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_request_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 struct ceph_osd_request *rbd_osd_req_create(
1315 struct rbd_device *rbd_dev,
1317 struct rbd_obj_request *obj_request,
1318 struct ceph_osd_req_op *op)
1320 struct rbd_img_request *img_request = obj_request->img_request;
1321 struct ceph_snap_context *snapc = NULL;
1322 struct ceph_osd_client *osdc;
1323 struct ceph_osd_request *osd_req;
1324 struct ceph_osd_data *osd_data;
1325 struct timespec now;
1326 struct timespec *mtime;
1327 u64 snap_id = CEPH_NOSNAP;
1328 u64 offset = obj_request->offset;
1329 u64 length = obj_request->length;
1332 rbd_assert(img_request->write_request == write_request);
1333 if (img_request->write_request)
1334 snapc = img_request->snapc;
1336 snap_id = img_request->snap_id;
1339 /* Allocate and initialize the request, for the single op */
1341 osdc = &rbd_dev->rbd_client->client->osdc;
1342 osd_req = ceph_osdc_alloc_request(osdc, snapc, 1, false, GFP_ATOMIC);
1344 return NULL; /* ENOMEM */
1345 osd_data = write_request ? &osd_req->r_data_out : &osd_req->r_data_in;
1347 rbd_assert(obj_request_type_valid(obj_request->type));
1348 switch (obj_request->type) {
1349 case OBJ_REQUEST_NODATA:
1350 break; /* Nothing to do */
1351 case OBJ_REQUEST_BIO:
1352 rbd_assert(obj_request->bio_list != NULL);
1353 osd_data->type = CEPH_OSD_DATA_TYPE_BIO;
1354 osd_data->bio = obj_request->bio_list;
1355 osd_data->bio_length = obj_request->length;
1357 case OBJ_REQUEST_PAGES:
1358 osd_data->type = CEPH_OSD_DATA_TYPE_PAGES;
1359 osd_data->pages = obj_request->pages;
1360 osd_data->length = obj_request->length;
1361 osd_data->alignment = offset & ~PAGE_MASK;
1362 osd_data->pages_from_pool = false;
1363 osd_data->own_pages = false;
1367 if (write_request) {
1368 osd_req->r_flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK;
1372 osd_req->r_flags = CEPH_OSD_FLAG_READ;
1373 mtime = NULL; /* not needed for reads */
1374 offset = 0; /* These are not used... */
1375 length = 0; /* ...for osd read requests */
1378 osd_req->r_callback = rbd_osd_req_callback;
1379 osd_req->r_priv = obj_request;
1381 osd_req->r_oid_len = strlen(obj_request->object_name);
1382 rbd_assert(osd_req->r_oid_len < sizeof (osd_req->r_oid));
1383 memcpy(osd_req->r_oid, obj_request->object_name, osd_req->r_oid_len);
1385 osd_req->r_file_layout = rbd_dev->layout; /* struct */
1387 /* osd_req will get its own reference to snapc (if non-null) */
1389 ceph_osdc_build_request(osd_req, offset, 1, op,
1390 snapc, snap_id, mtime);
1395 static void rbd_osd_req_destroy(struct ceph_osd_request *osd_req)
1397 ceph_osdc_put_request(osd_req);
1400 /* object_name is assumed to be a non-null pointer and NUL-terminated */
1402 static struct rbd_obj_request *rbd_obj_request_create(const char *object_name,
1403 u64 offset, u64 length,
1404 enum obj_request_type type)
1406 struct rbd_obj_request *obj_request;
1410 rbd_assert(obj_request_type_valid(type));
1412 size = strlen(object_name) + 1;
1413 obj_request = kzalloc(sizeof (*obj_request) + size, GFP_KERNEL);
1417 name = (char *)(obj_request + 1);
1418 obj_request->object_name = memcpy(name, object_name, size);
1419 obj_request->offset = offset;
1420 obj_request->length = length;
1421 obj_request->which = BAD_WHICH;
1422 obj_request->type = type;
1423 INIT_LIST_HEAD(&obj_request->links);
1424 obj_request_done_init(obj_request);
1425 init_completion(&obj_request->completion);
1426 kref_init(&obj_request->kref);
1428 dout("%s: \"%s\" %llu/%llu %d -> obj %p\n", __func__, object_name,
1429 offset, length, (int)type, obj_request);
1434 static void rbd_obj_request_destroy(struct kref *kref)
1436 struct rbd_obj_request *obj_request;
1438 obj_request = container_of(kref, struct rbd_obj_request, kref);
1440 dout("%s: obj %p\n", __func__, obj_request);
1442 rbd_assert(obj_request->img_request == NULL);
1443 rbd_assert(obj_request->which == BAD_WHICH);
1445 if (obj_request->osd_req)
1446 rbd_osd_req_destroy(obj_request->osd_req);
1448 rbd_assert(obj_request_type_valid(obj_request->type));
1449 switch (obj_request->type) {
1450 case OBJ_REQUEST_NODATA:
1451 break; /* Nothing to do */
1452 case OBJ_REQUEST_BIO:
1453 if (obj_request->bio_list)
1454 bio_chain_put(obj_request->bio_list);
1456 case OBJ_REQUEST_PAGES:
1457 if (obj_request->pages)
1458 ceph_release_page_vector(obj_request->pages,
1459 obj_request->page_count);
1467 * Caller is responsible for filling in the list of object requests
1468 * that comprises the image request, and the Linux request pointer
1469 * (if there is one).
1471 static struct rbd_img_request *rbd_img_request_create(
1472 struct rbd_device *rbd_dev,
1473 u64 offset, u64 length,
1476 struct rbd_img_request *img_request;
1477 struct ceph_snap_context *snapc = NULL;
1479 img_request = kmalloc(sizeof (*img_request), GFP_ATOMIC);
1483 if (write_request) {
1484 down_read(&rbd_dev->header_rwsem);
1485 snapc = ceph_get_snap_context(rbd_dev->header.snapc);
1486 up_read(&rbd_dev->header_rwsem);
1487 if (WARN_ON(!snapc)) {
1489 return NULL; /* Shouldn't happen */
1493 img_request->rq = NULL;
1494 img_request->rbd_dev = rbd_dev;
1495 img_request->offset = offset;
1496 img_request->length = length;
1497 img_request->write_request = write_request;
1499 img_request->snapc = snapc;
1501 img_request->snap_id = rbd_dev->spec->snap_id;
1502 spin_lock_init(&img_request->completion_lock);
1503 img_request->next_completion = 0;
1504 img_request->callback = NULL;
1505 img_request->obj_request_count = 0;
1506 INIT_LIST_HEAD(&img_request->obj_requests);
1507 kref_init(&img_request->kref);
1509 rbd_img_request_get(img_request); /* Avoid a warning */
1510 rbd_img_request_put(img_request); /* TEMPORARY */
1512 dout("%s: rbd_dev %p %s %llu/%llu -> img %p\n", __func__, rbd_dev,
1513 write_request ? "write" : "read", offset, length,
1519 static void rbd_img_request_destroy(struct kref *kref)
1521 struct rbd_img_request *img_request;
1522 struct rbd_obj_request *obj_request;
1523 struct rbd_obj_request *next_obj_request;
1525 img_request = container_of(kref, struct rbd_img_request, kref);
1527 dout("%s: img %p\n", __func__, img_request);
1529 for_each_obj_request_safe(img_request, obj_request, next_obj_request)
1530 rbd_img_obj_request_del(img_request, obj_request);
1531 rbd_assert(img_request->obj_request_count == 0);
1533 if (img_request->write_request)
1534 ceph_put_snap_context(img_request->snapc);
1539 static int rbd_img_request_fill_bio(struct rbd_img_request *img_request,
1540 struct bio *bio_list)
1542 struct rbd_device *rbd_dev = img_request->rbd_dev;
1543 struct rbd_obj_request *obj_request = NULL;
1544 struct rbd_obj_request *next_obj_request;
1545 unsigned int bio_offset;
1550 dout("%s: img %p bio %p\n", __func__, img_request, bio_list);
1552 opcode = img_request->write_request ? CEPH_OSD_OP_WRITE
1555 image_offset = img_request->offset;
1556 rbd_assert(image_offset == bio_list->bi_sector << SECTOR_SHIFT);
1557 resid = img_request->length;
1558 rbd_assert(resid > 0);
1560 const char *object_name;
1561 unsigned int clone_size;
1562 struct ceph_osd_req_op op;
1566 object_name = rbd_segment_name(rbd_dev, image_offset);
1569 offset = rbd_segment_offset(rbd_dev, image_offset);
1570 length = rbd_segment_length(rbd_dev, image_offset, resid);
1571 obj_request = rbd_obj_request_create(object_name,
1574 kfree(object_name); /* object request has its own copy */
1578 rbd_assert(length <= (u64) UINT_MAX);
1579 clone_size = (unsigned int) length;
1580 obj_request->bio_list = bio_chain_clone_range(&bio_list,
1581 &bio_offset, clone_size,
1583 if (!obj_request->bio_list)
1587 * Build up the op to use in building the osd
1588 * request. Note that the contents of the op are
1589 * copied by rbd_osd_req_create().
1591 osd_req_op_extent_init(&op, opcode, offset, length, 0, 0);
1592 obj_request->osd_req = rbd_osd_req_create(rbd_dev,
1593 img_request->write_request,
1595 if (!obj_request->osd_req)
1597 /* status and version are initially zero-filled */
1599 rbd_img_obj_request_add(img_request, obj_request);
1601 image_offset += length;
1608 rbd_obj_request_put(obj_request);
1610 for_each_obj_request_safe(img_request, obj_request, next_obj_request)
1611 rbd_obj_request_put(obj_request);
1616 static void rbd_img_obj_callback(struct rbd_obj_request *obj_request)
1618 struct rbd_img_request *img_request;
1619 u32 which = obj_request->which;
1622 img_request = obj_request->img_request;
1624 dout("%s: img %p obj %p\n", __func__, img_request, obj_request);
1625 rbd_assert(img_request != NULL);
1626 rbd_assert(img_request->rq != NULL);
1627 rbd_assert(img_request->obj_request_count > 0);
1628 rbd_assert(which != BAD_WHICH);
1629 rbd_assert(which < img_request->obj_request_count);
1630 rbd_assert(which >= img_request->next_completion);
1632 spin_lock_irq(&img_request->completion_lock);
1633 if (which != img_request->next_completion)
1636 for_each_obj_request_from(img_request, obj_request) {
1637 unsigned int xferred;
1641 rbd_assert(which < img_request->obj_request_count);
1643 if (!obj_request_done_test(obj_request))
1646 rbd_assert(obj_request->xferred <= (u64) UINT_MAX);
1647 xferred = (unsigned int) obj_request->xferred;
1648 result = (int) obj_request->result;
1650 rbd_warn(NULL, "obj_request %s result %d xferred %u\n",
1651 img_request->write_request ? "write" : "read",
1654 more = blk_end_request(img_request->rq, result, xferred);
1658 rbd_assert(more ^ (which == img_request->obj_request_count));
1659 img_request->next_completion = which;
1661 spin_unlock_irq(&img_request->completion_lock);
1664 rbd_img_request_complete(img_request);
1667 static int rbd_img_request_submit(struct rbd_img_request *img_request)
1669 struct rbd_device *rbd_dev = img_request->rbd_dev;
1670 struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
1671 struct rbd_obj_request *obj_request;
1672 struct rbd_obj_request *next_obj_request;
1674 dout("%s: img %p\n", __func__, img_request);
1675 for_each_obj_request_safe(img_request, obj_request, next_obj_request) {
1678 obj_request->callback = rbd_img_obj_callback;
1679 ret = rbd_obj_request_submit(osdc, obj_request);
1683 * The image request has its own reference to each
1684 * of its object requests, so we can safely drop the
1687 rbd_obj_request_put(obj_request);
1693 static int rbd_obj_notify_ack(struct rbd_device *rbd_dev,
1694 u64 ver, u64 notify_id)
1696 struct rbd_obj_request *obj_request;
1697 struct ceph_osd_req_op op;
1698 struct ceph_osd_client *osdc;
1701 obj_request = rbd_obj_request_create(rbd_dev->header_name, 0, 0,
1702 OBJ_REQUEST_NODATA);
1707 osd_req_op_watch_init(&op, CEPH_OSD_OP_NOTIFY_ACK, notify_id, ver, 0);
1708 obj_request->osd_req = rbd_osd_req_create(rbd_dev, false,
1710 if (!obj_request->osd_req)
1713 osdc = &rbd_dev->rbd_client->client->osdc;
1714 obj_request->callback = rbd_obj_request_put;
1715 ret = rbd_obj_request_submit(osdc, obj_request);
1718 rbd_obj_request_put(obj_request);
1723 static void rbd_watch_cb(u64 ver, u64 notify_id, u8 opcode, void *data)
1725 struct rbd_device *rbd_dev = (struct rbd_device *)data;
1732 dout("%s: \"%s\" notify_id %llu opcode %u\n", __func__,
1733 rbd_dev->header_name, (unsigned long long) notify_id,
1734 (unsigned int) opcode);
1735 rc = rbd_dev_refresh(rbd_dev, &hver);
1737 rbd_warn(rbd_dev, "got notification but failed to "
1738 " update snaps: %d\n", rc);
1740 rbd_obj_notify_ack(rbd_dev, hver, notify_id);
1744 * Request sync osd watch/unwatch. The value of "start" determines
1745 * whether a watch request is being initiated or torn down.
1747 static int rbd_dev_header_watch_sync(struct rbd_device *rbd_dev, int start)
1749 struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
1750 struct rbd_obj_request *obj_request;
1751 struct ceph_osd_req_op op;
1754 rbd_assert(start ^ !!rbd_dev->watch_event);
1755 rbd_assert(start ^ !!rbd_dev->watch_request);
1758 ret = ceph_osdc_create_event(osdc, rbd_watch_cb, rbd_dev,
1759 &rbd_dev->watch_event);
1762 rbd_assert(rbd_dev->watch_event != NULL);
1766 obj_request = rbd_obj_request_create(rbd_dev->header_name, 0, 0,
1767 OBJ_REQUEST_NODATA);
1771 osd_req_op_watch_init(&op, CEPH_OSD_OP_WATCH,
1772 rbd_dev->watch_event->cookie,
1773 rbd_dev->header.obj_version, start);
1774 obj_request->osd_req = rbd_osd_req_create(rbd_dev, true,
1776 if (!obj_request->osd_req)
1780 ceph_osdc_set_request_linger(osdc, obj_request->osd_req);
1782 ceph_osdc_unregister_linger_request(osdc,
1783 rbd_dev->watch_request->osd_req);
1784 ret = rbd_obj_request_submit(osdc, obj_request);
1787 ret = rbd_obj_request_wait(obj_request);
1790 ret = obj_request->result;
1795 * A watch request is set to linger, so the underlying osd
1796 * request won't go away until we unregister it. We retain
1797 * a pointer to the object request during that time (in
1798 * rbd_dev->watch_request), so we'll keep a reference to
1799 * it. We'll drop that reference (below) after we've
1803 rbd_dev->watch_request = obj_request;
1808 /* We have successfully torn down the watch request */
1810 rbd_obj_request_put(rbd_dev->watch_request);
1811 rbd_dev->watch_request = NULL;
1813 /* Cancel the event if we're tearing down, or on error */
1814 ceph_osdc_cancel_event(rbd_dev->watch_event);
1815 rbd_dev->watch_event = NULL;
1817 rbd_obj_request_put(obj_request);
1823 * Synchronous osd object method call
1825 static int rbd_obj_method_sync(struct rbd_device *rbd_dev,
1826 const char *object_name,
1827 const char *class_name,
1828 const char *method_name,
1829 const char *outbound,
1830 size_t outbound_size,
1832 size_t inbound_size,
1835 struct rbd_obj_request *obj_request;
1836 struct ceph_osd_client *osdc;
1837 struct ceph_osd_req_op op;
1838 struct page **pages;
1843 * Method calls are ultimately read operations but they
1844 * don't involve object data (so no offset or length).
1845 * The result should placed into the inbound buffer
1846 * provided. They also supply outbound data--parameters for
1847 * the object method. Currently if this is present it will
1850 page_count = (u32) calc_pages_for(0, inbound_size);
1851 pages = ceph_alloc_page_vector(page_count, GFP_KERNEL);
1853 return PTR_ERR(pages);
1856 obj_request = rbd_obj_request_create(object_name, 0, 0,
1861 obj_request->pages = pages;
1862 obj_request->page_count = page_count;
1864 osd_req_op_cls_init(&op, CEPH_OSD_OP_CALL, class_name, method_name,
1865 outbound, outbound_size);
1866 obj_request->osd_req = rbd_osd_req_create(rbd_dev, false,
1868 if (!obj_request->osd_req)
1871 osdc = &rbd_dev->rbd_client->client->osdc;
1872 ret = rbd_obj_request_submit(osdc, obj_request);
1875 ret = rbd_obj_request_wait(obj_request);
1879 ret = obj_request->result;
1883 ceph_copy_from_page_vector(pages, inbound, 0, obj_request->xferred);
1885 *version = obj_request->version;
1888 rbd_obj_request_put(obj_request);
1890 ceph_release_page_vector(pages, page_count);
1895 static void rbd_request_fn(struct request_queue *q)
1896 __releases(q->queue_lock) __acquires(q->queue_lock)
1898 struct rbd_device *rbd_dev = q->queuedata;
1899 bool read_only = rbd_dev->mapping.read_only;
1903 while ((rq = blk_fetch_request(q))) {
1904 bool write_request = rq_data_dir(rq) == WRITE;
1905 struct rbd_img_request *img_request;
1909 /* Ignore any non-FS requests that filter through. */
1911 if (rq->cmd_type != REQ_TYPE_FS) {
1912 dout("%s: non-fs request type %d\n", __func__,
1913 (int) rq->cmd_type);
1914 __blk_end_request_all(rq, 0);
1918 /* Ignore/skip any zero-length requests */
1920 offset = (u64) blk_rq_pos(rq) << SECTOR_SHIFT;
1921 length = (u64) blk_rq_bytes(rq);
1924 dout("%s: zero-length request\n", __func__);
1925 __blk_end_request_all(rq, 0);
1929 spin_unlock_irq(q->queue_lock);
1931 /* Disallow writes to a read-only device */
1933 if (write_request) {
1937 rbd_assert(rbd_dev->spec->snap_id == CEPH_NOSNAP);
1941 * Quit early if the mapped snapshot no longer
1942 * exists. It's still possible the snapshot will
1943 * have disappeared by the time our request arrives
1944 * at the osd, but there's no sense in sending it if
1947 if (!test_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags)) {
1948 dout("request for non-existent snapshot");
1949 rbd_assert(rbd_dev->spec->snap_id != CEPH_NOSNAP);
1955 if (WARN_ON(offset && length > U64_MAX - offset + 1))
1956 goto end_request; /* Shouldn't happen */
1959 img_request = rbd_img_request_create(rbd_dev, offset, length,
1964 img_request->rq = rq;
1966 result = rbd_img_request_fill_bio(img_request, rq->bio);
1968 result = rbd_img_request_submit(img_request);
1970 rbd_img_request_put(img_request);
1972 spin_lock_irq(q->queue_lock);
1974 rbd_warn(rbd_dev, "obj_request %s result %d\n",
1975 write_request ? "write" : "read", result);
1976 __blk_end_request_all(rq, result);
1982 * a queue callback. Makes sure that we don't create a bio that spans across
1983 * multiple osd objects. One exception would be with a single page bios,
1984 * which we handle later at bio_chain_clone_range()
1986 static int rbd_merge_bvec(struct request_queue *q, struct bvec_merge_data *bmd,
1987 struct bio_vec *bvec)
1989 struct rbd_device *rbd_dev = q->queuedata;
1990 sector_t sector_offset;
1991 sector_t sectors_per_obj;
1992 sector_t obj_sector_offset;
1996 * Find how far into its rbd object the partition-relative
1997 * bio start sector is to offset relative to the enclosing
2000 sector_offset = get_start_sect(bmd->bi_bdev) + bmd->bi_sector;
2001 sectors_per_obj = 1 << (rbd_dev->header.obj_order - SECTOR_SHIFT);
2002 obj_sector_offset = sector_offset & (sectors_per_obj - 1);
2005 * Compute the number of bytes from that offset to the end
2006 * of the object. Account for what's already used by the bio.
2008 ret = (int) (sectors_per_obj - obj_sector_offset) << SECTOR_SHIFT;
2009 if (ret > bmd->bi_size)
2010 ret -= bmd->bi_size;
2015 * Don't send back more than was asked for. And if the bio
2016 * was empty, let the whole thing through because: "Note
2017 * that a block device *must* allow a single page to be
2018 * added to an empty bio."
2020 rbd_assert(bvec->bv_len <= PAGE_SIZE);
2021 if (ret > (int) bvec->bv_len || !bmd->bi_size)
2022 ret = (int) bvec->bv_len;
2027 static void rbd_free_disk(struct rbd_device *rbd_dev)
2029 struct gendisk *disk = rbd_dev->disk;
2034 if (disk->flags & GENHD_FL_UP)
2037 blk_cleanup_queue(disk->queue);
2041 static int rbd_obj_read_sync(struct rbd_device *rbd_dev,
2042 const char *object_name,
2043 u64 offset, u64 length,
2044 char *buf, u64 *version)
2047 struct ceph_osd_req_op op;
2048 struct rbd_obj_request *obj_request;
2049 struct ceph_osd_client *osdc;
2050 struct page **pages = NULL;
2055 page_count = (u32) calc_pages_for(offset, length);
2056 pages = ceph_alloc_page_vector(page_count, GFP_KERNEL);
2058 ret = PTR_ERR(pages);
2061 obj_request = rbd_obj_request_create(object_name, offset, length,
2066 obj_request->pages = pages;
2067 obj_request->page_count = page_count;
2069 osd_req_op_extent_init(&op, CEPH_OSD_OP_READ, offset, length, 0, 0);
2070 obj_request->osd_req = rbd_osd_req_create(rbd_dev, false,
2072 if (!obj_request->osd_req)
2075 osdc = &rbd_dev->rbd_client->client->osdc;
2076 ret = rbd_obj_request_submit(osdc, obj_request);
2079 ret = rbd_obj_request_wait(obj_request);
2083 ret = obj_request->result;
2087 rbd_assert(obj_request->xferred <= (u64) SIZE_MAX);
2088 size = (size_t) obj_request->xferred;
2089 ceph_copy_from_page_vector(pages, buf, 0, size);
2090 rbd_assert(size <= (size_t) INT_MAX);
2093 *version = obj_request->version;
2096 rbd_obj_request_put(obj_request);
2098 ceph_release_page_vector(pages, page_count);
2104 * Read the complete header for the given rbd device.
2106 * Returns a pointer to a dynamically-allocated buffer containing
2107 * the complete and validated header. Caller can pass the address
2108 * of a variable that will be filled in with the version of the
2109 * header object at the time it was read.
2111 * Returns a pointer-coded errno if a failure occurs.
2113 static struct rbd_image_header_ondisk *
2114 rbd_dev_v1_header_read(struct rbd_device *rbd_dev, u64 *version)
2116 struct rbd_image_header_ondisk *ondisk = NULL;
2123 * The complete header will include an array of its 64-bit
2124 * snapshot ids, followed by the names of those snapshots as
2125 * a contiguous block of NUL-terminated strings. Note that
2126 * the number of snapshots could change by the time we read
2127 * it in, in which case we re-read it.
2134 size = sizeof (*ondisk);
2135 size += snap_count * sizeof (struct rbd_image_snap_ondisk);
2137 ondisk = kmalloc(size, GFP_KERNEL);
2139 return ERR_PTR(-ENOMEM);
2141 ret = rbd_obj_read_sync(rbd_dev, rbd_dev->header_name,
2143 (char *) ondisk, version);
2146 if (WARN_ON((size_t) ret < size)) {
2148 rbd_warn(rbd_dev, "short header read (want %zd got %d)",
2152 if (!rbd_dev_ondisk_valid(ondisk)) {
2154 rbd_warn(rbd_dev, "invalid header");
2158 names_size = le64_to_cpu(ondisk->snap_names_len);
2159 want_count = snap_count;
2160 snap_count = le32_to_cpu(ondisk->snap_count);
2161 } while (snap_count != want_count);
2168 return ERR_PTR(ret);
2172 * reload the ondisk the header
2174 static int rbd_read_header(struct rbd_device *rbd_dev,
2175 struct rbd_image_header *header)
2177 struct rbd_image_header_ondisk *ondisk;
2181 ondisk = rbd_dev_v1_header_read(rbd_dev, &ver);
2183 return PTR_ERR(ondisk);
2184 ret = rbd_header_from_disk(header, ondisk);
2186 header->obj_version = ver;
2192 static void rbd_remove_all_snaps(struct rbd_device *rbd_dev)
2194 struct rbd_snap *snap;
2195 struct rbd_snap *next;
2197 list_for_each_entry_safe(snap, next, &rbd_dev->snaps, node)
2198 rbd_remove_snap_dev(snap);
2201 static void rbd_update_mapping_size(struct rbd_device *rbd_dev)
2205 if (rbd_dev->spec->snap_id != CEPH_NOSNAP)
2208 size = (sector_t) rbd_dev->header.image_size / SECTOR_SIZE;
2209 dout("setting size to %llu sectors", (unsigned long long) size);
2210 rbd_dev->mapping.size = (u64) size;
2211 set_capacity(rbd_dev->disk, size);
2215 * only read the first part of the ondisk header, without the snaps info
2217 static int rbd_dev_v1_refresh(struct rbd_device *rbd_dev, u64 *hver)
2220 struct rbd_image_header h;
2222 ret = rbd_read_header(rbd_dev, &h);
2226 down_write(&rbd_dev->header_rwsem);
2228 /* Update image size, and check for resize of mapped image */
2229 rbd_dev->header.image_size = h.image_size;
2230 rbd_update_mapping_size(rbd_dev);
2232 /* rbd_dev->header.object_prefix shouldn't change */
2233 kfree(rbd_dev->header.snap_sizes);
2234 kfree(rbd_dev->header.snap_names);
2235 /* osd requests may still refer to snapc */
2236 ceph_put_snap_context(rbd_dev->header.snapc);
2239 *hver = h.obj_version;
2240 rbd_dev->header.obj_version = h.obj_version;
2241 rbd_dev->header.image_size = h.image_size;
2242 rbd_dev->header.snapc = h.snapc;
2243 rbd_dev->header.snap_names = h.snap_names;
2244 rbd_dev->header.snap_sizes = h.snap_sizes;
2245 /* Free the extra copy of the object prefix */
2246 WARN_ON(strcmp(rbd_dev->header.object_prefix, h.object_prefix));
2247 kfree(h.object_prefix);
2249 ret = rbd_dev_snaps_update(rbd_dev);
2251 ret = rbd_dev_snaps_register(rbd_dev);
2253 up_write(&rbd_dev->header_rwsem);
2258 static int rbd_dev_refresh(struct rbd_device *rbd_dev, u64 *hver)
2262 rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
2263 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
2264 if (rbd_dev->image_format == 1)
2265 ret = rbd_dev_v1_refresh(rbd_dev, hver);
2267 ret = rbd_dev_v2_refresh(rbd_dev, hver);
2268 mutex_unlock(&ctl_mutex);
2273 static int rbd_init_disk(struct rbd_device *rbd_dev)
2275 struct gendisk *disk;
2276 struct request_queue *q;
2279 /* create gendisk info */
2280 disk = alloc_disk(RBD_MINORS_PER_MAJOR);
2284 snprintf(disk->disk_name, sizeof(disk->disk_name), RBD_DRV_NAME "%d",
2286 disk->major = rbd_dev->major;
2287 disk->first_minor = 0;
2288 disk->fops = &rbd_bd_ops;
2289 disk->private_data = rbd_dev;
2291 q = blk_init_queue(rbd_request_fn, &rbd_dev->lock);
2295 /* We use the default size, but let's be explicit about it. */
2296 blk_queue_physical_block_size(q, SECTOR_SIZE);
2298 /* set io sizes to object size */
2299 segment_size = rbd_obj_bytes(&rbd_dev->header);
2300 blk_queue_max_hw_sectors(q, segment_size / SECTOR_SIZE);
2301 blk_queue_max_segment_size(q, segment_size);
2302 blk_queue_io_min(q, segment_size);
2303 blk_queue_io_opt(q, segment_size);
2305 blk_queue_merge_bvec(q, rbd_merge_bvec);
2308 q->queuedata = rbd_dev;
2310 rbd_dev->disk = disk;
2312 set_capacity(rbd_dev->disk, rbd_dev->mapping.size / SECTOR_SIZE);
2325 static struct rbd_device *dev_to_rbd_dev(struct device *dev)
2327 return container_of(dev, struct rbd_device, dev);
2330 static ssize_t rbd_size_show(struct device *dev,
2331 struct device_attribute *attr, char *buf)
2333 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
2336 down_read(&rbd_dev->header_rwsem);
2337 size = get_capacity(rbd_dev->disk);
2338 up_read(&rbd_dev->header_rwsem);
2340 return sprintf(buf, "%llu\n", (unsigned long long) size * SECTOR_SIZE);
2344 * Note this shows the features for whatever's mapped, which is not
2345 * necessarily the base image.
2347 static ssize_t rbd_features_show(struct device *dev,
2348 struct device_attribute *attr, char *buf)
2350 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
2352 return sprintf(buf, "0x%016llx\n",
2353 (unsigned long long) rbd_dev->mapping.features);
2356 static ssize_t rbd_major_show(struct device *dev,
2357 struct device_attribute *attr, char *buf)
2359 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
2361 return sprintf(buf, "%d\n", rbd_dev->major);
2364 static ssize_t rbd_client_id_show(struct device *dev,
2365 struct device_attribute *attr, char *buf)
2367 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
2369 return sprintf(buf, "client%lld\n",
2370 ceph_client_id(rbd_dev->rbd_client->client));
2373 static ssize_t rbd_pool_show(struct device *dev,
2374 struct device_attribute *attr, char *buf)
2376 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
2378 return sprintf(buf, "%s\n", rbd_dev->spec->pool_name);
2381 static ssize_t rbd_pool_id_show(struct device *dev,
2382 struct device_attribute *attr, char *buf)
2384 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
2386 return sprintf(buf, "%llu\n",
2387 (unsigned long long) rbd_dev->spec->pool_id);
2390 static ssize_t rbd_name_show(struct device *dev,
2391 struct device_attribute *attr, char *buf)
2393 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
2395 if (rbd_dev->spec->image_name)
2396 return sprintf(buf, "%s\n", rbd_dev->spec->image_name);
2398 return sprintf(buf, "(unknown)\n");
2401 static ssize_t rbd_image_id_show(struct device *dev,
2402 struct device_attribute *attr, char *buf)
2404 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
2406 return sprintf(buf, "%s\n", rbd_dev->spec->image_id);
2410 * Shows the name of the currently-mapped snapshot (or
2411 * RBD_SNAP_HEAD_NAME for the base image).
2413 static ssize_t rbd_snap_show(struct device *dev,
2414 struct device_attribute *attr,
2417 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
2419 return sprintf(buf, "%s\n", rbd_dev->spec->snap_name);
2423 * For an rbd v2 image, shows the pool id, image id, and snapshot id
2424 * for the parent image. If there is no parent, simply shows
2425 * "(no parent image)".
2427 static ssize_t rbd_parent_show(struct device *dev,
2428 struct device_attribute *attr,
2431 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
2432 struct rbd_spec *spec = rbd_dev->parent_spec;
2437 return sprintf(buf, "(no parent image)\n");
2439 count = sprintf(bufp, "pool_id %llu\npool_name %s\n",
2440 (unsigned long long) spec->pool_id, spec->pool_name);
2445 count = sprintf(bufp, "image_id %s\nimage_name %s\n", spec->image_id,
2446 spec->image_name ? spec->image_name : "(unknown)");
2451 count = sprintf(bufp, "snap_id %llu\nsnap_name %s\n",
2452 (unsigned long long) spec->snap_id, spec->snap_name);
2457 count = sprintf(bufp, "overlap %llu\n", rbd_dev->parent_overlap);
2462 return (ssize_t) (bufp - buf);
2465 static ssize_t rbd_image_refresh(struct device *dev,
2466 struct device_attribute *attr,
2470 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
2473 ret = rbd_dev_refresh(rbd_dev, NULL);
2475 return ret < 0 ? ret : size;
2478 static DEVICE_ATTR(size, S_IRUGO, rbd_size_show, NULL);
2479 static DEVICE_ATTR(features, S_IRUGO, rbd_features_show, NULL);
2480 static DEVICE_ATTR(major, S_IRUGO, rbd_major_show, NULL);
2481 static DEVICE_ATTR(client_id, S_IRUGO, rbd_client_id_show, NULL);
2482 static DEVICE_ATTR(pool, S_IRUGO, rbd_pool_show, NULL);
2483 static DEVICE_ATTR(pool_id, S_IRUGO, rbd_pool_id_show, NULL);
2484 static DEVICE_ATTR(name, S_IRUGO, rbd_name_show, NULL);
2485 static DEVICE_ATTR(image_id, S_IRUGO, rbd_image_id_show, NULL);
2486 static DEVICE_ATTR(refresh, S_IWUSR, NULL, rbd_image_refresh);
2487 static DEVICE_ATTR(current_snap, S_IRUGO, rbd_snap_show, NULL);
2488 static DEVICE_ATTR(parent, S_IRUGO, rbd_parent_show, NULL);
2490 static struct attribute *rbd_attrs[] = {
2491 &dev_attr_size.attr,
2492 &dev_attr_features.attr,
2493 &dev_attr_major.attr,
2494 &dev_attr_client_id.attr,
2495 &dev_attr_pool.attr,
2496 &dev_attr_pool_id.attr,
2497 &dev_attr_name.attr,
2498 &dev_attr_image_id.attr,
2499 &dev_attr_current_snap.attr,
2500 &dev_attr_parent.attr,
2501 &dev_attr_refresh.attr,
2505 static struct attribute_group rbd_attr_group = {
2509 static const struct attribute_group *rbd_attr_groups[] = {
2514 static void rbd_sysfs_dev_release(struct device *dev)
2518 static struct device_type rbd_device_type = {
2520 .groups = rbd_attr_groups,
2521 .release = rbd_sysfs_dev_release,
2529 static ssize_t rbd_snap_size_show(struct device *dev,
2530 struct device_attribute *attr,
2533 struct rbd_snap *snap = container_of(dev, struct rbd_snap, dev);
2535 return sprintf(buf, "%llu\n", (unsigned long long)snap->size);
2538 static ssize_t rbd_snap_id_show(struct device *dev,
2539 struct device_attribute *attr,
2542 struct rbd_snap *snap = container_of(dev, struct rbd_snap, dev);
2544 return sprintf(buf, "%llu\n", (unsigned long long)snap->id);
2547 static ssize_t rbd_snap_features_show(struct device *dev,
2548 struct device_attribute *attr,
2551 struct rbd_snap *snap = container_of(dev, struct rbd_snap, dev);
2553 return sprintf(buf, "0x%016llx\n",
2554 (unsigned long long) snap->features);
2557 static DEVICE_ATTR(snap_size, S_IRUGO, rbd_snap_size_show, NULL);
2558 static DEVICE_ATTR(snap_id, S_IRUGO, rbd_snap_id_show, NULL);
2559 static DEVICE_ATTR(snap_features, S_IRUGO, rbd_snap_features_show, NULL);
2561 static struct attribute *rbd_snap_attrs[] = {
2562 &dev_attr_snap_size.attr,
2563 &dev_attr_snap_id.attr,
2564 &dev_attr_snap_features.attr,
2568 static struct attribute_group rbd_snap_attr_group = {
2569 .attrs = rbd_snap_attrs,
2572 static void rbd_snap_dev_release(struct device *dev)
2574 struct rbd_snap *snap = container_of(dev, struct rbd_snap, dev);
2579 static const struct attribute_group *rbd_snap_attr_groups[] = {
2580 &rbd_snap_attr_group,
2584 static struct device_type rbd_snap_device_type = {
2585 .groups = rbd_snap_attr_groups,
2586 .release = rbd_snap_dev_release,
2589 static struct rbd_spec *rbd_spec_get(struct rbd_spec *spec)
2591 kref_get(&spec->kref);
2596 static void rbd_spec_free(struct kref *kref);
2597 static void rbd_spec_put(struct rbd_spec *spec)
2600 kref_put(&spec->kref, rbd_spec_free);
2603 static struct rbd_spec *rbd_spec_alloc(void)
2605 struct rbd_spec *spec;
2607 spec = kzalloc(sizeof (*spec), GFP_KERNEL);
2610 kref_init(&spec->kref);
2612 rbd_spec_put(rbd_spec_get(spec)); /* TEMPORARY */
2617 static void rbd_spec_free(struct kref *kref)
2619 struct rbd_spec *spec = container_of(kref, struct rbd_spec, kref);
2621 kfree(spec->pool_name);
2622 kfree(spec->image_id);
2623 kfree(spec->image_name);
2624 kfree(spec->snap_name);
2628 static struct rbd_device *rbd_dev_create(struct rbd_client *rbdc,
2629 struct rbd_spec *spec)
2631 struct rbd_device *rbd_dev;
2633 rbd_dev = kzalloc(sizeof (*rbd_dev), GFP_KERNEL);
2637 spin_lock_init(&rbd_dev->lock);
2639 INIT_LIST_HEAD(&rbd_dev->node);
2640 INIT_LIST_HEAD(&rbd_dev->snaps);
2641 init_rwsem(&rbd_dev->header_rwsem);
2643 rbd_dev->spec = spec;
2644 rbd_dev->rbd_client = rbdc;
2646 /* Initialize the layout used for all rbd requests */
2648 rbd_dev->layout.fl_stripe_unit = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER);
2649 rbd_dev->layout.fl_stripe_count = cpu_to_le32(1);
2650 rbd_dev->layout.fl_object_size = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER);
2651 rbd_dev->layout.fl_pg_pool = cpu_to_le32((u32) spec->pool_id);
2656 static void rbd_dev_destroy(struct rbd_device *rbd_dev)
2658 rbd_spec_put(rbd_dev->parent_spec);
2659 kfree(rbd_dev->header_name);
2660 rbd_put_client(rbd_dev->rbd_client);
2661 rbd_spec_put(rbd_dev->spec);
2665 static bool rbd_snap_registered(struct rbd_snap *snap)
2667 bool ret = snap->dev.type == &rbd_snap_device_type;
2668 bool reg = device_is_registered(&snap->dev);
2670 rbd_assert(!ret ^ reg);
2675 static void rbd_remove_snap_dev(struct rbd_snap *snap)
2677 list_del(&snap->node);
2678 if (device_is_registered(&snap->dev))
2679 device_unregister(&snap->dev);
2682 static int rbd_register_snap_dev(struct rbd_snap *snap,
2683 struct device *parent)
2685 struct device *dev = &snap->dev;
2688 dev->type = &rbd_snap_device_type;
2689 dev->parent = parent;
2690 dev->release = rbd_snap_dev_release;
2691 dev_set_name(dev, "%s%s", RBD_SNAP_DEV_NAME_PREFIX, snap->name);
2692 dout("%s: registering device for snapshot %s\n", __func__, snap->name);
2694 ret = device_register(dev);
2699 static struct rbd_snap *__rbd_add_snap_dev(struct rbd_device *rbd_dev,
2700 const char *snap_name,
2701 u64 snap_id, u64 snap_size,
2704 struct rbd_snap *snap;
2707 snap = kzalloc(sizeof (*snap), GFP_KERNEL);
2709 return ERR_PTR(-ENOMEM);
2712 snap->name = kstrdup(snap_name, GFP_KERNEL);
2717 snap->size = snap_size;
2718 snap->features = snap_features;
2726 return ERR_PTR(ret);
2729 static char *rbd_dev_v1_snap_info(struct rbd_device *rbd_dev, u32 which,
2730 u64 *snap_size, u64 *snap_features)
2734 rbd_assert(which < rbd_dev->header.snapc->num_snaps);
2736 *snap_size = rbd_dev->header.snap_sizes[which];
2737 *snap_features = 0; /* No features for v1 */
2739 /* Skip over names until we find the one we are looking for */
2741 snap_name = rbd_dev->header.snap_names;
2743 snap_name += strlen(snap_name) + 1;
2749 * Get the size and object order for an image snapshot, or if
2750 * snap_id is CEPH_NOSNAP, gets this information for the base
2753 static int _rbd_dev_v2_snap_size(struct rbd_device *rbd_dev, u64 snap_id,
2754 u8 *order, u64 *snap_size)
2756 __le64 snapid = cpu_to_le64(snap_id);
2761 } __attribute__ ((packed)) size_buf = { 0 };
2763 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
2765 (char *) &snapid, sizeof (snapid),
2766 (char *) &size_buf, sizeof (size_buf), NULL);
2767 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
2771 *order = size_buf.order;
2772 *snap_size = le64_to_cpu(size_buf.size);
2774 dout(" snap_id 0x%016llx order = %u, snap_size = %llu\n",
2775 (unsigned long long) snap_id, (unsigned int) *order,
2776 (unsigned long long) *snap_size);
2781 static int rbd_dev_v2_image_size(struct rbd_device *rbd_dev)
2783 return _rbd_dev_v2_snap_size(rbd_dev, CEPH_NOSNAP,
2784 &rbd_dev->header.obj_order,
2785 &rbd_dev->header.image_size);
2788 static int rbd_dev_v2_object_prefix(struct rbd_device *rbd_dev)
2794 reply_buf = kzalloc(RBD_OBJ_PREFIX_LEN_MAX, GFP_KERNEL);
2798 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
2799 "rbd", "get_object_prefix",
2801 reply_buf, RBD_OBJ_PREFIX_LEN_MAX, NULL);
2802 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
2807 rbd_dev->header.object_prefix = ceph_extract_encoded_string(&p,
2808 p + RBD_OBJ_PREFIX_LEN_MAX,
2811 if (IS_ERR(rbd_dev->header.object_prefix)) {
2812 ret = PTR_ERR(rbd_dev->header.object_prefix);
2813 rbd_dev->header.object_prefix = NULL;
2815 dout(" object_prefix = %s\n", rbd_dev->header.object_prefix);
2824 static int _rbd_dev_v2_snap_features(struct rbd_device *rbd_dev, u64 snap_id,
2827 __le64 snapid = cpu_to_le64(snap_id);
2831 } features_buf = { 0 };
2835 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
2836 "rbd", "get_features",
2837 (char *) &snapid, sizeof (snapid),
2838 (char *) &features_buf, sizeof (features_buf),
2840 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
2844 incompat = le64_to_cpu(features_buf.incompat);
2845 if (incompat & ~RBD_FEATURES_ALL)
2848 *snap_features = le64_to_cpu(features_buf.features);
2850 dout(" snap_id 0x%016llx features = 0x%016llx incompat = 0x%016llx\n",
2851 (unsigned long long) snap_id,
2852 (unsigned long long) *snap_features,
2853 (unsigned long long) le64_to_cpu(features_buf.incompat));
2858 static int rbd_dev_v2_features(struct rbd_device *rbd_dev)
2860 return _rbd_dev_v2_snap_features(rbd_dev, CEPH_NOSNAP,
2861 &rbd_dev->header.features);
2864 static int rbd_dev_v2_parent_info(struct rbd_device *rbd_dev)
2866 struct rbd_spec *parent_spec;
2868 void *reply_buf = NULL;
2876 parent_spec = rbd_spec_alloc();
2880 size = sizeof (__le64) + /* pool_id */
2881 sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX + /* image_id */
2882 sizeof (__le64) + /* snap_id */
2883 sizeof (__le64); /* overlap */
2884 reply_buf = kmalloc(size, GFP_KERNEL);
2890 snapid = cpu_to_le64(CEPH_NOSNAP);
2891 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
2892 "rbd", "get_parent",
2893 (char *) &snapid, sizeof (snapid),
2894 (char *) reply_buf, size, NULL);
2895 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
2901 end = (char *) reply_buf + size;
2902 ceph_decode_64_safe(&p, end, parent_spec->pool_id, out_err);
2903 if (parent_spec->pool_id == CEPH_NOPOOL)
2904 goto out; /* No parent? No problem. */
2906 /* The ceph file layout needs to fit pool id in 32 bits */
2909 if (WARN_ON(parent_spec->pool_id > (u64) U32_MAX))
2912 image_id = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL);
2913 if (IS_ERR(image_id)) {
2914 ret = PTR_ERR(image_id);
2917 parent_spec->image_id = image_id;
2918 ceph_decode_64_safe(&p, end, parent_spec->snap_id, out_err);
2919 ceph_decode_64_safe(&p, end, overlap, out_err);
2921 rbd_dev->parent_overlap = overlap;
2922 rbd_dev->parent_spec = parent_spec;
2923 parent_spec = NULL; /* rbd_dev now owns this */
2928 rbd_spec_put(parent_spec);
2933 static char *rbd_dev_image_name(struct rbd_device *rbd_dev)
2935 size_t image_id_size;
2940 void *reply_buf = NULL;
2942 char *image_name = NULL;
2945 rbd_assert(!rbd_dev->spec->image_name);
2947 len = strlen(rbd_dev->spec->image_id);
2948 image_id_size = sizeof (__le32) + len;
2949 image_id = kmalloc(image_id_size, GFP_KERNEL);
2954 end = (char *) image_id + image_id_size;
2955 ceph_encode_string(&p, end, rbd_dev->spec->image_id, (u32) len);
2957 size = sizeof (__le32) + RBD_IMAGE_NAME_LEN_MAX;
2958 reply_buf = kmalloc(size, GFP_KERNEL);
2962 ret = rbd_obj_method_sync(rbd_dev, RBD_DIRECTORY,
2963 "rbd", "dir_get_name",
2964 image_id, image_id_size,
2965 (char *) reply_buf, size, NULL);
2969 end = (char *) reply_buf + size;
2970 image_name = ceph_extract_encoded_string(&p, end, &len, GFP_KERNEL);
2971 if (IS_ERR(image_name))
2974 dout("%s: name is %s len is %zd\n", __func__, image_name, len);
2983 * When a parent image gets probed, we only have the pool, image,
2984 * and snapshot ids but not the names of any of them. This call
2985 * is made later to fill in those names. It has to be done after
2986 * rbd_dev_snaps_update() has completed because some of the
2987 * information (in particular, snapshot name) is not available
2990 static int rbd_dev_probe_update_spec(struct rbd_device *rbd_dev)
2992 struct ceph_osd_client *osdc;
2994 void *reply_buf = NULL;
2997 if (rbd_dev->spec->pool_name)
2998 return 0; /* Already have the names */
3000 /* Look up the pool name */
3002 osdc = &rbd_dev->rbd_client->client->osdc;
3003 name = ceph_pg_pool_name_by_id(osdc->osdmap, rbd_dev->spec->pool_id);
3005 rbd_warn(rbd_dev, "there is no pool with id %llu",
3006 rbd_dev->spec->pool_id); /* Really a BUG() */
3010 rbd_dev->spec->pool_name = kstrdup(name, GFP_KERNEL);
3011 if (!rbd_dev->spec->pool_name)
3014 /* Fetch the image name; tolerate failure here */
3016 name = rbd_dev_image_name(rbd_dev);
3018 rbd_dev->spec->image_name = (char *) name;
3020 rbd_warn(rbd_dev, "unable to get image name");
3022 /* Look up the snapshot name. */
3024 name = rbd_snap_name(rbd_dev, rbd_dev->spec->snap_id);
3026 rbd_warn(rbd_dev, "no snapshot with id %llu",
3027 rbd_dev->spec->snap_id); /* Really a BUG() */
3031 rbd_dev->spec->snap_name = kstrdup(name, GFP_KERNEL);
3032 if(!rbd_dev->spec->snap_name)
3038 kfree(rbd_dev->spec->pool_name);
3039 rbd_dev->spec->pool_name = NULL;
3044 static int rbd_dev_v2_snap_context(struct rbd_device *rbd_dev, u64 *ver)
3053 struct ceph_snap_context *snapc;
3057 * We'll need room for the seq value (maximum snapshot id),
3058 * snapshot count, and array of that many snapshot ids.
3059 * For now we have a fixed upper limit on the number we're
3060 * prepared to receive.
3062 size = sizeof (__le64) + sizeof (__le32) +
3063 RBD_MAX_SNAP_COUNT * sizeof (__le64);
3064 reply_buf = kzalloc(size, GFP_KERNEL);
3068 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
3069 "rbd", "get_snapcontext",
3071 reply_buf, size, ver);
3072 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
3078 end = (char *) reply_buf + size;
3079 ceph_decode_64_safe(&p, end, seq, out);
3080 ceph_decode_32_safe(&p, end, snap_count, out);
3083 * Make sure the reported number of snapshot ids wouldn't go
3084 * beyond the end of our buffer. But before checking that,
3085 * make sure the computed size of the snapshot context we
3086 * allocate is representable in a size_t.
3088 if (snap_count > (SIZE_MAX - sizeof (struct ceph_snap_context))
3093 if (!ceph_has_room(&p, end, snap_count * sizeof (__le64)))
3096 size = sizeof (struct ceph_snap_context) +
3097 snap_count * sizeof (snapc->snaps[0]);
3098 snapc = kmalloc(size, GFP_KERNEL);
3104 atomic_set(&snapc->nref, 1);
3106 snapc->num_snaps = snap_count;
3107 for (i = 0; i < snap_count; i++)
3108 snapc->snaps[i] = ceph_decode_64(&p);
3110 rbd_dev->header.snapc = snapc;
3112 dout(" snap context seq = %llu, snap_count = %u\n",
3113 (unsigned long long) seq, (unsigned int) snap_count);
3121 static char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev, u32 which)
3131 size = sizeof (__le32) + RBD_MAX_SNAP_NAME_LEN;
3132 reply_buf = kmalloc(size, GFP_KERNEL);
3134 return ERR_PTR(-ENOMEM);
3136 snap_id = cpu_to_le64(rbd_dev->header.snapc->snaps[which]);
3137 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
3138 "rbd", "get_snapshot_name",
3139 (char *) &snap_id, sizeof (snap_id),
3140 reply_buf, size, NULL);
3141 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
3146 end = (char *) reply_buf + size;
3147 snap_name = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL);
3148 if (IS_ERR(snap_name)) {
3149 ret = PTR_ERR(snap_name);
3152 dout(" snap_id 0x%016llx snap_name = %s\n",
3153 (unsigned long long) le64_to_cpu(snap_id), snap_name);
3161 return ERR_PTR(ret);
3164 static char *rbd_dev_v2_snap_info(struct rbd_device *rbd_dev, u32 which,
3165 u64 *snap_size, u64 *snap_features)
3171 snap_id = rbd_dev->header.snapc->snaps[which];
3172 ret = _rbd_dev_v2_snap_size(rbd_dev, snap_id, &order, snap_size);
3174 return ERR_PTR(ret);
3175 ret = _rbd_dev_v2_snap_features(rbd_dev, snap_id, snap_features);
3177 return ERR_PTR(ret);
3179 return rbd_dev_v2_snap_name(rbd_dev, which);
3182 static char *rbd_dev_snap_info(struct rbd_device *rbd_dev, u32 which,
3183 u64 *snap_size, u64 *snap_features)
3185 if (rbd_dev->image_format == 1)
3186 return rbd_dev_v1_snap_info(rbd_dev, which,
3187 snap_size, snap_features);
3188 if (rbd_dev->image_format == 2)
3189 return rbd_dev_v2_snap_info(rbd_dev, which,
3190 snap_size, snap_features);
3191 return ERR_PTR(-EINVAL);
3194 static int rbd_dev_v2_refresh(struct rbd_device *rbd_dev, u64 *hver)
3199 down_write(&rbd_dev->header_rwsem);
3201 /* Grab old order first, to see if it changes */
3203 obj_order = rbd_dev->header.obj_order,
3204 ret = rbd_dev_v2_image_size(rbd_dev);
3207 if (rbd_dev->header.obj_order != obj_order) {
3211 rbd_update_mapping_size(rbd_dev);
3213 ret = rbd_dev_v2_snap_context(rbd_dev, hver);
3214 dout("rbd_dev_v2_snap_context returned %d\n", ret);
3217 ret = rbd_dev_snaps_update(rbd_dev);
3218 dout("rbd_dev_snaps_update returned %d\n", ret);
3221 ret = rbd_dev_snaps_register(rbd_dev);
3222 dout("rbd_dev_snaps_register returned %d\n", ret);
3224 up_write(&rbd_dev->header_rwsem);
3230 * Scan the rbd device's current snapshot list and compare it to the
3231 * newly-received snapshot context. Remove any existing snapshots
3232 * not present in the new snapshot context. Add a new snapshot for
3233 * any snaphots in the snapshot context not in the current list.
3234 * And verify there are no changes to snapshots we already know
3237 * Assumes the snapshots in the snapshot context are sorted by
3238 * snapshot id, highest id first. (Snapshots in the rbd_dev's list
3239 * are also maintained in that order.)
3241 static int rbd_dev_snaps_update(struct rbd_device *rbd_dev)
3243 struct ceph_snap_context *snapc = rbd_dev->header.snapc;
3244 const u32 snap_count = snapc->num_snaps;
3245 struct list_head *head = &rbd_dev->snaps;
3246 struct list_head *links = head->next;
3249 dout("%s: snap count is %u\n", __func__, (unsigned int) snap_count);
3250 while (index < snap_count || links != head) {
3252 struct rbd_snap *snap;
3255 u64 snap_features = 0;
3257 snap_id = index < snap_count ? snapc->snaps[index]
3259 snap = links != head ? list_entry(links, struct rbd_snap, node)
3261 rbd_assert(!snap || snap->id != CEPH_NOSNAP);
3263 if (snap_id == CEPH_NOSNAP || (snap && snap->id > snap_id)) {
3264 struct list_head *next = links->next;
3267 * A previously-existing snapshot is not in
3268 * the new snap context.
3270 * If the now missing snapshot is the one the
3271 * image is mapped to, clear its exists flag
3272 * so we can avoid sending any more requests
3275 if (rbd_dev->spec->snap_id == snap->id)
3276 clear_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags);
3277 rbd_remove_snap_dev(snap);
3278 dout("%ssnap id %llu has been removed\n",
3279 rbd_dev->spec->snap_id == snap->id ?
3281 (unsigned long long) snap->id);
3283 /* Done with this list entry; advance */
3289 snap_name = rbd_dev_snap_info(rbd_dev, index,
3290 &snap_size, &snap_features);
3291 if (IS_ERR(snap_name))
3292 return PTR_ERR(snap_name);
3294 dout("entry %u: snap_id = %llu\n", (unsigned int) snap_count,
3295 (unsigned long long) snap_id);
3296 if (!snap || (snap_id != CEPH_NOSNAP && snap->id < snap_id)) {
3297 struct rbd_snap *new_snap;
3299 /* We haven't seen this snapshot before */
3301 new_snap = __rbd_add_snap_dev(rbd_dev, snap_name,
3302 snap_id, snap_size, snap_features);
3303 if (IS_ERR(new_snap)) {
3304 int err = PTR_ERR(new_snap);
3306 dout(" failed to add dev, error %d\n", err);
3311 /* New goes before existing, or at end of list */
3313 dout(" added dev%s\n", snap ? "" : " at end\n");
3315 list_add_tail(&new_snap->node, &snap->node);
3317 list_add_tail(&new_snap->node, head);
3319 /* Already have this one */
3321 dout(" already present\n");
3323 rbd_assert(snap->size == snap_size);
3324 rbd_assert(!strcmp(snap->name, snap_name));
3325 rbd_assert(snap->features == snap_features);
3327 /* Done with this list entry; advance */
3329 links = links->next;
3332 /* Advance to the next entry in the snapshot context */
3336 dout("%s: done\n", __func__);
3342 * Scan the list of snapshots and register the devices for any that
3343 * have not already been registered.
3345 static int rbd_dev_snaps_register(struct rbd_device *rbd_dev)
3347 struct rbd_snap *snap;
3350 dout("%s:\n", __func__);
3351 if (WARN_ON(!device_is_registered(&rbd_dev->dev)))
3354 list_for_each_entry(snap, &rbd_dev->snaps, node) {
3355 if (!rbd_snap_registered(snap)) {
3356 ret = rbd_register_snap_dev(snap, &rbd_dev->dev);
3361 dout("%s: returning %d\n", __func__, ret);
3366 static int rbd_bus_add_dev(struct rbd_device *rbd_dev)
3371 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
3373 dev = &rbd_dev->dev;
3374 dev->bus = &rbd_bus_type;
3375 dev->type = &rbd_device_type;
3376 dev->parent = &rbd_root_dev;
3377 dev->release = rbd_dev_release;
3378 dev_set_name(dev, "%d", rbd_dev->dev_id);
3379 ret = device_register(dev);
3381 mutex_unlock(&ctl_mutex);
3386 static void rbd_bus_del_dev(struct rbd_device *rbd_dev)
3388 device_unregister(&rbd_dev->dev);
3391 static atomic64_t rbd_dev_id_max = ATOMIC64_INIT(0);
3394 * Get a unique rbd identifier for the given new rbd_dev, and add
3395 * the rbd_dev to the global list. The minimum rbd id is 1.
3397 static void rbd_dev_id_get(struct rbd_device *rbd_dev)
3399 rbd_dev->dev_id = atomic64_inc_return(&rbd_dev_id_max);
3401 spin_lock(&rbd_dev_list_lock);
3402 list_add_tail(&rbd_dev->node, &rbd_dev_list);
3403 spin_unlock(&rbd_dev_list_lock);
3404 dout("rbd_dev %p given dev id %llu\n", rbd_dev,
3405 (unsigned long long) rbd_dev->dev_id);
3409 * Remove an rbd_dev from the global list, and record that its
3410 * identifier is no longer in use.
3412 static void rbd_dev_id_put(struct rbd_device *rbd_dev)
3414 struct list_head *tmp;
3415 int rbd_id = rbd_dev->dev_id;
3418 rbd_assert(rbd_id > 0);
3420 dout("rbd_dev %p released dev id %llu\n", rbd_dev,
3421 (unsigned long long) rbd_dev->dev_id);
3422 spin_lock(&rbd_dev_list_lock);
3423 list_del_init(&rbd_dev->node);
3426 * If the id being "put" is not the current maximum, there
3427 * is nothing special we need to do.
3429 if (rbd_id != atomic64_read(&rbd_dev_id_max)) {
3430 spin_unlock(&rbd_dev_list_lock);
3435 * We need to update the current maximum id. Search the
3436 * list to find out what it is. We're more likely to find
3437 * the maximum at the end, so search the list backward.
3440 list_for_each_prev(tmp, &rbd_dev_list) {
3441 struct rbd_device *rbd_dev;
3443 rbd_dev = list_entry(tmp, struct rbd_device, node);
3444 if (rbd_dev->dev_id > max_id)
3445 max_id = rbd_dev->dev_id;
3447 spin_unlock(&rbd_dev_list_lock);
3450 * The max id could have been updated by rbd_dev_id_get(), in
3451 * which case it now accurately reflects the new maximum.
3452 * Be careful not to overwrite the maximum value in that
3455 atomic64_cmpxchg(&rbd_dev_id_max, rbd_id, max_id);
3456 dout(" max dev id has been reset\n");
3460 * Skips over white space at *buf, and updates *buf to point to the
3461 * first found non-space character (if any). Returns the length of
3462 * the token (string of non-white space characters) found. Note
3463 * that *buf must be terminated with '\0'.
3465 static inline size_t next_token(const char **buf)
3468 * These are the characters that produce nonzero for
3469 * isspace() in the "C" and "POSIX" locales.
3471 const char *spaces = " \f\n\r\t\v";
3473 *buf += strspn(*buf, spaces); /* Find start of token */
3475 return strcspn(*buf, spaces); /* Return token length */
3479 * Finds the next token in *buf, and if the provided token buffer is
3480 * big enough, copies the found token into it. The result, if
3481 * copied, is guaranteed to be terminated with '\0'. Note that *buf
3482 * must be terminated with '\0' on entry.
3484 * Returns the length of the token found (not including the '\0').
3485 * Return value will be 0 if no token is found, and it will be >=
3486 * token_size if the token would not fit.
3488 * The *buf pointer will be updated to point beyond the end of the
3489 * found token. Note that this occurs even if the token buffer is
3490 * too small to hold it.
3492 static inline size_t copy_token(const char **buf,
3498 len = next_token(buf);
3499 if (len < token_size) {
3500 memcpy(token, *buf, len);
3501 *(token + len) = '\0';
3509 * Finds the next token in *buf, dynamically allocates a buffer big
3510 * enough to hold a copy of it, and copies the token into the new
3511 * buffer. The copy is guaranteed to be terminated with '\0'. Note
3512 * that a duplicate buffer is created even for a zero-length token.
3514 * Returns a pointer to the newly-allocated duplicate, or a null
3515 * pointer if memory for the duplicate was not available. If
3516 * the lenp argument is a non-null pointer, the length of the token
3517 * (not including the '\0') is returned in *lenp.
3519 * If successful, the *buf pointer will be updated to point beyond
3520 * the end of the found token.
3522 * Note: uses GFP_KERNEL for allocation.
3524 static inline char *dup_token(const char **buf, size_t *lenp)
3529 len = next_token(buf);
3530 dup = kmemdup(*buf, len + 1, GFP_KERNEL);
3533 *(dup + len) = '\0';
3543 * Parse the options provided for an "rbd add" (i.e., rbd image
3544 * mapping) request. These arrive via a write to /sys/bus/rbd/add,
3545 * and the data written is passed here via a NUL-terminated buffer.
3546 * Returns 0 if successful or an error code otherwise.
3548 * The information extracted from these options is recorded in
3549 * the other parameters which return dynamically-allocated
3552 * The address of a pointer that will refer to a ceph options
3553 * structure. Caller must release the returned pointer using
3554 * ceph_destroy_options() when it is no longer needed.
3556 * Address of an rbd options pointer. Fully initialized by
3557 * this function; caller must release with kfree().
3559 * Address of an rbd image specification pointer. Fully
3560 * initialized by this function based on parsed options.
3561 * Caller must release with rbd_spec_put().
3563 * The options passed take this form:
3564 * <mon_addrs> <options> <pool_name> <image_name> [<snap_id>]
3567 * A comma-separated list of one or more monitor addresses.
3568 * A monitor address is an ip address, optionally followed
3569 * by a port number (separated by a colon).
3570 * I.e.: ip1[:port1][,ip2[:port2]...]
3572 * A comma-separated list of ceph and/or rbd options.
3574 * The name of the rados pool containing the rbd image.
3576 * The name of the image in that pool to map.
3578 * An optional snapshot id. If provided, the mapping will
3579 * present data from the image at the time that snapshot was
3580 * created. The image head is used if no snapshot id is
3581 * provided. Snapshot mappings are always read-only.
3583 static int rbd_add_parse_args(const char *buf,
3584 struct ceph_options **ceph_opts,
3585 struct rbd_options **opts,
3586 struct rbd_spec **rbd_spec)
3590 const char *mon_addrs;
3591 size_t mon_addrs_size;
3592 struct rbd_spec *spec = NULL;
3593 struct rbd_options *rbd_opts = NULL;
3594 struct ceph_options *copts;
3597 /* The first four tokens are required */
3599 len = next_token(&buf);
3601 rbd_warn(NULL, "no monitor address(es) provided");
3605 mon_addrs_size = len + 1;
3609 options = dup_token(&buf, NULL);
3613 rbd_warn(NULL, "no options provided");
3617 spec = rbd_spec_alloc();
3621 spec->pool_name = dup_token(&buf, NULL);
3622 if (!spec->pool_name)
3624 if (!*spec->pool_name) {
3625 rbd_warn(NULL, "no pool name provided");
3629 spec->image_name = dup_token(&buf, NULL);
3630 if (!spec->image_name)
3632 if (!*spec->image_name) {
3633 rbd_warn(NULL, "no image name provided");
3638 * Snapshot name is optional; default is to use "-"
3639 * (indicating the head/no snapshot).
3641 len = next_token(&buf);
3643 buf = RBD_SNAP_HEAD_NAME; /* No snapshot supplied */
3644 len = sizeof (RBD_SNAP_HEAD_NAME) - 1;
3645 } else if (len > RBD_MAX_SNAP_NAME_LEN) {
3646 ret = -ENAMETOOLONG;
3649 spec->snap_name = kmemdup(buf, len + 1, GFP_KERNEL);
3650 if (!spec->snap_name)
3652 *(spec->snap_name + len) = '\0';
3654 /* Initialize all rbd options to the defaults */
3656 rbd_opts = kzalloc(sizeof (*rbd_opts), GFP_KERNEL);
3660 rbd_opts->read_only = RBD_READ_ONLY_DEFAULT;
3662 copts = ceph_parse_options(options, mon_addrs,
3663 mon_addrs + mon_addrs_size - 1,
3664 parse_rbd_opts_token, rbd_opts);
3665 if (IS_ERR(copts)) {
3666 ret = PTR_ERR(copts);
3687 * An rbd format 2 image has a unique identifier, distinct from the
3688 * name given to it by the user. Internally, that identifier is
3689 * what's used to specify the names of objects related to the image.
3691 * A special "rbd id" object is used to map an rbd image name to its
3692 * id. If that object doesn't exist, then there is no v2 rbd image
3693 * with the supplied name.
3695 * This function will record the given rbd_dev's image_id field if
3696 * it can be determined, and in that case will return 0. If any
3697 * errors occur a negative errno will be returned and the rbd_dev's
3698 * image_id field will be unchanged (and should be NULL).
3700 static int rbd_dev_image_id(struct rbd_device *rbd_dev)
3709 * When probing a parent image, the image id is already
3710 * known (and the image name likely is not). There's no
3711 * need to fetch the image id again in this case.
3713 if (rbd_dev->spec->image_id)
3717 * First, see if the format 2 image id file exists, and if
3718 * so, get the image's persistent id from it.
3720 size = sizeof (RBD_ID_PREFIX) + strlen(rbd_dev->spec->image_name);
3721 object_name = kmalloc(size, GFP_NOIO);
3724 sprintf(object_name, "%s%s", RBD_ID_PREFIX, rbd_dev->spec->image_name);
3725 dout("rbd id object name is %s\n", object_name);
3727 /* Response will be an encoded string, which includes a length */
3729 size = sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX;
3730 response = kzalloc(size, GFP_NOIO);
3736 ret = rbd_obj_method_sync(rbd_dev, object_name,
3739 response, RBD_IMAGE_ID_LEN_MAX, NULL);
3740 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
3745 rbd_dev->spec->image_id = ceph_extract_encoded_string(&p,
3746 p + RBD_IMAGE_ID_LEN_MAX,
3748 if (IS_ERR(rbd_dev->spec->image_id)) {
3749 ret = PTR_ERR(rbd_dev->spec->image_id);
3750 rbd_dev->spec->image_id = NULL;
3752 dout("image_id is %s\n", rbd_dev->spec->image_id);
3761 static int rbd_dev_v1_probe(struct rbd_device *rbd_dev)
3766 /* Version 1 images have no id; empty string is used */
3768 rbd_dev->spec->image_id = kstrdup("", GFP_KERNEL);
3769 if (!rbd_dev->spec->image_id)
3772 /* Record the header object name for this rbd image. */
3774 size = strlen(rbd_dev->spec->image_name) + sizeof (RBD_SUFFIX);
3775 rbd_dev->header_name = kmalloc(size, GFP_KERNEL);
3776 if (!rbd_dev->header_name) {
3780 sprintf(rbd_dev->header_name, "%s%s",
3781 rbd_dev->spec->image_name, RBD_SUFFIX);
3783 /* Populate rbd image metadata */
3785 ret = rbd_read_header(rbd_dev, &rbd_dev->header);
3789 /* Version 1 images have no parent (no layering) */
3791 rbd_dev->parent_spec = NULL;
3792 rbd_dev->parent_overlap = 0;
3794 rbd_dev->image_format = 1;
3796 dout("discovered version 1 image, header name is %s\n",
3797 rbd_dev->header_name);
3802 kfree(rbd_dev->header_name);
3803 rbd_dev->header_name = NULL;
3804 kfree(rbd_dev->spec->image_id);
3805 rbd_dev->spec->image_id = NULL;
3810 static int rbd_dev_v2_probe(struct rbd_device *rbd_dev)
3817 * Image id was filled in by the caller. Record the header
3818 * object name for this rbd image.
3820 size = sizeof (RBD_HEADER_PREFIX) + strlen(rbd_dev->spec->image_id);
3821 rbd_dev->header_name = kmalloc(size, GFP_KERNEL);
3822 if (!rbd_dev->header_name)
3824 sprintf(rbd_dev->header_name, "%s%s",
3825 RBD_HEADER_PREFIX, rbd_dev->spec->image_id);
3827 /* Get the size and object order for the image */
3829 ret = rbd_dev_v2_image_size(rbd_dev);
3833 /* Get the object prefix (a.k.a. block_name) for the image */
3835 ret = rbd_dev_v2_object_prefix(rbd_dev);
3839 /* Get the and check features for the image */
3841 ret = rbd_dev_v2_features(rbd_dev);
3845 /* If the image supports layering, get the parent info */
3847 if (rbd_dev->header.features & RBD_FEATURE_LAYERING) {
3848 ret = rbd_dev_v2_parent_info(rbd_dev);
3853 /* crypto and compression type aren't (yet) supported for v2 images */
3855 rbd_dev->header.crypt_type = 0;
3856 rbd_dev->header.comp_type = 0;
3858 /* Get the snapshot context, plus the header version */
3860 ret = rbd_dev_v2_snap_context(rbd_dev, &ver);
3863 rbd_dev->header.obj_version = ver;
3865 rbd_dev->image_format = 2;
3867 dout("discovered version 2 image, header name is %s\n",
3868 rbd_dev->header_name);
3872 rbd_dev->parent_overlap = 0;
3873 rbd_spec_put(rbd_dev->parent_spec);
3874 rbd_dev->parent_spec = NULL;
3875 kfree(rbd_dev->header_name);
3876 rbd_dev->header_name = NULL;
3877 kfree(rbd_dev->header.object_prefix);
3878 rbd_dev->header.object_prefix = NULL;
3883 static int rbd_dev_probe_finish(struct rbd_device *rbd_dev)
3887 /* no need to lock here, as rbd_dev is not registered yet */
3888 ret = rbd_dev_snaps_update(rbd_dev);
3892 ret = rbd_dev_probe_update_spec(rbd_dev);
3896 ret = rbd_dev_set_mapping(rbd_dev);
3900 /* generate unique id: find highest unique id, add one */
3901 rbd_dev_id_get(rbd_dev);
3903 /* Fill in the device name, now that we have its id. */
3904 BUILD_BUG_ON(DEV_NAME_LEN
3905 < sizeof (RBD_DRV_NAME) + MAX_INT_FORMAT_WIDTH);
3906 sprintf(rbd_dev->name, "%s%d", RBD_DRV_NAME, rbd_dev->dev_id);
3908 /* Get our block major device number. */
3910 ret = register_blkdev(0, rbd_dev->name);
3913 rbd_dev->major = ret;
3915 /* Set up the blkdev mapping. */
3917 ret = rbd_init_disk(rbd_dev);
3919 goto err_out_blkdev;
3921 ret = rbd_bus_add_dev(rbd_dev);
3926 * At this point cleanup in the event of an error is the job
3927 * of the sysfs code (initiated by rbd_bus_del_dev()).
3929 down_write(&rbd_dev->header_rwsem);
3930 ret = rbd_dev_snaps_register(rbd_dev);
3931 up_write(&rbd_dev->header_rwsem);
3935 ret = rbd_dev_header_watch_sync(rbd_dev, 1);
3939 /* Everything's ready. Announce the disk to the world. */
3941 add_disk(rbd_dev->disk);
3943 pr_info("%s: added with size 0x%llx\n", rbd_dev->disk->disk_name,
3944 (unsigned long long) rbd_dev->mapping.size);
3948 /* this will also clean up rest of rbd_dev stuff */
3950 rbd_bus_del_dev(rbd_dev);
3954 rbd_free_disk(rbd_dev);
3956 unregister_blkdev(rbd_dev->major, rbd_dev->name);
3958 rbd_dev_id_put(rbd_dev);
3960 rbd_remove_all_snaps(rbd_dev);
3966 * Probe for the existence of the header object for the given rbd
3967 * device. For format 2 images this includes determining the image
3970 static int rbd_dev_probe(struct rbd_device *rbd_dev)
3975 * Get the id from the image id object. If it's not a
3976 * format 2 image, we'll get ENOENT back, and we'll assume
3977 * it's a format 1 image.
3979 ret = rbd_dev_image_id(rbd_dev);
3981 ret = rbd_dev_v1_probe(rbd_dev);
3983 ret = rbd_dev_v2_probe(rbd_dev);
3985 dout("probe failed, returning %d\n", ret);
3990 ret = rbd_dev_probe_finish(rbd_dev);
3992 rbd_header_free(&rbd_dev->header);
3997 static ssize_t rbd_add(struct bus_type *bus,
4001 struct rbd_device *rbd_dev = NULL;
4002 struct ceph_options *ceph_opts = NULL;
4003 struct rbd_options *rbd_opts = NULL;
4004 struct rbd_spec *spec = NULL;
4005 struct rbd_client *rbdc;
4006 struct ceph_osd_client *osdc;
4009 if (!try_module_get(THIS_MODULE))
4012 /* parse add command */
4013 rc = rbd_add_parse_args(buf, &ceph_opts, &rbd_opts, &spec);
4015 goto err_out_module;
4017 rbdc = rbd_get_client(ceph_opts);
4022 ceph_opts = NULL; /* rbd_dev client now owns this */
4025 osdc = &rbdc->client->osdc;
4026 rc = ceph_pg_poolid_by_name(osdc->osdmap, spec->pool_name);
4028 goto err_out_client;
4029 spec->pool_id = (u64) rc;
4031 /* The ceph file layout needs to fit pool id in 32 bits */
4033 if (WARN_ON(spec->pool_id > (u64) U32_MAX)) {
4035 goto err_out_client;
4038 rbd_dev = rbd_dev_create(rbdc, spec);
4040 goto err_out_client;
4041 rbdc = NULL; /* rbd_dev now owns this */
4042 spec = NULL; /* rbd_dev now owns this */
4044 rbd_dev->mapping.read_only = rbd_opts->read_only;
4046 rbd_opts = NULL; /* done with this */
4048 rc = rbd_dev_probe(rbd_dev);
4050 goto err_out_rbd_dev;
4054 rbd_dev_destroy(rbd_dev);
4056 rbd_put_client(rbdc);
4059 ceph_destroy_options(ceph_opts);
4063 module_put(THIS_MODULE);
4065 dout("Error adding device %s\n", buf);
4067 return (ssize_t) rc;
4070 static struct rbd_device *__rbd_get_dev(unsigned long dev_id)
4072 struct list_head *tmp;
4073 struct rbd_device *rbd_dev;
4075 spin_lock(&rbd_dev_list_lock);
4076 list_for_each(tmp, &rbd_dev_list) {
4077 rbd_dev = list_entry(tmp, struct rbd_device, node);
4078 if (rbd_dev->dev_id == dev_id) {
4079 spin_unlock(&rbd_dev_list_lock);
4083 spin_unlock(&rbd_dev_list_lock);
4087 static void rbd_dev_release(struct device *dev)
4089 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
4091 if (rbd_dev->watch_event)
4092 rbd_dev_header_watch_sync(rbd_dev, 0);
4094 /* clean up and free blkdev */
4095 rbd_free_disk(rbd_dev);
4096 unregister_blkdev(rbd_dev->major, rbd_dev->name);
4098 /* release allocated disk header fields */
4099 rbd_header_free(&rbd_dev->header);
4101 /* done with the id, and with the rbd_dev */
4102 rbd_dev_id_put(rbd_dev);
4103 rbd_assert(rbd_dev->rbd_client != NULL);
4104 rbd_dev_destroy(rbd_dev);
4106 /* release module ref */
4107 module_put(THIS_MODULE);
4110 static ssize_t rbd_remove(struct bus_type *bus,
4114 struct rbd_device *rbd_dev = NULL;
4119 rc = strict_strtoul(buf, 10, &ul);
4123 /* convert to int; abort if we lost anything in the conversion */
4124 target_id = (int) ul;
4125 if (target_id != ul)
4128 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
4130 rbd_dev = __rbd_get_dev(target_id);
4136 spin_lock_irq(&rbd_dev->lock);
4137 if (rbd_dev->open_count)
4140 set_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags);
4141 spin_unlock_irq(&rbd_dev->lock);
4145 rbd_remove_all_snaps(rbd_dev);
4146 rbd_bus_del_dev(rbd_dev);
4149 mutex_unlock(&ctl_mutex);
4155 * create control files in sysfs
4158 static int rbd_sysfs_init(void)
4162 ret = device_register(&rbd_root_dev);
4166 ret = bus_register(&rbd_bus_type);
4168 device_unregister(&rbd_root_dev);
4173 static void rbd_sysfs_cleanup(void)
4175 bus_unregister(&rbd_bus_type);
4176 device_unregister(&rbd_root_dev);
4179 static int __init rbd_init(void)
4183 if (!libceph_compatible(NULL)) {
4184 rbd_warn(NULL, "libceph incompatibility (quitting)");
4188 rc = rbd_sysfs_init();
4191 pr_info("loaded " RBD_DRV_NAME_LONG "\n");
4195 static void __exit rbd_exit(void)
4197 rbd_sysfs_cleanup();
4200 module_init(rbd_init);
4201 module_exit(rbd_exit);
4203 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
4204 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
4205 MODULE_DESCRIPTION("rados block device");
4207 /* following authorship retained from original osdblk.c */
4208 MODULE_AUTHOR("Jeff Garzik <jeff@garzik.org>");
4210 MODULE_LICENSE("GPL");