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"
44 #define RBD_DEBUG /* Activate rbd_assert() calls */
47 * The basic unit of block I/O is a sector. It is interpreted in a
48 * number of contexts in Linux (blk, bio, genhd), but the default is
49 * universally 512 bytes. These symbols are just slightly more
50 * meaningful than the bare numbers they represent.
52 #define SECTOR_SHIFT 9
53 #define SECTOR_SIZE (1ULL << SECTOR_SHIFT)
55 #define RBD_DRV_NAME "rbd"
56 #define RBD_DRV_NAME_LONG "rbd (rados block device)"
58 #define RBD_MINORS_PER_MAJOR 256 /* max minors per blkdev */
60 #define RBD_SNAP_DEV_NAME_PREFIX "snap_"
61 #define RBD_MAX_SNAP_NAME_LEN \
62 (NAME_MAX - (sizeof (RBD_SNAP_DEV_NAME_PREFIX) - 1))
64 #define RBD_MAX_SNAP_COUNT 510 /* allows max snapc to fit in 4KB */
66 #define RBD_SNAP_HEAD_NAME "-"
68 /* This allows a single page to hold an image name sent by OSD */
69 #define RBD_IMAGE_NAME_LEN_MAX (PAGE_SIZE - sizeof (__le32) - 1)
70 #define RBD_IMAGE_ID_LEN_MAX 64
72 #define RBD_OBJ_PREFIX_LEN_MAX 64
76 #define RBD_FEATURE_LAYERING (1<<0)
77 #define RBD_FEATURE_STRIPINGV2 (1<<1)
78 #define RBD_FEATURES_ALL \
79 (RBD_FEATURE_LAYERING | RBD_FEATURE_STRIPINGV2)
81 /* Features supported by this (client software) implementation. */
83 #define RBD_FEATURES_SUPPORTED (0)
86 * An RBD device name will be "rbd#", where the "rbd" comes from
87 * RBD_DRV_NAME above, and # is a unique integer identifier.
88 * MAX_INT_FORMAT_WIDTH is used in ensuring DEV_NAME_LEN is big
89 * enough to hold all possible device names.
91 #define DEV_NAME_LEN 32
92 #define MAX_INT_FORMAT_WIDTH ((5 * sizeof (int)) / 2 + 1)
95 * block device image metadata (in-memory version)
97 struct rbd_image_header {
98 /* These four fields never change for a given rbd image */
105 /* The remaining fields need to be updated occasionally */
107 struct ceph_snap_context *snapc;
115 * An rbd image specification.
117 * The tuple (pool_id, image_id, snap_id) is sufficient to uniquely
118 * identify an image. Each rbd_dev structure includes a pointer to
119 * an rbd_spec structure that encapsulates this identity.
121 * Each of the id's in an rbd_spec has an associated name. For a
122 * user-mapped image, the names are supplied and the id's associated
123 * with them are looked up. For a layered image, a parent image is
124 * defined by the tuple, and the names are looked up.
126 * An rbd_dev structure contains a parent_spec pointer which is
127 * non-null if the image it represents is a child in a layered
128 * image. This pointer will refer to the rbd_spec structure used
129 * by the parent rbd_dev for its own identity (i.e., the structure
130 * is shared between the parent and child).
132 * Since these structures are populated once, during the discovery
133 * phase of image construction, they are effectively immutable so
134 * we make no effort to synchronize access to them.
136 * Note that code herein does not assume the image name is known (it
137 * could be a null pointer).
153 * an instance of the client. multiple devices may share an rbd client.
156 struct ceph_client *client;
158 struct list_head node;
161 struct rbd_img_request;
162 typedef void (*rbd_img_callback_t)(struct rbd_img_request *);
164 #define BAD_WHICH U32_MAX /* Good which or bad which, which? */
166 struct rbd_obj_request;
167 typedef void (*rbd_obj_callback_t)(struct rbd_obj_request *);
169 enum obj_request_type {
170 OBJ_REQUEST_NODATA, OBJ_REQUEST_BIO, OBJ_REQUEST_PAGES
174 OBJ_REQ_DONE, /* completion flag: not done = 0, done = 1 */
175 OBJ_REQ_IMG_DATA, /* object usage: standalone = 0, image = 1 */
176 OBJ_REQ_KNOWN, /* EXISTS flag valid: no = 0, yes = 1 */
177 OBJ_REQ_EXISTS, /* target exists: no = 0, yes = 1 */
180 struct rbd_obj_request {
181 const char *object_name;
182 u64 offset; /* object start byte */
183 u64 length; /* bytes from offset */
187 * An object request associated with an image will have its
188 * img_data flag set; a standalone object request will not.
190 * A standalone object request will have which == BAD_WHICH
191 * and a null obj_request pointer.
193 * An object request initiated in support of a layered image
194 * object (to check for its existence before a write) will
195 * have which == BAD_WHICH and a non-null obj_request pointer.
197 * Finally, an object request for rbd image data will have
198 * which != BAD_WHICH, and will have a non-null img_request
199 * pointer. The value of which will be in the range
200 * 0..(img_request->obj_request_count-1).
203 struct rbd_obj_request *obj_request; /* STAT op */
205 struct rbd_img_request *img_request;
207 /* links for img_request->obj_requests list */
208 struct list_head links;
211 u32 which; /* posn image request list */
213 enum obj_request_type type;
215 struct bio *bio_list;
222 struct ceph_osd_request *osd_req;
224 u64 xferred; /* bytes transferred */
228 rbd_obj_callback_t callback;
229 struct completion completion;
235 IMG_REQ_WRITE, /* I/O direction: read = 0, write = 1 */
236 IMG_REQ_CHILD, /* initiator: block = 0, child image = 1 */
237 IMG_REQ_LAYERED, /* ENOENT handling: normal = 0, layered = 1 */
240 struct rbd_img_request {
241 struct rbd_device *rbd_dev;
242 u64 offset; /* starting image byte offset */
243 u64 length; /* byte count from offset */
246 u64 snap_id; /* for reads */
247 struct ceph_snap_context *snapc; /* for writes */
250 struct request *rq; /* block request */
251 struct rbd_obj_request *obj_request; /* obj req initiator */
253 spinlock_t completion_lock;/* protects next_completion */
255 rbd_img_callback_t callback;
256 u64 xferred;/* aggregate bytes transferred */
257 int result; /* first nonzero obj_request result */
259 u32 obj_request_count;
260 struct list_head obj_requests; /* rbd_obj_request structs */
265 #define for_each_obj_request(ireq, oreq) \
266 list_for_each_entry(oreq, &(ireq)->obj_requests, links)
267 #define for_each_obj_request_from(ireq, oreq) \
268 list_for_each_entry_from(oreq, &(ireq)->obj_requests, links)
269 #define for_each_obj_request_safe(ireq, oreq, n) \
270 list_for_each_entry_safe_reverse(oreq, n, &(ireq)->obj_requests, links)
276 struct list_head node;
291 int dev_id; /* blkdev unique id */
293 int major; /* blkdev assigned major */
294 struct gendisk *disk; /* blkdev's gendisk and rq */
296 u32 image_format; /* Either 1 or 2 */
297 struct rbd_client *rbd_client;
299 char name[DEV_NAME_LEN]; /* blkdev name, e.g. rbd3 */
301 spinlock_t lock; /* queue, flags, open_count */
303 struct rbd_image_header header;
304 unsigned long flags; /* possibly lock protected */
305 struct rbd_spec *spec;
309 struct ceph_file_layout layout;
311 struct ceph_osd_event *watch_event;
312 struct rbd_obj_request *watch_request;
314 struct rbd_spec *parent_spec;
316 struct rbd_device *parent;
318 /* protects updating the header */
319 struct rw_semaphore header_rwsem;
321 struct rbd_mapping mapping;
323 struct list_head node;
325 /* list of snapshots */
326 struct list_head snaps;
330 unsigned long open_count; /* protected by lock */
334 * Flag bits for rbd_dev->flags. If atomicity is required,
335 * rbd_dev->lock is used to protect access.
337 * Currently, only the "removing" flag (which is coupled with the
338 * "open_count" field) requires atomic access.
341 RBD_DEV_FLAG_EXISTS, /* mapped snapshot has not been deleted */
342 RBD_DEV_FLAG_REMOVING, /* this mapping is being removed */
345 static DEFINE_MUTEX(ctl_mutex); /* Serialize open/close/setup/teardown */
347 static LIST_HEAD(rbd_dev_list); /* devices */
348 static DEFINE_SPINLOCK(rbd_dev_list_lock);
350 static LIST_HEAD(rbd_client_list); /* clients */
351 static DEFINE_SPINLOCK(rbd_client_list_lock);
353 static int rbd_dev_snaps_update(struct rbd_device *rbd_dev);
354 static int rbd_dev_snaps_register(struct rbd_device *rbd_dev);
356 static void rbd_dev_release(struct device *dev);
357 static void rbd_remove_snap_dev(struct rbd_snap *snap);
359 static ssize_t rbd_add(struct bus_type *bus, const char *buf,
361 static ssize_t rbd_remove(struct bus_type *bus, const char *buf,
363 static int rbd_dev_probe(struct rbd_device *rbd_dev);
365 static struct bus_attribute rbd_bus_attrs[] = {
366 __ATTR(add, S_IWUSR, NULL, rbd_add),
367 __ATTR(remove, S_IWUSR, NULL, rbd_remove),
371 static struct bus_type rbd_bus_type = {
373 .bus_attrs = rbd_bus_attrs,
376 static void rbd_root_dev_release(struct device *dev)
380 static struct device rbd_root_dev = {
382 .release = rbd_root_dev_release,
385 static __printf(2, 3)
386 void rbd_warn(struct rbd_device *rbd_dev, const char *fmt, ...)
388 struct va_format vaf;
396 printk(KERN_WARNING "%s: %pV\n", RBD_DRV_NAME, &vaf);
397 else if (rbd_dev->disk)
398 printk(KERN_WARNING "%s: %s: %pV\n",
399 RBD_DRV_NAME, rbd_dev->disk->disk_name, &vaf);
400 else if (rbd_dev->spec && rbd_dev->spec->image_name)
401 printk(KERN_WARNING "%s: image %s: %pV\n",
402 RBD_DRV_NAME, rbd_dev->spec->image_name, &vaf);
403 else if (rbd_dev->spec && rbd_dev->spec->image_id)
404 printk(KERN_WARNING "%s: id %s: %pV\n",
405 RBD_DRV_NAME, rbd_dev->spec->image_id, &vaf);
407 printk(KERN_WARNING "%s: rbd_dev %p: %pV\n",
408 RBD_DRV_NAME, rbd_dev, &vaf);
413 #define rbd_assert(expr) \
414 if (unlikely(!(expr))) { \
415 printk(KERN_ERR "\nAssertion failure in %s() " \
417 "\trbd_assert(%s);\n\n", \
418 __func__, __LINE__, #expr); \
421 #else /* !RBD_DEBUG */
422 # define rbd_assert(expr) ((void) 0)
423 #endif /* !RBD_DEBUG */
425 static void rbd_img_parent_read(struct rbd_obj_request *obj_request);
426 static int rbd_img_obj_request_submit(struct rbd_obj_request *obj_request);
428 static int rbd_dev_refresh(struct rbd_device *rbd_dev, u64 *hver);
429 static int rbd_dev_v2_refresh(struct rbd_device *rbd_dev, u64 *hver);
431 static int rbd_open(struct block_device *bdev, fmode_t mode)
433 struct rbd_device *rbd_dev = bdev->bd_disk->private_data;
434 bool removing = false;
436 if ((mode & FMODE_WRITE) && rbd_dev->mapping.read_only)
439 spin_lock_irq(&rbd_dev->lock);
440 if (test_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags))
443 rbd_dev->open_count++;
444 spin_unlock_irq(&rbd_dev->lock);
448 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
449 (void) get_device(&rbd_dev->dev);
450 set_device_ro(bdev, rbd_dev->mapping.read_only);
451 mutex_unlock(&ctl_mutex);
456 static int rbd_release(struct gendisk *disk, fmode_t mode)
458 struct rbd_device *rbd_dev = disk->private_data;
459 unsigned long open_count_before;
461 spin_lock_irq(&rbd_dev->lock);
462 open_count_before = rbd_dev->open_count--;
463 spin_unlock_irq(&rbd_dev->lock);
464 rbd_assert(open_count_before > 0);
466 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
467 put_device(&rbd_dev->dev);
468 mutex_unlock(&ctl_mutex);
473 static const struct block_device_operations rbd_bd_ops = {
474 .owner = THIS_MODULE,
476 .release = rbd_release,
480 * Initialize an rbd client instance.
483 static struct rbd_client *rbd_client_create(struct ceph_options *ceph_opts)
485 struct rbd_client *rbdc;
488 dout("%s:\n", __func__);
489 rbdc = kmalloc(sizeof(struct rbd_client), GFP_KERNEL);
493 kref_init(&rbdc->kref);
494 INIT_LIST_HEAD(&rbdc->node);
496 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
498 rbdc->client = ceph_create_client(ceph_opts, rbdc, 0, 0);
499 if (IS_ERR(rbdc->client))
501 ceph_opts = NULL; /* Now rbdc->client is responsible for ceph_opts */
503 ret = ceph_open_session(rbdc->client);
507 spin_lock(&rbd_client_list_lock);
508 list_add_tail(&rbdc->node, &rbd_client_list);
509 spin_unlock(&rbd_client_list_lock);
511 mutex_unlock(&ctl_mutex);
512 dout("%s: rbdc %p\n", __func__, rbdc);
517 ceph_destroy_client(rbdc->client);
519 mutex_unlock(&ctl_mutex);
523 ceph_destroy_options(ceph_opts);
524 dout("%s: error %d\n", __func__, ret);
529 static struct rbd_client *__rbd_get_client(struct rbd_client *rbdc)
531 kref_get(&rbdc->kref);
537 * Find a ceph client with specific addr and configuration. If
538 * found, bump its reference count.
540 static struct rbd_client *rbd_client_find(struct ceph_options *ceph_opts)
542 struct rbd_client *client_node;
545 if (ceph_opts->flags & CEPH_OPT_NOSHARE)
548 spin_lock(&rbd_client_list_lock);
549 list_for_each_entry(client_node, &rbd_client_list, node) {
550 if (!ceph_compare_options(ceph_opts, client_node->client)) {
551 __rbd_get_client(client_node);
557 spin_unlock(&rbd_client_list_lock);
559 return found ? client_node : NULL;
569 /* string args above */
572 /* Boolean args above */
576 static match_table_t rbd_opts_tokens = {
578 /* string args above */
579 {Opt_read_only, "read_only"},
580 {Opt_read_only, "ro"}, /* Alternate spelling */
581 {Opt_read_write, "read_write"},
582 {Opt_read_write, "rw"}, /* Alternate spelling */
583 /* Boolean args above */
591 #define RBD_READ_ONLY_DEFAULT false
593 static int parse_rbd_opts_token(char *c, void *private)
595 struct rbd_options *rbd_opts = private;
596 substring_t argstr[MAX_OPT_ARGS];
597 int token, intval, ret;
599 token = match_token(c, rbd_opts_tokens, argstr);
603 if (token < Opt_last_int) {
604 ret = match_int(&argstr[0], &intval);
606 pr_err("bad mount option arg (not int) "
610 dout("got int token %d val %d\n", token, intval);
611 } else if (token > Opt_last_int && token < Opt_last_string) {
612 dout("got string token %d val %s\n", token,
614 } else if (token > Opt_last_string && token < Opt_last_bool) {
615 dout("got Boolean token %d\n", token);
617 dout("got token %d\n", token);
622 rbd_opts->read_only = true;
625 rbd_opts->read_only = false;
635 * Get a ceph client with specific addr and configuration, if one does
636 * not exist create it.
638 static struct rbd_client *rbd_get_client(struct ceph_options *ceph_opts)
640 struct rbd_client *rbdc;
642 rbdc = rbd_client_find(ceph_opts);
643 if (rbdc) /* using an existing client */
644 ceph_destroy_options(ceph_opts);
646 rbdc = rbd_client_create(ceph_opts);
652 * Destroy ceph client
654 * Caller must hold rbd_client_list_lock.
656 static void rbd_client_release(struct kref *kref)
658 struct rbd_client *rbdc = container_of(kref, struct rbd_client, kref);
660 dout("%s: rbdc %p\n", __func__, rbdc);
661 spin_lock(&rbd_client_list_lock);
662 list_del(&rbdc->node);
663 spin_unlock(&rbd_client_list_lock);
665 ceph_destroy_client(rbdc->client);
670 * Drop reference to ceph client node. If it's not referenced anymore, release
673 static void rbd_put_client(struct rbd_client *rbdc)
676 kref_put(&rbdc->kref, rbd_client_release);
679 static bool rbd_image_format_valid(u32 image_format)
681 return image_format == 1 || image_format == 2;
684 static bool rbd_dev_ondisk_valid(struct rbd_image_header_ondisk *ondisk)
689 /* The header has to start with the magic rbd header text */
690 if (memcmp(&ondisk->text, RBD_HEADER_TEXT, sizeof (RBD_HEADER_TEXT)))
693 /* The bio layer requires at least sector-sized I/O */
695 if (ondisk->options.order < SECTOR_SHIFT)
698 /* If we use u64 in a few spots we may be able to loosen this */
700 if (ondisk->options.order > 8 * sizeof (int) - 1)
704 * The size of a snapshot header has to fit in a size_t, and
705 * that limits the number of snapshots.
707 snap_count = le32_to_cpu(ondisk->snap_count);
708 size = SIZE_MAX - sizeof (struct ceph_snap_context);
709 if (snap_count > size / sizeof (__le64))
713 * Not only that, but the size of the entire the snapshot
714 * header must also be representable in a size_t.
716 size -= snap_count * sizeof (__le64);
717 if ((u64) size < le64_to_cpu(ondisk->snap_names_len))
724 * Create a new header structure, translate header format from the on-disk
727 static int rbd_header_from_disk(struct rbd_image_header *header,
728 struct rbd_image_header_ondisk *ondisk)
735 memset(header, 0, sizeof (*header));
737 snap_count = le32_to_cpu(ondisk->snap_count);
739 len = strnlen(ondisk->object_prefix, sizeof (ondisk->object_prefix));
740 header->object_prefix = kmalloc(len + 1, GFP_KERNEL);
741 if (!header->object_prefix)
743 memcpy(header->object_prefix, ondisk->object_prefix, len);
744 header->object_prefix[len] = '\0';
747 u64 snap_names_len = le64_to_cpu(ondisk->snap_names_len);
749 /* Save a copy of the snapshot names */
751 if (snap_names_len > (u64) SIZE_MAX)
753 header->snap_names = kmalloc(snap_names_len, GFP_KERNEL);
754 if (!header->snap_names)
757 * Note that rbd_dev_v1_header_read() guarantees
758 * the ondisk buffer we're working with has
759 * snap_names_len bytes beyond the end of the
760 * snapshot id array, this memcpy() is safe.
762 memcpy(header->snap_names, &ondisk->snaps[snap_count],
765 /* Record each snapshot's size */
767 size = snap_count * sizeof (*header->snap_sizes);
768 header->snap_sizes = kmalloc(size, GFP_KERNEL);
769 if (!header->snap_sizes)
771 for (i = 0; i < snap_count; i++)
772 header->snap_sizes[i] =
773 le64_to_cpu(ondisk->snaps[i].image_size);
775 WARN_ON(ondisk->snap_names_len);
776 header->snap_names = NULL;
777 header->snap_sizes = NULL;
780 header->features = 0; /* No features support in v1 images */
781 header->obj_order = ondisk->options.order;
782 header->crypt_type = ondisk->options.crypt_type;
783 header->comp_type = ondisk->options.comp_type;
785 /* Allocate and fill in the snapshot context */
787 header->image_size = le64_to_cpu(ondisk->image_size);
788 size = sizeof (struct ceph_snap_context);
789 size += snap_count * sizeof (header->snapc->snaps[0]);
790 header->snapc = kzalloc(size, GFP_KERNEL);
794 atomic_set(&header->snapc->nref, 1);
795 header->snapc->seq = le64_to_cpu(ondisk->snap_seq);
796 header->snapc->num_snaps = snap_count;
797 for (i = 0; i < snap_count; i++)
798 header->snapc->snaps[i] =
799 le64_to_cpu(ondisk->snaps[i].id);
804 kfree(header->snap_sizes);
805 header->snap_sizes = NULL;
806 kfree(header->snap_names);
807 header->snap_names = NULL;
808 kfree(header->object_prefix);
809 header->object_prefix = NULL;
814 static const char *rbd_snap_name(struct rbd_device *rbd_dev, u64 snap_id)
816 struct rbd_snap *snap;
818 if (snap_id == CEPH_NOSNAP)
819 return RBD_SNAP_HEAD_NAME;
821 list_for_each_entry(snap, &rbd_dev->snaps, node)
822 if (snap_id == snap->id)
828 static int snap_by_name(struct rbd_device *rbd_dev, const char *snap_name)
831 struct rbd_snap *snap;
833 list_for_each_entry(snap, &rbd_dev->snaps, node) {
834 if (!strcmp(snap_name, snap->name)) {
835 rbd_dev->spec->snap_id = snap->id;
836 rbd_dev->mapping.size = snap->size;
837 rbd_dev->mapping.features = snap->features;
846 static int rbd_dev_set_mapping(struct rbd_device *rbd_dev)
850 if (!memcmp(rbd_dev->spec->snap_name, RBD_SNAP_HEAD_NAME,
851 sizeof (RBD_SNAP_HEAD_NAME))) {
852 rbd_dev->spec->snap_id = CEPH_NOSNAP;
853 rbd_dev->mapping.size = rbd_dev->header.image_size;
854 rbd_dev->mapping.features = rbd_dev->header.features;
857 ret = snap_by_name(rbd_dev, rbd_dev->spec->snap_name);
860 rbd_dev->mapping.read_only = true;
862 set_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags);
868 static void rbd_header_free(struct rbd_image_header *header)
870 kfree(header->object_prefix);
871 header->object_prefix = NULL;
872 kfree(header->snap_sizes);
873 header->snap_sizes = NULL;
874 kfree(header->snap_names);
875 header->snap_names = NULL;
876 ceph_put_snap_context(header->snapc);
877 header->snapc = NULL;
880 static const char *rbd_segment_name(struct rbd_device *rbd_dev, u64 offset)
886 name = kmalloc(MAX_OBJ_NAME_SIZE + 1, GFP_NOIO);
889 segment = offset >> rbd_dev->header.obj_order;
890 ret = snprintf(name, MAX_OBJ_NAME_SIZE + 1, "%s.%012llx",
891 rbd_dev->header.object_prefix, segment);
892 if (ret < 0 || ret > MAX_OBJ_NAME_SIZE) {
893 pr_err("error formatting segment name for #%llu (%d)\n",
902 static u64 rbd_segment_offset(struct rbd_device *rbd_dev, u64 offset)
904 u64 segment_size = (u64) 1 << rbd_dev->header.obj_order;
906 return offset & (segment_size - 1);
909 static u64 rbd_segment_length(struct rbd_device *rbd_dev,
910 u64 offset, u64 length)
912 u64 segment_size = (u64) 1 << rbd_dev->header.obj_order;
914 offset &= segment_size - 1;
916 rbd_assert(length <= U64_MAX - offset);
917 if (offset + length > segment_size)
918 length = segment_size - offset;
924 * returns the size of an object in the image
926 static u64 rbd_obj_bytes(struct rbd_image_header *header)
928 return 1 << header->obj_order;
935 static void bio_chain_put(struct bio *chain)
941 chain = chain->bi_next;
947 * zeros a bio chain, starting at specific offset
949 static void zero_bio_chain(struct bio *chain, int start_ofs)
958 bio_for_each_segment(bv, chain, i) {
959 if (pos + bv->bv_len > start_ofs) {
960 int remainder = max(start_ofs - pos, 0);
961 buf = bvec_kmap_irq(bv, &flags);
962 memset(buf + remainder, 0,
963 bv->bv_len - remainder);
964 bvec_kunmap_irq(buf, &flags);
969 chain = chain->bi_next;
974 * Clone a portion of a bio, starting at the given byte offset
975 * and continuing for the number of bytes indicated.
977 static struct bio *bio_clone_range(struct bio *bio_src,
986 unsigned short end_idx;
990 /* Handle the easy case for the caller */
992 if (!offset && len == bio_src->bi_size)
993 return bio_clone(bio_src, gfpmask);
995 if (WARN_ON_ONCE(!len))
997 if (WARN_ON_ONCE(len > bio_src->bi_size))
999 if (WARN_ON_ONCE(offset > bio_src->bi_size - len))
1002 /* Find first affected segment... */
1005 __bio_for_each_segment(bv, bio_src, idx, 0) {
1006 if (resid < bv->bv_len)
1008 resid -= bv->bv_len;
1012 /* ...and the last affected segment */
1015 __bio_for_each_segment(bv, bio_src, end_idx, idx) {
1016 if (resid <= bv->bv_len)
1018 resid -= bv->bv_len;
1020 vcnt = end_idx - idx + 1;
1022 /* Build the clone */
1024 bio = bio_alloc(gfpmask, (unsigned int) vcnt);
1026 return NULL; /* ENOMEM */
1028 bio->bi_bdev = bio_src->bi_bdev;
1029 bio->bi_sector = bio_src->bi_sector + (offset >> SECTOR_SHIFT);
1030 bio->bi_rw = bio_src->bi_rw;
1031 bio->bi_flags |= 1 << BIO_CLONED;
1034 * Copy over our part of the bio_vec, then update the first
1035 * and last (or only) entries.
1037 memcpy(&bio->bi_io_vec[0], &bio_src->bi_io_vec[idx],
1038 vcnt * sizeof (struct bio_vec));
1039 bio->bi_io_vec[0].bv_offset += voff;
1041 bio->bi_io_vec[0].bv_len -= voff;
1042 bio->bi_io_vec[vcnt - 1].bv_len = resid;
1044 bio->bi_io_vec[0].bv_len = len;
1047 bio->bi_vcnt = vcnt;
1055 * Clone a portion of a bio chain, starting at the given byte offset
1056 * into the first bio in the source chain and continuing for the
1057 * number of bytes indicated. The result is another bio chain of
1058 * exactly the given length, or a null pointer on error.
1060 * The bio_src and offset parameters are both in-out. On entry they
1061 * refer to the first source bio and the offset into that bio where
1062 * the start of data to be cloned is located.
1064 * On return, bio_src is updated to refer to the bio in the source
1065 * chain that contains first un-cloned byte, and *offset will
1066 * contain the offset of that byte within that bio.
1068 static struct bio *bio_chain_clone_range(struct bio **bio_src,
1069 unsigned int *offset,
1073 struct bio *bi = *bio_src;
1074 unsigned int off = *offset;
1075 struct bio *chain = NULL;
1078 /* Build up a chain of clone bios up to the limit */
1080 if (!bi || off >= bi->bi_size || !len)
1081 return NULL; /* Nothing to clone */
1085 unsigned int bi_size;
1089 rbd_warn(NULL, "bio_chain exhausted with %u left", len);
1090 goto out_err; /* EINVAL; ran out of bio's */
1092 bi_size = min_t(unsigned int, bi->bi_size - off, len);
1093 bio = bio_clone_range(bi, off, bi_size, gfpmask);
1095 goto out_err; /* ENOMEM */
1098 end = &bio->bi_next;
1101 if (off == bi->bi_size) {
1112 bio_chain_put(chain);
1118 * The default/initial value for all object request flags is 0. For
1119 * each flag, once its value is set to 1 it is never reset to 0
1122 static void obj_request_img_data_set(struct rbd_obj_request *obj_request)
1124 if (test_and_set_bit(OBJ_REQ_IMG_DATA, &obj_request->flags)) {
1125 struct rbd_device *rbd_dev;
1127 rbd_dev = obj_request->img_request->rbd_dev;
1128 rbd_warn(rbd_dev, "obj_request %p already marked img_data\n",
1133 static bool obj_request_img_data_test(struct rbd_obj_request *obj_request)
1136 return test_bit(OBJ_REQ_IMG_DATA, &obj_request->flags) != 0;
1139 static void obj_request_done_set(struct rbd_obj_request *obj_request)
1141 if (test_and_set_bit(OBJ_REQ_DONE, &obj_request->flags)) {
1142 struct rbd_device *rbd_dev = NULL;
1144 if (obj_request_img_data_test(obj_request))
1145 rbd_dev = obj_request->img_request->rbd_dev;
1146 rbd_warn(rbd_dev, "obj_request %p already marked done\n",
1151 static bool obj_request_done_test(struct rbd_obj_request *obj_request)
1154 return test_bit(OBJ_REQ_DONE, &obj_request->flags) != 0;
1158 * This sets the KNOWN flag after (possibly) setting the EXISTS
1159 * flag. The latter is set based on the "exists" value provided.
1161 * Note that for our purposes once an object exists it never goes
1162 * away again. It's possible that the response from two existence
1163 * checks are separated by the creation of the target object, and
1164 * the first ("doesn't exist") response arrives *after* the second
1165 * ("does exist"). In that case we ignore the second one.
1167 static void obj_request_existence_set(struct rbd_obj_request *obj_request,
1171 set_bit(OBJ_REQ_EXISTS, &obj_request->flags);
1172 set_bit(OBJ_REQ_KNOWN, &obj_request->flags);
1176 static bool obj_request_known_test(struct rbd_obj_request *obj_request)
1179 return test_bit(OBJ_REQ_KNOWN, &obj_request->flags) != 0;
1182 static bool obj_request_exists_test(struct rbd_obj_request *obj_request)
1185 return test_bit(OBJ_REQ_EXISTS, &obj_request->flags) != 0;
1188 static void rbd_obj_request_get(struct rbd_obj_request *obj_request)
1190 dout("%s: obj %p (was %d)\n", __func__, obj_request,
1191 atomic_read(&obj_request->kref.refcount));
1192 kref_get(&obj_request->kref);
1195 static void rbd_obj_request_destroy(struct kref *kref);
1196 static void rbd_obj_request_put(struct rbd_obj_request *obj_request)
1198 rbd_assert(obj_request != NULL);
1199 dout("%s: obj %p (was %d)\n", __func__, obj_request,
1200 atomic_read(&obj_request->kref.refcount));
1201 kref_put(&obj_request->kref, rbd_obj_request_destroy);
1204 static void rbd_img_request_get(struct rbd_img_request *img_request)
1206 dout("%s: img %p (was %d)\n", __func__, img_request,
1207 atomic_read(&img_request->kref.refcount));
1208 kref_get(&img_request->kref);
1211 static void rbd_img_request_destroy(struct kref *kref);
1212 static void rbd_img_request_put(struct rbd_img_request *img_request)
1214 rbd_assert(img_request != NULL);
1215 dout("%s: img %p (was %d)\n", __func__, img_request,
1216 atomic_read(&img_request->kref.refcount));
1217 kref_put(&img_request->kref, rbd_img_request_destroy);
1220 static inline void rbd_img_obj_request_add(struct rbd_img_request *img_request,
1221 struct rbd_obj_request *obj_request)
1223 rbd_assert(obj_request->img_request == NULL);
1225 /* Image request now owns object's original reference */
1226 obj_request->img_request = img_request;
1227 obj_request->which = img_request->obj_request_count;
1228 rbd_assert(!obj_request_img_data_test(obj_request));
1229 obj_request_img_data_set(obj_request);
1230 rbd_assert(obj_request->which != BAD_WHICH);
1231 img_request->obj_request_count++;
1232 list_add_tail(&obj_request->links, &img_request->obj_requests);
1233 dout("%s: img %p obj %p w=%u\n", __func__, img_request, obj_request,
1234 obj_request->which);
1237 static inline void rbd_img_obj_request_del(struct rbd_img_request *img_request,
1238 struct rbd_obj_request *obj_request)
1240 rbd_assert(obj_request->which != BAD_WHICH);
1242 dout("%s: img %p obj %p w=%u\n", __func__, img_request, obj_request,
1243 obj_request->which);
1244 list_del(&obj_request->links);
1245 rbd_assert(img_request->obj_request_count > 0);
1246 img_request->obj_request_count--;
1247 rbd_assert(obj_request->which == img_request->obj_request_count);
1248 obj_request->which = BAD_WHICH;
1249 rbd_assert(obj_request_img_data_test(obj_request));
1250 rbd_assert(obj_request->img_request == img_request);
1251 obj_request->img_request = NULL;
1252 obj_request->callback = NULL;
1253 rbd_obj_request_put(obj_request);
1256 static bool obj_request_type_valid(enum obj_request_type type)
1259 case OBJ_REQUEST_NODATA:
1260 case OBJ_REQUEST_BIO:
1261 case OBJ_REQUEST_PAGES:
1268 static int rbd_obj_request_submit(struct ceph_osd_client *osdc,
1269 struct rbd_obj_request *obj_request)
1271 dout("%s: osdc %p obj %p\n", __func__, osdc, obj_request);
1273 return ceph_osdc_start_request(osdc, obj_request->osd_req, false);
1276 static void rbd_img_request_complete(struct rbd_img_request *img_request)
1279 dout("%s: img %p\n", __func__, img_request);
1282 * If no error occurred, compute the aggregate transfer
1283 * count for the image request. We could instead use
1284 * atomic64_cmpxchg() to update it as each object request
1285 * completes; not clear which way is better off hand.
1287 if (!img_request->result) {
1288 struct rbd_obj_request *obj_request;
1291 for_each_obj_request(img_request, obj_request)
1292 xferred += obj_request->xferred;
1293 img_request->xferred = xferred;
1296 if (img_request->callback)
1297 img_request->callback(img_request);
1299 rbd_img_request_put(img_request);
1302 /* Caller is responsible for rbd_obj_request_destroy(obj_request) */
1304 static int rbd_obj_request_wait(struct rbd_obj_request *obj_request)
1306 dout("%s: obj %p\n", __func__, obj_request);
1308 return wait_for_completion_interruptible(&obj_request->completion);
1312 * The default/initial value for all image request flags is 0. Each
1313 * is conditionally set to 1 at image request initialization time
1314 * and currently never change thereafter.
1316 static void img_request_write_set(struct rbd_img_request *img_request)
1318 set_bit(IMG_REQ_WRITE, &img_request->flags);
1322 static bool img_request_write_test(struct rbd_img_request *img_request)
1325 return test_bit(IMG_REQ_WRITE, &img_request->flags) != 0;
1328 static void img_request_child_set(struct rbd_img_request *img_request)
1330 set_bit(IMG_REQ_CHILD, &img_request->flags);
1334 static bool img_request_child_test(struct rbd_img_request *img_request)
1337 return test_bit(IMG_REQ_CHILD, &img_request->flags) != 0;
1340 static void img_request_layered_set(struct rbd_img_request *img_request)
1342 set_bit(IMG_REQ_LAYERED, &img_request->flags);
1346 static bool img_request_layered_test(struct rbd_img_request *img_request)
1349 return test_bit(IMG_REQ_LAYERED, &img_request->flags) != 0;
1353 rbd_img_obj_request_read_callback(struct rbd_obj_request *obj_request)
1355 dout("%s: obj %p img %p result %d %llu/%llu\n", __func__,
1356 obj_request, obj_request->img_request, obj_request->result,
1357 obj_request->xferred, obj_request->length);
1359 * ENOENT means a hole in the image. We zero-fill the
1360 * entire length of the request. A short read also implies
1361 * zero-fill to the end of the request. Either way we
1362 * update the xferred count to indicate the whole request
1365 BUG_ON(obj_request->type != OBJ_REQUEST_BIO);
1366 if (obj_request->result == -ENOENT) {
1367 zero_bio_chain(obj_request->bio_list, 0);
1368 obj_request->result = 0;
1369 obj_request->xferred = obj_request->length;
1370 } else if (obj_request->xferred < obj_request->length &&
1371 !obj_request->result) {
1372 zero_bio_chain(obj_request->bio_list, obj_request->xferred);
1373 obj_request->xferred = obj_request->length;
1375 obj_request_done_set(obj_request);
1378 static void rbd_obj_request_complete(struct rbd_obj_request *obj_request)
1380 dout("%s: obj %p cb %p\n", __func__, obj_request,
1381 obj_request->callback);
1382 if (obj_request->callback)
1383 obj_request->callback(obj_request);
1385 complete_all(&obj_request->completion);
1388 static void rbd_osd_trivial_callback(struct rbd_obj_request *obj_request)
1390 dout("%s: obj %p\n", __func__, obj_request);
1391 obj_request_done_set(obj_request);
1394 static void rbd_osd_read_callback(struct rbd_obj_request *obj_request)
1396 struct rbd_img_request *img_request = NULL;
1397 bool layered = false;
1399 if (obj_request_img_data_test(obj_request)) {
1400 img_request = obj_request->img_request;
1401 layered = img_request && img_request_layered_test(img_request);
1407 dout("%s: obj %p img %p result %d %llu/%llu\n", __func__,
1408 obj_request, img_request, obj_request->result,
1409 obj_request->xferred, obj_request->length);
1410 if (layered && obj_request->result == -ENOENT)
1411 rbd_img_parent_read(obj_request);
1412 else if (img_request)
1413 rbd_img_obj_request_read_callback(obj_request);
1415 obj_request_done_set(obj_request);
1418 static void rbd_osd_write_callback(struct rbd_obj_request *obj_request)
1420 dout("%s: obj %p result %d %llu\n", __func__, obj_request,
1421 obj_request->result, obj_request->length);
1423 * There is no such thing as a successful short write. Set
1424 * it to our originally-requested length.
1426 obj_request->xferred = obj_request->length;
1427 obj_request_done_set(obj_request);
1431 * For a simple stat call there's nothing to do. We'll do more if
1432 * this is part of a write sequence for a layered image.
1434 static void rbd_osd_stat_callback(struct rbd_obj_request *obj_request)
1436 dout("%s: obj %p\n", __func__, obj_request);
1437 obj_request_done_set(obj_request);
1440 static void rbd_osd_req_callback(struct ceph_osd_request *osd_req,
1441 struct ceph_msg *msg)
1443 struct rbd_obj_request *obj_request = osd_req->r_priv;
1446 dout("%s: osd_req %p msg %p\n", __func__, osd_req, msg);
1447 rbd_assert(osd_req == obj_request->osd_req);
1448 if (obj_request_img_data_test(obj_request)) {
1449 rbd_assert(obj_request->img_request);
1450 rbd_assert(obj_request->which != BAD_WHICH);
1452 rbd_assert(obj_request->which == BAD_WHICH);
1455 if (osd_req->r_result < 0)
1456 obj_request->result = osd_req->r_result;
1457 obj_request->version = le64_to_cpu(osd_req->r_reassert_version.version);
1459 WARN_ON(osd_req->r_num_ops != 1); /* For now */
1462 * We support a 64-bit length, but ultimately it has to be
1463 * passed to blk_end_request(), which takes an unsigned int.
1465 obj_request->xferred = osd_req->r_reply_op_len[0];
1466 rbd_assert(obj_request->xferred < (u64)UINT_MAX);
1467 opcode = osd_req->r_ops[0].op;
1469 case CEPH_OSD_OP_READ:
1470 rbd_osd_read_callback(obj_request);
1472 case CEPH_OSD_OP_WRITE:
1473 rbd_osd_write_callback(obj_request);
1475 case CEPH_OSD_OP_STAT:
1476 rbd_osd_stat_callback(obj_request);
1478 case CEPH_OSD_OP_CALL:
1479 case CEPH_OSD_OP_NOTIFY_ACK:
1480 case CEPH_OSD_OP_WATCH:
1481 rbd_osd_trivial_callback(obj_request);
1484 rbd_warn(NULL, "%s: unsupported op %hu\n",
1485 obj_request->object_name, (unsigned short) opcode);
1489 if (obj_request_done_test(obj_request))
1490 rbd_obj_request_complete(obj_request);
1493 static void rbd_osd_req_format_read(struct rbd_obj_request *obj_request)
1495 struct rbd_img_request *img_request = obj_request->img_request;
1496 struct ceph_osd_request *osd_req = obj_request->osd_req;
1499 rbd_assert(osd_req != NULL);
1501 snap_id = img_request ? img_request->snap_id : CEPH_NOSNAP;
1502 ceph_osdc_build_request(osd_req, obj_request->offset,
1503 NULL, snap_id, NULL);
1506 static void rbd_osd_req_format_write(struct rbd_obj_request *obj_request)
1508 struct rbd_img_request *img_request = obj_request->img_request;
1509 struct ceph_osd_request *osd_req = obj_request->osd_req;
1510 struct ceph_snap_context *snapc;
1511 struct timespec mtime = CURRENT_TIME;
1513 rbd_assert(osd_req != NULL);
1515 snapc = img_request ? img_request->snapc : NULL;
1516 ceph_osdc_build_request(osd_req, obj_request->offset,
1517 snapc, CEPH_NOSNAP, &mtime);
1520 static struct ceph_osd_request *rbd_osd_req_create(
1521 struct rbd_device *rbd_dev,
1523 struct rbd_obj_request *obj_request)
1525 struct ceph_snap_context *snapc = NULL;
1526 struct ceph_osd_client *osdc;
1527 struct ceph_osd_request *osd_req;
1529 if (obj_request_img_data_test(obj_request)) {
1530 struct rbd_img_request *img_request = obj_request->img_request;
1532 rbd_assert(write_request ==
1533 img_request_write_test(img_request));
1535 snapc = img_request->snapc;
1538 /* Allocate and initialize the request, for the single op */
1540 osdc = &rbd_dev->rbd_client->client->osdc;
1541 osd_req = ceph_osdc_alloc_request(osdc, snapc, 1, false, GFP_ATOMIC);
1543 return NULL; /* ENOMEM */
1546 osd_req->r_flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK;
1548 osd_req->r_flags = CEPH_OSD_FLAG_READ;
1550 osd_req->r_callback = rbd_osd_req_callback;
1551 osd_req->r_priv = obj_request;
1553 osd_req->r_oid_len = strlen(obj_request->object_name);
1554 rbd_assert(osd_req->r_oid_len < sizeof (osd_req->r_oid));
1555 memcpy(osd_req->r_oid, obj_request->object_name, osd_req->r_oid_len);
1557 osd_req->r_file_layout = rbd_dev->layout; /* struct */
1562 static void rbd_osd_req_destroy(struct ceph_osd_request *osd_req)
1564 ceph_osdc_put_request(osd_req);
1567 /* object_name is assumed to be a non-null pointer and NUL-terminated */
1569 static struct rbd_obj_request *rbd_obj_request_create(const char *object_name,
1570 u64 offset, u64 length,
1571 enum obj_request_type type)
1573 struct rbd_obj_request *obj_request;
1577 rbd_assert(obj_request_type_valid(type));
1579 size = strlen(object_name) + 1;
1580 obj_request = kzalloc(sizeof (*obj_request) + size, GFP_KERNEL);
1584 name = (char *)(obj_request + 1);
1585 obj_request->object_name = memcpy(name, object_name, size);
1586 obj_request->offset = offset;
1587 obj_request->length = length;
1588 obj_request->flags = 0;
1589 obj_request->which = BAD_WHICH;
1590 obj_request->type = type;
1591 INIT_LIST_HEAD(&obj_request->links);
1592 init_completion(&obj_request->completion);
1593 kref_init(&obj_request->kref);
1595 dout("%s: \"%s\" %llu/%llu %d -> obj %p\n", __func__, object_name,
1596 offset, length, (int)type, obj_request);
1601 static void rbd_obj_request_destroy(struct kref *kref)
1603 struct rbd_obj_request *obj_request;
1605 obj_request = container_of(kref, struct rbd_obj_request, kref);
1607 dout("%s: obj %p\n", __func__, obj_request);
1609 rbd_assert(obj_request->img_request == NULL);
1610 rbd_assert(obj_request->which == BAD_WHICH);
1612 if (obj_request->osd_req)
1613 rbd_osd_req_destroy(obj_request->osd_req);
1615 rbd_assert(obj_request_type_valid(obj_request->type));
1616 switch (obj_request->type) {
1617 case OBJ_REQUEST_NODATA:
1618 break; /* Nothing to do */
1619 case OBJ_REQUEST_BIO:
1620 if (obj_request->bio_list)
1621 bio_chain_put(obj_request->bio_list);
1623 case OBJ_REQUEST_PAGES:
1624 if (obj_request->pages)
1625 ceph_release_page_vector(obj_request->pages,
1626 obj_request->page_count);
1634 * Caller is responsible for filling in the list of object requests
1635 * that comprises the image request, and the Linux request pointer
1636 * (if there is one).
1638 static struct rbd_img_request *rbd_img_request_create(
1639 struct rbd_device *rbd_dev,
1640 u64 offset, u64 length,
1644 struct rbd_img_request *img_request;
1645 struct ceph_snap_context *snapc = NULL;
1647 img_request = kmalloc(sizeof (*img_request), GFP_ATOMIC);
1651 if (write_request) {
1652 down_read(&rbd_dev->header_rwsem);
1653 snapc = ceph_get_snap_context(rbd_dev->header.snapc);
1654 up_read(&rbd_dev->header_rwsem);
1655 if (WARN_ON(!snapc)) {
1657 return NULL; /* Shouldn't happen */
1662 img_request->rq = NULL;
1663 img_request->rbd_dev = rbd_dev;
1664 img_request->offset = offset;
1665 img_request->length = length;
1666 img_request->flags = 0;
1667 if (write_request) {
1668 img_request_write_set(img_request);
1669 img_request->snapc = snapc;
1671 img_request->snap_id = rbd_dev->spec->snap_id;
1674 img_request_child_set(img_request);
1675 if (rbd_dev->parent_spec)
1676 img_request_layered_set(img_request);
1677 spin_lock_init(&img_request->completion_lock);
1678 img_request->next_completion = 0;
1679 img_request->callback = NULL;
1680 img_request->result = 0;
1681 img_request->obj_request_count = 0;
1682 INIT_LIST_HEAD(&img_request->obj_requests);
1683 kref_init(&img_request->kref);
1685 rbd_img_request_get(img_request); /* Avoid a warning */
1686 rbd_img_request_put(img_request); /* TEMPORARY */
1688 dout("%s: rbd_dev %p %s %llu/%llu -> img %p\n", __func__, rbd_dev,
1689 write_request ? "write" : "read", offset, length,
1695 static void rbd_img_request_destroy(struct kref *kref)
1697 struct rbd_img_request *img_request;
1698 struct rbd_obj_request *obj_request;
1699 struct rbd_obj_request *next_obj_request;
1701 img_request = container_of(kref, struct rbd_img_request, kref);
1703 dout("%s: img %p\n", __func__, img_request);
1705 for_each_obj_request_safe(img_request, obj_request, next_obj_request)
1706 rbd_img_obj_request_del(img_request, obj_request);
1707 rbd_assert(img_request->obj_request_count == 0);
1709 if (img_request_write_test(img_request))
1710 ceph_put_snap_context(img_request->snapc);
1712 if (img_request_child_test(img_request))
1713 rbd_obj_request_put(img_request->obj_request);
1718 static bool rbd_img_obj_end_request(struct rbd_obj_request *obj_request)
1720 struct rbd_img_request *img_request;
1721 unsigned int xferred;
1725 rbd_assert(obj_request_img_data_test(obj_request));
1726 img_request = obj_request->img_request;
1728 rbd_assert(obj_request->xferred <= (u64)UINT_MAX);
1729 xferred = (unsigned int)obj_request->xferred;
1730 result = obj_request->result;
1732 struct rbd_device *rbd_dev = img_request->rbd_dev;
1734 rbd_warn(rbd_dev, "%s %llx at %llx (%llx)\n",
1735 img_request_write_test(img_request) ? "write" : "read",
1736 obj_request->length, obj_request->img_offset,
1737 obj_request->offset);
1738 rbd_warn(rbd_dev, " result %d xferred %x\n",
1740 if (!img_request->result)
1741 img_request->result = result;
1744 if (img_request_child_test(img_request)) {
1745 rbd_assert(img_request->obj_request != NULL);
1746 more = obj_request->which < img_request->obj_request_count - 1;
1748 rbd_assert(img_request->rq != NULL);
1749 more = blk_end_request(img_request->rq, result, xferred);
1755 static void rbd_img_obj_callback(struct rbd_obj_request *obj_request)
1757 struct rbd_img_request *img_request;
1758 u32 which = obj_request->which;
1761 rbd_assert(obj_request_img_data_test(obj_request));
1762 img_request = obj_request->img_request;
1764 dout("%s: img %p obj %p\n", __func__, img_request, obj_request);
1765 rbd_assert(img_request != NULL);
1766 rbd_assert(img_request->obj_request_count > 0);
1767 rbd_assert(which != BAD_WHICH);
1768 rbd_assert(which < img_request->obj_request_count);
1769 rbd_assert(which >= img_request->next_completion);
1771 spin_lock_irq(&img_request->completion_lock);
1772 if (which != img_request->next_completion)
1775 for_each_obj_request_from(img_request, obj_request) {
1777 rbd_assert(which < img_request->obj_request_count);
1779 if (!obj_request_done_test(obj_request))
1781 more = rbd_img_obj_end_request(obj_request);
1785 rbd_assert(more ^ (which == img_request->obj_request_count));
1786 img_request->next_completion = which;
1788 spin_unlock_irq(&img_request->completion_lock);
1791 rbd_img_request_complete(img_request);
1794 static int rbd_img_request_fill_bio(struct rbd_img_request *img_request,
1795 struct bio *bio_list)
1797 struct rbd_device *rbd_dev = img_request->rbd_dev;
1798 struct rbd_obj_request *obj_request = NULL;
1799 struct rbd_obj_request *next_obj_request;
1800 bool write_request = img_request_write_test(img_request);
1801 unsigned int bio_offset;
1806 dout("%s: img %p bio %p\n", __func__, img_request, bio_list);
1808 opcode = write_request ? CEPH_OSD_OP_WRITE : CEPH_OSD_OP_READ;
1810 img_offset = img_request->offset;
1811 rbd_assert(img_offset == bio_list->bi_sector << SECTOR_SHIFT);
1812 resid = img_request->length;
1813 rbd_assert(resid > 0);
1815 struct ceph_osd_request *osd_req;
1816 const char *object_name;
1817 unsigned int clone_size;
1821 object_name = rbd_segment_name(rbd_dev, img_offset);
1824 offset = rbd_segment_offset(rbd_dev, img_offset);
1825 length = rbd_segment_length(rbd_dev, img_offset, resid);
1826 obj_request = rbd_obj_request_create(object_name,
1829 kfree(object_name); /* object request has its own copy */
1833 rbd_assert(length <= (u64) UINT_MAX);
1834 clone_size = (unsigned int) length;
1835 obj_request->bio_list = bio_chain_clone_range(&bio_list,
1836 &bio_offset, clone_size,
1838 if (!obj_request->bio_list)
1841 osd_req = rbd_osd_req_create(rbd_dev, write_request,
1845 obj_request->osd_req = osd_req;
1846 obj_request->callback = rbd_img_obj_callback;
1848 osd_req_op_extent_init(osd_req, 0, opcode, offset, length,
1850 osd_req_op_extent_osd_data_bio(osd_req, 0,
1851 obj_request->bio_list, obj_request->length);
1854 rbd_osd_req_format_write(obj_request);
1856 rbd_osd_req_format_read(obj_request);
1858 obj_request->img_offset = img_offset;
1859 rbd_img_obj_request_add(img_request, obj_request);
1861 img_offset += length;
1868 rbd_obj_request_put(obj_request);
1870 for_each_obj_request_safe(img_request, obj_request, next_obj_request)
1871 rbd_obj_request_put(obj_request);
1876 static void rbd_img_obj_exists_callback(struct rbd_obj_request *obj_request)
1878 struct rbd_obj_request *orig_request;
1881 rbd_assert(!obj_request_img_data_test(obj_request));
1884 * All we need from the object request is the original
1885 * request and the result of the STAT op. Grab those, then
1886 * we're done with the request.
1888 orig_request = obj_request->obj_request;
1889 obj_request->obj_request = NULL;
1890 rbd_assert(orig_request);
1891 rbd_assert(orig_request->img_request);
1893 result = obj_request->result;
1894 obj_request->result = 0;
1896 dout("%s: obj %p for obj %p result %d %llu/%llu\n", __func__,
1897 obj_request, orig_request, result,
1898 obj_request->xferred, obj_request->length);
1899 rbd_obj_request_put(obj_request);
1901 rbd_assert(orig_request);
1902 rbd_assert(orig_request->img_request);
1905 * Our only purpose here is to determine whether the object
1906 * exists, and we don't want to treat the non-existence as
1907 * an error. If something else comes back, transfer the
1908 * error to the original request and complete it now.
1911 obj_request_existence_set(orig_request, true);
1912 } else if (result == -ENOENT) {
1913 obj_request_existence_set(orig_request, false);
1914 } else if (result) {
1915 orig_request->result = result;
1920 * Resubmit the original request now that we have recorded
1921 * whether the target object exists.
1923 orig_request->result = rbd_img_obj_request_submit(orig_request);
1925 if (orig_request->result)
1926 rbd_obj_request_complete(orig_request);
1927 rbd_obj_request_put(orig_request);
1930 static int rbd_img_obj_exists_submit(struct rbd_obj_request *obj_request)
1932 struct rbd_obj_request *stat_request;
1933 struct rbd_device *rbd_dev;
1934 struct ceph_osd_client *osdc;
1935 struct page **pages = NULL;
1941 * The response data for a STAT call consists of:
1948 size = sizeof (__le64) + sizeof (__le32) + sizeof (__le32);
1949 page_count = (u32)calc_pages_for(0, size);
1950 pages = ceph_alloc_page_vector(page_count, GFP_KERNEL);
1952 return PTR_ERR(pages);
1955 stat_request = rbd_obj_request_create(obj_request->object_name, 0, 0,
1960 rbd_obj_request_get(obj_request);
1961 stat_request->obj_request = obj_request;
1962 stat_request->pages = pages;
1963 stat_request->page_count = page_count;
1965 rbd_assert(obj_request->img_request);
1966 rbd_dev = obj_request->img_request->rbd_dev;
1967 stat_request->osd_req = rbd_osd_req_create(rbd_dev, false,
1969 if (!stat_request->osd_req)
1971 stat_request->callback = rbd_img_obj_exists_callback;
1973 osd_req_op_init(stat_request->osd_req, 0, CEPH_OSD_OP_STAT);
1974 osd_req_op_raw_data_in_pages(stat_request->osd_req, 0, pages, size, 0,
1976 rbd_osd_req_format_read(stat_request);
1978 osdc = &rbd_dev->rbd_client->client->osdc;
1979 ret = rbd_obj_request_submit(osdc, stat_request);
1982 rbd_obj_request_put(obj_request);
1987 static int rbd_img_obj_request_submit(struct rbd_obj_request *obj_request)
1989 struct rbd_img_request *img_request;
1991 rbd_assert(obj_request_img_data_test(obj_request));
1993 img_request = obj_request->img_request;
1994 rbd_assert(img_request);
1996 /* (At the moment we don't care whether it exists or not...) */
1997 (void) obj_request_exists_test;
2000 * Only layered writes need special handling. If it's not a
2001 * layered write, or it is a layered write but we know the
2002 * target object exists, it's no different from any other
2005 if (!img_request_write_test(img_request) ||
2006 !img_request_layered_test(img_request) ||
2007 obj_request_known_test(obj_request)) {
2009 struct rbd_device *rbd_dev;
2010 struct ceph_osd_client *osdc;
2012 rbd_dev = obj_request->img_request->rbd_dev;
2013 osdc = &rbd_dev->rbd_client->client->osdc;
2015 return rbd_obj_request_submit(osdc, obj_request);
2019 * It's a layered write and we don't know whether the target
2020 * exists. Issue existence check; once that completes the
2021 * original request will be submitted again.
2024 return rbd_img_obj_exists_submit(obj_request);
2027 static int rbd_img_request_submit(struct rbd_img_request *img_request)
2029 struct rbd_obj_request *obj_request;
2030 struct rbd_obj_request *next_obj_request;
2032 dout("%s: img %p\n", __func__, img_request);
2033 for_each_obj_request_safe(img_request, obj_request, next_obj_request) {
2036 ret = rbd_img_obj_request_submit(obj_request);
2044 static void rbd_img_parent_read_callback(struct rbd_img_request *img_request)
2046 struct rbd_obj_request *obj_request;
2048 rbd_assert(img_request_child_test(img_request));
2050 obj_request = img_request->obj_request;
2051 rbd_assert(obj_request != NULL);
2052 obj_request->result = img_request->result;
2053 obj_request->xferred = img_request->xferred;
2055 rbd_img_obj_request_read_callback(obj_request);
2056 rbd_obj_request_complete(obj_request);
2059 static void rbd_img_parent_read(struct rbd_obj_request *obj_request)
2061 struct rbd_device *rbd_dev;
2062 struct rbd_img_request *img_request;
2065 rbd_assert(obj_request_img_data_test(obj_request));
2066 rbd_assert(obj_request->img_request != NULL);
2067 rbd_assert(obj_request->result == (s32) -ENOENT);
2068 rbd_assert(obj_request->type == OBJ_REQUEST_BIO);
2070 rbd_dev = obj_request->img_request->rbd_dev;
2071 rbd_assert(rbd_dev->parent != NULL);
2072 /* rbd_read_finish(obj_request, obj_request->length); */
2073 img_request = rbd_img_request_create(rbd_dev->parent,
2074 obj_request->img_offset,
2075 obj_request->length,
2081 rbd_obj_request_get(obj_request);
2082 img_request->obj_request = obj_request;
2084 result = rbd_img_request_fill_bio(img_request, obj_request->bio_list);
2088 img_request->callback = rbd_img_parent_read_callback;
2089 result = rbd_img_request_submit(img_request);
2096 rbd_img_request_put(img_request);
2097 obj_request->result = result;
2098 obj_request->xferred = 0;
2099 obj_request_done_set(obj_request);
2102 static int rbd_obj_notify_ack(struct rbd_device *rbd_dev,
2103 u64 ver, u64 notify_id)
2105 struct rbd_obj_request *obj_request;
2106 struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
2109 obj_request = rbd_obj_request_create(rbd_dev->header_name, 0, 0,
2110 OBJ_REQUEST_NODATA);
2115 obj_request->osd_req = rbd_osd_req_create(rbd_dev, false, obj_request);
2116 if (!obj_request->osd_req)
2118 obj_request->callback = rbd_obj_request_put;
2120 osd_req_op_watch_init(obj_request->osd_req, 0, CEPH_OSD_OP_NOTIFY_ACK,
2122 rbd_osd_req_format_read(obj_request);
2124 ret = rbd_obj_request_submit(osdc, obj_request);
2127 rbd_obj_request_put(obj_request);
2132 static void rbd_watch_cb(u64 ver, u64 notify_id, u8 opcode, void *data)
2134 struct rbd_device *rbd_dev = (struct rbd_device *)data;
2141 dout("%s: \"%s\" notify_id %llu opcode %u\n", __func__,
2142 rbd_dev->header_name, (unsigned long long) notify_id,
2143 (unsigned int) opcode);
2144 rc = rbd_dev_refresh(rbd_dev, &hver);
2146 rbd_warn(rbd_dev, "got notification but failed to "
2147 " update snaps: %d\n", rc);
2149 rbd_obj_notify_ack(rbd_dev, hver, notify_id);
2153 * Request sync osd watch/unwatch. The value of "start" determines
2154 * whether a watch request is being initiated or torn down.
2156 static int rbd_dev_header_watch_sync(struct rbd_device *rbd_dev, int start)
2158 struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
2159 struct rbd_obj_request *obj_request;
2162 rbd_assert(start ^ !!rbd_dev->watch_event);
2163 rbd_assert(start ^ !!rbd_dev->watch_request);
2166 ret = ceph_osdc_create_event(osdc, rbd_watch_cb, rbd_dev,
2167 &rbd_dev->watch_event);
2170 rbd_assert(rbd_dev->watch_event != NULL);
2174 obj_request = rbd_obj_request_create(rbd_dev->header_name, 0, 0,
2175 OBJ_REQUEST_NODATA);
2179 obj_request->osd_req = rbd_osd_req_create(rbd_dev, true, obj_request);
2180 if (!obj_request->osd_req)
2184 ceph_osdc_set_request_linger(osdc, obj_request->osd_req);
2186 ceph_osdc_unregister_linger_request(osdc,
2187 rbd_dev->watch_request->osd_req);
2189 osd_req_op_watch_init(obj_request->osd_req, 0, CEPH_OSD_OP_WATCH,
2190 rbd_dev->watch_event->cookie,
2191 rbd_dev->header.obj_version, start);
2192 rbd_osd_req_format_write(obj_request);
2194 ret = rbd_obj_request_submit(osdc, obj_request);
2197 ret = rbd_obj_request_wait(obj_request);
2200 ret = obj_request->result;
2205 * A watch request is set to linger, so the underlying osd
2206 * request won't go away until we unregister it. We retain
2207 * a pointer to the object request during that time (in
2208 * rbd_dev->watch_request), so we'll keep a reference to
2209 * it. We'll drop that reference (below) after we've
2213 rbd_dev->watch_request = obj_request;
2218 /* We have successfully torn down the watch request */
2220 rbd_obj_request_put(rbd_dev->watch_request);
2221 rbd_dev->watch_request = NULL;
2223 /* Cancel the event if we're tearing down, or on error */
2224 ceph_osdc_cancel_event(rbd_dev->watch_event);
2225 rbd_dev->watch_event = NULL;
2227 rbd_obj_request_put(obj_request);
2233 * Synchronous osd object method call
2235 static int rbd_obj_method_sync(struct rbd_device *rbd_dev,
2236 const char *object_name,
2237 const char *class_name,
2238 const char *method_name,
2239 const char *outbound,
2240 size_t outbound_size,
2242 size_t inbound_size,
2245 struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
2246 struct rbd_obj_request *obj_request;
2247 struct page **pages;
2252 * Method calls are ultimately read operations. The result
2253 * should placed into the inbound buffer provided. They
2254 * also supply outbound data--parameters for the object
2255 * method. Currently if this is present it will be a
2258 page_count = (u32) calc_pages_for(0, inbound_size);
2259 pages = ceph_alloc_page_vector(page_count, GFP_KERNEL);
2261 return PTR_ERR(pages);
2264 obj_request = rbd_obj_request_create(object_name, 0, inbound_size,
2269 obj_request->pages = pages;
2270 obj_request->page_count = page_count;
2272 obj_request->osd_req = rbd_osd_req_create(rbd_dev, false, obj_request);
2273 if (!obj_request->osd_req)
2276 osd_req_op_cls_init(obj_request->osd_req, 0, CEPH_OSD_OP_CALL,
2277 class_name, method_name);
2278 if (outbound_size) {
2279 struct ceph_pagelist *pagelist;
2281 pagelist = kmalloc(sizeof (*pagelist), GFP_NOFS);
2285 ceph_pagelist_init(pagelist);
2286 ceph_pagelist_append(pagelist, outbound, outbound_size);
2287 osd_req_op_cls_request_data_pagelist(obj_request->osd_req, 0,
2290 osd_req_op_cls_response_data_pages(obj_request->osd_req, 0,
2291 obj_request->pages, inbound_size,
2293 rbd_osd_req_format_read(obj_request);
2295 ret = rbd_obj_request_submit(osdc, obj_request);
2298 ret = rbd_obj_request_wait(obj_request);
2302 ret = obj_request->result;
2306 ceph_copy_from_page_vector(pages, inbound, 0, obj_request->xferred);
2308 *version = obj_request->version;
2311 rbd_obj_request_put(obj_request);
2313 ceph_release_page_vector(pages, page_count);
2318 static void rbd_request_fn(struct request_queue *q)
2319 __releases(q->queue_lock) __acquires(q->queue_lock)
2321 struct rbd_device *rbd_dev = q->queuedata;
2322 bool read_only = rbd_dev->mapping.read_only;
2326 while ((rq = blk_fetch_request(q))) {
2327 bool write_request = rq_data_dir(rq) == WRITE;
2328 struct rbd_img_request *img_request;
2332 /* Ignore any non-FS requests that filter through. */
2334 if (rq->cmd_type != REQ_TYPE_FS) {
2335 dout("%s: non-fs request type %d\n", __func__,
2336 (int) rq->cmd_type);
2337 __blk_end_request_all(rq, 0);
2341 /* Ignore/skip any zero-length requests */
2343 offset = (u64) blk_rq_pos(rq) << SECTOR_SHIFT;
2344 length = (u64) blk_rq_bytes(rq);
2347 dout("%s: zero-length request\n", __func__);
2348 __blk_end_request_all(rq, 0);
2352 spin_unlock_irq(q->queue_lock);
2354 /* Disallow writes to a read-only device */
2356 if (write_request) {
2360 rbd_assert(rbd_dev->spec->snap_id == CEPH_NOSNAP);
2364 * Quit early if the mapped snapshot no longer
2365 * exists. It's still possible the snapshot will
2366 * have disappeared by the time our request arrives
2367 * at the osd, but there's no sense in sending it if
2370 if (!test_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags)) {
2371 dout("request for non-existent snapshot");
2372 rbd_assert(rbd_dev->spec->snap_id != CEPH_NOSNAP);
2378 if (WARN_ON(offset && length > U64_MAX - offset + 1))
2379 goto end_request; /* Shouldn't happen */
2382 img_request = rbd_img_request_create(rbd_dev, offset, length,
2383 write_request, false);
2387 img_request->rq = rq;
2389 result = rbd_img_request_fill_bio(img_request, rq->bio);
2391 result = rbd_img_request_submit(img_request);
2393 rbd_img_request_put(img_request);
2395 spin_lock_irq(q->queue_lock);
2397 rbd_warn(rbd_dev, "%s %llx at %llx result %d\n",
2398 write_request ? "write" : "read",
2399 length, offset, result);
2401 __blk_end_request_all(rq, result);
2407 * a queue callback. Makes sure that we don't create a bio that spans across
2408 * multiple osd objects. One exception would be with a single page bios,
2409 * which we handle later at bio_chain_clone_range()
2411 static int rbd_merge_bvec(struct request_queue *q, struct bvec_merge_data *bmd,
2412 struct bio_vec *bvec)
2414 struct rbd_device *rbd_dev = q->queuedata;
2415 sector_t sector_offset;
2416 sector_t sectors_per_obj;
2417 sector_t obj_sector_offset;
2421 * Find how far into its rbd object the partition-relative
2422 * bio start sector is to offset relative to the enclosing
2425 sector_offset = get_start_sect(bmd->bi_bdev) + bmd->bi_sector;
2426 sectors_per_obj = 1 << (rbd_dev->header.obj_order - SECTOR_SHIFT);
2427 obj_sector_offset = sector_offset & (sectors_per_obj - 1);
2430 * Compute the number of bytes from that offset to the end
2431 * of the object. Account for what's already used by the bio.
2433 ret = (int) (sectors_per_obj - obj_sector_offset) << SECTOR_SHIFT;
2434 if (ret > bmd->bi_size)
2435 ret -= bmd->bi_size;
2440 * Don't send back more than was asked for. And if the bio
2441 * was empty, let the whole thing through because: "Note
2442 * that a block device *must* allow a single page to be
2443 * added to an empty bio."
2445 rbd_assert(bvec->bv_len <= PAGE_SIZE);
2446 if (ret > (int) bvec->bv_len || !bmd->bi_size)
2447 ret = (int) bvec->bv_len;
2452 static void rbd_free_disk(struct rbd_device *rbd_dev)
2454 struct gendisk *disk = rbd_dev->disk;
2459 if (disk->flags & GENHD_FL_UP)
2462 blk_cleanup_queue(disk->queue);
2466 static int rbd_obj_read_sync(struct rbd_device *rbd_dev,
2467 const char *object_name,
2468 u64 offset, u64 length,
2469 char *buf, u64 *version)
2472 struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
2473 struct rbd_obj_request *obj_request;
2474 struct page **pages = NULL;
2479 page_count = (u32) calc_pages_for(offset, length);
2480 pages = ceph_alloc_page_vector(page_count, GFP_KERNEL);
2482 ret = PTR_ERR(pages);
2485 obj_request = rbd_obj_request_create(object_name, offset, length,
2490 obj_request->pages = pages;
2491 obj_request->page_count = page_count;
2493 obj_request->osd_req = rbd_osd_req_create(rbd_dev, false, obj_request);
2494 if (!obj_request->osd_req)
2497 osd_req_op_extent_init(obj_request->osd_req, 0, CEPH_OSD_OP_READ,
2498 offset, length, 0, 0);
2499 osd_req_op_extent_osd_data_pages(obj_request->osd_req, 0,
2501 obj_request->length,
2502 obj_request->offset & ~PAGE_MASK,
2504 rbd_osd_req_format_read(obj_request);
2506 ret = rbd_obj_request_submit(osdc, obj_request);
2509 ret = rbd_obj_request_wait(obj_request);
2513 ret = obj_request->result;
2517 rbd_assert(obj_request->xferred <= (u64) SIZE_MAX);
2518 size = (size_t) obj_request->xferred;
2519 ceph_copy_from_page_vector(pages, buf, 0, size);
2520 rbd_assert(size <= (size_t) INT_MAX);
2523 *version = obj_request->version;
2526 rbd_obj_request_put(obj_request);
2528 ceph_release_page_vector(pages, page_count);
2534 * Read the complete header for the given rbd device.
2536 * Returns a pointer to a dynamically-allocated buffer containing
2537 * the complete and validated header. Caller can pass the address
2538 * of a variable that will be filled in with the version of the
2539 * header object at the time it was read.
2541 * Returns a pointer-coded errno if a failure occurs.
2543 static struct rbd_image_header_ondisk *
2544 rbd_dev_v1_header_read(struct rbd_device *rbd_dev, u64 *version)
2546 struct rbd_image_header_ondisk *ondisk = NULL;
2553 * The complete header will include an array of its 64-bit
2554 * snapshot ids, followed by the names of those snapshots as
2555 * a contiguous block of NUL-terminated strings. Note that
2556 * the number of snapshots could change by the time we read
2557 * it in, in which case we re-read it.
2564 size = sizeof (*ondisk);
2565 size += snap_count * sizeof (struct rbd_image_snap_ondisk);
2567 ondisk = kmalloc(size, GFP_KERNEL);
2569 return ERR_PTR(-ENOMEM);
2571 ret = rbd_obj_read_sync(rbd_dev, rbd_dev->header_name,
2573 (char *) ondisk, version);
2576 if (WARN_ON((size_t) ret < size)) {
2578 rbd_warn(rbd_dev, "short header read (want %zd got %d)",
2582 if (!rbd_dev_ondisk_valid(ondisk)) {
2584 rbd_warn(rbd_dev, "invalid header");
2588 names_size = le64_to_cpu(ondisk->snap_names_len);
2589 want_count = snap_count;
2590 snap_count = le32_to_cpu(ondisk->snap_count);
2591 } while (snap_count != want_count);
2598 return ERR_PTR(ret);
2602 * reload the ondisk the header
2604 static int rbd_read_header(struct rbd_device *rbd_dev,
2605 struct rbd_image_header *header)
2607 struct rbd_image_header_ondisk *ondisk;
2611 ondisk = rbd_dev_v1_header_read(rbd_dev, &ver);
2613 return PTR_ERR(ondisk);
2614 ret = rbd_header_from_disk(header, ondisk);
2616 header->obj_version = ver;
2622 static void rbd_remove_all_snaps(struct rbd_device *rbd_dev)
2624 struct rbd_snap *snap;
2625 struct rbd_snap *next;
2627 list_for_each_entry_safe(snap, next, &rbd_dev->snaps, node)
2628 rbd_remove_snap_dev(snap);
2631 static void rbd_update_mapping_size(struct rbd_device *rbd_dev)
2635 if (rbd_dev->spec->snap_id != CEPH_NOSNAP)
2638 size = (sector_t) rbd_dev->header.image_size / SECTOR_SIZE;
2639 dout("setting size to %llu sectors", (unsigned long long) size);
2640 rbd_dev->mapping.size = (u64) size;
2641 set_capacity(rbd_dev->disk, size);
2645 * only read the first part of the ondisk header, without the snaps info
2647 static int rbd_dev_v1_refresh(struct rbd_device *rbd_dev, u64 *hver)
2650 struct rbd_image_header h;
2652 ret = rbd_read_header(rbd_dev, &h);
2656 down_write(&rbd_dev->header_rwsem);
2658 /* Update image size, and check for resize of mapped image */
2659 rbd_dev->header.image_size = h.image_size;
2660 rbd_update_mapping_size(rbd_dev);
2662 /* rbd_dev->header.object_prefix shouldn't change */
2663 kfree(rbd_dev->header.snap_sizes);
2664 kfree(rbd_dev->header.snap_names);
2665 /* osd requests may still refer to snapc */
2666 ceph_put_snap_context(rbd_dev->header.snapc);
2669 *hver = h.obj_version;
2670 rbd_dev->header.obj_version = h.obj_version;
2671 rbd_dev->header.image_size = h.image_size;
2672 rbd_dev->header.snapc = h.snapc;
2673 rbd_dev->header.snap_names = h.snap_names;
2674 rbd_dev->header.snap_sizes = h.snap_sizes;
2675 /* Free the extra copy of the object prefix */
2676 WARN_ON(strcmp(rbd_dev->header.object_prefix, h.object_prefix));
2677 kfree(h.object_prefix);
2679 ret = rbd_dev_snaps_update(rbd_dev);
2681 ret = rbd_dev_snaps_register(rbd_dev);
2683 up_write(&rbd_dev->header_rwsem);
2688 static int rbd_dev_refresh(struct rbd_device *rbd_dev, u64 *hver)
2692 rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
2693 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
2694 if (rbd_dev->image_format == 1)
2695 ret = rbd_dev_v1_refresh(rbd_dev, hver);
2697 ret = rbd_dev_v2_refresh(rbd_dev, hver);
2698 mutex_unlock(&ctl_mutex);
2703 static int rbd_init_disk(struct rbd_device *rbd_dev)
2705 struct gendisk *disk;
2706 struct request_queue *q;
2709 /* create gendisk info */
2710 disk = alloc_disk(RBD_MINORS_PER_MAJOR);
2714 snprintf(disk->disk_name, sizeof(disk->disk_name), RBD_DRV_NAME "%d",
2716 disk->major = rbd_dev->major;
2717 disk->first_minor = 0;
2718 disk->fops = &rbd_bd_ops;
2719 disk->private_data = rbd_dev;
2721 q = blk_init_queue(rbd_request_fn, &rbd_dev->lock);
2725 /* We use the default size, but let's be explicit about it. */
2726 blk_queue_physical_block_size(q, SECTOR_SIZE);
2728 /* set io sizes to object size */
2729 segment_size = rbd_obj_bytes(&rbd_dev->header);
2730 blk_queue_max_hw_sectors(q, segment_size / SECTOR_SIZE);
2731 blk_queue_max_segment_size(q, segment_size);
2732 blk_queue_io_min(q, segment_size);
2733 blk_queue_io_opt(q, segment_size);
2735 blk_queue_merge_bvec(q, rbd_merge_bvec);
2738 q->queuedata = rbd_dev;
2740 rbd_dev->disk = disk;
2742 set_capacity(rbd_dev->disk, rbd_dev->mapping.size / SECTOR_SIZE);
2755 static struct rbd_device *dev_to_rbd_dev(struct device *dev)
2757 return container_of(dev, struct rbd_device, dev);
2760 static ssize_t rbd_size_show(struct device *dev,
2761 struct device_attribute *attr, char *buf)
2763 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
2766 down_read(&rbd_dev->header_rwsem);
2767 size = get_capacity(rbd_dev->disk);
2768 up_read(&rbd_dev->header_rwsem);
2770 return sprintf(buf, "%llu\n", (unsigned long long) size * SECTOR_SIZE);
2774 * Note this shows the features for whatever's mapped, which is not
2775 * necessarily the base image.
2777 static ssize_t rbd_features_show(struct device *dev,
2778 struct device_attribute *attr, char *buf)
2780 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
2782 return sprintf(buf, "0x%016llx\n",
2783 (unsigned long long) rbd_dev->mapping.features);
2786 static ssize_t rbd_major_show(struct device *dev,
2787 struct device_attribute *attr, char *buf)
2789 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
2791 return sprintf(buf, "%d\n", rbd_dev->major);
2794 static ssize_t rbd_client_id_show(struct device *dev,
2795 struct device_attribute *attr, char *buf)
2797 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
2799 return sprintf(buf, "client%lld\n",
2800 ceph_client_id(rbd_dev->rbd_client->client));
2803 static ssize_t rbd_pool_show(struct device *dev,
2804 struct device_attribute *attr, char *buf)
2806 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
2808 return sprintf(buf, "%s\n", rbd_dev->spec->pool_name);
2811 static ssize_t rbd_pool_id_show(struct device *dev,
2812 struct device_attribute *attr, char *buf)
2814 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
2816 return sprintf(buf, "%llu\n",
2817 (unsigned long long) rbd_dev->spec->pool_id);
2820 static ssize_t rbd_name_show(struct device *dev,
2821 struct device_attribute *attr, char *buf)
2823 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
2825 if (rbd_dev->spec->image_name)
2826 return sprintf(buf, "%s\n", rbd_dev->spec->image_name);
2828 return sprintf(buf, "(unknown)\n");
2831 static ssize_t rbd_image_id_show(struct device *dev,
2832 struct device_attribute *attr, char *buf)
2834 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
2836 return sprintf(buf, "%s\n", rbd_dev->spec->image_id);
2840 * Shows the name of the currently-mapped snapshot (or
2841 * RBD_SNAP_HEAD_NAME for the base image).
2843 static ssize_t rbd_snap_show(struct device *dev,
2844 struct device_attribute *attr,
2847 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
2849 return sprintf(buf, "%s\n", rbd_dev->spec->snap_name);
2853 * For an rbd v2 image, shows the pool id, image id, and snapshot id
2854 * for the parent image. If there is no parent, simply shows
2855 * "(no parent image)".
2857 static ssize_t rbd_parent_show(struct device *dev,
2858 struct device_attribute *attr,
2861 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
2862 struct rbd_spec *spec = rbd_dev->parent_spec;
2867 return sprintf(buf, "(no parent image)\n");
2869 count = sprintf(bufp, "pool_id %llu\npool_name %s\n",
2870 (unsigned long long) spec->pool_id, spec->pool_name);
2875 count = sprintf(bufp, "image_id %s\nimage_name %s\n", spec->image_id,
2876 spec->image_name ? spec->image_name : "(unknown)");
2881 count = sprintf(bufp, "snap_id %llu\nsnap_name %s\n",
2882 (unsigned long long) spec->snap_id, spec->snap_name);
2887 count = sprintf(bufp, "overlap %llu\n", rbd_dev->parent_overlap);
2892 return (ssize_t) (bufp - buf);
2895 static ssize_t rbd_image_refresh(struct device *dev,
2896 struct device_attribute *attr,
2900 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
2903 ret = rbd_dev_refresh(rbd_dev, NULL);
2905 return ret < 0 ? ret : size;
2908 static DEVICE_ATTR(size, S_IRUGO, rbd_size_show, NULL);
2909 static DEVICE_ATTR(features, S_IRUGO, rbd_features_show, NULL);
2910 static DEVICE_ATTR(major, S_IRUGO, rbd_major_show, NULL);
2911 static DEVICE_ATTR(client_id, S_IRUGO, rbd_client_id_show, NULL);
2912 static DEVICE_ATTR(pool, S_IRUGO, rbd_pool_show, NULL);
2913 static DEVICE_ATTR(pool_id, S_IRUGO, rbd_pool_id_show, NULL);
2914 static DEVICE_ATTR(name, S_IRUGO, rbd_name_show, NULL);
2915 static DEVICE_ATTR(image_id, S_IRUGO, rbd_image_id_show, NULL);
2916 static DEVICE_ATTR(refresh, S_IWUSR, NULL, rbd_image_refresh);
2917 static DEVICE_ATTR(current_snap, S_IRUGO, rbd_snap_show, NULL);
2918 static DEVICE_ATTR(parent, S_IRUGO, rbd_parent_show, NULL);
2920 static struct attribute *rbd_attrs[] = {
2921 &dev_attr_size.attr,
2922 &dev_attr_features.attr,
2923 &dev_attr_major.attr,
2924 &dev_attr_client_id.attr,
2925 &dev_attr_pool.attr,
2926 &dev_attr_pool_id.attr,
2927 &dev_attr_name.attr,
2928 &dev_attr_image_id.attr,
2929 &dev_attr_current_snap.attr,
2930 &dev_attr_parent.attr,
2931 &dev_attr_refresh.attr,
2935 static struct attribute_group rbd_attr_group = {
2939 static const struct attribute_group *rbd_attr_groups[] = {
2944 static void rbd_sysfs_dev_release(struct device *dev)
2948 static struct device_type rbd_device_type = {
2950 .groups = rbd_attr_groups,
2951 .release = rbd_sysfs_dev_release,
2959 static ssize_t rbd_snap_size_show(struct device *dev,
2960 struct device_attribute *attr,
2963 struct rbd_snap *snap = container_of(dev, struct rbd_snap, dev);
2965 return sprintf(buf, "%llu\n", (unsigned long long)snap->size);
2968 static ssize_t rbd_snap_id_show(struct device *dev,
2969 struct device_attribute *attr,
2972 struct rbd_snap *snap = container_of(dev, struct rbd_snap, dev);
2974 return sprintf(buf, "%llu\n", (unsigned long long)snap->id);
2977 static ssize_t rbd_snap_features_show(struct device *dev,
2978 struct device_attribute *attr,
2981 struct rbd_snap *snap = container_of(dev, struct rbd_snap, dev);
2983 return sprintf(buf, "0x%016llx\n",
2984 (unsigned long long) snap->features);
2987 static DEVICE_ATTR(snap_size, S_IRUGO, rbd_snap_size_show, NULL);
2988 static DEVICE_ATTR(snap_id, S_IRUGO, rbd_snap_id_show, NULL);
2989 static DEVICE_ATTR(snap_features, S_IRUGO, rbd_snap_features_show, NULL);
2991 static struct attribute *rbd_snap_attrs[] = {
2992 &dev_attr_snap_size.attr,
2993 &dev_attr_snap_id.attr,
2994 &dev_attr_snap_features.attr,
2998 static struct attribute_group rbd_snap_attr_group = {
2999 .attrs = rbd_snap_attrs,
3002 static void rbd_snap_dev_release(struct device *dev)
3004 struct rbd_snap *snap = container_of(dev, struct rbd_snap, dev);
3009 static const struct attribute_group *rbd_snap_attr_groups[] = {
3010 &rbd_snap_attr_group,
3014 static struct device_type rbd_snap_device_type = {
3015 .groups = rbd_snap_attr_groups,
3016 .release = rbd_snap_dev_release,
3019 static struct rbd_spec *rbd_spec_get(struct rbd_spec *spec)
3021 kref_get(&spec->kref);
3026 static void rbd_spec_free(struct kref *kref);
3027 static void rbd_spec_put(struct rbd_spec *spec)
3030 kref_put(&spec->kref, rbd_spec_free);
3033 static struct rbd_spec *rbd_spec_alloc(void)
3035 struct rbd_spec *spec;
3037 spec = kzalloc(sizeof (*spec), GFP_KERNEL);
3040 kref_init(&spec->kref);
3045 static void rbd_spec_free(struct kref *kref)
3047 struct rbd_spec *spec = container_of(kref, struct rbd_spec, kref);
3049 kfree(spec->pool_name);
3050 kfree(spec->image_id);
3051 kfree(spec->image_name);
3052 kfree(spec->snap_name);
3056 static struct rbd_device *rbd_dev_create(struct rbd_client *rbdc,
3057 struct rbd_spec *spec)
3059 struct rbd_device *rbd_dev;
3061 rbd_dev = kzalloc(sizeof (*rbd_dev), GFP_KERNEL);
3065 spin_lock_init(&rbd_dev->lock);
3067 INIT_LIST_HEAD(&rbd_dev->node);
3068 INIT_LIST_HEAD(&rbd_dev->snaps);
3069 init_rwsem(&rbd_dev->header_rwsem);
3071 rbd_dev->spec = spec;
3072 rbd_dev->rbd_client = rbdc;
3074 /* Initialize the layout used for all rbd requests */
3076 rbd_dev->layout.fl_stripe_unit = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER);
3077 rbd_dev->layout.fl_stripe_count = cpu_to_le32(1);
3078 rbd_dev->layout.fl_object_size = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER);
3079 rbd_dev->layout.fl_pg_pool = cpu_to_le32((u32) spec->pool_id);
3084 static void rbd_dev_destroy(struct rbd_device *rbd_dev)
3086 rbd_spec_put(rbd_dev->parent_spec);
3087 kfree(rbd_dev->header_name);
3088 rbd_put_client(rbd_dev->rbd_client);
3089 rbd_spec_put(rbd_dev->spec);
3093 static bool rbd_snap_registered(struct rbd_snap *snap)
3095 bool ret = snap->dev.type == &rbd_snap_device_type;
3096 bool reg = device_is_registered(&snap->dev);
3098 rbd_assert(!ret ^ reg);
3103 static void rbd_remove_snap_dev(struct rbd_snap *snap)
3105 list_del(&snap->node);
3106 if (device_is_registered(&snap->dev))
3107 device_unregister(&snap->dev);
3110 static int rbd_register_snap_dev(struct rbd_snap *snap,
3111 struct device *parent)
3113 struct device *dev = &snap->dev;
3116 dev->type = &rbd_snap_device_type;
3117 dev->parent = parent;
3118 dev->release = rbd_snap_dev_release;
3119 dev_set_name(dev, "%s%s", RBD_SNAP_DEV_NAME_PREFIX, snap->name);
3120 dout("%s: registering device for snapshot %s\n", __func__, snap->name);
3122 ret = device_register(dev);
3127 static struct rbd_snap *__rbd_add_snap_dev(struct rbd_device *rbd_dev,
3128 const char *snap_name,
3129 u64 snap_id, u64 snap_size,
3132 struct rbd_snap *snap;
3135 snap = kzalloc(sizeof (*snap), GFP_KERNEL);
3137 return ERR_PTR(-ENOMEM);
3140 snap->name = kstrdup(snap_name, GFP_KERNEL);
3145 snap->size = snap_size;
3146 snap->features = snap_features;
3154 return ERR_PTR(ret);
3157 static char *rbd_dev_v1_snap_info(struct rbd_device *rbd_dev, u32 which,
3158 u64 *snap_size, u64 *snap_features)
3162 rbd_assert(which < rbd_dev->header.snapc->num_snaps);
3164 *snap_size = rbd_dev->header.snap_sizes[which];
3165 *snap_features = 0; /* No features for v1 */
3167 /* Skip over names until we find the one we are looking for */
3169 snap_name = rbd_dev->header.snap_names;
3171 snap_name += strlen(snap_name) + 1;
3177 * Get the size and object order for an image snapshot, or if
3178 * snap_id is CEPH_NOSNAP, gets this information for the base
3181 static int _rbd_dev_v2_snap_size(struct rbd_device *rbd_dev, u64 snap_id,
3182 u8 *order, u64 *snap_size)
3184 __le64 snapid = cpu_to_le64(snap_id);
3189 } __attribute__ ((packed)) size_buf = { 0 };
3191 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
3193 (char *) &snapid, sizeof (snapid),
3194 (char *) &size_buf, sizeof (size_buf), NULL);
3195 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
3199 *order = size_buf.order;
3200 *snap_size = le64_to_cpu(size_buf.size);
3202 dout(" snap_id 0x%016llx order = %u, snap_size = %llu\n",
3203 (unsigned long long) snap_id, (unsigned int) *order,
3204 (unsigned long long) *snap_size);
3209 static int rbd_dev_v2_image_size(struct rbd_device *rbd_dev)
3211 return _rbd_dev_v2_snap_size(rbd_dev, CEPH_NOSNAP,
3212 &rbd_dev->header.obj_order,
3213 &rbd_dev->header.image_size);
3216 static int rbd_dev_v2_object_prefix(struct rbd_device *rbd_dev)
3222 reply_buf = kzalloc(RBD_OBJ_PREFIX_LEN_MAX, GFP_KERNEL);
3226 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
3227 "rbd", "get_object_prefix",
3229 reply_buf, RBD_OBJ_PREFIX_LEN_MAX, NULL);
3230 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
3235 rbd_dev->header.object_prefix = ceph_extract_encoded_string(&p,
3236 p + RBD_OBJ_PREFIX_LEN_MAX,
3239 if (IS_ERR(rbd_dev->header.object_prefix)) {
3240 ret = PTR_ERR(rbd_dev->header.object_prefix);
3241 rbd_dev->header.object_prefix = NULL;
3243 dout(" object_prefix = %s\n", rbd_dev->header.object_prefix);
3252 static int _rbd_dev_v2_snap_features(struct rbd_device *rbd_dev, u64 snap_id,
3255 __le64 snapid = cpu_to_le64(snap_id);
3259 } features_buf = { 0 };
3263 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
3264 "rbd", "get_features",
3265 (char *) &snapid, sizeof (snapid),
3266 (char *) &features_buf, sizeof (features_buf),
3268 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
3272 incompat = le64_to_cpu(features_buf.incompat);
3273 if (incompat & ~RBD_FEATURES_SUPPORTED)
3276 *snap_features = le64_to_cpu(features_buf.features);
3278 dout(" snap_id 0x%016llx features = 0x%016llx incompat = 0x%016llx\n",
3279 (unsigned long long) snap_id,
3280 (unsigned long long) *snap_features,
3281 (unsigned long long) le64_to_cpu(features_buf.incompat));
3286 static int rbd_dev_v2_features(struct rbd_device *rbd_dev)
3288 return _rbd_dev_v2_snap_features(rbd_dev, CEPH_NOSNAP,
3289 &rbd_dev->header.features);
3292 static int rbd_dev_v2_parent_info(struct rbd_device *rbd_dev)
3294 struct rbd_spec *parent_spec;
3296 void *reply_buf = NULL;
3304 parent_spec = rbd_spec_alloc();
3308 size = sizeof (__le64) + /* pool_id */
3309 sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX + /* image_id */
3310 sizeof (__le64) + /* snap_id */
3311 sizeof (__le64); /* overlap */
3312 reply_buf = kmalloc(size, GFP_KERNEL);
3318 snapid = cpu_to_le64(CEPH_NOSNAP);
3319 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
3320 "rbd", "get_parent",
3321 (char *) &snapid, sizeof (snapid),
3322 (char *) reply_buf, size, NULL);
3323 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
3329 end = (char *) reply_buf + size;
3330 ceph_decode_64_safe(&p, end, parent_spec->pool_id, out_err);
3331 if (parent_spec->pool_id == CEPH_NOPOOL)
3332 goto out; /* No parent? No problem. */
3334 /* The ceph file layout needs to fit pool id in 32 bits */
3337 if (WARN_ON(parent_spec->pool_id > (u64) U32_MAX))
3340 image_id = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL);
3341 if (IS_ERR(image_id)) {
3342 ret = PTR_ERR(image_id);
3345 parent_spec->image_id = image_id;
3346 ceph_decode_64_safe(&p, end, parent_spec->snap_id, out_err);
3347 ceph_decode_64_safe(&p, end, overlap, out_err);
3349 rbd_dev->parent_overlap = overlap;
3350 rbd_dev->parent_spec = parent_spec;
3351 parent_spec = NULL; /* rbd_dev now owns this */
3356 rbd_spec_put(parent_spec);
3361 static char *rbd_dev_image_name(struct rbd_device *rbd_dev)
3363 size_t image_id_size;
3368 void *reply_buf = NULL;
3370 char *image_name = NULL;
3373 rbd_assert(!rbd_dev->spec->image_name);
3375 len = strlen(rbd_dev->spec->image_id);
3376 image_id_size = sizeof (__le32) + len;
3377 image_id = kmalloc(image_id_size, GFP_KERNEL);
3382 end = (char *) image_id + image_id_size;
3383 ceph_encode_string(&p, end, rbd_dev->spec->image_id, (u32) len);
3385 size = sizeof (__le32) + RBD_IMAGE_NAME_LEN_MAX;
3386 reply_buf = kmalloc(size, GFP_KERNEL);
3390 ret = rbd_obj_method_sync(rbd_dev, RBD_DIRECTORY,
3391 "rbd", "dir_get_name",
3392 image_id, image_id_size,
3393 (char *) reply_buf, size, NULL);
3397 end = (char *) reply_buf + size;
3398 image_name = ceph_extract_encoded_string(&p, end, &len, GFP_KERNEL);
3399 if (IS_ERR(image_name))
3402 dout("%s: name is %s len is %zd\n", __func__, image_name, len);
3411 * When a parent image gets probed, we only have the pool, image,
3412 * and snapshot ids but not the names of any of them. This call
3413 * is made later to fill in those names. It has to be done after
3414 * rbd_dev_snaps_update() has completed because some of the
3415 * information (in particular, snapshot name) is not available
3418 static int rbd_dev_probe_update_spec(struct rbd_device *rbd_dev)
3420 struct ceph_osd_client *osdc;
3422 void *reply_buf = NULL;
3425 if (rbd_dev->spec->pool_name)
3426 return 0; /* Already have the names */
3428 /* Look up the pool name */
3430 osdc = &rbd_dev->rbd_client->client->osdc;
3431 name = ceph_pg_pool_name_by_id(osdc->osdmap, rbd_dev->spec->pool_id);
3433 rbd_warn(rbd_dev, "there is no pool with id %llu",
3434 rbd_dev->spec->pool_id); /* Really a BUG() */
3438 rbd_dev->spec->pool_name = kstrdup(name, GFP_KERNEL);
3439 if (!rbd_dev->spec->pool_name)
3442 /* Fetch the image name; tolerate failure here */
3444 name = rbd_dev_image_name(rbd_dev);
3446 rbd_dev->spec->image_name = (char *) name;
3448 rbd_warn(rbd_dev, "unable to get image name");
3450 /* Look up the snapshot name. */
3452 name = rbd_snap_name(rbd_dev, rbd_dev->spec->snap_id);
3454 rbd_warn(rbd_dev, "no snapshot with id %llu",
3455 rbd_dev->spec->snap_id); /* Really a BUG() */
3459 rbd_dev->spec->snap_name = kstrdup(name, GFP_KERNEL);
3460 if(!rbd_dev->spec->snap_name)
3466 kfree(rbd_dev->spec->pool_name);
3467 rbd_dev->spec->pool_name = NULL;
3472 static int rbd_dev_v2_snap_context(struct rbd_device *rbd_dev, u64 *ver)
3481 struct ceph_snap_context *snapc;
3485 * We'll need room for the seq value (maximum snapshot id),
3486 * snapshot count, and array of that many snapshot ids.
3487 * For now we have a fixed upper limit on the number we're
3488 * prepared to receive.
3490 size = sizeof (__le64) + sizeof (__le32) +
3491 RBD_MAX_SNAP_COUNT * sizeof (__le64);
3492 reply_buf = kzalloc(size, GFP_KERNEL);
3496 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
3497 "rbd", "get_snapcontext",
3499 reply_buf, size, ver);
3500 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
3506 end = (char *) reply_buf + size;
3507 ceph_decode_64_safe(&p, end, seq, out);
3508 ceph_decode_32_safe(&p, end, snap_count, out);
3511 * Make sure the reported number of snapshot ids wouldn't go
3512 * beyond the end of our buffer. But before checking that,
3513 * make sure the computed size of the snapshot context we
3514 * allocate is representable in a size_t.
3516 if (snap_count > (SIZE_MAX - sizeof (struct ceph_snap_context))
3521 if (!ceph_has_room(&p, end, snap_count * sizeof (__le64)))
3524 size = sizeof (struct ceph_snap_context) +
3525 snap_count * sizeof (snapc->snaps[0]);
3526 snapc = kmalloc(size, GFP_KERNEL);
3532 atomic_set(&snapc->nref, 1);
3534 snapc->num_snaps = snap_count;
3535 for (i = 0; i < snap_count; i++)
3536 snapc->snaps[i] = ceph_decode_64(&p);
3538 rbd_dev->header.snapc = snapc;
3540 dout(" snap context seq = %llu, snap_count = %u\n",
3541 (unsigned long long) seq, (unsigned int) snap_count);
3549 static char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev, u32 which)
3559 size = sizeof (__le32) + RBD_MAX_SNAP_NAME_LEN;
3560 reply_buf = kmalloc(size, GFP_KERNEL);
3562 return ERR_PTR(-ENOMEM);
3564 snap_id = cpu_to_le64(rbd_dev->header.snapc->snaps[which]);
3565 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
3566 "rbd", "get_snapshot_name",
3567 (char *) &snap_id, sizeof (snap_id),
3568 reply_buf, size, NULL);
3569 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
3574 end = (char *) reply_buf + size;
3575 snap_name = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL);
3576 if (IS_ERR(snap_name)) {
3577 ret = PTR_ERR(snap_name);
3580 dout(" snap_id 0x%016llx snap_name = %s\n",
3581 (unsigned long long) le64_to_cpu(snap_id), snap_name);
3589 return ERR_PTR(ret);
3592 static char *rbd_dev_v2_snap_info(struct rbd_device *rbd_dev, u32 which,
3593 u64 *snap_size, u64 *snap_features)
3599 snap_id = rbd_dev->header.snapc->snaps[which];
3600 ret = _rbd_dev_v2_snap_size(rbd_dev, snap_id, &order, snap_size);
3602 return ERR_PTR(ret);
3603 ret = _rbd_dev_v2_snap_features(rbd_dev, snap_id, snap_features);
3605 return ERR_PTR(ret);
3607 return rbd_dev_v2_snap_name(rbd_dev, which);
3610 static char *rbd_dev_snap_info(struct rbd_device *rbd_dev, u32 which,
3611 u64 *snap_size, u64 *snap_features)
3613 if (rbd_dev->image_format == 1)
3614 return rbd_dev_v1_snap_info(rbd_dev, which,
3615 snap_size, snap_features);
3616 if (rbd_dev->image_format == 2)
3617 return rbd_dev_v2_snap_info(rbd_dev, which,
3618 snap_size, snap_features);
3619 return ERR_PTR(-EINVAL);
3622 static int rbd_dev_v2_refresh(struct rbd_device *rbd_dev, u64 *hver)
3627 down_write(&rbd_dev->header_rwsem);
3629 /* Grab old order first, to see if it changes */
3631 obj_order = rbd_dev->header.obj_order,
3632 ret = rbd_dev_v2_image_size(rbd_dev);
3635 if (rbd_dev->header.obj_order != obj_order) {
3639 rbd_update_mapping_size(rbd_dev);
3641 ret = rbd_dev_v2_snap_context(rbd_dev, hver);
3642 dout("rbd_dev_v2_snap_context returned %d\n", ret);
3645 ret = rbd_dev_snaps_update(rbd_dev);
3646 dout("rbd_dev_snaps_update returned %d\n", ret);
3649 ret = rbd_dev_snaps_register(rbd_dev);
3650 dout("rbd_dev_snaps_register returned %d\n", ret);
3652 up_write(&rbd_dev->header_rwsem);
3658 * Scan the rbd device's current snapshot list and compare it to the
3659 * newly-received snapshot context. Remove any existing snapshots
3660 * not present in the new snapshot context. Add a new snapshot for
3661 * any snaphots in the snapshot context not in the current list.
3662 * And verify there are no changes to snapshots we already know
3665 * Assumes the snapshots in the snapshot context are sorted by
3666 * snapshot id, highest id first. (Snapshots in the rbd_dev's list
3667 * are also maintained in that order.)
3669 static int rbd_dev_snaps_update(struct rbd_device *rbd_dev)
3671 struct ceph_snap_context *snapc = rbd_dev->header.snapc;
3672 const u32 snap_count = snapc->num_snaps;
3673 struct list_head *head = &rbd_dev->snaps;
3674 struct list_head *links = head->next;
3677 dout("%s: snap count is %u\n", __func__, (unsigned int) snap_count);
3678 while (index < snap_count || links != head) {
3680 struct rbd_snap *snap;
3683 u64 snap_features = 0;
3685 snap_id = index < snap_count ? snapc->snaps[index]
3687 snap = links != head ? list_entry(links, struct rbd_snap, node)
3689 rbd_assert(!snap || snap->id != CEPH_NOSNAP);
3691 if (snap_id == CEPH_NOSNAP || (snap && snap->id > snap_id)) {
3692 struct list_head *next = links->next;
3695 * A previously-existing snapshot is not in
3696 * the new snap context.
3698 * If the now missing snapshot is the one the
3699 * image is mapped to, clear its exists flag
3700 * so we can avoid sending any more requests
3703 if (rbd_dev->spec->snap_id == snap->id)
3704 clear_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags);
3705 rbd_remove_snap_dev(snap);
3706 dout("%ssnap id %llu has been removed\n",
3707 rbd_dev->spec->snap_id == snap->id ?
3709 (unsigned long long) snap->id);
3711 /* Done with this list entry; advance */
3717 snap_name = rbd_dev_snap_info(rbd_dev, index,
3718 &snap_size, &snap_features);
3719 if (IS_ERR(snap_name))
3720 return PTR_ERR(snap_name);
3722 dout("entry %u: snap_id = %llu\n", (unsigned int) snap_count,
3723 (unsigned long long) snap_id);
3724 if (!snap || (snap_id != CEPH_NOSNAP && snap->id < snap_id)) {
3725 struct rbd_snap *new_snap;
3727 /* We haven't seen this snapshot before */
3729 new_snap = __rbd_add_snap_dev(rbd_dev, snap_name,
3730 snap_id, snap_size, snap_features);
3731 if (IS_ERR(new_snap)) {
3732 int err = PTR_ERR(new_snap);
3734 dout(" failed to add dev, error %d\n", err);
3739 /* New goes before existing, or at end of list */
3741 dout(" added dev%s\n", snap ? "" : " at end\n");
3743 list_add_tail(&new_snap->node, &snap->node);
3745 list_add_tail(&new_snap->node, head);
3747 /* Already have this one */
3749 dout(" already present\n");
3751 rbd_assert(snap->size == snap_size);
3752 rbd_assert(!strcmp(snap->name, snap_name));
3753 rbd_assert(snap->features == snap_features);
3755 /* Done with this list entry; advance */
3757 links = links->next;
3760 /* Advance to the next entry in the snapshot context */
3764 dout("%s: done\n", __func__);
3770 * Scan the list of snapshots and register the devices for any that
3771 * have not already been registered.
3773 static int rbd_dev_snaps_register(struct rbd_device *rbd_dev)
3775 struct rbd_snap *snap;
3778 dout("%s:\n", __func__);
3779 if (WARN_ON(!device_is_registered(&rbd_dev->dev)))
3782 list_for_each_entry(snap, &rbd_dev->snaps, node) {
3783 if (!rbd_snap_registered(snap)) {
3784 ret = rbd_register_snap_dev(snap, &rbd_dev->dev);
3789 dout("%s: returning %d\n", __func__, ret);
3794 static int rbd_bus_add_dev(struct rbd_device *rbd_dev)
3799 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
3801 dev = &rbd_dev->dev;
3802 dev->bus = &rbd_bus_type;
3803 dev->type = &rbd_device_type;
3804 dev->parent = &rbd_root_dev;
3805 dev->release = rbd_dev_release;
3806 dev_set_name(dev, "%d", rbd_dev->dev_id);
3807 ret = device_register(dev);
3809 mutex_unlock(&ctl_mutex);
3814 static void rbd_bus_del_dev(struct rbd_device *rbd_dev)
3816 device_unregister(&rbd_dev->dev);
3819 static atomic64_t rbd_dev_id_max = ATOMIC64_INIT(0);
3822 * Get a unique rbd identifier for the given new rbd_dev, and add
3823 * the rbd_dev to the global list. The minimum rbd id is 1.
3825 static void rbd_dev_id_get(struct rbd_device *rbd_dev)
3827 rbd_dev->dev_id = atomic64_inc_return(&rbd_dev_id_max);
3829 spin_lock(&rbd_dev_list_lock);
3830 list_add_tail(&rbd_dev->node, &rbd_dev_list);
3831 spin_unlock(&rbd_dev_list_lock);
3832 dout("rbd_dev %p given dev id %llu\n", rbd_dev,
3833 (unsigned long long) rbd_dev->dev_id);
3837 * Remove an rbd_dev from the global list, and record that its
3838 * identifier is no longer in use.
3840 static void rbd_dev_id_put(struct rbd_device *rbd_dev)
3842 struct list_head *tmp;
3843 int rbd_id = rbd_dev->dev_id;
3846 rbd_assert(rbd_id > 0);
3848 dout("rbd_dev %p released dev id %llu\n", rbd_dev,
3849 (unsigned long long) rbd_dev->dev_id);
3850 spin_lock(&rbd_dev_list_lock);
3851 list_del_init(&rbd_dev->node);
3854 * If the id being "put" is not the current maximum, there
3855 * is nothing special we need to do.
3857 if (rbd_id != atomic64_read(&rbd_dev_id_max)) {
3858 spin_unlock(&rbd_dev_list_lock);
3863 * We need to update the current maximum id. Search the
3864 * list to find out what it is. We're more likely to find
3865 * the maximum at the end, so search the list backward.
3868 list_for_each_prev(tmp, &rbd_dev_list) {
3869 struct rbd_device *rbd_dev;
3871 rbd_dev = list_entry(tmp, struct rbd_device, node);
3872 if (rbd_dev->dev_id > max_id)
3873 max_id = rbd_dev->dev_id;
3875 spin_unlock(&rbd_dev_list_lock);
3878 * The max id could have been updated by rbd_dev_id_get(), in
3879 * which case it now accurately reflects the new maximum.
3880 * Be careful not to overwrite the maximum value in that
3883 atomic64_cmpxchg(&rbd_dev_id_max, rbd_id, max_id);
3884 dout(" max dev id has been reset\n");
3888 * Skips over white space at *buf, and updates *buf to point to the
3889 * first found non-space character (if any). Returns the length of
3890 * the token (string of non-white space characters) found. Note
3891 * that *buf must be terminated with '\0'.
3893 static inline size_t next_token(const char **buf)
3896 * These are the characters that produce nonzero for
3897 * isspace() in the "C" and "POSIX" locales.
3899 const char *spaces = " \f\n\r\t\v";
3901 *buf += strspn(*buf, spaces); /* Find start of token */
3903 return strcspn(*buf, spaces); /* Return token length */
3907 * Finds the next token in *buf, and if the provided token buffer is
3908 * big enough, copies the found token into it. The result, if
3909 * copied, is guaranteed to be terminated with '\0'. Note that *buf
3910 * must be terminated with '\0' on entry.
3912 * Returns the length of the token found (not including the '\0').
3913 * Return value will be 0 if no token is found, and it will be >=
3914 * token_size if the token would not fit.
3916 * The *buf pointer will be updated to point beyond the end of the
3917 * found token. Note that this occurs even if the token buffer is
3918 * too small to hold it.
3920 static inline size_t copy_token(const char **buf,
3926 len = next_token(buf);
3927 if (len < token_size) {
3928 memcpy(token, *buf, len);
3929 *(token + len) = '\0';
3937 * Finds the next token in *buf, dynamically allocates a buffer big
3938 * enough to hold a copy of it, and copies the token into the new
3939 * buffer. The copy is guaranteed to be terminated with '\0'. Note
3940 * that a duplicate buffer is created even for a zero-length token.
3942 * Returns a pointer to the newly-allocated duplicate, or a null
3943 * pointer if memory for the duplicate was not available. If
3944 * the lenp argument is a non-null pointer, the length of the token
3945 * (not including the '\0') is returned in *lenp.
3947 * If successful, the *buf pointer will be updated to point beyond
3948 * the end of the found token.
3950 * Note: uses GFP_KERNEL for allocation.
3952 static inline char *dup_token(const char **buf, size_t *lenp)
3957 len = next_token(buf);
3958 dup = kmemdup(*buf, len + 1, GFP_KERNEL);
3961 *(dup + len) = '\0';
3971 * Parse the options provided for an "rbd add" (i.e., rbd image
3972 * mapping) request. These arrive via a write to /sys/bus/rbd/add,
3973 * and the data written is passed here via a NUL-terminated buffer.
3974 * Returns 0 if successful or an error code otherwise.
3976 * The information extracted from these options is recorded in
3977 * the other parameters which return dynamically-allocated
3980 * The address of a pointer that will refer to a ceph options
3981 * structure. Caller must release the returned pointer using
3982 * ceph_destroy_options() when it is no longer needed.
3984 * Address of an rbd options pointer. Fully initialized by
3985 * this function; caller must release with kfree().
3987 * Address of an rbd image specification pointer. Fully
3988 * initialized by this function based on parsed options.
3989 * Caller must release with rbd_spec_put().
3991 * The options passed take this form:
3992 * <mon_addrs> <options> <pool_name> <image_name> [<snap_id>]
3995 * A comma-separated list of one or more monitor addresses.
3996 * A monitor address is an ip address, optionally followed
3997 * by a port number (separated by a colon).
3998 * I.e.: ip1[:port1][,ip2[:port2]...]
4000 * A comma-separated list of ceph and/or rbd options.
4002 * The name of the rados pool containing the rbd image.
4004 * The name of the image in that pool to map.
4006 * An optional snapshot id. If provided, the mapping will
4007 * present data from the image at the time that snapshot was
4008 * created. The image head is used if no snapshot id is
4009 * provided. Snapshot mappings are always read-only.
4011 static int rbd_add_parse_args(const char *buf,
4012 struct ceph_options **ceph_opts,
4013 struct rbd_options **opts,
4014 struct rbd_spec **rbd_spec)
4018 const char *mon_addrs;
4019 size_t mon_addrs_size;
4020 struct rbd_spec *spec = NULL;
4021 struct rbd_options *rbd_opts = NULL;
4022 struct ceph_options *copts;
4025 /* The first four tokens are required */
4027 len = next_token(&buf);
4029 rbd_warn(NULL, "no monitor address(es) provided");
4033 mon_addrs_size = len + 1;
4037 options = dup_token(&buf, NULL);
4041 rbd_warn(NULL, "no options provided");
4045 spec = rbd_spec_alloc();
4049 spec->pool_name = dup_token(&buf, NULL);
4050 if (!spec->pool_name)
4052 if (!*spec->pool_name) {
4053 rbd_warn(NULL, "no pool name provided");
4057 spec->image_name = dup_token(&buf, NULL);
4058 if (!spec->image_name)
4060 if (!*spec->image_name) {
4061 rbd_warn(NULL, "no image name provided");
4066 * Snapshot name is optional; default is to use "-"
4067 * (indicating the head/no snapshot).
4069 len = next_token(&buf);
4071 buf = RBD_SNAP_HEAD_NAME; /* No snapshot supplied */
4072 len = sizeof (RBD_SNAP_HEAD_NAME) - 1;
4073 } else if (len > RBD_MAX_SNAP_NAME_LEN) {
4074 ret = -ENAMETOOLONG;
4077 spec->snap_name = kmemdup(buf, len + 1, GFP_KERNEL);
4078 if (!spec->snap_name)
4080 *(spec->snap_name + len) = '\0';
4082 /* Initialize all rbd options to the defaults */
4084 rbd_opts = kzalloc(sizeof (*rbd_opts), GFP_KERNEL);
4088 rbd_opts->read_only = RBD_READ_ONLY_DEFAULT;
4090 copts = ceph_parse_options(options, mon_addrs,
4091 mon_addrs + mon_addrs_size - 1,
4092 parse_rbd_opts_token, rbd_opts);
4093 if (IS_ERR(copts)) {
4094 ret = PTR_ERR(copts);
4115 * An rbd format 2 image has a unique identifier, distinct from the
4116 * name given to it by the user. Internally, that identifier is
4117 * what's used to specify the names of objects related to the image.
4119 * A special "rbd id" object is used to map an rbd image name to its
4120 * id. If that object doesn't exist, then there is no v2 rbd image
4121 * with the supplied name.
4123 * This function will record the given rbd_dev's image_id field if
4124 * it can be determined, and in that case will return 0. If any
4125 * errors occur a negative errno will be returned and the rbd_dev's
4126 * image_id field will be unchanged (and should be NULL).
4128 static int rbd_dev_image_id(struct rbd_device *rbd_dev)
4136 /* If we already have it we don't need to look it up */
4138 if (rbd_dev->spec->image_id)
4142 * When probing a parent image, the image id is already
4143 * known (and the image name likely is not). There's no
4144 * need to fetch the image id again in this case.
4146 if (rbd_dev->spec->image_id)
4150 * First, see if the format 2 image id file exists, and if
4151 * so, get the image's persistent id from it.
4153 size = sizeof (RBD_ID_PREFIX) + strlen(rbd_dev->spec->image_name);
4154 object_name = kmalloc(size, GFP_NOIO);
4157 sprintf(object_name, "%s%s", RBD_ID_PREFIX, rbd_dev->spec->image_name);
4158 dout("rbd id object name is %s\n", object_name);
4160 /* Response will be an encoded string, which includes a length */
4162 size = sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX;
4163 response = kzalloc(size, GFP_NOIO);
4169 ret = rbd_obj_method_sync(rbd_dev, object_name,
4172 response, RBD_IMAGE_ID_LEN_MAX, NULL);
4173 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
4178 rbd_dev->spec->image_id = ceph_extract_encoded_string(&p,
4179 p + RBD_IMAGE_ID_LEN_MAX,
4181 if (IS_ERR(rbd_dev->spec->image_id)) {
4182 ret = PTR_ERR(rbd_dev->spec->image_id);
4183 rbd_dev->spec->image_id = NULL;
4185 dout("image_id is %s\n", rbd_dev->spec->image_id);
4194 static int rbd_dev_v1_probe(struct rbd_device *rbd_dev)
4199 /* Version 1 images have no id; empty string is used */
4201 rbd_dev->spec->image_id = kstrdup("", GFP_KERNEL);
4202 if (!rbd_dev->spec->image_id)
4205 /* Record the header object name for this rbd image. */
4207 size = strlen(rbd_dev->spec->image_name) + sizeof (RBD_SUFFIX);
4208 rbd_dev->header_name = kmalloc(size, GFP_KERNEL);
4209 if (!rbd_dev->header_name) {
4213 sprintf(rbd_dev->header_name, "%s%s",
4214 rbd_dev->spec->image_name, RBD_SUFFIX);
4216 /* Populate rbd image metadata */
4218 ret = rbd_read_header(rbd_dev, &rbd_dev->header);
4222 /* Version 1 images have no parent (no layering) */
4224 rbd_dev->parent_spec = NULL;
4225 rbd_dev->parent_overlap = 0;
4227 rbd_dev->image_format = 1;
4229 dout("discovered version 1 image, header name is %s\n",
4230 rbd_dev->header_name);
4235 kfree(rbd_dev->header_name);
4236 rbd_dev->header_name = NULL;
4237 kfree(rbd_dev->spec->image_id);
4238 rbd_dev->spec->image_id = NULL;
4243 static int rbd_dev_v2_probe(struct rbd_device *rbd_dev)
4250 * Image id was filled in by the caller. Record the header
4251 * object name for this rbd image.
4253 size = sizeof (RBD_HEADER_PREFIX) + strlen(rbd_dev->spec->image_id);
4254 rbd_dev->header_name = kmalloc(size, GFP_KERNEL);
4255 if (!rbd_dev->header_name)
4257 sprintf(rbd_dev->header_name, "%s%s",
4258 RBD_HEADER_PREFIX, rbd_dev->spec->image_id);
4260 /* Get the size and object order for the image */
4262 ret = rbd_dev_v2_image_size(rbd_dev);
4266 /* Get the object prefix (a.k.a. block_name) for the image */
4268 ret = rbd_dev_v2_object_prefix(rbd_dev);
4272 /* Get the and check features for the image */
4274 ret = rbd_dev_v2_features(rbd_dev);
4278 /* If the image supports layering, get the parent info */
4280 if (rbd_dev->header.features & RBD_FEATURE_LAYERING) {
4281 ret = rbd_dev_v2_parent_info(rbd_dev);
4286 /* crypto and compression type aren't (yet) supported for v2 images */
4288 rbd_dev->header.crypt_type = 0;
4289 rbd_dev->header.comp_type = 0;
4291 /* Get the snapshot context, plus the header version */
4293 ret = rbd_dev_v2_snap_context(rbd_dev, &ver);
4296 rbd_dev->header.obj_version = ver;
4298 rbd_dev->image_format = 2;
4300 dout("discovered version 2 image, header name is %s\n",
4301 rbd_dev->header_name);
4305 rbd_dev->parent_overlap = 0;
4306 rbd_spec_put(rbd_dev->parent_spec);
4307 rbd_dev->parent_spec = NULL;
4308 kfree(rbd_dev->header_name);
4309 rbd_dev->header_name = NULL;
4310 kfree(rbd_dev->header.object_prefix);
4311 rbd_dev->header.object_prefix = NULL;
4316 static int rbd_dev_probe_finish(struct rbd_device *rbd_dev)
4318 struct rbd_device *parent = NULL;
4319 struct rbd_spec *parent_spec = NULL;
4320 struct rbd_client *rbdc = NULL;
4323 /* no need to lock here, as rbd_dev is not registered yet */
4324 ret = rbd_dev_snaps_update(rbd_dev);
4328 ret = rbd_dev_probe_update_spec(rbd_dev);
4332 ret = rbd_dev_set_mapping(rbd_dev);
4336 /* generate unique id: find highest unique id, add one */
4337 rbd_dev_id_get(rbd_dev);
4339 /* Fill in the device name, now that we have its id. */
4340 BUILD_BUG_ON(DEV_NAME_LEN
4341 < sizeof (RBD_DRV_NAME) + MAX_INT_FORMAT_WIDTH);
4342 sprintf(rbd_dev->name, "%s%d", RBD_DRV_NAME, rbd_dev->dev_id);
4344 /* Get our block major device number. */
4346 ret = register_blkdev(0, rbd_dev->name);
4349 rbd_dev->major = ret;
4351 /* Set up the blkdev mapping. */
4353 ret = rbd_init_disk(rbd_dev);
4355 goto err_out_blkdev;
4357 ret = rbd_bus_add_dev(rbd_dev);
4362 * At this point cleanup in the event of an error is the job
4363 * of the sysfs code (initiated by rbd_bus_del_dev()).
4365 /* Probe the parent if there is one */
4367 if (rbd_dev->parent_spec) {
4369 * We need to pass a reference to the client and the
4370 * parent spec when creating the parent rbd_dev.
4371 * Images related by parent/child relationships
4372 * always share both.
4374 parent_spec = rbd_spec_get(rbd_dev->parent_spec);
4375 rbdc = __rbd_get_client(rbd_dev->rbd_client);
4377 parent = rbd_dev_create(rbdc, parent_spec);
4382 rbdc = NULL; /* parent now owns reference */
4383 parent_spec = NULL; /* parent now owns reference */
4384 ret = rbd_dev_probe(parent);
4386 goto err_out_parent;
4387 rbd_dev->parent = parent;
4390 down_write(&rbd_dev->header_rwsem);
4391 ret = rbd_dev_snaps_register(rbd_dev);
4392 up_write(&rbd_dev->header_rwsem);
4396 ret = rbd_dev_header_watch_sync(rbd_dev, 1);
4400 /* Everything's ready. Announce the disk to the world. */
4402 add_disk(rbd_dev->disk);
4404 pr_info("%s: added with size 0x%llx\n", rbd_dev->disk->disk_name,
4405 (unsigned long long) rbd_dev->mapping.size);
4410 rbd_dev_destroy(parent);
4412 rbd_spec_put(parent_spec);
4413 rbd_put_client(rbdc);
4415 /* this will also clean up rest of rbd_dev stuff */
4417 rbd_bus_del_dev(rbd_dev);
4421 rbd_free_disk(rbd_dev);
4423 unregister_blkdev(rbd_dev->major, rbd_dev->name);
4425 rbd_dev_id_put(rbd_dev);
4427 rbd_remove_all_snaps(rbd_dev);
4433 * Probe for the existence of the header object for the given rbd
4434 * device. For format 2 images this includes determining the image
4437 static int rbd_dev_probe(struct rbd_device *rbd_dev)
4442 * Get the id from the image id object. If it's not a
4443 * format 2 image, we'll get ENOENT back, and we'll assume
4444 * it's a format 1 image.
4446 ret = rbd_dev_image_id(rbd_dev);
4448 ret = rbd_dev_v1_probe(rbd_dev);
4450 ret = rbd_dev_v2_probe(rbd_dev);
4452 dout("probe failed, returning %d\n", ret);
4457 ret = rbd_dev_probe_finish(rbd_dev);
4459 rbd_header_free(&rbd_dev->header);
4464 static ssize_t rbd_add(struct bus_type *bus,
4468 struct rbd_device *rbd_dev = NULL;
4469 struct ceph_options *ceph_opts = NULL;
4470 struct rbd_options *rbd_opts = NULL;
4471 struct rbd_spec *spec = NULL;
4472 struct rbd_client *rbdc;
4473 struct ceph_osd_client *osdc;
4476 if (!try_module_get(THIS_MODULE))
4479 /* parse add command */
4480 rc = rbd_add_parse_args(buf, &ceph_opts, &rbd_opts, &spec);
4482 goto err_out_module;
4484 rbdc = rbd_get_client(ceph_opts);
4489 ceph_opts = NULL; /* rbd_dev client now owns this */
4492 osdc = &rbdc->client->osdc;
4493 rc = ceph_pg_poolid_by_name(osdc->osdmap, spec->pool_name);
4495 goto err_out_client;
4496 spec->pool_id = (u64) rc;
4498 /* The ceph file layout needs to fit pool id in 32 bits */
4500 if (WARN_ON(spec->pool_id > (u64) U32_MAX)) {
4502 goto err_out_client;
4505 rbd_dev = rbd_dev_create(rbdc, spec);
4507 goto err_out_client;
4508 rbdc = NULL; /* rbd_dev now owns this */
4509 spec = NULL; /* rbd_dev now owns this */
4511 rbd_dev->mapping.read_only = rbd_opts->read_only;
4513 rbd_opts = NULL; /* done with this */
4515 rc = rbd_dev_probe(rbd_dev);
4517 goto err_out_rbd_dev;
4521 rbd_dev_destroy(rbd_dev);
4523 rbd_put_client(rbdc);
4526 ceph_destroy_options(ceph_opts);
4530 module_put(THIS_MODULE);
4532 dout("Error adding device %s\n", buf);
4534 return (ssize_t) rc;
4537 static struct rbd_device *__rbd_get_dev(unsigned long dev_id)
4539 struct list_head *tmp;
4540 struct rbd_device *rbd_dev;
4542 spin_lock(&rbd_dev_list_lock);
4543 list_for_each(tmp, &rbd_dev_list) {
4544 rbd_dev = list_entry(tmp, struct rbd_device, node);
4545 if (rbd_dev->dev_id == dev_id) {
4546 spin_unlock(&rbd_dev_list_lock);
4550 spin_unlock(&rbd_dev_list_lock);
4554 static void rbd_dev_release(struct device *dev)
4556 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
4558 if (rbd_dev->watch_event)
4559 rbd_dev_header_watch_sync(rbd_dev, 0);
4561 /* clean up and free blkdev */
4562 rbd_free_disk(rbd_dev);
4563 unregister_blkdev(rbd_dev->major, rbd_dev->name);
4565 /* release allocated disk header fields */
4566 rbd_header_free(&rbd_dev->header);
4568 /* done with the id, and with the rbd_dev */
4569 rbd_dev_id_put(rbd_dev);
4570 rbd_assert(rbd_dev->rbd_client != NULL);
4571 rbd_dev_destroy(rbd_dev);
4573 /* release module ref */
4574 module_put(THIS_MODULE);
4577 static void __rbd_remove(struct rbd_device *rbd_dev)
4579 rbd_remove_all_snaps(rbd_dev);
4580 rbd_bus_del_dev(rbd_dev);
4583 static ssize_t rbd_remove(struct bus_type *bus,
4587 struct rbd_device *rbd_dev = NULL;
4592 rc = strict_strtoul(buf, 10, &ul);
4596 /* convert to int; abort if we lost anything in the conversion */
4597 target_id = (int) ul;
4598 if (target_id != ul)
4601 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
4603 rbd_dev = __rbd_get_dev(target_id);
4609 spin_lock_irq(&rbd_dev->lock);
4610 if (rbd_dev->open_count)
4613 set_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags);
4614 spin_unlock_irq(&rbd_dev->lock);
4618 while (rbd_dev->parent_spec) {
4619 struct rbd_device *first = rbd_dev;
4620 struct rbd_device *second = first->parent;
4621 struct rbd_device *third;
4624 * Follow to the parent with no grandparent and
4627 while (second && (third = second->parent)) {
4631 __rbd_remove(second);
4632 rbd_spec_put(first->parent_spec);
4633 first->parent_spec = NULL;
4634 first->parent_overlap = 0;
4635 first->parent = NULL;
4637 __rbd_remove(rbd_dev);
4640 mutex_unlock(&ctl_mutex);
4646 * create control files in sysfs
4649 static int rbd_sysfs_init(void)
4653 ret = device_register(&rbd_root_dev);
4657 ret = bus_register(&rbd_bus_type);
4659 device_unregister(&rbd_root_dev);
4664 static void rbd_sysfs_cleanup(void)
4666 bus_unregister(&rbd_bus_type);
4667 device_unregister(&rbd_root_dev);
4670 static int __init rbd_init(void)
4674 if (!libceph_compatible(NULL)) {
4675 rbd_warn(NULL, "libceph incompatibility (quitting)");
4679 rc = rbd_sysfs_init();
4682 pr_info("loaded " RBD_DRV_NAME_LONG "\n");
4686 static void __exit rbd_exit(void)
4688 rbd_sysfs_cleanup();
4691 module_init(rbd_init);
4692 module_exit(rbd_exit);
4694 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
4695 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
4696 MODULE_DESCRIPTION("rados block device");
4698 /* following authorship retained from original osdblk.c */
4699 MODULE_AUTHOR("Jeff Garzik <jeff@garzik.org>");
4701 MODULE_LICENSE("GPL");