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 u32 cur_snap; /* index+1 of current snapshot within snap context
182 struct list_head node;
184 /* list of snapshots */
185 struct list_head snaps;
191 static DEFINE_MUTEX(ctl_mutex); /* Serialize open/close/setup/teardown */
193 static LIST_HEAD(rbd_dev_list); /* devices */
194 static DEFINE_SPINLOCK(rbd_dev_list_lock);
196 static LIST_HEAD(rbd_client_list); /* clients */
197 static DEFINE_SPINLOCK(rbd_client_list_lock);
199 static int __rbd_init_snaps_header(struct rbd_device *rbd_dev);
200 static void rbd_dev_release(struct device *dev);
201 static ssize_t rbd_snap_add(struct device *dev,
202 struct device_attribute *attr,
205 static void __rbd_remove_snap_dev(struct rbd_device *rbd_dev,
206 struct rbd_snap *snap);
208 static ssize_t rbd_add(struct bus_type *bus, const char *buf,
210 static ssize_t rbd_remove(struct bus_type *bus, const char *buf,
213 static struct bus_attribute rbd_bus_attrs[] = {
214 __ATTR(add, S_IWUSR, NULL, rbd_add),
215 __ATTR(remove, S_IWUSR, NULL, rbd_remove),
219 static struct bus_type rbd_bus_type = {
221 .bus_attrs = rbd_bus_attrs,
224 static void rbd_root_dev_release(struct device *dev)
228 static struct device rbd_root_dev = {
230 .release = rbd_root_dev_release,
234 static struct device *rbd_get_dev(struct rbd_device *rbd_dev)
236 return get_device(&rbd_dev->dev);
239 static void rbd_put_dev(struct rbd_device *rbd_dev)
241 put_device(&rbd_dev->dev);
244 static int __rbd_update_snaps(struct rbd_device *rbd_dev);
246 static int rbd_open(struct block_device *bdev, fmode_t mode)
248 struct rbd_device *rbd_dev = bdev->bd_disk->private_data;
250 rbd_get_dev(rbd_dev);
252 set_device_ro(bdev, rbd_dev->read_only);
254 if ((mode & FMODE_WRITE) && rbd_dev->read_only)
260 static int rbd_release(struct gendisk *disk, fmode_t mode)
262 struct rbd_device *rbd_dev = disk->private_data;
264 rbd_put_dev(rbd_dev);
269 static const struct block_device_operations rbd_bd_ops = {
270 .owner = THIS_MODULE,
272 .release = rbd_release,
276 * Initialize an rbd client instance.
279 static struct rbd_client *rbd_client_create(struct ceph_options *opt,
280 struct rbd_options *rbd_opts)
282 struct rbd_client *rbdc;
285 dout("rbd_client_create\n");
286 rbdc = kmalloc(sizeof(struct rbd_client), GFP_KERNEL);
290 kref_init(&rbdc->kref);
291 INIT_LIST_HEAD(&rbdc->node);
293 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
295 rbdc->client = ceph_create_client(opt, rbdc, 0, 0);
296 if (IS_ERR(rbdc->client))
298 opt = NULL; /* Now rbdc->client is responsible for opt */
300 ret = ceph_open_session(rbdc->client);
304 rbdc->rbd_opts = rbd_opts;
306 spin_lock(&rbd_client_list_lock);
307 list_add_tail(&rbdc->node, &rbd_client_list);
308 spin_unlock(&rbd_client_list_lock);
310 mutex_unlock(&ctl_mutex);
312 dout("rbd_client_create created %p\n", rbdc);
316 ceph_destroy_client(rbdc->client);
318 mutex_unlock(&ctl_mutex);
322 ceph_destroy_options(opt);
327 * Find a ceph client with specific addr and configuration.
329 static struct rbd_client *__rbd_client_find(struct ceph_options *opt)
331 struct rbd_client *client_node;
333 if (opt->flags & CEPH_OPT_NOSHARE)
336 list_for_each_entry(client_node, &rbd_client_list, node)
337 if (ceph_compare_options(opt, client_node->client) == 0)
350 /* string args above */
353 static match_table_t rbdopt_tokens = {
354 {Opt_notify_timeout, "notify_timeout=%d"},
356 /* string args above */
360 static int parse_rbd_opts_token(char *c, void *private)
362 struct rbd_options *rbdopt = private;
363 substring_t argstr[MAX_OPT_ARGS];
364 int token, intval, ret;
366 token = match_token(c, rbdopt_tokens, argstr);
370 if (token < Opt_last_int) {
371 ret = match_int(&argstr[0], &intval);
373 pr_err("bad mount option arg (not int) "
377 dout("got int token %d val %d\n", token, intval);
378 } else if (token > Opt_last_int && token < Opt_last_string) {
379 dout("got string token %d val %s\n", token,
382 dout("got token %d\n", token);
386 case Opt_notify_timeout:
387 rbdopt->notify_timeout = intval;
396 * Get a ceph client with specific addr and configuration, if one does
397 * not exist create it.
399 static struct rbd_client *rbd_get_client(const char *mon_addr,
403 struct rbd_client *rbdc;
404 struct ceph_options *opt;
405 struct rbd_options *rbd_opts;
407 rbd_opts = kzalloc(sizeof(*rbd_opts), GFP_KERNEL);
409 return ERR_PTR(-ENOMEM);
411 rbd_opts->notify_timeout = RBD_NOTIFY_TIMEOUT_DEFAULT;
413 opt = ceph_parse_options(options, mon_addr,
414 mon_addr + mon_addr_len,
415 parse_rbd_opts_token, rbd_opts);
418 return ERR_CAST(opt);
421 spin_lock(&rbd_client_list_lock);
422 rbdc = __rbd_client_find(opt);
424 /* using an existing client */
425 kref_get(&rbdc->kref);
426 spin_unlock(&rbd_client_list_lock);
428 ceph_destroy_options(opt);
433 spin_unlock(&rbd_client_list_lock);
435 rbdc = rbd_client_create(opt, rbd_opts);
444 * Destroy ceph client
446 * Caller must hold rbd_client_list_lock.
448 static void rbd_client_release(struct kref *kref)
450 struct rbd_client *rbdc = container_of(kref, struct rbd_client, kref);
452 dout("rbd_release_client %p\n", rbdc);
453 spin_lock(&rbd_client_list_lock);
454 list_del(&rbdc->node);
455 spin_unlock(&rbd_client_list_lock);
457 ceph_destroy_client(rbdc->client);
458 kfree(rbdc->rbd_opts);
463 * Drop reference to ceph client node. If it's not referenced anymore, release
466 static void rbd_put_client(struct rbd_device *rbd_dev)
468 kref_put(&rbd_dev->rbd_client->kref, rbd_client_release);
469 rbd_dev->rbd_client = NULL;
473 * Destroy requests collection
475 static void rbd_coll_release(struct kref *kref)
477 struct rbd_req_coll *coll =
478 container_of(kref, struct rbd_req_coll, kref);
480 dout("rbd_coll_release %p\n", coll);
485 * Create a new header structure, translate header format from the on-disk
488 static int rbd_header_from_disk(struct rbd_image_header *header,
489 struct rbd_image_header_ondisk *ondisk,
495 if (memcmp(ondisk, RBD_HEADER_TEXT, sizeof(RBD_HEADER_TEXT)))
498 snap_count = le32_to_cpu(ondisk->snap_count);
499 if (snap_count > (UINT_MAX - sizeof(struct ceph_snap_context))
502 header->snapc = kmalloc(sizeof(struct ceph_snap_context) +
503 snap_count * sizeof (*ondisk),
508 header->snap_names_len = le64_to_cpu(ondisk->snap_names_len);
510 header->snap_names = kmalloc(header->snap_names_len,
512 if (!header->snap_names)
514 header->snap_sizes = kmalloc(snap_count * sizeof(u64),
516 if (!header->snap_sizes)
519 header->snap_names = NULL;
520 header->snap_sizes = NULL;
522 memcpy(header->block_name, ondisk->block_name,
523 sizeof(ondisk->block_name));
525 header->image_size = le64_to_cpu(ondisk->image_size);
526 header->obj_order = ondisk->options.order;
527 header->crypt_type = ondisk->options.crypt_type;
528 header->comp_type = ondisk->options.comp_type;
530 atomic_set(&header->snapc->nref, 1);
531 header->snap_seq = le64_to_cpu(ondisk->snap_seq);
532 header->snapc->num_snaps = snap_count;
533 header->total_snaps = snap_count;
535 if (snap_count && allocated_snaps == snap_count) {
536 for (i = 0; i < snap_count; i++) {
537 header->snapc->snaps[i] =
538 le64_to_cpu(ondisk->snaps[i].id);
539 header->snap_sizes[i] =
540 le64_to_cpu(ondisk->snaps[i].image_size);
543 /* copy snapshot names */
544 memcpy(header->snap_names, &ondisk->snaps[i],
545 header->snap_names_len);
551 kfree(header->snap_names);
553 kfree(header->snapc);
557 static int snap_index(struct rbd_image_header *header, int snap_num)
559 return header->total_snaps - snap_num;
562 static u64 cur_snap_id(struct rbd_device *rbd_dev)
564 struct rbd_image_header *header = &rbd_dev->header;
566 if (!rbd_dev->cur_snap)
569 return header->snapc->snaps[snap_index(header, rbd_dev->cur_snap)];
572 static int snap_by_name(struct rbd_image_header *header, const char *snap_name,
576 char *p = header->snap_names;
578 for (i = 0; i < header->total_snaps; i++) {
579 if (!strcmp(snap_name, p)) {
581 /* Found it. Pass back its id and/or size */
584 *seq = header->snapc->snaps[i];
586 *size = header->snap_sizes[i];
589 p += strlen(p) + 1; /* Skip ahead to the next name */
594 static int rbd_header_set_snap(struct rbd_device *dev, u64 *size)
596 struct rbd_image_header *header = &dev->header;
597 struct ceph_snap_context *snapc = header->snapc;
600 BUILD_BUG_ON(sizeof (dev->snap_name) < sizeof (RBD_SNAP_HEAD_NAME));
602 down_write(&dev->header_rwsem);
604 if (!memcmp(dev->snap_name, RBD_SNAP_HEAD_NAME,
605 sizeof (RBD_SNAP_HEAD_NAME))) {
606 if (header->total_snaps)
607 snapc->seq = header->snap_seq;
613 *size = header->image_size;
615 ret = snap_by_name(header, dev->snap_name, &snapc->seq, size);
619 dev->cur_snap = header->total_snaps - ret;
625 up_write(&dev->header_rwsem);
629 static void rbd_header_free(struct rbd_image_header *header)
631 kfree(header->snapc);
632 kfree(header->snap_names);
633 kfree(header->snap_sizes);
637 * get the actual striped segment name, offset and length
639 static u64 rbd_get_segment(struct rbd_image_header *header,
640 const char *block_name,
642 char *seg_name, u64 *segofs)
644 u64 seg = ofs >> header->obj_order;
647 snprintf(seg_name, RBD_MAX_SEG_NAME_LEN,
648 "%s.%012llx", block_name, seg);
650 ofs = ofs & ((1 << header->obj_order) - 1);
651 len = min_t(u64, len, (1 << header->obj_order) - ofs);
659 static int rbd_get_num_segments(struct rbd_image_header *header,
662 u64 start_seg = ofs >> header->obj_order;
663 u64 end_seg = (ofs + len - 1) >> header->obj_order;
664 return end_seg - start_seg + 1;
668 * returns the size of an object in the image
670 static u64 rbd_obj_bytes(struct rbd_image_header *header)
672 return 1 << header->obj_order;
679 static void bio_chain_put(struct bio *chain)
685 chain = chain->bi_next;
691 * zeros a bio chain, starting at specific offset
693 static void zero_bio_chain(struct bio *chain, int start_ofs)
702 bio_for_each_segment(bv, chain, i) {
703 if (pos + bv->bv_len > start_ofs) {
704 int remainder = max(start_ofs - pos, 0);
705 buf = bvec_kmap_irq(bv, &flags);
706 memset(buf + remainder, 0,
707 bv->bv_len - remainder);
708 bvec_kunmap_irq(buf, &flags);
713 chain = chain->bi_next;
718 * bio_chain_clone - clone a chain of bios up to a certain length.
719 * might return a bio_pair that will need to be released.
721 static struct bio *bio_chain_clone(struct bio **old, struct bio **next,
722 struct bio_pair **bp,
723 int len, gfp_t gfpmask)
725 struct bio *tmp, *old_chain = *old, *new_chain = NULL, *tail = NULL;
729 bio_pair_release(*bp);
733 while (old_chain && (total < len)) {
734 tmp = bio_kmalloc(gfpmask, old_chain->bi_max_vecs);
738 if (total + old_chain->bi_size > len) {
742 * this split can only happen with a single paged bio,
743 * split_bio will BUG_ON if this is not the case
745 dout("bio_chain_clone split! total=%d remaining=%d"
747 (int)total, (int)len-total,
748 (int)old_chain->bi_size);
750 /* split the bio. We'll release it either in the next
751 call, or it will have to be released outside */
752 bp = bio_split(old_chain, (len - total) / SECTOR_SIZE);
756 __bio_clone(tmp, &bp->bio1);
760 __bio_clone(tmp, old_chain);
761 *next = old_chain->bi_next;
765 gfpmask &= ~__GFP_WAIT;
769 new_chain = tail = tmp;
774 old_chain = old_chain->bi_next;
776 total += tmp->bi_size;
782 tail->bi_next = NULL;
789 dout("bio_chain_clone with err\n");
790 bio_chain_put(new_chain);
795 * helpers for osd request op vectors.
797 static int rbd_create_rw_ops(struct ceph_osd_req_op **ops,
802 *ops = kzalloc(sizeof(struct ceph_osd_req_op) * (num_ops + 1),
806 (*ops)[0].op = opcode;
808 * op extent offset and length will be set later on
809 * in calc_raw_layout()
811 (*ops)[0].payload_len = payload_len;
815 static void rbd_destroy_ops(struct ceph_osd_req_op *ops)
820 static void rbd_coll_end_req_index(struct request *rq,
821 struct rbd_req_coll *coll,
825 struct request_queue *q;
828 dout("rbd_coll_end_req_index %p index %d ret %d len %lld\n",
829 coll, index, ret, len);
835 blk_end_request(rq, ret, len);
841 spin_lock_irq(q->queue_lock);
842 coll->status[index].done = 1;
843 coll->status[index].rc = ret;
844 coll->status[index].bytes = len;
845 max = min = coll->num_done;
846 while (max < coll->total && coll->status[max].done)
849 for (i = min; i<max; i++) {
850 __blk_end_request(rq, coll->status[i].rc,
851 coll->status[i].bytes);
853 kref_put(&coll->kref, rbd_coll_release);
855 spin_unlock_irq(q->queue_lock);
858 static void rbd_coll_end_req(struct rbd_request *req,
861 rbd_coll_end_req_index(req->rq, req->coll, req->coll_index, ret, len);
865 * Send ceph osd request
867 static int rbd_do_request(struct request *rq,
868 struct rbd_device *dev,
869 struct ceph_snap_context *snapc,
871 const char *obj, u64 ofs, u64 len,
876 struct ceph_osd_req_op *ops,
878 struct rbd_req_coll *coll,
880 void (*rbd_cb)(struct ceph_osd_request *req,
881 struct ceph_msg *msg),
882 struct ceph_osd_request **linger_req,
885 struct ceph_osd_request *req;
886 struct ceph_file_layout *layout;
889 struct timespec mtime = CURRENT_TIME;
890 struct rbd_request *req_data;
891 struct ceph_osd_request_head *reqhead;
892 struct ceph_osd_client *osdc;
894 req_data = kzalloc(sizeof(*req_data), GFP_NOIO);
897 rbd_coll_end_req_index(rq, coll, coll_index,
903 req_data->coll = coll;
904 req_data->coll_index = coll_index;
907 dout("rbd_do_request obj=%s ofs=%lld len=%lld\n", obj, len, ofs);
909 down_read(&dev->header_rwsem);
911 osdc = &dev->rbd_client->client->osdc;
912 req = ceph_osdc_alloc_request(osdc, flags, snapc, ops,
913 false, GFP_NOIO, pages, bio);
915 up_read(&dev->header_rwsem);
920 req->r_callback = rbd_cb;
924 req_data->pages = pages;
927 req->r_priv = req_data;
929 reqhead = req->r_request->front.iov_base;
930 reqhead->snapid = cpu_to_le64(CEPH_NOSNAP);
932 strncpy(req->r_oid, obj, sizeof(req->r_oid));
933 req->r_oid_len = strlen(req->r_oid);
935 layout = &req->r_file_layout;
936 memset(layout, 0, sizeof(*layout));
937 layout->fl_stripe_unit = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER);
938 layout->fl_stripe_count = cpu_to_le32(1);
939 layout->fl_object_size = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER);
940 layout->fl_pg_pool = cpu_to_le32(dev->poolid);
941 ceph_calc_raw_layout(osdc, layout, snapid, ofs, &len, &bno,
944 ceph_osdc_build_request(req, ofs, &len,
948 req->r_oid, req->r_oid_len);
949 up_read(&dev->header_rwsem);
952 ceph_osdc_set_request_linger(osdc, req);
956 ret = ceph_osdc_start_request(osdc, req, false);
961 ret = ceph_osdc_wait_request(osdc, req);
963 *ver = le64_to_cpu(req->r_reassert_version.version);
964 dout("reassert_ver=%lld\n",
965 le64_to_cpu(req->r_reassert_version.version));
966 ceph_osdc_put_request(req);
971 bio_chain_put(req_data->bio);
972 ceph_osdc_put_request(req);
974 rbd_coll_end_req(req_data, ret, len);
980 * Ceph osd op callback
982 static void rbd_req_cb(struct ceph_osd_request *req, struct ceph_msg *msg)
984 struct rbd_request *req_data = req->r_priv;
985 struct ceph_osd_reply_head *replyhead;
986 struct ceph_osd_op *op;
992 replyhead = msg->front.iov_base;
993 WARN_ON(le32_to_cpu(replyhead->num_ops) == 0);
994 op = (void *)(replyhead + 1);
995 rc = le32_to_cpu(replyhead->result);
996 bytes = le64_to_cpu(op->extent.length);
997 read_op = (le32_to_cpu(op->op) == CEPH_OSD_OP_READ);
999 dout("rbd_req_cb bytes=%lld readop=%d rc=%d\n", bytes, read_op, rc);
1001 if (rc == -ENOENT && read_op) {
1002 zero_bio_chain(req_data->bio, 0);
1004 } else if (rc == 0 && read_op && bytes < req_data->len) {
1005 zero_bio_chain(req_data->bio, bytes);
1006 bytes = req_data->len;
1009 rbd_coll_end_req(req_data, rc, bytes);
1012 bio_chain_put(req_data->bio);
1014 ceph_osdc_put_request(req);
1018 static void rbd_simple_req_cb(struct ceph_osd_request *req, struct ceph_msg *msg)
1020 ceph_osdc_put_request(req);
1024 * Do a synchronous ceph osd operation
1026 static int rbd_req_sync_op(struct rbd_device *dev,
1027 struct ceph_snap_context *snapc,
1031 struct ceph_osd_req_op *orig_ops,
1036 struct ceph_osd_request **linger_req,
1040 struct page **pages;
1042 struct ceph_osd_req_op *ops = orig_ops;
1045 num_pages = calc_pages_for(ofs , len);
1046 pages = ceph_alloc_page_vector(num_pages, GFP_KERNEL);
1048 return PTR_ERR(pages);
1051 payload_len = (flags & CEPH_OSD_FLAG_WRITE ? len : 0);
1052 ret = rbd_create_rw_ops(&ops, 1, opcode, payload_len);
1056 if ((flags & CEPH_OSD_FLAG_WRITE) && buf) {
1057 ret = ceph_copy_to_page_vector(pages, buf, ofs, len);
1063 ret = rbd_do_request(NULL, dev, snapc, snapid,
1064 obj, ofs, len, NULL,
1075 if ((flags & CEPH_OSD_FLAG_READ) && buf)
1076 ret = ceph_copy_from_page_vector(pages, buf, ofs, ret);
1080 rbd_destroy_ops(ops);
1082 ceph_release_page_vector(pages, num_pages);
1087 * Do an asynchronous ceph osd operation
1089 static int rbd_do_op(struct request *rq,
1090 struct rbd_device *rbd_dev ,
1091 struct ceph_snap_context *snapc,
1093 int opcode, int flags, int num_reply,
1096 struct rbd_req_coll *coll,
1103 struct ceph_osd_req_op *ops;
1106 seg_name = kmalloc(RBD_MAX_SEG_NAME_LEN + 1, GFP_NOIO);
1110 seg_len = rbd_get_segment(&rbd_dev->header,
1111 rbd_dev->header.block_name,
1113 seg_name, &seg_ofs);
1115 payload_len = (flags & CEPH_OSD_FLAG_WRITE ? seg_len : 0);
1117 ret = rbd_create_rw_ops(&ops, 1, opcode, payload_len);
1121 /* we've taken care of segment sizes earlier when we
1122 cloned the bios. We should never have a segment
1123 truncated at this point */
1124 BUG_ON(seg_len < len);
1126 ret = rbd_do_request(rq, rbd_dev, snapc, snapid,
1127 seg_name, seg_ofs, seg_len,
1134 rbd_req_cb, 0, NULL);
1136 rbd_destroy_ops(ops);
1143 * Request async osd write
1145 static int rbd_req_write(struct request *rq,
1146 struct rbd_device *rbd_dev,
1147 struct ceph_snap_context *snapc,
1150 struct rbd_req_coll *coll,
1153 return rbd_do_op(rq, rbd_dev, snapc, CEPH_NOSNAP,
1155 CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK,
1157 ofs, len, bio, coll, coll_index);
1161 * Request async osd read
1163 static int rbd_req_read(struct request *rq,
1164 struct rbd_device *rbd_dev,
1168 struct rbd_req_coll *coll,
1171 return rbd_do_op(rq, rbd_dev, NULL,
1172 (snapid ? snapid : CEPH_NOSNAP),
1176 ofs, len, bio, coll, coll_index);
1180 * Request sync osd read
1182 static int rbd_req_sync_read(struct rbd_device *dev,
1183 struct ceph_snap_context *snapc,
1190 return rbd_req_sync_op(dev, NULL,
1191 (snapid ? snapid : CEPH_NOSNAP),
1195 1, obj, ofs, len, buf, NULL, ver);
1199 * Request sync osd watch
1201 static int rbd_req_sync_notify_ack(struct rbd_device *dev,
1206 struct ceph_osd_req_op *ops;
1207 struct page **pages = NULL;
1210 ret = rbd_create_rw_ops(&ops, 1, CEPH_OSD_OP_NOTIFY_ACK, 0);
1214 ops[0].watch.ver = cpu_to_le64(dev->header.obj_version);
1215 ops[0].watch.cookie = notify_id;
1216 ops[0].watch.flag = 0;
1218 ret = rbd_do_request(NULL, dev, NULL, CEPH_NOSNAP,
1225 rbd_simple_req_cb, 0, NULL);
1227 rbd_destroy_ops(ops);
1231 static void rbd_watch_cb(u64 ver, u64 notify_id, u8 opcode, void *data)
1233 struct rbd_device *dev = (struct rbd_device *)data;
1239 dout("rbd_watch_cb %s notify_id=%lld opcode=%d\n", dev->obj_md_name,
1240 notify_id, (int)opcode);
1241 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
1242 rc = __rbd_update_snaps(dev);
1243 mutex_unlock(&ctl_mutex);
1245 pr_warning(RBD_DRV_NAME "%d got notification but failed to "
1246 " update snaps: %d\n", dev->major, rc);
1248 rbd_req_sync_notify_ack(dev, ver, notify_id, dev->obj_md_name);
1252 * Request sync osd watch
1254 static int rbd_req_sync_watch(struct rbd_device *dev,
1258 struct ceph_osd_req_op *ops;
1259 struct ceph_osd_client *osdc = &dev->rbd_client->client->osdc;
1261 int ret = rbd_create_rw_ops(&ops, 1, CEPH_OSD_OP_WATCH, 0);
1265 ret = ceph_osdc_create_event(osdc, rbd_watch_cb, 0,
1266 (void *)dev, &dev->watch_event);
1270 ops[0].watch.ver = cpu_to_le64(ver);
1271 ops[0].watch.cookie = cpu_to_le64(dev->watch_event->cookie);
1272 ops[0].watch.flag = 1;
1274 ret = rbd_req_sync_op(dev, NULL,
1277 CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK,
1280 &dev->watch_request, NULL);
1285 rbd_destroy_ops(ops);
1289 ceph_osdc_cancel_event(dev->watch_event);
1290 dev->watch_event = NULL;
1292 rbd_destroy_ops(ops);
1297 * Request sync osd unwatch
1299 static int rbd_req_sync_unwatch(struct rbd_device *dev,
1302 struct ceph_osd_req_op *ops;
1304 int ret = rbd_create_rw_ops(&ops, 1, CEPH_OSD_OP_WATCH, 0);
1308 ops[0].watch.ver = 0;
1309 ops[0].watch.cookie = cpu_to_le64(dev->watch_event->cookie);
1310 ops[0].watch.flag = 0;
1312 ret = rbd_req_sync_op(dev, NULL,
1315 CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK,
1317 1, obj, 0, 0, NULL, NULL, NULL);
1319 rbd_destroy_ops(ops);
1320 ceph_osdc_cancel_event(dev->watch_event);
1321 dev->watch_event = NULL;
1325 struct rbd_notify_info {
1326 struct rbd_device *dev;
1329 static void rbd_notify_cb(u64 ver, u64 notify_id, u8 opcode, void *data)
1331 struct rbd_device *dev = (struct rbd_device *)data;
1335 dout("rbd_notify_cb %s notify_id=%lld opcode=%d\n", dev->obj_md_name,
1336 notify_id, (int)opcode);
1340 * Request sync osd notify
1342 static int rbd_req_sync_notify(struct rbd_device *dev,
1345 struct ceph_osd_req_op *ops;
1346 struct ceph_osd_client *osdc = &dev->rbd_client->client->osdc;
1347 struct ceph_osd_event *event;
1348 struct rbd_notify_info info;
1349 int payload_len = sizeof(u32) + sizeof(u32);
1352 ret = rbd_create_rw_ops(&ops, 1, CEPH_OSD_OP_NOTIFY, payload_len);
1358 ret = ceph_osdc_create_event(osdc, rbd_notify_cb, 1,
1359 (void *)&info, &event);
1363 ops[0].watch.ver = 1;
1364 ops[0].watch.flag = 1;
1365 ops[0].watch.cookie = event->cookie;
1366 ops[0].watch.prot_ver = RADOS_NOTIFY_VER;
1367 ops[0].watch.timeout = 12;
1369 ret = rbd_req_sync_op(dev, NULL,
1372 CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK,
1374 1, obj, 0, 0, NULL, NULL, NULL);
1378 ret = ceph_osdc_wait_event(event, CEPH_OSD_TIMEOUT_DEFAULT);
1379 dout("ceph_osdc_wait_event returned %d\n", ret);
1380 rbd_destroy_ops(ops);
1384 ceph_osdc_cancel_event(event);
1386 rbd_destroy_ops(ops);
1391 * Request sync osd read
1393 static int rbd_req_sync_exec(struct rbd_device *dev,
1401 struct ceph_osd_req_op *ops;
1402 int cls_len = strlen(cls);
1403 int method_len = strlen(method);
1404 int ret = rbd_create_rw_ops(&ops, 1, CEPH_OSD_OP_CALL,
1405 cls_len + method_len + len);
1409 ops[0].cls.class_name = cls;
1410 ops[0].cls.class_len = (__u8)cls_len;
1411 ops[0].cls.method_name = method;
1412 ops[0].cls.method_len = (__u8)method_len;
1413 ops[0].cls.argc = 0;
1414 ops[0].cls.indata = data;
1415 ops[0].cls.indata_len = len;
1417 ret = rbd_req_sync_op(dev, NULL,
1420 CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK,
1422 1, obj, 0, 0, NULL, NULL, ver);
1424 rbd_destroy_ops(ops);
1426 dout("cls_exec returned %d\n", ret);
1430 static struct rbd_req_coll *rbd_alloc_coll(int num_reqs)
1432 struct rbd_req_coll *coll =
1433 kzalloc(sizeof(struct rbd_req_coll) +
1434 sizeof(struct rbd_req_status) * num_reqs,
1439 coll->total = num_reqs;
1440 kref_init(&coll->kref);
1445 * block device queue callback
1447 static void rbd_rq_fn(struct request_queue *q)
1449 struct rbd_device *rbd_dev = q->queuedata;
1451 struct bio_pair *bp = NULL;
1453 while ((rq = blk_fetch_request(q))) {
1455 struct bio *rq_bio, *next_bio = NULL;
1457 int size, op_size = 0;
1459 int num_segs, cur_seg = 0;
1460 struct rbd_req_coll *coll;
1462 /* peek at request from block layer */
1466 dout("fetched request\n");
1468 /* filter out block requests we don't understand */
1469 if ((rq->cmd_type != REQ_TYPE_FS)) {
1470 __blk_end_request_all(rq, 0);
1474 /* deduce our operation (read, write) */
1475 do_write = (rq_data_dir(rq) == WRITE);
1477 size = blk_rq_bytes(rq);
1478 ofs = blk_rq_pos(rq) * SECTOR_SIZE;
1480 if (do_write && rbd_dev->read_only) {
1481 __blk_end_request_all(rq, -EROFS);
1485 spin_unlock_irq(q->queue_lock);
1487 dout("%s 0x%x bytes at 0x%llx\n",
1488 do_write ? "write" : "read",
1489 size, blk_rq_pos(rq) * SECTOR_SIZE);
1491 num_segs = rbd_get_num_segments(&rbd_dev->header, ofs, size);
1492 coll = rbd_alloc_coll(num_segs);
1494 spin_lock_irq(q->queue_lock);
1495 __blk_end_request_all(rq, -ENOMEM);
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.block_name,
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,
1519 rbd_dev->header.snapc,
1524 rbd_req_read(rq, rbd_dev,
1525 cur_snap_id(rbd_dev),
1537 kref_put(&coll->kref, rbd_coll_release);
1540 bio_pair_release(bp);
1541 spin_lock_irq(q->queue_lock);
1546 * a queue callback. Makes sure that we don't create a bio that spans across
1547 * multiple osd objects. One exception would be with a single page bios,
1548 * which we handle later at bio_chain_clone
1550 static int rbd_merge_bvec(struct request_queue *q, struct bvec_merge_data *bmd,
1551 struct bio_vec *bvec)
1553 struct rbd_device *rbd_dev = q->queuedata;
1554 unsigned int chunk_sectors;
1556 unsigned int bio_sectors;
1559 chunk_sectors = 1 << (rbd_dev->header.obj_order - SECTOR_SHIFT);
1560 sector = bmd->bi_sector + get_start_sect(bmd->bi_bdev);
1561 bio_sectors = bmd->bi_size >> SECTOR_SHIFT;
1563 max = (chunk_sectors - ((sector & (chunk_sectors - 1))
1564 + bio_sectors)) << SECTOR_SHIFT;
1566 max = 0; /* bio_add cannot handle a negative return */
1567 if (max <= bvec->bv_len && bio_sectors == 0)
1568 return bvec->bv_len;
1572 static void rbd_free_disk(struct rbd_device *rbd_dev)
1574 struct gendisk *disk = rbd_dev->disk;
1579 rbd_header_free(&rbd_dev->header);
1581 if (disk->flags & GENHD_FL_UP)
1584 blk_cleanup_queue(disk->queue);
1589 * reload the ondisk the header
1591 static int rbd_read_header(struct rbd_device *rbd_dev,
1592 struct rbd_image_header *header)
1595 struct rbd_image_header_ondisk *dh;
1601 * First reads the fixed-size header to determine the number
1602 * of snapshots, then re-reads it, along with all snapshot
1603 * records as well as their stored names.
1607 dh = kmalloc(len, GFP_KERNEL);
1611 rc = rbd_req_sync_read(rbd_dev,
1613 rbd_dev->obj_md_name,
1619 rc = rbd_header_from_disk(header, dh, snap_count, GFP_KERNEL);
1622 pr_warning("unrecognized header format"
1623 " for image %s", rbd_dev->obj);
1627 if (snap_count == header->total_snaps)
1630 snap_count = header->total_snaps;
1631 len = sizeof (*dh) +
1632 snap_count * sizeof(struct rbd_image_snap_ondisk) +
1633 header->snap_names_len;
1635 rbd_header_free(header);
1638 header->obj_version = ver;
1648 static int rbd_header_add_snap(struct rbd_device *dev,
1649 const char *snap_name,
1652 int name_len = strlen(snap_name);
1657 struct ceph_mon_client *monc;
1659 /* we should create a snapshot only if we're pointing at the head */
1663 monc = &dev->rbd_client->client->monc;
1664 ret = ceph_monc_create_snapid(monc, dev->poolid, &new_snapid);
1665 dout("created snapid=%lld\n", new_snapid);
1669 data = kmalloc(name_len + 16, gfp_flags);
1674 e = data + name_len + 16;
1676 ceph_encode_string_safe(&p, e, snap_name, name_len, bad);
1677 ceph_encode_64_safe(&p, e, new_snapid, bad);
1679 ret = rbd_req_sync_exec(dev, dev->obj_md_name, "rbd", "snap_add",
1680 data, p - data, &ver);
1687 dev->header.snapc->seq = new_snapid;
1694 static void __rbd_remove_all_snaps(struct rbd_device *rbd_dev)
1696 struct rbd_snap *snap;
1698 while (!list_empty(&rbd_dev->snaps)) {
1699 snap = list_first_entry(&rbd_dev->snaps, struct rbd_snap, node);
1700 __rbd_remove_snap_dev(rbd_dev, snap);
1705 * only read the first part of the ondisk header, without the snaps info
1707 static int __rbd_update_snaps(struct rbd_device *rbd_dev)
1710 struct rbd_image_header h;
1714 ret = rbd_read_header(rbd_dev, &h);
1719 set_capacity(rbd_dev->disk, h.image_size / SECTOR_SIZE);
1721 down_write(&rbd_dev->header_rwsem);
1723 snap_seq = rbd_dev->header.snapc->seq;
1724 if (rbd_dev->header.total_snaps &&
1725 rbd_dev->header.snapc->snaps[0] == snap_seq)
1726 /* pointing at the head, will need to follow that
1730 kfree(rbd_dev->header.snapc);
1731 kfree(rbd_dev->header.snap_names);
1732 kfree(rbd_dev->header.snap_sizes);
1734 rbd_dev->header.total_snaps = h.total_snaps;
1735 rbd_dev->header.snapc = h.snapc;
1736 rbd_dev->header.snap_names = h.snap_names;
1737 rbd_dev->header.snap_names_len = h.snap_names_len;
1738 rbd_dev->header.snap_sizes = h.snap_sizes;
1740 rbd_dev->header.snapc->seq = rbd_dev->header.snapc->snaps[0];
1742 rbd_dev->header.snapc->seq = snap_seq;
1744 ret = __rbd_init_snaps_header(rbd_dev);
1746 up_write(&rbd_dev->header_rwsem);
1751 static int rbd_init_disk(struct rbd_device *rbd_dev)
1753 struct gendisk *disk;
1754 struct request_queue *q;
1759 /* contact OSD, request size info about the object being mapped */
1760 rc = rbd_read_header(rbd_dev, &rbd_dev->header);
1764 /* no need to lock here, as rbd_dev is not registered yet */
1765 rc = __rbd_init_snaps_header(rbd_dev);
1769 rc = rbd_header_set_snap(rbd_dev, &total_size);
1773 /* create gendisk info */
1775 disk = alloc_disk(RBD_MINORS_PER_MAJOR);
1779 snprintf(disk->disk_name, sizeof(disk->disk_name), RBD_DRV_NAME "%d",
1781 disk->major = rbd_dev->major;
1782 disk->first_minor = 0;
1783 disk->fops = &rbd_bd_ops;
1784 disk->private_data = rbd_dev;
1788 q = blk_init_queue(rbd_rq_fn, &rbd_dev->lock);
1792 /* We use the default size, but let's be explicit about it. */
1793 blk_queue_physical_block_size(q, SECTOR_SIZE);
1795 /* set io sizes to object size */
1796 segment_size = rbd_obj_bytes(&rbd_dev->header);
1797 blk_queue_max_hw_sectors(q, segment_size / SECTOR_SIZE);
1798 blk_queue_max_segment_size(q, segment_size);
1799 blk_queue_io_min(q, segment_size);
1800 blk_queue_io_opt(q, segment_size);
1802 blk_queue_merge_bvec(q, rbd_merge_bvec);
1805 q->queuedata = rbd_dev;
1807 rbd_dev->disk = disk;
1810 /* finally, announce the disk to the world */
1811 set_capacity(disk, total_size / SECTOR_SIZE);
1814 pr_info("%s: added with size 0x%llx\n",
1815 disk->disk_name, (unsigned long long)total_size);
1828 static struct rbd_device *dev_to_rbd_dev(struct device *dev)
1830 return container_of(dev, struct rbd_device, dev);
1833 static ssize_t rbd_size_show(struct device *dev,
1834 struct device_attribute *attr, char *buf)
1836 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
1838 return sprintf(buf, "%llu\n", (unsigned long long)rbd_dev->header.image_size);
1841 static ssize_t rbd_major_show(struct device *dev,
1842 struct device_attribute *attr, char *buf)
1844 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
1846 return sprintf(buf, "%d\n", rbd_dev->major);
1849 static ssize_t rbd_client_id_show(struct device *dev,
1850 struct device_attribute *attr, char *buf)
1852 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
1854 return sprintf(buf, "client%lld\n",
1855 ceph_client_id(rbd_dev->rbd_client->client));
1858 static ssize_t rbd_pool_show(struct device *dev,
1859 struct device_attribute *attr, char *buf)
1861 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
1863 return sprintf(buf, "%s\n", rbd_dev->pool_name);
1866 static ssize_t rbd_name_show(struct device *dev,
1867 struct device_attribute *attr, char *buf)
1869 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
1871 return sprintf(buf, "%s\n", rbd_dev->obj);
1874 static ssize_t rbd_snap_show(struct device *dev,
1875 struct device_attribute *attr,
1878 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
1880 return sprintf(buf, "%s\n", rbd_dev->snap_name);
1883 static ssize_t rbd_image_refresh(struct device *dev,
1884 struct device_attribute *attr,
1888 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
1892 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
1894 rc = __rbd_update_snaps(rbd_dev);
1898 mutex_unlock(&ctl_mutex);
1902 static DEVICE_ATTR(size, S_IRUGO, rbd_size_show, NULL);
1903 static DEVICE_ATTR(major, S_IRUGO, rbd_major_show, NULL);
1904 static DEVICE_ATTR(client_id, S_IRUGO, rbd_client_id_show, NULL);
1905 static DEVICE_ATTR(pool, S_IRUGO, rbd_pool_show, NULL);
1906 static DEVICE_ATTR(name, S_IRUGO, rbd_name_show, NULL);
1907 static DEVICE_ATTR(refresh, S_IWUSR, NULL, rbd_image_refresh);
1908 static DEVICE_ATTR(current_snap, S_IRUGO, rbd_snap_show, NULL);
1909 static DEVICE_ATTR(create_snap, S_IWUSR, NULL, rbd_snap_add);
1911 static struct attribute *rbd_attrs[] = {
1912 &dev_attr_size.attr,
1913 &dev_attr_major.attr,
1914 &dev_attr_client_id.attr,
1915 &dev_attr_pool.attr,
1916 &dev_attr_name.attr,
1917 &dev_attr_current_snap.attr,
1918 &dev_attr_refresh.attr,
1919 &dev_attr_create_snap.attr,
1923 static struct attribute_group rbd_attr_group = {
1927 static const struct attribute_group *rbd_attr_groups[] = {
1932 static void rbd_sysfs_dev_release(struct device *dev)
1936 static struct device_type rbd_device_type = {
1938 .groups = rbd_attr_groups,
1939 .release = rbd_sysfs_dev_release,
1947 static ssize_t rbd_snap_size_show(struct device *dev,
1948 struct device_attribute *attr,
1951 struct rbd_snap *snap = container_of(dev, struct rbd_snap, dev);
1953 return sprintf(buf, "%zd\n", snap->size);
1956 static ssize_t rbd_snap_id_show(struct device *dev,
1957 struct device_attribute *attr,
1960 struct rbd_snap *snap = container_of(dev, struct rbd_snap, dev);
1962 return sprintf(buf, "%llu\n", (unsigned long long) snap->id);
1965 static DEVICE_ATTR(snap_size, S_IRUGO, rbd_snap_size_show, NULL);
1966 static DEVICE_ATTR(snap_id, S_IRUGO, rbd_snap_id_show, NULL);
1968 static struct attribute *rbd_snap_attrs[] = {
1969 &dev_attr_snap_size.attr,
1970 &dev_attr_snap_id.attr,
1974 static struct attribute_group rbd_snap_attr_group = {
1975 .attrs = rbd_snap_attrs,
1978 static void rbd_snap_dev_release(struct device *dev)
1980 struct rbd_snap *snap = container_of(dev, struct rbd_snap, dev);
1985 static const struct attribute_group *rbd_snap_attr_groups[] = {
1986 &rbd_snap_attr_group,
1990 static struct device_type rbd_snap_device_type = {
1991 .groups = rbd_snap_attr_groups,
1992 .release = rbd_snap_dev_release,
1995 static void __rbd_remove_snap_dev(struct rbd_device *rbd_dev,
1996 struct rbd_snap *snap)
1998 list_del(&snap->node);
1999 device_unregister(&snap->dev);
2002 static int rbd_register_snap_dev(struct rbd_device *rbd_dev,
2003 struct rbd_snap *snap,
2004 struct device *parent)
2006 struct device *dev = &snap->dev;
2009 dev->type = &rbd_snap_device_type;
2010 dev->parent = parent;
2011 dev->release = rbd_snap_dev_release;
2012 dev_set_name(dev, "snap_%s", snap->name);
2013 ret = device_register(dev);
2018 static int __rbd_add_snap_dev(struct rbd_device *rbd_dev,
2019 int i, const char *name,
2020 struct rbd_snap **snapp)
2023 struct rbd_snap *snap = kzalloc(sizeof(*snap), GFP_KERNEL);
2026 snap->name = kstrdup(name, GFP_KERNEL);
2027 snap->size = rbd_dev->header.snap_sizes[i];
2028 snap->id = rbd_dev->header.snapc->snaps[i];
2029 if (device_is_registered(&rbd_dev->dev)) {
2030 ret = rbd_register_snap_dev(rbd_dev, snap,
2044 * search for the previous snap in a null delimited string list
2046 const char *rbd_prev_snap_name(const char *name, const char *start)
2048 if (name < start + 2)
2061 * compare the old list of snapshots that we have to what's in the header
2062 * and update it accordingly. Note that the header holds the snapshots
2063 * in a reverse order (from newest to oldest) and we need to go from
2064 * older to new so that we don't get a duplicate snap name when
2065 * doing the process (e.g., removed snapshot and recreated a new
2066 * one with the same name.
2068 static int __rbd_init_snaps_header(struct rbd_device *rbd_dev)
2070 const char *name, *first_name;
2071 int i = rbd_dev->header.total_snaps;
2072 struct rbd_snap *snap, *old_snap = NULL;
2074 struct list_head *p, *n;
2076 first_name = rbd_dev->header.snap_names;
2077 name = first_name + rbd_dev->header.snap_names_len;
2079 list_for_each_prev_safe(p, n, &rbd_dev->snaps) {
2082 old_snap = list_entry(p, struct rbd_snap, node);
2085 cur_id = rbd_dev->header.snapc->snaps[i - 1];
2087 if (!i || old_snap->id < cur_id) {
2088 /* old_snap->id was skipped, thus was removed */
2089 __rbd_remove_snap_dev(rbd_dev, old_snap);
2092 if (old_snap->id == cur_id) {
2093 /* we have this snapshot already */
2095 name = rbd_prev_snap_name(name, first_name);
2099 i--, name = rbd_prev_snap_name(name, first_name)) {
2104 cur_id = rbd_dev->header.snapc->snaps[i];
2105 /* snapshot removal? handle it above */
2106 if (cur_id >= old_snap->id)
2108 /* a new snapshot */
2109 ret = __rbd_add_snap_dev(rbd_dev, i - 1, name, &snap);
2113 /* note that we add it backward so using n and not p */
2114 list_add(&snap->node, n);
2118 /* we're done going over the old snap list, just add what's left */
2119 for (; i > 0; i--) {
2120 name = rbd_prev_snap_name(name, first_name);
2125 ret = __rbd_add_snap_dev(rbd_dev, i - 1, name, &snap);
2128 list_add(&snap->node, &rbd_dev->snaps);
2134 static int rbd_bus_add_dev(struct rbd_device *rbd_dev)
2138 struct rbd_snap *snap;
2140 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
2141 dev = &rbd_dev->dev;
2143 dev->bus = &rbd_bus_type;
2144 dev->type = &rbd_device_type;
2145 dev->parent = &rbd_root_dev;
2146 dev->release = rbd_dev_release;
2147 dev_set_name(dev, "%d", rbd_dev->id);
2148 ret = device_register(dev);
2152 list_for_each_entry(snap, &rbd_dev->snaps, node) {
2153 ret = rbd_register_snap_dev(rbd_dev, snap,
2159 mutex_unlock(&ctl_mutex);
2163 static void rbd_bus_del_dev(struct rbd_device *rbd_dev)
2165 device_unregister(&rbd_dev->dev);
2168 static int rbd_init_watch_dev(struct rbd_device *rbd_dev)
2173 ret = rbd_req_sync_watch(rbd_dev, rbd_dev->obj_md_name,
2174 rbd_dev->header.obj_version);
2175 if (ret == -ERANGE) {
2176 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
2177 rc = __rbd_update_snaps(rbd_dev);
2178 mutex_unlock(&ctl_mutex);
2182 } while (ret == -ERANGE);
2187 static atomic64_t rbd_id_max = ATOMIC64_INIT(0);
2190 * Get a unique rbd identifier for the given new rbd_dev, and add
2191 * the rbd_dev to the global list. The minimum rbd id is 1.
2193 static void rbd_id_get(struct rbd_device *rbd_dev)
2195 rbd_dev->id = atomic64_inc_return(&rbd_id_max);
2197 spin_lock(&rbd_dev_list_lock);
2198 list_add_tail(&rbd_dev->node, &rbd_dev_list);
2199 spin_unlock(&rbd_dev_list_lock);
2203 * Remove an rbd_dev from the global list, and record that its
2204 * identifier is no longer in use.
2206 static void rbd_id_put(struct rbd_device *rbd_dev)
2208 struct list_head *tmp;
2209 int rbd_id = rbd_dev->id;
2214 spin_lock(&rbd_dev_list_lock);
2215 list_del_init(&rbd_dev->node);
2218 * If the id being "put" is not the current maximum, there
2219 * is nothing special we need to do.
2221 if (rbd_id != atomic64_read(&rbd_id_max)) {
2222 spin_unlock(&rbd_dev_list_lock);
2227 * We need to update the current maximum id. Search the
2228 * list to find out what it is. We're more likely to find
2229 * the maximum at the end, so search the list backward.
2232 list_for_each_prev(tmp, &rbd_dev_list) {
2233 struct rbd_device *rbd_dev;
2235 rbd_dev = list_entry(tmp, struct rbd_device, node);
2236 if (rbd_id > max_id)
2239 spin_unlock(&rbd_dev_list_lock);
2242 * The max id could have been updated by rbd_id_get(), in
2243 * which case it now accurately reflects the new maximum.
2244 * Be careful not to overwrite the maximum value in that
2247 atomic64_cmpxchg(&rbd_id_max, rbd_id, max_id);
2251 * Skips over white space at *buf, and updates *buf to point to the
2252 * first found non-space character (if any). Returns the length of
2253 * the token (string of non-white space characters) found. Note
2254 * that *buf must be terminated with '\0'.
2256 static inline size_t next_token(const char **buf)
2259 * These are the characters that produce nonzero for
2260 * isspace() in the "C" and "POSIX" locales.
2262 const char *spaces = " \f\n\r\t\v";
2264 *buf += strspn(*buf, spaces); /* Find start of token */
2266 return strcspn(*buf, spaces); /* Return token length */
2270 * Finds the next token in *buf, and if the provided token buffer is
2271 * big enough, copies the found token into it. The result, if
2272 * copied, is guaranteed to be terminated with '\0'. Note that *buf
2273 * must be terminated with '\0' on entry.
2275 * Returns the length of the token found (not including the '\0').
2276 * Return value will be 0 if no token is found, and it will be >=
2277 * token_size if the token would not fit.
2279 * The *buf pointer will be updated to point beyond the end of the
2280 * found token. Note that this occurs even if the token buffer is
2281 * too small to hold it.
2283 static inline size_t copy_token(const char **buf,
2289 len = next_token(buf);
2290 if (len < token_size) {
2291 memcpy(token, *buf, len);
2292 *(token + len) = '\0';
2300 * This fills in the pool_name, obj, obj_len, snap_name, obj_len,
2301 * rbd_dev, rbd_md_name, and name fields of the given rbd_dev, based
2302 * on the list of monitor addresses and other options provided via
2305 static int rbd_add_parse_args(struct rbd_device *rbd_dev,
2307 const char **mon_addrs,
2308 size_t *mon_addrs_size,
2310 size_t options_size)
2314 /* The first four tokens are required */
2316 len = next_token(&buf);
2319 *mon_addrs_size = len + 1;
2324 len = copy_token(&buf, options, options_size);
2325 if (!len || len >= options_size)
2328 len = copy_token(&buf, rbd_dev->pool_name, sizeof (rbd_dev->pool_name));
2329 if (!len || len >= sizeof (rbd_dev->pool_name))
2332 len = copy_token(&buf, rbd_dev->obj, sizeof (rbd_dev->obj));
2333 if (!len || len >= sizeof (rbd_dev->obj))
2336 /* We have the object length in hand, save it. */
2338 rbd_dev->obj_len = len;
2340 BUILD_BUG_ON(RBD_MAX_MD_NAME_LEN
2341 < RBD_MAX_OBJ_NAME_LEN + sizeof (RBD_SUFFIX));
2342 sprintf(rbd_dev->obj_md_name, "%s%s", rbd_dev->obj, RBD_SUFFIX);
2345 * The snapshot name is optional, but it's an error if it's
2346 * too long. If no snapshot is supplied, fill in the default.
2348 len = copy_token(&buf, rbd_dev->snap_name, sizeof (rbd_dev->snap_name));
2350 memcpy(rbd_dev->snap_name, RBD_SNAP_HEAD_NAME,
2351 sizeof (RBD_SNAP_HEAD_NAME));
2352 else if (len >= sizeof (rbd_dev->snap_name))
2358 static ssize_t rbd_add(struct bus_type *bus,
2362 struct rbd_device *rbd_dev;
2363 const char *mon_addrs = NULL;
2364 size_t mon_addrs_size = 0;
2365 char *options = NULL;
2366 struct ceph_osd_client *osdc;
2369 if (!try_module_get(THIS_MODULE))
2372 rbd_dev = kzalloc(sizeof(*rbd_dev), GFP_KERNEL);
2375 options = kmalloc(count, GFP_KERNEL);
2379 /* static rbd_device initialization */
2380 spin_lock_init(&rbd_dev->lock);
2381 INIT_LIST_HEAD(&rbd_dev->node);
2382 INIT_LIST_HEAD(&rbd_dev->snaps);
2383 init_rwsem(&rbd_dev->header_rwsem);
2385 init_rwsem(&rbd_dev->header_rwsem);
2387 /* generate unique id: find highest unique id, add one */
2388 rbd_id_get(rbd_dev);
2390 /* Fill in the device name, now that we have its id. */
2391 BUILD_BUG_ON(DEV_NAME_LEN
2392 < sizeof (RBD_DRV_NAME) + MAX_INT_FORMAT_WIDTH);
2393 sprintf(rbd_dev->name, "%s%d", RBD_DRV_NAME, rbd_dev->id);
2395 /* parse add command */
2396 rc = rbd_add_parse_args(rbd_dev, buf, &mon_addrs, &mon_addrs_size,
2401 rbd_dev->rbd_client = rbd_get_client(mon_addrs, mon_addrs_size - 1,
2403 if (IS_ERR(rbd_dev->rbd_client)) {
2404 rc = PTR_ERR(rbd_dev->rbd_client);
2409 osdc = &rbd_dev->rbd_client->client->osdc;
2410 rc = ceph_pg_poolid_by_name(osdc->osdmap, rbd_dev->pool_name);
2412 goto err_out_client;
2413 rbd_dev->poolid = rc;
2415 /* register our block device */
2416 rc = register_blkdev(0, rbd_dev->name);
2418 goto err_out_client;
2419 rbd_dev->major = rc;
2421 rc = rbd_bus_add_dev(rbd_dev);
2423 goto err_out_blkdev;
2426 * At this point cleanup in the event of an error is the job
2427 * of the sysfs code (initiated by rbd_bus_del_dev()).
2429 * Set up and announce blkdev mapping.
2431 rc = rbd_init_disk(rbd_dev);
2435 rc = rbd_init_watch_dev(rbd_dev);
2442 /* this will also clean up rest of rbd_dev stuff */
2444 rbd_bus_del_dev(rbd_dev);
2449 unregister_blkdev(rbd_dev->major, rbd_dev->name);
2451 rbd_put_client(rbd_dev);
2453 rbd_id_put(rbd_dev);
2458 dout("Error adding device %s\n", buf);
2459 module_put(THIS_MODULE);
2461 return (ssize_t) rc;
2464 static struct rbd_device *__rbd_get_dev(unsigned long id)
2466 struct list_head *tmp;
2467 struct rbd_device *rbd_dev;
2469 spin_lock(&rbd_dev_list_lock);
2470 list_for_each(tmp, &rbd_dev_list) {
2471 rbd_dev = list_entry(tmp, struct rbd_device, node);
2472 if (rbd_dev->id == id) {
2473 spin_unlock(&rbd_dev_list_lock);
2477 spin_unlock(&rbd_dev_list_lock);
2481 static void rbd_dev_release(struct device *dev)
2483 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
2485 if (rbd_dev->watch_request) {
2486 struct ceph_client *client = rbd_dev->rbd_client->client;
2488 ceph_osdc_unregister_linger_request(&client->osdc,
2489 rbd_dev->watch_request);
2491 if (rbd_dev->watch_event)
2492 rbd_req_sync_unwatch(rbd_dev, rbd_dev->obj_md_name);
2494 rbd_put_client(rbd_dev);
2496 /* clean up and free blkdev */
2497 rbd_free_disk(rbd_dev);
2498 unregister_blkdev(rbd_dev->major, rbd_dev->name);
2500 /* done with the id, and with the rbd_dev */
2501 rbd_id_put(rbd_dev);
2504 /* release module ref */
2505 module_put(THIS_MODULE);
2508 static ssize_t rbd_remove(struct bus_type *bus,
2512 struct rbd_device *rbd_dev = NULL;
2517 rc = strict_strtoul(buf, 10, &ul);
2521 /* convert to int; abort if we lost anything in the conversion */
2522 target_id = (int) ul;
2523 if (target_id != ul)
2526 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
2528 rbd_dev = __rbd_get_dev(target_id);
2534 __rbd_remove_all_snaps(rbd_dev);
2535 rbd_bus_del_dev(rbd_dev);
2538 mutex_unlock(&ctl_mutex);
2542 static ssize_t rbd_snap_add(struct device *dev,
2543 struct device_attribute *attr,
2547 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
2549 char *name = kmalloc(count + 1, GFP_KERNEL);
2553 snprintf(name, count, "%s", buf);
2555 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
2557 ret = rbd_header_add_snap(rbd_dev,
2562 ret = __rbd_update_snaps(rbd_dev);
2566 /* shouldn't hold ctl_mutex when notifying.. notify might
2567 trigger a watch callback that would need to get that mutex */
2568 mutex_unlock(&ctl_mutex);
2570 /* make a best effort, don't error if failed */
2571 rbd_req_sync_notify(rbd_dev, rbd_dev->obj_md_name);
2578 mutex_unlock(&ctl_mutex);
2584 * create control files in sysfs
2587 static int rbd_sysfs_init(void)
2591 ret = device_register(&rbd_root_dev);
2595 ret = bus_register(&rbd_bus_type);
2597 device_unregister(&rbd_root_dev);
2602 static void rbd_sysfs_cleanup(void)
2604 bus_unregister(&rbd_bus_type);
2605 device_unregister(&rbd_root_dev);
2608 int __init rbd_init(void)
2612 rc = rbd_sysfs_init();
2615 pr_info("loaded " RBD_DRV_NAME_LONG "\n");
2619 void __exit rbd_exit(void)
2621 rbd_sysfs_cleanup();
2624 module_init(rbd_init);
2625 module_exit(rbd_exit);
2627 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
2628 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
2629 MODULE_DESCRIPTION("rados block device");
2631 /* following authorship retained from original osdblk.c */
2632 MODULE_AUTHOR("Jeff Garzik <jeff@garzik.org>");
2634 MODULE_LICENSE("GPL");