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)
593 down_write(&rbd_dev->header_rwsem);
595 if (!memcmp(rbd_dev->snap_name, RBD_SNAP_HEAD_NAME,
596 sizeof (RBD_SNAP_HEAD_NAME))) {
597 rbd_dev->snap_id = CEPH_NOSNAP;
598 rbd_dev->snap_exists = false;
599 rbd_dev->read_only = 0;
601 *size = rbd_dev->header.image_size;
605 ret = snap_by_name(&rbd_dev->header, rbd_dev->snap_name,
609 rbd_dev->snap_id = snap_id;
610 rbd_dev->snap_exists = true;
611 rbd_dev->read_only = 1;
616 up_write(&rbd_dev->header_rwsem);
620 static void rbd_header_free(struct rbd_image_header *header)
622 kfree(header->object_prefix);
623 kfree(header->snap_sizes);
624 kfree(header->snap_names);
625 ceph_put_snap_context(header->snapc);
629 * get the actual striped segment name, offset and length
631 static u64 rbd_get_segment(struct rbd_image_header *header,
632 const char *object_prefix,
634 char *seg_name, u64 *segofs)
636 u64 seg = ofs >> header->obj_order;
639 snprintf(seg_name, RBD_MAX_SEG_NAME_LEN,
640 "%s.%012llx", object_prefix, seg);
642 ofs = ofs & ((1 << header->obj_order) - 1);
643 len = min_t(u64, len, (1 << header->obj_order) - ofs);
651 static int rbd_get_num_segments(struct rbd_image_header *header,
654 u64 start_seg = ofs >> header->obj_order;
655 u64 end_seg = (ofs + len - 1) >> header->obj_order;
656 return end_seg - start_seg + 1;
660 * returns the size of an object in the image
662 static u64 rbd_obj_bytes(struct rbd_image_header *header)
664 return 1 << header->obj_order;
671 static void bio_chain_put(struct bio *chain)
677 chain = chain->bi_next;
683 * zeros a bio chain, starting at specific offset
685 static void zero_bio_chain(struct bio *chain, int start_ofs)
694 bio_for_each_segment(bv, chain, i) {
695 if (pos + bv->bv_len > start_ofs) {
696 int remainder = max(start_ofs - pos, 0);
697 buf = bvec_kmap_irq(bv, &flags);
698 memset(buf + remainder, 0,
699 bv->bv_len - remainder);
700 bvec_kunmap_irq(buf, &flags);
705 chain = chain->bi_next;
710 * bio_chain_clone - clone a chain of bios up to a certain length.
711 * might return a bio_pair that will need to be released.
713 static struct bio *bio_chain_clone(struct bio **old, struct bio **next,
714 struct bio_pair **bp,
715 int len, gfp_t gfpmask)
717 struct bio *tmp, *old_chain = *old, *new_chain = NULL, *tail = NULL;
721 bio_pair_release(*bp);
725 while (old_chain && (total < len)) {
726 tmp = bio_kmalloc(gfpmask, old_chain->bi_max_vecs);
730 if (total + old_chain->bi_size > len) {
734 * this split can only happen with a single paged bio,
735 * split_bio will BUG_ON if this is not the case
737 dout("bio_chain_clone split! total=%d remaining=%d"
739 (int)total, (int)len-total,
740 (int)old_chain->bi_size);
742 /* split the bio. We'll release it either in the next
743 call, or it will have to be released outside */
744 bp = bio_split(old_chain, (len - total) / SECTOR_SIZE);
748 __bio_clone(tmp, &bp->bio1);
752 __bio_clone(tmp, old_chain);
753 *next = old_chain->bi_next;
757 gfpmask &= ~__GFP_WAIT;
761 new_chain = tail = tmp;
766 old_chain = old_chain->bi_next;
768 total += tmp->bi_size;
774 tail->bi_next = NULL;
781 dout("bio_chain_clone with err\n");
782 bio_chain_put(new_chain);
787 * helpers for osd request op vectors.
789 static int rbd_create_rw_ops(struct ceph_osd_req_op **ops,
794 *ops = kzalloc(sizeof(struct ceph_osd_req_op) * (num_ops + 1),
798 (*ops)[0].op = opcode;
800 * op extent offset and length will be set later on
801 * in calc_raw_layout()
803 (*ops)[0].payload_len = payload_len;
807 static void rbd_destroy_ops(struct ceph_osd_req_op *ops)
812 static void rbd_coll_end_req_index(struct request *rq,
813 struct rbd_req_coll *coll,
817 struct request_queue *q;
820 dout("rbd_coll_end_req_index %p index %d ret %d len %lld\n",
821 coll, index, ret, len);
827 blk_end_request(rq, ret, len);
833 spin_lock_irq(q->queue_lock);
834 coll->status[index].done = 1;
835 coll->status[index].rc = ret;
836 coll->status[index].bytes = len;
837 max = min = coll->num_done;
838 while (max < coll->total && coll->status[max].done)
841 for (i = min; i<max; i++) {
842 __blk_end_request(rq, coll->status[i].rc,
843 coll->status[i].bytes);
845 kref_put(&coll->kref, rbd_coll_release);
847 spin_unlock_irq(q->queue_lock);
850 static void rbd_coll_end_req(struct rbd_request *req,
853 rbd_coll_end_req_index(req->rq, req->coll, req->coll_index, ret, len);
857 * Send ceph osd request
859 static int rbd_do_request(struct request *rq,
860 struct rbd_device *rbd_dev,
861 struct ceph_snap_context *snapc,
863 const char *object_name, u64 ofs, u64 len,
868 struct ceph_osd_req_op *ops,
869 struct rbd_req_coll *coll,
871 void (*rbd_cb)(struct ceph_osd_request *req,
872 struct ceph_msg *msg),
873 struct ceph_osd_request **linger_req,
876 struct ceph_osd_request *req;
877 struct ceph_file_layout *layout;
880 struct timespec mtime = CURRENT_TIME;
881 struct rbd_request *req_data;
882 struct ceph_osd_request_head *reqhead;
883 struct ceph_osd_client *osdc;
885 req_data = kzalloc(sizeof(*req_data), GFP_NOIO);
888 rbd_coll_end_req_index(rq, coll, coll_index,
894 req_data->coll = coll;
895 req_data->coll_index = coll_index;
898 dout("rbd_do_request object_name=%s ofs=%lld len=%lld\n",
899 object_name, len, ofs);
901 osdc = &rbd_dev->rbd_client->client->osdc;
902 req = ceph_osdc_alloc_request(osdc, flags, snapc, ops,
903 false, GFP_NOIO, pages, bio);
909 req->r_callback = rbd_cb;
913 req_data->pages = pages;
916 req->r_priv = req_data;
918 reqhead = req->r_request->front.iov_base;
919 reqhead->snapid = cpu_to_le64(CEPH_NOSNAP);
921 strncpy(req->r_oid, object_name, sizeof(req->r_oid));
922 req->r_oid_len = strlen(req->r_oid);
924 layout = &req->r_file_layout;
925 memset(layout, 0, sizeof(*layout));
926 layout->fl_stripe_unit = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER);
927 layout->fl_stripe_count = cpu_to_le32(1);
928 layout->fl_object_size = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER);
929 layout->fl_pg_pool = cpu_to_le32(rbd_dev->pool_id);
930 ceph_calc_raw_layout(osdc, layout, snapid, ofs, &len, &bno,
933 ceph_osdc_build_request(req, ofs, &len,
937 req->r_oid, req->r_oid_len);
940 ceph_osdc_set_request_linger(osdc, req);
944 ret = ceph_osdc_start_request(osdc, req, false);
949 ret = ceph_osdc_wait_request(osdc, req);
951 *ver = le64_to_cpu(req->r_reassert_version.version);
952 dout("reassert_ver=%lld\n",
953 le64_to_cpu(req->r_reassert_version.version));
954 ceph_osdc_put_request(req);
959 bio_chain_put(req_data->bio);
960 ceph_osdc_put_request(req);
962 rbd_coll_end_req(req_data, ret, len);
968 * Ceph osd op callback
970 static void rbd_req_cb(struct ceph_osd_request *req, struct ceph_msg *msg)
972 struct rbd_request *req_data = req->r_priv;
973 struct ceph_osd_reply_head *replyhead;
974 struct ceph_osd_op *op;
980 replyhead = msg->front.iov_base;
981 WARN_ON(le32_to_cpu(replyhead->num_ops) == 0);
982 op = (void *)(replyhead + 1);
983 rc = le32_to_cpu(replyhead->result);
984 bytes = le64_to_cpu(op->extent.length);
985 read_op = (le16_to_cpu(op->op) == CEPH_OSD_OP_READ);
987 dout("rbd_req_cb bytes=%lld readop=%d rc=%d\n", bytes, read_op, rc);
989 if (rc == -ENOENT && read_op) {
990 zero_bio_chain(req_data->bio, 0);
992 } else if (rc == 0 && read_op && bytes < req_data->len) {
993 zero_bio_chain(req_data->bio, bytes);
994 bytes = req_data->len;
997 rbd_coll_end_req(req_data, rc, bytes);
1000 bio_chain_put(req_data->bio);
1002 ceph_osdc_put_request(req);
1006 static void rbd_simple_req_cb(struct ceph_osd_request *req, struct ceph_msg *msg)
1008 ceph_osdc_put_request(req);
1012 * Do a synchronous ceph osd operation
1014 static int rbd_req_sync_op(struct rbd_device *rbd_dev,
1015 struct ceph_snap_context *snapc,
1019 struct ceph_osd_req_op *orig_ops,
1020 const char *object_name,
1023 struct ceph_osd_request **linger_req,
1027 struct page **pages;
1029 struct ceph_osd_req_op *ops = orig_ops;
1032 num_pages = calc_pages_for(ofs , len);
1033 pages = ceph_alloc_page_vector(num_pages, GFP_KERNEL);
1035 return PTR_ERR(pages);
1038 payload_len = (flags & CEPH_OSD_FLAG_WRITE ? len : 0);
1039 ret = rbd_create_rw_ops(&ops, 1, opcode, payload_len);
1043 if ((flags & CEPH_OSD_FLAG_WRITE) && buf) {
1044 ret = ceph_copy_to_page_vector(pages, buf, ofs, len);
1050 ret = rbd_do_request(NULL, rbd_dev, snapc, snapid,
1051 object_name, ofs, len, NULL,
1061 if ((flags & CEPH_OSD_FLAG_READ) && buf)
1062 ret = ceph_copy_from_page_vector(pages, buf, ofs, ret);
1066 rbd_destroy_ops(ops);
1068 ceph_release_page_vector(pages, num_pages);
1073 * Do an asynchronous ceph osd operation
1075 static int rbd_do_op(struct request *rq,
1076 struct rbd_device *rbd_dev,
1077 struct ceph_snap_context *snapc,
1079 int opcode, int flags,
1082 struct rbd_req_coll *coll,
1089 struct ceph_osd_req_op *ops;
1092 seg_name = kmalloc(RBD_MAX_SEG_NAME_LEN + 1, GFP_NOIO);
1096 seg_len = rbd_get_segment(&rbd_dev->header,
1097 rbd_dev->header.object_prefix,
1099 seg_name, &seg_ofs);
1101 payload_len = (flags & CEPH_OSD_FLAG_WRITE ? seg_len : 0);
1103 ret = rbd_create_rw_ops(&ops, 1, opcode, payload_len);
1107 /* we've taken care of segment sizes earlier when we
1108 cloned the bios. We should never have a segment
1109 truncated at this point */
1110 BUG_ON(seg_len < len);
1112 ret = rbd_do_request(rq, rbd_dev, snapc, snapid,
1113 seg_name, seg_ofs, seg_len,
1119 rbd_req_cb, 0, NULL);
1121 rbd_destroy_ops(ops);
1128 * Request async osd write
1130 static int rbd_req_write(struct request *rq,
1131 struct rbd_device *rbd_dev,
1132 struct ceph_snap_context *snapc,
1135 struct rbd_req_coll *coll,
1138 return rbd_do_op(rq, rbd_dev, snapc, CEPH_NOSNAP,
1140 CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK,
1141 ofs, len, bio, coll, coll_index);
1145 * Request async osd read
1147 static int rbd_req_read(struct request *rq,
1148 struct rbd_device *rbd_dev,
1152 struct rbd_req_coll *coll,
1155 return rbd_do_op(rq, rbd_dev, NULL,
1159 ofs, len, bio, coll, coll_index);
1163 * Request sync osd read
1165 static int rbd_req_sync_read(struct rbd_device *rbd_dev,
1166 struct ceph_snap_context *snapc,
1168 const char *object_name,
1173 return rbd_req_sync_op(rbd_dev, NULL,
1178 object_name, ofs, len, buf, NULL, ver);
1182 * Request sync osd watch
1184 static int rbd_req_sync_notify_ack(struct rbd_device *rbd_dev,
1187 const char *object_name)
1189 struct ceph_osd_req_op *ops;
1192 ret = rbd_create_rw_ops(&ops, 1, CEPH_OSD_OP_NOTIFY_ACK, 0);
1196 ops[0].watch.ver = cpu_to_le64(ver);
1197 ops[0].watch.cookie = notify_id;
1198 ops[0].watch.flag = 0;
1200 ret = rbd_do_request(NULL, rbd_dev, NULL, CEPH_NOSNAP,
1201 object_name, 0, 0, NULL,
1206 rbd_simple_req_cb, 0, NULL);
1208 rbd_destroy_ops(ops);
1212 static void rbd_watch_cb(u64 ver, u64 notify_id, u8 opcode, void *data)
1214 struct rbd_device *rbd_dev = (struct rbd_device *)data;
1221 dout("rbd_watch_cb %s notify_id=%lld opcode=%d\n",
1222 rbd_dev->header_name, notify_id, (int) opcode);
1223 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
1224 rc = __rbd_refresh_header(rbd_dev);
1225 hver = rbd_dev->header.obj_version;
1226 mutex_unlock(&ctl_mutex);
1228 pr_warning(RBD_DRV_NAME "%d got notification but failed to "
1229 " update snaps: %d\n", rbd_dev->major, rc);
1231 rbd_req_sync_notify_ack(rbd_dev, hver, notify_id, rbd_dev->header_name);
1235 * Request sync osd watch
1237 static int rbd_req_sync_watch(struct rbd_device *rbd_dev,
1238 const char *object_name,
1241 struct ceph_osd_req_op *ops;
1242 struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
1244 int ret = rbd_create_rw_ops(&ops, 1, CEPH_OSD_OP_WATCH, 0);
1248 ret = ceph_osdc_create_event(osdc, rbd_watch_cb, 0,
1249 (void *)rbd_dev, &rbd_dev->watch_event);
1253 ops[0].watch.ver = cpu_to_le64(ver);
1254 ops[0].watch.cookie = cpu_to_le64(rbd_dev->watch_event->cookie);
1255 ops[0].watch.flag = 1;
1257 ret = rbd_req_sync_op(rbd_dev, NULL,
1260 CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK,
1262 object_name, 0, 0, NULL,
1263 &rbd_dev->watch_request, NULL);
1268 rbd_destroy_ops(ops);
1272 ceph_osdc_cancel_event(rbd_dev->watch_event);
1273 rbd_dev->watch_event = NULL;
1275 rbd_destroy_ops(ops);
1280 * Request sync osd unwatch
1282 static int rbd_req_sync_unwatch(struct rbd_device *rbd_dev,
1283 const char *object_name)
1285 struct ceph_osd_req_op *ops;
1287 int ret = rbd_create_rw_ops(&ops, 1, CEPH_OSD_OP_WATCH, 0);
1291 ops[0].watch.ver = 0;
1292 ops[0].watch.cookie = cpu_to_le64(rbd_dev->watch_event->cookie);
1293 ops[0].watch.flag = 0;
1295 ret = rbd_req_sync_op(rbd_dev, NULL,
1298 CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK,
1300 object_name, 0, 0, NULL, NULL, NULL);
1302 rbd_destroy_ops(ops);
1303 ceph_osdc_cancel_event(rbd_dev->watch_event);
1304 rbd_dev->watch_event = NULL;
1308 struct rbd_notify_info {
1309 struct rbd_device *rbd_dev;
1312 static void rbd_notify_cb(u64 ver, u64 notify_id, u8 opcode, void *data)
1314 struct rbd_device *rbd_dev = (struct rbd_device *)data;
1318 dout("rbd_notify_cb %s notify_id=%lld opcode=%d\n",
1319 rbd_dev->header_name,
1320 notify_id, (int)opcode);
1324 * Request sync osd notify
1326 static int rbd_req_sync_notify(struct rbd_device *rbd_dev,
1327 const char *object_name)
1329 struct ceph_osd_req_op *ops;
1330 struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
1331 struct ceph_osd_event *event;
1332 struct rbd_notify_info info;
1333 int payload_len = sizeof(u32) + sizeof(u32);
1336 ret = rbd_create_rw_ops(&ops, 1, CEPH_OSD_OP_NOTIFY, payload_len);
1340 info.rbd_dev = rbd_dev;
1342 ret = ceph_osdc_create_event(osdc, rbd_notify_cb, 1,
1343 (void *)&info, &event);
1347 ops[0].watch.ver = 1;
1348 ops[0].watch.flag = 1;
1349 ops[0].watch.cookie = event->cookie;
1350 ops[0].watch.prot_ver = RADOS_NOTIFY_VER;
1351 ops[0].watch.timeout = 12;
1353 ret = rbd_req_sync_op(rbd_dev, NULL,
1356 CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK,
1358 object_name, 0, 0, NULL, NULL, NULL);
1362 ret = ceph_osdc_wait_event(event, CEPH_OSD_TIMEOUT_DEFAULT);
1363 dout("ceph_osdc_wait_event returned %d\n", ret);
1364 rbd_destroy_ops(ops);
1368 ceph_osdc_cancel_event(event);
1370 rbd_destroy_ops(ops);
1375 * Request sync osd read
1377 static int rbd_req_sync_exec(struct rbd_device *rbd_dev,
1378 const char *object_name,
1379 const char *class_name,
1380 const char *method_name,
1385 struct ceph_osd_req_op *ops;
1386 int class_name_len = strlen(class_name);
1387 int method_name_len = strlen(method_name);
1388 int ret = rbd_create_rw_ops(&ops, 1, CEPH_OSD_OP_CALL,
1389 class_name_len + method_name_len + len);
1393 ops[0].cls.class_name = class_name;
1394 ops[0].cls.class_len = (__u8) class_name_len;
1395 ops[0].cls.method_name = method_name;
1396 ops[0].cls.method_len = (__u8) method_name_len;
1397 ops[0].cls.argc = 0;
1398 ops[0].cls.indata = data;
1399 ops[0].cls.indata_len = len;
1401 ret = rbd_req_sync_op(rbd_dev, NULL,
1404 CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK,
1406 object_name, 0, 0, NULL, NULL, ver);
1408 rbd_destroy_ops(ops);
1410 dout("cls_exec returned %d\n", ret);
1414 static struct rbd_req_coll *rbd_alloc_coll(int num_reqs)
1416 struct rbd_req_coll *coll =
1417 kzalloc(sizeof(struct rbd_req_coll) +
1418 sizeof(struct rbd_req_status) * num_reqs,
1423 coll->total = num_reqs;
1424 kref_init(&coll->kref);
1429 * block device queue callback
1431 static void rbd_rq_fn(struct request_queue *q)
1433 struct rbd_device *rbd_dev = q->queuedata;
1435 struct bio_pair *bp = NULL;
1437 while ((rq = blk_fetch_request(q))) {
1439 struct bio *rq_bio, *next_bio = NULL;
1441 int size, op_size = 0;
1443 int num_segs, cur_seg = 0;
1444 struct rbd_req_coll *coll;
1445 struct ceph_snap_context *snapc;
1447 /* peek at request from block layer */
1451 dout("fetched request\n");
1453 /* filter out block requests we don't understand */
1454 if ((rq->cmd_type != REQ_TYPE_FS)) {
1455 __blk_end_request_all(rq, 0);
1459 /* deduce our operation (read, write) */
1460 do_write = (rq_data_dir(rq) == WRITE);
1462 size = blk_rq_bytes(rq);
1463 ofs = blk_rq_pos(rq) * SECTOR_SIZE;
1465 if (do_write && rbd_dev->read_only) {
1466 __blk_end_request_all(rq, -EROFS);
1470 spin_unlock_irq(q->queue_lock);
1472 down_read(&rbd_dev->header_rwsem);
1474 if (rbd_dev->snap_id != CEPH_NOSNAP && !rbd_dev->snap_exists) {
1475 up_read(&rbd_dev->header_rwsem);
1476 dout("request for non-existent snapshot");
1477 spin_lock_irq(q->queue_lock);
1478 __blk_end_request_all(rq, -ENXIO);
1482 snapc = ceph_get_snap_context(rbd_dev->header.snapc);
1484 up_read(&rbd_dev->header_rwsem);
1486 dout("%s 0x%x bytes at 0x%llx\n",
1487 do_write ? "write" : "read",
1488 size, blk_rq_pos(rq) * SECTOR_SIZE);
1490 num_segs = rbd_get_num_segments(&rbd_dev->header, ofs, size);
1491 coll = rbd_alloc_coll(num_segs);
1493 spin_lock_irq(q->queue_lock);
1494 __blk_end_request_all(rq, -ENOMEM);
1495 ceph_put_snap_context(snapc);
1500 /* a bio clone to be passed down to OSD req */
1501 dout("rq->bio->bi_vcnt=%d\n", rq->bio->bi_vcnt);
1502 op_size = rbd_get_segment(&rbd_dev->header,
1503 rbd_dev->header.object_prefix,
1506 kref_get(&coll->kref);
1507 bio = bio_chain_clone(&rq_bio, &next_bio, &bp,
1508 op_size, GFP_ATOMIC);
1510 rbd_coll_end_req_index(rq, coll, cur_seg,
1516 /* init OSD command: write or read */
1518 rbd_req_write(rq, rbd_dev,
1524 rbd_req_read(rq, rbd_dev,
1537 kref_put(&coll->kref, rbd_coll_release);
1540 bio_pair_release(bp);
1541 spin_lock_irq(q->queue_lock);
1543 ceph_put_snap_context(snapc);
1548 * a queue callback. Makes sure that we don't create a bio that spans across
1549 * multiple osd objects. One exception would be with a single page bios,
1550 * which we handle later at bio_chain_clone
1552 static int rbd_merge_bvec(struct request_queue *q, struct bvec_merge_data *bmd,
1553 struct bio_vec *bvec)
1555 struct rbd_device *rbd_dev = q->queuedata;
1556 unsigned int chunk_sectors;
1558 unsigned int bio_sectors;
1561 chunk_sectors = 1 << (rbd_dev->header.obj_order - SECTOR_SHIFT);
1562 sector = bmd->bi_sector + get_start_sect(bmd->bi_bdev);
1563 bio_sectors = bmd->bi_size >> SECTOR_SHIFT;
1565 max = (chunk_sectors - ((sector & (chunk_sectors - 1))
1566 + bio_sectors)) << SECTOR_SHIFT;
1568 max = 0; /* bio_add cannot handle a negative return */
1569 if (max <= bvec->bv_len && bio_sectors == 0)
1570 return bvec->bv_len;
1574 static void rbd_free_disk(struct rbd_device *rbd_dev)
1576 struct gendisk *disk = rbd_dev->disk;
1581 rbd_header_free(&rbd_dev->header);
1583 if (disk->flags & GENHD_FL_UP)
1586 blk_cleanup_queue(disk->queue);
1591 * reload the ondisk the header
1593 static int rbd_read_header(struct rbd_device *rbd_dev,
1594 struct rbd_image_header *header)
1597 struct rbd_image_header_ondisk *dh;
1603 * First reads the fixed-size header to determine the number
1604 * of snapshots, then re-reads it, along with all snapshot
1605 * records as well as their stored names.
1609 dh = kmalloc(len, GFP_KERNEL);
1613 rc = rbd_req_sync_read(rbd_dev,
1615 rbd_dev->header_name,
1621 rc = rbd_header_from_disk(header, dh, snap_count, GFP_KERNEL);
1624 pr_warning("unrecognized header format"
1626 rbd_dev->image_name);
1630 if (snap_count == header->total_snaps)
1633 snap_count = header->total_snaps;
1634 len = sizeof (*dh) +
1635 snap_count * sizeof(struct rbd_image_snap_ondisk) +
1636 header->snap_names_len;
1638 rbd_header_free(header);
1641 header->obj_version = ver;
1651 static int rbd_header_add_snap(struct rbd_device *rbd_dev,
1652 const char *snap_name,
1655 int name_len = strlen(snap_name);
1660 struct ceph_mon_client *monc;
1662 /* we should create a snapshot only if we're pointing at the head */
1663 if (rbd_dev->snap_id != CEPH_NOSNAP)
1666 monc = &rbd_dev->rbd_client->client->monc;
1667 ret = ceph_monc_create_snapid(monc, rbd_dev->pool_id, &new_snapid);
1668 dout("created snapid=%lld\n", new_snapid);
1672 data = kmalloc(name_len + 16, gfp_flags);
1677 e = data + name_len + 16;
1679 ceph_encode_string_safe(&p, e, snap_name, name_len, bad);
1680 ceph_encode_64_safe(&p, e, new_snapid, bad);
1682 ret = rbd_req_sync_exec(rbd_dev, rbd_dev->header_name,
1684 data, p - data, &ver);
1691 down_write(&rbd_dev->header_rwsem);
1692 rbd_dev->header.snapc->seq = new_snapid;
1693 up_write(&rbd_dev->header_rwsem);
1700 static void __rbd_remove_all_snaps(struct rbd_device *rbd_dev)
1702 struct rbd_snap *snap;
1704 while (!list_empty(&rbd_dev->snaps)) {
1705 snap = list_first_entry(&rbd_dev->snaps, struct rbd_snap, node);
1706 __rbd_remove_snap_dev(rbd_dev, snap);
1711 * only read the first part of the ondisk header, without the snaps info
1713 static int __rbd_refresh_header(struct rbd_device *rbd_dev)
1716 struct rbd_image_header h;
1718 ret = rbd_read_header(rbd_dev, &h);
1722 down_write(&rbd_dev->header_rwsem);
1725 if (rbd_dev->snap_id == CEPH_NOSNAP) {
1726 sector_t size = (sector_t) h.image_size / SECTOR_SIZE;
1728 dout("setting size to %llu sectors", (unsigned long long) size);
1729 set_capacity(rbd_dev->disk, size);
1732 /* rbd_dev->header.object_prefix shouldn't change */
1733 kfree(rbd_dev->header.snap_sizes);
1734 kfree(rbd_dev->header.snap_names);
1735 /* osd requests may still refer to snapc */
1736 ceph_put_snap_context(rbd_dev->header.snapc);
1738 rbd_dev->header.obj_version = h.obj_version;
1739 rbd_dev->header.image_size = h.image_size;
1740 rbd_dev->header.total_snaps = h.total_snaps;
1741 rbd_dev->header.snapc = h.snapc;
1742 rbd_dev->header.snap_names = h.snap_names;
1743 rbd_dev->header.snap_names_len = h.snap_names_len;
1744 rbd_dev->header.snap_sizes = h.snap_sizes;
1745 /* Free the extra copy of the object prefix */
1746 WARN_ON(strcmp(rbd_dev->header.object_prefix, h.object_prefix));
1747 kfree(h.object_prefix);
1749 ret = __rbd_init_snaps_header(rbd_dev);
1751 up_write(&rbd_dev->header_rwsem);
1756 static int rbd_init_disk(struct rbd_device *rbd_dev)
1758 struct gendisk *disk;
1759 struct request_queue *q;
1764 /* contact OSD, request size info about the object being mapped */
1765 rc = rbd_read_header(rbd_dev, &rbd_dev->header);
1769 /* no need to lock here, as rbd_dev is not registered yet */
1770 rc = __rbd_init_snaps_header(rbd_dev);
1774 rc = rbd_header_set_snap(rbd_dev, &total_size);
1778 /* create gendisk info */
1780 disk = alloc_disk(RBD_MINORS_PER_MAJOR);
1784 snprintf(disk->disk_name, sizeof(disk->disk_name), RBD_DRV_NAME "%d",
1786 disk->major = rbd_dev->major;
1787 disk->first_minor = 0;
1788 disk->fops = &rbd_bd_ops;
1789 disk->private_data = rbd_dev;
1793 q = blk_init_queue(rbd_rq_fn, &rbd_dev->lock);
1797 /* We use the default size, but let's be explicit about it. */
1798 blk_queue_physical_block_size(q, SECTOR_SIZE);
1800 /* set io sizes to object size */
1801 segment_size = rbd_obj_bytes(&rbd_dev->header);
1802 blk_queue_max_hw_sectors(q, segment_size / SECTOR_SIZE);
1803 blk_queue_max_segment_size(q, segment_size);
1804 blk_queue_io_min(q, segment_size);
1805 blk_queue_io_opt(q, segment_size);
1807 blk_queue_merge_bvec(q, rbd_merge_bvec);
1810 q->queuedata = rbd_dev;
1812 rbd_dev->disk = disk;
1815 /* finally, announce the disk to the world */
1816 set_capacity(disk, total_size / SECTOR_SIZE);
1819 pr_info("%s: added with size 0x%llx\n",
1820 disk->disk_name, (unsigned long long)total_size);
1833 static struct rbd_device *dev_to_rbd_dev(struct device *dev)
1835 return container_of(dev, struct rbd_device, dev);
1838 static ssize_t rbd_size_show(struct device *dev,
1839 struct device_attribute *attr, char *buf)
1841 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
1844 down_read(&rbd_dev->header_rwsem);
1845 size = get_capacity(rbd_dev->disk);
1846 up_read(&rbd_dev->header_rwsem);
1848 return sprintf(buf, "%llu\n", (unsigned long long) size * SECTOR_SIZE);
1851 static ssize_t rbd_major_show(struct device *dev,
1852 struct device_attribute *attr, char *buf)
1854 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
1856 return sprintf(buf, "%d\n", rbd_dev->major);
1859 static ssize_t rbd_client_id_show(struct device *dev,
1860 struct device_attribute *attr, char *buf)
1862 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
1864 return sprintf(buf, "client%lld\n",
1865 ceph_client_id(rbd_dev->rbd_client->client));
1868 static ssize_t rbd_pool_show(struct device *dev,
1869 struct device_attribute *attr, char *buf)
1871 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
1873 return sprintf(buf, "%s\n", rbd_dev->pool_name);
1876 static ssize_t rbd_pool_id_show(struct device *dev,
1877 struct device_attribute *attr, char *buf)
1879 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
1881 return sprintf(buf, "%d\n", rbd_dev->pool_id);
1884 static ssize_t rbd_name_show(struct device *dev,
1885 struct device_attribute *attr, char *buf)
1887 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
1889 return sprintf(buf, "%s\n", rbd_dev->image_name);
1892 static ssize_t rbd_snap_show(struct device *dev,
1893 struct device_attribute *attr,
1896 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
1898 return sprintf(buf, "%s\n", rbd_dev->snap_name);
1901 static ssize_t rbd_image_refresh(struct device *dev,
1902 struct device_attribute *attr,
1906 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
1910 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
1912 rc = __rbd_refresh_header(rbd_dev);
1916 mutex_unlock(&ctl_mutex);
1920 static DEVICE_ATTR(size, S_IRUGO, rbd_size_show, NULL);
1921 static DEVICE_ATTR(major, S_IRUGO, rbd_major_show, NULL);
1922 static DEVICE_ATTR(client_id, S_IRUGO, rbd_client_id_show, NULL);
1923 static DEVICE_ATTR(pool, S_IRUGO, rbd_pool_show, NULL);
1924 static DEVICE_ATTR(pool_id, S_IRUGO, rbd_pool_id_show, NULL);
1925 static DEVICE_ATTR(name, S_IRUGO, rbd_name_show, NULL);
1926 static DEVICE_ATTR(refresh, S_IWUSR, NULL, rbd_image_refresh);
1927 static DEVICE_ATTR(current_snap, S_IRUGO, rbd_snap_show, NULL);
1928 static DEVICE_ATTR(create_snap, S_IWUSR, NULL, rbd_snap_add);
1930 static struct attribute *rbd_attrs[] = {
1931 &dev_attr_size.attr,
1932 &dev_attr_major.attr,
1933 &dev_attr_client_id.attr,
1934 &dev_attr_pool.attr,
1935 &dev_attr_pool_id.attr,
1936 &dev_attr_name.attr,
1937 &dev_attr_current_snap.attr,
1938 &dev_attr_refresh.attr,
1939 &dev_attr_create_snap.attr,
1943 static struct attribute_group rbd_attr_group = {
1947 static const struct attribute_group *rbd_attr_groups[] = {
1952 static void rbd_sysfs_dev_release(struct device *dev)
1956 static struct device_type rbd_device_type = {
1958 .groups = rbd_attr_groups,
1959 .release = rbd_sysfs_dev_release,
1967 static ssize_t rbd_snap_size_show(struct device *dev,
1968 struct device_attribute *attr,
1971 struct rbd_snap *snap = container_of(dev, struct rbd_snap, dev);
1973 return sprintf(buf, "%llu\n", (unsigned long long)snap->size);
1976 static ssize_t rbd_snap_id_show(struct device *dev,
1977 struct device_attribute *attr,
1980 struct rbd_snap *snap = container_of(dev, struct rbd_snap, dev);
1982 return sprintf(buf, "%llu\n", (unsigned long long)snap->id);
1985 static DEVICE_ATTR(snap_size, S_IRUGO, rbd_snap_size_show, NULL);
1986 static DEVICE_ATTR(snap_id, S_IRUGO, rbd_snap_id_show, NULL);
1988 static struct attribute *rbd_snap_attrs[] = {
1989 &dev_attr_snap_size.attr,
1990 &dev_attr_snap_id.attr,
1994 static struct attribute_group rbd_snap_attr_group = {
1995 .attrs = rbd_snap_attrs,
1998 static void rbd_snap_dev_release(struct device *dev)
2000 struct rbd_snap *snap = container_of(dev, struct rbd_snap, dev);
2005 static const struct attribute_group *rbd_snap_attr_groups[] = {
2006 &rbd_snap_attr_group,
2010 static struct device_type rbd_snap_device_type = {
2011 .groups = rbd_snap_attr_groups,
2012 .release = rbd_snap_dev_release,
2015 static void __rbd_remove_snap_dev(struct rbd_device *rbd_dev,
2016 struct rbd_snap *snap)
2018 list_del(&snap->node);
2019 device_unregister(&snap->dev);
2022 static int rbd_register_snap_dev(struct rbd_device *rbd_dev,
2023 struct rbd_snap *snap,
2024 struct device *parent)
2026 struct device *dev = &snap->dev;
2029 dev->type = &rbd_snap_device_type;
2030 dev->parent = parent;
2031 dev->release = rbd_snap_dev_release;
2032 dev_set_name(dev, "snap_%s", snap->name);
2033 ret = device_register(dev);
2038 static int __rbd_add_snap_dev(struct rbd_device *rbd_dev,
2039 int i, const char *name,
2040 struct rbd_snap **snapp)
2043 struct rbd_snap *snap = kzalloc(sizeof(*snap), GFP_KERNEL);
2046 snap->name = kstrdup(name, GFP_KERNEL);
2047 snap->size = rbd_dev->header.snap_sizes[i];
2048 snap->id = rbd_dev->header.snapc->snaps[i];
2049 if (device_is_registered(&rbd_dev->dev)) {
2050 ret = rbd_register_snap_dev(rbd_dev, snap,
2064 * search for the previous snap in a null delimited string list
2066 const char *rbd_prev_snap_name(const char *name, const char *start)
2068 if (name < start + 2)
2081 * compare the old list of snapshots that we have to what's in the header
2082 * and update it accordingly. Note that the header holds the snapshots
2083 * in a reverse order (from newest to oldest) and we need to go from
2084 * older to new so that we don't get a duplicate snap name when
2085 * doing the process (e.g., removed snapshot and recreated a new
2086 * one with the same name.
2088 static int __rbd_init_snaps_header(struct rbd_device *rbd_dev)
2090 const char *name, *first_name;
2091 int i = rbd_dev->header.total_snaps;
2092 struct rbd_snap *snap, *old_snap = NULL;
2094 struct list_head *p, *n;
2096 first_name = rbd_dev->header.snap_names;
2097 name = first_name + rbd_dev->header.snap_names_len;
2099 list_for_each_prev_safe(p, n, &rbd_dev->snaps) {
2102 old_snap = list_entry(p, struct rbd_snap, node);
2105 cur_id = rbd_dev->header.snapc->snaps[i - 1];
2107 if (!i || old_snap->id < cur_id) {
2109 * old_snap->id was skipped, thus was
2110 * removed. If this rbd_dev is mapped to
2111 * the removed snapshot, record that it no
2112 * longer exists, to prevent further I/O.
2114 if (rbd_dev->snap_id == old_snap->id)
2115 rbd_dev->snap_exists = false;
2116 __rbd_remove_snap_dev(rbd_dev, old_snap);
2119 if (old_snap->id == cur_id) {
2120 /* we have this snapshot already */
2122 name = rbd_prev_snap_name(name, first_name);
2126 i--, name = rbd_prev_snap_name(name, first_name)) {
2131 cur_id = rbd_dev->header.snapc->snaps[i];
2132 /* snapshot removal? handle it above */
2133 if (cur_id >= old_snap->id)
2135 /* a new snapshot */
2136 ret = __rbd_add_snap_dev(rbd_dev, i - 1, name, &snap);
2140 /* note that we add it backward so using n and not p */
2141 list_add(&snap->node, n);
2145 /* we're done going over the old snap list, just add what's left */
2146 for (; i > 0; i--) {
2147 name = rbd_prev_snap_name(name, first_name);
2152 ret = __rbd_add_snap_dev(rbd_dev, i - 1, name, &snap);
2155 list_add(&snap->node, &rbd_dev->snaps);
2161 static int rbd_bus_add_dev(struct rbd_device *rbd_dev)
2165 struct rbd_snap *snap;
2167 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
2168 dev = &rbd_dev->dev;
2170 dev->bus = &rbd_bus_type;
2171 dev->type = &rbd_device_type;
2172 dev->parent = &rbd_root_dev;
2173 dev->release = rbd_dev_release;
2174 dev_set_name(dev, "%d", rbd_dev->id);
2175 ret = device_register(dev);
2179 list_for_each_entry(snap, &rbd_dev->snaps, node) {
2180 ret = rbd_register_snap_dev(rbd_dev, snap,
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, rbd_dev->header_name,
2201 rbd_dev->header.obj_version);
2202 if (ret == -ERANGE) {
2203 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
2204 rc = __rbd_refresh_header(rbd_dev);
2205 mutex_unlock(&ctl_mutex);
2209 } while (ret == -ERANGE);
2214 static atomic64_t rbd_id_max = ATOMIC64_INIT(0);
2217 * Get a unique rbd identifier for the given new rbd_dev, and add
2218 * the rbd_dev to the global list. The minimum rbd id is 1.
2220 static void rbd_id_get(struct rbd_device *rbd_dev)
2222 rbd_dev->id = atomic64_inc_return(&rbd_id_max);
2224 spin_lock(&rbd_dev_list_lock);
2225 list_add_tail(&rbd_dev->node, &rbd_dev_list);
2226 spin_unlock(&rbd_dev_list_lock);
2230 * Remove an rbd_dev from the global list, and record that its
2231 * identifier is no longer in use.
2233 static void rbd_id_put(struct rbd_device *rbd_dev)
2235 struct list_head *tmp;
2236 int rbd_id = rbd_dev->id;
2241 spin_lock(&rbd_dev_list_lock);
2242 list_del_init(&rbd_dev->node);
2245 * If the id being "put" is not the current maximum, there
2246 * is nothing special we need to do.
2248 if (rbd_id != atomic64_read(&rbd_id_max)) {
2249 spin_unlock(&rbd_dev_list_lock);
2254 * We need to update the current maximum id. Search the
2255 * list to find out what it is. We're more likely to find
2256 * the maximum at the end, so search the list backward.
2259 list_for_each_prev(tmp, &rbd_dev_list) {
2260 struct rbd_device *rbd_dev;
2262 rbd_dev = list_entry(tmp, struct rbd_device, node);
2263 if (rbd_id > max_id)
2266 spin_unlock(&rbd_dev_list_lock);
2269 * The max id could have been updated by rbd_id_get(), in
2270 * which case it now accurately reflects the new maximum.
2271 * Be careful not to overwrite the maximum value in that
2274 atomic64_cmpxchg(&rbd_id_max, rbd_id, max_id);
2278 * Skips over white space at *buf, and updates *buf to point to the
2279 * first found non-space character (if any). Returns the length of
2280 * the token (string of non-white space characters) found. Note
2281 * that *buf must be terminated with '\0'.
2283 static inline size_t next_token(const char **buf)
2286 * These are the characters that produce nonzero for
2287 * isspace() in the "C" and "POSIX" locales.
2289 const char *spaces = " \f\n\r\t\v";
2291 *buf += strspn(*buf, spaces); /* Find start of token */
2293 return strcspn(*buf, spaces); /* Return token length */
2297 * Finds the next token in *buf, and if the provided token buffer is
2298 * big enough, copies the found token into it. The result, if
2299 * copied, is guaranteed to be terminated with '\0'. Note that *buf
2300 * must be terminated with '\0' on entry.
2302 * Returns the length of the token found (not including the '\0').
2303 * Return value will be 0 if no token is found, and it will be >=
2304 * token_size if the token would not fit.
2306 * The *buf pointer will be updated to point beyond the end of the
2307 * found token. Note that this occurs even if the token buffer is
2308 * too small to hold it.
2310 static inline size_t copy_token(const char **buf,
2316 len = next_token(buf);
2317 if (len < token_size) {
2318 memcpy(token, *buf, len);
2319 *(token + len) = '\0';
2327 * Finds the next token in *buf, dynamically allocates a buffer big
2328 * enough to hold a copy of it, and copies the token into the new
2329 * buffer. The copy is guaranteed to be terminated with '\0'. Note
2330 * that a duplicate buffer is created even for a zero-length token.
2332 * Returns a pointer to the newly-allocated duplicate, or a null
2333 * pointer if memory for the duplicate was not available. If
2334 * the lenp argument is a non-null pointer, the length of the token
2335 * (not including the '\0') is returned in *lenp.
2337 * If successful, the *buf pointer will be updated to point beyond
2338 * the end of the found token.
2340 * Note: uses GFP_KERNEL for allocation.
2342 static inline char *dup_token(const char **buf, size_t *lenp)
2347 len = next_token(buf);
2348 dup = kmalloc(len + 1, GFP_KERNEL);
2352 memcpy(dup, *buf, len);
2353 *(dup + len) = '\0';
2363 * This fills in the pool_name, image_name, image_name_len, snap_name,
2364 * rbd_dev, rbd_md_name, and name fields of the given rbd_dev, based
2365 * on the list of monitor addresses and other options provided via
2368 * Note: rbd_dev is assumed to have been initially zero-filled.
2370 static int rbd_add_parse_args(struct rbd_device *rbd_dev,
2372 const char **mon_addrs,
2373 size_t *mon_addrs_size,
2375 size_t options_size)
2380 /* The first four tokens are required */
2382 len = next_token(&buf);
2385 *mon_addrs_size = len + 1;
2390 len = copy_token(&buf, options, options_size);
2391 if (!len || len >= options_size)
2395 rbd_dev->pool_name = dup_token(&buf, NULL);
2396 if (!rbd_dev->pool_name)
2399 rbd_dev->image_name = dup_token(&buf, &rbd_dev->image_name_len);
2400 if (!rbd_dev->image_name)
2403 /* Create the name of the header object */
2405 rbd_dev->header_name = kmalloc(rbd_dev->image_name_len
2406 + sizeof (RBD_SUFFIX),
2408 if (!rbd_dev->header_name)
2410 sprintf(rbd_dev->header_name, "%s%s", rbd_dev->image_name, RBD_SUFFIX);
2413 * The snapshot name is optional. If none is is supplied,
2414 * we use the default value.
2416 rbd_dev->snap_name = dup_token(&buf, &len);
2417 if (!rbd_dev->snap_name)
2420 /* Replace the empty name with the default */
2421 kfree(rbd_dev->snap_name);
2423 = kmalloc(sizeof (RBD_SNAP_HEAD_NAME), GFP_KERNEL);
2424 if (!rbd_dev->snap_name)
2427 memcpy(rbd_dev->snap_name, RBD_SNAP_HEAD_NAME,
2428 sizeof (RBD_SNAP_HEAD_NAME));
2434 kfree(rbd_dev->header_name);
2435 kfree(rbd_dev->image_name);
2436 kfree(rbd_dev->pool_name);
2437 rbd_dev->pool_name = NULL;
2442 static ssize_t rbd_add(struct bus_type *bus,
2447 struct rbd_device *rbd_dev = NULL;
2448 const char *mon_addrs = NULL;
2449 size_t mon_addrs_size = 0;
2450 struct ceph_osd_client *osdc;
2453 if (!try_module_get(THIS_MODULE))
2456 options = kmalloc(count, GFP_KERNEL);
2459 rbd_dev = kzalloc(sizeof(*rbd_dev), GFP_KERNEL);
2463 /* static rbd_device initialization */
2464 spin_lock_init(&rbd_dev->lock);
2465 INIT_LIST_HEAD(&rbd_dev->node);
2466 INIT_LIST_HEAD(&rbd_dev->snaps);
2467 init_rwsem(&rbd_dev->header_rwsem);
2469 init_rwsem(&rbd_dev->header_rwsem);
2471 /* generate unique id: find highest unique id, add one */
2472 rbd_id_get(rbd_dev);
2474 /* Fill in the device name, now that we have its id. */
2475 BUILD_BUG_ON(DEV_NAME_LEN
2476 < sizeof (RBD_DRV_NAME) + MAX_INT_FORMAT_WIDTH);
2477 sprintf(rbd_dev->name, "%s%d", RBD_DRV_NAME, rbd_dev->id);
2479 /* parse add command */
2480 rc = rbd_add_parse_args(rbd_dev, buf, &mon_addrs, &mon_addrs_size,
2485 rbd_dev->rbd_client = rbd_get_client(mon_addrs, mon_addrs_size - 1,
2487 if (IS_ERR(rbd_dev->rbd_client)) {
2488 rc = PTR_ERR(rbd_dev->rbd_client);
2493 osdc = &rbd_dev->rbd_client->client->osdc;
2494 rc = ceph_pg_poolid_by_name(osdc->osdmap, rbd_dev->pool_name);
2496 goto err_out_client;
2497 rbd_dev->pool_id = rc;
2499 /* register our block device */
2500 rc = register_blkdev(0, rbd_dev->name);
2502 goto err_out_client;
2503 rbd_dev->major = rc;
2505 rc = rbd_bus_add_dev(rbd_dev);
2507 goto err_out_blkdev;
2510 * At this point cleanup in the event of an error is the job
2511 * of the sysfs code (initiated by rbd_bus_del_dev()).
2513 * Set up and announce blkdev mapping.
2515 rc = rbd_init_disk(rbd_dev);
2519 rc = rbd_init_watch_dev(rbd_dev);
2526 /* this will also clean up rest of rbd_dev stuff */
2528 rbd_bus_del_dev(rbd_dev);
2533 unregister_blkdev(rbd_dev->major, rbd_dev->name);
2535 rbd_put_client(rbd_dev);
2537 if (rbd_dev->pool_name) {
2538 kfree(rbd_dev->snap_name);
2539 kfree(rbd_dev->header_name);
2540 kfree(rbd_dev->image_name);
2541 kfree(rbd_dev->pool_name);
2543 rbd_id_put(rbd_dev);
2548 dout("Error adding device %s\n", buf);
2549 module_put(THIS_MODULE);
2551 return (ssize_t) rc;
2554 static struct rbd_device *__rbd_get_dev(unsigned long id)
2556 struct list_head *tmp;
2557 struct rbd_device *rbd_dev;
2559 spin_lock(&rbd_dev_list_lock);
2560 list_for_each(tmp, &rbd_dev_list) {
2561 rbd_dev = list_entry(tmp, struct rbd_device, node);
2562 if (rbd_dev->id == id) {
2563 spin_unlock(&rbd_dev_list_lock);
2567 spin_unlock(&rbd_dev_list_lock);
2571 static void rbd_dev_release(struct device *dev)
2573 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
2575 if (rbd_dev->watch_request) {
2576 struct ceph_client *client = rbd_dev->rbd_client->client;
2578 ceph_osdc_unregister_linger_request(&client->osdc,
2579 rbd_dev->watch_request);
2581 if (rbd_dev->watch_event)
2582 rbd_req_sync_unwatch(rbd_dev, rbd_dev->header_name);
2584 rbd_put_client(rbd_dev);
2586 /* clean up and free blkdev */
2587 rbd_free_disk(rbd_dev);
2588 unregister_blkdev(rbd_dev->major, rbd_dev->name);
2590 /* done with the id, and with the rbd_dev */
2591 kfree(rbd_dev->snap_name);
2592 kfree(rbd_dev->header_name);
2593 kfree(rbd_dev->pool_name);
2594 kfree(rbd_dev->image_name);
2595 rbd_id_put(rbd_dev);
2598 /* release module ref */
2599 module_put(THIS_MODULE);
2602 static ssize_t rbd_remove(struct bus_type *bus,
2606 struct rbd_device *rbd_dev = NULL;
2611 rc = strict_strtoul(buf, 10, &ul);
2615 /* convert to int; abort if we lost anything in the conversion */
2616 target_id = (int) ul;
2617 if (target_id != ul)
2620 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
2622 rbd_dev = __rbd_get_dev(target_id);
2628 __rbd_remove_all_snaps(rbd_dev);
2629 rbd_bus_del_dev(rbd_dev);
2632 mutex_unlock(&ctl_mutex);
2636 static ssize_t rbd_snap_add(struct device *dev,
2637 struct device_attribute *attr,
2641 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
2643 char *name = kmalloc(count + 1, GFP_KERNEL);
2647 snprintf(name, count, "%s", buf);
2649 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
2651 ret = rbd_header_add_snap(rbd_dev,
2656 ret = __rbd_refresh_header(rbd_dev);
2660 /* shouldn't hold ctl_mutex when notifying.. notify might
2661 trigger a watch callback that would need to get that mutex */
2662 mutex_unlock(&ctl_mutex);
2664 /* make a best effort, don't error if failed */
2665 rbd_req_sync_notify(rbd_dev, rbd_dev->header_name);
2672 mutex_unlock(&ctl_mutex);
2678 * create control files in sysfs
2681 static int rbd_sysfs_init(void)
2685 ret = device_register(&rbd_root_dev);
2689 ret = bus_register(&rbd_bus_type);
2691 device_unregister(&rbd_root_dev);
2696 static void rbd_sysfs_cleanup(void)
2698 bus_unregister(&rbd_bus_type);
2699 device_unregister(&rbd_root_dev);
2702 int __init rbd_init(void)
2706 rc = rbd_sysfs_init();
2709 pr_info("loaded " RBD_DRV_NAME_LONG "\n");
2713 void __exit rbd_exit(void)
2715 rbd_sysfs_cleanup();
2718 module_init(rbd_init);
2719 module_exit(rbd_exit);
2721 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
2722 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
2723 MODULE_DESCRIPTION("rados block device");
2725 /* following authorship retained from original osdblk.c */
2726 MODULE_AUTHOR("Jeff Garzik <jeff@garzik.org>");
2728 MODULE_LICENSE("GPL");