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 dev_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_snap *snap);
207 static ssize_t rbd_add(struct bus_type *bus, const char *buf,
209 static ssize_t rbd_remove(struct bus_type *bus, const char *buf,
212 static struct bus_attribute rbd_bus_attrs[] = {
213 __ATTR(add, S_IWUSR, NULL, rbd_add),
214 __ATTR(remove, S_IWUSR, NULL, rbd_remove),
218 static struct bus_type rbd_bus_type = {
220 .bus_attrs = rbd_bus_attrs,
223 static void rbd_root_dev_release(struct device *dev)
227 static struct device rbd_root_dev = {
229 .release = rbd_root_dev_release,
233 static struct device *rbd_get_dev(struct rbd_device *rbd_dev)
235 return get_device(&rbd_dev->dev);
238 static void rbd_put_dev(struct rbd_device *rbd_dev)
240 put_device(&rbd_dev->dev);
243 static int __rbd_refresh_header(struct rbd_device *rbd_dev);
245 static int rbd_open(struct block_device *bdev, fmode_t mode)
247 struct rbd_device *rbd_dev = bdev->bd_disk->private_data;
249 rbd_get_dev(rbd_dev);
251 set_device_ro(bdev, rbd_dev->read_only);
253 if ((mode & FMODE_WRITE) && rbd_dev->read_only)
259 static int rbd_release(struct gendisk *disk, fmode_t mode)
261 struct rbd_device *rbd_dev = disk->private_data;
263 rbd_put_dev(rbd_dev);
268 static const struct block_device_operations rbd_bd_ops = {
269 .owner = THIS_MODULE,
271 .release = rbd_release,
275 * Initialize an rbd client instance.
278 static struct rbd_client *rbd_client_create(struct ceph_options *ceph_opts,
279 struct rbd_options *rbd_opts)
281 struct rbd_client *rbdc;
284 dout("rbd_client_create\n");
285 rbdc = kmalloc(sizeof(struct rbd_client), GFP_KERNEL);
289 kref_init(&rbdc->kref);
290 INIT_LIST_HEAD(&rbdc->node);
292 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
294 rbdc->client = ceph_create_client(ceph_opts, rbdc, 0, 0);
295 if (IS_ERR(rbdc->client))
297 ceph_opts = NULL; /* Now rbdc->client is responsible for ceph_opts */
299 ret = ceph_open_session(rbdc->client);
303 rbdc->rbd_opts = rbd_opts;
305 spin_lock(&rbd_client_list_lock);
306 list_add_tail(&rbdc->node, &rbd_client_list);
307 spin_unlock(&rbd_client_list_lock);
309 mutex_unlock(&ctl_mutex);
311 dout("rbd_client_create created %p\n", rbdc);
315 ceph_destroy_client(rbdc->client);
317 mutex_unlock(&ctl_mutex);
321 ceph_destroy_options(ceph_opts);
326 * Find a ceph client with specific addr and configuration.
328 static struct rbd_client *__rbd_client_find(struct ceph_options *ceph_opts)
330 struct rbd_client *client_node;
332 if (ceph_opts->flags & CEPH_OPT_NOSHARE)
335 list_for_each_entry(client_node, &rbd_client_list, node)
336 if (!ceph_compare_options(ceph_opts, client_node->client))
349 /* string args above */
352 static match_table_t rbd_opts_tokens = {
353 {Opt_notify_timeout, "notify_timeout=%d"},
355 /* string args above */
359 static int parse_rbd_opts_token(char *c, void *private)
361 struct rbd_options *rbd_opts = private;
362 substring_t argstr[MAX_OPT_ARGS];
363 int token, intval, ret;
365 token = match_token(c, rbd_opts_tokens, argstr);
369 if (token < Opt_last_int) {
370 ret = match_int(&argstr[0], &intval);
372 pr_err("bad mount option arg (not int) "
376 dout("got int token %d val %d\n", token, intval);
377 } else if (token > Opt_last_int && token < Opt_last_string) {
378 dout("got string token %d val %s\n", token,
381 dout("got token %d\n", token);
385 case Opt_notify_timeout:
386 rbd_opts->notify_timeout = intval;
395 * Get a ceph client with specific addr and configuration, if one does
396 * not exist create it.
398 static struct rbd_client *rbd_get_client(const char *mon_addr,
402 struct rbd_client *rbdc;
403 struct ceph_options *ceph_opts;
404 struct rbd_options *rbd_opts;
406 rbd_opts = kzalloc(sizeof(*rbd_opts), GFP_KERNEL);
408 return ERR_PTR(-ENOMEM);
410 rbd_opts->notify_timeout = RBD_NOTIFY_TIMEOUT_DEFAULT;
412 ceph_opts = ceph_parse_options(options, mon_addr,
413 mon_addr + mon_addr_len,
414 parse_rbd_opts_token, rbd_opts);
415 if (IS_ERR(ceph_opts)) {
417 return ERR_CAST(ceph_opts);
420 spin_lock(&rbd_client_list_lock);
421 rbdc = __rbd_client_find(ceph_opts);
423 /* using an existing client */
424 kref_get(&rbdc->kref);
425 spin_unlock(&rbd_client_list_lock);
427 ceph_destroy_options(ceph_opts);
432 spin_unlock(&rbd_client_list_lock);
434 rbdc = rbd_client_create(ceph_opts, rbd_opts);
443 * Destroy ceph client
445 * Caller must hold rbd_client_list_lock.
447 static void rbd_client_release(struct kref *kref)
449 struct rbd_client *rbdc = container_of(kref, struct rbd_client, kref);
451 dout("rbd_release_client %p\n", rbdc);
452 spin_lock(&rbd_client_list_lock);
453 list_del(&rbdc->node);
454 spin_unlock(&rbd_client_list_lock);
456 ceph_destroy_client(rbdc->client);
457 kfree(rbdc->rbd_opts);
462 * Drop reference to ceph client node. If it's not referenced anymore, release
465 static void rbd_put_client(struct rbd_device *rbd_dev)
467 kref_put(&rbd_dev->rbd_client->kref, rbd_client_release);
468 rbd_dev->rbd_client = NULL;
472 * Destroy requests collection
474 static void rbd_coll_release(struct kref *kref)
476 struct rbd_req_coll *coll =
477 container_of(kref, struct rbd_req_coll, kref);
479 dout("rbd_coll_release %p\n", coll);
483 static bool rbd_dev_ondisk_valid(struct rbd_image_header_ondisk *ondisk)
485 return !memcmp(&ondisk->text,
486 RBD_HEADER_TEXT, sizeof (RBD_HEADER_TEXT));
490 * Create a new header structure, translate header format from the on-disk
493 static int rbd_header_from_disk(struct rbd_image_header *header,
494 struct rbd_image_header_ondisk *ondisk,
499 if (!rbd_dev_ondisk_valid(ondisk))
502 snap_count = le32_to_cpu(ondisk->snap_count);
503 if (snap_count > (SIZE_MAX - sizeof(struct ceph_snap_context))
506 header->snapc = kmalloc(sizeof(struct ceph_snap_context) +
507 snap_count * sizeof(u64),
513 header->snap_names_len = le64_to_cpu(ondisk->snap_names_len);
514 header->snap_names = kmalloc(header->snap_names_len,
516 if (!header->snap_names)
518 header->snap_sizes = kmalloc(snap_count * sizeof(u64),
520 if (!header->snap_sizes)
523 WARN_ON(ondisk->snap_names_len);
524 header->snap_names_len = 0;
525 header->snap_names = NULL;
526 header->snap_sizes = NULL;
529 header->object_prefix = kmalloc(sizeof (ondisk->block_name) + 1,
531 if (!header->object_prefix)
534 memcpy(header->object_prefix, ondisk->block_name,
535 sizeof(ondisk->block_name));
536 header->object_prefix[sizeof (ondisk->block_name)] = '\0';
538 header->image_size = le64_to_cpu(ondisk->image_size);
539 header->obj_order = ondisk->options.order;
540 header->crypt_type = ondisk->options.crypt_type;
541 header->comp_type = ondisk->options.comp_type;
543 atomic_set(&header->snapc->nref, 1);
544 header->snapc->seq = le64_to_cpu(ondisk->snap_seq);
545 header->snapc->num_snaps = snap_count;
546 header->total_snaps = snap_count;
548 if (snap_count && allocated_snaps == snap_count) {
551 for (i = 0; i < snap_count; i++) {
552 header->snapc->snaps[i] =
553 le64_to_cpu(ondisk->snaps[i].id);
554 header->snap_sizes[i] =
555 le64_to_cpu(ondisk->snaps[i].image_size);
558 /* copy snapshot names */
559 memcpy(header->snap_names, &ondisk->snaps[snap_count],
560 header->snap_names_len);
566 kfree(header->snap_sizes);
567 header->snap_sizes = NULL;
569 kfree(header->snap_names);
570 header->snap_names = NULL;
572 kfree(header->snapc);
573 header->snapc = NULL;
578 static int snap_by_name(struct rbd_image_header *header, const char *snap_name,
582 char *p = header->snap_names;
584 for (i = 0; i < header->total_snaps; i++) {
585 if (!strcmp(snap_name, p)) {
587 /* Found it. Pass back its id and/or size */
590 *seq = header->snapc->snaps[i];
592 *size = header->snap_sizes[i];
595 p += strlen(p) + 1; /* Skip ahead to the next name */
600 static int rbd_header_set_snap(struct rbd_device *rbd_dev, u64 *size)
604 down_write(&rbd_dev->header_rwsem);
606 if (!memcmp(rbd_dev->snap_name, RBD_SNAP_HEAD_NAME,
607 sizeof (RBD_SNAP_HEAD_NAME))) {
608 rbd_dev->snap_id = CEPH_NOSNAP;
609 rbd_dev->snap_exists = false;
610 rbd_dev->read_only = 0;
612 *size = rbd_dev->header.image_size;
616 ret = snap_by_name(&rbd_dev->header, rbd_dev->snap_name,
620 rbd_dev->snap_id = snap_id;
621 rbd_dev->snap_exists = true;
622 rbd_dev->read_only = 1;
627 up_write(&rbd_dev->header_rwsem);
631 static void rbd_header_free(struct rbd_image_header *header)
633 kfree(header->object_prefix);
634 kfree(header->snap_sizes);
635 kfree(header->snap_names);
636 ceph_put_snap_context(header->snapc);
640 * get the actual striped segment name, offset and length
642 static u64 rbd_get_segment(struct rbd_image_header *header,
643 const char *object_prefix,
645 char *seg_name, u64 *segofs)
647 u64 seg = ofs >> header->obj_order;
650 snprintf(seg_name, RBD_MAX_SEG_NAME_LEN,
651 "%s.%012llx", object_prefix, seg);
653 ofs = ofs & ((1 << header->obj_order) - 1);
654 len = min_t(u64, len, (1 << header->obj_order) - ofs);
662 static int rbd_get_num_segments(struct rbd_image_header *header,
665 u64 start_seg = ofs >> header->obj_order;
666 u64 end_seg = (ofs + len - 1) >> header->obj_order;
667 return end_seg - start_seg + 1;
671 * returns the size of an object in the image
673 static u64 rbd_obj_bytes(struct rbd_image_header *header)
675 return 1 << header->obj_order;
682 static void bio_chain_put(struct bio *chain)
688 chain = chain->bi_next;
694 * zeros a bio chain, starting at specific offset
696 static void zero_bio_chain(struct bio *chain, int start_ofs)
705 bio_for_each_segment(bv, chain, i) {
706 if (pos + bv->bv_len > start_ofs) {
707 int remainder = max(start_ofs - pos, 0);
708 buf = bvec_kmap_irq(bv, &flags);
709 memset(buf + remainder, 0,
710 bv->bv_len - remainder);
711 bvec_kunmap_irq(buf, &flags);
716 chain = chain->bi_next;
721 * bio_chain_clone - clone a chain of bios up to a certain length.
722 * might return a bio_pair that will need to be released.
724 static struct bio *bio_chain_clone(struct bio **old, struct bio **next,
725 struct bio_pair **bp,
726 int len, gfp_t gfpmask)
728 struct bio *tmp, *old_chain = *old, *new_chain = NULL, *tail = NULL;
732 bio_pair_release(*bp);
736 while (old_chain && (total < len)) {
737 tmp = bio_kmalloc(gfpmask, old_chain->bi_max_vecs);
741 if (total + old_chain->bi_size > len) {
745 * this split can only happen with a single paged bio,
746 * split_bio will BUG_ON if this is not the case
748 dout("bio_chain_clone split! total=%d remaining=%d"
750 total, len - total, old_chain->bi_size);
752 /* split the bio. We'll release it either in the next
753 call, or it will have to be released outside */
754 bp = bio_split(old_chain, (len - total) / SECTOR_SIZE);
758 __bio_clone(tmp, &bp->bio1);
762 __bio_clone(tmp, old_chain);
763 *next = old_chain->bi_next;
767 gfpmask &= ~__GFP_WAIT;
771 new_chain = tail = tmp;
776 old_chain = old_chain->bi_next;
778 total += tmp->bi_size;
784 tail->bi_next = NULL;
791 dout("bio_chain_clone with err\n");
792 bio_chain_put(new_chain);
797 * helpers for osd request op vectors.
799 static struct ceph_osd_req_op *rbd_create_rw_ops(int num_ops,
800 int opcode, u32 payload_len)
802 struct ceph_osd_req_op *ops;
804 ops = kzalloc(sizeof (*ops) * (num_ops + 1), GFP_NOIO);
811 * op extent offset and length will be set later on
812 * in calc_raw_layout()
814 ops[0].payload_len = payload_len;
819 static void rbd_destroy_ops(struct ceph_osd_req_op *ops)
824 static void rbd_coll_end_req_index(struct request *rq,
825 struct rbd_req_coll *coll,
829 struct request_queue *q;
832 dout("rbd_coll_end_req_index %p index %d ret %d len %llu\n",
833 coll, index, ret, (unsigned long long) len);
839 blk_end_request(rq, ret, len);
845 spin_lock_irq(q->queue_lock);
846 coll->status[index].done = 1;
847 coll->status[index].rc = ret;
848 coll->status[index].bytes = len;
849 max = min = coll->num_done;
850 while (max < coll->total && coll->status[max].done)
853 for (i = min; i<max; i++) {
854 __blk_end_request(rq, coll->status[i].rc,
855 coll->status[i].bytes);
857 kref_put(&coll->kref, rbd_coll_release);
859 spin_unlock_irq(q->queue_lock);
862 static void rbd_coll_end_req(struct rbd_request *req,
865 rbd_coll_end_req_index(req->rq, req->coll, req->coll_index, ret, len);
869 * Send ceph osd request
871 static int rbd_do_request(struct request *rq,
872 struct rbd_device *rbd_dev,
873 struct ceph_snap_context *snapc,
875 const char *object_name, u64 ofs, u64 len,
880 struct ceph_osd_req_op *ops,
881 struct rbd_req_coll *coll,
883 void (*rbd_cb)(struct ceph_osd_request *req,
884 struct ceph_msg *msg),
885 struct ceph_osd_request **linger_req,
888 struct ceph_osd_request *req;
889 struct ceph_file_layout *layout;
892 struct timespec mtime = CURRENT_TIME;
893 struct rbd_request *req_data;
894 struct ceph_osd_request_head *reqhead;
895 struct ceph_osd_client *osdc;
897 req_data = kzalloc(sizeof(*req_data), GFP_NOIO);
900 rbd_coll_end_req_index(rq, coll, coll_index,
906 req_data->coll = coll;
907 req_data->coll_index = coll_index;
910 dout("rbd_do_request object_name=%s ofs=%llu len=%llu\n", object_name,
911 (unsigned long long) ofs, (unsigned long long) len);
913 osdc = &rbd_dev->rbd_client->client->osdc;
914 req = ceph_osdc_alloc_request(osdc, flags, snapc, ops,
915 false, GFP_NOIO, pages, bio);
921 req->r_callback = rbd_cb;
925 req_data->pages = pages;
928 req->r_priv = req_data;
930 reqhead = req->r_request->front.iov_base;
931 reqhead->snapid = cpu_to_le64(CEPH_NOSNAP);
933 strncpy(req->r_oid, object_name, sizeof(req->r_oid));
934 req->r_oid_len = strlen(req->r_oid);
936 layout = &req->r_file_layout;
937 memset(layout, 0, sizeof(*layout));
938 layout->fl_stripe_unit = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER);
939 layout->fl_stripe_count = cpu_to_le32(1);
940 layout->fl_object_size = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER);
941 layout->fl_pg_pool = cpu_to_le32(rbd_dev->pool_id);
942 ceph_calc_raw_layout(osdc, layout, snapid, ofs, &len, &bno,
945 ceph_osdc_build_request(req, ofs, &len,
949 req->r_oid, req->r_oid_len);
952 ceph_osdc_set_request_linger(osdc, req);
956 ret = ceph_osdc_start_request(osdc, req, false);
961 ret = ceph_osdc_wait_request(osdc, req);
963 *ver = le64_to_cpu(req->r_reassert_version.version);
964 dout("reassert_ver=%llu\n",
966 le64_to_cpu(req->r_reassert_version.version));
967 ceph_osdc_put_request(req);
972 bio_chain_put(req_data->bio);
973 ceph_osdc_put_request(req);
975 rbd_coll_end_req(req_data, ret, len);
981 * Ceph osd op callback
983 static void rbd_req_cb(struct ceph_osd_request *req, struct ceph_msg *msg)
985 struct rbd_request *req_data = req->r_priv;
986 struct ceph_osd_reply_head *replyhead;
987 struct ceph_osd_op *op;
993 replyhead = msg->front.iov_base;
994 WARN_ON(le32_to_cpu(replyhead->num_ops) == 0);
995 op = (void *)(replyhead + 1);
996 rc = le32_to_cpu(replyhead->result);
997 bytes = le64_to_cpu(op->extent.length);
998 read_op = (le16_to_cpu(op->op) == CEPH_OSD_OP_READ);
1000 dout("rbd_req_cb bytes=%llu readop=%d rc=%d\n",
1001 (unsigned long long) bytes, read_op, (int) rc);
1003 if (rc == -ENOENT && read_op) {
1004 zero_bio_chain(req_data->bio, 0);
1006 } else if (rc == 0 && read_op && bytes < req_data->len) {
1007 zero_bio_chain(req_data->bio, bytes);
1008 bytes = req_data->len;
1011 rbd_coll_end_req(req_data, rc, bytes);
1014 bio_chain_put(req_data->bio);
1016 ceph_osdc_put_request(req);
1020 static void rbd_simple_req_cb(struct ceph_osd_request *req, struct ceph_msg *msg)
1022 ceph_osdc_put_request(req);
1026 * Do a synchronous ceph osd operation
1028 static int rbd_req_sync_op(struct rbd_device *rbd_dev,
1029 struct ceph_snap_context *snapc,
1032 struct ceph_osd_req_op *ops,
1033 const char *object_name,
1036 struct ceph_osd_request **linger_req,
1040 struct page **pages;
1043 BUG_ON(ops == NULL);
1045 num_pages = calc_pages_for(ofs , len);
1046 pages = ceph_alloc_page_vector(num_pages, GFP_KERNEL);
1048 return PTR_ERR(pages);
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);
1065 ceph_release_page_vector(pages, num_pages);
1070 * Do an asynchronous ceph osd operation
1072 static int rbd_do_op(struct request *rq,
1073 struct rbd_device *rbd_dev,
1074 struct ceph_snap_context *snapc,
1076 int opcode, int flags,
1079 struct rbd_req_coll *coll,
1086 struct ceph_osd_req_op *ops;
1089 seg_name = kmalloc(RBD_MAX_SEG_NAME_LEN + 1, GFP_NOIO);
1093 seg_len = rbd_get_segment(&rbd_dev->header,
1094 rbd_dev->header.object_prefix,
1096 seg_name, &seg_ofs);
1098 payload_len = (flags & CEPH_OSD_FLAG_WRITE ? seg_len : 0);
1101 ops = rbd_create_rw_ops(1, opcode, payload_len);
1105 /* we've taken care of segment sizes earlier when we
1106 cloned the bios. We should never have a segment
1107 truncated at this point */
1108 BUG_ON(seg_len < len);
1110 ret = rbd_do_request(rq, rbd_dev, snapc, snapid,
1111 seg_name, seg_ofs, seg_len,
1117 rbd_req_cb, 0, NULL);
1119 rbd_destroy_ops(ops);
1126 * Request async osd write
1128 static int rbd_req_write(struct request *rq,
1129 struct rbd_device *rbd_dev,
1130 struct ceph_snap_context *snapc,
1133 struct rbd_req_coll *coll,
1136 return rbd_do_op(rq, rbd_dev, snapc, CEPH_NOSNAP,
1138 CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK,
1139 ofs, len, bio, coll, coll_index);
1143 * Request async osd read
1145 static int rbd_req_read(struct request *rq,
1146 struct rbd_device *rbd_dev,
1150 struct rbd_req_coll *coll,
1153 return rbd_do_op(rq, rbd_dev, NULL,
1157 ofs, len, bio, coll, coll_index);
1161 * Request sync osd read
1163 static int rbd_req_sync_read(struct rbd_device *rbd_dev,
1165 const char *object_name,
1170 struct ceph_osd_req_op *ops;
1173 ops = rbd_create_rw_ops(1, CEPH_OSD_OP_READ, 0);
1177 ret = rbd_req_sync_op(rbd_dev, NULL,
1180 ops, object_name, ofs, len, buf, NULL, ver);
1181 rbd_destroy_ops(ops);
1187 * Request sync osd watch
1189 static int rbd_req_sync_notify_ack(struct rbd_device *rbd_dev,
1193 struct ceph_osd_req_op *ops;
1196 ops = rbd_create_rw_ops(1, CEPH_OSD_OP_NOTIFY_ACK, 0);
1200 ops[0].watch.ver = cpu_to_le64(ver);
1201 ops[0].watch.cookie = notify_id;
1202 ops[0].watch.flag = 0;
1204 ret = rbd_do_request(NULL, rbd_dev, NULL, CEPH_NOSNAP,
1205 rbd_dev->header_name, 0, 0, NULL,
1210 rbd_simple_req_cb, 0, NULL);
1212 rbd_destroy_ops(ops);
1216 static void rbd_watch_cb(u64 ver, u64 notify_id, u8 opcode, void *data)
1218 struct rbd_device *rbd_dev = (struct rbd_device *)data;
1225 dout("rbd_watch_cb %s notify_id=%llu opcode=%u\n",
1226 rbd_dev->header_name, (unsigned long long) notify_id,
1227 (unsigned int) opcode);
1228 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
1229 rc = __rbd_refresh_header(rbd_dev);
1230 hver = rbd_dev->header.obj_version;
1231 mutex_unlock(&ctl_mutex);
1233 pr_warning(RBD_DRV_NAME "%d got notification but failed to "
1234 " update snaps: %d\n", rbd_dev->major, rc);
1236 rbd_req_sync_notify_ack(rbd_dev, hver, notify_id);
1240 * Request sync osd watch
1242 static int rbd_req_sync_watch(struct rbd_device *rbd_dev)
1244 struct ceph_osd_req_op *ops;
1245 struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
1248 ops = rbd_create_rw_ops(1, CEPH_OSD_OP_WATCH, 0);
1252 ret = ceph_osdc_create_event(osdc, rbd_watch_cb, 0,
1253 (void *)rbd_dev, &rbd_dev->watch_event);
1257 ops[0].watch.ver = cpu_to_le64(rbd_dev->header.obj_version);
1258 ops[0].watch.cookie = cpu_to_le64(rbd_dev->watch_event->cookie);
1259 ops[0].watch.flag = 1;
1261 ret = rbd_req_sync_op(rbd_dev, NULL,
1263 CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK,
1265 rbd_dev->header_name,
1267 &rbd_dev->watch_request, NULL);
1272 rbd_destroy_ops(ops);
1276 ceph_osdc_cancel_event(rbd_dev->watch_event);
1277 rbd_dev->watch_event = NULL;
1279 rbd_destroy_ops(ops);
1284 * Request sync osd unwatch
1286 static int rbd_req_sync_unwatch(struct rbd_device *rbd_dev)
1288 struct ceph_osd_req_op *ops;
1291 ops = rbd_create_rw_ops(1, CEPH_OSD_OP_WATCH, 0);
1295 ops[0].watch.ver = 0;
1296 ops[0].watch.cookie = cpu_to_le64(rbd_dev->watch_event->cookie);
1297 ops[0].watch.flag = 0;
1299 ret = rbd_req_sync_op(rbd_dev, NULL,
1301 CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK,
1303 rbd_dev->header_name,
1304 0, 0, NULL, NULL, NULL);
1307 rbd_destroy_ops(ops);
1308 ceph_osdc_cancel_event(rbd_dev->watch_event);
1309 rbd_dev->watch_event = NULL;
1313 struct rbd_notify_info {
1314 struct rbd_device *rbd_dev;
1317 static void rbd_notify_cb(u64 ver, u64 notify_id, u8 opcode, void *data)
1319 struct rbd_device *rbd_dev = (struct rbd_device *)data;
1323 dout("rbd_notify_cb %s notify_id=%llu opcode=%u\n",
1324 rbd_dev->header_name, (unsigned long long) notify_id,
1325 (unsigned int) opcode);
1329 * Request sync osd notify
1331 static int rbd_req_sync_notify(struct rbd_device *rbd_dev)
1333 struct ceph_osd_req_op *ops;
1334 struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
1335 struct ceph_osd_event *event;
1336 struct rbd_notify_info info;
1337 int payload_len = sizeof(u32) + sizeof(u32);
1340 ops = rbd_create_rw_ops(1, CEPH_OSD_OP_NOTIFY, payload_len);
1344 info.rbd_dev = rbd_dev;
1346 ret = ceph_osdc_create_event(osdc, rbd_notify_cb, 1,
1347 (void *)&info, &event);
1351 ops[0].watch.ver = 1;
1352 ops[0].watch.flag = 1;
1353 ops[0].watch.cookie = event->cookie;
1354 ops[0].watch.prot_ver = RADOS_NOTIFY_VER;
1355 ops[0].watch.timeout = 12;
1357 ret = rbd_req_sync_op(rbd_dev, NULL,
1359 CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK,
1361 rbd_dev->header_name,
1362 0, 0, NULL, NULL, NULL);
1366 ret = ceph_osdc_wait_event(event, CEPH_OSD_TIMEOUT_DEFAULT);
1367 dout("ceph_osdc_wait_event returned %d\n", ret);
1368 rbd_destroy_ops(ops);
1372 ceph_osdc_cancel_event(event);
1374 rbd_destroy_ops(ops);
1379 * Request sync osd read
1381 static int rbd_req_sync_exec(struct rbd_device *rbd_dev,
1382 const char *object_name,
1383 const char *class_name,
1384 const char *method_name,
1389 struct ceph_osd_req_op *ops;
1390 int class_name_len = strlen(class_name);
1391 int method_name_len = strlen(method_name);
1394 ops = rbd_create_rw_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,
1409 CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK,
1411 object_name, 0, 0, NULL, NULL, ver);
1413 rbd_destroy_ops(ops);
1415 dout("cls_exec returned %d\n", ret);
1419 static struct rbd_req_coll *rbd_alloc_coll(int num_reqs)
1421 struct rbd_req_coll *coll =
1422 kzalloc(sizeof(struct rbd_req_coll) +
1423 sizeof(struct rbd_req_status) * num_reqs,
1428 coll->total = num_reqs;
1429 kref_init(&coll->kref);
1434 * block device queue callback
1436 static void rbd_rq_fn(struct request_queue *q)
1438 struct rbd_device *rbd_dev = q->queuedata;
1440 struct bio_pair *bp = NULL;
1442 while ((rq = blk_fetch_request(q))) {
1444 struct bio *rq_bio, *next_bio = NULL;
1449 int num_segs, cur_seg = 0;
1450 struct rbd_req_coll *coll;
1451 struct ceph_snap_context *snapc;
1453 /* peek at request from block layer */
1457 dout("fetched request\n");
1459 /* filter out block requests we don't understand */
1460 if ((rq->cmd_type != REQ_TYPE_FS)) {
1461 __blk_end_request_all(rq, 0);
1465 /* deduce our operation (read, write) */
1466 do_write = (rq_data_dir(rq) == WRITE);
1468 size = blk_rq_bytes(rq);
1469 ofs = blk_rq_pos(rq) * SECTOR_SIZE;
1471 if (do_write && rbd_dev->read_only) {
1472 __blk_end_request_all(rq, -EROFS);
1476 spin_unlock_irq(q->queue_lock);
1478 down_read(&rbd_dev->header_rwsem);
1480 if (rbd_dev->snap_id != CEPH_NOSNAP && !rbd_dev->snap_exists) {
1481 up_read(&rbd_dev->header_rwsem);
1482 dout("request for non-existent snapshot");
1483 spin_lock_irq(q->queue_lock);
1484 __blk_end_request_all(rq, -ENXIO);
1488 snapc = ceph_get_snap_context(rbd_dev->header.snapc);
1490 up_read(&rbd_dev->header_rwsem);
1492 dout("%s 0x%x bytes at 0x%llx\n",
1493 do_write ? "write" : "read",
1494 size, (unsigned long long) 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);
1501 ceph_put_snap_context(snapc);
1506 /* a bio clone to be passed down to OSD req */
1507 dout("rq->bio->bi_vcnt=%hu\n", rq->bio->bi_vcnt);
1508 op_size = rbd_get_segment(&rbd_dev->header,
1509 rbd_dev->header.object_prefix,
1512 kref_get(&coll->kref);
1513 bio = bio_chain_clone(&rq_bio, &next_bio, &bp,
1514 op_size, GFP_ATOMIC);
1516 rbd_coll_end_req_index(rq, coll, cur_seg,
1522 /* init OSD command: write or read */
1524 rbd_req_write(rq, rbd_dev,
1530 rbd_req_read(rq, rbd_dev,
1543 kref_put(&coll->kref, rbd_coll_release);
1546 bio_pair_release(bp);
1547 spin_lock_irq(q->queue_lock);
1549 ceph_put_snap_context(snapc);
1554 * a queue callback. Makes sure that we don't create a bio that spans across
1555 * multiple osd objects. One exception would be with a single page bios,
1556 * which we handle later at bio_chain_clone
1558 static int rbd_merge_bvec(struct request_queue *q, struct bvec_merge_data *bmd,
1559 struct bio_vec *bvec)
1561 struct rbd_device *rbd_dev = q->queuedata;
1562 unsigned int chunk_sectors;
1564 unsigned int bio_sectors;
1567 chunk_sectors = 1 << (rbd_dev->header.obj_order - SECTOR_SHIFT);
1568 sector = bmd->bi_sector + get_start_sect(bmd->bi_bdev);
1569 bio_sectors = bmd->bi_size >> SECTOR_SHIFT;
1571 max = (chunk_sectors - ((sector & (chunk_sectors - 1))
1572 + bio_sectors)) << SECTOR_SHIFT;
1574 max = 0; /* bio_add cannot handle a negative return */
1575 if (max <= bvec->bv_len && bio_sectors == 0)
1576 return bvec->bv_len;
1580 static void rbd_free_disk(struct rbd_device *rbd_dev)
1582 struct gendisk *disk = rbd_dev->disk;
1587 rbd_header_free(&rbd_dev->header);
1589 if (disk->flags & GENHD_FL_UP)
1592 blk_cleanup_queue(disk->queue);
1597 * reload the ondisk the header
1599 static int rbd_read_header(struct rbd_device *rbd_dev,
1600 struct rbd_image_header *header)
1603 struct rbd_image_header_ondisk *dh;
1609 * First reads the fixed-size header to determine the number
1610 * of snapshots, then re-reads it, along with all snapshot
1611 * records as well as their stored names.
1615 dh = kmalloc(len, GFP_KERNEL);
1619 rc = rbd_req_sync_read(rbd_dev,
1621 rbd_dev->header_name,
1627 rc = rbd_header_from_disk(header, dh, snap_count);
1630 pr_warning("unrecognized header format"
1632 rbd_dev->image_name);
1636 if (snap_count == header->total_snaps)
1639 snap_count = header->total_snaps;
1640 len = sizeof (*dh) +
1641 snap_count * sizeof(struct rbd_image_snap_ondisk) +
1642 header->snap_names_len;
1644 rbd_header_free(header);
1647 header->obj_version = ver;
1657 static int rbd_header_add_snap(struct rbd_device *rbd_dev,
1658 const char *snap_name,
1661 int name_len = strlen(snap_name);
1665 struct ceph_mon_client *monc;
1667 /* we should create a snapshot only if we're pointing at the head */
1668 if (rbd_dev->snap_id != CEPH_NOSNAP)
1671 monc = &rbd_dev->rbd_client->client->monc;
1672 ret = ceph_monc_create_snapid(monc, rbd_dev->pool_id, &new_snapid);
1673 dout("created snapid=%llu\n", (unsigned long long) new_snapid);
1677 data = kmalloc(name_len + 16, gfp_flags);
1682 e = data + name_len + 16;
1684 ceph_encode_string_safe(&p, e, snap_name, name_len, bad);
1685 ceph_encode_64_safe(&p, e, new_snapid, bad);
1687 ret = rbd_req_sync_exec(rbd_dev, rbd_dev->header_name,
1689 data, p - data, NULL);
1693 return ret < 0 ? ret : 0;
1698 static void __rbd_remove_all_snaps(struct rbd_device *rbd_dev)
1700 struct rbd_snap *snap;
1701 struct rbd_snap *next;
1703 list_for_each_entry_safe(snap, next, &rbd_dev->snaps, node)
1704 __rbd_remove_snap_dev(snap);
1708 * only read the first part of the ondisk header, without the snaps info
1710 static int __rbd_refresh_header(struct rbd_device *rbd_dev)
1713 struct rbd_image_header h;
1715 ret = rbd_read_header(rbd_dev, &h);
1719 down_write(&rbd_dev->header_rwsem);
1722 if (rbd_dev->snap_id == CEPH_NOSNAP) {
1723 sector_t size = (sector_t) h.image_size / SECTOR_SIZE;
1725 dout("setting size to %llu sectors", (unsigned long long) size);
1726 set_capacity(rbd_dev->disk, size);
1729 /* rbd_dev->header.object_prefix shouldn't change */
1730 kfree(rbd_dev->header.snap_sizes);
1731 kfree(rbd_dev->header.snap_names);
1732 /* osd requests may still refer to snapc */
1733 ceph_put_snap_context(rbd_dev->header.snapc);
1735 rbd_dev->header.obj_version = h.obj_version;
1736 rbd_dev->header.image_size = h.image_size;
1737 rbd_dev->header.total_snaps = h.total_snaps;
1738 rbd_dev->header.snapc = h.snapc;
1739 rbd_dev->header.snap_names = h.snap_names;
1740 rbd_dev->header.snap_names_len = h.snap_names_len;
1741 rbd_dev->header.snap_sizes = h.snap_sizes;
1742 /* Free the extra copy of the object prefix */
1743 WARN_ON(strcmp(rbd_dev->header.object_prefix, h.object_prefix));
1744 kfree(h.object_prefix);
1746 ret = __rbd_init_snaps_header(rbd_dev);
1748 up_write(&rbd_dev->header_rwsem);
1753 static int rbd_init_disk(struct rbd_device *rbd_dev)
1755 struct gendisk *disk;
1756 struct request_queue *q;
1761 /* contact OSD, request size info about the object being mapped */
1762 rc = rbd_read_header(rbd_dev, &rbd_dev->header);
1766 /* no need to lock here, as rbd_dev is not registered yet */
1767 rc = __rbd_init_snaps_header(rbd_dev);
1771 rc = rbd_header_set_snap(rbd_dev, &total_size);
1775 /* create gendisk info */
1777 disk = alloc_disk(RBD_MINORS_PER_MAJOR);
1781 snprintf(disk->disk_name, sizeof(disk->disk_name), RBD_DRV_NAME "%d",
1783 disk->major = rbd_dev->major;
1784 disk->first_minor = 0;
1785 disk->fops = &rbd_bd_ops;
1786 disk->private_data = rbd_dev;
1790 q = blk_init_queue(rbd_rq_fn, &rbd_dev->lock);
1794 /* We use the default size, but let's be explicit about it. */
1795 blk_queue_physical_block_size(q, SECTOR_SIZE);
1797 /* set io sizes to object size */
1798 segment_size = rbd_obj_bytes(&rbd_dev->header);
1799 blk_queue_max_hw_sectors(q, segment_size / SECTOR_SIZE);
1800 blk_queue_max_segment_size(q, segment_size);
1801 blk_queue_io_min(q, segment_size);
1802 blk_queue_io_opt(q, segment_size);
1804 blk_queue_merge_bvec(q, rbd_merge_bvec);
1807 q->queuedata = rbd_dev;
1809 rbd_dev->disk = disk;
1812 /* finally, announce the disk to the world */
1813 set_capacity(disk, total_size / SECTOR_SIZE);
1816 pr_info("%s: added with size 0x%llx\n",
1817 disk->disk_name, (unsigned long long)total_size);
1830 static struct rbd_device *dev_to_rbd_dev(struct device *dev)
1832 return container_of(dev, struct rbd_device, dev);
1835 static ssize_t rbd_size_show(struct device *dev,
1836 struct device_attribute *attr, char *buf)
1838 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
1841 down_read(&rbd_dev->header_rwsem);
1842 size = get_capacity(rbd_dev->disk);
1843 up_read(&rbd_dev->header_rwsem);
1845 return sprintf(buf, "%llu\n", (unsigned long long) size * SECTOR_SIZE);
1848 static ssize_t rbd_major_show(struct device *dev,
1849 struct device_attribute *attr, char *buf)
1851 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
1853 return sprintf(buf, "%d\n", rbd_dev->major);
1856 static ssize_t rbd_client_id_show(struct device *dev,
1857 struct device_attribute *attr, char *buf)
1859 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
1861 return sprintf(buf, "client%lld\n",
1862 ceph_client_id(rbd_dev->rbd_client->client));
1865 static ssize_t rbd_pool_show(struct device *dev,
1866 struct device_attribute *attr, char *buf)
1868 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
1870 return sprintf(buf, "%s\n", rbd_dev->pool_name);
1873 static ssize_t rbd_pool_id_show(struct device *dev,
1874 struct device_attribute *attr, char *buf)
1876 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
1878 return sprintf(buf, "%d\n", rbd_dev->pool_id);
1881 static ssize_t rbd_name_show(struct device *dev,
1882 struct device_attribute *attr, char *buf)
1884 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
1886 return sprintf(buf, "%s\n", rbd_dev->image_name);
1889 static ssize_t rbd_snap_show(struct device *dev,
1890 struct device_attribute *attr,
1893 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
1895 return sprintf(buf, "%s\n", rbd_dev->snap_name);
1898 static ssize_t rbd_image_refresh(struct device *dev,
1899 struct device_attribute *attr,
1903 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
1907 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
1909 rc = __rbd_refresh_header(rbd_dev);
1913 mutex_unlock(&ctl_mutex);
1917 static DEVICE_ATTR(size, S_IRUGO, rbd_size_show, NULL);
1918 static DEVICE_ATTR(major, S_IRUGO, rbd_major_show, NULL);
1919 static DEVICE_ATTR(client_id, S_IRUGO, rbd_client_id_show, NULL);
1920 static DEVICE_ATTR(pool, S_IRUGO, rbd_pool_show, NULL);
1921 static DEVICE_ATTR(pool_id, S_IRUGO, rbd_pool_id_show, NULL);
1922 static DEVICE_ATTR(name, S_IRUGO, rbd_name_show, NULL);
1923 static DEVICE_ATTR(refresh, S_IWUSR, NULL, rbd_image_refresh);
1924 static DEVICE_ATTR(current_snap, S_IRUGO, rbd_snap_show, NULL);
1925 static DEVICE_ATTR(create_snap, S_IWUSR, NULL, rbd_snap_add);
1927 static struct attribute *rbd_attrs[] = {
1928 &dev_attr_size.attr,
1929 &dev_attr_major.attr,
1930 &dev_attr_client_id.attr,
1931 &dev_attr_pool.attr,
1932 &dev_attr_pool_id.attr,
1933 &dev_attr_name.attr,
1934 &dev_attr_current_snap.attr,
1935 &dev_attr_refresh.attr,
1936 &dev_attr_create_snap.attr,
1940 static struct attribute_group rbd_attr_group = {
1944 static const struct attribute_group *rbd_attr_groups[] = {
1949 static void rbd_sysfs_dev_release(struct device *dev)
1953 static struct device_type rbd_device_type = {
1955 .groups = rbd_attr_groups,
1956 .release = rbd_sysfs_dev_release,
1964 static ssize_t rbd_snap_size_show(struct device *dev,
1965 struct device_attribute *attr,
1968 struct rbd_snap *snap = container_of(dev, struct rbd_snap, dev);
1970 return sprintf(buf, "%llu\n", (unsigned long long)snap->size);
1973 static ssize_t rbd_snap_id_show(struct device *dev,
1974 struct device_attribute *attr,
1977 struct rbd_snap *snap = container_of(dev, struct rbd_snap, dev);
1979 return sprintf(buf, "%llu\n", (unsigned long long)snap->id);
1982 static DEVICE_ATTR(snap_size, S_IRUGO, rbd_snap_size_show, NULL);
1983 static DEVICE_ATTR(snap_id, S_IRUGO, rbd_snap_id_show, NULL);
1985 static struct attribute *rbd_snap_attrs[] = {
1986 &dev_attr_snap_size.attr,
1987 &dev_attr_snap_id.attr,
1991 static struct attribute_group rbd_snap_attr_group = {
1992 .attrs = rbd_snap_attrs,
1995 static void rbd_snap_dev_release(struct device *dev)
1997 struct rbd_snap *snap = container_of(dev, struct rbd_snap, dev);
2002 static const struct attribute_group *rbd_snap_attr_groups[] = {
2003 &rbd_snap_attr_group,
2007 static struct device_type rbd_snap_device_type = {
2008 .groups = rbd_snap_attr_groups,
2009 .release = rbd_snap_dev_release,
2012 static void __rbd_remove_snap_dev(struct rbd_snap *snap)
2014 list_del(&snap->node);
2015 device_unregister(&snap->dev);
2018 static int rbd_register_snap_dev(struct rbd_snap *snap,
2019 struct device *parent)
2021 struct device *dev = &snap->dev;
2024 dev->type = &rbd_snap_device_type;
2025 dev->parent = parent;
2026 dev->release = rbd_snap_dev_release;
2027 dev_set_name(dev, "snap_%s", snap->name);
2028 ret = device_register(dev);
2033 static struct rbd_snap *__rbd_add_snap_dev(struct rbd_device *rbd_dev,
2034 int i, const char *name)
2036 struct rbd_snap *snap;
2039 snap = kzalloc(sizeof (*snap), GFP_KERNEL);
2041 return ERR_PTR(-ENOMEM);
2044 snap->name = kstrdup(name, GFP_KERNEL);
2048 snap->size = rbd_dev->header.snap_sizes[i];
2049 snap->id = rbd_dev->header.snapc->snaps[i];
2050 if (device_is_registered(&rbd_dev->dev)) {
2051 ret = rbd_register_snap_dev(snap, &rbd_dev->dev);
2062 return ERR_PTR(ret);
2066 * search for the previous snap in a null delimited string list
2068 const char *rbd_prev_snap_name(const char *name, const char *start)
2070 if (name < start + 2)
2083 * compare the old list of snapshots that we have to what's in the header
2084 * and update it accordingly. Note that the header holds the snapshots
2085 * in a reverse order (from newest to oldest) and we need to go from
2086 * older to new so that we don't get a duplicate snap name when
2087 * doing the process (e.g., removed snapshot and recreated a new
2088 * one with the same name.
2090 static int __rbd_init_snaps_header(struct rbd_device *rbd_dev)
2092 const char *name, *first_name;
2093 int i = rbd_dev->header.total_snaps;
2094 struct rbd_snap *snap, *old_snap = NULL;
2095 struct list_head *p, *n;
2097 first_name = rbd_dev->header.snap_names;
2098 name = first_name + rbd_dev->header.snap_names_len;
2100 list_for_each_prev_safe(p, n, &rbd_dev->snaps) {
2103 old_snap = list_entry(p, struct rbd_snap, node);
2106 cur_id = rbd_dev->header.snapc->snaps[i - 1];
2108 if (!i || old_snap->id < cur_id) {
2110 * old_snap->id was skipped, thus was
2111 * removed. If this rbd_dev is mapped to
2112 * the removed snapshot, record that it no
2113 * longer exists, to prevent further I/O.
2115 if (rbd_dev->snap_id == old_snap->id)
2116 rbd_dev->snap_exists = false;
2117 __rbd_remove_snap_dev(old_snap);
2120 if (old_snap->id == cur_id) {
2121 /* we have this snapshot already */
2123 name = rbd_prev_snap_name(name, first_name);
2127 i--, name = rbd_prev_snap_name(name, first_name)) {
2132 cur_id = rbd_dev->header.snapc->snaps[i];
2133 /* snapshot removal? handle it above */
2134 if (cur_id >= old_snap->id)
2136 /* a new snapshot */
2137 snap = __rbd_add_snap_dev(rbd_dev, i - 1, name);
2139 return PTR_ERR(snap);
2141 /* note that we add it backward so using n and not p */
2142 list_add(&snap->node, n);
2146 /* we're done going over the old snap list, just add what's left */
2147 for (; i > 0; i--) {
2148 name = rbd_prev_snap_name(name, first_name);
2153 snap = __rbd_add_snap_dev(rbd_dev, i - 1, name);
2155 return PTR_ERR(snap);
2156 list_add(&snap->node, &rbd_dev->snaps);
2162 static int rbd_bus_add_dev(struct rbd_device *rbd_dev)
2166 struct rbd_snap *snap;
2168 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
2169 dev = &rbd_dev->dev;
2171 dev->bus = &rbd_bus_type;
2172 dev->type = &rbd_device_type;
2173 dev->parent = &rbd_root_dev;
2174 dev->release = rbd_dev_release;
2175 dev_set_name(dev, "%d", rbd_dev->dev_id);
2176 ret = device_register(dev);
2180 list_for_each_entry(snap, &rbd_dev->snaps, node) {
2181 ret = rbd_register_snap_dev(snap, &rbd_dev->dev);
2186 mutex_unlock(&ctl_mutex);
2190 static void rbd_bus_del_dev(struct rbd_device *rbd_dev)
2192 device_unregister(&rbd_dev->dev);
2195 static int rbd_init_watch_dev(struct rbd_device *rbd_dev)
2200 ret = rbd_req_sync_watch(rbd_dev);
2201 if (ret == -ERANGE) {
2202 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
2203 rc = __rbd_refresh_header(rbd_dev);
2204 mutex_unlock(&ctl_mutex);
2208 } while (ret == -ERANGE);
2213 static atomic64_t rbd_id_max = ATOMIC64_INIT(0);
2216 * Get a unique rbd identifier for the given new rbd_dev, and add
2217 * the rbd_dev to the global list. The minimum rbd id is 1.
2219 static void rbd_id_get(struct rbd_device *rbd_dev)
2221 rbd_dev->dev_id = atomic64_inc_return(&rbd_id_max);
2223 spin_lock(&rbd_dev_list_lock);
2224 list_add_tail(&rbd_dev->node, &rbd_dev_list);
2225 spin_unlock(&rbd_dev_list_lock);
2229 * Remove an rbd_dev from the global list, and record that its
2230 * identifier is no longer in use.
2232 static void rbd_id_put(struct rbd_device *rbd_dev)
2234 struct list_head *tmp;
2235 int rbd_id = rbd_dev->dev_id;
2240 spin_lock(&rbd_dev_list_lock);
2241 list_del_init(&rbd_dev->node);
2244 * If the id being "put" is not the current maximum, there
2245 * is nothing special we need to do.
2247 if (rbd_id != atomic64_read(&rbd_id_max)) {
2248 spin_unlock(&rbd_dev_list_lock);
2253 * We need to update the current maximum id. Search the
2254 * list to find out what it is. We're more likely to find
2255 * the maximum at the end, so search the list backward.
2258 list_for_each_prev(tmp, &rbd_dev_list) {
2259 struct rbd_device *rbd_dev;
2261 rbd_dev = list_entry(tmp, struct rbd_device, node);
2262 if (rbd_id > max_id)
2265 spin_unlock(&rbd_dev_list_lock);
2268 * The max id could have been updated by rbd_id_get(), in
2269 * which case it now accurately reflects the new maximum.
2270 * Be careful not to overwrite the maximum value in that
2273 atomic64_cmpxchg(&rbd_id_max, rbd_id, max_id);
2277 * Skips over white space at *buf, and updates *buf to point to the
2278 * first found non-space character (if any). Returns the length of
2279 * the token (string of non-white space characters) found. Note
2280 * that *buf must be terminated with '\0'.
2282 static inline size_t next_token(const char **buf)
2285 * These are the characters that produce nonzero for
2286 * isspace() in the "C" and "POSIX" locales.
2288 const char *spaces = " \f\n\r\t\v";
2290 *buf += strspn(*buf, spaces); /* Find start of token */
2292 return strcspn(*buf, spaces); /* Return token length */
2296 * Finds the next token in *buf, and if the provided token buffer is
2297 * big enough, copies the found token into it. The result, if
2298 * copied, is guaranteed to be terminated with '\0'. Note that *buf
2299 * must be terminated with '\0' on entry.
2301 * Returns the length of the token found (not including the '\0').
2302 * Return value will be 0 if no token is found, and it will be >=
2303 * token_size if the token would not fit.
2305 * The *buf pointer will be updated to point beyond the end of the
2306 * found token. Note that this occurs even if the token buffer is
2307 * too small to hold it.
2309 static inline size_t copy_token(const char **buf,
2315 len = next_token(buf);
2316 if (len < token_size) {
2317 memcpy(token, *buf, len);
2318 *(token + len) = '\0';
2326 * Finds the next token in *buf, dynamically allocates a buffer big
2327 * enough to hold a copy of it, and copies the token into the new
2328 * buffer. The copy is guaranteed to be terminated with '\0'. Note
2329 * that a duplicate buffer is created even for a zero-length token.
2331 * Returns a pointer to the newly-allocated duplicate, or a null
2332 * pointer if memory for the duplicate was not available. If
2333 * the lenp argument is a non-null pointer, the length of the token
2334 * (not including the '\0') is returned in *lenp.
2336 * If successful, the *buf pointer will be updated to point beyond
2337 * the end of the found token.
2339 * Note: uses GFP_KERNEL for allocation.
2341 static inline char *dup_token(const char **buf, size_t *lenp)
2346 len = next_token(buf);
2347 dup = kmalloc(len + 1, GFP_KERNEL);
2351 memcpy(dup, *buf, len);
2352 *(dup + len) = '\0';
2362 * This fills in the pool_name, image_name, image_name_len, snap_name,
2363 * rbd_dev, rbd_md_name, and name fields of the given rbd_dev, based
2364 * on the list of monitor addresses and other options provided via
2367 * Note: rbd_dev is assumed to have been initially zero-filled.
2369 static int rbd_add_parse_args(struct rbd_device *rbd_dev,
2371 const char **mon_addrs,
2372 size_t *mon_addrs_size,
2374 size_t options_size)
2379 /* The first four tokens are required */
2381 len = next_token(&buf);
2384 *mon_addrs_size = len + 1;
2389 len = copy_token(&buf, options, options_size);
2390 if (!len || len >= options_size)
2394 rbd_dev->pool_name = dup_token(&buf, NULL);
2395 if (!rbd_dev->pool_name)
2398 rbd_dev->image_name = dup_token(&buf, &rbd_dev->image_name_len);
2399 if (!rbd_dev->image_name)
2402 /* Create the name of the header object */
2404 rbd_dev->header_name = kmalloc(rbd_dev->image_name_len
2405 + sizeof (RBD_SUFFIX),
2407 if (!rbd_dev->header_name)
2409 sprintf(rbd_dev->header_name, "%s%s", rbd_dev->image_name, RBD_SUFFIX);
2412 * The snapshot name is optional. If none is is supplied,
2413 * we use the default value.
2415 rbd_dev->snap_name = dup_token(&buf, &len);
2416 if (!rbd_dev->snap_name)
2419 /* Replace the empty name with the default */
2420 kfree(rbd_dev->snap_name);
2422 = kmalloc(sizeof (RBD_SNAP_HEAD_NAME), GFP_KERNEL);
2423 if (!rbd_dev->snap_name)
2426 memcpy(rbd_dev->snap_name, RBD_SNAP_HEAD_NAME,
2427 sizeof (RBD_SNAP_HEAD_NAME));
2433 kfree(rbd_dev->header_name);
2434 kfree(rbd_dev->image_name);
2435 kfree(rbd_dev->pool_name);
2436 rbd_dev->pool_name = NULL;
2441 static ssize_t rbd_add(struct bus_type *bus,
2446 struct rbd_device *rbd_dev = NULL;
2447 const char *mon_addrs = NULL;
2448 size_t mon_addrs_size = 0;
2449 struct ceph_osd_client *osdc;
2452 if (!try_module_get(THIS_MODULE))
2455 options = kmalloc(count, GFP_KERNEL);
2458 rbd_dev = kzalloc(sizeof(*rbd_dev), GFP_KERNEL);
2462 /* static rbd_device initialization */
2463 spin_lock_init(&rbd_dev->lock);
2464 INIT_LIST_HEAD(&rbd_dev->node);
2465 INIT_LIST_HEAD(&rbd_dev->snaps);
2466 init_rwsem(&rbd_dev->header_rwsem);
2468 /* generate unique id: find highest unique id, add one */
2469 rbd_id_get(rbd_dev);
2471 /* Fill in the device name, now that we have its id. */
2472 BUILD_BUG_ON(DEV_NAME_LEN
2473 < sizeof (RBD_DRV_NAME) + MAX_INT_FORMAT_WIDTH);
2474 sprintf(rbd_dev->name, "%s%d", RBD_DRV_NAME, rbd_dev->dev_id);
2476 /* parse add command */
2477 rc = rbd_add_parse_args(rbd_dev, buf, &mon_addrs, &mon_addrs_size,
2482 rbd_dev->rbd_client = rbd_get_client(mon_addrs, mon_addrs_size - 1,
2484 if (IS_ERR(rbd_dev->rbd_client)) {
2485 rc = PTR_ERR(rbd_dev->rbd_client);
2490 osdc = &rbd_dev->rbd_client->client->osdc;
2491 rc = ceph_pg_poolid_by_name(osdc->osdmap, rbd_dev->pool_name);
2493 goto err_out_client;
2494 rbd_dev->pool_id = rc;
2496 /* register our block device */
2497 rc = register_blkdev(0, rbd_dev->name);
2499 goto err_out_client;
2500 rbd_dev->major = rc;
2502 rc = rbd_bus_add_dev(rbd_dev);
2504 goto err_out_blkdev;
2507 * At this point cleanup in the event of an error is the job
2508 * of the sysfs code (initiated by rbd_bus_del_dev()).
2510 * Set up and announce blkdev mapping.
2512 rc = rbd_init_disk(rbd_dev);
2516 rc = rbd_init_watch_dev(rbd_dev);
2523 /* this will also clean up rest of rbd_dev stuff */
2525 rbd_bus_del_dev(rbd_dev);
2530 unregister_blkdev(rbd_dev->major, rbd_dev->name);
2532 rbd_put_client(rbd_dev);
2534 if (rbd_dev->pool_name) {
2535 kfree(rbd_dev->snap_name);
2536 kfree(rbd_dev->header_name);
2537 kfree(rbd_dev->image_name);
2538 kfree(rbd_dev->pool_name);
2540 rbd_id_put(rbd_dev);
2545 dout("Error adding device %s\n", buf);
2546 module_put(THIS_MODULE);
2548 return (ssize_t) rc;
2551 static struct rbd_device *__rbd_get_dev(unsigned long dev_id)
2553 struct list_head *tmp;
2554 struct rbd_device *rbd_dev;
2556 spin_lock(&rbd_dev_list_lock);
2557 list_for_each(tmp, &rbd_dev_list) {
2558 rbd_dev = list_entry(tmp, struct rbd_device, node);
2559 if (rbd_dev->dev_id == dev_id) {
2560 spin_unlock(&rbd_dev_list_lock);
2564 spin_unlock(&rbd_dev_list_lock);
2568 static void rbd_dev_release(struct device *dev)
2570 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
2572 if (rbd_dev->watch_request) {
2573 struct ceph_client *client = rbd_dev->rbd_client->client;
2575 ceph_osdc_unregister_linger_request(&client->osdc,
2576 rbd_dev->watch_request);
2578 if (rbd_dev->watch_event)
2579 rbd_req_sync_unwatch(rbd_dev);
2581 rbd_put_client(rbd_dev);
2583 /* clean up and free blkdev */
2584 rbd_free_disk(rbd_dev);
2585 unregister_blkdev(rbd_dev->major, rbd_dev->name);
2587 /* done with the id, and with the rbd_dev */
2588 kfree(rbd_dev->snap_name);
2589 kfree(rbd_dev->header_name);
2590 kfree(rbd_dev->pool_name);
2591 kfree(rbd_dev->image_name);
2592 rbd_id_put(rbd_dev);
2595 /* release module ref */
2596 module_put(THIS_MODULE);
2599 static ssize_t rbd_remove(struct bus_type *bus,
2603 struct rbd_device *rbd_dev = NULL;
2608 rc = strict_strtoul(buf, 10, &ul);
2612 /* convert to int; abort if we lost anything in the conversion */
2613 target_id = (int) ul;
2614 if (target_id != ul)
2617 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
2619 rbd_dev = __rbd_get_dev(target_id);
2625 __rbd_remove_all_snaps(rbd_dev);
2626 rbd_bus_del_dev(rbd_dev);
2629 mutex_unlock(&ctl_mutex);
2633 static ssize_t rbd_snap_add(struct device *dev,
2634 struct device_attribute *attr,
2638 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
2640 char *name = kmalloc(count + 1, GFP_KERNEL);
2644 snprintf(name, count, "%s", buf);
2646 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
2648 ret = rbd_header_add_snap(rbd_dev,
2653 ret = __rbd_refresh_header(rbd_dev);
2657 /* shouldn't hold ctl_mutex when notifying.. notify might
2658 trigger a watch callback that would need to get that mutex */
2659 mutex_unlock(&ctl_mutex);
2661 /* make a best effort, don't error if failed */
2662 rbd_req_sync_notify(rbd_dev);
2669 mutex_unlock(&ctl_mutex);
2675 * create control files in sysfs
2678 static int rbd_sysfs_init(void)
2682 ret = device_register(&rbd_root_dev);
2686 ret = bus_register(&rbd_bus_type);
2688 device_unregister(&rbd_root_dev);
2693 static void rbd_sysfs_cleanup(void)
2695 bus_unregister(&rbd_bus_type);
2696 device_unregister(&rbd_root_dev);
2699 int __init rbd_init(void)
2703 rc = rbd_sysfs_init();
2706 pr_info("loaded " RBD_DRV_NAME_LONG "\n");
2710 void __exit rbd_exit(void)
2712 rbd_sysfs_cleanup();
2715 module_init(rbd_init);
2716 module_exit(rbd_exit);
2718 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
2719 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
2720 MODULE_DESCRIPTION("rados block device");
2722 /* following authorship retained from original osdblk.c */
2723 MODULE_AUTHOR("Jeff Garzik <jeff@garzik.org>");
2725 MODULE_LICENSE("GPL");