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"
45 * The basic unit of block I/O is a sector. It is interpreted in a
46 * number of contexts in Linux (blk, bio, genhd), but the default is
47 * universally 512 bytes. These symbols are just slightly more
48 * meaningful than the bare numbers they represent.
50 #define SECTOR_SHIFT 9
51 #define SECTOR_SIZE (1ULL << SECTOR_SHIFT)
53 #define RBD_DRV_NAME "rbd"
54 #define RBD_DRV_NAME_LONG "rbd (rados block device)"
56 #define RBD_MINORS_PER_MAJOR 256 /* max minors per blkdev */
58 #define RBD_MAX_SNAP_NAME_LEN 32
59 #define RBD_MAX_OPT_LEN 1024
61 #define RBD_SNAP_HEAD_NAME "-"
64 * An RBD device name will be "rbd#", where the "rbd" comes from
65 * RBD_DRV_NAME above, and # is a unique integer identifier.
66 * MAX_INT_FORMAT_WIDTH is used in ensuring DEV_NAME_LEN is big
67 * enough to hold all possible device names.
69 #define DEV_NAME_LEN 32
70 #define MAX_INT_FORMAT_WIDTH ((5 * sizeof (int)) / 2 + 1)
72 #define RBD_NOTIFY_TIMEOUT_DEFAULT 10
75 * block device image metadata (in-memory version)
77 struct rbd_image_header {
83 struct ceph_snap_context *snapc;
84 size_t snap_names_len;
98 * an instance of the client. multiple devices may share an rbd client.
101 struct ceph_client *client;
102 struct rbd_options *rbd_opts;
104 struct list_head node;
108 * a request completion status
110 struct rbd_req_status {
117 * a collection of requests
119 struct rbd_req_coll {
123 struct rbd_req_status status[0];
127 * a single io request
130 struct request *rq; /* blk layer request */
131 struct bio *bio; /* cloned bio */
132 struct page **pages; /* list of used pages */
135 struct rbd_req_coll *coll;
142 struct list_head node;
150 int id; /* blkdev unique id */
152 int major; /* blkdev assigned major */
153 struct gendisk *disk; /* blkdev's gendisk and rq */
154 struct request_queue *q;
156 struct rbd_client *rbd_client;
158 char name[DEV_NAME_LEN]; /* blkdev name, e.g. rbd3 */
160 spinlock_t lock; /* queue lock */
162 struct rbd_image_header header;
164 size_t image_name_len;
169 struct ceph_osd_event *watch_event;
170 struct ceph_osd_request *watch_request;
172 /* protects updating the header */
173 struct rw_semaphore header_rwsem;
174 /* name of the snapshot this device reads from */
176 /* id of the snapshot this device reads from */
177 u64 snap_id; /* current snapshot id */
178 /* whether the snap_id this device reads from still exists */
182 struct list_head node;
184 /* list of snapshots */
185 struct list_head snaps;
191 static DEFINE_MUTEX(ctl_mutex); /* Serialize open/close/setup/teardown */
193 static LIST_HEAD(rbd_dev_list); /* devices */
194 static DEFINE_SPINLOCK(rbd_dev_list_lock);
196 static LIST_HEAD(rbd_client_list); /* clients */
197 static DEFINE_SPINLOCK(rbd_client_list_lock);
199 static int __rbd_init_snaps_header(struct rbd_device *rbd_dev);
200 static void rbd_dev_release(struct device *dev);
201 static ssize_t rbd_snap_add(struct device *dev,
202 struct device_attribute *attr,
205 static void __rbd_remove_snap_dev(struct rbd_device *rbd_dev,
206 struct rbd_snap *snap);
208 static ssize_t rbd_add(struct bus_type *bus, const char *buf,
210 static ssize_t rbd_remove(struct bus_type *bus, const char *buf,
213 static struct bus_attribute rbd_bus_attrs[] = {
214 __ATTR(add, S_IWUSR, NULL, rbd_add),
215 __ATTR(remove, S_IWUSR, NULL, rbd_remove),
219 static struct bus_type rbd_bus_type = {
221 .bus_attrs = rbd_bus_attrs,
224 static void rbd_root_dev_release(struct device *dev)
228 static struct device rbd_root_dev = {
230 .release = rbd_root_dev_release,
234 static struct device *rbd_get_dev(struct rbd_device *rbd_dev)
236 return get_device(&rbd_dev->dev);
239 static void rbd_put_dev(struct rbd_device *rbd_dev)
241 put_device(&rbd_dev->dev);
244 static int __rbd_refresh_header(struct rbd_device *rbd_dev);
246 static int rbd_open(struct block_device *bdev, fmode_t mode)
248 struct rbd_device *rbd_dev = bdev->bd_disk->private_data;
250 rbd_get_dev(rbd_dev);
252 set_device_ro(bdev, rbd_dev->read_only);
254 if ((mode & FMODE_WRITE) && rbd_dev->read_only)
260 static int rbd_release(struct gendisk *disk, fmode_t mode)
262 struct rbd_device *rbd_dev = disk->private_data;
264 rbd_put_dev(rbd_dev);
269 static const struct block_device_operations rbd_bd_ops = {
270 .owner = THIS_MODULE,
272 .release = rbd_release,
276 * Initialize an rbd client instance.
279 static struct rbd_client *rbd_client_create(struct ceph_options *ceph_opts,
280 struct rbd_options *rbd_opts)
282 struct rbd_client *rbdc;
285 dout("rbd_client_create\n");
286 rbdc = kmalloc(sizeof(struct rbd_client), GFP_KERNEL);
290 kref_init(&rbdc->kref);
291 INIT_LIST_HEAD(&rbdc->node);
293 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
295 rbdc->client = ceph_create_client(ceph_opts, rbdc, 0, 0);
296 if (IS_ERR(rbdc->client))
298 ceph_opts = NULL; /* Now rbdc->client is responsible for ceph_opts */
300 ret = ceph_open_session(rbdc->client);
304 rbdc->rbd_opts = rbd_opts;
306 spin_lock(&rbd_client_list_lock);
307 list_add_tail(&rbdc->node, &rbd_client_list);
308 spin_unlock(&rbd_client_list_lock);
310 mutex_unlock(&ctl_mutex);
312 dout("rbd_client_create created %p\n", rbdc);
316 ceph_destroy_client(rbdc->client);
318 mutex_unlock(&ctl_mutex);
322 ceph_destroy_options(ceph_opts);
327 * Find a ceph client with specific addr and configuration.
329 static struct rbd_client *__rbd_client_find(struct ceph_options *ceph_opts)
331 struct rbd_client *client_node;
333 if (ceph_opts->flags & CEPH_OPT_NOSHARE)
336 list_for_each_entry(client_node, &rbd_client_list, node)
337 if (!ceph_compare_options(ceph_opts, client_node->client))
350 /* string args above */
353 static match_table_t rbd_opts_tokens = {
354 {Opt_notify_timeout, "notify_timeout=%d"},
356 /* string args above */
360 static int parse_rbd_opts_token(char *c, void *private)
362 struct rbd_options *rbd_opts = private;
363 substring_t argstr[MAX_OPT_ARGS];
364 int token, intval, ret;
366 token = match_token(c, rbd_opts_tokens, argstr);
370 if (token < Opt_last_int) {
371 ret = match_int(&argstr[0], &intval);
373 pr_err("bad mount option arg (not int) "
377 dout("got int token %d val %d\n", token, intval);
378 } else if (token > Opt_last_int && token < Opt_last_string) {
379 dout("got string token %d val %s\n", token,
382 dout("got token %d\n", token);
386 case Opt_notify_timeout:
387 rbd_opts->notify_timeout = intval;
396 * Get a ceph client with specific addr and configuration, if one does
397 * not exist create it.
399 static struct rbd_client *rbd_get_client(const char *mon_addr,
403 struct rbd_client *rbdc;
404 struct ceph_options *ceph_opts;
405 struct rbd_options *rbd_opts;
407 rbd_opts = kzalloc(sizeof(*rbd_opts), GFP_KERNEL);
409 return ERR_PTR(-ENOMEM);
411 rbd_opts->notify_timeout = RBD_NOTIFY_TIMEOUT_DEFAULT;
413 ceph_opts = ceph_parse_options(options, mon_addr,
414 mon_addr + mon_addr_len,
415 parse_rbd_opts_token, rbd_opts);
416 if (IS_ERR(ceph_opts)) {
418 return ERR_CAST(ceph_opts);
421 spin_lock(&rbd_client_list_lock);
422 rbdc = __rbd_client_find(ceph_opts);
424 /* using an existing client */
425 kref_get(&rbdc->kref);
426 spin_unlock(&rbd_client_list_lock);
428 ceph_destroy_options(ceph_opts);
433 spin_unlock(&rbd_client_list_lock);
435 rbdc = rbd_client_create(ceph_opts, rbd_opts);
444 * Destroy ceph client
446 * Caller must hold rbd_client_list_lock.
448 static void rbd_client_release(struct kref *kref)
450 struct rbd_client *rbdc = container_of(kref, struct rbd_client, kref);
452 dout("rbd_release_client %p\n", rbdc);
453 spin_lock(&rbd_client_list_lock);
454 list_del(&rbdc->node);
455 spin_unlock(&rbd_client_list_lock);
457 ceph_destroy_client(rbdc->client);
458 kfree(rbdc->rbd_opts);
463 * Drop reference to ceph client node. If it's not referenced anymore, release
466 static void rbd_put_client(struct rbd_device *rbd_dev)
468 kref_put(&rbd_dev->rbd_client->kref, rbd_client_release);
469 rbd_dev->rbd_client = NULL;
473 * Destroy requests collection
475 static void rbd_coll_release(struct kref *kref)
477 struct rbd_req_coll *coll =
478 container_of(kref, struct rbd_req_coll, kref);
480 dout("rbd_coll_release %p\n", coll);
485 * Create a new header structure, translate header format from the on-disk
488 static int rbd_header_from_disk(struct rbd_image_header *header,
489 struct rbd_image_header_ondisk *ondisk,
495 if (memcmp(ondisk, RBD_HEADER_TEXT, sizeof(RBD_HEADER_TEXT)))
498 snap_count = le32_to_cpu(ondisk->snap_count);
499 if (snap_count > (UINT_MAX - sizeof(struct ceph_snap_context))
502 header->snapc = kmalloc(sizeof(struct ceph_snap_context) +
503 snap_count * sizeof(u64),
508 header->snap_names_len = le64_to_cpu(ondisk->snap_names_len);
510 header->snap_names = kmalloc(header->snap_names_len,
512 if (!header->snap_names)
514 header->snap_sizes = kmalloc(snap_count * sizeof(u64),
516 if (!header->snap_sizes)
519 header->snap_names = NULL;
520 header->snap_sizes = NULL;
523 header->object_prefix = kmalloc(sizeof (ondisk->block_name) + 1,
525 if (!header->object_prefix)
528 memcpy(header->object_prefix, ondisk->block_name,
529 sizeof(ondisk->block_name));
530 header->object_prefix[sizeof (ondisk->block_name)] = '\0';
532 header->image_size = le64_to_cpu(ondisk->image_size);
533 header->obj_order = ondisk->options.order;
534 header->crypt_type = ondisk->options.crypt_type;
535 header->comp_type = ondisk->options.comp_type;
537 atomic_set(&header->snapc->nref, 1);
538 header->snapc->seq = le64_to_cpu(ondisk->snap_seq);
539 header->snapc->num_snaps = snap_count;
540 header->total_snaps = snap_count;
542 if (snap_count && allocated_snaps == snap_count) {
543 for (i = 0; i < snap_count; i++) {
544 header->snapc->snaps[i] =
545 le64_to_cpu(ondisk->snaps[i].id);
546 header->snap_sizes[i] =
547 le64_to_cpu(ondisk->snaps[i].image_size);
550 /* copy snapshot names */
551 memcpy(header->snap_names, &ondisk->snaps[i],
552 header->snap_names_len);
558 kfree(header->snap_sizes);
560 kfree(header->snap_names);
562 kfree(header->snapc);
566 static int snap_by_name(struct rbd_image_header *header, const char *snap_name,
570 char *p = header->snap_names;
572 for (i = 0; i < header->total_snaps; i++) {
573 if (!strcmp(snap_name, p)) {
575 /* Found it. Pass back its id and/or size */
578 *seq = header->snapc->snaps[i];
580 *size = header->snap_sizes[i];
583 p += strlen(p) + 1; /* Skip ahead to the next name */
588 static int rbd_header_set_snap(struct rbd_device *rbd_dev, u64 *size)
592 down_write(&rbd_dev->header_rwsem);
594 if (!memcmp(rbd_dev->snap_name, RBD_SNAP_HEAD_NAME,
595 sizeof (RBD_SNAP_HEAD_NAME))) {
596 rbd_dev->snap_id = CEPH_NOSNAP;
597 rbd_dev->snap_exists = false;
598 rbd_dev->read_only = 0;
600 *size = rbd_dev->header.image_size;
604 ret = snap_by_name(&rbd_dev->header, rbd_dev->snap_name,
608 rbd_dev->snap_id = snap_id;
609 rbd_dev->snap_exists = true;
610 rbd_dev->read_only = 1;
615 up_write(&rbd_dev->header_rwsem);
619 static void rbd_header_free(struct rbd_image_header *header)
621 kfree(header->object_prefix);
622 kfree(header->snap_sizes);
623 kfree(header->snap_names);
624 ceph_put_snap_context(header->snapc);
628 * get the actual striped segment name, offset and length
630 static u64 rbd_get_segment(struct rbd_image_header *header,
631 const char *object_prefix,
633 char *seg_name, u64 *segofs)
635 u64 seg = ofs >> header->obj_order;
638 snprintf(seg_name, RBD_MAX_SEG_NAME_LEN,
639 "%s.%012llx", object_prefix, seg);
641 ofs = ofs & ((1 << header->obj_order) - 1);
642 len = min_t(u64, len, (1 << header->obj_order) - ofs);
650 static int rbd_get_num_segments(struct rbd_image_header *header,
653 u64 start_seg = ofs >> header->obj_order;
654 u64 end_seg = (ofs + len - 1) >> header->obj_order;
655 return end_seg - start_seg + 1;
659 * returns the size of an object in the image
661 static u64 rbd_obj_bytes(struct rbd_image_header *header)
663 return 1 << header->obj_order;
670 static void bio_chain_put(struct bio *chain)
676 chain = chain->bi_next;
682 * zeros a bio chain, starting at specific offset
684 static void zero_bio_chain(struct bio *chain, int start_ofs)
693 bio_for_each_segment(bv, chain, i) {
694 if (pos + bv->bv_len > start_ofs) {
695 int remainder = max(start_ofs - pos, 0);
696 buf = bvec_kmap_irq(bv, &flags);
697 memset(buf + remainder, 0,
698 bv->bv_len - remainder);
699 bvec_kunmap_irq(buf, &flags);
704 chain = chain->bi_next;
709 * bio_chain_clone - clone a chain of bios up to a certain length.
710 * might return a bio_pair that will need to be released.
712 static struct bio *bio_chain_clone(struct bio **old, struct bio **next,
713 struct bio_pair **bp,
714 int len, gfp_t gfpmask)
716 struct bio *tmp, *old_chain = *old, *new_chain = NULL, *tail = NULL;
720 bio_pair_release(*bp);
724 while (old_chain && (total < len)) {
725 tmp = bio_kmalloc(gfpmask, old_chain->bi_max_vecs);
729 if (total + old_chain->bi_size > len) {
733 * this split can only happen with a single paged bio,
734 * split_bio will BUG_ON if this is not the case
736 dout("bio_chain_clone split! total=%d remaining=%d"
738 total, len - total, old_chain->bi_size);
740 /* split the bio. We'll release it either in the next
741 call, or it will have to be released outside */
742 bp = bio_split(old_chain, (len - total) / SECTOR_SIZE);
746 __bio_clone(tmp, &bp->bio1);
750 __bio_clone(tmp, old_chain);
751 *next = old_chain->bi_next;
755 gfpmask &= ~__GFP_WAIT;
759 new_chain = tail = tmp;
764 old_chain = old_chain->bi_next;
766 total += tmp->bi_size;
772 tail->bi_next = NULL;
779 dout("bio_chain_clone with err\n");
780 bio_chain_put(new_chain);
785 * helpers for osd request op vectors.
787 static int rbd_create_rw_ops(struct ceph_osd_req_op **ops,
792 *ops = kzalloc(sizeof(struct ceph_osd_req_op) * (num_ops + 1),
796 (*ops)[0].op = opcode;
798 * op extent offset and length will be set later on
799 * in calc_raw_layout()
801 (*ops)[0].payload_len = payload_len;
805 static void rbd_destroy_ops(struct ceph_osd_req_op *ops)
810 static void rbd_coll_end_req_index(struct request *rq,
811 struct rbd_req_coll *coll,
815 struct request_queue *q;
818 dout("rbd_coll_end_req_index %p index %d ret %d len %llu\n",
819 coll, index, ret, (unsigned long long) len);
825 blk_end_request(rq, ret, len);
831 spin_lock_irq(q->queue_lock);
832 coll->status[index].done = 1;
833 coll->status[index].rc = ret;
834 coll->status[index].bytes = len;
835 max = min = coll->num_done;
836 while (max < coll->total && coll->status[max].done)
839 for (i = min; i<max; i++) {
840 __blk_end_request(rq, coll->status[i].rc,
841 coll->status[i].bytes);
843 kref_put(&coll->kref, rbd_coll_release);
845 spin_unlock_irq(q->queue_lock);
848 static void rbd_coll_end_req(struct rbd_request *req,
851 rbd_coll_end_req_index(req->rq, req->coll, req->coll_index, ret, len);
855 * Send ceph osd request
857 static int rbd_do_request(struct request *rq,
858 struct rbd_device *rbd_dev,
859 struct ceph_snap_context *snapc,
861 const char *object_name, u64 ofs, u64 len,
866 struct ceph_osd_req_op *ops,
867 struct rbd_req_coll *coll,
869 void (*rbd_cb)(struct ceph_osd_request *req,
870 struct ceph_msg *msg),
871 struct ceph_osd_request **linger_req,
874 struct ceph_osd_request *req;
875 struct ceph_file_layout *layout;
878 struct timespec mtime = CURRENT_TIME;
879 struct rbd_request *req_data;
880 struct ceph_osd_request_head *reqhead;
881 struct ceph_osd_client *osdc;
883 req_data = kzalloc(sizeof(*req_data), GFP_NOIO);
886 rbd_coll_end_req_index(rq, coll, coll_index,
892 req_data->coll = coll;
893 req_data->coll_index = coll_index;
896 dout("rbd_do_request object_name=%s ofs=%llu len=%llu\n", object_name,
897 (unsigned long long) ofs, (unsigned long long) len);
899 osdc = &rbd_dev->rbd_client->client->osdc;
900 req = ceph_osdc_alloc_request(osdc, flags, snapc, ops,
901 false, GFP_NOIO, pages, bio);
907 req->r_callback = rbd_cb;
911 req_data->pages = pages;
914 req->r_priv = req_data;
916 reqhead = req->r_request->front.iov_base;
917 reqhead->snapid = cpu_to_le64(CEPH_NOSNAP);
919 strncpy(req->r_oid, object_name, sizeof(req->r_oid));
920 req->r_oid_len = strlen(req->r_oid);
922 layout = &req->r_file_layout;
923 memset(layout, 0, sizeof(*layout));
924 layout->fl_stripe_unit = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER);
925 layout->fl_stripe_count = cpu_to_le32(1);
926 layout->fl_object_size = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER);
927 layout->fl_pg_pool = cpu_to_le32(rbd_dev->pool_id);
928 ceph_calc_raw_layout(osdc, layout, snapid, ofs, &len, &bno,
931 ceph_osdc_build_request(req, ofs, &len,
935 req->r_oid, req->r_oid_len);
938 ceph_osdc_set_request_linger(osdc, req);
942 ret = ceph_osdc_start_request(osdc, req, false);
947 ret = ceph_osdc_wait_request(osdc, req);
949 *ver = le64_to_cpu(req->r_reassert_version.version);
950 dout("reassert_ver=%llu\n",
952 le64_to_cpu(req->r_reassert_version.version));
953 ceph_osdc_put_request(req);
958 bio_chain_put(req_data->bio);
959 ceph_osdc_put_request(req);
961 rbd_coll_end_req(req_data, ret, len);
967 * Ceph osd op callback
969 static void rbd_req_cb(struct ceph_osd_request *req, struct ceph_msg *msg)
971 struct rbd_request *req_data = req->r_priv;
972 struct ceph_osd_reply_head *replyhead;
973 struct ceph_osd_op *op;
979 replyhead = msg->front.iov_base;
980 WARN_ON(le32_to_cpu(replyhead->num_ops) == 0);
981 op = (void *)(replyhead + 1);
982 rc = le32_to_cpu(replyhead->result);
983 bytes = le64_to_cpu(op->extent.length);
984 read_op = (le16_to_cpu(op->op) == CEPH_OSD_OP_READ);
986 dout("rbd_req_cb bytes=%llu readop=%d rc=%d\n",
987 (unsigned long long) bytes, read_op, (int) rc);
989 if (rc == -ENOENT && read_op) {
990 zero_bio_chain(req_data->bio, 0);
992 } else if (rc == 0 && read_op && bytes < req_data->len) {
993 zero_bio_chain(req_data->bio, bytes);
994 bytes = req_data->len;
997 rbd_coll_end_req(req_data, rc, bytes);
1000 bio_chain_put(req_data->bio);
1002 ceph_osdc_put_request(req);
1006 static void rbd_simple_req_cb(struct ceph_osd_request *req, struct ceph_msg *msg)
1008 ceph_osdc_put_request(req);
1012 * Do a synchronous ceph osd operation
1014 static int rbd_req_sync_op(struct rbd_device *rbd_dev,
1015 struct ceph_snap_context *snapc,
1019 struct ceph_osd_req_op *orig_ops,
1020 const char *object_name,
1023 struct ceph_osd_request **linger_req,
1027 struct page **pages;
1029 struct ceph_osd_req_op *ops = orig_ops;
1032 num_pages = calc_pages_for(ofs , len);
1033 pages = ceph_alloc_page_vector(num_pages, GFP_KERNEL);
1035 return PTR_ERR(pages);
1038 payload_len = (flags & CEPH_OSD_FLAG_WRITE ? len : 0);
1039 ret = rbd_create_rw_ops(&ops, 1, opcode, payload_len);
1043 if ((flags & CEPH_OSD_FLAG_WRITE) && buf) {
1044 ret = ceph_copy_to_page_vector(pages, buf, ofs, len);
1050 ret = rbd_do_request(NULL, rbd_dev, snapc, snapid,
1051 object_name, ofs, len, NULL,
1061 if ((flags & CEPH_OSD_FLAG_READ) && buf)
1062 ret = ceph_copy_from_page_vector(pages, buf, ofs, ret);
1066 rbd_destroy_ops(ops);
1068 ceph_release_page_vector(pages, num_pages);
1073 * Do an asynchronous ceph osd operation
1075 static int rbd_do_op(struct request *rq,
1076 struct rbd_device *rbd_dev,
1077 struct ceph_snap_context *snapc,
1079 int opcode, int flags,
1082 struct rbd_req_coll *coll,
1089 struct ceph_osd_req_op *ops;
1092 seg_name = kmalloc(RBD_MAX_SEG_NAME_LEN + 1, GFP_NOIO);
1096 seg_len = rbd_get_segment(&rbd_dev->header,
1097 rbd_dev->header.object_prefix,
1099 seg_name, &seg_ofs);
1101 payload_len = (flags & CEPH_OSD_FLAG_WRITE ? seg_len : 0);
1103 ret = rbd_create_rw_ops(&ops, 1, opcode, payload_len);
1107 /* we've taken care of segment sizes earlier when we
1108 cloned the bios. We should never have a segment
1109 truncated at this point */
1110 BUG_ON(seg_len < len);
1112 ret = rbd_do_request(rq, rbd_dev, snapc, snapid,
1113 seg_name, seg_ofs, seg_len,
1119 rbd_req_cb, 0, NULL);
1121 rbd_destroy_ops(ops);
1128 * Request async osd write
1130 static int rbd_req_write(struct request *rq,
1131 struct rbd_device *rbd_dev,
1132 struct ceph_snap_context *snapc,
1135 struct rbd_req_coll *coll,
1138 return rbd_do_op(rq, rbd_dev, snapc, CEPH_NOSNAP,
1140 CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK,
1141 ofs, len, bio, coll, coll_index);
1145 * Request async osd read
1147 static int rbd_req_read(struct request *rq,
1148 struct rbd_device *rbd_dev,
1152 struct rbd_req_coll *coll,
1155 return rbd_do_op(rq, rbd_dev, NULL,
1159 ofs, len, bio, coll, coll_index);
1163 * Request sync osd read
1165 static int rbd_req_sync_read(struct rbd_device *rbd_dev,
1166 struct ceph_snap_context *snapc,
1168 const char *object_name,
1173 return rbd_req_sync_op(rbd_dev, NULL,
1178 object_name, ofs, len, buf, NULL, ver);
1182 * Request sync osd watch
1184 static int rbd_req_sync_notify_ack(struct rbd_device *rbd_dev,
1187 const char *object_name)
1189 struct ceph_osd_req_op *ops;
1192 ret = rbd_create_rw_ops(&ops, 1, CEPH_OSD_OP_NOTIFY_ACK, 0);
1196 ops[0].watch.ver = cpu_to_le64(ver);
1197 ops[0].watch.cookie = notify_id;
1198 ops[0].watch.flag = 0;
1200 ret = rbd_do_request(NULL, rbd_dev, NULL, CEPH_NOSNAP,
1201 object_name, 0, 0, NULL,
1206 rbd_simple_req_cb, 0, NULL);
1208 rbd_destroy_ops(ops);
1212 static void rbd_watch_cb(u64 ver, u64 notify_id, u8 opcode, void *data)
1214 struct rbd_device *rbd_dev = (struct rbd_device *)data;
1221 dout("rbd_watch_cb %s notify_id=%llu opcode=%u\n",
1222 rbd_dev->header_name, (unsigned long long) notify_id,
1223 (unsigned int) opcode);
1224 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
1225 rc = __rbd_refresh_header(rbd_dev);
1226 hver = rbd_dev->header.obj_version;
1227 mutex_unlock(&ctl_mutex);
1229 pr_warning(RBD_DRV_NAME "%d got notification but failed to "
1230 " update snaps: %d\n", rbd_dev->major, rc);
1232 rbd_req_sync_notify_ack(rbd_dev, hver, notify_id, rbd_dev->header_name);
1236 * Request sync osd watch
1238 static int rbd_req_sync_watch(struct rbd_device *rbd_dev,
1239 const char *object_name,
1242 struct ceph_osd_req_op *ops;
1243 struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
1245 int ret = rbd_create_rw_ops(&ops, 1, CEPH_OSD_OP_WATCH, 0);
1249 ret = ceph_osdc_create_event(osdc, rbd_watch_cb, 0,
1250 (void *)rbd_dev, &rbd_dev->watch_event);
1254 ops[0].watch.ver = cpu_to_le64(ver);
1255 ops[0].watch.cookie = cpu_to_le64(rbd_dev->watch_event->cookie);
1256 ops[0].watch.flag = 1;
1258 ret = rbd_req_sync_op(rbd_dev, NULL,
1261 CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK,
1263 object_name, 0, 0, NULL,
1264 &rbd_dev->watch_request, NULL);
1269 rbd_destroy_ops(ops);
1273 ceph_osdc_cancel_event(rbd_dev->watch_event);
1274 rbd_dev->watch_event = NULL;
1276 rbd_destroy_ops(ops);
1281 * Request sync osd unwatch
1283 static int rbd_req_sync_unwatch(struct rbd_device *rbd_dev,
1284 const char *object_name)
1286 struct ceph_osd_req_op *ops;
1288 int ret = rbd_create_rw_ops(&ops, 1, CEPH_OSD_OP_WATCH, 0);
1292 ops[0].watch.ver = 0;
1293 ops[0].watch.cookie = cpu_to_le64(rbd_dev->watch_event->cookie);
1294 ops[0].watch.flag = 0;
1296 ret = rbd_req_sync_op(rbd_dev, NULL,
1299 CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK,
1301 object_name, 0, 0, NULL, NULL, NULL);
1303 rbd_destroy_ops(ops);
1304 ceph_osdc_cancel_event(rbd_dev->watch_event);
1305 rbd_dev->watch_event = NULL;
1309 struct rbd_notify_info {
1310 struct rbd_device *rbd_dev;
1313 static void rbd_notify_cb(u64 ver, u64 notify_id, u8 opcode, void *data)
1315 struct rbd_device *rbd_dev = (struct rbd_device *)data;
1319 dout("rbd_notify_cb %s notify_id=%llu opcode=%u\n",
1320 rbd_dev->header_name, (unsigned long long) notify_id,
1321 (unsigned int) opcode);
1325 * Request sync osd notify
1327 static int rbd_req_sync_notify(struct rbd_device *rbd_dev,
1328 const char *object_name)
1330 struct ceph_osd_req_op *ops;
1331 struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
1332 struct ceph_osd_event *event;
1333 struct rbd_notify_info info;
1334 int payload_len = sizeof(u32) + sizeof(u32);
1337 ret = rbd_create_rw_ops(&ops, 1, CEPH_OSD_OP_NOTIFY, payload_len);
1341 info.rbd_dev = rbd_dev;
1343 ret = ceph_osdc_create_event(osdc, rbd_notify_cb, 1,
1344 (void *)&info, &event);
1348 ops[0].watch.ver = 1;
1349 ops[0].watch.flag = 1;
1350 ops[0].watch.cookie = event->cookie;
1351 ops[0].watch.prot_ver = RADOS_NOTIFY_VER;
1352 ops[0].watch.timeout = 12;
1354 ret = rbd_req_sync_op(rbd_dev, NULL,
1357 CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK,
1359 object_name, 0, 0, NULL, NULL, NULL);
1363 ret = ceph_osdc_wait_event(event, CEPH_OSD_TIMEOUT_DEFAULT);
1364 dout("ceph_osdc_wait_event returned %d\n", ret);
1365 rbd_destroy_ops(ops);
1369 ceph_osdc_cancel_event(event);
1371 rbd_destroy_ops(ops);
1376 * Request sync osd read
1378 static int rbd_req_sync_exec(struct rbd_device *rbd_dev,
1379 const char *object_name,
1380 const char *class_name,
1381 const char *method_name,
1386 struct ceph_osd_req_op *ops;
1387 int class_name_len = strlen(class_name);
1388 int method_name_len = strlen(method_name);
1389 int ret = rbd_create_rw_ops(&ops, 1, CEPH_OSD_OP_CALL,
1390 class_name_len + method_name_len + len);
1394 ops[0].cls.class_name = class_name;
1395 ops[0].cls.class_len = (__u8) class_name_len;
1396 ops[0].cls.method_name = method_name;
1397 ops[0].cls.method_len = (__u8) method_name_len;
1398 ops[0].cls.argc = 0;
1399 ops[0].cls.indata = data;
1400 ops[0].cls.indata_len = len;
1402 ret = rbd_req_sync_op(rbd_dev, NULL,
1405 CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK,
1407 object_name, 0, 0, NULL, NULL, ver);
1409 rbd_destroy_ops(ops);
1411 dout("cls_exec returned %d\n", ret);
1415 static struct rbd_req_coll *rbd_alloc_coll(int num_reqs)
1417 struct rbd_req_coll *coll =
1418 kzalloc(sizeof(struct rbd_req_coll) +
1419 sizeof(struct rbd_req_status) * num_reqs,
1424 coll->total = num_reqs;
1425 kref_init(&coll->kref);
1430 * block device queue callback
1432 static void rbd_rq_fn(struct request_queue *q)
1434 struct rbd_device *rbd_dev = q->queuedata;
1436 struct bio_pair *bp = NULL;
1438 while ((rq = blk_fetch_request(q))) {
1440 struct bio *rq_bio, *next_bio = NULL;
1445 int num_segs, cur_seg = 0;
1446 struct rbd_req_coll *coll;
1447 struct ceph_snap_context *snapc;
1449 /* peek at request from block layer */
1453 dout("fetched request\n");
1455 /* filter out block requests we don't understand */
1456 if ((rq->cmd_type != REQ_TYPE_FS)) {
1457 __blk_end_request_all(rq, 0);
1461 /* deduce our operation (read, write) */
1462 do_write = (rq_data_dir(rq) == WRITE);
1464 size = blk_rq_bytes(rq);
1465 ofs = blk_rq_pos(rq) * SECTOR_SIZE;
1467 if (do_write && rbd_dev->read_only) {
1468 __blk_end_request_all(rq, -EROFS);
1472 spin_unlock_irq(q->queue_lock);
1474 down_read(&rbd_dev->header_rwsem);
1476 if (rbd_dev->snap_id != CEPH_NOSNAP && !rbd_dev->snap_exists) {
1477 up_read(&rbd_dev->header_rwsem);
1478 dout("request for non-existent snapshot");
1479 spin_lock_irq(q->queue_lock);
1480 __blk_end_request_all(rq, -ENXIO);
1484 snapc = ceph_get_snap_context(rbd_dev->header.snapc);
1486 up_read(&rbd_dev->header_rwsem);
1488 dout("%s 0x%x bytes at 0x%llx\n",
1489 do_write ? "write" : "read",
1490 size, (unsigned long long) blk_rq_pos(rq) * SECTOR_SIZE);
1492 num_segs = rbd_get_num_segments(&rbd_dev->header, ofs, size);
1493 coll = rbd_alloc_coll(num_segs);
1495 spin_lock_irq(q->queue_lock);
1496 __blk_end_request_all(rq, -ENOMEM);
1497 ceph_put_snap_context(snapc);
1502 /* a bio clone to be passed down to OSD req */
1503 dout("rq->bio->bi_vcnt=%hu\n", rq->bio->bi_vcnt);
1504 op_size = rbd_get_segment(&rbd_dev->header,
1505 rbd_dev->header.object_prefix,
1508 kref_get(&coll->kref);
1509 bio = bio_chain_clone(&rq_bio, &next_bio, &bp,
1510 op_size, GFP_ATOMIC);
1512 rbd_coll_end_req_index(rq, coll, cur_seg,
1518 /* init OSD command: write or read */
1520 rbd_req_write(rq, rbd_dev,
1526 rbd_req_read(rq, rbd_dev,
1539 kref_put(&coll->kref, rbd_coll_release);
1542 bio_pair_release(bp);
1543 spin_lock_irq(q->queue_lock);
1545 ceph_put_snap_context(snapc);
1550 * a queue callback. Makes sure that we don't create a bio that spans across
1551 * multiple osd objects. One exception would be with a single page bios,
1552 * which we handle later at bio_chain_clone
1554 static int rbd_merge_bvec(struct request_queue *q, struct bvec_merge_data *bmd,
1555 struct bio_vec *bvec)
1557 struct rbd_device *rbd_dev = q->queuedata;
1558 unsigned int chunk_sectors;
1560 unsigned int bio_sectors;
1563 chunk_sectors = 1 << (rbd_dev->header.obj_order - SECTOR_SHIFT);
1564 sector = bmd->bi_sector + get_start_sect(bmd->bi_bdev);
1565 bio_sectors = bmd->bi_size >> SECTOR_SHIFT;
1567 max = (chunk_sectors - ((sector & (chunk_sectors - 1))
1568 + bio_sectors)) << SECTOR_SHIFT;
1570 max = 0; /* bio_add cannot handle a negative return */
1571 if (max <= bvec->bv_len && bio_sectors == 0)
1572 return bvec->bv_len;
1576 static void rbd_free_disk(struct rbd_device *rbd_dev)
1578 struct gendisk *disk = rbd_dev->disk;
1583 rbd_header_free(&rbd_dev->header);
1585 if (disk->flags & GENHD_FL_UP)
1588 blk_cleanup_queue(disk->queue);
1593 * reload the ondisk the header
1595 static int rbd_read_header(struct rbd_device *rbd_dev,
1596 struct rbd_image_header *header)
1599 struct rbd_image_header_ondisk *dh;
1605 * First reads the fixed-size header to determine the number
1606 * of snapshots, then re-reads it, along with all snapshot
1607 * records as well as their stored names.
1611 dh = kmalloc(len, GFP_KERNEL);
1615 rc = rbd_req_sync_read(rbd_dev,
1617 rbd_dev->header_name,
1623 rc = rbd_header_from_disk(header, dh, snap_count, GFP_KERNEL);
1626 pr_warning("unrecognized header format"
1628 rbd_dev->image_name);
1632 if (snap_count == header->total_snaps)
1635 snap_count = header->total_snaps;
1636 len = sizeof (*dh) +
1637 snap_count * sizeof(struct rbd_image_snap_ondisk) +
1638 header->snap_names_len;
1640 rbd_header_free(header);
1643 header->obj_version = ver;
1653 static int rbd_header_add_snap(struct rbd_device *rbd_dev,
1654 const char *snap_name,
1657 int name_len = strlen(snap_name);
1662 struct ceph_mon_client *monc;
1664 /* we should create a snapshot only if we're pointing at the head */
1665 if (rbd_dev->snap_id != CEPH_NOSNAP)
1668 monc = &rbd_dev->rbd_client->client->monc;
1669 ret = ceph_monc_create_snapid(monc, rbd_dev->pool_id, &new_snapid);
1670 dout("created snapid=%llu\n", (unsigned long long) new_snapid);
1674 data = kmalloc(name_len + 16, gfp_flags);
1679 e = data + name_len + 16;
1681 ceph_encode_string_safe(&p, e, snap_name, name_len, bad);
1682 ceph_encode_64_safe(&p, e, new_snapid, bad);
1684 ret = rbd_req_sync_exec(rbd_dev, rbd_dev->header_name,
1686 data, p - data, &ver);
1690 return ret < 0 ? ret : 0;
1695 static void __rbd_remove_all_snaps(struct rbd_device *rbd_dev)
1697 struct rbd_snap *snap;
1698 struct rbd_snap *next;
1700 list_for_each_entry_safe(snap, next, &rbd_dev->snaps, node)
1701 __rbd_remove_snap_dev(rbd_dev, snap);
1705 * only read the first part of the ondisk header, without the snaps info
1707 static int __rbd_refresh_header(struct rbd_device *rbd_dev)
1710 struct rbd_image_header h;
1712 ret = rbd_read_header(rbd_dev, &h);
1716 down_write(&rbd_dev->header_rwsem);
1719 if (rbd_dev->snap_id == CEPH_NOSNAP) {
1720 sector_t size = (sector_t) h.image_size / SECTOR_SIZE;
1722 dout("setting size to %llu sectors", (unsigned long long) size);
1723 set_capacity(rbd_dev->disk, size);
1726 /* rbd_dev->header.object_prefix shouldn't change */
1727 kfree(rbd_dev->header.snap_sizes);
1728 kfree(rbd_dev->header.snap_names);
1729 /* osd requests may still refer to snapc */
1730 ceph_put_snap_context(rbd_dev->header.snapc);
1732 rbd_dev->header.obj_version = h.obj_version;
1733 rbd_dev->header.image_size = h.image_size;
1734 rbd_dev->header.total_snaps = h.total_snaps;
1735 rbd_dev->header.snapc = h.snapc;
1736 rbd_dev->header.snap_names = h.snap_names;
1737 rbd_dev->header.snap_names_len = h.snap_names_len;
1738 rbd_dev->header.snap_sizes = h.snap_sizes;
1739 /* Free the extra copy of the object prefix */
1740 WARN_ON(strcmp(rbd_dev->header.object_prefix, h.object_prefix));
1741 kfree(h.object_prefix);
1743 ret = __rbd_init_snaps_header(rbd_dev);
1745 up_write(&rbd_dev->header_rwsem);
1750 static int rbd_init_disk(struct rbd_device *rbd_dev)
1752 struct gendisk *disk;
1753 struct request_queue *q;
1758 /* contact OSD, request size info about the object being mapped */
1759 rc = rbd_read_header(rbd_dev, &rbd_dev->header);
1763 /* no need to lock here, as rbd_dev is not registered yet */
1764 rc = __rbd_init_snaps_header(rbd_dev);
1768 rc = rbd_header_set_snap(rbd_dev, &total_size);
1772 /* create gendisk info */
1774 disk = alloc_disk(RBD_MINORS_PER_MAJOR);
1778 snprintf(disk->disk_name, sizeof(disk->disk_name), RBD_DRV_NAME "%d",
1780 disk->major = rbd_dev->major;
1781 disk->first_minor = 0;
1782 disk->fops = &rbd_bd_ops;
1783 disk->private_data = rbd_dev;
1787 q = blk_init_queue(rbd_rq_fn, &rbd_dev->lock);
1791 /* We use the default size, but let's be explicit about it. */
1792 blk_queue_physical_block_size(q, SECTOR_SIZE);
1794 /* set io sizes to object size */
1795 segment_size = rbd_obj_bytes(&rbd_dev->header);
1796 blk_queue_max_hw_sectors(q, segment_size / SECTOR_SIZE);
1797 blk_queue_max_segment_size(q, segment_size);
1798 blk_queue_io_min(q, segment_size);
1799 blk_queue_io_opt(q, segment_size);
1801 blk_queue_merge_bvec(q, rbd_merge_bvec);
1804 q->queuedata = rbd_dev;
1806 rbd_dev->disk = disk;
1809 /* finally, announce the disk to the world */
1810 set_capacity(disk, total_size / SECTOR_SIZE);
1813 pr_info("%s: added with size 0x%llx\n",
1814 disk->disk_name, (unsigned long long)total_size);
1827 static struct rbd_device *dev_to_rbd_dev(struct device *dev)
1829 return container_of(dev, struct rbd_device, dev);
1832 static ssize_t rbd_size_show(struct device *dev,
1833 struct device_attribute *attr, char *buf)
1835 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
1838 down_read(&rbd_dev->header_rwsem);
1839 size = get_capacity(rbd_dev->disk);
1840 up_read(&rbd_dev->header_rwsem);
1842 return sprintf(buf, "%llu\n", (unsigned long long) size * SECTOR_SIZE);
1845 static ssize_t rbd_major_show(struct device *dev,
1846 struct device_attribute *attr, char *buf)
1848 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
1850 return sprintf(buf, "%d\n", rbd_dev->major);
1853 static ssize_t rbd_client_id_show(struct device *dev,
1854 struct device_attribute *attr, char *buf)
1856 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
1858 return sprintf(buf, "client%lld\n",
1859 ceph_client_id(rbd_dev->rbd_client->client));
1862 static ssize_t rbd_pool_show(struct device *dev,
1863 struct device_attribute *attr, char *buf)
1865 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
1867 return sprintf(buf, "%s\n", rbd_dev->pool_name);
1870 static ssize_t rbd_pool_id_show(struct device *dev,
1871 struct device_attribute *attr, char *buf)
1873 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
1875 return sprintf(buf, "%d\n", rbd_dev->pool_id);
1878 static ssize_t rbd_name_show(struct device *dev,
1879 struct device_attribute *attr, char *buf)
1881 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
1883 return sprintf(buf, "%s\n", rbd_dev->image_name);
1886 static ssize_t rbd_snap_show(struct device *dev,
1887 struct device_attribute *attr,
1890 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
1892 return sprintf(buf, "%s\n", rbd_dev->snap_name);
1895 static ssize_t rbd_image_refresh(struct device *dev,
1896 struct device_attribute *attr,
1900 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
1904 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
1906 rc = __rbd_refresh_header(rbd_dev);
1910 mutex_unlock(&ctl_mutex);
1914 static DEVICE_ATTR(size, S_IRUGO, rbd_size_show, NULL);
1915 static DEVICE_ATTR(major, S_IRUGO, rbd_major_show, NULL);
1916 static DEVICE_ATTR(client_id, S_IRUGO, rbd_client_id_show, NULL);
1917 static DEVICE_ATTR(pool, S_IRUGO, rbd_pool_show, NULL);
1918 static DEVICE_ATTR(pool_id, S_IRUGO, rbd_pool_id_show, NULL);
1919 static DEVICE_ATTR(name, S_IRUGO, rbd_name_show, NULL);
1920 static DEVICE_ATTR(refresh, S_IWUSR, NULL, rbd_image_refresh);
1921 static DEVICE_ATTR(current_snap, S_IRUGO, rbd_snap_show, NULL);
1922 static DEVICE_ATTR(create_snap, S_IWUSR, NULL, rbd_snap_add);
1924 static struct attribute *rbd_attrs[] = {
1925 &dev_attr_size.attr,
1926 &dev_attr_major.attr,
1927 &dev_attr_client_id.attr,
1928 &dev_attr_pool.attr,
1929 &dev_attr_pool_id.attr,
1930 &dev_attr_name.attr,
1931 &dev_attr_current_snap.attr,
1932 &dev_attr_refresh.attr,
1933 &dev_attr_create_snap.attr,
1937 static struct attribute_group rbd_attr_group = {
1941 static const struct attribute_group *rbd_attr_groups[] = {
1946 static void rbd_sysfs_dev_release(struct device *dev)
1950 static struct device_type rbd_device_type = {
1952 .groups = rbd_attr_groups,
1953 .release = rbd_sysfs_dev_release,
1961 static ssize_t rbd_snap_size_show(struct device *dev,
1962 struct device_attribute *attr,
1965 struct rbd_snap *snap = container_of(dev, struct rbd_snap, dev);
1967 return sprintf(buf, "%llu\n", (unsigned long long)snap->size);
1970 static ssize_t rbd_snap_id_show(struct device *dev,
1971 struct device_attribute *attr,
1974 struct rbd_snap *snap = container_of(dev, struct rbd_snap, dev);
1976 return sprintf(buf, "%llu\n", (unsigned long long)snap->id);
1979 static DEVICE_ATTR(snap_size, S_IRUGO, rbd_snap_size_show, NULL);
1980 static DEVICE_ATTR(snap_id, S_IRUGO, rbd_snap_id_show, NULL);
1982 static struct attribute *rbd_snap_attrs[] = {
1983 &dev_attr_snap_size.attr,
1984 &dev_attr_snap_id.attr,
1988 static struct attribute_group rbd_snap_attr_group = {
1989 .attrs = rbd_snap_attrs,
1992 static void rbd_snap_dev_release(struct device *dev)
1994 struct rbd_snap *snap = container_of(dev, struct rbd_snap, dev);
1999 static const struct attribute_group *rbd_snap_attr_groups[] = {
2000 &rbd_snap_attr_group,
2004 static struct device_type rbd_snap_device_type = {
2005 .groups = rbd_snap_attr_groups,
2006 .release = rbd_snap_dev_release,
2009 static void __rbd_remove_snap_dev(struct rbd_device *rbd_dev,
2010 struct rbd_snap *snap)
2012 list_del(&snap->node);
2013 device_unregister(&snap->dev);
2016 static int rbd_register_snap_dev(struct rbd_device *rbd_dev,
2017 struct rbd_snap *snap,
2018 struct device *parent)
2020 struct device *dev = &snap->dev;
2023 dev->type = &rbd_snap_device_type;
2024 dev->parent = parent;
2025 dev->release = rbd_snap_dev_release;
2026 dev_set_name(dev, "snap_%s", snap->name);
2027 ret = device_register(dev);
2032 static int __rbd_add_snap_dev(struct rbd_device *rbd_dev,
2033 int i, const char *name,
2034 struct rbd_snap **snapp)
2037 struct rbd_snap *snap = kzalloc(sizeof(*snap), GFP_KERNEL);
2040 snap->name = kstrdup(name, GFP_KERNEL);
2041 snap->size = rbd_dev->header.snap_sizes[i];
2042 snap->id = rbd_dev->header.snapc->snaps[i];
2043 if (device_is_registered(&rbd_dev->dev)) {
2044 ret = rbd_register_snap_dev(rbd_dev, snap,
2058 * search for the previous snap in a null delimited string list
2060 const char *rbd_prev_snap_name(const char *name, const char *start)
2062 if (name < start + 2)
2075 * compare the old list of snapshots that we have to what's in the header
2076 * and update it accordingly. Note that the header holds the snapshots
2077 * in a reverse order (from newest to oldest) and we need to go from
2078 * older to new so that we don't get a duplicate snap name when
2079 * doing the process (e.g., removed snapshot and recreated a new
2080 * one with the same name.
2082 static int __rbd_init_snaps_header(struct rbd_device *rbd_dev)
2084 const char *name, *first_name;
2085 int i = rbd_dev->header.total_snaps;
2086 struct rbd_snap *snap, *old_snap = NULL;
2088 struct list_head *p, *n;
2090 first_name = rbd_dev->header.snap_names;
2091 name = first_name + rbd_dev->header.snap_names_len;
2093 list_for_each_prev_safe(p, n, &rbd_dev->snaps) {
2096 old_snap = list_entry(p, struct rbd_snap, node);
2099 cur_id = rbd_dev->header.snapc->snaps[i - 1];
2101 if (!i || old_snap->id < cur_id) {
2103 * old_snap->id was skipped, thus was
2104 * removed. If this rbd_dev is mapped to
2105 * the removed snapshot, record that it no
2106 * longer exists, to prevent further I/O.
2108 if (rbd_dev->snap_id == old_snap->id)
2109 rbd_dev->snap_exists = false;
2110 __rbd_remove_snap_dev(rbd_dev, old_snap);
2113 if (old_snap->id == cur_id) {
2114 /* we have this snapshot already */
2116 name = rbd_prev_snap_name(name, first_name);
2120 i--, name = rbd_prev_snap_name(name, first_name)) {
2125 cur_id = rbd_dev->header.snapc->snaps[i];
2126 /* snapshot removal? handle it above */
2127 if (cur_id >= old_snap->id)
2129 /* a new snapshot */
2130 ret = __rbd_add_snap_dev(rbd_dev, i - 1, name, &snap);
2134 /* note that we add it backward so using n and not p */
2135 list_add(&snap->node, n);
2139 /* we're done going over the old snap list, just add what's left */
2140 for (; i > 0; i--) {
2141 name = rbd_prev_snap_name(name, first_name);
2146 ret = __rbd_add_snap_dev(rbd_dev, i - 1, name, &snap);
2149 list_add(&snap->node, &rbd_dev->snaps);
2155 static int rbd_bus_add_dev(struct rbd_device *rbd_dev)
2159 struct rbd_snap *snap;
2161 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
2162 dev = &rbd_dev->dev;
2164 dev->bus = &rbd_bus_type;
2165 dev->type = &rbd_device_type;
2166 dev->parent = &rbd_root_dev;
2167 dev->release = rbd_dev_release;
2168 dev_set_name(dev, "%d", rbd_dev->id);
2169 ret = device_register(dev);
2173 list_for_each_entry(snap, &rbd_dev->snaps, node) {
2174 ret = rbd_register_snap_dev(rbd_dev, snap,
2180 mutex_unlock(&ctl_mutex);
2184 static void rbd_bus_del_dev(struct rbd_device *rbd_dev)
2186 device_unregister(&rbd_dev->dev);
2189 static int rbd_init_watch_dev(struct rbd_device *rbd_dev)
2194 ret = rbd_req_sync_watch(rbd_dev, rbd_dev->header_name,
2195 rbd_dev->header.obj_version);
2196 if (ret == -ERANGE) {
2197 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
2198 rc = __rbd_refresh_header(rbd_dev);
2199 mutex_unlock(&ctl_mutex);
2203 } while (ret == -ERANGE);
2208 static atomic64_t rbd_id_max = ATOMIC64_INIT(0);
2211 * Get a unique rbd identifier for the given new rbd_dev, and add
2212 * the rbd_dev to the global list. The minimum rbd id is 1.
2214 static void rbd_id_get(struct rbd_device *rbd_dev)
2216 rbd_dev->id = atomic64_inc_return(&rbd_id_max);
2218 spin_lock(&rbd_dev_list_lock);
2219 list_add_tail(&rbd_dev->node, &rbd_dev_list);
2220 spin_unlock(&rbd_dev_list_lock);
2224 * Remove an rbd_dev from the global list, and record that its
2225 * identifier is no longer in use.
2227 static void rbd_id_put(struct rbd_device *rbd_dev)
2229 struct list_head *tmp;
2230 int rbd_id = rbd_dev->id;
2235 spin_lock(&rbd_dev_list_lock);
2236 list_del_init(&rbd_dev->node);
2239 * If the id being "put" is not the current maximum, there
2240 * is nothing special we need to do.
2242 if (rbd_id != atomic64_read(&rbd_id_max)) {
2243 spin_unlock(&rbd_dev_list_lock);
2248 * We need to update the current maximum id. Search the
2249 * list to find out what it is. We're more likely to find
2250 * the maximum at the end, so search the list backward.
2253 list_for_each_prev(tmp, &rbd_dev_list) {
2254 struct rbd_device *rbd_dev;
2256 rbd_dev = list_entry(tmp, struct rbd_device, node);
2257 if (rbd_id > max_id)
2260 spin_unlock(&rbd_dev_list_lock);
2263 * The max id could have been updated by rbd_id_get(), in
2264 * which case it now accurately reflects the new maximum.
2265 * Be careful not to overwrite the maximum value in that
2268 atomic64_cmpxchg(&rbd_id_max, rbd_id, max_id);
2272 * Skips over white space at *buf, and updates *buf to point to the
2273 * first found non-space character (if any). Returns the length of
2274 * the token (string of non-white space characters) found. Note
2275 * that *buf must be terminated with '\0'.
2277 static inline size_t next_token(const char **buf)
2280 * These are the characters that produce nonzero for
2281 * isspace() in the "C" and "POSIX" locales.
2283 const char *spaces = " \f\n\r\t\v";
2285 *buf += strspn(*buf, spaces); /* Find start of token */
2287 return strcspn(*buf, spaces); /* Return token length */
2291 * Finds the next token in *buf, and if the provided token buffer is
2292 * big enough, copies the found token into it. The result, if
2293 * copied, is guaranteed to be terminated with '\0'. Note that *buf
2294 * must be terminated with '\0' on entry.
2296 * Returns the length of the token found (not including the '\0').
2297 * Return value will be 0 if no token is found, and it will be >=
2298 * token_size if the token would not fit.
2300 * The *buf pointer will be updated to point beyond the end of the
2301 * found token. Note that this occurs even if the token buffer is
2302 * too small to hold it.
2304 static inline size_t copy_token(const char **buf,
2310 len = next_token(buf);
2311 if (len < token_size) {
2312 memcpy(token, *buf, len);
2313 *(token + len) = '\0';
2321 * Finds the next token in *buf, dynamically allocates a buffer big
2322 * enough to hold a copy of it, and copies the token into the new
2323 * buffer. The copy is guaranteed to be terminated with '\0'. Note
2324 * that a duplicate buffer is created even for a zero-length token.
2326 * Returns a pointer to the newly-allocated duplicate, or a null
2327 * pointer if memory for the duplicate was not available. If
2328 * the lenp argument is a non-null pointer, the length of the token
2329 * (not including the '\0') is returned in *lenp.
2331 * If successful, the *buf pointer will be updated to point beyond
2332 * the end of the found token.
2334 * Note: uses GFP_KERNEL for allocation.
2336 static inline char *dup_token(const char **buf, size_t *lenp)
2341 len = next_token(buf);
2342 dup = kmalloc(len + 1, GFP_KERNEL);
2346 memcpy(dup, *buf, len);
2347 *(dup + len) = '\0';
2357 * This fills in the pool_name, image_name, image_name_len, snap_name,
2358 * rbd_dev, rbd_md_name, and name fields of the given rbd_dev, based
2359 * on the list of monitor addresses and other options provided via
2362 * Note: rbd_dev is assumed to have been initially zero-filled.
2364 static int rbd_add_parse_args(struct rbd_device *rbd_dev,
2366 const char **mon_addrs,
2367 size_t *mon_addrs_size,
2369 size_t options_size)
2374 /* The first four tokens are required */
2376 len = next_token(&buf);
2379 *mon_addrs_size = len + 1;
2384 len = copy_token(&buf, options, options_size);
2385 if (!len || len >= options_size)
2389 rbd_dev->pool_name = dup_token(&buf, NULL);
2390 if (!rbd_dev->pool_name)
2393 rbd_dev->image_name = dup_token(&buf, &rbd_dev->image_name_len);
2394 if (!rbd_dev->image_name)
2397 /* Create the name of the header object */
2399 rbd_dev->header_name = kmalloc(rbd_dev->image_name_len
2400 + sizeof (RBD_SUFFIX),
2402 if (!rbd_dev->header_name)
2404 sprintf(rbd_dev->header_name, "%s%s", rbd_dev->image_name, RBD_SUFFIX);
2407 * The snapshot name is optional. If none is is supplied,
2408 * we use the default value.
2410 rbd_dev->snap_name = dup_token(&buf, &len);
2411 if (!rbd_dev->snap_name)
2414 /* Replace the empty name with the default */
2415 kfree(rbd_dev->snap_name);
2417 = kmalloc(sizeof (RBD_SNAP_HEAD_NAME), GFP_KERNEL);
2418 if (!rbd_dev->snap_name)
2421 memcpy(rbd_dev->snap_name, RBD_SNAP_HEAD_NAME,
2422 sizeof (RBD_SNAP_HEAD_NAME));
2428 kfree(rbd_dev->header_name);
2429 kfree(rbd_dev->image_name);
2430 kfree(rbd_dev->pool_name);
2431 rbd_dev->pool_name = NULL;
2436 static ssize_t rbd_add(struct bus_type *bus,
2441 struct rbd_device *rbd_dev = NULL;
2442 const char *mon_addrs = NULL;
2443 size_t mon_addrs_size = 0;
2444 struct ceph_osd_client *osdc;
2447 if (!try_module_get(THIS_MODULE))
2450 options = kmalloc(count, GFP_KERNEL);
2453 rbd_dev = kzalloc(sizeof(*rbd_dev), GFP_KERNEL);
2457 /* static rbd_device initialization */
2458 spin_lock_init(&rbd_dev->lock);
2459 INIT_LIST_HEAD(&rbd_dev->node);
2460 INIT_LIST_HEAD(&rbd_dev->snaps);
2461 init_rwsem(&rbd_dev->header_rwsem);
2463 /* generate unique id: find highest unique id, add one */
2464 rbd_id_get(rbd_dev);
2466 /* Fill in the device name, now that we have its id. */
2467 BUILD_BUG_ON(DEV_NAME_LEN
2468 < sizeof (RBD_DRV_NAME) + MAX_INT_FORMAT_WIDTH);
2469 sprintf(rbd_dev->name, "%s%d", RBD_DRV_NAME, rbd_dev->id);
2471 /* parse add command */
2472 rc = rbd_add_parse_args(rbd_dev, buf, &mon_addrs, &mon_addrs_size,
2477 rbd_dev->rbd_client = rbd_get_client(mon_addrs, mon_addrs_size - 1,
2479 if (IS_ERR(rbd_dev->rbd_client)) {
2480 rc = PTR_ERR(rbd_dev->rbd_client);
2485 osdc = &rbd_dev->rbd_client->client->osdc;
2486 rc = ceph_pg_poolid_by_name(osdc->osdmap, rbd_dev->pool_name);
2488 goto err_out_client;
2489 rbd_dev->pool_id = rc;
2491 /* register our block device */
2492 rc = register_blkdev(0, rbd_dev->name);
2494 goto err_out_client;
2495 rbd_dev->major = rc;
2497 rc = rbd_bus_add_dev(rbd_dev);
2499 goto err_out_blkdev;
2502 * At this point cleanup in the event of an error is the job
2503 * of the sysfs code (initiated by rbd_bus_del_dev()).
2505 * Set up and announce blkdev mapping.
2507 rc = rbd_init_disk(rbd_dev);
2511 rc = rbd_init_watch_dev(rbd_dev);
2518 /* this will also clean up rest of rbd_dev stuff */
2520 rbd_bus_del_dev(rbd_dev);
2525 unregister_blkdev(rbd_dev->major, rbd_dev->name);
2527 rbd_put_client(rbd_dev);
2529 if (rbd_dev->pool_name) {
2530 kfree(rbd_dev->snap_name);
2531 kfree(rbd_dev->header_name);
2532 kfree(rbd_dev->image_name);
2533 kfree(rbd_dev->pool_name);
2535 rbd_id_put(rbd_dev);
2540 dout("Error adding device %s\n", buf);
2541 module_put(THIS_MODULE);
2543 return (ssize_t) rc;
2546 static struct rbd_device *__rbd_get_dev(unsigned long id)
2548 struct list_head *tmp;
2549 struct rbd_device *rbd_dev;
2551 spin_lock(&rbd_dev_list_lock);
2552 list_for_each(tmp, &rbd_dev_list) {
2553 rbd_dev = list_entry(tmp, struct rbd_device, node);
2554 if (rbd_dev->id == id) {
2555 spin_unlock(&rbd_dev_list_lock);
2559 spin_unlock(&rbd_dev_list_lock);
2563 static void rbd_dev_release(struct device *dev)
2565 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
2567 if (rbd_dev->watch_request) {
2568 struct ceph_client *client = rbd_dev->rbd_client->client;
2570 ceph_osdc_unregister_linger_request(&client->osdc,
2571 rbd_dev->watch_request);
2573 if (rbd_dev->watch_event)
2574 rbd_req_sync_unwatch(rbd_dev, rbd_dev->header_name);
2576 rbd_put_client(rbd_dev);
2578 /* clean up and free blkdev */
2579 rbd_free_disk(rbd_dev);
2580 unregister_blkdev(rbd_dev->major, rbd_dev->name);
2582 /* done with the id, and with the rbd_dev */
2583 kfree(rbd_dev->snap_name);
2584 kfree(rbd_dev->header_name);
2585 kfree(rbd_dev->pool_name);
2586 kfree(rbd_dev->image_name);
2587 rbd_id_put(rbd_dev);
2590 /* release module ref */
2591 module_put(THIS_MODULE);
2594 static ssize_t rbd_remove(struct bus_type *bus,
2598 struct rbd_device *rbd_dev = NULL;
2603 rc = strict_strtoul(buf, 10, &ul);
2607 /* convert to int; abort if we lost anything in the conversion */
2608 target_id = (int) ul;
2609 if (target_id != ul)
2612 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
2614 rbd_dev = __rbd_get_dev(target_id);
2620 __rbd_remove_all_snaps(rbd_dev);
2621 rbd_bus_del_dev(rbd_dev);
2624 mutex_unlock(&ctl_mutex);
2628 static ssize_t rbd_snap_add(struct device *dev,
2629 struct device_attribute *attr,
2633 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
2635 char *name = kmalloc(count + 1, GFP_KERNEL);
2639 snprintf(name, count, "%s", buf);
2641 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
2643 ret = rbd_header_add_snap(rbd_dev,
2648 ret = __rbd_refresh_header(rbd_dev);
2652 /* shouldn't hold ctl_mutex when notifying.. notify might
2653 trigger a watch callback that would need to get that mutex */
2654 mutex_unlock(&ctl_mutex);
2656 /* make a best effort, don't error if failed */
2657 rbd_req_sync_notify(rbd_dev, rbd_dev->header_name);
2664 mutex_unlock(&ctl_mutex);
2670 * create control files in sysfs
2673 static int rbd_sysfs_init(void)
2677 ret = device_register(&rbd_root_dev);
2681 ret = bus_register(&rbd_bus_type);
2683 device_unregister(&rbd_root_dev);
2688 static void rbd_sysfs_cleanup(void)
2690 bus_unregister(&rbd_bus_type);
2691 device_unregister(&rbd_root_dev);
2694 int __init rbd_init(void)
2698 rc = rbd_sysfs_init();
2701 pr_info("loaded " RBD_DRV_NAME_LONG "\n");
2705 void __exit rbd_exit(void)
2707 rbd_sysfs_cleanup();
2710 module_init(rbd_init);
2711 module_exit(rbd_exit);
2713 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
2714 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
2715 MODULE_DESCRIPTION("rados block device");
2717 /* following authorship retained from original osdblk.c */
2718 MODULE_AUTHOR("Jeff Garzik <jeff@garzik.org>");
2720 MODULE_LICENSE("GPL");