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;
99 * an instance of the client. multiple devices may share an rbd client.
102 struct ceph_client *client;
103 struct rbd_options *rbd_opts;
105 struct list_head node;
109 * a request completion status
111 struct rbd_req_status {
118 * a collection of requests
120 struct rbd_req_coll {
124 struct rbd_req_status status[0];
128 * a single io request
131 struct request *rq; /* blk layer request */
132 struct bio *bio; /* cloned bio */
133 struct page **pages; /* list of used pages */
136 struct rbd_req_coll *coll;
143 struct list_head node;
151 int id; /* blkdev unique id */
153 int major; /* blkdev assigned major */
154 struct gendisk *disk; /* blkdev's gendisk and rq */
155 struct request_queue *q;
157 struct rbd_client *rbd_client;
159 char name[DEV_NAME_LEN]; /* blkdev name, e.g. rbd3 */
161 spinlock_t lock; /* queue lock */
163 struct rbd_image_header header;
165 size_t image_name_len;
170 struct ceph_osd_event *watch_event;
171 struct ceph_osd_request *watch_request;
173 /* protects updating the header */
174 struct rw_semaphore header_rwsem;
175 /* name of the snapshot this device reads from */
177 /* id of the snapshot this device reads from */
178 u64 snap_id; /* current snapshot id */
179 /* whether the snap_id this device reads from still exists */
183 struct list_head node;
185 /* list of snapshots */
186 struct list_head snaps;
192 static DEFINE_MUTEX(ctl_mutex); /* Serialize open/close/setup/teardown */
194 static LIST_HEAD(rbd_dev_list); /* devices */
195 static DEFINE_SPINLOCK(rbd_dev_list_lock);
197 static LIST_HEAD(rbd_client_list); /* clients */
198 static DEFINE_SPINLOCK(rbd_client_list_lock);
200 static int __rbd_init_snaps_header(struct rbd_device *rbd_dev);
201 static void rbd_dev_release(struct device *dev);
202 static ssize_t rbd_snap_add(struct device *dev,
203 struct device_attribute *attr,
206 static void __rbd_remove_snap_dev(struct rbd_device *rbd_dev,
207 struct rbd_snap *snap);
209 static ssize_t rbd_add(struct bus_type *bus, const char *buf,
211 static ssize_t rbd_remove(struct bus_type *bus, const char *buf,
214 static struct bus_attribute rbd_bus_attrs[] = {
215 __ATTR(add, S_IWUSR, NULL, rbd_add),
216 __ATTR(remove, S_IWUSR, NULL, rbd_remove),
220 static struct bus_type rbd_bus_type = {
222 .bus_attrs = rbd_bus_attrs,
225 static void rbd_root_dev_release(struct device *dev)
229 static struct device rbd_root_dev = {
231 .release = rbd_root_dev_release,
235 static struct device *rbd_get_dev(struct rbd_device *rbd_dev)
237 return get_device(&rbd_dev->dev);
240 static void rbd_put_dev(struct rbd_device *rbd_dev)
242 put_device(&rbd_dev->dev);
245 static int __rbd_refresh_header(struct rbd_device *rbd_dev);
247 static int rbd_open(struct block_device *bdev, fmode_t mode)
249 struct rbd_device *rbd_dev = bdev->bd_disk->private_data;
251 rbd_get_dev(rbd_dev);
253 set_device_ro(bdev, rbd_dev->read_only);
255 if ((mode & FMODE_WRITE) && rbd_dev->read_only)
261 static int rbd_release(struct gendisk *disk, fmode_t mode)
263 struct rbd_device *rbd_dev = disk->private_data;
265 rbd_put_dev(rbd_dev);
270 static const struct block_device_operations rbd_bd_ops = {
271 .owner = THIS_MODULE,
273 .release = rbd_release,
277 * Initialize an rbd client instance.
280 static struct rbd_client *rbd_client_create(struct ceph_options *ceph_opts,
281 struct rbd_options *rbd_opts)
283 struct rbd_client *rbdc;
286 dout("rbd_client_create\n");
287 rbdc = kmalloc(sizeof(struct rbd_client), GFP_KERNEL);
291 kref_init(&rbdc->kref);
292 INIT_LIST_HEAD(&rbdc->node);
294 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
296 rbdc->client = ceph_create_client(ceph_opts, rbdc, 0, 0);
297 if (IS_ERR(rbdc->client))
299 ceph_opts = NULL; /* Now rbdc->client is responsible for ceph_opts */
301 ret = ceph_open_session(rbdc->client);
305 rbdc->rbd_opts = rbd_opts;
307 spin_lock(&rbd_client_list_lock);
308 list_add_tail(&rbdc->node, &rbd_client_list);
309 spin_unlock(&rbd_client_list_lock);
311 mutex_unlock(&ctl_mutex);
313 dout("rbd_client_create created %p\n", rbdc);
317 ceph_destroy_client(rbdc->client);
319 mutex_unlock(&ctl_mutex);
323 ceph_destroy_options(ceph_opts);
328 * Find a ceph client with specific addr and configuration.
330 static struct rbd_client *__rbd_client_find(struct ceph_options *ceph_opts)
332 struct rbd_client *client_node;
334 if (ceph_opts->flags & CEPH_OPT_NOSHARE)
337 list_for_each_entry(client_node, &rbd_client_list, node)
338 if (!ceph_compare_options(ceph_opts, client_node->client))
351 /* string args above */
354 static match_table_t rbd_opts_tokens = {
355 {Opt_notify_timeout, "notify_timeout=%d"},
357 /* string args above */
361 static int parse_rbd_opts_token(char *c, void *private)
363 struct rbd_options *rbd_opts = private;
364 substring_t argstr[MAX_OPT_ARGS];
365 int token, intval, ret;
367 token = match_token(c, rbd_opts_tokens, argstr);
371 if (token < Opt_last_int) {
372 ret = match_int(&argstr[0], &intval);
374 pr_err("bad mount option arg (not int) "
378 dout("got int token %d val %d\n", token, intval);
379 } else if (token > Opt_last_int && token < Opt_last_string) {
380 dout("got string token %d val %s\n", token,
383 dout("got token %d\n", token);
387 case Opt_notify_timeout:
388 rbd_opts->notify_timeout = intval;
397 * Get a ceph client with specific addr and configuration, if one does
398 * not exist create it.
400 static struct rbd_client *rbd_get_client(const char *mon_addr,
404 struct rbd_client *rbdc;
405 struct ceph_options *ceph_opts;
406 struct rbd_options *rbd_opts;
408 rbd_opts = kzalloc(sizeof(*rbd_opts), GFP_KERNEL);
410 return ERR_PTR(-ENOMEM);
412 rbd_opts->notify_timeout = RBD_NOTIFY_TIMEOUT_DEFAULT;
414 ceph_opts = ceph_parse_options(options, mon_addr,
415 mon_addr + mon_addr_len,
416 parse_rbd_opts_token, rbd_opts);
417 if (IS_ERR(ceph_opts)) {
419 return ERR_CAST(ceph_opts);
422 spin_lock(&rbd_client_list_lock);
423 rbdc = __rbd_client_find(ceph_opts);
425 /* using an existing client */
426 kref_get(&rbdc->kref);
427 spin_unlock(&rbd_client_list_lock);
429 ceph_destroy_options(ceph_opts);
434 spin_unlock(&rbd_client_list_lock);
436 rbdc = rbd_client_create(ceph_opts, rbd_opts);
445 * Destroy ceph client
447 * Caller must hold rbd_client_list_lock.
449 static void rbd_client_release(struct kref *kref)
451 struct rbd_client *rbdc = container_of(kref, struct rbd_client, kref);
453 dout("rbd_release_client %p\n", rbdc);
454 spin_lock(&rbd_client_list_lock);
455 list_del(&rbdc->node);
456 spin_unlock(&rbd_client_list_lock);
458 ceph_destroy_client(rbdc->client);
459 kfree(rbdc->rbd_opts);
464 * Drop reference to ceph client node. If it's not referenced anymore, release
467 static void rbd_put_client(struct rbd_device *rbd_dev)
469 kref_put(&rbd_dev->rbd_client->kref, rbd_client_release);
470 rbd_dev->rbd_client = NULL;
474 * Destroy requests collection
476 static void rbd_coll_release(struct kref *kref)
478 struct rbd_req_coll *coll =
479 container_of(kref, struct rbd_req_coll, kref);
481 dout("rbd_coll_release %p\n", coll);
486 * Create a new header structure, translate header format from the on-disk
489 static int rbd_header_from_disk(struct rbd_image_header *header,
490 struct rbd_image_header_ondisk *ondisk,
496 if (memcmp(ondisk, RBD_HEADER_TEXT, sizeof(RBD_HEADER_TEXT)))
499 snap_count = le32_to_cpu(ondisk->snap_count);
500 if (snap_count > (UINT_MAX - sizeof(struct ceph_snap_context))
503 header->snapc = kmalloc(sizeof(struct ceph_snap_context) +
504 snap_count * sizeof(u64),
509 header->snap_names_len = le64_to_cpu(ondisk->snap_names_len);
511 header->snap_names = kmalloc(header->snap_names_len,
513 if (!header->snap_names)
515 header->snap_sizes = kmalloc(snap_count * sizeof(u64),
517 if (!header->snap_sizes)
520 header->snap_names = NULL;
521 header->snap_sizes = NULL;
524 header->object_prefix = kmalloc(sizeof (ondisk->block_name) + 1,
526 if (!header->object_prefix)
529 memcpy(header->object_prefix, ondisk->block_name,
530 sizeof(ondisk->block_name));
531 header->object_prefix[sizeof (ondisk->block_name)] = '\0';
533 header->image_size = le64_to_cpu(ondisk->image_size);
534 header->obj_order = ondisk->options.order;
535 header->crypt_type = ondisk->options.crypt_type;
536 header->comp_type = ondisk->options.comp_type;
538 atomic_set(&header->snapc->nref, 1);
539 header->snap_seq = le64_to_cpu(ondisk->snap_seq);
540 header->snapc->num_snaps = snap_count;
541 header->total_snaps = snap_count;
543 if (snap_count && allocated_snaps == snap_count) {
544 for (i = 0; i < snap_count; i++) {
545 header->snapc->snaps[i] =
546 le64_to_cpu(ondisk->snaps[i].id);
547 header->snap_sizes[i] =
548 le64_to_cpu(ondisk->snaps[i].image_size);
551 /* copy snapshot names */
552 memcpy(header->snap_names, &ondisk->snaps[i],
553 header->snap_names_len);
559 kfree(header->snap_sizes);
561 kfree(header->snap_names);
563 kfree(header->snapc);
567 static int snap_by_name(struct rbd_image_header *header, const char *snap_name,
571 char *p = header->snap_names;
573 for (i = 0; i < header->total_snaps; i++) {
574 if (!strcmp(snap_name, p)) {
576 /* Found it. Pass back its id and/or size */
579 *seq = header->snapc->snaps[i];
581 *size = header->snap_sizes[i];
584 p += strlen(p) + 1; /* Skip ahead to the next name */
589 static int rbd_header_set_snap(struct rbd_device *rbd_dev, u64 *size)
591 struct rbd_image_header *header = &rbd_dev->header;
592 struct ceph_snap_context *snapc = header->snapc;
595 down_write(&rbd_dev->header_rwsem);
597 if (!memcmp(rbd_dev->snap_name, RBD_SNAP_HEAD_NAME,
598 sizeof (RBD_SNAP_HEAD_NAME))) {
599 if (header->total_snaps)
600 snapc->seq = header->snap_seq;
603 rbd_dev->snap_id = CEPH_NOSNAP;
604 rbd_dev->snap_exists = false;
605 rbd_dev->read_only = 0;
607 *size = header->image_size;
609 ret = snap_by_name(header, rbd_dev->snap_name,
613 rbd_dev->snap_id = snapc->seq;
614 rbd_dev->snap_exists = true;
615 rbd_dev->read_only = 1;
620 up_write(&rbd_dev->header_rwsem);
624 static void rbd_header_free(struct rbd_image_header *header)
626 kfree(header->object_prefix);
627 kfree(header->snap_sizes);
628 kfree(header->snap_names);
629 kfree(header->snapc);
633 * get the actual striped segment name, offset and length
635 static u64 rbd_get_segment(struct rbd_image_header *header,
636 const char *object_prefix,
638 char *seg_name, u64 *segofs)
640 u64 seg = ofs >> header->obj_order;
643 snprintf(seg_name, RBD_MAX_SEG_NAME_LEN,
644 "%s.%012llx", object_prefix, seg);
646 ofs = ofs & ((1 << header->obj_order) - 1);
647 len = min_t(u64, len, (1 << header->obj_order) - ofs);
655 static int rbd_get_num_segments(struct rbd_image_header *header,
658 u64 start_seg = ofs >> header->obj_order;
659 u64 end_seg = (ofs + len - 1) >> header->obj_order;
660 return end_seg - start_seg + 1;
664 * returns the size of an object in the image
666 static u64 rbd_obj_bytes(struct rbd_image_header *header)
668 return 1 << header->obj_order;
675 static void bio_chain_put(struct bio *chain)
681 chain = chain->bi_next;
687 * zeros a bio chain, starting at specific offset
689 static void zero_bio_chain(struct bio *chain, int start_ofs)
698 bio_for_each_segment(bv, chain, i) {
699 if (pos + bv->bv_len > start_ofs) {
700 int remainder = max(start_ofs - pos, 0);
701 buf = bvec_kmap_irq(bv, &flags);
702 memset(buf + remainder, 0,
703 bv->bv_len - remainder);
704 bvec_kunmap_irq(buf, &flags);
709 chain = chain->bi_next;
714 * bio_chain_clone - clone a chain of bios up to a certain length.
715 * might return a bio_pair that will need to be released.
717 static struct bio *bio_chain_clone(struct bio **old, struct bio **next,
718 struct bio_pair **bp,
719 int len, gfp_t gfpmask)
721 struct bio *tmp, *old_chain = *old, *new_chain = NULL, *tail = NULL;
725 bio_pair_release(*bp);
729 while (old_chain && (total < len)) {
730 tmp = bio_kmalloc(gfpmask, old_chain->bi_max_vecs);
734 if (total + old_chain->bi_size > len) {
738 * this split can only happen with a single paged bio,
739 * split_bio will BUG_ON if this is not the case
741 dout("bio_chain_clone split! total=%d remaining=%d"
743 (int)total, (int)len-total,
744 (int)old_chain->bi_size);
746 /* split the bio. We'll release it either in the next
747 call, or it will have to be released outside */
748 bp = bio_split(old_chain, (len - total) / SECTOR_SIZE);
752 __bio_clone(tmp, &bp->bio1);
756 __bio_clone(tmp, old_chain);
757 *next = old_chain->bi_next;
761 gfpmask &= ~__GFP_WAIT;
765 new_chain = tail = tmp;
770 old_chain = old_chain->bi_next;
772 total += tmp->bi_size;
778 tail->bi_next = NULL;
785 dout("bio_chain_clone with err\n");
786 bio_chain_put(new_chain);
791 * helpers for osd request op vectors.
793 static int rbd_create_rw_ops(struct ceph_osd_req_op **ops,
798 *ops = kzalloc(sizeof(struct ceph_osd_req_op) * (num_ops + 1),
802 (*ops)[0].op = opcode;
804 * op extent offset and length will be set later on
805 * in calc_raw_layout()
807 (*ops)[0].payload_len = payload_len;
811 static void rbd_destroy_ops(struct ceph_osd_req_op *ops)
816 static void rbd_coll_end_req_index(struct request *rq,
817 struct rbd_req_coll *coll,
821 struct request_queue *q;
824 dout("rbd_coll_end_req_index %p index %d ret %d len %lld\n",
825 coll, index, ret, len);
831 blk_end_request(rq, ret, len);
837 spin_lock_irq(q->queue_lock);
838 coll->status[index].done = 1;
839 coll->status[index].rc = ret;
840 coll->status[index].bytes = len;
841 max = min = coll->num_done;
842 while (max < coll->total && coll->status[max].done)
845 for (i = min; i<max; i++) {
846 __blk_end_request(rq, coll->status[i].rc,
847 coll->status[i].bytes);
849 kref_put(&coll->kref, rbd_coll_release);
851 spin_unlock_irq(q->queue_lock);
854 static void rbd_coll_end_req(struct rbd_request *req,
857 rbd_coll_end_req_index(req->rq, req->coll, req->coll_index, ret, len);
861 * Send ceph osd request
863 static int rbd_do_request(struct request *rq,
864 struct rbd_device *rbd_dev,
865 struct ceph_snap_context *snapc,
867 const char *object_name, u64 ofs, u64 len,
872 struct ceph_osd_req_op *ops,
873 struct rbd_req_coll *coll,
875 void (*rbd_cb)(struct ceph_osd_request *req,
876 struct ceph_msg *msg),
877 struct ceph_osd_request **linger_req,
880 struct ceph_osd_request *req;
881 struct ceph_file_layout *layout;
884 struct timespec mtime = CURRENT_TIME;
885 struct rbd_request *req_data;
886 struct ceph_osd_request_head *reqhead;
887 struct ceph_osd_client *osdc;
889 req_data = kzalloc(sizeof(*req_data), GFP_NOIO);
892 rbd_coll_end_req_index(rq, coll, coll_index,
898 req_data->coll = coll;
899 req_data->coll_index = coll_index;
902 dout("rbd_do_request object_name=%s ofs=%lld len=%lld\n",
903 object_name, len, ofs);
905 down_read(&rbd_dev->header_rwsem);
907 osdc = &rbd_dev->rbd_client->client->osdc;
908 req = ceph_osdc_alloc_request(osdc, flags, snapc, ops,
909 false, GFP_NOIO, pages, bio);
911 up_read(&rbd_dev->header_rwsem);
916 req->r_callback = rbd_cb;
920 req_data->pages = pages;
923 req->r_priv = req_data;
925 reqhead = req->r_request->front.iov_base;
926 reqhead->snapid = cpu_to_le64(CEPH_NOSNAP);
928 strncpy(req->r_oid, object_name, sizeof(req->r_oid));
929 req->r_oid_len = strlen(req->r_oid);
931 layout = &req->r_file_layout;
932 memset(layout, 0, sizeof(*layout));
933 layout->fl_stripe_unit = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER);
934 layout->fl_stripe_count = cpu_to_le32(1);
935 layout->fl_object_size = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER);
936 layout->fl_pg_pool = cpu_to_le32(rbd_dev->pool_id);
937 ceph_calc_raw_layout(osdc, layout, snapid, ofs, &len, &bno,
940 ceph_osdc_build_request(req, ofs, &len,
944 req->r_oid, req->r_oid_len);
945 up_read(&rbd_dev->header_rwsem);
948 ceph_osdc_set_request_linger(osdc, req);
952 ret = ceph_osdc_start_request(osdc, req, false);
957 ret = ceph_osdc_wait_request(osdc, req);
959 *ver = le64_to_cpu(req->r_reassert_version.version);
960 dout("reassert_ver=%lld\n",
961 le64_to_cpu(req->r_reassert_version.version));
962 ceph_osdc_put_request(req);
967 bio_chain_put(req_data->bio);
968 ceph_osdc_put_request(req);
970 rbd_coll_end_req(req_data, ret, len);
976 * Ceph osd op callback
978 static void rbd_req_cb(struct ceph_osd_request *req, struct ceph_msg *msg)
980 struct rbd_request *req_data = req->r_priv;
981 struct ceph_osd_reply_head *replyhead;
982 struct ceph_osd_op *op;
988 replyhead = msg->front.iov_base;
989 WARN_ON(le32_to_cpu(replyhead->num_ops) == 0);
990 op = (void *)(replyhead + 1);
991 rc = le32_to_cpu(replyhead->result);
992 bytes = le64_to_cpu(op->extent.length);
993 read_op = (le16_to_cpu(op->op) == CEPH_OSD_OP_READ);
995 dout("rbd_req_cb bytes=%lld readop=%d rc=%d\n", bytes, read_op, rc);
997 if (rc == -ENOENT && read_op) {
998 zero_bio_chain(req_data->bio, 0);
1000 } else if (rc == 0 && read_op && bytes < req_data->len) {
1001 zero_bio_chain(req_data->bio, bytes);
1002 bytes = req_data->len;
1005 rbd_coll_end_req(req_data, rc, bytes);
1008 bio_chain_put(req_data->bio);
1010 ceph_osdc_put_request(req);
1014 static void rbd_simple_req_cb(struct ceph_osd_request *req, struct ceph_msg *msg)
1016 ceph_osdc_put_request(req);
1020 * Do a synchronous ceph osd operation
1022 static int rbd_req_sync_op(struct rbd_device *rbd_dev,
1023 struct ceph_snap_context *snapc,
1027 struct ceph_osd_req_op *orig_ops,
1028 const char *object_name,
1031 struct ceph_osd_request **linger_req,
1035 struct page **pages;
1037 struct ceph_osd_req_op *ops = orig_ops;
1040 num_pages = calc_pages_for(ofs , len);
1041 pages = ceph_alloc_page_vector(num_pages, GFP_KERNEL);
1043 return PTR_ERR(pages);
1046 payload_len = (flags & CEPH_OSD_FLAG_WRITE ? len : 0);
1047 ret = rbd_create_rw_ops(&ops, 1, opcode, payload_len);
1051 if ((flags & CEPH_OSD_FLAG_WRITE) && buf) {
1052 ret = ceph_copy_to_page_vector(pages, buf, ofs, len);
1058 ret = rbd_do_request(NULL, rbd_dev, snapc, snapid,
1059 object_name, ofs, len, NULL,
1069 if ((flags & CEPH_OSD_FLAG_READ) && buf)
1070 ret = ceph_copy_from_page_vector(pages, buf, ofs, ret);
1074 rbd_destroy_ops(ops);
1076 ceph_release_page_vector(pages, num_pages);
1081 * Do an asynchronous ceph osd operation
1083 static int rbd_do_op(struct request *rq,
1084 struct rbd_device *rbd_dev,
1085 struct ceph_snap_context *snapc,
1087 int opcode, int flags,
1090 struct rbd_req_coll *coll,
1097 struct ceph_osd_req_op *ops;
1100 seg_name = kmalloc(RBD_MAX_SEG_NAME_LEN + 1, GFP_NOIO);
1104 seg_len = rbd_get_segment(&rbd_dev->header,
1105 rbd_dev->header.object_prefix,
1107 seg_name, &seg_ofs);
1109 payload_len = (flags & CEPH_OSD_FLAG_WRITE ? seg_len : 0);
1111 ret = rbd_create_rw_ops(&ops, 1, opcode, payload_len);
1115 /* we've taken care of segment sizes earlier when we
1116 cloned the bios. We should never have a segment
1117 truncated at this point */
1118 BUG_ON(seg_len < len);
1120 ret = rbd_do_request(rq, rbd_dev, snapc, snapid,
1121 seg_name, seg_ofs, seg_len,
1127 rbd_req_cb, 0, NULL);
1129 rbd_destroy_ops(ops);
1136 * Request async osd write
1138 static int rbd_req_write(struct request *rq,
1139 struct rbd_device *rbd_dev,
1140 struct ceph_snap_context *snapc,
1143 struct rbd_req_coll *coll,
1146 return rbd_do_op(rq, rbd_dev, snapc, CEPH_NOSNAP,
1148 CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK,
1149 ofs, len, bio, coll, coll_index);
1153 * Request async osd read
1155 static int rbd_req_read(struct request *rq,
1156 struct rbd_device *rbd_dev,
1160 struct rbd_req_coll *coll,
1163 return rbd_do_op(rq, rbd_dev, NULL,
1167 ofs, len, bio, coll, coll_index);
1171 * Request sync osd read
1173 static int rbd_req_sync_read(struct rbd_device *rbd_dev,
1174 struct ceph_snap_context *snapc,
1176 const char *object_name,
1181 return rbd_req_sync_op(rbd_dev, NULL,
1186 object_name, ofs, len, buf, NULL, ver);
1190 * Request sync osd watch
1192 static int rbd_req_sync_notify_ack(struct rbd_device *rbd_dev,
1195 const char *object_name)
1197 struct ceph_osd_req_op *ops;
1200 ret = rbd_create_rw_ops(&ops, 1, CEPH_OSD_OP_NOTIFY_ACK, 0);
1204 ops[0].watch.ver = cpu_to_le64(rbd_dev->header.obj_version);
1205 ops[0].watch.cookie = notify_id;
1206 ops[0].watch.flag = 0;
1208 ret = rbd_do_request(NULL, rbd_dev, NULL, CEPH_NOSNAP,
1209 object_name, 0, 0, NULL,
1214 rbd_simple_req_cb, 0, NULL);
1216 rbd_destroy_ops(ops);
1220 static void rbd_watch_cb(u64 ver, u64 notify_id, u8 opcode, void *data)
1222 struct rbd_device *rbd_dev = (struct rbd_device *)data;
1228 dout("rbd_watch_cb %s notify_id=%lld opcode=%d\n",
1229 rbd_dev->header_name, notify_id, (int) opcode);
1230 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
1231 rc = __rbd_refresh_header(rbd_dev);
1232 mutex_unlock(&ctl_mutex);
1234 pr_warning(RBD_DRV_NAME "%d got notification but failed to "
1235 " update snaps: %d\n", rbd_dev->major, rc);
1237 rbd_req_sync_notify_ack(rbd_dev, ver, notify_id, rbd_dev->header_name);
1241 * Request sync osd watch
1243 static int rbd_req_sync_watch(struct rbd_device *rbd_dev,
1244 const char *object_name,
1247 struct ceph_osd_req_op *ops;
1248 struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
1250 int ret = rbd_create_rw_ops(&ops, 1, CEPH_OSD_OP_WATCH, 0);
1254 ret = ceph_osdc_create_event(osdc, rbd_watch_cb, 0,
1255 (void *)rbd_dev, &rbd_dev->watch_event);
1259 ops[0].watch.ver = cpu_to_le64(ver);
1260 ops[0].watch.cookie = cpu_to_le64(rbd_dev->watch_event->cookie);
1261 ops[0].watch.flag = 1;
1263 ret = rbd_req_sync_op(rbd_dev, NULL,
1266 CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK,
1268 object_name, 0, 0, NULL,
1269 &rbd_dev->watch_request, NULL);
1274 rbd_destroy_ops(ops);
1278 ceph_osdc_cancel_event(rbd_dev->watch_event);
1279 rbd_dev->watch_event = NULL;
1281 rbd_destroy_ops(ops);
1286 * Request sync osd unwatch
1288 static int rbd_req_sync_unwatch(struct rbd_device *rbd_dev,
1289 const char *object_name)
1291 struct ceph_osd_req_op *ops;
1293 int ret = rbd_create_rw_ops(&ops, 1, CEPH_OSD_OP_WATCH, 0);
1297 ops[0].watch.ver = 0;
1298 ops[0].watch.cookie = cpu_to_le64(rbd_dev->watch_event->cookie);
1299 ops[0].watch.flag = 0;
1301 ret = rbd_req_sync_op(rbd_dev, NULL,
1304 CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK,
1306 object_name, 0, 0, NULL, NULL, NULL);
1308 rbd_destroy_ops(ops);
1309 ceph_osdc_cancel_event(rbd_dev->watch_event);
1310 rbd_dev->watch_event = NULL;
1314 struct rbd_notify_info {
1315 struct rbd_device *rbd_dev;
1318 static void rbd_notify_cb(u64 ver, u64 notify_id, u8 opcode, void *data)
1320 struct rbd_device *rbd_dev = (struct rbd_device *)data;
1324 dout("rbd_notify_cb %s notify_id=%lld opcode=%d\n",
1325 rbd_dev->header_name,
1326 notify_id, (int)opcode);
1330 * Request sync osd notify
1332 static int rbd_req_sync_notify(struct rbd_device *rbd_dev,
1333 const char *object_name)
1335 struct ceph_osd_req_op *ops;
1336 struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
1337 struct ceph_osd_event *event;
1338 struct rbd_notify_info info;
1339 int payload_len = sizeof(u32) + sizeof(u32);
1342 ret = rbd_create_rw_ops(&ops, 1, CEPH_OSD_OP_NOTIFY, payload_len);
1346 info.rbd_dev = rbd_dev;
1348 ret = ceph_osdc_create_event(osdc, rbd_notify_cb, 1,
1349 (void *)&info, &event);
1353 ops[0].watch.ver = 1;
1354 ops[0].watch.flag = 1;
1355 ops[0].watch.cookie = event->cookie;
1356 ops[0].watch.prot_ver = RADOS_NOTIFY_VER;
1357 ops[0].watch.timeout = 12;
1359 ret = rbd_req_sync_op(rbd_dev, NULL,
1362 CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK,
1364 object_name, 0, 0, NULL, NULL, NULL);
1368 ret = ceph_osdc_wait_event(event, CEPH_OSD_TIMEOUT_DEFAULT);
1369 dout("ceph_osdc_wait_event returned %d\n", ret);
1370 rbd_destroy_ops(ops);
1374 ceph_osdc_cancel_event(event);
1376 rbd_destroy_ops(ops);
1381 * Request sync osd read
1383 static int rbd_req_sync_exec(struct rbd_device *rbd_dev,
1384 const char *object_name,
1385 const char *class_name,
1386 const char *method_name,
1391 struct ceph_osd_req_op *ops;
1392 int class_name_len = strlen(class_name);
1393 int method_name_len = strlen(method_name);
1394 int ret = rbd_create_rw_ops(&ops, 1, CEPH_OSD_OP_CALL,
1395 class_name_len + method_name_len + len);
1399 ops[0].cls.class_name = class_name;
1400 ops[0].cls.class_len = (__u8) class_name_len;
1401 ops[0].cls.method_name = method_name;
1402 ops[0].cls.method_len = (__u8) method_name_len;
1403 ops[0].cls.argc = 0;
1404 ops[0].cls.indata = data;
1405 ops[0].cls.indata_len = len;
1407 ret = rbd_req_sync_op(rbd_dev, NULL,
1410 CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK,
1412 object_name, 0, 0, NULL, NULL, ver);
1414 rbd_destroy_ops(ops);
1416 dout("cls_exec returned %d\n", ret);
1420 static struct rbd_req_coll *rbd_alloc_coll(int num_reqs)
1422 struct rbd_req_coll *coll =
1423 kzalloc(sizeof(struct rbd_req_coll) +
1424 sizeof(struct rbd_req_status) * num_reqs,
1429 coll->total = num_reqs;
1430 kref_init(&coll->kref);
1435 * block device queue callback
1437 static void rbd_rq_fn(struct request_queue *q)
1439 struct rbd_device *rbd_dev = q->queuedata;
1441 struct bio_pair *bp = NULL;
1443 while ((rq = blk_fetch_request(q))) {
1445 struct bio *rq_bio, *next_bio = NULL;
1447 int size, op_size = 0;
1449 int num_segs, cur_seg = 0;
1450 struct rbd_req_coll *coll;
1452 /* peek at request from block layer */
1456 dout("fetched request\n");
1458 /* filter out block requests we don't understand */
1459 if ((rq->cmd_type != REQ_TYPE_FS)) {
1460 __blk_end_request_all(rq, 0);
1464 /* deduce our operation (read, write) */
1465 do_write = (rq_data_dir(rq) == WRITE);
1467 size = blk_rq_bytes(rq);
1468 ofs = blk_rq_pos(rq) * SECTOR_SIZE;
1470 if (do_write && rbd_dev->read_only) {
1471 __blk_end_request_all(rq, -EROFS);
1475 spin_unlock_irq(q->queue_lock);
1477 if (rbd_dev->snap_id != CEPH_NOSNAP) {
1480 down_read(&rbd_dev->header_rwsem);
1481 snap_exists = rbd_dev->snap_exists;
1482 up_read(&rbd_dev->header_rwsem);
1485 dout("request for non-existent snapshot");
1486 spin_lock_irq(q->queue_lock);
1487 __blk_end_request_all(rq, -ENXIO);
1492 dout("%s 0x%x bytes at 0x%llx\n",
1493 do_write ? "write" : "read",
1494 size, blk_rq_pos(rq) * SECTOR_SIZE);
1496 num_segs = rbd_get_num_segments(&rbd_dev->header, ofs, size);
1497 coll = rbd_alloc_coll(num_segs);
1499 spin_lock_irq(q->queue_lock);
1500 __blk_end_request_all(rq, -ENOMEM);
1505 /* a bio clone to be passed down to OSD req */
1506 dout("rq->bio->bi_vcnt=%d\n", rq->bio->bi_vcnt);
1507 op_size = rbd_get_segment(&rbd_dev->header,
1508 rbd_dev->header.object_prefix,
1511 kref_get(&coll->kref);
1512 bio = bio_chain_clone(&rq_bio, &next_bio, &bp,
1513 op_size, GFP_ATOMIC);
1515 rbd_coll_end_req_index(rq, coll, cur_seg,
1521 /* init OSD command: write or read */
1523 rbd_req_write(rq, rbd_dev,
1524 rbd_dev->header.snapc,
1529 rbd_req_read(rq, rbd_dev,
1542 kref_put(&coll->kref, rbd_coll_release);
1545 bio_pair_release(bp);
1546 spin_lock_irq(q->queue_lock);
1551 * a queue callback. Makes sure that we don't create a bio that spans across
1552 * multiple osd objects. One exception would be with a single page bios,
1553 * which we handle later at bio_chain_clone
1555 static int rbd_merge_bvec(struct request_queue *q, struct bvec_merge_data *bmd,
1556 struct bio_vec *bvec)
1558 struct rbd_device *rbd_dev = q->queuedata;
1559 unsigned int chunk_sectors;
1561 unsigned int bio_sectors;
1564 chunk_sectors = 1 << (rbd_dev->header.obj_order - SECTOR_SHIFT);
1565 sector = bmd->bi_sector + get_start_sect(bmd->bi_bdev);
1566 bio_sectors = bmd->bi_size >> SECTOR_SHIFT;
1568 max = (chunk_sectors - ((sector & (chunk_sectors - 1))
1569 + bio_sectors)) << SECTOR_SHIFT;
1571 max = 0; /* bio_add cannot handle a negative return */
1572 if (max <= bvec->bv_len && bio_sectors == 0)
1573 return bvec->bv_len;
1577 static void rbd_free_disk(struct rbd_device *rbd_dev)
1579 struct gendisk *disk = rbd_dev->disk;
1584 rbd_header_free(&rbd_dev->header);
1586 if (disk->flags & GENHD_FL_UP)
1589 blk_cleanup_queue(disk->queue);
1594 * reload the ondisk the header
1596 static int rbd_read_header(struct rbd_device *rbd_dev,
1597 struct rbd_image_header *header)
1600 struct rbd_image_header_ondisk *dh;
1606 * First reads the fixed-size header to determine the number
1607 * of snapshots, then re-reads it, along with all snapshot
1608 * records as well as their stored names.
1612 dh = kmalloc(len, GFP_KERNEL);
1616 rc = rbd_req_sync_read(rbd_dev,
1618 rbd_dev->header_name,
1624 rc = rbd_header_from_disk(header, dh, snap_count, GFP_KERNEL);
1627 pr_warning("unrecognized header format"
1629 rbd_dev->image_name);
1633 if (snap_count == header->total_snaps)
1636 snap_count = header->total_snaps;
1637 len = sizeof (*dh) +
1638 snap_count * sizeof(struct rbd_image_snap_ondisk) +
1639 header->snap_names_len;
1641 rbd_header_free(header);
1644 header->obj_version = ver;
1654 static int rbd_header_add_snap(struct rbd_device *rbd_dev,
1655 const char *snap_name,
1658 int name_len = strlen(snap_name);
1663 struct ceph_mon_client *monc;
1665 /* we should create a snapshot only if we're pointing at the head */
1666 if (rbd_dev->snap_id != CEPH_NOSNAP)
1669 monc = &rbd_dev->rbd_client->client->monc;
1670 ret = ceph_monc_create_snapid(monc, rbd_dev->pool_id, &new_snapid);
1671 dout("created snapid=%lld\n", new_snapid);
1675 data = kmalloc(name_len + 16, gfp_flags);
1680 e = data + name_len + 16;
1682 ceph_encode_string_safe(&p, e, snap_name, name_len, bad);
1683 ceph_encode_64_safe(&p, e, new_snapid, bad);
1685 ret = rbd_req_sync_exec(rbd_dev, rbd_dev->header_name,
1687 data, p - data, &ver);
1694 down_write(&rbd_dev->header_rwsem);
1695 rbd_dev->header.snapc->seq = new_snapid;
1696 up_write(&rbd_dev->header_rwsem);
1703 static void __rbd_remove_all_snaps(struct rbd_device *rbd_dev)
1705 struct rbd_snap *snap;
1707 while (!list_empty(&rbd_dev->snaps)) {
1708 snap = list_first_entry(&rbd_dev->snaps, struct rbd_snap, node);
1709 __rbd_remove_snap_dev(rbd_dev, snap);
1714 * only read the first part of the ondisk header, without the snaps info
1716 static int __rbd_refresh_header(struct rbd_device *rbd_dev)
1719 struct rbd_image_header h;
1723 ret = rbd_read_header(rbd_dev, &h);
1727 down_write(&rbd_dev->header_rwsem);
1730 if (rbd_dev->snap_id == CEPH_NOSNAP) {
1731 sector_t size = (sector_t) h.image_size / SECTOR_SIZE;
1733 dout("setting size to %llu sectors", (unsigned long long) size);
1734 set_capacity(rbd_dev->disk, size);
1737 snap_seq = rbd_dev->header.snapc->seq;
1738 if (rbd_dev->header.total_snaps &&
1739 rbd_dev->header.snapc->snaps[0] == snap_seq)
1740 /* pointing at the head, will need to follow that
1744 /* rbd_dev->header.object_prefix shouldn't change */
1745 kfree(rbd_dev->header.snap_sizes);
1746 kfree(rbd_dev->header.snap_names);
1747 kfree(rbd_dev->header.snapc);
1749 rbd_dev->header.image_size = h.image_size;
1750 rbd_dev->header.total_snaps = h.total_snaps;
1751 rbd_dev->header.snapc = h.snapc;
1752 rbd_dev->header.snap_names = h.snap_names;
1753 rbd_dev->header.snap_names_len = h.snap_names_len;
1754 rbd_dev->header.snap_sizes = h.snap_sizes;
1755 /* Free the extra copy of the object prefix */
1756 WARN_ON(strcmp(rbd_dev->header.object_prefix, h.object_prefix));
1757 kfree(h.object_prefix);
1760 rbd_dev->header.snapc->seq = rbd_dev->header.snapc->snaps[0];
1762 rbd_dev->header.snapc->seq = snap_seq;
1764 ret = __rbd_init_snaps_header(rbd_dev);
1766 up_write(&rbd_dev->header_rwsem);
1771 static int rbd_init_disk(struct rbd_device *rbd_dev)
1773 struct gendisk *disk;
1774 struct request_queue *q;
1779 /* contact OSD, request size info about the object being mapped */
1780 rc = rbd_read_header(rbd_dev, &rbd_dev->header);
1784 /* no need to lock here, as rbd_dev is not registered yet */
1785 rc = __rbd_init_snaps_header(rbd_dev);
1789 rc = rbd_header_set_snap(rbd_dev, &total_size);
1793 /* create gendisk info */
1795 disk = alloc_disk(RBD_MINORS_PER_MAJOR);
1799 snprintf(disk->disk_name, sizeof(disk->disk_name), RBD_DRV_NAME "%d",
1801 disk->major = rbd_dev->major;
1802 disk->first_minor = 0;
1803 disk->fops = &rbd_bd_ops;
1804 disk->private_data = rbd_dev;
1808 q = blk_init_queue(rbd_rq_fn, &rbd_dev->lock);
1812 /* We use the default size, but let's be explicit about it. */
1813 blk_queue_physical_block_size(q, SECTOR_SIZE);
1815 /* set io sizes to object size */
1816 segment_size = rbd_obj_bytes(&rbd_dev->header);
1817 blk_queue_max_hw_sectors(q, segment_size / SECTOR_SIZE);
1818 blk_queue_max_segment_size(q, segment_size);
1819 blk_queue_io_min(q, segment_size);
1820 blk_queue_io_opt(q, segment_size);
1822 blk_queue_merge_bvec(q, rbd_merge_bvec);
1825 q->queuedata = rbd_dev;
1827 rbd_dev->disk = disk;
1830 /* finally, announce the disk to the world */
1831 set_capacity(disk, total_size / SECTOR_SIZE);
1834 pr_info("%s: added with size 0x%llx\n",
1835 disk->disk_name, (unsigned long long)total_size);
1848 static struct rbd_device *dev_to_rbd_dev(struct device *dev)
1850 return container_of(dev, struct rbd_device, dev);
1853 static ssize_t rbd_size_show(struct device *dev,
1854 struct device_attribute *attr, char *buf)
1856 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
1859 down_read(&rbd_dev->header_rwsem);
1860 size = get_capacity(rbd_dev->disk);
1861 up_read(&rbd_dev->header_rwsem);
1863 return sprintf(buf, "%llu\n", (unsigned long long) size * SECTOR_SIZE);
1866 static ssize_t rbd_major_show(struct device *dev,
1867 struct device_attribute *attr, char *buf)
1869 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
1871 return sprintf(buf, "%d\n", rbd_dev->major);
1874 static ssize_t rbd_client_id_show(struct device *dev,
1875 struct device_attribute *attr, char *buf)
1877 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
1879 return sprintf(buf, "client%lld\n",
1880 ceph_client_id(rbd_dev->rbd_client->client));
1883 static ssize_t rbd_pool_show(struct device *dev,
1884 struct device_attribute *attr, char *buf)
1886 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
1888 return sprintf(buf, "%s\n", rbd_dev->pool_name);
1891 static ssize_t rbd_pool_id_show(struct device *dev,
1892 struct device_attribute *attr, char *buf)
1894 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
1896 return sprintf(buf, "%d\n", rbd_dev->pool_id);
1899 static ssize_t rbd_name_show(struct device *dev,
1900 struct device_attribute *attr, char *buf)
1902 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
1904 return sprintf(buf, "%s\n", rbd_dev->image_name);
1907 static ssize_t rbd_snap_show(struct device *dev,
1908 struct device_attribute *attr,
1911 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
1913 return sprintf(buf, "%s\n", rbd_dev->snap_name);
1916 static ssize_t rbd_image_refresh(struct device *dev,
1917 struct device_attribute *attr,
1921 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
1925 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
1927 rc = __rbd_refresh_header(rbd_dev);
1931 mutex_unlock(&ctl_mutex);
1935 static DEVICE_ATTR(size, S_IRUGO, rbd_size_show, NULL);
1936 static DEVICE_ATTR(major, S_IRUGO, rbd_major_show, NULL);
1937 static DEVICE_ATTR(client_id, S_IRUGO, rbd_client_id_show, NULL);
1938 static DEVICE_ATTR(pool, S_IRUGO, rbd_pool_show, NULL);
1939 static DEVICE_ATTR(pool_id, S_IRUGO, rbd_pool_id_show, NULL);
1940 static DEVICE_ATTR(name, S_IRUGO, rbd_name_show, NULL);
1941 static DEVICE_ATTR(refresh, S_IWUSR, NULL, rbd_image_refresh);
1942 static DEVICE_ATTR(current_snap, S_IRUGO, rbd_snap_show, NULL);
1943 static DEVICE_ATTR(create_snap, S_IWUSR, NULL, rbd_snap_add);
1945 static struct attribute *rbd_attrs[] = {
1946 &dev_attr_size.attr,
1947 &dev_attr_major.attr,
1948 &dev_attr_client_id.attr,
1949 &dev_attr_pool.attr,
1950 &dev_attr_pool_id.attr,
1951 &dev_attr_name.attr,
1952 &dev_attr_current_snap.attr,
1953 &dev_attr_refresh.attr,
1954 &dev_attr_create_snap.attr,
1958 static struct attribute_group rbd_attr_group = {
1962 static const struct attribute_group *rbd_attr_groups[] = {
1967 static void rbd_sysfs_dev_release(struct device *dev)
1971 static struct device_type rbd_device_type = {
1973 .groups = rbd_attr_groups,
1974 .release = rbd_sysfs_dev_release,
1982 static ssize_t rbd_snap_size_show(struct device *dev,
1983 struct device_attribute *attr,
1986 struct rbd_snap *snap = container_of(dev, struct rbd_snap, dev);
1988 return sprintf(buf, "%llu\n", (unsigned long long)snap->size);
1991 static ssize_t rbd_snap_id_show(struct device *dev,
1992 struct device_attribute *attr,
1995 struct rbd_snap *snap = container_of(dev, struct rbd_snap, dev);
1997 return sprintf(buf, "%llu\n", (unsigned long long)snap->id);
2000 static DEVICE_ATTR(snap_size, S_IRUGO, rbd_snap_size_show, NULL);
2001 static DEVICE_ATTR(snap_id, S_IRUGO, rbd_snap_id_show, NULL);
2003 static struct attribute *rbd_snap_attrs[] = {
2004 &dev_attr_snap_size.attr,
2005 &dev_attr_snap_id.attr,
2009 static struct attribute_group rbd_snap_attr_group = {
2010 .attrs = rbd_snap_attrs,
2013 static void rbd_snap_dev_release(struct device *dev)
2015 struct rbd_snap *snap = container_of(dev, struct rbd_snap, dev);
2020 static const struct attribute_group *rbd_snap_attr_groups[] = {
2021 &rbd_snap_attr_group,
2025 static struct device_type rbd_snap_device_type = {
2026 .groups = rbd_snap_attr_groups,
2027 .release = rbd_snap_dev_release,
2030 static void __rbd_remove_snap_dev(struct rbd_device *rbd_dev,
2031 struct rbd_snap *snap)
2033 list_del(&snap->node);
2034 device_unregister(&snap->dev);
2037 static int rbd_register_snap_dev(struct rbd_device *rbd_dev,
2038 struct rbd_snap *snap,
2039 struct device *parent)
2041 struct device *dev = &snap->dev;
2044 dev->type = &rbd_snap_device_type;
2045 dev->parent = parent;
2046 dev->release = rbd_snap_dev_release;
2047 dev_set_name(dev, "snap_%s", snap->name);
2048 ret = device_register(dev);
2053 static int __rbd_add_snap_dev(struct rbd_device *rbd_dev,
2054 int i, const char *name,
2055 struct rbd_snap **snapp)
2058 struct rbd_snap *snap = kzalloc(sizeof(*snap), GFP_KERNEL);
2061 snap->name = kstrdup(name, GFP_KERNEL);
2062 snap->size = rbd_dev->header.snap_sizes[i];
2063 snap->id = rbd_dev->header.snapc->snaps[i];
2064 if (device_is_registered(&rbd_dev->dev)) {
2065 ret = rbd_register_snap_dev(rbd_dev, snap,
2079 * search for the previous snap in a null delimited string list
2081 const char *rbd_prev_snap_name(const char *name, const char *start)
2083 if (name < start + 2)
2096 * compare the old list of snapshots that we have to what's in the header
2097 * and update it accordingly. Note that the header holds the snapshots
2098 * in a reverse order (from newest to oldest) and we need to go from
2099 * older to new so that we don't get a duplicate snap name when
2100 * doing the process (e.g., removed snapshot and recreated a new
2101 * one with the same name.
2103 static int __rbd_init_snaps_header(struct rbd_device *rbd_dev)
2105 const char *name, *first_name;
2106 int i = rbd_dev->header.total_snaps;
2107 struct rbd_snap *snap, *old_snap = NULL;
2109 struct list_head *p, *n;
2111 first_name = rbd_dev->header.snap_names;
2112 name = first_name + rbd_dev->header.snap_names_len;
2114 list_for_each_prev_safe(p, n, &rbd_dev->snaps) {
2117 old_snap = list_entry(p, struct rbd_snap, node);
2120 cur_id = rbd_dev->header.snapc->snaps[i - 1];
2122 if (!i || old_snap->id < cur_id) {
2124 * old_snap->id was skipped, thus was
2125 * removed. If this rbd_dev is mapped to
2126 * the removed snapshot, record that it no
2127 * longer exists, to prevent further I/O.
2129 if (rbd_dev->snap_id == old_snap->id)
2130 rbd_dev->snap_exists = false;
2131 __rbd_remove_snap_dev(rbd_dev, old_snap);
2134 if (old_snap->id == cur_id) {
2135 /* we have this snapshot already */
2137 name = rbd_prev_snap_name(name, first_name);
2141 i--, name = rbd_prev_snap_name(name, first_name)) {
2146 cur_id = rbd_dev->header.snapc->snaps[i];
2147 /* snapshot removal? handle it above */
2148 if (cur_id >= old_snap->id)
2150 /* a new snapshot */
2151 ret = __rbd_add_snap_dev(rbd_dev, i - 1, name, &snap);
2155 /* note that we add it backward so using n and not p */
2156 list_add(&snap->node, n);
2160 /* we're done going over the old snap list, just add what's left */
2161 for (; i > 0; i--) {
2162 name = rbd_prev_snap_name(name, first_name);
2167 ret = __rbd_add_snap_dev(rbd_dev, i - 1, name, &snap);
2170 list_add(&snap->node, &rbd_dev->snaps);
2176 static int rbd_bus_add_dev(struct rbd_device *rbd_dev)
2180 struct rbd_snap *snap;
2182 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
2183 dev = &rbd_dev->dev;
2185 dev->bus = &rbd_bus_type;
2186 dev->type = &rbd_device_type;
2187 dev->parent = &rbd_root_dev;
2188 dev->release = rbd_dev_release;
2189 dev_set_name(dev, "%d", rbd_dev->id);
2190 ret = device_register(dev);
2194 list_for_each_entry(snap, &rbd_dev->snaps, node) {
2195 ret = rbd_register_snap_dev(rbd_dev, snap,
2201 mutex_unlock(&ctl_mutex);
2205 static void rbd_bus_del_dev(struct rbd_device *rbd_dev)
2207 device_unregister(&rbd_dev->dev);
2210 static int rbd_init_watch_dev(struct rbd_device *rbd_dev)
2215 ret = rbd_req_sync_watch(rbd_dev, rbd_dev->header_name,
2216 rbd_dev->header.obj_version);
2217 if (ret == -ERANGE) {
2218 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
2219 rc = __rbd_refresh_header(rbd_dev);
2220 mutex_unlock(&ctl_mutex);
2224 } while (ret == -ERANGE);
2229 static atomic64_t rbd_id_max = ATOMIC64_INIT(0);
2232 * Get a unique rbd identifier for the given new rbd_dev, and add
2233 * the rbd_dev to the global list. The minimum rbd id is 1.
2235 static void rbd_id_get(struct rbd_device *rbd_dev)
2237 rbd_dev->id = atomic64_inc_return(&rbd_id_max);
2239 spin_lock(&rbd_dev_list_lock);
2240 list_add_tail(&rbd_dev->node, &rbd_dev_list);
2241 spin_unlock(&rbd_dev_list_lock);
2245 * Remove an rbd_dev from the global list, and record that its
2246 * identifier is no longer in use.
2248 static void rbd_id_put(struct rbd_device *rbd_dev)
2250 struct list_head *tmp;
2251 int rbd_id = rbd_dev->id;
2256 spin_lock(&rbd_dev_list_lock);
2257 list_del_init(&rbd_dev->node);
2260 * If the id being "put" is not the current maximum, there
2261 * is nothing special we need to do.
2263 if (rbd_id != atomic64_read(&rbd_id_max)) {
2264 spin_unlock(&rbd_dev_list_lock);
2269 * We need to update the current maximum id. Search the
2270 * list to find out what it is. We're more likely to find
2271 * the maximum at the end, so search the list backward.
2274 list_for_each_prev(tmp, &rbd_dev_list) {
2275 struct rbd_device *rbd_dev;
2277 rbd_dev = list_entry(tmp, struct rbd_device, node);
2278 if (rbd_id > max_id)
2281 spin_unlock(&rbd_dev_list_lock);
2284 * The max id could have been updated by rbd_id_get(), in
2285 * which case it now accurately reflects the new maximum.
2286 * Be careful not to overwrite the maximum value in that
2289 atomic64_cmpxchg(&rbd_id_max, rbd_id, max_id);
2293 * Skips over white space at *buf, and updates *buf to point to the
2294 * first found non-space character (if any). Returns the length of
2295 * the token (string of non-white space characters) found. Note
2296 * that *buf must be terminated with '\0'.
2298 static inline size_t next_token(const char **buf)
2301 * These are the characters that produce nonzero for
2302 * isspace() in the "C" and "POSIX" locales.
2304 const char *spaces = " \f\n\r\t\v";
2306 *buf += strspn(*buf, spaces); /* Find start of token */
2308 return strcspn(*buf, spaces); /* Return token length */
2312 * Finds the next token in *buf, and if the provided token buffer is
2313 * big enough, copies the found token into it. The result, if
2314 * copied, is guaranteed to be terminated with '\0'. Note that *buf
2315 * must be terminated with '\0' on entry.
2317 * Returns the length of the token found (not including the '\0').
2318 * Return value will be 0 if no token is found, and it will be >=
2319 * token_size if the token would not fit.
2321 * The *buf pointer will be updated to point beyond the end of the
2322 * found token. Note that this occurs even if the token buffer is
2323 * too small to hold it.
2325 static inline size_t copy_token(const char **buf,
2331 len = next_token(buf);
2332 if (len < token_size) {
2333 memcpy(token, *buf, len);
2334 *(token + len) = '\0';
2342 * Finds the next token in *buf, dynamically allocates a buffer big
2343 * enough to hold a copy of it, and copies the token into the new
2344 * buffer. The copy is guaranteed to be terminated with '\0'. Note
2345 * that a duplicate buffer is created even for a zero-length token.
2347 * Returns a pointer to the newly-allocated duplicate, or a null
2348 * pointer if memory for the duplicate was not available. If
2349 * the lenp argument is a non-null pointer, the length of the token
2350 * (not including the '\0') is returned in *lenp.
2352 * If successful, the *buf pointer will be updated to point beyond
2353 * the end of the found token.
2355 * Note: uses GFP_KERNEL for allocation.
2357 static inline char *dup_token(const char **buf, size_t *lenp)
2362 len = next_token(buf);
2363 dup = kmalloc(len + 1, GFP_KERNEL);
2367 memcpy(dup, *buf, len);
2368 *(dup + len) = '\0';
2378 * This fills in the pool_name, image_name, image_name_len, snap_name,
2379 * rbd_dev, rbd_md_name, and name fields of the given rbd_dev, based
2380 * on the list of monitor addresses and other options provided via
2383 * Note: rbd_dev is assumed to have been initially zero-filled.
2385 static int rbd_add_parse_args(struct rbd_device *rbd_dev,
2387 const char **mon_addrs,
2388 size_t *mon_addrs_size,
2390 size_t options_size)
2395 /* The first four tokens are required */
2397 len = next_token(&buf);
2400 *mon_addrs_size = len + 1;
2405 len = copy_token(&buf, options, options_size);
2406 if (!len || len >= options_size)
2410 rbd_dev->pool_name = dup_token(&buf, NULL);
2411 if (!rbd_dev->pool_name)
2414 rbd_dev->image_name = dup_token(&buf, &rbd_dev->image_name_len);
2415 if (!rbd_dev->image_name)
2418 /* Create the name of the header object */
2420 rbd_dev->header_name = kmalloc(rbd_dev->image_name_len
2421 + sizeof (RBD_SUFFIX),
2423 if (!rbd_dev->header_name)
2425 sprintf(rbd_dev->header_name, "%s%s", rbd_dev->image_name, RBD_SUFFIX);
2428 * The snapshot name is optional. If none is is supplied,
2429 * we use the default value.
2431 rbd_dev->snap_name = dup_token(&buf, &len);
2432 if (!rbd_dev->snap_name)
2435 /* Replace the empty name with the default */
2436 kfree(rbd_dev->snap_name);
2438 = kmalloc(sizeof (RBD_SNAP_HEAD_NAME), GFP_KERNEL);
2439 if (!rbd_dev->snap_name)
2442 memcpy(rbd_dev->snap_name, RBD_SNAP_HEAD_NAME,
2443 sizeof (RBD_SNAP_HEAD_NAME));
2449 kfree(rbd_dev->header_name);
2450 kfree(rbd_dev->image_name);
2451 kfree(rbd_dev->pool_name);
2452 rbd_dev->pool_name = NULL;
2457 static ssize_t rbd_add(struct bus_type *bus,
2462 struct rbd_device *rbd_dev = NULL;
2463 const char *mon_addrs = NULL;
2464 size_t mon_addrs_size = 0;
2465 struct ceph_osd_client *osdc;
2468 if (!try_module_get(THIS_MODULE))
2471 options = kmalloc(count, GFP_KERNEL);
2474 rbd_dev = kzalloc(sizeof(*rbd_dev), GFP_KERNEL);
2478 /* static rbd_device initialization */
2479 spin_lock_init(&rbd_dev->lock);
2480 INIT_LIST_HEAD(&rbd_dev->node);
2481 INIT_LIST_HEAD(&rbd_dev->snaps);
2482 init_rwsem(&rbd_dev->header_rwsem);
2484 init_rwsem(&rbd_dev->header_rwsem);
2486 /* generate unique id: find highest unique id, add one */
2487 rbd_id_get(rbd_dev);
2489 /* Fill in the device name, now that we have its id. */
2490 BUILD_BUG_ON(DEV_NAME_LEN
2491 < sizeof (RBD_DRV_NAME) + MAX_INT_FORMAT_WIDTH);
2492 sprintf(rbd_dev->name, "%s%d", RBD_DRV_NAME, rbd_dev->id);
2494 /* parse add command */
2495 rc = rbd_add_parse_args(rbd_dev, buf, &mon_addrs, &mon_addrs_size,
2500 rbd_dev->rbd_client = rbd_get_client(mon_addrs, mon_addrs_size - 1,
2502 if (IS_ERR(rbd_dev->rbd_client)) {
2503 rc = PTR_ERR(rbd_dev->rbd_client);
2508 osdc = &rbd_dev->rbd_client->client->osdc;
2509 rc = ceph_pg_poolid_by_name(osdc->osdmap, rbd_dev->pool_name);
2511 goto err_out_client;
2512 rbd_dev->pool_id = rc;
2514 /* register our block device */
2515 rc = register_blkdev(0, rbd_dev->name);
2517 goto err_out_client;
2518 rbd_dev->major = rc;
2520 rc = rbd_bus_add_dev(rbd_dev);
2522 goto err_out_blkdev;
2525 * At this point cleanup in the event of an error is the job
2526 * of the sysfs code (initiated by rbd_bus_del_dev()).
2528 * Set up and announce blkdev mapping.
2530 rc = rbd_init_disk(rbd_dev);
2534 rc = rbd_init_watch_dev(rbd_dev);
2541 /* this will also clean up rest of rbd_dev stuff */
2543 rbd_bus_del_dev(rbd_dev);
2548 unregister_blkdev(rbd_dev->major, rbd_dev->name);
2550 rbd_put_client(rbd_dev);
2552 if (rbd_dev->pool_name) {
2553 kfree(rbd_dev->snap_name);
2554 kfree(rbd_dev->header_name);
2555 kfree(rbd_dev->image_name);
2556 kfree(rbd_dev->pool_name);
2558 rbd_id_put(rbd_dev);
2563 dout("Error adding device %s\n", buf);
2564 module_put(THIS_MODULE);
2566 return (ssize_t) rc;
2569 static struct rbd_device *__rbd_get_dev(unsigned long id)
2571 struct list_head *tmp;
2572 struct rbd_device *rbd_dev;
2574 spin_lock(&rbd_dev_list_lock);
2575 list_for_each(tmp, &rbd_dev_list) {
2576 rbd_dev = list_entry(tmp, struct rbd_device, node);
2577 if (rbd_dev->id == id) {
2578 spin_unlock(&rbd_dev_list_lock);
2582 spin_unlock(&rbd_dev_list_lock);
2586 static void rbd_dev_release(struct device *dev)
2588 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
2590 if (rbd_dev->watch_request) {
2591 struct ceph_client *client = rbd_dev->rbd_client->client;
2593 ceph_osdc_unregister_linger_request(&client->osdc,
2594 rbd_dev->watch_request);
2596 if (rbd_dev->watch_event)
2597 rbd_req_sync_unwatch(rbd_dev, rbd_dev->header_name);
2599 rbd_put_client(rbd_dev);
2601 /* clean up and free blkdev */
2602 rbd_free_disk(rbd_dev);
2603 unregister_blkdev(rbd_dev->major, rbd_dev->name);
2605 /* done with the id, and with the rbd_dev */
2606 kfree(rbd_dev->snap_name);
2607 kfree(rbd_dev->header_name);
2608 kfree(rbd_dev->pool_name);
2609 kfree(rbd_dev->image_name);
2610 rbd_id_put(rbd_dev);
2613 /* release module ref */
2614 module_put(THIS_MODULE);
2617 static ssize_t rbd_remove(struct bus_type *bus,
2621 struct rbd_device *rbd_dev = NULL;
2626 rc = strict_strtoul(buf, 10, &ul);
2630 /* convert to int; abort if we lost anything in the conversion */
2631 target_id = (int) ul;
2632 if (target_id != ul)
2635 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
2637 rbd_dev = __rbd_get_dev(target_id);
2643 __rbd_remove_all_snaps(rbd_dev);
2644 rbd_bus_del_dev(rbd_dev);
2647 mutex_unlock(&ctl_mutex);
2651 static ssize_t rbd_snap_add(struct device *dev,
2652 struct device_attribute *attr,
2656 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
2658 char *name = kmalloc(count + 1, GFP_KERNEL);
2662 snprintf(name, count, "%s", buf);
2664 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
2666 ret = rbd_header_add_snap(rbd_dev,
2671 ret = __rbd_refresh_header(rbd_dev);
2675 /* shouldn't hold ctl_mutex when notifying.. notify might
2676 trigger a watch callback that would need to get that mutex */
2677 mutex_unlock(&ctl_mutex);
2679 /* make a best effort, don't error if failed */
2680 rbd_req_sync_notify(rbd_dev, rbd_dev->header_name);
2687 mutex_unlock(&ctl_mutex);
2693 * create control files in sysfs
2696 static int rbd_sysfs_init(void)
2700 ret = device_register(&rbd_root_dev);
2704 ret = bus_register(&rbd_bus_type);
2706 device_unregister(&rbd_root_dev);
2711 static void rbd_sysfs_cleanup(void)
2713 bus_unregister(&rbd_bus_type);
2714 device_unregister(&rbd_root_dev);
2717 int __init rbd_init(void)
2721 rc = rbd_sysfs_init();
2724 pr_info("loaded " RBD_DRV_NAME_LONG "\n");
2728 void __exit rbd_exit(void)
2730 rbd_sysfs_cleanup();
2733 module_init(rbd_init);
2734 module_exit(rbd_exit);
2736 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
2737 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
2738 MODULE_DESCRIPTION("rados block device");
2740 /* following authorship retained from original osdblk.c */
2741 MODULE_AUTHOR("Jeff Garzik <jeff@garzik.org>");
2743 MODULE_LICENSE("GPL");