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 #define RBD_DRV_NAME "rbd"
56 #define RBD_DRV_NAME_LONG "rbd (rados block device)"
58 #define RBD_MINORS_PER_MAJOR 256 /* max minors per blkdev */
60 #define RBD_SNAP_DEV_NAME_PREFIX "snap_"
61 #define RBD_MAX_SNAP_NAME_LEN \
62 (NAME_MAX - (sizeof (RBD_SNAP_DEV_NAME_PREFIX) - 1))
64 #define RBD_MAX_SNAP_COUNT 510 /* allows max snapc to fit in 4KB */
66 #define RBD_SNAP_HEAD_NAME "-"
68 /* This allows a single page to hold an image name sent by OSD */
69 #define RBD_IMAGE_NAME_LEN_MAX (PAGE_SIZE - sizeof (__le32) - 1)
70 #define RBD_IMAGE_ID_LEN_MAX 64
72 #define RBD_OBJ_PREFIX_LEN_MAX 64
76 #define RBD_FEATURE_LAYERING 1
78 /* Features supported by this (client software) implementation. */
80 #define RBD_FEATURES_ALL (0)
83 * An RBD device name will be "rbd#", where the "rbd" comes from
84 * RBD_DRV_NAME above, and # is a unique integer identifier.
85 * MAX_INT_FORMAT_WIDTH is used in ensuring DEV_NAME_LEN is big
86 * enough to hold all possible device names.
88 #define DEV_NAME_LEN 32
89 #define MAX_INT_FORMAT_WIDTH ((5 * sizeof (int)) / 2 + 1)
92 * block device image metadata (in-memory version)
94 struct rbd_image_header {
95 /* These four fields never change for a given rbd image */
102 /* The remaining fields need to be updated occasionally */
104 struct ceph_snap_context *snapc;
112 * An rbd image specification.
114 * The tuple (pool_id, image_id, snap_id) is sufficient to uniquely
115 * identify an image. Each rbd_dev structure includes a pointer to
116 * an rbd_spec structure that encapsulates this identity.
118 * Each of the id's in an rbd_spec has an associated name. For a
119 * user-mapped image, the names are supplied and the id's associated
120 * with them are looked up. For a layered image, a parent image is
121 * defined by the tuple, and the names are looked up.
123 * An rbd_dev structure contains a parent_spec pointer which is
124 * non-null if the image it represents is a child in a layered
125 * image. This pointer will refer to the rbd_spec structure used
126 * by the parent rbd_dev for its own identity (i.e., the structure
127 * is shared between the parent and child).
129 * Since these structures are populated once, during the discovery
130 * phase of image construction, they are effectively immutable so
131 * we make no effort to synchronize access to them.
133 * Note that code herein does not assume the image name is known (it
134 * could be a null pointer).
150 * an instance of the client. multiple devices may share an rbd client.
153 struct ceph_client *client;
155 struct list_head node;
158 struct rbd_img_request;
159 typedef void (*rbd_img_callback_t)(struct rbd_img_request *);
161 #define BAD_WHICH U32_MAX /* Good which or bad which, which? */
163 struct rbd_obj_request;
164 typedef void (*rbd_obj_callback_t)(struct rbd_obj_request *);
166 enum obj_request_type {
167 OBJ_REQUEST_NODATA, OBJ_REQUEST_BIO, OBJ_REQUEST_PAGES
170 struct rbd_obj_request {
171 const char *object_name;
172 u64 offset; /* object start byte */
173 u64 length; /* bytes from offset */
175 struct rbd_img_request *img_request;
176 struct list_head links; /* img_request->obj_requests */
177 u32 which; /* posn image request list */
179 enum obj_request_type type;
181 struct bio *bio_list;
188 struct ceph_osd_request *osd_req;
190 u64 xferred; /* bytes transferred */
195 rbd_obj_callback_t callback;
196 struct completion completion;
201 struct rbd_img_request {
203 struct rbd_device *rbd_dev;
204 u64 offset; /* starting image byte offset */
205 u64 length; /* byte count from offset */
206 bool write_request; /* false for read */
208 struct ceph_snap_context *snapc; /* for writes */
209 u64 snap_id; /* for reads */
211 spinlock_t completion_lock;/* protects next_completion */
213 rbd_img_callback_t callback;
215 u32 obj_request_count;
216 struct list_head obj_requests; /* rbd_obj_request structs */
221 #define for_each_obj_request(ireq, oreq) \
222 list_for_each_entry(oreq, &(ireq)->obj_requests, links)
223 #define for_each_obj_request_from(ireq, oreq) \
224 list_for_each_entry_from(oreq, &(ireq)->obj_requests, links)
225 #define for_each_obj_request_safe(ireq, oreq, n) \
226 list_for_each_entry_safe_reverse(oreq, n, &(ireq)->obj_requests, links)
232 struct list_head node;
247 int dev_id; /* blkdev unique id */
249 int major; /* blkdev assigned major */
250 struct gendisk *disk; /* blkdev's gendisk and rq */
252 u32 image_format; /* Either 1 or 2 */
253 struct rbd_client *rbd_client;
255 char name[DEV_NAME_LEN]; /* blkdev name, e.g. rbd3 */
257 spinlock_t lock; /* queue, flags, open_count */
259 struct rbd_image_header header;
260 unsigned long flags; /* possibly lock protected */
261 struct rbd_spec *spec;
265 struct ceph_file_layout layout;
267 struct ceph_osd_event *watch_event;
268 struct rbd_obj_request *watch_request;
270 struct rbd_spec *parent_spec;
273 /* protects updating the header */
274 struct rw_semaphore header_rwsem;
276 struct rbd_mapping mapping;
278 struct list_head node;
280 /* list of snapshots */
281 struct list_head snaps;
285 unsigned long open_count; /* protected by lock */
289 * Flag bits for rbd_dev->flags. If atomicity is required,
290 * rbd_dev->lock is used to protect access.
292 * Currently, only the "removing" flag (which is coupled with the
293 * "open_count" field) requires atomic access.
296 RBD_DEV_FLAG_EXISTS, /* mapped snapshot has not been deleted */
297 RBD_DEV_FLAG_REMOVING, /* this mapping is being removed */
300 static DEFINE_MUTEX(ctl_mutex); /* Serialize open/close/setup/teardown */
302 static LIST_HEAD(rbd_dev_list); /* devices */
303 static DEFINE_SPINLOCK(rbd_dev_list_lock);
305 static LIST_HEAD(rbd_client_list); /* clients */
306 static DEFINE_SPINLOCK(rbd_client_list_lock);
308 static int rbd_dev_snaps_update(struct rbd_device *rbd_dev);
309 static int rbd_dev_snaps_register(struct rbd_device *rbd_dev);
311 static void rbd_dev_release(struct device *dev);
312 static void rbd_remove_snap_dev(struct rbd_snap *snap);
314 static ssize_t rbd_add(struct bus_type *bus, const char *buf,
316 static ssize_t rbd_remove(struct bus_type *bus, const char *buf,
319 static struct bus_attribute rbd_bus_attrs[] = {
320 __ATTR(add, S_IWUSR, NULL, rbd_add),
321 __ATTR(remove, S_IWUSR, NULL, rbd_remove),
325 static struct bus_type rbd_bus_type = {
327 .bus_attrs = rbd_bus_attrs,
330 static void rbd_root_dev_release(struct device *dev)
334 static struct device rbd_root_dev = {
336 .release = rbd_root_dev_release,
339 static __printf(2, 3)
340 void rbd_warn(struct rbd_device *rbd_dev, const char *fmt, ...)
342 struct va_format vaf;
350 printk(KERN_WARNING "%s: %pV\n", RBD_DRV_NAME, &vaf);
351 else if (rbd_dev->disk)
352 printk(KERN_WARNING "%s: %s: %pV\n",
353 RBD_DRV_NAME, rbd_dev->disk->disk_name, &vaf);
354 else if (rbd_dev->spec && rbd_dev->spec->image_name)
355 printk(KERN_WARNING "%s: image %s: %pV\n",
356 RBD_DRV_NAME, rbd_dev->spec->image_name, &vaf);
357 else if (rbd_dev->spec && rbd_dev->spec->image_id)
358 printk(KERN_WARNING "%s: id %s: %pV\n",
359 RBD_DRV_NAME, rbd_dev->spec->image_id, &vaf);
361 printk(KERN_WARNING "%s: rbd_dev %p: %pV\n",
362 RBD_DRV_NAME, rbd_dev, &vaf);
367 #define rbd_assert(expr) \
368 if (unlikely(!(expr))) { \
369 printk(KERN_ERR "\nAssertion failure in %s() " \
371 "\trbd_assert(%s);\n\n", \
372 __func__, __LINE__, #expr); \
375 #else /* !RBD_DEBUG */
376 # define rbd_assert(expr) ((void) 0)
377 #endif /* !RBD_DEBUG */
379 static int rbd_dev_refresh(struct rbd_device *rbd_dev, u64 *hver);
380 static int rbd_dev_v2_refresh(struct rbd_device *rbd_dev, u64 *hver);
382 static int rbd_open(struct block_device *bdev, fmode_t mode)
384 struct rbd_device *rbd_dev = bdev->bd_disk->private_data;
385 bool removing = false;
387 if ((mode & FMODE_WRITE) && rbd_dev->mapping.read_only)
390 spin_lock_irq(&rbd_dev->lock);
391 if (test_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags))
394 rbd_dev->open_count++;
395 spin_unlock_irq(&rbd_dev->lock);
399 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
400 (void) get_device(&rbd_dev->dev);
401 set_device_ro(bdev, rbd_dev->mapping.read_only);
402 mutex_unlock(&ctl_mutex);
407 static int rbd_release(struct gendisk *disk, fmode_t mode)
409 struct rbd_device *rbd_dev = disk->private_data;
410 unsigned long open_count_before;
412 spin_lock_irq(&rbd_dev->lock);
413 open_count_before = rbd_dev->open_count--;
414 spin_unlock_irq(&rbd_dev->lock);
415 rbd_assert(open_count_before > 0);
417 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
418 put_device(&rbd_dev->dev);
419 mutex_unlock(&ctl_mutex);
424 static const struct block_device_operations rbd_bd_ops = {
425 .owner = THIS_MODULE,
427 .release = rbd_release,
431 * Initialize an rbd client instance.
434 static struct rbd_client *rbd_client_create(struct ceph_options *ceph_opts)
436 struct rbd_client *rbdc;
439 dout("%s:\n", __func__);
440 rbdc = kmalloc(sizeof(struct rbd_client), GFP_KERNEL);
444 kref_init(&rbdc->kref);
445 INIT_LIST_HEAD(&rbdc->node);
447 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
449 rbdc->client = ceph_create_client(ceph_opts, rbdc, 0, 0);
450 if (IS_ERR(rbdc->client))
452 ceph_opts = NULL; /* Now rbdc->client is responsible for ceph_opts */
454 ret = ceph_open_session(rbdc->client);
458 spin_lock(&rbd_client_list_lock);
459 list_add_tail(&rbdc->node, &rbd_client_list);
460 spin_unlock(&rbd_client_list_lock);
462 mutex_unlock(&ctl_mutex);
463 dout("%s: rbdc %p\n", __func__, rbdc);
468 ceph_destroy_client(rbdc->client);
470 mutex_unlock(&ctl_mutex);
474 ceph_destroy_options(ceph_opts);
475 dout("%s: error %d\n", __func__, ret);
481 * Find a ceph client with specific addr and configuration. If
482 * found, bump its reference count.
484 static struct rbd_client *rbd_client_find(struct ceph_options *ceph_opts)
486 struct rbd_client *client_node;
489 if (ceph_opts->flags & CEPH_OPT_NOSHARE)
492 spin_lock(&rbd_client_list_lock);
493 list_for_each_entry(client_node, &rbd_client_list, node) {
494 if (!ceph_compare_options(ceph_opts, client_node->client)) {
495 kref_get(&client_node->kref);
500 spin_unlock(&rbd_client_list_lock);
502 return found ? client_node : NULL;
512 /* string args above */
515 /* Boolean args above */
519 static match_table_t rbd_opts_tokens = {
521 /* string args above */
522 {Opt_read_only, "read_only"},
523 {Opt_read_only, "ro"}, /* Alternate spelling */
524 {Opt_read_write, "read_write"},
525 {Opt_read_write, "rw"}, /* Alternate spelling */
526 /* Boolean args above */
534 #define RBD_READ_ONLY_DEFAULT false
536 static int parse_rbd_opts_token(char *c, void *private)
538 struct rbd_options *rbd_opts = private;
539 substring_t argstr[MAX_OPT_ARGS];
540 int token, intval, ret;
542 token = match_token(c, rbd_opts_tokens, argstr);
546 if (token < Opt_last_int) {
547 ret = match_int(&argstr[0], &intval);
549 pr_err("bad mount option arg (not int) "
553 dout("got int token %d val %d\n", token, intval);
554 } else if (token > Opt_last_int && token < Opt_last_string) {
555 dout("got string token %d val %s\n", token,
557 } else if (token > Opt_last_string && token < Opt_last_bool) {
558 dout("got Boolean token %d\n", token);
560 dout("got token %d\n", token);
565 rbd_opts->read_only = true;
568 rbd_opts->read_only = false;
578 * Get a ceph client with specific addr and configuration, if one does
579 * not exist create it.
581 static struct rbd_client *rbd_get_client(struct ceph_options *ceph_opts)
583 struct rbd_client *rbdc;
585 rbdc = rbd_client_find(ceph_opts);
586 if (rbdc) /* using an existing client */
587 ceph_destroy_options(ceph_opts);
589 rbdc = rbd_client_create(ceph_opts);
595 * Destroy ceph client
597 * Caller must hold rbd_client_list_lock.
599 static void rbd_client_release(struct kref *kref)
601 struct rbd_client *rbdc = container_of(kref, struct rbd_client, kref);
603 dout("%s: rbdc %p\n", __func__, rbdc);
604 spin_lock(&rbd_client_list_lock);
605 list_del(&rbdc->node);
606 spin_unlock(&rbd_client_list_lock);
608 ceph_destroy_client(rbdc->client);
613 * Drop reference to ceph client node. If it's not referenced anymore, release
616 static void rbd_put_client(struct rbd_client *rbdc)
619 kref_put(&rbdc->kref, rbd_client_release);
622 static bool rbd_image_format_valid(u32 image_format)
624 return image_format == 1 || image_format == 2;
627 static bool rbd_dev_ondisk_valid(struct rbd_image_header_ondisk *ondisk)
632 /* The header has to start with the magic rbd header text */
633 if (memcmp(&ondisk->text, RBD_HEADER_TEXT, sizeof (RBD_HEADER_TEXT)))
636 /* The bio layer requires at least sector-sized I/O */
638 if (ondisk->options.order < SECTOR_SHIFT)
641 /* If we use u64 in a few spots we may be able to loosen this */
643 if (ondisk->options.order > 8 * sizeof (int) - 1)
647 * The size of a snapshot header has to fit in a size_t, and
648 * that limits the number of snapshots.
650 snap_count = le32_to_cpu(ondisk->snap_count);
651 size = SIZE_MAX - sizeof (struct ceph_snap_context);
652 if (snap_count > size / sizeof (__le64))
656 * Not only that, but the size of the entire the snapshot
657 * header must also be representable in a size_t.
659 size -= snap_count * sizeof (__le64);
660 if ((u64) size < le64_to_cpu(ondisk->snap_names_len))
667 * Create a new header structure, translate header format from the on-disk
670 static int rbd_header_from_disk(struct rbd_image_header *header,
671 struct rbd_image_header_ondisk *ondisk)
678 memset(header, 0, sizeof (*header));
680 snap_count = le32_to_cpu(ondisk->snap_count);
682 len = strnlen(ondisk->object_prefix, sizeof (ondisk->object_prefix));
683 header->object_prefix = kmalloc(len + 1, GFP_KERNEL);
684 if (!header->object_prefix)
686 memcpy(header->object_prefix, ondisk->object_prefix, len);
687 header->object_prefix[len] = '\0';
690 u64 snap_names_len = le64_to_cpu(ondisk->snap_names_len);
692 /* Save a copy of the snapshot names */
694 if (snap_names_len > (u64) SIZE_MAX)
696 header->snap_names = kmalloc(snap_names_len, GFP_KERNEL);
697 if (!header->snap_names)
700 * Note that rbd_dev_v1_header_read() guarantees
701 * the ondisk buffer we're working with has
702 * snap_names_len bytes beyond the end of the
703 * snapshot id array, this memcpy() is safe.
705 memcpy(header->snap_names, &ondisk->snaps[snap_count],
708 /* Record each snapshot's size */
710 size = snap_count * sizeof (*header->snap_sizes);
711 header->snap_sizes = kmalloc(size, GFP_KERNEL);
712 if (!header->snap_sizes)
714 for (i = 0; i < snap_count; i++)
715 header->snap_sizes[i] =
716 le64_to_cpu(ondisk->snaps[i].image_size);
718 WARN_ON(ondisk->snap_names_len);
719 header->snap_names = NULL;
720 header->snap_sizes = NULL;
723 header->features = 0; /* No features support in v1 images */
724 header->obj_order = ondisk->options.order;
725 header->crypt_type = ondisk->options.crypt_type;
726 header->comp_type = ondisk->options.comp_type;
728 /* Allocate and fill in the snapshot context */
730 header->image_size = le64_to_cpu(ondisk->image_size);
731 size = sizeof (struct ceph_snap_context);
732 size += snap_count * sizeof (header->snapc->snaps[0]);
733 header->snapc = kzalloc(size, GFP_KERNEL);
737 atomic_set(&header->snapc->nref, 1);
738 header->snapc->seq = le64_to_cpu(ondisk->snap_seq);
739 header->snapc->num_snaps = snap_count;
740 for (i = 0; i < snap_count; i++)
741 header->snapc->snaps[i] =
742 le64_to_cpu(ondisk->snaps[i].id);
747 kfree(header->snap_sizes);
748 header->snap_sizes = NULL;
749 kfree(header->snap_names);
750 header->snap_names = NULL;
751 kfree(header->object_prefix);
752 header->object_prefix = NULL;
757 static const char *rbd_snap_name(struct rbd_device *rbd_dev, u64 snap_id)
759 struct rbd_snap *snap;
761 if (snap_id == CEPH_NOSNAP)
762 return RBD_SNAP_HEAD_NAME;
764 list_for_each_entry(snap, &rbd_dev->snaps, node)
765 if (snap_id == snap->id)
771 static int snap_by_name(struct rbd_device *rbd_dev, const char *snap_name)
774 struct rbd_snap *snap;
776 list_for_each_entry(snap, &rbd_dev->snaps, node) {
777 if (!strcmp(snap_name, snap->name)) {
778 rbd_dev->spec->snap_id = snap->id;
779 rbd_dev->mapping.size = snap->size;
780 rbd_dev->mapping.features = snap->features;
789 static int rbd_dev_set_mapping(struct rbd_device *rbd_dev)
793 if (!memcmp(rbd_dev->spec->snap_name, RBD_SNAP_HEAD_NAME,
794 sizeof (RBD_SNAP_HEAD_NAME))) {
795 rbd_dev->spec->snap_id = CEPH_NOSNAP;
796 rbd_dev->mapping.size = rbd_dev->header.image_size;
797 rbd_dev->mapping.features = rbd_dev->header.features;
800 ret = snap_by_name(rbd_dev, rbd_dev->spec->snap_name);
803 rbd_dev->mapping.read_only = true;
805 set_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags);
811 static void rbd_header_free(struct rbd_image_header *header)
813 kfree(header->object_prefix);
814 header->object_prefix = NULL;
815 kfree(header->snap_sizes);
816 header->snap_sizes = NULL;
817 kfree(header->snap_names);
818 header->snap_names = NULL;
819 ceph_put_snap_context(header->snapc);
820 header->snapc = NULL;
823 static const char *rbd_segment_name(struct rbd_device *rbd_dev, u64 offset)
829 name = kmalloc(MAX_OBJ_NAME_SIZE + 1, GFP_NOIO);
832 segment = offset >> rbd_dev->header.obj_order;
833 ret = snprintf(name, MAX_OBJ_NAME_SIZE + 1, "%s.%012llx",
834 rbd_dev->header.object_prefix, segment);
835 if (ret < 0 || ret > MAX_OBJ_NAME_SIZE) {
836 pr_err("error formatting segment name for #%llu (%d)\n",
845 static u64 rbd_segment_offset(struct rbd_device *rbd_dev, u64 offset)
847 u64 segment_size = (u64) 1 << rbd_dev->header.obj_order;
849 return offset & (segment_size - 1);
852 static u64 rbd_segment_length(struct rbd_device *rbd_dev,
853 u64 offset, u64 length)
855 u64 segment_size = (u64) 1 << rbd_dev->header.obj_order;
857 offset &= segment_size - 1;
859 rbd_assert(length <= U64_MAX - offset);
860 if (offset + length > segment_size)
861 length = segment_size - offset;
867 * returns the size of an object in the image
869 static u64 rbd_obj_bytes(struct rbd_image_header *header)
871 return 1 << header->obj_order;
878 static void bio_chain_put(struct bio *chain)
884 chain = chain->bi_next;
890 * zeros a bio chain, starting at specific offset
892 static void zero_bio_chain(struct bio *chain, int start_ofs)
901 bio_for_each_segment(bv, chain, i) {
902 if (pos + bv->bv_len > start_ofs) {
903 int remainder = max(start_ofs - pos, 0);
904 buf = bvec_kmap_irq(bv, &flags);
905 memset(buf + remainder, 0,
906 bv->bv_len - remainder);
907 bvec_kunmap_irq(buf, &flags);
912 chain = chain->bi_next;
917 * Clone a portion of a bio, starting at the given byte offset
918 * and continuing for the number of bytes indicated.
920 static struct bio *bio_clone_range(struct bio *bio_src,
929 unsigned short end_idx;
933 /* Handle the easy case for the caller */
935 if (!offset && len == bio_src->bi_size)
936 return bio_clone(bio_src, gfpmask);
938 if (WARN_ON_ONCE(!len))
940 if (WARN_ON_ONCE(len > bio_src->bi_size))
942 if (WARN_ON_ONCE(offset > bio_src->bi_size - len))
945 /* Find first affected segment... */
948 __bio_for_each_segment(bv, bio_src, idx, 0) {
949 if (resid < bv->bv_len)
955 /* ...and the last affected segment */
958 __bio_for_each_segment(bv, bio_src, end_idx, idx) {
959 if (resid <= bv->bv_len)
963 vcnt = end_idx - idx + 1;
965 /* Build the clone */
967 bio = bio_alloc(gfpmask, (unsigned int) vcnt);
969 return NULL; /* ENOMEM */
971 bio->bi_bdev = bio_src->bi_bdev;
972 bio->bi_sector = bio_src->bi_sector + (offset >> SECTOR_SHIFT);
973 bio->bi_rw = bio_src->bi_rw;
974 bio->bi_flags |= 1 << BIO_CLONED;
977 * Copy over our part of the bio_vec, then update the first
978 * and last (or only) entries.
980 memcpy(&bio->bi_io_vec[0], &bio_src->bi_io_vec[idx],
981 vcnt * sizeof (struct bio_vec));
982 bio->bi_io_vec[0].bv_offset += voff;
984 bio->bi_io_vec[0].bv_len -= voff;
985 bio->bi_io_vec[vcnt - 1].bv_len = resid;
987 bio->bi_io_vec[0].bv_len = len;
998 * Clone a portion of a bio chain, starting at the given byte offset
999 * into the first bio in the source chain and continuing for the
1000 * number of bytes indicated. The result is another bio chain of
1001 * exactly the given length, or a null pointer on error.
1003 * The bio_src and offset parameters are both in-out. On entry they
1004 * refer to the first source bio and the offset into that bio where
1005 * the start of data to be cloned is located.
1007 * On return, bio_src is updated to refer to the bio in the source
1008 * chain that contains first un-cloned byte, and *offset will
1009 * contain the offset of that byte within that bio.
1011 static struct bio *bio_chain_clone_range(struct bio **bio_src,
1012 unsigned int *offset,
1016 struct bio *bi = *bio_src;
1017 unsigned int off = *offset;
1018 struct bio *chain = NULL;
1021 /* Build up a chain of clone bios up to the limit */
1023 if (!bi || off >= bi->bi_size || !len)
1024 return NULL; /* Nothing to clone */
1028 unsigned int bi_size;
1032 rbd_warn(NULL, "bio_chain exhausted with %u left", len);
1033 goto out_err; /* EINVAL; ran out of bio's */
1035 bi_size = min_t(unsigned int, bi->bi_size - off, len);
1036 bio = bio_clone_range(bi, off, bi_size, gfpmask);
1038 goto out_err; /* ENOMEM */
1041 end = &bio->bi_next;
1044 if (off == bi->bi_size) {
1055 bio_chain_put(chain);
1060 static void rbd_obj_request_get(struct rbd_obj_request *obj_request)
1062 dout("%s: obj %p (was %d)\n", __func__, obj_request,
1063 atomic_read(&obj_request->kref.refcount));
1064 kref_get(&obj_request->kref);
1067 static void rbd_obj_request_destroy(struct kref *kref);
1068 static void rbd_obj_request_put(struct rbd_obj_request *obj_request)
1070 rbd_assert(obj_request != NULL);
1071 dout("%s: obj %p (was %d)\n", __func__, obj_request,
1072 atomic_read(&obj_request->kref.refcount));
1073 kref_put(&obj_request->kref, rbd_obj_request_destroy);
1076 static void rbd_img_request_get(struct rbd_img_request *img_request)
1078 dout("%s: img %p (was %d)\n", __func__, img_request,
1079 atomic_read(&img_request->kref.refcount));
1080 kref_get(&img_request->kref);
1083 static void rbd_img_request_destroy(struct kref *kref);
1084 static void rbd_img_request_put(struct rbd_img_request *img_request)
1086 rbd_assert(img_request != NULL);
1087 dout("%s: img %p (was %d)\n", __func__, img_request,
1088 atomic_read(&img_request->kref.refcount));
1089 kref_put(&img_request->kref, rbd_img_request_destroy);
1092 static inline void rbd_img_obj_request_add(struct rbd_img_request *img_request,
1093 struct rbd_obj_request *obj_request)
1095 rbd_assert(obj_request->img_request == NULL);
1097 rbd_obj_request_get(obj_request);
1098 obj_request->img_request = img_request;
1099 obj_request->which = img_request->obj_request_count;
1100 rbd_assert(obj_request->which != BAD_WHICH);
1101 img_request->obj_request_count++;
1102 list_add_tail(&obj_request->links, &img_request->obj_requests);
1103 dout("%s: img %p obj %p w=%u\n", __func__, img_request, obj_request,
1104 obj_request->which);
1107 static inline void rbd_img_obj_request_del(struct rbd_img_request *img_request,
1108 struct rbd_obj_request *obj_request)
1110 rbd_assert(obj_request->which != BAD_WHICH);
1112 dout("%s: img %p obj %p w=%u\n", __func__, img_request, obj_request,
1113 obj_request->which);
1114 list_del(&obj_request->links);
1115 rbd_assert(img_request->obj_request_count > 0);
1116 img_request->obj_request_count--;
1117 rbd_assert(obj_request->which == img_request->obj_request_count);
1118 obj_request->which = BAD_WHICH;
1119 rbd_assert(obj_request->img_request == img_request);
1120 obj_request->img_request = NULL;
1121 obj_request->callback = NULL;
1122 rbd_obj_request_put(obj_request);
1125 static bool obj_request_type_valid(enum obj_request_type type)
1128 case OBJ_REQUEST_NODATA:
1129 case OBJ_REQUEST_BIO:
1130 case OBJ_REQUEST_PAGES:
1137 static int rbd_obj_request_submit(struct ceph_osd_client *osdc,
1138 struct rbd_obj_request *obj_request)
1140 dout("%s: osdc %p obj %p\n", __func__, osdc, obj_request);
1142 return ceph_osdc_start_request(osdc, obj_request->osd_req, false);
1145 static void rbd_img_request_complete(struct rbd_img_request *img_request)
1147 dout("%s: img %p\n", __func__, img_request);
1148 if (img_request->callback)
1149 img_request->callback(img_request);
1151 rbd_img_request_put(img_request);
1154 /* Caller is responsible for rbd_obj_request_destroy(obj_request) */
1156 static int rbd_obj_request_wait(struct rbd_obj_request *obj_request)
1158 dout("%s: obj %p\n", __func__, obj_request);
1160 return wait_for_completion_interruptible(&obj_request->completion);
1163 static void obj_request_done_init(struct rbd_obj_request *obj_request)
1165 atomic_set(&obj_request->done, 0);
1169 static void obj_request_done_set(struct rbd_obj_request *obj_request)
1173 done = atomic_inc_return(&obj_request->done);
1175 struct rbd_img_request *img_request = obj_request->img_request;
1176 struct rbd_device *rbd_dev;
1178 rbd_dev = img_request ? img_request->rbd_dev : NULL;
1179 rbd_warn(rbd_dev, "obj_request %p was already done\n",
1184 static bool obj_request_done_test(struct rbd_obj_request *obj_request)
1187 return atomic_read(&obj_request->done) != 0;
1191 rbd_img_obj_request_read_callback(struct rbd_obj_request *obj_request)
1193 dout("%s: obj %p img %p result %d %llu/%llu\n", __func__,
1194 obj_request, obj_request->img_request, obj_request->result,
1195 obj_request->xferred, obj_request->length);
1197 * ENOENT means a hole in the image. We zero-fill the
1198 * entire length of the request. A short read also implies
1199 * zero-fill to the end of the request. Either way we
1200 * update the xferred count to indicate the whole request
1203 BUG_ON(obj_request->type != OBJ_REQUEST_BIO);
1204 if (obj_request->result == -ENOENT) {
1205 zero_bio_chain(obj_request->bio_list, 0);
1206 obj_request->result = 0;
1207 obj_request->xferred = obj_request->length;
1208 } else if (obj_request->xferred < obj_request->length &&
1209 !obj_request->result) {
1210 zero_bio_chain(obj_request->bio_list, obj_request->xferred);
1211 obj_request->xferred = obj_request->length;
1213 obj_request_done_set(obj_request);
1216 static void rbd_obj_request_complete(struct rbd_obj_request *obj_request)
1218 dout("%s: obj %p cb %p\n", __func__, obj_request,
1219 obj_request->callback);
1220 if (obj_request->callback)
1221 obj_request->callback(obj_request);
1223 complete_all(&obj_request->completion);
1226 static void rbd_osd_trivial_callback(struct rbd_obj_request *obj_request)
1228 dout("%s: obj %p\n", __func__, obj_request);
1229 obj_request_done_set(obj_request);
1232 static void rbd_osd_read_callback(struct rbd_obj_request *obj_request)
1234 dout("%s: obj %p result %d %llu/%llu\n", __func__, obj_request,
1235 obj_request->result, obj_request->xferred, obj_request->length);
1236 if (obj_request->img_request)
1237 rbd_img_obj_request_read_callback(obj_request);
1239 obj_request_done_set(obj_request);
1242 static void rbd_osd_write_callback(struct rbd_obj_request *obj_request)
1244 dout("%s: obj %p result %d %llu\n", __func__, obj_request,
1245 obj_request->result, obj_request->length);
1247 * There is no such thing as a successful short write.
1248 * Our xferred value is the number of bytes transferred
1249 * back. Set it to our originally-requested length.
1251 obj_request->xferred = obj_request->length;
1252 obj_request_done_set(obj_request);
1256 * For a simple stat call there's nothing to do. We'll do more if
1257 * this is part of a write sequence for a layered image.
1259 static void rbd_osd_stat_callback(struct rbd_obj_request *obj_request)
1261 dout("%s: obj %p\n", __func__, obj_request);
1262 obj_request_done_set(obj_request);
1265 static void rbd_osd_req_callback(struct ceph_osd_request *osd_req,
1266 struct ceph_msg *msg)
1268 struct rbd_obj_request *obj_request = osd_req->r_priv;
1271 dout("%s: osd_req %p msg %p\n", __func__, osd_req, msg);
1272 rbd_assert(osd_req == obj_request->osd_req);
1273 rbd_assert(!!obj_request->img_request ^
1274 (obj_request->which == BAD_WHICH));
1276 if (osd_req->r_result < 0)
1277 obj_request->result = osd_req->r_result;
1278 obj_request->version = le64_to_cpu(osd_req->r_reassert_version.version);
1280 WARN_ON(osd_req->r_num_ops != 1); /* For now */
1283 * We support a 64-bit length, but ultimately it has to be
1284 * passed to blk_end_request(), which takes an unsigned int.
1286 obj_request->xferred = osd_req->r_reply_op_len[0];
1287 rbd_assert(obj_request->xferred < (u64) UINT_MAX);
1288 opcode = osd_req->r_ops[0].op;
1290 case CEPH_OSD_OP_READ:
1291 rbd_osd_read_callback(obj_request);
1293 case CEPH_OSD_OP_WRITE:
1294 rbd_osd_write_callback(obj_request);
1296 case CEPH_OSD_OP_STAT:
1297 rbd_osd_stat_callback(obj_request);
1299 case CEPH_OSD_OP_CALL:
1300 case CEPH_OSD_OP_NOTIFY_ACK:
1301 case CEPH_OSD_OP_WATCH:
1302 rbd_osd_trivial_callback(obj_request);
1305 rbd_warn(NULL, "%s: unsupported op %hu\n",
1306 obj_request->object_name, (unsigned short) opcode);
1310 if (obj_request_done_test(obj_request))
1311 rbd_obj_request_complete(obj_request);
1314 static void rbd_osd_req_format(struct rbd_obj_request *obj_request,
1317 struct rbd_img_request *img_request = obj_request->img_request;
1318 struct ceph_osd_request *osd_req = obj_request->osd_req;
1319 struct ceph_snap_context *snapc = NULL;
1320 u64 snap_id = CEPH_NOSNAP;
1321 struct timespec *mtime = NULL;
1322 struct timespec now;
1324 rbd_assert(osd_req != NULL);
1326 if (write_request) {
1330 snapc = img_request->snapc;
1331 } else if (img_request) {
1332 snap_id = img_request->snap_id;
1334 ceph_osdc_build_request(osd_req, obj_request->offset,
1335 snapc, snap_id, mtime);
1338 static struct ceph_osd_request *rbd_osd_req_create(
1339 struct rbd_device *rbd_dev,
1341 struct rbd_obj_request *obj_request)
1343 struct rbd_img_request *img_request = obj_request->img_request;
1344 struct ceph_snap_context *snapc = NULL;
1345 struct ceph_osd_client *osdc;
1346 struct ceph_osd_request *osd_req;
1347 struct ceph_osd_data *osd_data;
1348 u64 offset = obj_request->offset;
1351 rbd_assert(img_request->write_request == write_request);
1352 if (img_request->write_request)
1353 snapc = img_request->snapc;
1356 /* Allocate and initialize the request, for the single op */
1358 osdc = &rbd_dev->rbd_client->client->osdc;
1359 osd_req = ceph_osdc_alloc_request(osdc, snapc, 1, false, GFP_ATOMIC);
1361 return NULL; /* ENOMEM */
1362 osd_data = write_request ? &osd_req->r_data_out : &osd_req->r_data_in;
1364 rbd_assert(obj_request_type_valid(obj_request->type));
1365 switch (obj_request->type) {
1366 case OBJ_REQUEST_NODATA:
1367 break; /* Nothing to do */
1368 case OBJ_REQUEST_BIO:
1369 rbd_assert(obj_request->bio_list != NULL);
1370 ceph_osd_data_bio_init(osd_data, obj_request->bio_list,
1371 obj_request->length);
1373 case OBJ_REQUEST_PAGES:
1374 ceph_osd_data_pages_init(osd_data, obj_request->pages,
1375 obj_request->length, offset & ~PAGE_MASK,
1381 osd_req->r_flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK;
1383 osd_req->r_flags = CEPH_OSD_FLAG_READ;
1385 osd_req->r_callback = rbd_osd_req_callback;
1386 osd_req->r_priv = obj_request;
1388 osd_req->r_oid_len = strlen(obj_request->object_name);
1389 rbd_assert(osd_req->r_oid_len < sizeof (osd_req->r_oid));
1390 memcpy(osd_req->r_oid, obj_request->object_name, osd_req->r_oid_len);
1392 osd_req->r_file_layout = rbd_dev->layout; /* struct */
1397 static void rbd_osd_req_destroy(struct ceph_osd_request *osd_req)
1399 ceph_osdc_put_request(osd_req);
1402 /* object_name is assumed to be a non-null pointer and NUL-terminated */
1404 static struct rbd_obj_request *rbd_obj_request_create(const char *object_name,
1405 u64 offset, u64 length,
1406 enum obj_request_type type)
1408 struct rbd_obj_request *obj_request;
1412 rbd_assert(obj_request_type_valid(type));
1414 size = strlen(object_name) + 1;
1415 obj_request = kzalloc(sizeof (*obj_request) + size, GFP_KERNEL);
1419 name = (char *)(obj_request + 1);
1420 obj_request->object_name = memcpy(name, object_name, size);
1421 obj_request->offset = offset;
1422 obj_request->length = length;
1423 obj_request->which = BAD_WHICH;
1424 obj_request->type = type;
1425 INIT_LIST_HEAD(&obj_request->links);
1426 obj_request_done_init(obj_request);
1427 init_completion(&obj_request->completion);
1428 kref_init(&obj_request->kref);
1430 dout("%s: \"%s\" %llu/%llu %d -> obj %p\n", __func__, object_name,
1431 offset, length, (int)type, obj_request);
1436 static void rbd_obj_request_destroy(struct kref *kref)
1438 struct rbd_obj_request *obj_request;
1440 obj_request = container_of(kref, struct rbd_obj_request, kref);
1442 dout("%s: obj %p\n", __func__, obj_request);
1444 rbd_assert(obj_request->img_request == NULL);
1445 rbd_assert(obj_request->which == BAD_WHICH);
1447 if (obj_request->osd_req)
1448 rbd_osd_req_destroy(obj_request->osd_req);
1450 rbd_assert(obj_request_type_valid(obj_request->type));
1451 switch (obj_request->type) {
1452 case OBJ_REQUEST_NODATA:
1453 break; /* Nothing to do */
1454 case OBJ_REQUEST_BIO:
1455 if (obj_request->bio_list)
1456 bio_chain_put(obj_request->bio_list);
1458 case OBJ_REQUEST_PAGES:
1459 if (obj_request->pages)
1460 ceph_release_page_vector(obj_request->pages,
1461 obj_request->page_count);
1469 * Caller is responsible for filling in the list of object requests
1470 * that comprises the image request, and the Linux request pointer
1471 * (if there is one).
1473 static struct rbd_img_request *rbd_img_request_create(
1474 struct rbd_device *rbd_dev,
1475 u64 offset, u64 length,
1478 struct rbd_img_request *img_request;
1479 struct ceph_snap_context *snapc = NULL;
1481 img_request = kmalloc(sizeof (*img_request), GFP_ATOMIC);
1485 if (write_request) {
1486 down_read(&rbd_dev->header_rwsem);
1487 snapc = ceph_get_snap_context(rbd_dev->header.snapc);
1488 up_read(&rbd_dev->header_rwsem);
1489 if (WARN_ON(!snapc)) {
1491 return NULL; /* Shouldn't happen */
1495 img_request->rq = NULL;
1496 img_request->rbd_dev = rbd_dev;
1497 img_request->offset = offset;
1498 img_request->length = length;
1499 img_request->write_request = write_request;
1501 img_request->snapc = snapc;
1503 img_request->snap_id = rbd_dev->spec->snap_id;
1504 spin_lock_init(&img_request->completion_lock);
1505 img_request->next_completion = 0;
1506 img_request->callback = NULL;
1507 img_request->obj_request_count = 0;
1508 INIT_LIST_HEAD(&img_request->obj_requests);
1509 kref_init(&img_request->kref);
1511 rbd_img_request_get(img_request); /* Avoid a warning */
1512 rbd_img_request_put(img_request); /* TEMPORARY */
1514 dout("%s: rbd_dev %p %s %llu/%llu -> img %p\n", __func__, rbd_dev,
1515 write_request ? "write" : "read", offset, length,
1521 static void rbd_img_request_destroy(struct kref *kref)
1523 struct rbd_img_request *img_request;
1524 struct rbd_obj_request *obj_request;
1525 struct rbd_obj_request *next_obj_request;
1527 img_request = container_of(kref, struct rbd_img_request, kref);
1529 dout("%s: img %p\n", __func__, img_request);
1531 for_each_obj_request_safe(img_request, obj_request, next_obj_request)
1532 rbd_img_obj_request_del(img_request, obj_request);
1533 rbd_assert(img_request->obj_request_count == 0);
1535 if (img_request->write_request)
1536 ceph_put_snap_context(img_request->snapc);
1541 static int rbd_img_request_fill_bio(struct rbd_img_request *img_request,
1542 struct bio *bio_list)
1544 struct rbd_device *rbd_dev = img_request->rbd_dev;
1545 struct rbd_obj_request *obj_request = NULL;
1546 struct rbd_obj_request *next_obj_request;
1547 bool write_request = img_request->write_request;
1548 unsigned int bio_offset;
1553 dout("%s: img %p bio %p\n", __func__, img_request, bio_list);
1555 opcode = write_request ? CEPH_OSD_OP_WRITE : CEPH_OSD_OP_READ;
1557 image_offset = img_request->offset;
1558 rbd_assert(image_offset == bio_list->bi_sector << SECTOR_SHIFT);
1559 resid = img_request->length;
1560 rbd_assert(resid > 0);
1562 struct ceph_osd_request *osd_req;
1563 struct ceph_osd_data *osd_data;
1564 const char *object_name;
1565 unsigned int clone_size;
1569 object_name = rbd_segment_name(rbd_dev, image_offset);
1572 offset = rbd_segment_offset(rbd_dev, image_offset);
1573 length = rbd_segment_length(rbd_dev, image_offset, resid);
1574 obj_request = rbd_obj_request_create(object_name,
1577 kfree(object_name); /* object request has its own copy */
1581 rbd_assert(length <= (u64) UINT_MAX);
1582 clone_size = (unsigned int) length;
1583 obj_request->bio_list = bio_chain_clone_range(&bio_list,
1584 &bio_offset, clone_size,
1586 if (!obj_request->bio_list)
1589 osd_req = rbd_osd_req_create(rbd_dev, write_request,
1593 obj_request->osd_req = osd_req;
1595 osd_data = write_request ? &osd_req->r_data_out
1596 : &osd_req->r_data_in;
1597 osd_req_op_extent_init(osd_req, 0, opcode, offset, length,
1599 osd_req_op_extent_osd_data(osd_req, 0, osd_data);
1600 rbd_osd_req_format(obj_request, write_request);
1602 /* status and version are initially zero-filled */
1604 rbd_img_obj_request_add(img_request, obj_request);
1606 image_offset += length;
1613 rbd_obj_request_put(obj_request);
1615 for_each_obj_request_safe(img_request, obj_request, next_obj_request)
1616 rbd_obj_request_put(obj_request);
1621 static void rbd_img_obj_callback(struct rbd_obj_request *obj_request)
1623 struct rbd_img_request *img_request;
1624 u32 which = obj_request->which;
1627 img_request = obj_request->img_request;
1629 dout("%s: img %p obj %p\n", __func__, img_request, obj_request);
1630 rbd_assert(img_request != NULL);
1631 rbd_assert(img_request->rq != NULL);
1632 rbd_assert(img_request->obj_request_count > 0);
1633 rbd_assert(which != BAD_WHICH);
1634 rbd_assert(which < img_request->obj_request_count);
1635 rbd_assert(which >= img_request->next_completion);
1637 spin_lock_irq(&img_request->completion_lock);
1638 if (which != img_request->next_completion)
1641 for_each_obj_request_from(img_request, obj_request) {
1642 unsigned int xferred;
1646 rbd_assert(which < img_request->obj_request_count);
1648 if (!obj_request_done_test(obj_request))
1651 rbd_assert(obj_request->xferred <= (u64) UINT_MAX);
1652 xferred = (unsigned int) obj_request->xferred;
1653 result = (int) obj_request->result;
1655 rbd_warn(NULL, "obj_request %s result %d xferred %u\n",
1656 img_request->write_request ? "write" : "read",
1659 more = blk_end_request(img_request->rq, result, xferred);
1663 rbd_assert(more ^ (which == img_request->obj_request_count));
1664 img_request->next_completion = which;
1666 spin_unlock_irq(&img_request->completion_lock);
1669 rbd_img_request_complete(img_request);
1672 static int rbd_img_request_submit(struct rbd_img_request *img_request)
1674 struct rbd_device *rbd_dev = img_request->rbd_dev;
1675 struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
1676 struct rbd_obj_request *obj_request;
1677 struct rbd_obj_request *next_obj_request;
1679 dout("%s: img %p\n", __func__, img_request);
1680 for_each_obj_request_safe(img_request, obj_request, next_obj_request) {
1683 obj_request->callback = rbd_img_obj_callback;
1684 ret = rbd_obj_request_submit(osdc, obj_request);
1688 * The image request has its own reference to each
1689 * of its object requests, so we can safely drop the
1692 rbd_obj_request_put(obj_request);
1698 static int rbd_obj_notify_ack(struct rbd_device *rbd_dev,
1699 u64 ver, u64 notify_id)
1701 struct rbd_obj_request *obj_request;
1702 struct ceph_osd_client *osdc;
1705 obj_request = rbd_obj_request_create(rbd_dev->header_name, 0, 0,
1706 OBJ_REQUEST_NODATA);
1711 obj_request->osd_req = rbd_osd_req_create(rbd_dev, false, obj_request);
1712 if (!obj_request->osd_req)
1715 osd_req_op_watch_init(obj_request->osd_req, 0, CEPH_OSD_OP_NOTIFY_ACK,
1717 rbd_osd_req_format(obj_request, false);
1719 osdc = &rbd_dev->rbd_client->client->osdc;
1720 obj_request->callback = rbd_obj_request_put;
1721 ret = rbd_obj_request_submit(osdc, obj_request);
1724 rbd_obj_request_put(obj_request);
1729 static void rbd_watch_cb(u64 ver, u64 notify_id, u8 opcode, void *data)
1731 struct rbd_device *rbd_dev = (struct rbd_device *)data;
1738 dout("%s: \"%s\" notify_id %llu opcode %u\n", __func__,
1739 rbd_dev->header_name, (unsigned long long) notify_id,
1740 (unsigned int) opcode);
1741 rc = rbd_dev_refresh(rbd_dev, &hver);
1743 rbd_warn(rbd_dev, "got notification but failed to "
1744 " update snaps: %d\n", rc);
1746 rbd_obj_notify_ack(rbd_dev, hver, notify_id);
1750 * Request sync osd watch/unwatch. The value of "start" determines
1751 * whether a watch request is being initiated or torn down.
1753 static int rbd_dev_header_watch_sync(struct rbd_device *rbd_dev, int start)
1755 struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
1756 struct rbd_obj_request *obj_request;
1759 rbd_assert(start ^ !!rbd_dev->watch_event);
1760 rbd_assert(start ^ !!rbd_dev->watch_request);
1763 ret = ceph_osdc_create_event(osdc, rbd_watch_cb, rbd_dev,
1764 &rbd_dev->watch_event);
1767 rbd_assert(rbd_dev->watch_event != NULL);
1771 obj_request = rbd_obj_request_create(rbd_dev->header_name, 0, 0,
1772 OBJ_REQUEST_NODATA);
1776 obj_request->osd_req = rbd_osd_req_create(rbd_dev, true, obj_request);
1777 if (!obj_request->osd_req)
1780 osd_req_op_watch_init(obj_request->osd_req, 0, CEPH_OSD_OP_WATCH,
1781 rbd_dev->watch_event->cookie,
1782 rbd_dev->header.obj_version, start);
1783 rbd_osd_req_format(obj_request, true);
1786 ceph_osdc_set_request_linger(osdc, obj_request->osd_req);
1788 ceph_osdc_unregister_linger_request(osdc,
1789 rbd_dev->watch_request->osd_req);
1790 ret = rbd_obj_request_submit(osdc, obj_request);
1793 ret = rbd_obj_request_wait(obj_request);
1796 ret = obj_request->result;
1801 * A watch request is set to linger, so the underlying osd
1802 * request won't go away until we unregister it. We retain
1803 * a pointer to the object request during that time (in
1804 * rbd_dev->watch_request), so we'll keep a reference to
1805 * it. We'll drop that reference (below) after we've
1809 rbd_dev->watch_request = obj_request;
1814 /* We have successfully torn down the watch request */
1816 rbd_obj_request_put(rbd_dev->watch_request);
1817 rbd_dev->watch_request = NULL;
1819 /* Cancel the event if we're tearing down, or on error */
1820 ceph_osdc_cancel_event(rbd_dev->watch_event);
1821 rbd_dev->watch_event = NULL;
1823 rbd_obj_request_put(obj_request);
1829 * Synchronous osd object method call
1831 static int rbd_obj_method_sync(struct rbd_device *rbd_dev,
1832 const char *object_name,
1833 const char *class_name,
1834 const char *method_name,
1835 const char *outbound,
1836 size_t outbound_size,
1838 size_t inbound_size,
1841 struct rbd_obj_request *obj_request;
1842 struct ceph_osd_data *osd_data;
1843 struct ceph_osd_client *osdc;
1844 struct page **pages;
1849 * Method calls are ultimately read operations. The result
1850 * should placed into the inbound buffer provided. They
1851 * also supply outbound data--parameters for the object
1852 * method. Currently if this is present it will be a
1855 page_count = (u32) calc_pages_for(0, inbound_size);
1856 pages = ceph_alloc_page_vector(page_count, GFP_KERNEL);
1858 return PTR_ERR(pages);
1861 obj_request = rbd_obj_request_create(object_name, 0, inbound_size,
1866 obj_request->pages = pages;
1867 obj_request->page_count = page_count;
1869 obj_request->osd_req = rbd_osd_req_create(rbd_dev, false, obj_request);
1870 if (!obj_request->osd_req)
1873 osd_data = &obj_request->osd_req->r_data_in;
1874 osd_req_op_cls_init(obj_request->osd_req, 0, CEPH_OSD_OP_CALL,
1875 class_name, method_name,
1876 outbound, outbound_size);
1877 osd_req_op_cls_response_data(obj_request->osd_req, 0, osd_data);
1878 rbd_osd_req_format(obj_request, false);
1880 osdc = &rbd_dev->rbd_client->client->osdc;
1881 ret = rbd_obj_request_submit(osdc, obj_request);
1884 ret = rbd_obj_request_wait(obj_request);
1888 ret = obj_request->result;
1892 ceph_copy_from_page_vector(pages, inbound, 0, obj_request->xferred);
1894 *version = obj_request->version;
1897 rbd_obj_request_put(obj_request);
1899 ceph_release_page_vector(pages, page_count);
1904 static void rbd_request_fn(struct request_queue *q)
1905 __releases(q->queue_lock) __acquires(q->queue_lock)
1907 struct rbd_device *rbd_dev = q->queuedata;
1908 bool read_only = rbd_dev->mapping.read_only;
1912 while ((rq = blk_fetch_request(q))) {
1913 bool write_request = rq_data_dir(rq) == WRITE;
1914 struct rbd_img_request *img_request;
1918 /* Ignore any non-FS requests that filter through. */
1920 if (rq->cmd_type != REQ_TYPE_FS) {
1921 dout("%s: non-fs request type %d\n", __func__,
1922 (int) rq->cmd_type);
1923 __blk_end_request_all(rq, 0);
1927 /* Ignore/skip any zero-length requests */
1929 offset = (u64) blk_rq_pos(rq) << SECTOR_SHIFT;
1930 length = (u64) blk_rq_bytes(rq);
1933 dout("%s: zero-length request\n", __func__);
1934 __blk_end_request_all(rq, 0);
1938 spin_unlock_irq(q->queue_lock);
1940 /* Disallow writes to a read-only device */
1942 if (write_request) {
1946 rbd_assert(rbd_dev->spec->snap_id == CEPH_NOSNAP);
1950 * Quit early if the mapped snapshot no longer
1951 * exists. It's still possible the snapshot will
1952 * have disappeared by the time our request arrives
1953 * at the osd, but there's no sense in sending it if
1956 if (!test_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags)) {
1957 dout("request for non-existent snapshot");
1958 rbd_assert(rbd_dev->spec->snap_id != CEPH_NOSNAP);
1964 if (WARN_ON(offset && length > U64_MAX - offset + 1))
1965 goto end_request; /* Shouldn't happen */
1968 img_request = rbd_img_request_create(rbd_dev, offset, length,
1973 img_request->rq = rq;
1975 result = rbd_img_request_fill_bio(img_request, rq->bio);
1977 result = rbd_img_request_submit(img_request);
1979 rbd_img_request_put(img_request);
1981 spin_lock_irq(q->queue_lock);
1983 rbd_warn(rbd_dev, "obj_request %s result %d\n",
1984 write_request ? "write" : "read", result);
1985 __blk_end_request_all(rq, result);
1991 * a queue callback. Makes sure that we don't create a bio that spans across
1992 * multiple osd objects. One exception would be with a single page bios,
1993 * which we handle later at bio_chain_clone_range()
1995 static int rbd_merge_bvec(struct request_queue *q, struct bvec_merge_data *bmd,
1996 struct bio_vec *bvec)
1998 struct rbd_device *rbd_dev = q->queuedata;
1999 sector_t sector_offset;
2000 sector_t sectors_per_obj;
2001 sector_t obj_sector_offset;
2005 * Find how far into its rbd object the partition-relative
2006 * bio start sector is to offset relative to the enclosing
2009 sector_offset = get_start_sect(bmd->bi_bdev) + bmd->bi_sector;
2010 sectors_per_obj = 1 << (rbd_dev->header.obj_order - SECTOR_SHIFT);
2011 obj_sector_offset = sector_offset & (sectors_per_obj - 1);
2014 * Compute the number of bytes from that offset to the end
2015 * of the object. Account for what's already used by the bio.
2017 ret = (int) (sectors_per_obj - obj_sector_offset) << SECTOR_SHIFT;
2018 if (ret > bmd->bi_size)
2019 ret -= bmd->bi_size;
2024 * Don't send back more than was asked for. And if the bio
2025 * was empty, let the whole thing through because: "Note
2026 * that a block device *must* allow a single page to be
2027 * added to an empty bio."
2029 rbd_assert(bvec->bv_len <= PAGE_SIZE);
2030 if (ret > (int) bvec->bv_len || !bmd->bi_size)
2031 ret = (int) bvec->bv_len;
2036 static void rbd_free_disk(struct rbd_device *rbd_dev)
2038 struct gendisk *disk = rbd_dev->disk;
2043 if (disk->flags & GENHD_FL_UP)
2046 blk_cleanup_queue(disk->queue);
2050 static int rbd_obj_read_sync(struct rbd_device *rbd_dev,
2051 const char *object_name,
2052 u64 offset, u64 length,
2053 char *buf, u64 *version)
2056 struct rbd_obj_request *obj_request;
2057 struct ceph_osd_data *osd_data;
2058 struct ceph_osd_client *osdc;
2059 struct page **pages = NULL;
2064 page_count = (u32) calc_pages_for(offset, length);
2065 pages = ceph_alloc_page_vector(page_count, GFP_KERNEL);
2067 ret = PTR_ERR(pages);
2070 obj_request = rbd_obj_request_create(object_name, offset, length,
2075 obj_request->pages = pages;
2076 obj_request->page_count = page_count;
2078 obj_request->osd_req = rbd_osd_req_create(rbd_dev, false, obj_request);
2079 if (!obj_request->osd_req)
2082 osd_data = &obj_request->osd_req->r_data_in;
2083 osd_req_op_extent_init(obj_request->osd_req, 0, CEPH_OSD_OP_READ,
2084 offset, length, 0, 0);
2085 osd_req_op_extent_osd_data(obj_request->osd_req, 0, osd_data);
2086 rbd_osd_req_format(obj_request, false);
2088 osdc = &rbd_dev->rbd_client->client->osdc;
2089 ret = rbd_obj_request_submit(osdc, obj_request);
2092 ret = rbd_obj_request_wait(obj_request);
2096 ret = obj_request->result;
2100 rbd_assert(obj_request->xferred <= (u64) SIZE_MAX);
2101 size = (size_t) obj_request->xferred;
2102 ceph_copy_from_page_vector(pages, buf, 0, size);
2103 rbd_assert(size <= (size_t) INT_MAX);
2106 *version = obj_request->version;
2109 rbd_obj_request_put(obj_request);
2111 ceph_release_page_vector(pages, page_count);
2117 * Read the complete header for the given rbd device.
2119 * Returns a pointer to a dynamically-allocated buffer containing
2120 * the complete and validated header. Caller can pass the address
2121 * of a variable that will be filled in with the version of the
2122 * header object at the time it was read.
2124 * Returns a pointer-coded errno if a failure occurs.
2126 static struct rbd_image_header_ondisk *
2127 rbd_dev_v1_header_read(struct rbd_device *rbd_dev, u64 *version)
2129 struct rbd_image_header_ondisk *ondisk = NULL;
2136 * The complete header will include an array of its 64-bit
2137 * snapshot ids, followed by the names of those snapshots as
2138 * a contiguous block of NUL-terminated strings. Note that
2139 * the number of snapshots could change by the time we read
2140 * it in, in which case we re-read it.
2147 size = sizeof (*ondisk);
2148 size += snap_count * sizeof (struct rbd_image_snap_ondisk);
2150 ondisk = kmalloc(size, GFP_KERNEL);
2152 return ERR_PTR(-ENOMEM);
2154 ret = rbd_obj_read_sync(rbd_dev, rbd_dev->header_name,
2156 (char *) ondisk, version);
2159 if (WARN_ON((size_t) ret < size)) {
2161 rbd_warn(rbd_dev, "short header read (want %zd got %d)",
2165 if (!rbd_dev_ondisk_valid(ondisk)) {
2167 rbd_warn(rbd_dev, "invalid header");
2171 names_size = le64_to_cpu(ondisk->snap_names_len);
2172 want_count = snap_count;
2173 snap_count = le32_to_cpu(ondisk->snap_count);
2174 } while (snap_count != want_count);
2181 return ERR_PTR(ret);
2185 * reload the ondisk the header
2187 static int rbd_read_header(struct rbd_device *rbd_dev,
2188 struct rbd_image_header *header)
2190 struct rbd_image_header_ondisk *ondisk;
2194 ondisk = rbd_dev_v1_header_read(rbd_dev, &ver);
2196 return PTR_ERR(ondisk);
2197 ret = rbd_header_from_disk(header, ondisk);
2199 header->obj_version = ver;
2205 static void rbd_remove_all_snaps(struct rbd_device *rbd_dev)
2207 struct rbd_snap *snap;
2208 struct rbd_snap *next;
2210 list_for_each_entry_safe(snap, next, &rbd_dev->snaps, node)
2211 rbd_remove_snap_dev(snap);
2214 static void rbd_update_mapping_size(struct rbd_device *rbd_dev)
2218 if (rbd_dev->spec->snap_id != CEPH_NOSNAP)
2221 size = (sector_t) rbd_dev->header.image_size / SECTOR_SIZE;
2222 dout("setting size to %llu sectors", (unsigned long long) size);
2223 rbd_dev->mapping.size = (u64) size;
2224 set_capacity(rbd_dev->disk, size);
2228 * only read the first part of the ondisk header, without the snaps info
2230 static int rbd_dev_v1_refresh(struct rbd_device *rbd_dev, u64 *hver)
2233 struct rbd_image_header h;
2235 ret = rbd_read_header(rbd_dev, &h);
2239 down_write(&rbd_dev->header_rwsem);
2241 /* Update image size, and check for resize of mapped image */
2242 rbd_dev->header.image_size = h.image_size;
2243 rbd_update_mapping_size(rbd_dev);
2245 /* rbd_dev->header.object_prefix shouldn't change */
2246 kfree(rbd_dev->header.snap_sizes);
2247 kfree(rbd_dev->header.snap_names);
2248 /* osd requests may still refer to snapc */
2249 ceph_put_snap_context(rbd_dev->header.snapc);
2252 *hver = h.obj_version;
2253 rbd_dev->header.obj_version = h.obj_version;
2254 rbd_dev->header.image_size = h.image_size;
2255 rbd_dev->header.snapc = h.snapc;
2256 rbd_dev->header.snap_names = h.snap_names;
2257 rbd_dev->header.snap_sizes = h.snap_sizes;
2258 /* Free the extra copy of the object prefix */
2259 WARN_ON(strcmp(rbd_dev->header.object_prefix, h.object_prefix));
2260 kfree(h.object_prefix);
2262 ret = rbd_dev_snaps_update(rbd_dev);
2264 ret = rbd_dev_snaps_register(rbd_dev);
2266 up_write(&rbd_dev->header_rwsem);
2271 static int rbd_dev_refresh(struct rbd_device *rbd_dev, u64 *hver)
2275 rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
2276 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
2277 if (rbd_dev->image_format == 1)
2278 ret = rbd_dev_v1_refresh(rbd_dev, hver);
2280 ret = rbd_dev_v2_refresh(rbd_dev, hver);
2281 mutex_unlock(&ctl_mutex);
2286 static int rbd_init_disk(struct rbd_device *rbd_dev)
2288 struct gendisk *disk;
2289 struct request_queue *q;
2292 /* create gendisk info */
2293 disk = alloc_disk(RBD_MINORS_PER_MAJOR);
2297 snprintf(disk->disk_name, sizeof(disk->disk_name), RBD_DRV_NAME "%d",
2299 disk->major = rbd_dev->major;
2300 disk->first_minor = 0;
2301 disk->fops = &rbd_bd_ops;
2302 disk->private_data = rbd_dev;
2304 q = blk_init_queue(rbd_request_fn, &rbd_dev->lock);
2308 /* We use the default size, but let's be explicit about it. */
2309 blk_queue_physical_block_size(q, SECTOR_SIZE);
2311 /* set io sizes to object size */
2312 segment_size = rbd_obj_bytes(&rbd_dev->header);
2313 blk_queue_max_hw_sectors(q, segment_size / SECTOR_SIZE);
2314 blk_queue_max_segment_size(q, segment_size);
2315 blk_queue_io_min(q, segment_size);
2316 blk_queue_io_opt(q, segment_size);
2318 blk_queue_merge_bvec(q, rbd_merge_bvec);
2321 q->queuedata = rbd_dev;
2323 rbd_dev->disk = disk;
2325 set_capacity(rbd_dev->disk, rbd_dev->mapping.size / SECTOR_SIZE);
2338 static struct rbd_device *dev_to_rbd_dev(struct device *dev)
2340 return container_of(dev, struct rbd_device, dev);
2343 static ssize_t rbd_size_show(struct device *dev,
2344 struct device_attribute *attr, char *buf)
2346 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
2349 down_read(&rbd_dev->header_rwsem);
2350 size = get_capacity(rbd_dev->disk);
2351 up_read(&rbd_dev->header_rwsem);
2353 return sprintf(buf, "%llu\n", (unsigned long long) size * SECTOR_SIZE);
2357 * Note this shows the features for whatever's mapped, which is not
2358 * necessarily the base image.
2360 static ssize_t rbd_features_show(struct device *dev,
2361 struct device_attribute *attr, char *buf)
2363 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
2365 return sprintf(buf, "0x%016llx\n",
2366 (unsigned long long) rbd_dev->mapping.features);
2369 static ssize_t rbd_major_show(struct device *dev,
2370 struct device_attribute *attr, char *buf)
2372 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
2374 return sprintf(buf, "%d\n", rbd_dev->major);
2377 static ssize_t rbd_client_id_show(struct device *dev,
2378 struct device_attribute *attr, char *buf)
2380 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
2382 return sprintf(buf, "client%lld\n",
2383 ceph_client_id(rbd_dev->rbd_client->client));
2386 static ssize_t rbd_pool_show(struct device *dev,
2387 struct device_attribute *attr, char *buf)
2389 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
2391 return sprintf(buf, "%s\n", rbd_dev->spec->pool_name);
2394 static ssize_t rbd_pool_id_show(struct device *dev,
2395 struct device_attribute *attr, char *buf)
2397 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
2399 return sprintf(buf, "%llu\n",
2400 (unsigned long long) rbd_dev->spec->pool_id);
2403 static ssize_t rbd_name_show(struct device *dev,
2404 struct device_attribute *attr, char *buf)
2406 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
2408 if (rbd_dev->spec->image_name)
2409 return sprintf(buf, "%s\n", rbd_dev->spec->image_name);
2411 return sprintf(buf, "(unknown)\n");
2414 static ssize_t rbd_image_id_show(struct device *dev,
2415 struct device_attribute *attr, char *buf)
2417 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
2419 return sprintf(buf, "%s\n", rbd_dev->spec->image_id);
2423 * Shows the name of the currently-mapped snapshot (or
2424 * RBD_SNAP_HEAD_NAME for the base image).
2426 static ssize_t rbd_snap_show(struct device *dev,
2427 struct device_attribute *attr,
2430 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
2432 return sprintf(buf, "%s\n", rbd_dev->spec->snap_name);
2436 * For an rbd v2 image, shows the pool id, image id, and snapshot id
2437 * for the parent image. If there is no parent, simply shows
2438 * "(no parent image)".
2440 static ssize_t rbd_parent_show(struct device *dev,
2441 struct device_attribute *attr,
2444 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
2445 struct rbd_spec *spec = rbd_dev->parent_spec;
2450 return sprintf(buf, "(no parent image)\n");
2452 count = sprintf(bufp, "pool_id %llu\npool_name %s\n",
2453 (unsigned long long) spec->pool_id, spec->pool_name);
2458 count = sprintf(bufp, "image_id %s\nimage_name %s\n", spec->image_id,
2459 spec->image_name ? spec->image_name : "(unknown)");
2464 count = sprintf(bufp, "snap_id %llu\nsnap_name %s\n",
2465 (unsigned long long) spec->snap_id, spec->snap_name);
2470 count = sprintf(bufp, "overlap %llu\n", rbd_dev->parent_overlap);
2475 return (ssize_t) (bufp - buf);
2478 static ssize_t rbd_image_refresh(struct device *dev,
2479 struct device_attribute *attr,
2483 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
2486 ret = rbd_dev_refresh(rbd_dev, NULL);
2488 return ret < 0 ? ret : size;
2491 static DEVICE_ATTR(size, S_IRUGO, rbd_size_show, NULL);
2492 static DEVICE_ATTR(features, S_IRUGO, rbd_features_show, NULL);
2493 static DEVICE_ATTR(major, S_IRUGO, rbd_major_show, NULL);
2494 static DEVICE_ATTR(client_id, S_IRUGO, rbd_client_id_show, NULL);
2495 static DEVICE_ATTR(pool, S_IRUGO, rbd_pool_show, NULL);
2496 static DEVICE_ATTR(pool_id, S_IRUGO, rbd_pool_id_show, NULL);
2497 static DEVICE_ATTR(name, S_IRUGO, rbd_name_show, NULL);
2498 static DEVICE_ATTR(image_id, S_IRUGO, rbd_image_id_show, NULL);
2499 static DEVICE_ATTR(refresh, S_IWUSR, NULL, rbd_image_refresh);
2500 static DEVICE_ATTR(current_snap, S_IRUGO, rbd_snap_show, NULL);
2501 static DEVICE_ATTR(parent, S_IRUGO, rbd_parent_show, NULL);
2503 static struct attribute *rbd_attrs[] = {
2504 &dev_attr_size.attr,
2505 &dev_attr_features.attr,
2506 &dev_attr_major.attr,
2507 &dev_attr_client_id.attr,
2508 &dev_attr_pool.attr,
2509 &dev_attr_pool_id.attr,
2510 &dev_attr_name.attr,
2511 &dev_attr_image_id.attr,
2512 &dev_attr_current_snap.attr,
2513 &dev_attr_parent.attr,
2514 &dev_attr_refresh.attr,
2518 static struct attribute_group rbd_attr_group = {
2522 static const struct attribute_group *rbd_attr_groups[] = {
2527 static void rbd_sysfs_dev_release(struct device *dev)
2531 static struct device_type rbd_device_type = {
2533 .groups = rbd_attr_groups,
2534 .release = rbd_sysfs_dev_release,
2542 static ssize_t rbd_snap_size_show(struct device *dev,
2543 struct device_attribute *attr,
2546 struct rbd_snap *snap = container_of(dev, struct rbd_snap, dev);
2548 return sprintf(buf, "%llu\n", (unsigned long long)snap->size);
2551 static ssize_t rbd_snap_id_show(struct device *dev,
2552 struct device_attribute *attr,
2555 struct rbd_snap *snap = container_of(dev, struct rbd_snap, dev);
2557 return sprintf(buf, "%llu\n", (unsigned long long)snap->id);
2560 static ssize_t rbd_snap_features_show(struct device *dev,
2561 struct device_attribute *attr,
2564 struct rbd_snap *snap = container_of(dev, struct rbd_snap, dev);
2566 return sprintf(buf, "0x%016llx\n",
2567 (unsigned long long) snap->features);
2570 static DEVICE_ATTR(snap_size, S_IRUGO, rbd_snap_size_show, NULL);
2571 static DEVICE_ATTR(snap_id, S_IRUGO, rbd_snap_id_show, NULL);
2572 static DEVICE_ATTR(snap_features, S_IRUGO, rbd_snap_features_show, NULL);
2574 static struct attribute *rbd_snap_attrs[] = {
2575 &dev_attr_snap_size.attr,
2576 &dev_attr_snap_id.attr,
2577 &dev_attr_snap_features.attr,
2581 static struct attribute_group rbd_snap_attr_group = {
2582 .attrs = rbd_snap_attrs,
2585 static void rbd_snap_dev_release(struct device *dev)
2587 struct rbd_snap *snap = container_of(dev, struct rbd_snap, dev);
2592 static const struct attribute_group *rbd_snap_attr_groups[] = {
2593 &rbd_snap_attr_group,
2597 static struct device_type rbd_snap_device_type = {
2598 .groups = rbd_snap_attr_groups,
2599 .release = rbd_snap_dev_release,
2602 static struct rbd_spec *rbd_spec_get(struct rbd_spec *spec)
2604 kref_get(&spec->kref);
2609 static void rbd_spec_free(struct kref *kref);
2610 static void rbd_spec_put(struct rbd_spec *spec)
2613 kref_put(&spec->kref, rbd_spec_free);
2616 static struct rbd_spec *rbd_spec_alloc(void)
2618 struct rbd_spec *spec;
2620 spec = kzalloc(sizeof (*spec), GFP_KERNEL);
2623 kref_init(&spec->kref);
2625 rbd_spec_put(rbd_spec_get(spec)); /* TEMPORARY */
2630 static void rbd_spec_free(struct kref *kref)
2632 struct rbd_spec *spec = container_of(kref, struct rbd_spec, kref);
2634 kfree(spec->pool_name);
2635 kfree(spec->image_id);
2636 kfree(spec->image_name);
2637 kfree(spec->snap_name);
2641 static struct rbd_device *rbd_dev_create(struct rbd_client *rbdc,
2642 struct rbd_spec *spec)
2644 struct rbd_device *rbd_dev;
2646 rbd_dev = kzalloc(sizeof (*rbd_dev), GFP_KERNEL);
2650 spin_lock_init(&rbd_dev->lock);
2652 INIT_LIST_HEAD(&rbd_dev->node);
2653 INIT_LIST_HEAD(&rbd_dev->snaps);
2654 init_rwsem(&rbd_dev->header_rwsem);
2656 rbd_dev->spec = spec;
2657 rbd_dev->rbd_client = rbdc;
2659 /* Initialize the layout used for all rbd requests */
2661 rbd_dev->layout.fl_stripe_unit = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER);
2662 rbd_dev->layout.fl_stripe_count = cpu_to_le32(1);
2663 rbd_dev->layout.fl_object_size = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER);
2664 rbd_dev->layout.fl_pg_pool = cpu_to_le32((u32) spec->pool_id);
2669 static void rbd_dev_destroy(struct rbd_device *rbd_dev)
2671 rbd_spec_put(rbd_dev->parent_spec);
2672 kfree(rbd_dev->header_name);
2673 rbd_put_client(rbd_dev->rbd_client);
2674 rbd_spec_put(rbd_dev->spec);
2678 static bool rbd_snap_registered(struct rbd_snap *snap)
2680 bool ret = snap->dev.type == &rbd_snap_device_type;
2681 bool reg = device_is_registered(&snap->dev);
2683 rbd_assert(!ret ^ reg);
2688 static void rbd_remove_snap_dev(struct rbd_snap *snap)
2690 list_del(&snap->node);
2691 if (device_is_registered(&snap->dev))
2692 device_unregister(&snap->dev);
2695 static int rbd_register_snap_dev(struct rbd_snap *snap,
2696 struct device *parent)
2698 struct device *dev = &snap->dev;
2701 dev->type = &rbd_snap_device_type;
2702 dev->parent = parent;
2703 dev->release = rbd_snap_dev_release;
2704 dev_set_name(dev, "%s%s", RBD_SNAP_DEV_NAME_PREFIX, snap->name);
2705 dout("%s: registering device for snapshot %s\n", __func__, snap->name);
2707 ret = device_register(dev);
2712 static struct rbd_snap *__rbd_add_snap_dev(struct rbd_device *rbd_dev,
2713 const char *snap_name,
2714 u64 snap_id, u64 snap_size,
2717 struct rbd_snap *snap;
2720 snap = kzalloc(sizeof (*snap), GFP_KERNEL);
2722 return ERR_PTR(-ENOMEM);
2725 snap->name = kstrdup(snap_name, GFP_KERNEL);
2730 snap->size = snap_size;
2731 snap->features = snap_features;
2739 return ERR_PTR(ret);
2742 static char *rbd_dev_v1_snap_info(struct rbd_device *rbd_dev, u32 which,
2743 u64 *snap_size, u64 *snap_features)
2747 rbd_assert(which < rbd_dev->header.snapc->num_snaps);
2749 *snap_size = rbd_dev->header.snap_sizes[which];
2750 *snap_features = 0; /* No features for v1 */
2752 /* Skip over names until we find the one we are looking for */
2754 snap_name = rbd_dev->header.snap_names;
2756 snap_name += strlen(snap_name) + 1;
2762 * Get the size and object order for an image snapshot, or if
2763 * snap_id is CEPH_NOSNAP, gets this information for the base
2766 static int _rbd_dev_v2_snap_size(struct rbd_device *rbd_dev, u64 snap_id,
2767 u8 *order, u64 *snap_size)
2769 __le64 snapid = cpu_to_le64(snap_id);
2774 } __attribute__ ((packed)) size_buf = { 0 };
2776 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
2778 (char *) &snapid, sizeof (snapid),
2779 (char *) &size_buf, sizeof (size_buf), NULL);
2780 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
2784 *order = size_buf.order;
2785 *snap_size = le64_to_cpu(size_buf.size);
2787 dout(" snap_id 0x%016llx order = %u, snap_size = %llu\n",
2788 (unsigned long long) snap_id, (unsigned int) *order,
2789 (unsigned long long) *snap_size);
2794 static int rbd_dev_v2_image_size(struct rbd_device *rbd_dev)
2796 return _rbd_dev_v2_snap_size(rbd_dev, CEPH_NOSNAP,
2797 &rbd_dev->header.obj_order,
2798 &rbd_dev->header.image_size);
2801 static int rbd_dev_v2_object_prefix(struct rbd_device *rbd_dev)
2807 reply_buf = kzalloc(RBD_OBJ_PREFIX_LEN_MAX, GFP_KERNEL);
2811 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
2812 "rbd", "get_object_prefix",
2814 reply_buf, RBD_OBJ_PREFIX_LEN_MAX, NULL);
2815 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
2820 rbd_dev->header.object_prefix = ceph_extract_encoded_string(&p,
2821 p + RBD_OBJ_PREFIX_LEN_MAX,
2824 if (IS_ERR(rbd_dev->header.object_prefix)) {
2825 ret = PTR_ERR(rbd_dev->header.object_prefix);
2826 rbd_dev->header.object_prefix = NULL;
2828 dout(" object_prefix = %s\n", rbd_dev->header.object_prefix);
2837 static int _rbd_dev_v2_snap_features(struct rbd_device *rbd_dev, u64 snap_id,
2840 __le64 snapid = cpu_to_le64(snap_id);
2844 } features_buf = { 0 };
2848 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
2849 "rbd", "get_features",
2850 (char *) &snapid, sizeof (snapid),
2851 (char *) &features_buf, sizeof (features_buf),
2853 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
2857 incompat = le64_to_cpu(features_buf.incompat);
2858 if (incompat & ~RBD_FEATURES_ALL)
2861 *snap_features = le64_to_cpu(features_buf.features);
2863 dout(" snap_id 0x%016llx features = 0x%016llx incompat = 0x%016llx\n",
2864 (unsigned long long) snap_id,
2865 (unsigned long long) *snap_features,
2866 (unsigned long long) le64_to_cpu(features_buf.incompat));
2871 static int rbd_dev_v2_features(struct rbd_device *rbd_dev)
2873 return _rbd_dev_v2_snap_features(rbd_dev, CEPH_NOSNAP,
2874 &rbd_dev->header.features);
2877 static int rbd_dev_v2_parent_info(struct rbd_device *rbd_dev)
2879 struct rbd_spec *parent_spec;
2881 void *reply_buf = NULL;
2889 parent_spec = rbd_spec_alloc();
2893 size = sizeof (__le64) + /* pool_id */
2894 sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX + /* image_id */
2895 sizeof (__le64) + /* snap_id */
2896 sizeof (__le64); /* overlap */
2897 reply_buf = kmalloc(size, GFP_KERNEL);
2903 snapid = cpu_to_le64(CEPH_NOSNAP);
2904 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
2905 "rbd", "get_parent",
2906 (char *) &snapid, sizeof (snapid),
2907 (char *) reply_buf, size, NULL);
2908 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
2914 end = (char *) reply_buf + size;
2915 ceph_decode_64_safe(&p, end, parent_spec->pool_id, out_err);
2916 if (parent_spec->pool_id == CEPH_NOPOOL)
2917 goto out; /* No parent? No problem. */
2919 /* The ceph file layout needs to fit pool id in 32 bits */
2922 if (WARN_ON(parent_spec->pool_id > (u64) U32_MAX))
2925 image_id = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL);
2926 if (IS_ERR(image_id)) {
2927 ret = PTR_ERR(image_id);
2930 parent_spec->image_id = image_id;
2931 ceph_decode_64_safe(&p, end, parent_spec->snap_id, out_err);
2932 ceph_decode_64_safe(&p, end, overlap, out_err);
2934 rbd_dev->parent_overlap = overlap;
2935 rbd_dev->parent_spec = parent_spec;
2936 parent_spec = NULL; /* rbd_dev now owns this */
2941 rbd_spec_put(parent_spec);
2946 static char *rbd_dev_image_name(struct rbd_device *rbd_dev)
2948 size_t image_id_size;
2953 void *reply_buf = NULL;
2955 char *image_name = NULL;
2958 rbd_assert(!rbd_dev->spec->image_name);
2960 len = strlen(rbd_dev->spec->image_id);
2961 image_id_size = sizeof (__le32) + len;
2962 image_id = kmalloc(image_id_size, GFP_KERNEL);
2967 end = (char *) image_id + image_id_size;
2968 ceph_encode_string(&p, end, rbd_dev->spec->image_id, (u32) len);
2970 size = sizeof (__le32) + RBD_IMAGE_NAME_LEN_MAX;
2971 reply_buf = kmalloc(size, GFP_KERNEL);
2975 ret = rbd_obj_method_sync(rbd_dev, RBD_DIRECTORY,
2976 "rbd", "dir_get_name",
2977 image_id, image_id_size,
2978 (char *) reply_buf, size, NULL);
2982 end = (char *) reply_buf + size;
2983 image_name = ceph_extract_encoded_string(&p, end, &len, GFP_KERNEL);
2984 if (IS_ERR(image_name))
2987 dout("%s: name is %s len is %zd\n", __func__, image_name, len);
2996 * When a parent image gets probed, we only have the pool, image,
2997 * and snapshot ids but not the names of any of them. This call
2998 * is made later to fill in those names. It has to be done after
2999 * rbd_dev_snaps_update() has completed because some of the
3000 * information (in particular, snapshot name) is not available
3003 static int rbd_dev_probe_update_spec(struct rbd_device *rbd_dev)
3005 struct ceph_osd_client *osdc;
3007 void *reply_buf = NULL;
3010 if (rbd_dev->spec->pool_name)
3011 return 0; /* Already have the names */
3013 /* Look up the pool name */
3015 osdc = &rbd_dev->rbd_client->client->osdc;
3016 name = ceph_pg_pool_name_by_id(osdc->osdmap, rbd_dev->spec->pool_id);
3018 rbd_warn(rbd_dev, "there is no pool with id %llu",
3019 rbd_dev->spec->pool_id); /* Really a BUG() */
3023 rbd_dev->spec->pool_name = kstrdup(name, GFP_KERNEL);
3024 if (!rbd_dev->spec->pool_name)
3027 /* Fetch the image name; tolerate failure here */
3029 name = rbd_dev_image_name(rbd_dev);
3031 rbd_dev->spec->image_name = (char *) name;
3033 rbd_warn(rbd_dev, "unable to get image name");
3035 /* Look up the snapshot name. */
3037 name = rbd_snap_name(rbd_dev, rbd_dev->spec->snap_id);
3039 rbd_warn(rbd_dev, "no snapshot with id %llu",
3040 rbd_dev->spec->snap_id); /* Really a BUG() */
3044 rbd_dev->spec->snap_name = kstrdup(name, GFP_KERNEL);
3045 if(!rbd_dev->spec->snap_name)
3051 kfree(rbd_dev->spec->pool_name);
3052 rbd_dev->spec->pool_name = NULL;
3057 static int rbd_dev_v2_snap_context(struct rbd_device *rbd_dev, u64 *ver)
3066 struct ceph_snap_context *snapc;
3070 * We'll need room for the seq value (maximum snapshot id),
3071 * snapshot count, and array of that many snapshot ids.
3072 * For now we have a fixed upper limit on the number we're
3073 * prepared to receive.
3075 size = sizeof (__le64) + sizeof (__le32) +
3076 RBD_MAX_SNAP_COUNT * sizeof (__le64);
3077 reply_buf = kzalloc(size, GFP_KERNEL);
3081 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
3082 "rbd", "get_snapcontext",
3084 reply_buf, size, ver);
3085 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
3091 end = (char *) reply_buf + size;
3092 ceph_decode_64_safe(&p, end, seq, out);
3093 ceph_decode_32_safe(&p, end, snap_count, out);
3096 * Make sure the reported number of snapshot ids wouldn't go
3097 * beyond the end of our buffer. But before checking that,
3098 * make sure the computed size of the snapshot context we
3099 * allocate is representable in a size_t.
3101 if (snap_count > (SIZE_MAX - sizeof (struct ceph_snap_context))
3106 if (!ceph_has_room(&p, end, snap_count * sizeof (__le64)))
3109 size = sizeof (struct ceph_snap_context) +
3110 snap_count * sizeof (snapc->snaps[0]);
3111 snapc = kmalloc(size, GFP_KERNEL);
3117 atomic_set(&snapc->nref, 1);
3119 snapc->num_snaps = snap_count;
3120 for (i = 0; i < snap_count; i++)
3121 snapc->snaps[i] = ceph_decode_64(&p);
3123 rbd_dev->header.snapc = snapc;
3125 dout(" snap context seq = %llu, snap_count = %u\n",
3126 (unsigned long long) seq, (unsigned int) snap_count);
3134 static char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev, u32 which)
3144 size = sizeof (__le32) + RBD_MAX_SNAP_NAME_LEN;
3145 reply_buf = kmalloc(size, GFP_KERNEL);
3147 return ERR_PTR(-ENOMEM);
3149 snap_id = cpu_to_le64(rbd_dev->header.snapc->snaps[which]);
3150 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
3151 "rbd", "get_snapshot_name",
3152 (char *) &snap_id, sizeof (snap_id),
3153 reply_buf, size, NULL);
3154 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
3159 end = (char *) reply_buf + size;
3160 snap_name = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL);
3161 if (IS_ERR(snap_name)) {
3162 ret = PTR_ERR(snap_name);
3165 dout(" snap_id 0x%016llx snap_name = %s\n",
3166 (unsigned long long) le64_to_cpu(snap_id), snap_name);
3174 return ERR_PTR(ret);
3177 static char *rbd_dev_v2_snap_info(struct rbd_device *rbd_dev, u32 which,
3178 u64 *snap_size, u64 *snap_features)
3184 snap_id = rbd_dev->header.snapc->snaps[which];
3185 ret = _rbd_dev_v2_snap_size(rbd_dev, snap_id, &order, snap_size);
3187 return ERR_PTR(ret);
3188 ret = _rbd_dev_v2_snap_features(rbd_dev, snap_id, snap_features);
3190 return ERR_PTR(ret);
3192 return rbd_dev_v2_snap_name(rbd_dev, which);
3195 static char *rbd_dev_snap_info(struct rbd_device *rbd_dev, u32 which,
3196 u64 *snap_size, u64 *snap_features)
3198 if (rbd_dev->image_format == 1)
3199 return rbd_dev_v1_snap_info(rbd_dev, which,
3200 snap_size, snap_features);
3201 if (rbd_dev->image_format == 2)
3202 return rbd_dev_v2_snap_info(rbd_dev, which,
3203 snap_size, snap_features);
3204 return ERR_PTR(-EINVAL);
3207 static int rbd_dev_v2_refresh(struct rbd_device *rbd_dev, u64 *hver)
3212 down_write(&rbd_dev->header_rwsem);
3214 /* Grab old order first, to see if it changes */
3216 obj_order = rbd_dev->header.obj_order,
3217 ret = rbd_dev_v2_image_size(rbd_dev);
3220 if (rbd_dev->header.obj_order != obj_order) {
3224 rbd_update_mapping_size(rbd_dev);
3226 ret = rbd_dev_v2_snap_context(rbd_dev, hver);
3227 dout("rbd_dev_v2_snap_context returned %d\n", ret);
3230 ret = rbd_dev_snaps_update(rbd_dev);
3231 dout("rbd_dev_snaps_update returned %d\n", ret);
3234 ret = rbd_dev_snaps_register(rbd_dev);
3235 dout("rbd_dev_snaps_register returned %d\n", ret);
3237 up_write(&rbd_dev->header_rwsem);
3243 * Scan the rbd device's current snapshot list and compare it to the
3244 * newly-received snapshot context. Remove any existing snapshots
3245 * not present in the new snapshot context. Add a new snapshot for
3246 * any snaphots in the snapshot context not in the current list.
3247 * And verify there are no changes to snapshots we already know
3250 * Assumes the snapshots in the snapshot context are sorted by
3251 * snapshot id, highest id first. (Snapshots in the rbd_dev's list
3252 * are also maintained in that order.)
3254 static int rbd_dev_snaps_update(struct rbd_device *rbd_dev)
3256 struct ceph_snap_context *snapc = rbd_dev->header.snapc;
3257 const u32 snap_count = snapc->num_snaps;
3258 struct list_head *head = &rbd_dev->snaps;
3259 struct list_head *links = head->next;
3262 dout("%s: snap count is %u\n", __func__, (unsigned int) snap_count);
3263 while (index < snap_count || links != head) {
3265 struct rbd_snap *snap;
3268 u64 snap_features = 0;
3270 snap_id = index < snap_count ? snapc->snaps[index]
3272 snap = links != head ? list_entry(links, struct rbd_snap, node)
3274 rbd_assert(!snap || snap->id != CEPH_NOSNAP);
3276 if (snap_id == CEPH_NOSNAP || (snap && snap->id > snap_id)) {
3277 struct list_head *next = links->next;
3280 * A previously-existing snapshot is not in
3281 * the new snap context.
3283 * If the now missing snapshot is the one the
3284 * image is mapped to, clear its exists flag
3285 * so we can avoid sending any more requests
3288 if (rbd_dev->spec->snap_id == snap->id)
3289 clear_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags);
3290 rbd_remove_snap_dev(snap);
3291 dout("%ssnap id %llu has been removed\n",
3292 rbd_dev->spec->snap_id == snap->id ?
3294 (unsigned long long) snap->id);
3296 /* Done with this list entry; advance */
3302 snap_name = rbd_dev_snap_info(rbd_dev, index,
3303 &snap_size, &snap_features);
3304 if (IS_ERR(snap_name))
3305 return PTR_ERR(snap_name);
3307 dout("entry %u: snap_id = %llu\n", (unsigned int) snap_count,
3308 (unsigned long long) snap_id);
3309 if (!snap || (snap_id != CEPH_NOSNAP && snap->id < snap_id)) {
3310 struct rbd_snap *new_snap;
3312 /* We haven't seen this snapshot before */
3314 new_snap = __rbd_add_snap_dev(rbd_dev, snap_name,
3315 snap_id, snap_size, snap_features);
3316 if (IS_ERR(new_snap)) {
3317 int err = PTR_ERR(new_snap);
3319 dout(" failed to add dev, error %d\n", err);
3324 /* New goes before existing, or at end of list */
3326 dout(" added dev%s\n", snap ? "" : " at end\n");
3328 list_add_tail(&new_snap->node, &snap->node);
3330 list_add_tail(&new_snap->node, head);
3332 /* Already have this one */
3334 dout(" already present\n");
3336 rbd_assert(snap->size == snap_size);
3337 rbd_assert(!strcmp(snap->name, snap_name));
3338 rbd_assert(snap->features == snap_features);
3340 /* Done with this list entry; advance */
3342 links = links->next;
3345 /* Advance to the next entry in the snapshot context */
3349 dout("%s: done\n", __func__);
3355 * Scan the list of snapshots and register the devices for any that
3356 * have not already been registered.
3358 static int rbd_dev_snaps_register(struct rbd_device *rbd_dev)
3360 struct rbd_snap *snap;
3363 dout("%s:\n", __func__);
3364 if (WARN_ON(!device_is_registered(&rbd_dev->dev)))
3367 list_for_each_entry(snap, &rbd_dev->snaps, node) {
3368 if (!rbd_snap_registered(snap)) {
3369 ret = rbd_register_snap_dev(snap, &rbd_dev->dev);
3374 dout("%s: returning %d\n", __func__, ret);
3379 static int rbd_bus_add_dev(struct rbd_device *rbd_dev)
3384 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
3386 dev = &rbd_dev->dev;
3387 dev->bus = &rbd_bus_type;
3388 dev->type = &rbd_device_type;
3389 dev->parent = &rbd_root_dev;
3390 dev->release = rbd_dev_release;
3391 dev_set_name(dev, "%d", rbd_dev->dev_id);
3392 ret = device_register(dev);
3394 mutex_unlock(&ctl_mutex);
3399 static void rbd_bus_del_dev(struct rbd_device *rbd_dev)
3401 device_unregister(&rbd_dev->dev);
3404 static atomic64_t rbd_dev_id_max = ATOMIC64_INIT(0);
3407 * Get a unique rbd identifier for the given new rbd_dev, and add
3408 * the rbd_dev to the global list. The minimum rbd id is 1.
3410 static void rbd_dev_id_get(struct rbd_device *rbd_dev)
3412 rbd_dev->dev_id = atomic64_inc_return(&rbd_dev_id_max);
3414 spin_lock(&rbd_dev_list_lock);
3415 list_add_tail(&rbd_dev->node, &rbd_dev_list);
3416 spin_unlock(&rbd_dev_list_lock);
3417 dout("rbd_dev %p given dev id %llu\n", rbd_dev,
3418 (unsigned long long) rbd_dev->dev_id);
3422 * Remove an rbd_dev from the global list, and record that its
3423 * identifier is no longer in use.
3425 static void rbd_dev_id_put(struct rbd_device *rbd_dev)
3427 struct list_head *tmp;
3428 int rbd_id = rbd_dev->dev_id;
3431 rbd_assert(rbd_id > 0);
3433 dout("rbd_dev %p released dev id %llu\n", rbd_dev,
3434 (unsigned long long) rbd_dev->dev_id);
3435 spin_lock(&rbd_dev_list_lock);
3436 list_del_init(&rbd_dev->node);
3439 * If the id being "put" is not the current maximum, there
3440 * is nothing special we need to do.
3442 if (rbd_id != atomic64_read(&rbd_dev_id_max)) {
3443 spin_unlock(&rbd_dev_list_lock);
3448 * We need to update the current maximum id. Search the
3449 * list to find out what it is. We're more likely to find
3450 * the maximum at the end, so search the list backward.
3453 list_for_each_prev(tmp, &rbd_dev_list) {
3454 struct rbd_device *rbd_dev;
3456 rbd_dev = list_entry(tmp, struct rbd_device, node);
3457 if (rbd_dev->dev_id > max_id)
3458 max_id = rbd_dev->dev_id;
3460 spin_unlock(&rbd_dev_list_lock);
3463 * The max id could have been updated by rbd_dev_id_get(), in
3464 * which case it now accurately reflects the new maximum.
3465 * Be careful not to overwrite the maximum value in that
3468 atomic64_cmpxchg(&rbd_dev_id_max, rbd_id, max_id);
3469 dout(" max dev id has been reset\n");
3473 * Skips over white space at *buf, and updates *buf to point to the
3474 * first found non-space character (if any). Returns the length of
3475 * the token (string of non-white space characters) found. Note
3476 * that *buf must be terminated with '\0'.
3478 static inline size_t next_token(const char **buf)
3481 * These are the characters that produce nonzero for
3482 * isspace() in the "C" and "POSIX" locales.
3484 const char *spaces = " \f\n\r\t\v";
3486 *buf += strspn(*buf, spaces); /* Find start of token */
3488 return strcspn(*buf, spaces); /* Return token length */
3492 * Finds the next token in *buf, and if the provided token buffer is
3493 * big enough, copies the found token into it. The result, if
3494 * copied, is guaranteed to be terminated with '\0'. Note that *buf
3495 * must be terminated with '\0' on entry.
3497 * Returns the length of the token found (not including the '\0').
3498 * Return value will be 0 if no token is found, and it will be >=
3499 * token_size if the token would not fit.
3501 * The *buf pointer will be updated to point beyond the end of the
3502 * found token. Note that this occurs even if the token buffer is
3503 * too small to hold it.
3505 static inline size_t copy_token(const char **buf,
3511 len = next_token(buf);
3512 if (len < token_size) {
3513 memcpy(token, *buf, len);
3514 *(token + len) = '\0';
3522 * Finds the next token in *buf, dynamically allocates a buffer big
3523 * enough to hold a copy of it, and copies the token into the new
3524 * buffer. The copy is guaranteed to be terminated with '\0'. Note
3525 * that a duplicate buffer is created even for a zero-length token.
3527 * Returns a pointer to the newly-allocated duplicate, or a null
3528 * pointer if memory for the duplicate was not available. If
3529 * the lenp argument is a non-null pointer, the length of the token
3530 * (not including the '\0') is returned in *lenp.
3532 * If successful, the *buf pointer will be updated to point beyond
3533 * the end of the found token.
3535 * Note: uses GFP_KERNEL for allocation.
3537 static inline char *dup_token(const char **buf, size_t *lenp)
3542 len = next_token(buf);
3543 dup = kmemdup(*buf, len + 1, GFP_KERNEL);
3546 *(dup + len) = '\0';
3556 * Parse the options provided for an "rbd add" (i.e., rbd image
3557 * mapping) request. These arrive via a write to /sys/bus/rbd/add,
3558 * and the data written is passed here via a NUL-terminated buffer.
3559 * Returns 0 if successful or an error code otherwise.
3561 * The information extracted from these options is recorded in
3562 * the other parameters which return dynamically-allocated
3565 * The address of a pointer that will refer to a ceph options
3566 * structure. Caller must release the returned pointer using
3567 * ceph_destroy_options() when it is no longer needed.
3569 * Address of an rbd options pointer. Fully initialized by
3570 * this function; caller must release with kfree().
3572 * Address of an rbd image specification pointer. Fully
3573 * initialized by this function based on parsed options.
3574 * Caller must release with rbd_spec_put().
3576 * The options passed take this form:
3577 * <mon_addrs> <options> <pool_name> <image_name> [<snap_id>]
3580 * A comma-separated list of one or more monitor addresses.
3581 * A monitor address is an ip address, optionally followed
3582 * by a port number (separated by a colon).
3583 * I.e.: ip1[:port1][,ip2[:port2]...]
3585 * A comma-separated list of ceph and/or rbd options.
3587 * The name of the rados pool containing the rbd image.
3589 * The name of the image in that pool to map.
3591 * An optional snapshot id. If provided, the mapping will
3592 * present data from the image at the time that snapshot was
3593 * created. The image head is used if no snapshot id is
3594 * provided. Snapshot mappings are always read-only.
3596 static int rbd_add_parse_args(const char *buf,
3597 struct ceph_options **ceph_opts,
3598 struct rbd_options **opts,
3599 struct rbd_spec **rbd_spec)
3603 const char *mon_addrs;
3604 size_t mon_addrs_size;
3605 struct rbd_spec *spec = NULL;
3606 struct rbd_options *rbd_opts = NULL;
3607 struct ceph_options *copts;
3610 /* The first four tokens are required */
3612 len = next_token(&buf);
3614 rbd_warn(NULL, "no monitor address(es) provided");
3618 mon_addrs_size = len + 1;
3622 options = dup_token(&buf, NULL);
3626 rbd_warn(NULL, "no options provided");
3630 spec = rbd_spec_alloc();
3634 spec->pool_name = dup_token(&buf, NULL);
3635 if (!spec->pool_name)
3637 if (!*spec->pool_name) {
3638 rbd_warn(NULL, "no pool name provided");
3642 spec->image_name = dup_token(&buf, NULL);
3643 if (!spec->image_name)
3645 if (!*spec->image_name) {
3646 rbd_warn(NULL, "no image name provided");
3651 * Snapshot name is optional; default is to use "-"
3652 * (indicating the head/no snapshot).
3654 len = next_token(&buf);
3656 buf = RBD_SNAP_HEAD_NAME; /* No snapshot supplied */
3657 len = sizeof (RBD_SNAP_HEAD_NAME) - 1;
3658 } else if (len > RBD_MAX_SNAP_NAME_LEN) {
3659 ret = -ENAMETOOLONG;
3662 spec->snap_name = kmemdup(buf, len + 1, GFP_KERNEL);
3663 if (!spec->snap_name)
3665 *(spec->snap_name + len) = '\0';
3667 /* Initialize all rbd options to the defaults */
3669 rbd_opts = kzalloc(sizeof (*rbd_opts), GFP_KERNEL);
3673 rbd_opts->read_only = RBD_READ_ONLY_DEFAULT;
3675 copts = ceph_parse_options(options, mon_addrs,
3676 mon_addrs + mon_addrs_size - 1,
3677 parse_rbd_opts_token, rbd_opts);
3678 if (IS_ERR(copts)) {
3679 ret = PTR_ERR(copts);
3700 * An rbd format 2 image has a unique identifier, distinct from the
3701 * name given to it by the user. Internally, that identifier is
3702 * what's used to specify the names of objects related to the image.
3704 * A special "rbd id" object is used to map an rbd image name to its
3705 * id. If that object doesn't exist, then there is no v2 rbd image
3706 * with the supplied name.
3708 * This function will record the given rbd_dev's image_id field if
3709 * it can be determined, and in that case will return 0. If any
3710 * errors occur a negative errno will be returned and the rbd_dev's
3711 * image_id field will be unchanged (and should be NULL).
3713 static int rbd_dev_image_id(struct rbd_device *rbd_dev)
3722 * When probing a parent image, the image id is already
3723 * known (and the image name likely is not). There's no
3724 * need to fetch the image id again in this case.
3726 if (rbd_dev->spec->image_id)
3730 * First, see if the format 2 image id file exists, and if
3731 * so, get the image's persistent id from it.
3733 size = sizeof (RBD_ID_PREFIX) + strlen(rbd_dev->spec->image_name);
3734 object_name = kmalloc(size, GFP_NOIO);
3737 sprintf(object_name, "%s%s", RBD_ID_PREFIX, rbd_dev->spec->image_name);
3738 dout("rbd id object name is %s\n", object_name);
3740 /* Response will be an encoded string, which includes a length */
3742 size = sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX;
3743 response = kzalloc(size, GFP_NOIO);
3749 ret = rbd_obj_method_sync(rbd_dev, object_name,
3752 response, RBD_IMAGE_ID_LEN_MAX, NULL);
3753 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
3758 rbd_dev->spec->image_id = ceph_extract_encoded_string(&p,
3759 p + RBD_IMAGE_ID_LEN_MAX,
3761 if (IS_ERR(rbd_dev->spec->image_id)) {
3762 ret = PTR_ERR(rbd_dev->spec->image_id);
3763 rbd_dev->spec->image_id = NULL;
3765 dout("image_id is %s\n", rbd_dev->spec->image_id);
3774 static int rbd_dev_v1_probe(struct rbd_device *rbd_dev)
3779 /* Version 1 images have no id; empty string is used */
3781 rbd_dev->spec->image_id = kstrdup("", GFP_KERNEL);
3782 if (!rbd_dev->spec->image_id)
3785 /* Record the header object name for this rbd image. */
3787 size = strlen(rbd_dev->spec->image_name) + sizeof (RBD_SUFFIX);
3788 rbd_dev->header_name = kmalloc(size, GFP_KERNEL);
3789 if (!rbd_dev->header_name) {
3793 sprintf(rbd_dev->header_name, "%s%s",
3794 rbd_dev->spec->image_name, RBD_SUFFIX);
3796 /* Populate rbd image metadata */
3798 ret = rbd_read_header(rbd_dev, &rbd_dev->header);
3802 /* Version 1 images have no parent (no layering) */
3804 rbd_dev->parent_spec = NULL;
3805 rbd_dev->parent_overlap = 0;
3807 rbd_dev->image_format = 1;
3809 dout("discovered version 1 image, header name is %s\n",
3810 rbd_dev->header_name);
3815 kfree(rbd_dev->header_name);
3816 rbd_dev->header_name = NULL;
3817 kfree(rbd_dev->spec->image_id);
3818 rbd_dev->spec->image_id = NULL;
3823 static int rbd_dev_v2_probe(struct rbd_device *rbd_dev)
3830 * Image id was filled in by the caller. Record the header
3831 * object name for this rbd image.
3833 size = sizeof (RBD_HEADER_PREFIX) + strlen(rbd_dev->spec->image_id);
3834 rbd_dev->header_name = kmalloc(size, GFP_KERNEL);
3835 if (!rbd_dev->header_name)
3837 sprintf(rbd_dev->header_name, "%s%s",
3838 RBD_HEADER_PREFIX, rbd_dev->spec->image_id);
3840 /* Get the size and object order for the image */
3842 ret = rbd_dev_v2_image_size(rbd_dev);
3846 /* Get the object prefix (a.k.a. block_name) for the image */
3848 ret = rbd_dev_v2_object_prefix(rbd_dev);
3852 /* Get the and check features for the image */
3854 ret = rbd_dev_v2_features(rbd_dev);
3858 /* If the image supports layering, get the parent info */
3860 if (rbd_dev->header.features & RBD_FEATURE_LAYERING) {
3861 ret = rbd_dev_v2_parent_info(rbd_dev);
3866 /* crypto and compression type aren't (yet) supported for v2 images */
3868 rbd_dev->header.crypt_type = 0;
3869 rbd_dev->header.comp_type = 0;
3871 /* Get the snapshot context, plus the header version */
3873 ret = rbd_dev_v2_snap_context(rbd_dev, &ver);
3876 rbd_dev->header.obj_version = ver;
3878 rbd_dev->image_format = 2;
3880 dout("discovered version 2 image, header name is %s\n",
3881 rbd_dev->header_name);
3885 rbd_dev->parent_overlap = 0;
3886 rbd_spec_put(rbd_dev->parent_spec);
3887 rbd_dev->parent_spec = NULL;
3888 kfree(rbd_dev->header_name);
3889 rbd_dev->header_name = NULL;
3890 kfree(rbd_dev->header.object_prefix);
3891 rbd_dev->header.object_prefix = NULL;
3896 static int rbd_dev_probe_finish(struct rbd_device *rbd_dev)
3900 /* no need to lock here, as rbd_dev is not registered yet */
3901 ret = rbd_dev_snaps_update(rbd_dev);
3905 ret = rbd_dev_probe_update_spec(rbd_dev);
3909 ret = rbd_dev_set_mapping(rbd_dev);
3913 /* generate unique id: find highest unique id, add one */
3914 rbd_dev_id_get(rbd_dev);
3916 /* Fill in the device name, now that we have its id. */
3917 BUILD_BUG_ON(DEV_NAME_LEN
3918 < sizeof (RBD_DRV_NAME) + MAX_INT_FORMAT_WIDTH);
3919 sprintf(rbd_dev->name, "%s%d", RBD_DRV_NAME, rbd_dev->dev_id);
3921 /* Get our block major device number. */
3923 ret = register_blkdev(0, rbd_dev->name);
3926 rbd_dev->major = ret;
3928 /* Set up the blkdev mapping. */
3930 ret = rbd_init_disk(rbd_dev);
3932 goto err_out_blkdev;
3934 ret = rbd_bus_add_dev(rbd_dev);
3939 * At this point cleanup in the event of an error is the job
3940 * of the sysfs code (initiated by rbd_bus_del_dev()).
3942 down_write(&rbd_dev->header_rwsem);
3943 ret = rbd_dev_snaps_register(rbd_dev);
3944 up_write(&rbd_dev->header_rwsem);
3948 ret = rbd_dev_header_watch_sync(rbd_dev, 1);
3952 /* Everything's ready. Announce the disk to the world. */
3954 add_disk(rbd_dev->disk);
3956 pr_info("%s: added with size 0x%llx\n", rbd_dev->disk->disk_name,
3957 (unsigned long long) rbd_dev->mapping.size);
3961 /* this will also clean up rest of rbd_dev stuff */
3963 rbd_bus_del_dev(rbd_dev);
3967 rbd_free_disk(rbd_dev);
3969 unregister_blkdev(rbd_dev->major, rbd_dev->name);
3971 rbd_dev_id_put(rbd_dev);
3973 rbd_remove_all_snaps(rbd_dev);
3979 * Probe for the existence of the header object for the given rbd
3980 * device. For format 2 images this includes determining the image
3983 static int rbd_dev_probe(struct rbd_device *rbd_dev)
3988 * Get the id from the image id object. If it's not a
3989 * format 2 image, we'll get ENOENT back, and we'll assume
3990 * it's a format 1 image.
3992 ret = rbd_dev_image_id(rbd_dev);
3994 ret = rbd_dev_v1_probe(rbd_dev);
3996 ret = rbd_dev_v2_probe(rbd_dev);
3998 dout("probe failed, returning %d\n", ret);
4003 ret = rbd_dev_probe_finish(rbd_dev);
4005 rbd_header_free(&rbd_dev->header);
4010 static ssize_t rbd_add(struct bus_type *bus,
4014 struct rbd_device *rbd_dev = NULL;
4015 struct ceph_options *ceph_opts = NULL;
4016 struct rbd_options *rbd_opts = NULL;
4017 struct rbd_spec *spec = NULL;
4018 struct rbd_client *rbdc;
4019 struct ceph_osd_client *osdc;
4022 if (!try_module_get(THIS_MODULE))
4025 /* parse add command */
4026 rc = rbd_add_parse_args(buf, &ceph_opts, &rbd_opts, &spec);
4028 goto err_out_module;
4030 rbdc = rbd_get_client(ceph_opts);
4035 ceph_opts = NULL; /* rbd_dev client now owns this */
4038 osdc = &rbdc->client->osdc;
4039 rc = ceph_pg_poolid_by_name(osdc->osdmap, spec->pool_name);
4041 goto err_out_client;
4042 spec->pool_id = (u64) rc;
4044 /* The ceph file layout needs to fit pool id in 32 bits */
4046 if (WARN_ON(spec->pool_id > (u64) U32_MAX)) {
4048 goto err_out_client;
4051 rbd_dev = rbd_dev_create(rbdc, spec);
4053 goto err_out_client;
4054 rbdc = NULL; /* rbd_dev now owns this */
4055 spec = NULL; /* rbd_dev now owns this */
4057 rbd_dev->mapping.read_only = rbd_opts->read_only;
4059 rbd_opts = NULL; /* done with this */
4061 rc = rbd_dev_probe(rbd_dev);
4063 goto err_out_rbd_dev;
4067 rbd_dev_destroy(rbd_dev);
4069 rbd_put_client(rbdc);
4072 ceph_destroy_options(ceph_opts);
4076 module_put(THIS_MODULE);
4078 dout("Error adding device %s\n", buf);
4080 return (ssize_t) rc;
4083 static struct rbd_device *__rbd_get_dev(unsigned long dev_id)
4085 struct list_head *tmp;
4086 struct rbd_device *rbd_dev;
4088 spin_lock(&rbd_dev_list_lock);
4089 list_for_each(tmp, &rbd_dev_list) {
4090 rbd_dev = list_entry(tmp, struct rbd_device, node);
4091 if (rbd_dev->dev_id == dev_id) {
4092 spin_unlock(&rbd_dev_list_lock);
4096 spin_unlock(&rbd_dev_list_lock);
4100 static void rbd_dev_release(struct device *dev)
4102 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
4104 if (rbd_dev->watch_event)
4105 rbd_dev_header_watch_sync(rbd_dev, 0);
4107 /* clean up and free blkdev */
4108 rbd_free_disk(rbd_dev);
4109 unregister_blkdev(rbd_dev->major, rbd_dev->name);
4111 /* release allocated disk header fields */
4112 rbd_header_free(&rbd_dev->header);
4114 /* done with the id, and with the rbd_dev */
4115 rbd_dev_id_put(rbd_dev);
4116 rbd_assert(rbd_dev->rbd_client != NULL);
4117 rbd_dev_destroy(rbd_dev);
4119 /* release module ref */
4120 module_put(THIS_MODULE);
4123 static ssize_t rbd_remove(struct bus_type *bus,
4127 struct rbd_device *rbd_dev = NULL;
4132 rc = strict_strtoul(buf, 10, &ul);
4136 /* convert to int; abort if we lost anything in the conversion */
4137 target_id = (int) ul;
4138 if (target_id != ul)
4141 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
4143 rbd_dev = __rbd_get_dev(target_id);
4149 spin_lock_irq(&rbd_dev->lock);
4150 if (rbd_dev->open_count)
4153 set_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags);
4154 spin_unlock_irq(&rbd_dev->lock);
4158 rbd_remove_all_snaps(rbd_dev);
4159 rbd_bus_del_dev(rbd_dev);
4162 mutex_unlock(&ctl_mutex);
4168 * create control files in sysfs
4171 static int rbd_sysfs_init(void)
4175 ret = device_register(&rbd_root_dev);
4179 ret = bus_register(&rbd_bus_type);
4181 device_unregister(&rbd_root_dev);
4186 static void rbd_sysfs_cleanup(void)
4188 bus_unregister(&rbd_bus_type);
4189 device_unregister(&rbd_root_dev);
4192 static int __init rbd_init(void)
4196 if (!libceph_compatible(NULL)) {
4197 rbd_warn(NULL, "libceph incompatibility (quitting)");
4201 rc = rbd_sysfs_init();
4204 pr_info("loaded " RBD_DRV_NAME_LONG "\n");
4208 static void __exit rbd_exit(void)
4210 rbd_sysfs_cleanup();
4213 module_init(rbd_init);
4214 module_exit(rbd_exit);
4216 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
4217 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
4218 MODULE_DESCRIPTION("rados block device");
4220 /* following authorship retained from original osdblk.c */
4221 MODULE_AUTHOR("Jeff Garzik <jeff@garzik.org>");
4223 MODULE_LICENSE("GPL");