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_MD_NAME_LEN (RBD_MAX_OBJ_NAME_LEN + sizeof(RBD_SUFFIX))
59 #define RBD_MAX_POOL_NAME_LEN 64
60 #define RBD_MAX_SNAP_NAME_LEN 32
61 #define RBD_MAX_OPT_LEN 1024
63 #define RBD_SNAP_HEAD_NAME "-"
66 * An RBD device name will be "rbd#", where the "rbd" comes from
67 * RBD_DRV_NAME above, and # is a unique integer identifier.
68 * MAX_INT_FORMAT_WIDTH is used in ensuring DEV_NAME_LEN is big
69 * enough to hold all possible device names.
71 #define DEV_NAME_LEN 32
72 #define MAX_INT_FORMAT_WIDTH ((5 * sizeof (int)) / 2 + 1)
74 #define RBD_NOTIFY_TIMEOUT_DEFAULT 10
77 * block device image metadata (in-memory version)
79 struct rbd_image_header {
85 struct ceph_snap_context *snapc;
86 size_t snap_names_len;
101 * an instance of the client. multiple devices may share an rbd client.
104 struct ceph_client *client;
105 struct rbd_options *rbd_opts;
107 struct list_head node;
111 * a request completion status
113 struct rbd_req_status {
120 * a collection of requests
122 struct rbd_req_coll {
126 struct rbd_req_status status[0];
130 * a single io request
133 struct request *rq; /* blk layer request */
134 struct bio *bio; /* cloned bio */
135 struct page **pages; /* list of used pages */
138 struct rbd_req_coll *coll;
145 struct list_head node;
153 int id; /* blkdev unique id */
155 int major; /* blkdev assigned major */
156 struct gendisk *disk; /* blkdev's gendisk and rq */
157 struct request_queue *q;
159 struct rbd_client *rbd_client;
161 char name[DEV_NAME_LEN]; /* blkdev name, e.g. rbd3 */
163 spinlock_t lock; /* queue lock */
165 struct rbd_image_header header;
166 char obj[RBD_MAX_OBJ_NAME_LEN]; /* rbd image name */
168 char obj_md_name[RBD_MAX_MD_NAME_LEN]; /* hdr nm. */
169 char pool_name[RBD_MAX_POOL_NAME_LEN];
172 struct ceph_osd_event *watch_event;
173 struct ceph_osd_request *watch_request;
175 /* protects updating the header */
176 struct rw_semaphore header_rwsem;
177 char snap_name[RBD_MAX_SNAP_NAME_LEN];
178 u64 snap_id; /* current snapshot id */
181 struct list_head node;
183 /* list of snapshots */
184 struct list_head snaps;
190 static DEFINE_MUTEX(ctl_mutex); /* Serialize open/close/setup/teardown */
192 static LIST_HEAD(rbd_dev_list); /* devices */
193 static DEFINE_SPINLOCK(rbd_dev_list_lock);
195 static LIST_HEAD(rbd_client_list); /* clients */
196 static DEFINE_SPINLOCK(rbd_client_list_lock);
198 static int __rbd_init_snaps_header(struct rbd_device *rbd_dev);
199 static void rbd_dev_release(struct device *dev);
200 static ssize_t rbd_snap_add(struct device *dev,
201 struct device_attribute *attr,
204 static void __rbd_remove_snap_dev(struct rbd_device *rbd_dev,
205 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 *opt,
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(opt, rbdc, 0, 0);
295 if (IS_ERR(rbdc->client))
297 opt = NULL; /* Now rbdc->client is responsible for opt */
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(opt);
326 * Find a ceph client with specific addr and configuration.
328 static struct rbd_client *__rbd_client_find(struct ceph_options *opt)
330 struct rbd_client *client_node;
332 if (opt->flags & CEPH_OPT_NOSHARE)
335 list_for_each_entry(client_node, &rbd_client_list, node)
336 if (ceph_compare_options(opt, client_node->client) == 0)
349 /* string args above */
352 static match_table_t rbdopt_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 *rbdopt = private;
362 substring_t argstr[MAX_OPT_ARGS];
363 int token, intval, ret;
365 token = match_token(c, rbdopt_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 rbdopt->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 *opt;
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 opt = ceph_parse_options(options, mon_addr,
413 mon_addr + mon_addr_len,
414 parse_rbd_opts_token, rbd_opts);
417 return ERR_CAST(opt);
420 spin_lock(&rbd_client_list_lock);
421 rbdc = __rbd_client_find(opt);
423 /* using an existing client */
424 kref_get(&rbdc->kref);
425 spin_unlock(&rbd_client_list_lock);
427 ceph_destroy_options(opt);
432 spin_unlock(&rbd_client_list_lock);
434 rbdc = rbd_client_create(opt, 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);
484 * Create a new header structure, translate header format from the on-disk
487 static int rbd_header_from_disk(struct rbd_image_header *header,
488 struct rbd_image_header_ondisk *ondisk,
494 if (memcmp(ondisk, RBD_HEADER_TEXT, sizeof(RBD_HEADER_TEXT)))
497 snap_count = le32_to_cpu(ondisk->snap_count);
498 if (snap_count > (UINT_MAX - sizeof(struct ceph_snap_context))
501 header->snapc = kmalloc(sizeof(struct ceph_snap_context) +
502 snap_count * sizeof (*ondisk),
507 header->snap_names_len = le64_to_cpu(ondisk->snap_names_len);
509 header->snap_names = kmalloc(header->snap_names_len,
511 if (!header->snap_names)
513 header->snap_sizes = kmalloc(snap_count * sizeof(u64),
515 if (!header->snap_sizes)
518 header->snap_names = NULL;
519 header->snap_sizes = NULL;
521 memcpy(header->block_name, ondisk->block_name,
522 sizeof(ondisk->block_name));
524 header->image_size = le64_to_cpu(ondisk->image_size);
525 header->obj_order = ondisk->options.order;
526 header->crypt_type = ondisk->options.crypt_type;
527 header->comp_type = ondisk->options.comp_type;
529 atomic_set(&header->snapc->nref, 1);
530 header->snap_seq = le64_to_cpu(ondisk->snap_seq);
531 header->snapc->num_snaps = snap_count;
532 header->total_snaps = snap_count;
534 if (snap_count && allocated_snaps == snap_count) {
535 for (i = 0; i < snap_count; i++) {
536 header->snapc->snaps[i] =
537 le64_to_cpu(ondisk->snaps[i].id);
538 header->snap_sizes[i] =
539 le64_to_cpu(ondisk->snaps[i].image_size);
542 /* copy snapshot names */
543 memcpy(header->snap_names, &ondisk->snaps[i],
544 header->snap_names_len);
550 kfree(header->snap_names);
552 kfree(header->snapc);
556 static int snap_by_name(struct rbd_image_header *header, const char *snap_name,
560 char *p = header->snap_names;
562 for (i = 0; i < header->total_snaps; i++) {
563 if (!strcmp(snap_name, p)) {
565 /* Found it. Pass back its id and/or size */
568 *seq = header->snapc->snaps[i];
570 *size = header->snap_sizes[i];
573 p += strlen(p) + 1; /* Skip ahead to the next name */
578 static int rbd_header_set_snap(struct rbd_device *dev, u64 *size)
580 struct rbd_image_header *header = &dev->header;
581 struct ceph_snap_context *snapc = header->snapc;
584 BUILD_BUG_ON(sizeof (dev->snap_name) < sizeof (RBD_SNAP_HEAD_NAME));
586 down_write(&dev->header_rwsem);
588 if (!memcmp(dev->snap_name, RBD_SNAP_HEAD_NAME,
589 sizeof (RBD_SNAP_HEAD_NAME))) {
590 if (header->total_snaps)
591 snapc->seq = header->snap_seq;
594 dev->snap_id = CEPH_NOSNAP;
597 *size = header->image_size;
599 ret = snap_by_name(header, dev->snap_name, &snapc->seq, size);
602 dev->snap_id = snapc->seq;
608 up_write(&dev->header_rwsem);
612 static void rbd_header_free(struct rbd_image_header *header)
614 kfree(header->snapc);
615 kfree(header->snap_names);
616 kfree(header->snap_sizes);
620 * get the actual striped segment name, offset and length
622 static u64 rbd_get_segment(struct rbd_image_header *header,
623 const char *block_name,
625 char *seg_name, u64 *segofs)
627 u64 seg = ofs >> header->obj_order;
630 snprintf(seg_name, RBD_MAX_SEG_NAME_LEN,
631 "%s.%012llx", block_name, seg);
633 ofs = ofs & ((1 << header->obj_order) - 1);
634 len = min_t(u64, len, (1 << header->obj_order) - ofs);
642 static int rbd_get_num_segments(struct rbd_image_header *header,
645 u64 start_seg = ofs >> header->obj_order;
646 u64 end_seg = (ofs + len - 1) >> header->obj_order;
647 return end_seg - start_seg + 1;
651 * returns the size of an object in the image
653 static u64 rbd_obj_bytes(struct rbd_image_header *header)
655 return 1 << header->obj_order;
662 static void bio_chain_put(struct bio *chain)
668 chain = chain->bi_next;
674 * zeros a bio chain, starting at specific offset
676 static void zero_bio_chain(struct bio *chain, int start_ofs)
685 bio_for_each_segment(bv, chain, i) {
686 if (pos + bv->bv_len > start_ofs) {
687 int remainder = max(start_ofs - pos, 0);
688 buf = bvec_kmap_irq(bv, &flags);
689 memset(buf + remainder, 0,
690 bv->bv_len - remainder);
691 bvec_kunmap_irq(buf, &flags);
696 chain = chain->bi_next;
701 * bio_chain_clone - clone a chain of bios up to a certain length.
702 * might return a bio_pair that will need to be released.
704 static struct bio *bio_chain_clone(struct bio **old, struct bio **next,
705 struct bio_pair **bp,
706 int len, gfp_t gfpmask)
708 struct bio *tmp, *old_chain = *old, *new_chain = NULL, *tail = NULL;
712 bio_pair_release(*bp);
716 while (old_chain && (total < len)) {
717 tmp = bio_kmalloc(gfpmask, old_chain->bi_max_vecs);
721 if (total + old_chain->bi_size > len) {
725 * this split can only happen with a single paged bio,
726 * split_bio will BUG_ON if this is not the case
728 dout("bio_chain_clone split! total=%d remaining=%d"
730 (int)total, (int)len-total,
731 (int)old_chain->bi_size);
733 /* split the bio. We'll release it either in the next
734 call, or it will have to be released outside */
735 bp = bio_split(old_chain, (len - total) / SECTOR_SIZE);
739 __bio_clone(tmp, &bp->bio1);
743 __bio_clone(tmp, old_chain);
744 *next = old_chain->bi_next;
748 gfpmask &= ~__GFP_WAIT;
752 new_chain = tail = tmp;
757 old_chain = old_chain->bi_next;
759 total += tmp->bi_size;
765 tail->bi_next = NULL;
772 dout("bio_chain_clone with err\n");
773 bio_chain_put(new_chain);
778 * helpers for osd request op vectors.
780 static int rbd_create_rw_ops(struct ceph_osd_req_op **ops,
785 *ops = kzalloc(sizeof(struct ceph_osd_req_op) * (num_ops + 1),
789 (*ops)[0].op = opcode;
791 * op extent offset and length will be set later on
792 * in calc_raw_layout()
794 (*ops)[0].payload_len = payload_len;
798 static void rbd_destroy_ops(struct ceph_osd_req_op *ops)
803 static void rbd_coll_end_req_index(struct request *rq,
804 struct rbd_req_coll *coll,
808 struct request_queue *q;
811 dout("rbd_coll_end_req_index %p index %d ret %d len %lld\n",
812 coll, index, ret, len);
818 blk_end_request(rq, ret, len);
824 spin_lock_irq(q->queue_lock);
825 coll->status[index].done = 1;
826 coll->status[index].rc = ret;
827 coll->status[index].bytes = len;
828 max = min = coll->num_done;
829 while (max < coll->total && coll->status[max].done)
832 for (i = min; i<max; i++) {
833 __blk_end_request(rq, coll->status[i].rc,
834 coll->status[i].bytes);
836 kref_put(&coll->kref, rbd_coll_release);
838 spin_unlock_irq(q->queue_lock);
841 static void rbd_coll_end_req(struct rbd_request *req,
844 rbd_coll_end_req_index(req->rq, req->coll, req->coll_index, ret, len);
848 * Send ceph osd request
850 static int rbd_do_request(struct request *rq,
851 struct rbd_device *dev,
852 struct ceph_snap_context *snapc,
854 const char *obj, u64 ofs, u64 len,
859 struct ceph_osd_req_op *ops,
861 struct rbd_req_coll *coll,
863 void (*rbd_cb)(struct ceph_osd_request *req,
864 struct ceph_msg *msg),
865 struct ceph_osd_request **linger_req,
868 struct ceph_osd_request *req;
869 struct ceph_file_layout *layout;
872 struct timespec mtime = CURRENT_TIME;
873 struct rbd_request *req_data;
874 struct ceph_osd_request_head *reqhead;
875 struct ceph_osd_client *osdc;
877 req_data = kzalloc(sizeof(*req_data), GFP_NOIO);
880 rbd_coll_end_req_index(rq, coll, coll_index,
886 req_data->coll = coll;
887 req_data->coll_index = coll_index;
890 dout("rbd_do_request obj=%s ofs=%lld len=%lld\n", obj, len, ofs);
892 down_read(&dev->header_rwsem);
894 osdc = &dev->rbd_client->client->osdc;
895 req = ceph_osdc_alloc_request(osdc, flags, snapc, ops,
896 false, GFP_NOIO, pages, bio);
898 up_read(&dev->header_rwsem);
903 req->r_callback = rbd_cb;
907 req_data->pages = pages;
910 req->r_priv = req_data;
912 reqhead = req->r_request->front.iov_base;
913 reqhead->snapid = cpu_to_le64(CEPH_NOSNAP);
915 strncpy(req->r_oid, obj, sizeof(req->r_oid));
916 req->r_oid_len = strlen(req->r_oid);
918 layout = &req->r_file_layout;
919 memset(layout, 0, sizeof(*layout));
920 layout->fl_stripe_unit = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER);
921 layout->fl_stripe_count = cpu_to_le32(1);
922 layout->fl_object_size = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER);
923 layout->fl_pg_pool = cpu_to_le32(dev->poolid);
924 ceph_calc_raw_layout(osdc, layout, snapid, ofs, &len, &bno,
927 ceph_osdc_build_request(req, ofs, &len,
931 req->r_oid, req->r_oid_len);
932 up_read(&dev->header_rwsem);
935 ceph_osdc_set_request_linger(osdc, req);
939 ret = ceph_osdc_start_request(osdc, req, false);
944 ret = ceph_osdc_wait_request(osdc, req);
946 *ver = le64_to_cpu(req->r_reassert_version.version);
947 dout("reassert_ver=%lld\n",
948 le64_to_cpu(req->r_reassert_version.version));
949 ceph_osdc_put_request(req);
954 bio_chain_put(req_data->bio);
955 ceph_osdc_put_request(req);
957 rbd_coll_end_req(req_data, ret, len);
963 * Ceph osd op callback
965 static void rbd_req_cb(struct ceph_osd_request *req, struct ceph_msg *msg)
967 struct rbd_request *req_data = req->r_priv;
968 struct ceph_osd_reply_head *replyhead;
969 struct ceph_osd_op *op;
975 replyhead = msg->front.iov_base;
976 WARN_ON(le32_to_cpu(replyhead->num_ops) == 0);
977 op = (void *)(replyhead + 1);
978 rc = le32_to_cpu(replyhead->result);
979 bytes = le64_to_cpu(op->extent.length);
980 read_op = (le32_to_cpu(op->op) == CEPH_OSD_OP_READ);
982 dout("rbd_req_cb bytes=%lld readop=%d rc=%d\n", bytes, read_op, rc);
984 if (rc == -ENOENT && read_op) {
985 zero_bio_chain(req_data->bio, 0);
987 } else if (rc == 0 && read_op && bytes < req_data->len) {
988 zero_bio_chain(req_data->bio, bytes);
989 bytes = req_data->len;
992 rbd_coll_end_req(req_data, rc, bytes);
995 bio_chain_put(req_data->bio);
997 ceph_osdc_put_request(req);
1001 static void rbd_simple_req_cb(struct ceph_osd_request *req, struct ceph_msg *msg)
1003 ceph_osdc_put_request(req);
1007 * Do a synchronous ceph osd operation
1009 static int rbd_req_sync_op(struct rbd_device *dev,
1010 struct ceph_snap_context *snapc,
1014 struct ceph_osd_req_op *orig_ops,
1019 struct ceph_osd_request **linger_req,
1023 struct page **pages;
1025 struct ceph_osd_req_op *ops = orig_ops;
1028 num_pages = calc_pages_for(ofs , len);
1029 pages = ceph_alloc_page_vector(num_pages, GFP_KERNEL);
1031 return PTR_ERR(pages);
1034 payload_len = (flags & CEPH_OSD_FLAG_WRITE ? len : 0);
1035 ret = rbd_create_rw_ops(&ops, 1, opcode, payload_len);
1039 if ((flags & CEPH_OSD_FLAG_WRITE) && buf) {
1040 ret = ceph_copy_to_page_vector(pages, buf, ofs, len);
1046 ret = rbd_do_request(NULL, dev, snapc, snapid,
1047 obj, ofs, len, NULL,
1058 if ((flags & CEPH_OSD_FLAG_READ) && buf)
1059 ret = ceph_copy_from_page_vector(pages, buf, ofs, ret);
1063 rbd_destroy_ops(ops);
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, int num_reply,
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.block_name,
1096 seg_name, &seg_ofs);
1098 payload_len = (flags & CEPH_OSD_FLAG_WRITE ? seg_len : 0);
1100 ret = rbd_create_rw_ops(&ops, 1, opcode, payload_len);
1104 /* we've taken care of segment sizes earlier when we
1105 cloned the bios. We should never have a segment
1106 truncated at this point */
1107 BUG_ON(seg_len < len);
1109 ret = rbd_do_request(rq, rbd_dev, snapc, snapid,
1110 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,
1140 ofs, len, bio, coll, coll_index);
1144 * Request async osd read
1146 static int rbd_req_read(struct request *rq,
1147 struct rbd_device *rbd_dev,
1151 struct rbd_req_coll *coll,
1154 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 *dev,
1166 struct ceph_snap_context *snapc,
1173 return rbd_req_sync_op(dev, NULL,
1178 1, obj, ofs, len, buf, NULL, ver);
1182 * Request sync osd watch
1184 static int rbd_req_sync_notify_ack(struct rbd_device *dev,
1189 struct ceph_osd_req_op *ops;
1190 struct page **pages = NULL;
1193 ret = rbd_create_rw_ops(&ops, 1, CEPH_OSD_OP_NOTIFY_ACK, 0);
1197 ops[0].watch.ver = cpu_to_le64(dev->header.obj_version);
1198 ops[0].watch.cookie = notify_id;
1199 ops[0].watch.flag = 0;
1201 ret = rbd_do_request(NULL, dev, NULL, CEPH_NOSNAP,
1208 rbd_simple_req_cb, 0, NULL);
1210 rbd_destroy_ops(ops);
1214 static void rbd_watch_cb(u64 ver, u64 notify_id, u8 opcode, void *data)
1216 struct rbd_device *dev = (struct rbd_device *)data;
1222 dout("rbd_watch_cb %s notify_id=%lld opcode=%d\n", dev->obj_md_name,
1223 notify_id, (int)opcode);
1224 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
1225 rc = __rbd_refresh_header(dev);
1226 mutex_unlock(&ctl_mutex);
1228 pr_warning(RBD_DRV_NAME "%d got notification but failed to "
1229 " update snaps: %d\n", dev->major, rc);
1231 rbd_req_sync_notify_ack(dev, ver, notify_id, dev->obj_md_name);
1235 * Request sync osd watch
1237 static int rbd_req_sync_watch(struct rbd_device *dev,
1241 struct ceph_osd_req_op *ops;
1242 struct ceph_osd_client *osdc = &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 *)dev, &dev->watch_event);
1253 ops[0].watch.ver = cpu_to_le64(ver);
1254 ops[0].watch.cookie = cpu_to_le64(dev->watch_event->cookie);
1255 ops[0].watch.flag = 1;
1257 ret = rbd_req_sync_op(dev, NULL,
1260 CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK,
1263 &dev->watch_request, NULL);
1268 rbd_destroy_ops(ops);
1272 ceph_osdc_cancel_event(dev->watch_event);
1273 dev->watch_event = NULL;
1275 rbd_destroy_ops(ops);
1280 * Request sync osd unwatch
1282 static int rbd_req_sync_unwatch(struct rbd_device *dev,
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(dev->watch_event->cookie);
1293 ops[0].watch.flag = 0;
1295 ret = rbd_req_sync_op(dev, NULL,
1298 CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK,
1300 1, obj, 0, 0, NULL, NULL, NULL);
1302 rbd_destroy_ops(ops);
1303 ceph_osdc_cancel_event(dev->watch_event);
1304 dev->watch_event = NULL;
1308 struct rbd_notify_info {
1309 struct rbd_device *dev;
1312 static void rbd_notify_cb(u64 ver, u64 notify_id, u8 opcode, void *data)
1314 struct rbd_device *dev = (struct rbd_device *)data;
1318 dout("rbd_notify_cb %s notify_id=%lld opcode=%d\n", dev->obj_md_name,
1319 notify_id, (int)opcode);
1323 * Request sync osd notify
1325 static int rbd_req_sync_notify(struct rbd_device *dev,
1328 struct ceph_osd_req_op *ops;
1329 struct ceph_osd_client *osdc = &dev->rbd_client->client->osdc;
1330 struct ceph_osd_event *event;
1331 struct rbd_notify_info info;
1332 int payload_len = sizeof(u32) + sizeof(u32);
1335 ret = rbd_create_rw_ops(&ops, 1, CEPH_OSD_OP_NOTIFY, payload_len);
1341 ret = ceph_osdc_create_event(osdc, rbd_notify_cb, 1,
1342 (void *)&info, &event);
1346 ops[0].watch.ver = 1;
1347 ops[0].watch.flag = 1;
1348 ops[0].watch.cookie = event->cookie;
1349 ops[0].watch.prot_ver = RADOS_NOTIFY_VER;
1350 ops[0].watch.timeout = 12;
1352 ret = rbd_req_sync_op(dev, NULL,
1355 CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK,
1357 1, obj, 0, 0, NULL, NULL, NULL);
1361 ret = ceph_osdc_wait_event(event, CEPH_OSD_TIMEOUT_DEFAULT);
1362 dout("ceph_osdc_wait_event returned %d\n", ret);
1363 rbd_destroy_ops(ops);
1367 ceph_osdc_cancel_event(event);
1369 rbd_destroy_ops(ops);
1374 * Request sync osd read
1376 static int rbd_req_sync_exec(struct rbd_device *dev,
1384 struct ceph_osd_req_op *ops;
1385 int cls_len = strlen(cls);
1386 int method_len = strlen(method);
1387 int ret = rbd_create_rw_ops(&ops, 1, CEPH_OSD_OP_CALL,
1388 cls_len + method_len + len);
1392 ops[0].cls.class_name = cls;
1393 ops[0].cls.class_len = (__u8)cls_len;
1394 ops[0].cls.method_name = method;
1395 ops[0].cls.method_len = (__u8)method_len;
1396 ops[0].cls.argc = 0;
1397 ops[0].cls.indata = data;
1398 ops[0].cls.indata_len = len;
1400 ret = rbd_req_sync_op(dev, NULL,
1403 CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK,
1405 1, obj, 0, 0, NULL, NULL, ver);
1407 rbd_destroy_ops(ops);
1409 dout("cls_exec returned %d\n", ret);
1413 static struct rbd_req_coll *rbd_alloc_coll(int num_reqs)
1415 struct rbd_req_coll *coll =
1416 kzalloc(sizeof(struct rbd_req_coll) +
1417 sizeof(struct rbd_req_status) * num_reqs,
1422 coll->total = num_reqs;
1423 kref_init(&coll->kref);
1428 * block device queue callback
1430 static void rbd_rq_fn(struct request_queue *q)
1432 struct rbd_device *rbd_dev = q->queuedata;
1434 struct bio_pair *bp = NULL;
1436 while ((rq = blk_fetch_request(q))) {
1438 struct bio *rq_bio, *next_bio = NULL;
1440 int size, op_size = 0;
1442 int num_segs, cur_seg = 0;
1443 struct rbd_req_coll *coll;
1445 /* peek at request from block layer */
1449 dout("fetched request\n");
1451 /* filter out block requests we don't understand */
1452 if ((rq->cmd_type != REQ_TYPE_FS)) {
1453 __blk_end_request_all(rq, 0);
1457 /* deduce our operation (read, write) */
1458 do_write = (rq_data_dir(rq) == WRITE);
1460 size = blk_rq_bytes(rq);
1461 ofs = blk_rq_pos(rq) * SECTOR_SIZE;
1463 if (do_write && rbd_dev->read_only) {
1464 __blk_end_request_all(rq, -EROFS);
1468 spin_unlock_irq(q->queue_lock);
1470 dout("%s 0x%x bytes at 0x%llx\n",
1471 do_write ? "write" : "read",
1472 size, blk_rq_pos(rq) * SECTOR_SIZE);
1474 num_segs = rbd_get_num_segments(&rbd_dev->header, ofs, size);
1475 coll = rbd_alloc_coll(num_segs);
1477 spin_lock_irq(q->queue_lock);
1478 __blk_end_request_all(rq, -ENOMEM);
1483 /* a bio clone to be passed down to OSD req */
1484 dout("rq->bio->bi_vcnt=%d\n", rq->bio->bi_vcnt);
1485 op_size = rbd_get_segment(&rbd_dev->header,
1486 rbd_dev->header.block_name,
1489 kref_get(&coll->kref);
1490 bio = bio_chain_clone(&rq_bio, &next_bio, &bp,
1491 op_size, GFP_ATOMIC);
1493 rbd_coll_end_req_index(rq, coll, cur_seg,
1499 /* init OSD command: write or read */
1501 rbd_req_write(rq, rbd_dev,
1502 rbd_dev->header.snapc,
1507 rbd_req_read(rq, rbd_dev,
1520 kref_put(&coll->kref, rbd_coll_release);
1523 bio_pair_release(bp);
1524 spin_lock_irq(q->queue_lock);
1529 * a queue callback. Makes sure that we don't create a bio that spans across
1530 * multiple osd objects. One exception would be with a single page bios,
1531 * which we handle later at bio_chain_clone
1533 static int rbd_merge_bvec(struct request_queue *q, struct bvec_merge_data *bmd,
1534 struct bio_vec *bvec)
1536 struct rbd_device *rbd_dev = q->queuedata;
1537 unsigned int chunk_sectors;
1539 unsigned int bio_sectors;
1542 chunk_sectors = 1 << (rbd_dev->header.obj_order - SECTOR_SHIFT);
1543 sector = bmd->bi_sector + get_start_sect(bmd->bi_bdev);
1544 bio_sectors = bmd->bi_size >> SECTOR_SHIFT;
1546 max = (chunk_sectors - ((sector & (chunk_sectors - 1))
1547 + bio_sectors)) << SECTOR_SHIFT;
1549 max = 0; /* bio_add cannot handle a negative return */
1550 if (max <= bvec->bv_len && bio_sectors == 0)
1551 return bvec->bv_len;
1555 static void rbd_free_disk(struct rbd_device *rbd_dev)
1557 struct gendisk *disk = rbd_dev->disk;
1562 rbd_header_free(&rbd_dev->header);
1564 if (disk->flags & GENHD_FL_UP)
1567 blk_cleanup_queue(disk->queue);
1572 * reload the ondisk the header
1574 static int rbd_read_header(struct rbd_device *rbd_dev,
1575 struct rbd_image_header *header)
1578 struct rbd_image_header_ondisk *dh;
1584 * First reads the fixed-size header to determine the number
1585 * of snapshots, then re-reads it, along with all snapshot
1586 * records as well as their stored names.
1590 dh = kmalloc(len, GFP_KERNEL);
1594 rc = rbd_req_sync_read(rbd_dev,
1596 rbd_dev->obj_md_name,
1602 rc = rbd_header_from_disk(header, dh, snap_count, GFP_KERNEL);
1605 pr_warning("unrecognized header format"
1606 " for image %s", rbd_dev->obj);
1610 if (snap_count == header->total_snaps)
1613 snap_count = header->total_snaps;
1614 len = sizeof (*dh) +
1615 snap_count * sizeof(struct rbd_image_snap_ondisk) +
1616 header->snap_names_len;
1618 rbd_header_free(header);
1621 header->obj_version = ver;
1631 static int rbd_header_add_snap(struct rbd_device *dev,
1632 const char *snap_name,
1635 int name_len = strlen(snap_name);
1640 struct ceph_mon_client *monc;
1642 /* we should create a snapshot only if we're pointing at the head */
1643 if (dev->snap_id != CEPH_NOSNAP)
1646 monc = &dev->rbd_client->client->monc;
1647 ret = ceph_monc_create_snapid(monc, dev->poolid, &new_snapid);
1648 dout("created snapid=%lld\n", new_snapid);
1652 data = kmalloc(name_len + 16, gfp_flags);
1657 e = data + name_len + 16;
1659 ceph_encode_string_safe(&p, e, snap_name, name_len, bad);
1660 ceph_encode_64_safe(&p, e, new_snapid, bad);
1662 ret = rbd_req_sync_exec(dev, dev->obj_md_name, "rbd", "snap_add",
1663 data, p - data, &ver);
1670 down_write(&dev->header_rwsem);
1671 dev->header.snapc->seq = new_snapid;
1672 up_write(&dev->header_rwsem);
1679 static void __rbd_remove_all_snaps(struct rbd_device *rbd_dev)
1681 struct rbd_snap *snap;
1683 while (!list_empty(&rbd_dev->snaps)) {
1684 snap = list_first_entry(&rbd_dev->snaps, struct rbd_snap, node);
1685 __rbd_remove_snap_dev(rbd_dev, snap);
1690 * only read the first part of the ondisk header, without the snaps info
1692 static int __rbd_refresh_header(struct rbd_device *rbd_dev)
1695 struct rbd_image_header h;
1699 ret = rbd_read_header(rbd_dev, &h);
1704 set_capacity(rbd_dev->disk, h.image_size / SECTOR_SIZE);
1706 down_write(&rbd_dev->header_rwsem);
1708 snap_seq = rbd_dev->header.snapc->seq;
1709 if (rbd_dev->header.total_snaps &&
1710 rbd_dev->header.snapc->snaps[0] == snap_seq)
1711 /* pointing at the head, will need to follow that
1715 kfree(rbd_dev->header.snapc);
1716 kfree(rbd_dev->header.snap_names);
1717 kfree(rbd_dev->header.snap_sizes);
1719 rbd_dev->header.total_snaps = h.total_snaps;
1720 rbd_dev->header.snapc = h.snapc;
1721 rbd_dev->header.snap_names = h.snap_names;
1722 rbd_dev->header.snap_names_len = h.snap_names_len;
1723 rbd_dev->header.snap_sizes = h.snap_sizes;
1725 rbd_dev->header.snapc->seq = rbd_dev->header.snapc->snaps[0];
1727 rbd_dev->header.snapc->seq = snap_seq;
1729 ret = __rbd_init_snaps_header(rbd_dev);
1731 up_write(&rbd_dev->header_rwsem);
1736 static int rbd_init_disk(struct rbd_device *rbd_dev)
1738 struct gendisk *disk;
1739 struct request_queue *q;
1744 /* contact OSD, request size info about the object being mapped */
1745 rc = rbd_read_header(rbd_dev, &rbd_dev->header);
1749 /* no need to lock here, as rbd_dev is not registered yet */
1750 rc = __rbd_init_snaps_header(rbd_dev);
1754 rc = rbd_header_set_snap(rbd_dev, &total_size);
1758 /* create gendisk info */
1760 disk = alloc_disk(RBD_MINORS_PER_MAJOR);
1764 snprintf(disk->disk_name, sizeof(disk->disk_name), RBD_DRV_NAME "%d",
1766 disk->major = rbd_dev->major;
1767 disk->first_minor = 0;
1768 disk->fops = &rbd_bd_ops;
1769 disk->private_data = rbd_dev;
1773 q = blk_init_queue(rbd_rq_fn, &rbd_dev->lock);
1777 /* We use the default size, but let's be explicit about it. */
1778 blk_queue_physical_block_size(q, SECTOR_SIZE);
1780 /* set io sizes to object size */
1781 segment_size = rbd_obj_bytes(&rbd_dev->header);
1782 blk_queue_max_hw_sectors(q, segment_size / SECTOR_SIZE);
1783 blk_queue_max_segment_size(q, segment_size);
1784 blk_queue_io_min(q, segment_size);
1785 blk_queue_io_opt(q, segment_size);
1787 blk_queue_merge_bvec(q, rbd_merge_bvec);
1790 q->queuedata = rbd_dev;
1792 rbd_dev->disk = disk;
1795 /* finally, announce the disk to the world */
1796 set_capacity(disk, total_size / SECTOR_SIZE);
1799 pr_info("%s: added with size 0x%llx\n",
1800 disk->disk_name, (unsigned long long)total_size);
1813 static struct rbd_device *dev_to_rbd_dev(struct device *dev)
1815 return container_of(dev, struct rbd_device, dev);
1818 static ssize_t rbd_size_show(struct device *dev,
1819 struct device_attribute *attr, char *buf)
1821 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
1823 return sprintf(buf, "%llu\n", (unsigned long long)rbd_dev->header.image_size);
1826 static ssize_t rbd_major_show(struct device *dev,
1827 struct device_attribute *attr, char *buf)
1829 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
1831 return sprintf(buf, "%d\n", rbd_dev->major);
1834 static ssize_t rbd_client_id_show(struct device *dev,
1835 struct device_attribute *attr, char *buf)
1837 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
1839 return sprintf(buf, "client%lld\n",
1840 ceph_client_id(rbd_dev->rbd_client->client));
1843 static ssize_t rbd_pool_show(struct device *dev,
1844 struct device_attribute *attr, char *buf)
1846 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
1848 return sprintf(buf, "%s\n", rbd_dev->pool_name);
1851 static ssize_t rbd_name_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, "%s\n", rbd_dev->obj);
1859 static ssize_t rbd_snap_show(struct device *dev,
1860 struct device_attribute *attr,
1863 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
1865 return sprintf(buf, "%s\n", rbd_dev->snap_name);
1868 static ssize_t rbd_image_refresh(struct device *dev,
1869 struct device_attribute *attr,
1873 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
1877 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
1879 rc = __rbd_refresh_header(rbd_dev);
1883 mutex_unlock(&ctl_mutex);
1887 static DEVICE_ATTR(size, S_IRUGO, rbd_size_show, NULL);
1888 static DEVICE_ATTR(major, S_IRUGO, rbd_major_show, NULL);
1889 static DEVICE_ATTR(client_id, S_IRUGO, rbd_client_id_show, NULL);
1890 static DEVICE_ATTR(pool, S_IRUGO, rbd_pool_show, NULL);
1891 static DEVICE_ATTR(name, S_IRUGO, rbd_name_show, NULL);
1892 static DEVICE_ATTR(refresh, S_IWUSR, NULL, rbd_image_refresh);
1893 static DEVICE_ATTR(current_snap, S_IRUGO, rbd_snap_show, NULL);
1894 static DEVICE_ATTR(create_snap, S_IWUSR, NULL, rbd_snap_add);
1896 static struct attribute *rbd_attrs[] = {
1897 &dev_attr_size.attr,
1898 &dev_attr_major.attr,
1899 &dev_attr_client_id.attr,
1900 &dev_attr_pool.attr,
1901 &dev_attr_name.attr,
1902 &dev_attr_current_snap.attr,
1903 &dev_attr_refresh.attr,
1904 &dev_attr_create_snap.attr,
1908 static struct attribute_group rbd_attr_group = {
1912 static const struct attribute_group *rbd_attr_groups[] = {
1917 static void rbd_sysfs_dev_release(struct device *dev)
1921 static struct device_type rbd_device_type = {
1923 .groups = rbd_attr_groups,
1924 .release = rbd_sysfs_dev_release,
1932 static ssize_t rbd_snap_size_show(struct device *dev,
1933 struct device_attribute *attr,
1936 struct rbd_snap *snap = container_of(dev, struct rbd_snap, dev);
1938 return sprintf(buf, "%llu\n", (unsigned long long)snap->size);
1941 static ssize_t rbd_snap_id_show(struct device *dev,
1942 struct device_attribute *attr,
1945 struct rbd_snap *snap = container_of(dev, struct rbd_snap, dev);
1947 return sprintf(buf, "%llu\n", (unsigned long long)snap->id);
1950 static DEVICE_ATTR(snap_size, S_IRUGO, rbd_snap_size_show, NULL);
1951 static DEVICE_ATTR(snap_id, S_IRUGO, rbd_snap_id_show, NULL);
1953 static struct attribute *rbd_snap_attrs[] = {
1954 &dev_attr_snap_size.attr,
1955 &dev_attr_snap_id.attr,
1959 static struct attribute_group rbd_snap_attr_group = {
1960 .attrs = rbd_snap_attrs,
1963 static void rbd_snap_dev_release(struct device *dev)
1965 struct rbd_snap *snap = container_of(dev, struct rbd_snap, dev);
1970 static const struct attribute_group *rbd_snap_attr_groups[] = {
1971 &rbd_snap_attr_group,
1975 static struct device_type rbd_snap_device_type = {
1976 .groups = rbd_snap_attr_groups,
1977 .release = rbd_snap_dev_release,
1980 static void __rbd_remove_snap_dev(struct rbd_device *rbd_dev,
1981 struct rbd_snap *snap)
1983 list_del(&snap->node);
1984 device_unregister(&snap->dev);
1987 static int rbd_register_snap_dev(struct rbd_device *rbd_dev,
1988 struct rbd_snap *snap,
1989 struct device *parent)
1991 struct device *dev = &snap->dev;
1994 dev->type = &rbd_snap_device_type;
1995 dev->parent = parent;
1996 dev->release = rbd_snap_dev_release;
1997 dev_set_name(dev, "snap_%s", snap->name);
1998 ret = device_register(dev);
2003 static int __rbd_add_snap_dev(struct rbd_device *rbd_dev,
2004 int i, const char *name,
2005 struct rbd_snap **snapp)
2008 struct rbd_snap *snap = kzalloc(sizeof(*snap), GFP_KERNEL);
2011 snap->name = kstrdup(name, GFP_KERNEL);
2012 snap->size = rbd_dev->header.snap_sizes[i];
2013 snap->id = rbd_dev->header.snapc->snaps[i];
2014 if (device_is_registered(&rbd_dev->dev)) {
2015 ret = rbd_register_snap_dev(rbd_dev, snap,
2029 * search for the previous snap in a null delimited string list
2031 const char *rbd_prev_snap_name(const char *name, const char *start)
2033 if (name < start + 2)
2046 * compare the old list of snapshots that we have to what's in the header
2047 * and update it accordingly. Note that the header holds the snapshots
2048 * in a reverse order (from newest to oldest) and we need to go from
2049 * older to new so that we don't get a duplicate snap name when
2050 * doing the process (e.g., removed snapshot and recreated a new
2051 * one with the same name.
2053 static int __rbd_init_snaps_header(struct rbd_device *rbd_dev)
2055 const char *name, *first_name;
2056 int i = rbd_dev->header.total_snaps;
2057 struct rbd_snap *snap, *old_snap = NULL;
2059 struct list_head *p, *n;
2061 first_name = rbd_dev->header.snap_names;
2062 name = first_name + rbd_dev->header.snap_names_len;
2064 list_for_each_prev_safe(p, n, &rbd_dev->snaps) {
2067 old_snap = list_entry(p, struct rbd_snap, node);
2070 cur_id = rbd_dev->header.snapc->snaps[i - 1];
2072 if (!i || old_snap->id < cur_id) {
2073 /* old_snap->id was skipped, thus was removed */
2074 __rbd_remove_snap_dev(rbd_dev, old_snap);
2077 if (old_snap->id == cur_id) {
2078 /* we have this snapshot already */
2080 name = rbd_prev_snap_name(name, first_name);
2084 i--, name = rbd_prev_snap_name(name, first_name)) {
2089 cur_id = rbd_dev->header.snapc->snaps[i];
2090 /* snapshot removal? handle it above */
2091 if (cur_id >= old_snap->id)
2093 /* a new snapshot */
2094 ret = __rbd_add_snap_dev(rbd_dev, i - 1, name, &snap);
2098 /* note that we add it backward so using n and not p */
2099 list_add(&snap->node, n);
2103 /* we're done going over the old snap list, just add what's left */
2104 for (; i > 0; i--) {
2105 name = rbd_prev_snap_name(name, first_name);
2110 ret = __rbd_add_snap_dev(rbd_dev, i - 1, name, &snap);
2113 list_add(&snap->node, &rbd_dev->snaps);
2119 static int rbd_bus_add_dev(struct rbd_device *rbd_dev)
2123 struct rbd_snap *snap;
2125 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
2126 dev = &rbd_dev->dev;
2128 dev->bus = &rbd_bus_type;
2129 dev->type = &rbd_device_type;
2130 dev->parent = &rbd_root_dev;
2131 dev->release = rbd_dev_release;
2132 dev_set_name(dev, "%d", rbd_dev->id);
2133 ret = device_register(dev);
2137 list_for_each_entry(snap, &rbd_dev->snaps, node) {
2138 ret = rbd_register_snap_dev(rbd_dev, snap,
2144 mutex_unlock(&ctl_mutex);
2148 static void rbd_bus_del_dev(struct rbd_device *rbd_dev)
2150 device_unregister(&rbd_dev->dev);
2153 static int rbd_init_watch_dev(struct rbd_device *rbd_dev)
2158 ret = rbd_req_sync_watch(rbd_dev, rbd_dev->obj_md_name,
2159 rbd_dev->header.obj_version);
2160 if (ret == -ERANGE) {
2161 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
2162 rc = __rbd_refresh_header(rbd_dev);
2163 mutex_unlock(&ctl_mutex);
2167 } while (ret == -ERANGE);
2172 static atomic64_t rbd_id_max = ATOMIC64_INIT(0);
2175 * Get a unique rbd identifier for the given new rbd_dev, and add
2176 * the rbd_dev to the global list. The minimum rbd id is 1.
2178 static void rbd_id_get(struct rbd_device *rbd_dev)
2180 rbd_dev->id = atomic64_inc_return(&rbd_id_max);
2182 spin_lock(&rbd_dev_list_lock);
2183 list_add_tail(&rbd_dev->node, &rbd_dev_list);
2184 spin_unlock(&rbd_dev_list_lock);
2188 * Remove an rbd_dev from the global list, and record that its
2189 * identifier is no longer in use.
2191 static void rbd_id_put(struct rbd_device *rbd_dev)
2193 struct list_head *tmp;
2194 int rbd_id = rbd_dev->id;
2199 spin_lock(&rbd_dev_list_lock);
2200 list_del_init(&rbd_dev->node);
2203 * If the id being "put" is not the current maximum, there
2204 * is nothing special we need to do.
2206 if (rbd_id != atomic64_read(&rbd_id_max)) {
2207 spin_unlock(&rbd_dev_list_lock);
2212 * We need to update the current maximum id. Search the
2213 * list to find out what it is. We're more likely to find
2214 * the maximum at the end, so search the list backward.
2217 list_for_each_prev(tmp, &rbd_dev_list) {
2218 struct rbd_device *rbd_dev;
2220 rbd_dev = list_entry(tmp, struct rbd_device, node);
2221 if (rbd_id > max_id)
2224 spin_unlock(&rbd_dev_list_lock);
2227 * The max id could have been updated by rbd_id_get(), in
2228 * which case it now accurately reflects the new maximum.
2229 * Be careful not to overwrite the maximum value in that
2232 atomic64_cmpxchg(&rbd_id_max, rbd_id, max_id);
2236 * Skips over white space at *buf, and updates *buf to point to the
2237 * first found non-space character (if any). Returns the length of
2238 * the token (string of non-white space characters) found. Note
2239 * that *buf must be terminated with '\0'.
2241 static inline size_t next_token(const char **buf)
2244 * These are the characters that produce nonzero for
2245 * isspace() in the "C" and "POSIX" locales.
2247 const char *spaces = " \f\n\r\t\v";
2249 *buf += strspn(*buf, spaces); /* Find start of token */
2251 return strcspn(*buf, spaces); /* Return token length */
2255 * Finds the next token in *buf, and if the provided token buffer is
2256 * big enough, copies the found token into it. The result, if
2257 * copied, is guaranteed to be terminated with '\0'. Note that *buf
2258 * must be terminated with '\0' on entry.
2260 * Returns the length of the token found (not including the '\0').
2261 * Return value will be 0 if no token is found, and it will be >=
2262 * token_size if the token would not fit.
2264 * The *buf pointer will be updated to point beyond the end of the
2265 * found token. Note that this occurs even if the token buffer is
2266 * too small to hold it.
2268 static inline size_t copy_token(const char **buf,
2274 len = next_token(buf);
2275 if (len < token_size) {
2276 memcpy(token, *buf, len);
2277 *(token + len) = '\0';
2285 * This fills in the pool_name, obj, obj_len, snap_name, obj_len,
2286 * rbd_dev, rbd_md_name, and name fields of the given rbd_dev, based
2287 * on the list of monitor addresses and other options provided via
2290 static int rbd_add_parse_args(struct rbd_device *rbd_dev,
2292 const char **mon_addrs,
2293 size_t *mon_addrs_size,
2295 size_t options_size)
2299 /* The first four tokens are required */
2301 len = next_token(&buf);
2304 *mon_addrs_size = len + 1;
2309 len = copy_token(&buf, options, options_size);
2310 if (!len || len >= options_size)
2313 len = copy_token(&buf, rbd_dev->pool_name, sizeof (rbd_dev->pool_name));
2314 if (!len || len >= sizeof (rbd_dev->pool_name))
2317 len = copy_token(&buf, rbd_dev->obj, sizeof (rbd_dev->obj));
2318 if (!len || len >= sizeof (rbd_dev->obj))
2321 /* We have the object length in hand, save it. */
2323 rbd_dev->obj_len = len;
2325 BUILD_BUG_ON(RBD_MAX_MD_NAME_LEN
2326 < RBD_MAX_OBJ_NAME_LEN + sizeof (RBD_SUFFIX));
2327 sprintf(rbd_dev->obj_md_name, "%s%s", rbd_dev->obj, RBD_SUFFIX);
2330 * The snapshot name is optional, but it's an error if it's
2331 * too long. If no snapshot is supplied, fill in the default.
2333 len = copy_token(&buf, rbd_dev->snap_name, sizeof (rbd_dev->snap_name));
2335 memcpy(rbd_dev->snap_name, RBD_SNAP_HEAD_NAME,
2336 sizeof (RBD_SNAP_HEAD_NAME));
2337 else if (len >= sizeof (rbd_dev->snap_name))
2343 static ssize_t rbd_add(struct bus_type *bus,
2347 struct rbd_device *rbd_dev;
2348 const char *mon_addrs = NULL;
2349 size_t mon_addrs_size = 0;
2350 char *options = NULL;
2351 struct ceph_osd_client *osdc;
2354 if (!try_module_get(THIS_MODULE))
2357 rbd_dev = kzalloc(sizeof(*rbd_dev), GFP_KERNEL);
2360 options = kmalloc(count, GFP_KERNEL);
2364 /* static rbd_device initialization */
2365 spin_lock_init(&rbd_dev->lock);
2366 INIT_LIST_HEAD(&rbd_dev->node);
2367 INIT_LIST_HEAD(&rbd_dev->snaps);
2368 init_rwsem(&rbd_dev->header_rwsem);
2370 init_rwsem(&rbd_dev->header_rwsem);
2372 /* generate unique id: find highest unique id, add one */
2373 rbd_id_get(rbd_dev);
2375 /* Fill in the device name, now that we have its id. */
2376 BUILD_BUG_ON(DEV_NAME_LEN
2377 < sizeof (RBD_DRV_NAME) + MAX_INT_FORMAT_WIDTH);
2378 sprintf(rbd_dev->name, "%s%d", RBD_DRV_NAME, rbd_dev->id);
2380 /* parse add command */
2381 rc = rbd_add_parse_args(rbd_dev, buf, &mon_addrs, &mon_addrs_size,
2386 rbd_dev->rbd_client = rbd_get_client(mon_addrs, mon_addrs_size - 1,
2388 if (IS_ERR(rbd_dev->rbd_client)) {
2389 rc = PTR_ERR(rbd_dev->rbd_client);
2394 osdc = &rbd_dev->rbd_client->client->osdc;
2395 rc = ceph_pg_poolid_by_name(osdc->osdmap, rbd_dev->pool_name);
2397 goto err_out_client;
2398 rbd_dev->poolid = rc;
2400 /* register our block device */
2401 rc = register_blkdev(0, rbd_dev->name);
2403 goto err_out_client;
2404 rbd_dev->major = rc;
2406 rc = rbd_bus_add_dev(rbd_dev);
2408 goto err_out_blkdev;
2411 * At this point cleanup in the event of an error is the job
2412 * of the sysfs code (initiated by rbd_bus_del_dev()).
2414 * Set up and announce blkdev mapping.
2416 rc = rbd_init_disk(rbd_dev);
2420 rc = rbd_init_watch_dev(rbd_dev);
2427 /* this will also clean up rest of rbd_dev stuff */
2429 rbd_bus_del_dev(rbd_dev);
2434 unregister_blkdev(rbd_dev->major, rbd_dev->name);
2436 rbd_put_client(rbd_dev);
2438 rbd_id_put(rbd_dev);
2443 dout("Error adding device %s\n", buf);
2444 module_put(THIS_MODULE);
2446 return (ssize_t) rc;
2449 static struct rbd_device *__rbd_get_dev(unsigned long id)
2451 struct list_head *tmp;
2452 struct rbd_device *rbd_dev;
2454 spin_lock(&rbd_dev_list_lock);
2455 list_for_each(tmp, &rbd_dev_list) {
2456 rbd_dev = list_entry(tmp, struct rbd_device, node);
2457 if (rbd_dev->id == id) {
2458 spin_unlock(&rbd_dev_list_lock);
2462 spin_unlock(&rbd_dev_list_lock);
2466 static void rbd_dev_release(struct device *dev)
2468 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
2470 if (rbd_dev->watch_request) {
2471 struct ceph_client *client = rbd_dev->rbd_client->client;
2473 ceph_osdc_unregister_linger_request(&client->osdc,
2474 rbd_dev->watch_request);
2476 if (rbd_dev->watch_event)
2477 rbd_req_sync_unwatch(rbd_dev, rbd_dev->obj_md_name);
2479 rbd_put_client(rbd_dev);
2481 /* clean up and free blkdev */
2482 rbd_free_disk(rbd_dev);
2483 unregister_blkdev(rbd_dev->major, rbd_dev->name);
2485 /* done with the id, and with the rbd_dev */
2486 rbd_id_put(rbd_dev);
2489 /* release module ref */
2490 module_put(THIS_MODULE);
2493 static ssize_t rbd_remove(struct bus_type *bus,
2497 struct rbd_device *rbd_dev = NULL;
2502 rc = strict_strtoul(buf, 10, &ul);
2506 /* convert to int; abort if we lost anything in the conversion */
2507 target_id = (int) ul;
2508 if (target_id != ul)
2511 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
2513 rbd_dev = __rbd_get_dev(target_id);
2519 __rbd_remove_all_snaps(rbd_dev);
2520 rbd_bus_del_dev(rbd_dev);
2523 mutex_unlock(&ctl_mutex);
2527 static ssize_t rbd_snap_add(struct device *dev,
2528 struct device_attribute *attr,
2532 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
2534 char *name = kmalloc(count + 1, GFP_KERNEL);
2538 snprintf(name, count, "%s", buf);
2540 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
2542 ret = rbd_header_add_snap(rbd_dev,
2547 ret = __rbd_refresh_header(rbd_dev);
2551 /* shouldn't hold ctl_mutex when notifying.. notify might
2552 trigger a watch callback that would need to get that mutex */
2553 mutex_unlock(&ctl_mutex);
2555 /* make a best effort, don't error if failed */
2556 rbd_req_sync_notify(rbd_dev, rbd_dev->obj_md_name);
2563 mutex_unlock(&ctl_mutex);
2569 * create control files in sysfs
2572 static int rbd_sysfs_init(void)
2576 ret = device_register(&rbd_root_dev);
2580 ret = bus_register(&rbd_bus_type);
2582 device_unregister(&rbd_root_dev);
2587 static void rbd_sysfs_cleanup(void)
2589 bus_unregister(&rbd_bus_type);
2590 device_unregister(&rbd_root_dev);
2593 int __init rbd_init(void)
2597 rc = rbd_sysfs_init();
2600 pr_info("loaded " RBD_DRV_NAME_LONG "\n");
2604 void __exit rbd_exit(void)
2606 rbd_sysfs_cleanup();
2609 module_init(rbd_init);
2610 module_exit(rbd_exit);
2612 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
2613 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
2614 MODULE_DESCRIPTION("rados block device");
2616 /* following authorship retained from original osdblk.c */
2617 MODULE_AUTHOR("Jeff Garzik <jeff@garzik.org>");
2619 MODULE_LICENSE("GPL");