, GFP_KERNEL);
3865 if (!rbd_dev->header_name) {
3869 sprintf(rbd_dev->header_name, "%s%s",
3870 rbd_dev->spec->image_name, RBD_SUFFIX);
3872 /* Populate rbd image metadata */
3874 ret = rbd_read_header(rbd_dev, &rbd_dev->header);
3878 /* Version 1 images have no parent (no layering) */
3880 rbd_dev->parent_spec = NULL;
3881 rbd_dev->parent_overlap = 0;
3883 rbd_dev->image_format = 1;
3885 dout("discovered version 1 image, header name is %s\n",
3886 rbd_dev->header_name);
3891 kfree(rbd_dev->header_name);
3892 rbd_dev->header_name = NULL;
3893 kfree(rbd_dev->spec->image_id);
3894 rbd_dev->spec->image_id = NULL;
3899 static int rbd_dev_v2_probe(struct rbd_device *rbd_dev)
3906 * Image id was filled in by the caller. Record the header
3907 * object name for this rbd image.
3909 size = sizeof (RBD_HEADER_PREFIX) + strlen(rbd_dev->spec->image_id);
3910 rbd_dev->header_name = kmalloc(size, GFP_KERNEL);
3911 if (!rbd_dev->header_name)
3913 sprintf(rbd_dev->header_name, "%s%s",
3914 RBD_HEADER_PREFIX, rbd_dev->spec->image_id);
3916 /* Get the size and object order for the image */
3918 ret = rbd_dev_v2_image_size(rbd_dev);
3922 /* Get the object prefix (a.k.a. block_name) for the image */
3924 ret = rbd_dev_v2_object_prefix(rbd_dev);
3928 /* Get the and check features for the image */
3930 ret = rbd_dev_v2_features(rbd_dev);
3934 /* If the image supports layering, get the parent info */
3936 if (rbd_dev->header.features & RBD_FEATURE_LAYERING) {
3937 ret = rbd_dev_v2_parent_info(rbd_dev);
3942 /* crypto and compression type aren't (yet) supported for v2 images */
3944 rbd_dev->header.crypt_type = 0;
3945 rbd_dev->header.comp_type = 0;
3947 /* Get the snapshot context, plus the header version */
3949 ret = rbd_dev_v2_snap_context(rbd_dev, &ver);
3952 rbd_dev->header.obj_version = ver;
3954 rbd_dev->image_format = 2;
3956 dout("discovered version 2 image, header name is %s\n",
3957 rbd_dev->header_name);
3961 rbd_dev->parent_overlap = 0;
3962 rbd_spec_put(rbd_dev->parent_spec);
3963 rbd_dev->parent_spec = NULL;
3964 kfree(rbd_dev->header_name);
3965 rbd_dev->header_name = NULL;
3966 kfree(rbd_dev->header.object_prefix);
3967 rbd_dev->header.object_prefix = NULL;
3972 static int rbd_dev_probe_finish(struct rbd_device *rbd_dev)
3976 /* no need to lock here, as rbd_dev is not registered yet */
3977 ret = rbd_dev_snaps_update(rbd_dev);
3981 ret = rbd_dev_probe_update_spec(rbd_dev);
3985 ret = rbd_dev_set_mapping(rbd_dev);
3989 /* generate unique id: find highest unique id, add one */
3990 rbd_dev_id_get(rbd_dev);
3992 /* Fill in the device name, now that we have its id. */
3993 BUILD_BUG_ON(DEV_NAME_LEN
3994 < sizeof (RBD_DRV_NAME) + MAX_INT_FORMAT_WIDTH);
3995 sprintf(rbd_dev->name, "%s%d", RBD_DRV_NAME, rbd_dev->dev_id);
3997 /* Get our block major device number. */
3999 ret = register_blkdev(0, rbd_dev->name);
4002 rbd_dev->major = ret;
4004 /* Set up the blkdev mapping. */
4006 ret = rbd_init_disk(rbd_dev);
4008 goto err_out_blkdev;
4010 ret = rbd_bus_add_dev(rbd_dev);
4015 * At this point cleanup in the event of an error is the job
4016 * of the sysfs code (initiated by rbd_bus_del_dev()).
4018 down_write(&rbd_dev->header_rwsem);
4019 ret = rbd_dev_snaps_register(rbd_dev);
4020 up_write(&rbd_dev->header_rwsem);
4024 ret = rbd_dev_header_watch_sync(rbd_dev, 1);
4028 /* Everything's ready. Announce the disk to the world. */
4030 add_disk(rbd_dev->disk);
4032 pr_info("%s: added with size 0x%llx\n", rbd_dev->disk->disk_name,
4033 (unsigned long long) rbd_dev->mapping.size);
4037 /* this will also clean up rest of rbd_dev stuff */
4039 rbd_bus_del_dev(rbd_dev);
4043 rbd_free_disk(rbd_dev);
4045 unregister_blkdev(rbd_dev->major, rbd_dev->name);
4047 rbd_dev_id_put(rbd_dev);
4049 rbd_remove_all_snaps(rbd_dev);
4055 * Probe for the existence of the header object for the given rbd
4056 * device. For format 2 images this includes determining the image
4059 static int rbd_dev_probe(struct rbd_device *rbd_dev)
4064 * Get the id from the image id object. If it's not a
4065 * format 2 image, we'll get ENOENT back, and we'll assume
4066 * it's a format 1 image.
4068 ret = rbd_dev_image_id(rbd_dev);
4070 ret = rbd_dev_v1_probe(rbd_dev);
4072 ret = rbd_dev_v2_probe(rbd_dev);
4074 dout("probe failed, returning %d\n", ret);
4079 ret = rbd_dev_probe_finish(rbd_dev);
4081 rbd_header_free(&rbd_dev->header);
4086 static ssize_t rbd_add(struct bus_type *bus,
4090 struct rbd_device *rbd_dev = NULL;
4091 struct ceph_options *ceph_opts = NULL;
4092 struct rbd_options *rbd_opts = NULL;
4093 struct rbd_spec *spec = NULL;
4094 struct rbd_client *rbdc;
4095 struct ceph_osd_client *osdc;
4098 if (!try_module_get(THIS_MODULE))
4101 /* parse add command */
4102 rc = rbd_add_parse_args(buf, &ceph_opts, &rbd_opts, &spec);
4104 goto err_out_module;
4106 rbdc = rbd_get_client(ceph_opts);
4111 ceph_opts = NULL; /* rbd_dev client now owns this */
4114 osdc = &rbdc->client->osdc;
4115 rc = ceph_pg_poolid_by_name(osdc->osdmap, spec->pool_name);
4117 goto err_out_client;
4118 spec->pool_id = (u64) rc;
4120 /* The ceph file layout needs to fit pool id in 32 bits */
4122 if (WARN_ON(spec->pool_id > (u64) U32_MAX)) {
4124 goto err_out_client;
4127 rbd_dev = rbd_dev_create(rbdc, spec);
4129 goto err_out_client;
4130 rbdc = NULL; /* rbd_dev now owns this */
4131 spec = NULL; /* rbd_dev now owns this */
4133 rbd_dev->mapping.read_only = rbd_opts->read_only;
4135 rbd_opts = NULL; /* done with this */
4137 rc = rbd_dev_probe(rbd_dev);
4139 goto err_out_rbd_dev;
4143 rbd_dev_destroy(rbd_dev);
4145 rbd_put_client(rbdc);
4148 ceph_destroy_options(ceph_opts);
4152 module_put(THIS_MODULE);
4154 dout("Error adding device %s\n", buf);
4156 return (ssize_t) rc;
4159 static struct rbd_device *__rbd_get_dev(unsigned long dev_id)
4161 struct list_head *tmp;
4162 struct rbd_device *rbd_dev;
4164 spin_lock(&rbd_dev_list_lock);
4165 list_for_each(tmp, &rbd_dev_list) {
4166 rbd_dev = list_entry(tmp, struct rbd_device, node);
4167 if (rbd_dev->dev_id == dev_id) {
4168 spin_unlock(&rbd_dev_list_lock);
4172 spin_unlock(&rbd_dev_list_lock);
4176 static void rbd_dev_release(struct device *dev)
4178 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
4180 if (rbd_dev->watch_event)
4181 rbd_dev_header_watch_sync(rbd_dev, 0);
4183 /* clean up and free blkdev */
4184 rbd_free_disk(rbd_dev);
4185 unregister_blkdev(rbd_dev->major, rbd_dev->name);
4187 /* release allocated disk header fields */
4188 rbd_header_free(&rbd_dev->header);
4190 /* done with the id, and with the rbd_dev */
4191 rbd_dev_id_put(rbd_dev);
4192 rbd_assert(rbd_dev->rbd_client != NULL);
4193 rbd_dev_destroy(rbd_dev);
4195 /* release module ref */
4196 module_put(THIS_MODULE);
4199 static ssize_t rbd_remove(struct bus_type *bus,
4203 struct rbd_device *rbd_dev = NULL;
4208 rc = strict_strtoul(buf, 10, &ul);
4212 /* convert to int; abort if we lost anything in the conversion */
4213 target_id = (int) ul;
4214 if (target_id != ul)
4217 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
4219 rbd_dev = __rbd_get_dev(target_id);
4225 spin_lock_irq(&rbd_dev->lock);
4226 if (rbd_dev->open_count)
4229 set_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags);
4230 spin_unlock_irq(&rbd_dev->lock);
4234 rbd_remove_all_snaps(rbd_dev);
4235 rbd_bus_del_dev(rbd_dev);
4238 mutex_unlock(&ctl_mutex);
4244 * create control files in sysfs
4247 static int rbd_sysfs_init(void)
4251 ret = device_register(&rbd_root_dev);
4255 ret = bus_register(&rbd_bus_type);
4257 device_unregister(&rbd_root_dev);
4262 static void rbd_sysfs_cleanup(void)
4264 bus_unregister(&rbd_bus_type);
4265 device_unregister(&rbd_root_dev);
4268 static int __init rbd_init(void)
4272 if (!libceph_compatible(NULL)) {
4273 rbd_warn(NULL, "libceph incompatibility (quitting)");
4277 rc = rbd_sysfs_init();
4280 pr_info("loaded " RBD_DRV_NAME_LONG "\n");
4284 static void __exit rbd_exit(void)
4286 rbd_sysfs_cleanup();
4289 module_init(rbd_init);
4290 module_exit(rbd_exit);
4292 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
4293 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
4294 MODULE_DESCRIPTION("rados block device");
4296 /* following authorship retained from original osdblk.c */
4297 MODULE_AUTHOR("Jeff Garzik <jeff@garzik.org>");
4299 MODULE_LICENSE("GPL");