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 (RBD_FEATURES_ALL)
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;
221 struct page **copyup_pages;
223 struct ceph_osd_request *osd_req;
225 u64 xferred; /* bytes transferred */
229 rbd_obj_callback_t callback;
230 struct completion completion;
236 IMG_REQ_WRITE, /* I/O direction: read = 0, write = 1 */
237 IMG_REQ_CHILD, /* initiator: block = 0, child image = 1 */
238 IMG_REQ_LAYERED, /* ENOENT handling: normal = 0, layered = 1 */
241 struct rbd_img_request {
242 struct rbd_device *rbd_dev;
243 u64 offset; /* starting image byte offset */
244 u64 length; /* byte count from offset */
247 u64 snap_id; /* for reads */
248 struct ceph_snap_context *snapc; /* for writes */
251 struct request *rq; /* block request */
252 struct rbd_obj_request *obj_request; /* obj req initiator */
254 struct page **copyup_pages;
255 spinlock_t completion_lock;/* protects next_completion */
257 rbd_img_callback_t callback;
258 u64 xferred;/* aggregate bytes transferred */
259 int result; /* first nonzero obj_request result */
261 u32 obj_request_count;
262 struct list_head obj_requests; /* rbd_obj_request structs */
267 #define for_each_obj_request(ireq, oreq) \
268 list_for_each_entry(oreq, &(ireq)->obj_requests, links)
269 #define for_each_obj_request_from(ireq, oreq) \
270 list_for_each_entry_from(oreq, &(ireq)->obj_requests, links)
271 #define for_each_obj_request_safe(ireq, oreq, n) \
272 list_for_each_entry_safe_reverse(oreq, n, &(ireq)->obj_requests, links)
277 struct list_head node;
292 int dev_id; /* blkdev unique id */
294 int major; /* blkdev assigned major */
295 struct gendisk *disk; /* blkdev's gendisk and rq */
297 u32 image_format; /* Either 1 or 2 */
298 struct rbd_client *rbd_client;
300 char name[DEV_NAME_LEN]; /* blkdev name, e.g. rbd3 */
302 spinlock_t lock; /* queue, flags, open_count */
304 struct rbd_image_header header;
305 unsigned long flags; /* possibly lock protected */
306 struct rbd_spec *spec;
310 struct ceph_file_layout layout;
312 struct ceph_osd_event *watch_event;
313 struct rbd_obj_request *watch_request;
315 struct rbd_spec *parent_spec;
317 struct rbd_device *parent;
322 /* protects updating the header */
323 struct rw_semaphore header_rwsem;
325 struct rbd_mapping mapping;
327 struct list_head node;
329 /* list of snapshots */
330 struct list_head snaps;
334 unsigned long open_count; /* protected by lock */
338 * Flag bits for rbd_dev->flags. If atomicity is required,
339 * rbd_dev->lock is used to protect access.
341 * Currently, only the "removing" flag (which is coupled with the
342 * "open_count" field) requires atomic access.
345 RBD_DEV_FLAG_EXISTS, /* mapped snapshot has not been deleted */
346 RBD_DEV_FLAG_REMOVING, /* this mapping is being removed */
349 static DEFINE_MUTEX(ctl_mutex); /* Serialize open/close/setup/teardown */
351 static LIST_HEAD(rbd_dev_list); /* devices */
352 static DEFINE_SPINLOCK(rbd_dev_list_lock);
354 static LIST_HEAD(rbd_client_list); /* clients */
355 static DEFINE_SPINLOCK(rbd_client_list_lock);
357 static int rbd_img_request_submit(struct rbd_img_request *img_request);
359 static int rbd_dev_snaps_update(struct rbd_device *rbd_dev);
361 static void rbd_dev_release(struct device *dev);
362 static void rbd_snap_destroy(struct rbd_snap *snap);
364 static ssize_t rbd_add(struct bus_type *bus, const char *buf,
366 static ssize_t rbd_remove(struct bus_type *bus, const char *buf,
368 static int rbd_dev_probe(struct rbd_device *rbd_dev);
370 static struct bus_attribute rbd_bus_attrs[] = {
371 __ATTR(add, S_IWUSR, NULL, rbd_add),
372 __ATTR(remove, S_IWUSR, NULL, rbd_remove),
376 static struct bus_type rbd_bus_type = {
378 .bus_attrs = rbd_bus_attrs,
381 static void rbd_root_dev_release(struct device *dev)
385 static struct device rbd_root_dev = {
387 .release = rbd_root_dev_release,
390 static __printf(2, 3)
391 void rbd_warn(struct rbd_device *rbd_dev, const char *fmt, ...)
393 struct va_format vaf;
401 printk(KERN_WARNING "%s: %pV\n", RBD_DRV_NAME, &vaf);
402 else if (rbd_dev->disk)
403 printk(KERN_WARNING "%s: %s: %pV\n",
404 RBD_DRV_NAME, rbd_dev->disk->disk_name, &vaf);
405 else if (rbd_dev->spec && rbd_dev->spec->image_name)
406 printk(KERN_WARNING "%s: image %s: %pV\n",
407 RBD_DRV_NAME, rbd_dev->spec->image_name, &vaf);
408 else if (rbd_dev->spec && rbd_dev->spec->image_id)
409 printk(KERN_WARNING "%s: id %s: %pV\n",
410 RBD_DRV_NAME, rbd_dev->spec->image_id, &vaf);
412 printk(KERN_WARNING "%s: rbd_dev %p: %pV\n",
413 RBD_DRV_NAME, rbd_dev, &vaf);
418 #define rbd_assert(expr) \
419 if (unlikely(!(expr))) { \
420 printk(KERN_ERR "\nAssertion failure in %s() " \
422 "\trbd_assert(%s);\n\n", \
423 __func__, __LINE__, #expr); \
426 #else /* !RBD_DEBUG */
427 # define rbd_assert(expr) ((void) 0)
428 #endif /* !RBD_DEBUG */
430 static void rbd_img_parent_read(struct rbd_obj_request *obj_request);
431 static int rbd_img_obj_request_submit(struct rbd_obj_request *obj_request);
433 static int rbd_dev_refresh(struct rbd_device *rbd_dev, u64 *hver);
434 static int rbd_dev_v2_refresh(struct rbd_device *rbd_dev, u64 *hver);
436 static int rbd_open(struct block_device *bdev, fmode_t mode)
438 struct rbd_device *rbd_dev = bdev->bd_disk->private_data;
439 bool removing = false;
441 if ((mode & FMODE_WRITE) && rbd_dev->mapping.read_only)
444 spin_lock_irq(&rbd_dev->lock);
445 if (test_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags))
448 rbd_dev->open_count++;
449 spin_unlock_irq(&rbd_dev->lock);
453 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
454 (void) get_device(&rbd_dev->dev);
455 set_device_ro(bdev, rbd_dev->mapping.read_only);
456 mutex_unlock(&ctl_mutex);
461 static int rbd_release(struct gendisk *disk, fmode_t mode)
463 struct rbd_device *rbd_dev = disk->private_data;
464 unsigned long open_count_before;
466 spin_lock_irq(&rbd_dev->lock);
467 open_count_before = rbd_dev->open_count--;
468 spin_unlock_irq(&rbd_dev->lock);
469 rbd_assert(open_count_before > 0);
471 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
472 put_device(&rbd_dev->dev);
473 mutex_unlock(&ctl_mutex);
478 static const struct block_device_operations rbd_bd_ops = {
479 .owner = THIS_MODULE,
481 .release = rbd_release,
485 * Initialize an rbd client instance.
488 static struct rbd_client *rbd_client_create(struct ceph_options *ceph_opts)
490 struct rbd_client *rbdc;
493 dout("%s:\n", __func__);
494 rbdc = kmalloc(sizeof(struct rbd_client), GFP_KERNEL);
498 kref_init(&rbdc->kref);
499 INIT_LIST_HEAD(&rbdc->node);
501 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
503 rbdc->client = ceph_create_client(ceph_opts, rbdc, 0, 0);
504 if (IS_ERR(rbdc->client))
506 ceph_opts = NULL; /* Now rbdc->client is responsible for ceph_opts */
508 ret = ceph_open_session(rbdc->client);
512 spin_lock(&rbd_client_list_lock);
513 list_add_tail(&rbdc->node, &rbd_client_list);
514 spin_unlock(&rbd_client_list_lock);
516 mutex_unlock(&ctl_mutex);
517 dout("%s: rbdc %p\n", __func__, rbdc);
522 ceph_destroy_client(rbdc->client);
524 mutex_unlock(&ctl_mutex);
528 ceph_destroy_options(ceph_opts);
529 dout("%s: error %d\n", __func__, ret);
534 static struct rbd_client *__rbd_get_client(struct rbd_client *rbdc)
536 kref_get(&rbdc->kref);
542 * Find a ceph client with specific addr and configuration. If
543 * found, bump its reference count.
545 static struct rbd_client *rbd_client_find(struct ceph_options *ceph_opts)
547 struct rbd_client *client_node;
550 if (ceph_opts->flags & CEPH_OPT_NOSHARE)
553 spin_lock(&rbd_client_list_lock);
554 list_for_each_entry(client_node, &rbd_client_list, node) {
555 if (!ceph_compare_options(ceph_opts, client_node->client)) {
556 __rbd_get_client(client_node);
562 spin_unlock(&rbd_client_list_lock);
564 return found ? client_node : NULL;
574 /* string args above */
577 /* Boolean args above */
581 static match_table_t rbd_opts_tokens = {
583 /* string args above */
584 {Opt_read_only, "read_only"},
585 {Opt_read_only, "ro"}, /* Alternate spelling */
586 {Opt_read_write, "read_write"},
587 {Opt_read_write, "rw"}, /* Alternate spelling */
588 /* Boolean args above */
596 #define RBD_READ_ONLY_DEFAULT false
598 static int parse_rbd_opts_token(char *c, void *private)
600 struct rbd_options *rbd_opts = private;
601 substring_t argstr[MAX_OPT_ARGS];
602 int token, intval, ret;
604 token = match_token(c, rbd_opts_tokens, argstr);
608 if (token < Opt_last_int) {
609 ret = match_int(&argstr[0], &intval);
611 pr_err("bad mount option arg (not int) "
615 dout("got int token %d val %d\n", token, intval);
616 } else if (token > Opt_last_int && token < Opt_last_string) {
617 dout("got string token %d val %s\n", token,
619 } else if (token > Opt_last_string && token < Opt_last_bool) {
620 dout("got Boolean token %d\n", token);
622 dout("got token %d\n", token);
627 rbd_opts->read_only = true;
630 rbd_opts->read_only = false;
640 * Get a ceph client with specific addr and configuration, if one does
641 * not exist create it.
643 static struct rbd_client *rbd_get_client(struct ceph_options *ceph_opts)
645 struct rbd_client *rbdc;
647 rbdc = rbd_client_find(ceph_opts);
648 if (rbdc) /* using an existing client */
649 ceph_destroy_options(ceph_opts);
651 rbdc = rbd_client_create(ceph_opts);
657 * Destroy ceph client
659 * Caller must hold rbd_client_list_lock.
661 static void rbd_client_release(struct kref *kref)
663 struct rbd_client *rbdc = container_of(kref, struct rbd_client, kref);
665 dout("%s: rbdc %p\n", __func__, rbdc);
666 spin_lock(&rbd_client_list_lock);
667 list_del(&rbdc->node);
668 spin_unlock(&rbd_client_list_lock);
670 ceph_destroy_client(rbdc->client);
675 * Drop reference to ceph client node. If it's not referenced anymore, release
678 static void rbd_put_client(struct rbd_client *rbdc)
681 kref_put(&rbdc->kref, rbd_client_release);
684 static bool rbd_image_format_valid(u32 image_format)
686 return image_format == 1 || image_format == 2;
689 static bool rbd_dev_ondisk_valid(struct rbd_image_header_ondisk *ondisk)
694 /* The header has to start with the magic rbd header text */
695 if (memcmp(&ondisk->text, RBD_HEADER_TEXT, sizeof (RBD_HEADER_TEXT)))
698 /* The bio layer requires at least sector-sized I/O */
700 if (ondisk->options.order < SECTOR_SHIFT)
703 /* If we use u64 in a few spots we may be able to loosen this */
705 if (ondisk->options.order > 8 * sizeof (int) - 1)
709 * The size of a snapshot header has to fit in a size_t, and
710 * that limits the number of snapshots.
712 snap_count = le32_to_cpu(ondisk->snap_count);
713 size = SIZE_MAX - sizeof (struct ceph_snap_context);
714 if (snap_count > size / sizeof (__le64))
718 * Not only that, but the size of the entire the snapshot
719 * header must also be representable in a size_t.
721 size -= snap_count * sizeof (__le64);
722 if ((u64) size < le64_to_cpu(ondisk->snap_names_len))
729 * Create a new header structure, translate header format from the on-disk
732 static int rbd_header_from_disk(struct rbd_image_header *header,
733 struct rbd_image_header_ondisk *ondisk)
740 memset(header, 0, sizeof (*header));
742 snap_count = le32_to_cpu(ondisk->snap_count);
744 len = strnlen(ondisk->object_prefix, sizeof (ondisk->object_prefix));
745 header->object_prefix = kmalloc(len + 1, GFP_KERNEL);
746 if (!header->object_prefix)
748 memcpy(header->object_prefix, ondisk->object_prefix, len);
749 header->object_prefix[len] = '\0';
752 u64 snap_names_len = le64_to_cpu(ondisk->snap_names_len);
754 /* Save a copy of the snapshot names */
756 if (snap_names_len > (u64) SIZE_MAX)
758 header->snap_names = kmalloc(snap_names_len, GFP_KERNEL);
759 if (!header->snap_names)
762 * Note that rbd_dev_v1_header_read() guarantees
763 * the ondisk buffer we're working with has
764 * snap_names_len bytes beyond the end of the
765 * snapshot id array, this memcpy() is safe.
767 memcpy(header->snap_names, &ondisk->snaps[snap_count],
770 /* Record each snapshot's size */
772 size = snap_count * sizeof (*header->snap_sizes);
773 header->snap_sizes = kmalloc(size, GFP_KERNEL);
774 if (!header->snap_sizes)
776 for (i = 0; i < snap_count; i++)
777 header->snap_sizes[i] =
778 le64_to_cpu(ondisk->snaps[i].image_size);
780 WARN_ON(ondisk->snap_names_len);
781 header->snap_names = NULL;
782 header->snap_sizes = NULL;
785 header->features = 0; /* No features support in v1 images */
786 header->obj_order = ondisk->options.order;
787 header->crypt_type = ondisk->options.crypt_type;
788 header->comp_type = ondisk->options.comp_type;
790 /* Allocate and fill in the snapshot context */
792 header->image_size = le64_to_cpu(ondisk->image_size);
793 size = sizeof (struct ceph_snap_context);
794 size += snap_count * sizeof (header->snapc->snaps[0]);
795 header->snapc = kzalloc(size, GFP_KERNEL);
799 atomic_set(&header->snapc->nref, 1);
800 header->snapc->seq = le64_to_cpu(ondisk->snap_seq);
801 header->snapc->num_snaps = snap_count;
802 for (i = 0; i < snap_count; i++)
803 header->snapc->snaps[i] =
804 le64_to_cpu(ondisk->snaps[i].id);
809 kfree(header->snap_sizes);
810 header->snap_sizes = NULL;
811 kfree(header->snap_names);
812 header->snap_names = NULL;
813 kfree(header->object_prefix);
814 header->object_prefix = NULL;
819 static const char *rbd_snap_name(struct rbd_device *rbd_dev, u64 snap_id)
821 struct rbd_snap *snap;
823 if (snap_id == CEPH_NOSNAP)
824 return RBD_SNAP_HEAD_NAME;
826 list_for_each_entry(snap, &rbd_dev->snaps, node)
827 if (snap_id == snap->id)
833 static int snap_by_name(struct rbd_device *rbd_dev, const char *snap_name)
836 struct rbd_snap *snap;
838 list_for_each_entry(snap, &rbd_dev->snaps, node) {
839 if (!strcmp(snap_name, snap->name)) {
840 rbd_dev->spec->snap_id = snap->id;
841 rbd_dev->mapping.size = snap->size;
842 rbd_dev->mapping.features = snap->features;
851 static int rbd_dev_set_mapping(struct rbd_device *rbd_dev)
855 if (!memcmp(rbd_dev->spec->snap_name, RBD_SNAP_HEAD_NAME,
856 sizeof (RBD_SNAP_HEAD_NAME))) {
857 rbd_dev->spec->snap_id = CEPH_NOSNAP;
858 rbd_dev->mapping.size = rbd_dev->header.image_size;
859 rbd_dev->mapping.features = rbd_dev->header.features;
862 ret = snap_by_name(rbd_dev, rbd_dev->spec->snap_name);
865 rbd_dev->mapping.read_only = true;
867 set_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags);
873 static void rbd_header_free(struct rbd_image_header *header)
875 kfree(header->object_prefix);
876 header->object_prefix = NULL;
877 kfree(header->snap_sizes);
878 header->snap_sizes = NULL;
879 kfree(header->snap_names);
880 header->snap_names = NULL;
881 ceph_put_snap_context(header->snapc);
882 header->snapc = NULL;
885 static const char *rbd_segment_name(struct rbd_device *rbd_dev, u64 offset)
891 name = kmalloc(MAX_OBJ_NAME_SIZE + 1, GFP_NOIO);
894 segment = offset >> rbd_dev->header.obj_order;
895 ret = snprintf(name, MAX_OBJ_NAME_SIZE + 1, "%s.%012llx",
896 rbd_dev->header.object_prefix, segment);
897 if (ret < 0 || ret > MAX_OBJ_NAME_SIZE) {
898 pr_err("error formatting segment name for #%llu (%d)\n",
907 static u64 rbd_segment_offset(struct rbd_device *rbd_dev, u64 offset)
909 u64 segment_size = (u64) 1 << rbd_dev->header.obj_order;
911 return offset & (segment_size - 1);
914 static u64 rbd_segment_length(struct rbd_device *rbd_dev,
915 u64 offset, u64 length)
917 u64 segment_size = (u64) 1 << rbd_dev->header.obj_order;
919 offset &= segment_size - 1;
921 rbd_assert(length <= U64_MAX - offset);
922 if (offset + length > segment_size)
923 length = segment_size - offset;
929 * returns the size of an object in the image
931 static u64 rbd_obj_bytes(struct rbd_image_header *header)
933 return 1 << header->obj_order;
940 static void bio_chain_put(struct bio *chain)
946 chain = chain->bi_next;
952 * zeros a bio chain, starting at specific offset
954 static void zero_bio_chain(struct bio *chain, int start_ofs)
963 bio_for_each_segment(bv, chain, i) {
964 if (pos + bv->bv_len > start_ofs) {
965 int remainder = max(start_ofs - pos, 0);
966 buf = bvec_kmap_irq(bv, &flags);
967 memset(buf + remainder, 0,
968 bv->bv_len - remainder);
969 bvec_kunmap_irq(buf, &flags);
974 chain = chain->bi_next;
979 * similar to zero_bio_chain(), zeros data defined by a page array,
980 * starting at the given byte offset from the start of the array and
981 * continuing up to the given end offset. The pages array is
982 * assumed to be big enough to hold all bytes up to the end.
984 static void zero_pages(struct page **pages, u64 offset, u64 end)
986 struct page **page = &pages[offset >> PAGE_SHIFT];
988 rbd_assert(end > offset);
989 rbd_assert(end - offset <= (u64)SIZE_MAX);
990 while (offset < end) {
996 page_offset = (size_t)(offset & ~PAGE_MASK);
997 length = min(PAGE_SIZE - page_offset, (size_t)(end - offset));
998 local_irq_save(flags);
999 kaddr = kmap_atomic(*page);
1000 memset(kaddr + page_offset, 0, length);
1001 kunmap_atomic(kaddr);
1002 local_irq_restore(flags);
1010 * Clone a portion of a bio, starting at the given byte offset
1011 * and continuing for the number of bytes indicated.
1013 static struct bio *bio_clone_range(struct bio *bio_src,
1014 unsigned int offset,
1022 unsigned short end_idx;
1023 unsigned short vcnt;
1026 /* Handle the easy case for the caller */
1028 if (!offset && len == bio_src->bi_size)
1029 return bio_clone(bio_src, gfpmask);
1031 if (WARN_ON_ONCE(!len))
1033 if (WARN_ON_ONCE(len > bio_src->bi_size))
1035 if (WARN_ON_ONCE(offset > bio_src->bi_size - len))
1038 /* Find first affected segment... */
1041 __bio_for_each_segment(bv, bio_src, idx, 0) {
1042 if (resid < bv->bv_len)
1044 resid -= bv->bv_len;
1048 /* ...and the last affected segment */
1051 __bio_for_each_segment(bv, bio_src, end_idx, idx) {
1052 if (resid <= bv->bv_len)
1054 resid -= bv->bv_len;
1056 vcnt = end_idx - idx + 1;
1058 /* Build the clone */
1060 bio = bio_alloc(gfpmask, (unsigned int) vcnt);
1062 return NULL; /* ENOMEM */
1064 bio->bi_bdev = bio_src->bi_bdev;
1065 bio->bi_sector = bio_src->bi_sector + (offset >> SECTOR_SHIFT);
1066 bio->bi_rw = bio_src->bi_rw;
1067 bio->bi_flags |= 1 << BIO_CLONED;
1070 * Copy over our part of the bio_vec, then update the first
1071 * and last (or only) entries.
1073 memcpy(&bio->bi_io_vec[0], &bio_src->bi_io_vec[idx],
1074 vcnt * sizeof (struct bio_vec));
1075 bio->bi_io_vec[0].bv_offset += voff;
1077 bio->bi_io_vec[0].bv_len -= voff;
1078 bio->bi_io_vec[vcnt - 1].bv_len = resid;
1080 bio->bi_io_vec[0].bv_len = len;
1083 bio->bi_vcnt = vcnt;
1091 * Clone a portion of a bio chain, starting at the given byte offset
1092 * into the first bio in the source chain and continuing for the
1093 * number of bytes indicated. The result is another bio chain of
1094 * exactly the given length, or a null pointer on error.
1096 * The bio_src and offset parameters are both in-out. On entry they
1097 * refer to the first source bio and the offset into that bio where
1098 * the start of data to be cloned is located.
1100 * On return, bio_src is updated to refer to the bio in the source
1101 * chain that contains first un-cloned byte, and *offset will
1102 * contain the offset of that byte within that bio.
1104 static struct bio *bio_chain_clone_range(struct bio **bio_src,
1105 unsigned int *offset,
1109 struct bio *bi = *bio_src;
1110 unsigned int off = *offset;
1111 struct bio *chain = NULL;
1114 /* Build up a chain of clone bios up to the limit */
1116 if (!bi || off >= bi->bi_size || !len)
1117 return NULL; /* Nothing to clone */
1121 unsigned int bi_size;
1125 rbd_warn(NULL, "bio_chain exhausted with %u left", len);
1126 goto out_err; /* EINVAL; ran out of bio's */
1128 bi_size = min_t(unsigned int, bi->bi_size - off, len);
1129 bio = bio_clone_range(bi, off, bi_size, gfpmask);
1131 goto out_err; /* ENOMEM */
1134 end = &bio->bi_next;
1137 if (off == bi->bi_size) {
1148 bio_chain_put(chain);
1154 * The default/initial value for all object request flags is 0. For
1155 * each flag, once its value is set to 1 it is never reset to 0
1158 static void obj_request_img_data_set(struct rbd_obj_request *obj_request)
1160 if (test_and_set_bit(OBJ_REQ_IMG_DATA, &obj_request->flags)) {
1161 struct rbd_device *rbd_dev;
1163 rbd_dev = obj_request->img_request->rbd_dev;
1164 rbd_warn(rbd_dev, "obj_request %p already marked img_data\n",
1169 static bool obj_request_img_data_test(struct rbd_obj_request *obj_request)
1172 return test_bit(OBJ_REQ_IMG_DATA, &obj_request->flags) != 0;
1175 static void obj_request_done_set(struct rbd_obj_request *obj_request)
1177 if (test_and_set_bit(OBJ_REQ_DONE, &obj_request->flags)) {
1178 struct rbd_device *rbd_dev = NULL;
1180 if (obj_request_img_data_test(obj_request))
1181 rbd_dev = obj_request->img_request->rbd_dev;
1182 rbd_warn(rbd_dev, "obj_request %p already marked done\n",
1187 static bool obj_request_done_test(struct rbd_obj_request *obj_request)
1190 return test_bit(OBJ_REQ_DONE, &obj_request->flags) != 0;
1194 * This sets the KNOWN flag after (possibly) setting the EXISTS
1195 * flag. The latter is set based on the "exists" value provided.
1197 * Note that for our purposes once an object exists it never goes
1198 * away again. It's possible that the response from two existence
1199 * checks are separated by the creation of the target object, and
1200 * the first ("doesn't exist") response arrives *after* the second
1201 * ("does exist"). In that case we ignore the second one.
1203 static void obj_request_existence_set(struct rbd_obj_request *obj_request,
1207 set_bit(OBJ_REQ_EXISTS, &obj_request->flags);
1208 set_bit(OBJ_REQ_KNOWN, &obj_request->flags);
1212 static bool obj_request_known_test(struct rbd_obj_request *obj_request)
1215 return test_bit(OBJ_REQ_KNOWN, &obj_request->flags) != 0;
1218 static bool obj_request_exists_test(struct rbd_obj_request *obj_request)
1221 return test_bit(OBJ_REQ_EXISTS, &obj_request->flags) != 0;
1224 static void rbd_obj_request_get(struct rbd_obj_request *obj_request)
1226 dout("%s: obj %p (was %d)\n", __func__, obj_request,
1227 atomic_read(&obj_request->kref.refcount));
1228 kref_get(&obj_request->kref);
1231 static void rbd_obj_request_destroy(struct kref *kref);
1232 static void rbd_obj_request_put(struct rbd_obj_request *obj_request)
1234 rbd_assert(obj_request != NULL);
1235 dout("%s: obj %p (was %d)\n", __func__, obj_request,
1236 atomic_read(&obj_request->kref.refcount));
1237 kref_put(&obj_request->kref, rbd_obj_request_destroy);
1240 static void rbd_img_request_get(struct rbd_img_request *img_request)
1242 dout("%s: img %p (was %d)\n", __func__, img_request,
1243 atomic_read(&img_request->kref.refcount));
1244 kref_get(&img_request->kref);
1247 static void rbd_img_request_destroy(struct kref *kref);
1248 static void rbd_img_request_put(struct rbd_img_request *img_request)
1250 rbd_assert(img_request != NULL);
1251 dout("%s: img %p (was %d)\n", __func__, img_request,
1252 atomic_read(&img_request->kref.refcount));
1253 kref_put(&img_request->kref, rbd_img_request_destroy);
1256 static inline void rbd_img_obj_request_add(struct rbd_img_request *img_request,
1257 struct rbd_obj_request *obj_request)
1259 rbd_assert(obj_request->img_request == NULL);
1261 /* Image request now owns object's original reference */
1262 obj_request->img_request = img_request;
1263 obj_request->which = img_request->obj_request_count;
1264 rbd_assert(!obj_request_img_data_test(obj_request));
1265 obj_request_img_data_set(obj_request);
1266 rbd_assert(obj_request->which != BAD_WHICH);
1267 img_request->obj_request_count++;
1268 list_add_tail(&obj_request->links, &img_request->obj_requests);
1269 dout("%s: img %p obj %p w=%u\n", __func__, img_request, obj_request,
1270 obj_request->which);
1273 static inline void rbd_img_obj_request_del(struct rbd_img_request *img_request,
1274 struct rbd_obj_request *obj_request)
1276 rbd_assert(obj_request->which != BAD_WHICH);
1278 dout("%s: img %p obj %p w=%u\n", __func__, img_request, obj_request,
1279 obj_request->which);
1280 list_del(&obj_request->links);
1281 rbd_assert(img_request->obj_request_count > 0);
1282 img_request->obj_request_count--;
1283 rbd_assert(obj_request->which == img_request->obj_request_count);
1284 obj_request->which = BAD_WHICH;
1285 rbd_assert(obj_request_img_data_test(obj_request));
1286 rbd_assert(obj_request->img_request == img_request);
1287 obj_request->img_request = NULL;
1288 obj_request->callback = NULL;
1289 rbd_obj_request_put(obj_request);
1292 static bool obj_request_type_valid(enum obj_request_type type)
1295 case OBJ_REQUEST_NODATA:
1296 case OBJ_REQUEST_BIO:
1297 case OBJ_REQUEST_PAGES:
1304 static int rbd_obj_request_submit(struct ceph_osd_client *osdc,
1305 struct rbd_obj_request *obj_request)
1307 dout("%s: osdc %p obj %p\n", __func__, osdc, obj_request);
1309 return ceph_osdc_start_request(osdc, obj_request->osd_req, false);
1312 static void rbd_img_request_complete(struct rbd_img_request *img_request)
1315 dout("%s: img %p\n", __func__, img_request);
1318 * If no error occurred, compute the aggregate transfer
1319 * count for the image request. We could instead use
1320 * atomic64_cmpxchg() to update it as each object request
1321 * completes; not clear which way is better off hand.
1323 if (!img_request->result) {
1324 struct rbd_obj_request *obj_request;
1327 for_each_obj_request(img_request, obj_request)
1328 xferred += obj_request->xferred;
1329 img_request->xferred = xferred;
1332 if (img_request->callback)
1333 img_request->callback(img_request);
1335 rbd_img_request_put(img_request);
1338 /* Caller is responsible for rbd_obj_request_destroy(obj_request) */
1340 static int rbd_obj_request_wait(struct rbd_obj_request *obj_request)
1342 dout("%s: obj %p\n", __func__, obj_request);
1344 return wait_for_completion_interruptible(&obj_request->completion);
1348 * The default/initial value for all image request flags is 0. Each
1349 * is conditionally set to 1 at image request initialization time
1350 * and currently never change thereafter.
1352 static void img_request_write_set(struct rbd_img_request *img_request)
1354 set_bit(IMG_REQ_WRITE, &img_request->flags);
1358 static bool img_request_write_test(struct rbd_img_request *img_request)
1361 return test_bit(IMG_REQ_WRITE, &img_request->flags) != 0;
1364 static void img_request_child_set(struct rbd_img_request *img_request)
1366 set_bit(IMG_REQ_CHILD, &img_request->flags);
1370 static bool img_request_child_test(struct rbd_img_request *img_request)
1373 return test_bit(IMG_REQ_CHILD, &img_request->flags) != 0;
1376 static void img_request_layered_set(struct rbd_img_request *img_request)
1378 set_bit(IMG_REQ_LAYERED, &img_request->flags);
1382 static bool img_request_layered_test(struct rbd_img_request *img_request)
1385 return test_bit(IMG_REQ_LAYERED, &img_request->flags) != 0;
1389 rbd_img_obj_request_read_callback(struct rbd_obj_request *obj_request)
1391 u64 xferred = obj_request->xferred;
1392 u64 length = obj_request->length;
1394 dout("%s: obj %p img %p result %d %llu/%llu\n", __func__,
1395 obj_request, obj_request->img_request, obj_request->result,
1398 * ENOENT means a hole in the image. We zero-fill the
1399 * entire length of the request. A short read also implies
1400 * zero-fill to the end of the request. Either way we
1401 * update the xferred count to indicate the whole request
1404 rbd_assert(obj_request->type != OBJ_REQUEST_NODATA);
1405 if (obj_request->result == -ENOENT) {
1406 if (obj_request->type == OBJ_REQUEST_BIO)
1407 zero_bio_chain(obj_request->bio_list, 0);
1409 zero_pages(obj_request->pages, 0, length);
1410 obj_request->result = 0;
1411 obj_request->xferred = length;
1412 } else if (xferred < length && !obj_request->result) {
1413 if (obj_request->type == OBJ_REQUEST_BIO)
1414 zero_bio_chain(obj_request->bio_list, xferred);
1416 zero_pages(obj_request->pages, xferred, length);
1417 obj_request->xferred = length;
1419 obj_request_done_set(obj_request);
1422 static void rbd_obj_request_complete(struct rbd_obj_request *obj_request)
1424 dout("%s: obj %p cb %p\n", __func__, obj_request,
1425 obj_request->callback);
1426 if (obj_request->callback)
1427 obj_request->callback(obj_request);
1429 complete_all(&obj_request->completion);
1432 static void rbd_osd_trivial_callback(struct rbd_obj_request *obj_request)
1434 dout("%s: obj %p\n", __func__, obj_request);
1435 obj_request_done_set(obj_request);
1438 static void rbd_osd_read_callback(struct rbd_obj_request *obj_request)
1440 struct rbd_img_request *img_request = NULL;
1441 struct rbd_device *rbd_dev = NULL;
1442 bool layered = false;
1444 if (obj_request_img_data_test(obj_request)) {
1445 img_request = obj_request->img_request;
1446 layered = img_request && img_request_layered_test(img_request);
1447 rbd_dev = img_request->rbd_dev;
1450 dout("%s: obj %p img %p result %d %llu/%llu\n", __func__,
1451 obj_request, img_request, obj_request->result,
1452 obj_request->xferred, obj_request->length);
1453 if (layered && obj_request->result == -ENOENT &&
1454 obj_request->img_offset < rbd_dev->parent_overlap)
1455 rbd_img_parent_read(obj_request);
1456 else if (img_request)
1457 rbd_img_obj_request_read_callback(obj_request);
1459 obj_request_done_set(obj_request);
1462 static void rbd_osd_write_callback(struct rbd_obj_request *obj_request)
1464 dout("%s: obj %p result %d %llu\n", __func__, obj_request,
1465 obj_request->result, obj_request->length);
1467 * There is no such thing as a successful short write. Set
1468 * it to our originally-requested length.
1470 obj_request->xferred = obj_request->length;
1471 obj_request_done_set(obj_request);
1475 * For a simple stat call there's nothing to do. We'll do more if
1476 * this is part of a write sequence for a layered image.
1478 static void rbd_osd_stat_callback(struct rbd_obj_request *obj_request)
1480 dout("%s: obj %p\n", __func__, obj_request);
1481 obj_request_done_set(obj_request);
1484 static void rbd_osd_req_callback(struct ceph_osd_request *osd_req,
1485 struct ceph_msg *msg)
1487 struct rbd_obj_request *obj_request = osd_req->r_priv;
1490 dout("%s: osd_req %p msg %p\n", __func__, osd_req, msg);
1491 rbd_assert(osd_req == obj_request->osd_req);
1492 if (obj_request_img_data_test(obj_request)) {
1493 rbd_assert(obj_request->img_request);
1494 rbd_assert(obj_request->which != BAD_WHICH);
1496 rbd_assert(obj_request->which == BAD_WHICH);
1499 if (osd_req->r_result < 0)
1500 obj_request->result = osd_req->r_result;
1501 obj_request->version = le64_to_cpu(osd_req->r_reassert_version.version);
1503 BUG_ON(osd_req->r_num_ops > 2);
1506 * We support a 64-bit length, but ultimately it has to be
1507 * passed to blk_end_request(), which takes an unsigned int.
1509 obj_request->xferred = osd_req->r_reply_op_len[0];
1510 rbd_assert(obj_request->xferred < (u64)UINT_MAX);
1511 opcode = osd_req->r_ops[0].op;
1513 case CEPH_OSD_OP_READ:
1514 rbd_osd_read_callback(obj_request);
1516 case CEPH_OSD_OP_WRITE:
1517 rbd_osd_write_callback(obj_request);
1519 case CEPH_OSD_OP_STAT:
1520 rbd_osd_stat_callback(obj_request);
1522 case CEPH_OSD_OP_CALL:
1523 case CEPH_OSD_OP_NOTIFY_ACK:
1524 case CEPH_OSD_OP_WATCH:
1525 rbd_osd_trivial_callback(obj_request);
1528 rbd_warn(NULL, "%s: unsupported op %hu\n",
1529 obj_request->object_name, (unsigned short) opcode);
1533 if (obj_request_done_test(obj_request))
1534 rbd_obj_request_complete(obj_request);
1537 static void rbd_osd_req_format_read(struct rbd_obj_request *obj_request)
1539 struct rbd_img_request *img_request = obj_request->img_request;
1540 struct ceph_osd_request *osd_req = obj_request->osd_req;
1543 rbd_assert(osd_req != NULL);
1545 snap_id = img_request ? img_request->snap_id : CEPH_NOSNAP;
1546 ceph_osdc_build_request(osd_req, obj_request->offset,
1547 NULL, snap_id, NULL);
1550 static void rbd_osd_req_format_write(struct rbd_obj_request *obj_request)
1552 struct rbd_img_request *img_request = obj_request->img_request;
1553 struct ceph_osd_request *osd_req = obj_request->osd_req;
1554 struct ceph_snap_context *snapc;
1555 struct timespec mtime = CURRENT_TIME;
1557 rbd_assert(osd_req != NULL);
1559 snapc = img_request ? img_request->snapc : NULL;
1560 ceph_osdc_build_request(osd_req, obj_request->offset,
1561 snapc, CEPH_NOSNAP, &mtime);
1564 static struct ceph_osd_request *rbd_osd_req_create(
1565 struct rbd_device *rbd_dev,
1567 struct rbd_obj_request *obj_request)
1569 struct ceph_snap_context *snapc = NULL;
1570 struct ceph_osd_client *osdc;
1571 struct ceph_osd_request *osd_req;
1573 if (obj_request_img_data_test(obj_request)) {
1574 struct rbd_img_request *img_request = obj_request->img_request;
1576 rbd_assert(write_request ==
1577 img_request_write_test(img_request));
1579 snapc = img_request->snapc;
1582 /* Allocate and initialize the request, for the single op */
1584 osdc = &rbd_dev->rbd_client->client->osdc;
1585 osd_req = ceph_osdc_alloc_request(osdc, snapc, 1, false, GFP_ATOMIC);
1587 return NULL; /* ENOMEM */
1590 osd_req->r_flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK;
1592 osd_req->r_flags = CEPH_OSD_FLAG_READ;
1594 osd_req->r_callback = rbd_osd_req_callback;
1595 osd_req->r_priv = obj_request;
1597 osd_req->r_oid_len = strlen(obj_request->object_name);
1598 rbd_assert(osd_req->r_oid_len < sizeof (osd_req->r_oid));
1599 memcpy(osd_req->r_oid, obj_request->object_name, osd_req->r_oid_len);
1601 osd_req->r_file_layout = rbd_dev->layout; /* struct */
1607 * Create a copyup osd request based on the information in the
1608 * object request supplied. A copyup request has two osd ops,
1609 * a copyup method call, and a "normal" write request.
1611 static struct ceph_osd_request *
1612 rbd_osd_req_create_copyup(struct rbd_obj_request *obj_request)
1614 struct rbd_img_request *img_request;
1615 struct ceph_snap_context *snapc;
1616 struct rbd_device *rbd_dev;
1617 struct ceph_osd_client *osdc;
1618 struct ceph_osd_request *osd_req;
1620 rbd_assert(obj_request_img_data_test(obj_request));
1621 img_request = obj_request->img_request;
1622 rbd_assert(img_request);
1623 rbd_assert(img_request_write_test(img_request));
1625 /* Allocate and initialize the request, for the two ops */
1627 snapc = img_request->snapc;
1628 rbd_dev = img_request->rbd_dev;
1629 osdc = &rbd_dev->rbd_client->client->osdc;
1630 osd_req = ceph_osdc_alloc_request(osdc, snapc, 2, false, GFP_ATOMIC);
1632 return NULL; /* ENOMEM */
1634 osd_req->r_flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK;
1635 osd_req->r_callback = rbd_osd_req_callback;
1636 osd_req->r_priv = obj_request;
1638 osd_req->r_oid_len = strlen(obj_request->object_name);
1639 rbd_assert(osd_req->r_oid_len < sizeof (osd_req->r_oid));
1640 memcpy(osd_req->r_oid, obj_request->object_name, osd_req->r_oid_len);
1642 osd_req->r_file_layout = rbd_dev->layout; /* struct */
1648 static void rbd_osd_req_destroy(struct ceph_osd_request *osd_req)
1650 ceph_osdc_put_request(osd_req);
1653 /* object_name is assumed to be a non-null pointer and NUL-terminated */
1655 static struct rbd_obj_request *rbd_obj_request_create(const char *object_name,
1656 u64 offset, u64 length,
1657 enum obj_request_type type)
1659 struct rbd_obj_request *obj_request;
1663 rbd_assert(obj_request_type_valid(type));
1665 size = strlen(object_name) + 1;
1666 obj_request = kzalloc(sizeof (*obj_request) + size, GFP_KERNEL);
1670 name = (char *)(obj_request + 1);
1671 obj_request->object_name = memcpy(name, object_name, size);
1672 obj_request->offset = offset;
1673 obj_request->length = length;
1674 obj_request->flags = 0;
1675 obj_request->which = BAD_WHICH;
1676 obj_request->type = type;
1677 INIT_LIST_HEAD(&obj_request->links);
1678 init_completion(&obj_request->completion);
1679 kref_init(&obj_request->kref);
1681 dout("%s: \"%s\" %llu/%llu %d -> obj %p\n", __func__, object_name,
1682 offset, length, (int)type, obj_request);
1687 static void rbd_obj_request_destroy(struct kref *kref)
1689 struct rbd_obj_request *obj_request;
1691 obj_request = container_of(kref, struct rbd_obj_request, kref);
1693 dout("%s: obj %p\n", __func__, obj_request);
1695 rbd_assert(obj_request->img_request == NULL);
1696 rbd_assert(obj_request->which == BAD_WHICH);
1698 if (obj_request->osd_req)
1699 rbd_osd_req_destroy(obj_request->osd_req);
1701 rbd_assert(obj_request_type_valid(obj_request->type));
1702 switch (obj_request->type) {
1703 case OBJ_REQUEST_NODATA:
1704 break; /* Nothing to do */
1705 case OBJ_REQUEST_BIO:
1706 if (obj_request->bio_list)
1707 bio_chain_put(obj_request->bio_list);
1709 case OBJ_REQUEST_PAGES:
1710 if (obj_request->pages)
1711 ceph_release_page_vector(obj_request->pages,
1712 obj_request->page_count);
1720 * Caller is responsible for filling in the list of object requests
1721 * that comprises the image request, and the Linux request pointer
1722 * (if there is one).
1724 static struct rbd_img_request *rbd_img_request_create(
1725 struct rbd_device *rbd_dev,
1726 u64 offset, u64 length,
1730 struct rbd_img_request *img_request;
1731 struct ceph_snap_context *snapc = NULL;
1733 img_request = kmalloc(sizeof (*img_request), GFP_ATOMIC);
1737 if (write_request) {
1738 down_read(&rbd_dev->header_rwsem);
1739 snapc = ceph_get_snap_context(rbd_dev->header.snapc);
1740 up_read(&rbd_dev->header_rwsem);
1741 if (WARN_ON(!snapc)) {
1743 return NULL; /* Shouldn't happen */
1748 img_request->rq = NULL;
1749 img_request->rbd_dev = rbd_dev;
1750 img_request->offset = offset;
1751 img_request->length = length;
1752 img_request->flags = 0;
1753 if (write_request) {
1754 img_request_write_set(img_request);
1755 img_request->snapc = snapc;
1757 img_request->snap_id = rbd_dev->spec->snap_id;
1760 img_request_child_set(img_request);
1761 if (rbd_dev->parent_spec)
1762 img_request_layered_set(img_request);
1763 spin_lock_init(&img_request->completion_lock);
1764 img_request->next_completion = 0;
1765 img_request->callback = NULL;
1766 img_request->result = 0;
1767 img_request->obj_request_count = 0;
1768 INIT_LIST_HEAD(&img_request->obj_requests);
1769 kref_init(&img_request->kref);
1771 rbd_img_request_get(img_request); /* Avoid a warning */
1772 rbd_img_request_put(img_request); /* TEMPORARY */
1774 dout("%s: rbd_dev %p %s %llu/%llu -> img %p\n", __func__, rbd_dev,
1775 write_request ? "write" : "read", offset, length,
1781 static void rbd_img_request_destroy(struct kref *kref)
1783 struct rbd_img_request *img_request;
1784 struct rbd_obj_request *obj_request;
1785 struct rbd_obj_request *next_obj_request;
1787 img_request = container_of(kref, struct rbd_img_request, kref);
1789 dout("%s: img %p\n", __func__, img_request);
1791 for_each_obj_request_safe(img_request, obj_request, next_obj_request)
1792 rbd_img_obj_request_del(img_request, obj_request);
1793 rbd_assert(img_request->obj_request_count == 0);
1795 if (img_request_write_test(img_request))
1796 ceph_put_snap_context(img_request->snapc);
1798 if (img_request_child_test(img_request))
1799 rbd_obj_request_put(img_request->obj_request);
1804 static bool rbd_img_obj_end_request(struct rbd_obj_request *obj_request)
1806 struct rbd_img_request *img_request;
1807 unsigned int xferred;
1811 rbd_assert(obj_request_img_data_test(obj_request));
1812 img_request = obj_request->img_request;
1814 rbd_assert(obj_request->xferred <= (u64)UINT_MAX);
1815 xferred = (unsigned int)obj_request->xferred;
1816 result = obj_request->result;
1818 struct rbd_device *rbd_dev = img_request->rbd_dev;
1820 rbd_warn(rbd_dev, "%s %llx at %llx (%llx)\n",
1821 img_request_write_test(img_request) ? "write" : "read",
1822 obj_request->length, obj_request->img_offset,
1823 obj_request->offset);
1824 rbd_warn(rbd_dev, " result %d xferred %x\n",
1826 if (!img_request->result)
1827 img_request->result = result;
1830 /* Image object requests don't own their page array */
1832 if (obj_request->type == OBJ_REQUEST_PAGES) {
1833 obj_request->pages = NULL;
1834 obj_request->page_count = 0;
1837 if (img_request_child_test(img_request)) {
1838 rbd_assert(img_request->obj_request != NULL);
1839 more = obj_request->which < img_request->obj_request_count - 1;
1841 rbd_assert(img_request->rq != NULL);
1842 more = blk_end_request(img_request->rq, result, xferred);
1848 static void rbd_img_obj_callback(struct rbd_obj_request *obj_request)
1850 struct rbd_img_request *img_request;
1851 u32 which = obj_request->which;
1854 rbd_assert(obj_request_img_data_test(obj_request));
1855 img_request = obj_request->img_request;
1857 dout("%s: img %p obj %p\n", __func__, img_request, obj_request);
1858 rbd_assert(img_request != NULL);
1859 rbd_assert(img_request->obj_request_count > 0);
1860 rbd_assert(which != BAD_WHICH);
1861 rbd_assert(which < img_request->obj_request_count);
1862 rbd_assert(which >= img_request->next_completion);
1864 spin_lock_irq(&img_request->completion_lock);
1865 if (which != img_request->next_completion)
1868 for_each_obj_request_from(img_request, obj_request) {
1870 rbd_assert(which < img_request->obj_request_count);
1872 if (!obj_request_done_test(obj_request))
1874 more = rbd_img_obj_end_request(obj_request);
1878 rbd_assert(more ^ (which == img_request->obj_request_count));
1879 img_request->next_completion = which;
1881 spin_unlock_irq(&img_request->completion_lock);
1884 rbd_img_request_complete(img_request);
1888 * Split up an image request into one or more object requests, each
1889 * to a different object. The "type" parameter indicates whether
1890 * "data_desc" is the pointer to the head of a list of bio
1891 * structures, or the base of a page array. In either case this
1892 * function assumes data_desc describes memory sufficient to hold
1893 * all data described by the image request.
1895 static int rbd_img_request_fill(struct rbd_img_request *img_request,
1896 enum obj_request_type type,
1899 struct rbd_device *rbd_dev = img_request->rbd_dev;
1900 struct rbd_obj_request *obj_request = NULL;
1901 struct rbd_obj_request *next_obj_request;
1902 bool write_request = img_request_write_test(img_request);
1903 struct bio *bio_list;
1904 unsigned int bio_offset = 0;
1905 struct page **pages;
1910 dout("%s: img %p type %d data_desc %p\n", __func__, img_request,
1911 (int)type, data_desc);
1913 opcode = write_request ? CEPH_OSD_OP_WRITE : CEPH_OSD_OP_READ;
1914 img_offset = img_request->offset;
1915 resid = img_request->length;
1916 rbd_assert(resid > 0);
1918 if (type == OBJ_REQUEST_BIO) {
1919 bio_list = data_desc;
1920 rbd_assert(img_offset == bio_list->bi_sector << SECTOR_SHIFT);
1922 rbd_assert(type == OBJ_REQUEST_PAGES);
1927 struct ceph_osd_request *osd_req;
1928 const char *object_name;
1932 object_name = rbd_segment_name(rbd_dev, img_offset);
1935 offset = rbd_segment_offset(rbd_dev, img_offset);
1936 length = rbd_segment_length(rbd_dev, img_offset, resid);
1937 obj_request = rbd_obj_request_create(object_name,
1938 offset, length, type);
1939 kfree(object_name); /* object request has its own copy */
1943 if (type == OBJ_REQUEST_BIO) {
1944 unsigned int clone_size;
1946 rbd_assert(length <= (u64)UINT_MAX);
1947 clone_size = (unsigned int)length;
1948 obj_request->bio_list =
1949 bio_chain_clone_range(&bio_list,
1953 if (!obj_request->bio_list)
1956 unsigned int page_count;
1958 obj_request->pages = pages;
1959 page_count = (u32)calc_pages_for(offset, length);
1960 obj_request->page_count = page_count;
1961 if ((offset + length) & ~PAGE_MASK)
1962 page_count--; /* more on last page */
1963 pages += page_count;
1966 osd_req = rbd_osd_req_create(rbd_dev, write_request,
1970 obj_request->osd_req = osd_req;
1971 obj_request->callback = rbd_img_obj_callback;
1973 osd_req_op_extent_init(osd_req, 0, opcode, offset, length,
1975 if (type == OBJ_REQUEST_BIO)
1976 osd_req_op_extent_osd_data_bio(osd_req, 0,
1977 obj_request->bio_list, length);
1979 osd_req_op_extent_osd_data_pages(osd_req, 0,
1980 obj_request->pages, length,
1981 offset & ~PAGE_MASK, false, false);
1984 rbd_osd_req_format_write(obj_request);
1986 rbd_osd_req_format_read(obj_request);
1988 obj_request->img_offset = img_offset;
1989 rbd_img_obj_request_add(img_request, obj_request);
1991 img_offset += length;
1998 rbd_obj_request_put(obj_request);
2000 for_each_obj_request_safe(img_request, obj_request, next_obj_request)
2001 rbd_obj_request_put(obj_request);
2007 rbd_img_obj_copyup_callback(struct rbd_obj_request *obj_request)
2009 struct rbd_img_request *img_request;
2010 struct rbd_device *rbd_dev;
2014 rbd_assert(obj_request->type == OBJ_REQUEST_BIO);
2015 rbd_assert(obj_request_img_data_test(obj_request));
2016 img_request = obj_request->img_request;
2017 rbd_assert(img_request);
2019 rbd_dev = img_request->rbd_dev;
2020 rbd_assert(rbd_dev);
2021 length = (u64)1 << rbd_dev->header.obj_order;
2022 page_count = (u32)calc_pages_for(0, length);
2024 rbd_assert(obj_request->copyup_pages);
2025 ceph_release_page_vector(obj_request->copyup_pages, page_count);
2026 obj_request->copyup_pages = NULL;
2029 * We want the transfer count to reflect the size of the
2030 * original write request. There is no such thing as a
2031 * successful short write, so if the request was successful
2032 * we can just set it to the originally-requested length.
2034 if (!obj_request->result)
2035 obj_request->xferred = obj_request->length;
2037 /* Finish up with the normal image object callback */
2039 rbd_img_obj_callback(obj_request);
2043 rbd_img_obj_parent_read_full_callback(struct rbd_img_request *img_request)
2045 struct rbd_obj_request *orig_request;
2046 struct ceph_osd_request *osd_req;
2047 struct ceph_osd_client *osdc;
2048 struct rbd_device *rbd_dev;
2049 struct page **pages;
2054 rbd_assert(img_request_child_test(img_request));
2056 /* First get what we need from the image request */
2058 pages = img_request->copyup_pages;
2059 rbd_assert(pages != NULL);
2060 img_request->copyup_pages = NULL;
2062 orig_request = img_request->obj_request;
2063 rbd_assert(orig_request != NULL);
2064 rbd_assert(orig_request->type == OBJ_REQUEST_BIO);
2065 result = img_request->result;
2066 obj_size = img_request->length;
2067 xferred = img_request->xferred;
2069 rbd_dev = img_request->rbd_dev;
2070 rbd_assert(rbd_dev);
2071 rbd_assert(obj_size == (u64)1 << rbd_dev->header.obj_order);
2073 rbd_img_request_put(img_request);
2078 /* Allocate the new copyup osd request for the original request */
2081 rbd_assert(!orig_request->osd_req);
2082 osd_req = rbd_osd_req_create_copyup(orig_request);
2085 orig_request->osd_req = osd_req;
2086 orig_request->copyup_pages = pages;
2088 /* Initialize the copyup op */
2090 osd_req_op_cls_init(osd_req, 0, CEPH_OSD_OP_CALL, "rbd", "copyup");
2091 osd_req_op_cls_request_data_pages(osd_req, 0, pages, obj_size, 0,
2094 /* Then the original write request op */
2096 osd_req_op_extent_init(osd_req, 1, CEPH_OSD_OP_WRITE,
2097 orig_request->offset,
2098 orig_request->length, 0, 0);
2099 osd_req_op_extent_osd_data_bio(osd_req, 1, orig_request->bio_list,
2100 orig_request->length);
2102 rbd_osd_req_format_write(orig_request);
2104 /* All set, send it off. */
2106 orig_request->callback = rbd_img_obj_copyup_callback;
2107 osdc = &rbd_dev->rbd_client->client->osdc;
2108 result = rbd_obj_request_submit(osdc, orig_request);
2112 /* Record the error code and complete the request */
2114 orig_request->result = result;
2115 orig_request->xferred = 0;
2116 obj_request_done_set(orig_request);
2117 rbd_obj_request_complete(orig_request);
2121 * Read from the parent image the range of data that covers the
2122 * entire target of the given object request. This is used for
2123 * satisfying a layered image write request when the target of an
2124 * object request from the image request does not exist.
2126 * A page array big enough to hold the returned data is allocated
2127 * and supplied to rbd_img_request_fill() as the "data descriptor."
2128 * When the read completes, this page array will be transferred to
2129 * the original object request for the copyup operation.
2131 * If an error occurs, record it as the result of the original
2132 * object request and mark it done so it gets completed.
2134 static int rbd_img_obj_parent_read_full(struct rbd_obj_request *obj_request)
2136 struct rbd_img_request *img_request = NULL;
2137 struct rbd_img_request *parent_request = NULL;
2138 struct rbd_device *rbd_dev;
2141 struct page **pages = NULL;
2145 rbd_assert(obj_request_img_data_test(obj_request));
2146 rbd_assert(obj_request->type == OBJ_REQUEST_BIO);
2148 img_request = obj_request->img_request;
2149 rbd_assert(img_request != NULL);
2150 rbd_dev = img_request->rbd_dev;
2151 rbd_assert(rbd_dev->parent != NULL);
2154 * First things first. The original osd request is of no
2155 * use to use any more, we'll need a new one that can hold
2156 * the two ops in a copyup request. We'll get that later,
2157 * but for now we can release the old one.
2159 rbd_osd_req_destroy(obj_request->osd_req);
2160 obj_request->osd_req = NULL;
2163 * Determine the byte range covered by the object in the
2164 * child image to which the original request was to be sent.
2166 img_offset = obj_request->img_offset - obj_request->offset;
2167 length = (u64)1 << rbd_dev->header.obj_order;
2170 * There is no defined parent data beyond the parent
2171 * overlap, so limit what we read at that boundary if
2174 if (img_offset + length > rbd_dev->parent_overlap) {
2175 rbd_assert(img_offset < rbd_dev->parent_overlap);
2176 length = rbd_dev->parent_overlap - img_offset;
2180 * Allocate a page array big enough to receive the data read
2183 page_count = (u32)calc_pages_for(0, length);
2184 pages = ceph_alloc_page_vector(page_count, GFP_KERNEL);
2185 if (IS_ERR(pages)) {
2186 result = PTR_ERR(pages);
2192 parent_request = rbd_img_request_create(rbd_dev->parent,
2195 if (!parent_request)
2197 rbd_obj_request_get(obj_request);
2198 parent_request->obj_request = obj_request;
2200 result = rbd_img_request_fill(parent_request, OBJ_REQUEST_PAGES, pages);
2203 parent_request->copyup_pages = pages;
2205 parent_request->callback = rbd_img_obj_parent_read_full_callback;
2206 result = rbd_img_request_submit(parent_request);
2210 parent_request->copyup_pages = NULL;
2211 parent_request->obj_request = NULL;
2212 rbd_obj_request_put(obj_request);
2215 ceph_release_page_vector(pages, page_count);
2217 rbd_img_request_put(parent_request);
2218 obj_request->result = result;
2219 obj_request->xferred = 0;
2220 obj_request_done_set(obj_request);
2225 static void rbd_img_obj_exists_callback(struct rbd_obj_request *obj_request)
2227 struct rbd_obj_request *orig_request;
2230 rbd_assert(!obj_request_img_data_test(obj_request));
2233 * All we need from the object request is the original
2234 * request and the result of the STAT op. Grab those, then
2235 * we're done with the request.
2237 orig_request = obj_request->obj_request;
2238 obj_request->obj_request = NULL;
2239 rbd_assert(orig_request);
2240 rbd_assert(orig_request->img_request);
2242 result = obj_request->result;
2243 obj_request->result = 0;
2245 dout("%s: obj %p for obj %p result %d %llu/%llu\n", __func__,
2246 obj_request, orig_request, result,
2247 obj_request->xferred, obj_request->length);
2248 rbd_obj_request_put(obj_request);
2250 rbd_assert(orig_request);
2251 rbd_assert(orig_request->img_request);
2254 * Our only purpose here is to determine whether the object
2255 * exists, and we don't want to treat the non-existence as
2256 * an error. If something else comes back, transfer the
2257 * error to the original request and complete it now.
2260 obj_request_existence_set(orig_request, true);
2261 } else if (result == -ENOENT) {
2262 obj_request_existence_set(orig_request, false);
2263 } else if (result) {
2264 orig_request->result = result;
2269 * Resubmit the original request now that we have recorded
2270 * whether the target object exists.
2272 orig_request->result = rbd_img_obj_request_submit(orig_request);
2274 if (orig_request->result)
2275 rbd_obj_request_complete(orig_request);
2276 rbd_obj_request_put(orig_request);
2279 static int rbd_img_obj_exists_submit(struct rbd_obj_request *obj_request)
2281 struct rbd_obj_request *stat_request;
2282 struct rbd_device *rbd_dev;
2283 struct ceph_osd_client *osdc;
2284 struct page **pages = NULL;
2290 * The response data for a STAT call consists of:
2297 size = sizeof (__le64) + sizeof (__le32) + sizeof (__le32);
2298 page_count = (u32)calc_pages_for(0, size);
2299 pages = ceph_alloc_page_vector(page_count, GFP_KERNEL);
2301 return PTR_ERR(pages);
2304 stat_request = rbd_obj_request_create(obj_request->object_name, 0, 0,
2309 rbd_obj_request_get(obj_request);
2310 stat_request->obj_request = obj_request;
2311 stat_request->pages = pages;
2312 stat_request->page_count = page_count;
2314 rbd_assert(obj_request->img_request);
2315 rbd_dev = obj_request->img_request->rbd_dev;
2316 stat_request->osd_req = rbd_osd_req_create(rbd_dev, false,
2318 if (!stat_request->osd_req)
2320 stat_request->callback = rbd_img_obj_exists_callback;
2322 osd_req_op_init(stat_request->osd_req, 0, CEPH_OSD_OP_STAT);
2323 osd_req_op_raw_data_in_pages(stat_request->osd_req, 0, pages, size, 0,
2325 rbd_osd_req_format_read(stat_request);
2327 osdc = &rbd_dev->rbd_client->client->osdc;
2328 ret = rbd_obj_request_submit(osdc, stat_request);
2331 rbd_obj_request_put(obj_request);
2336 static int rbd_img_obj_request_submit(struct rbd_obj_request *obj_request)
2338 struct rbd_img_request *img_request;
2339 struct rbd_device *rbd_dev;
2342 rbd_assert(obj_request_img_data_test(obj_request));
2344 img_request = obj_request->img_request;
2345 rbd_assert(img_request);
2346 rbd_dev = img_request->rbd_dev;
2349 * Only writes to layered images need special handling.
2350 * Reads and non-layered writes are simple object requests.
2351 * Layered writes that start beyond the end of the overlap
2352 * with the parent have no parent data, so they too are
2353 * simple object requests. Finally, if the target object is
2354 * known to already exist, its parent data has already been
2355 * copied, so a write to the object can also be handled as a
2356 * simple object request.
2358 if (!img_request_write_test(img_request) ||
2359 !img_request_layered_test(img_request) ||
2360 rbd_dev->parent_overlap <= obj_request->img_offset ||
2361 ((known = obj_request_known_test(obj_request)) &&
2362 obj_request_exists_test(obj_request))) {
2364 struct rbd_device *rbd_dev;
2365 struct ceph_osd_client *osdc;
2367 rbd_dev = obj_request->img_request->rbd_dev;
2368 osdc = &rbd_dev->rbd_client->client->osdc;
2370 return rbd_obj_request_submit(osdc, obj_request);
2374 * It's a layered write. The target object might exist but
2375 * we may not know that yet. If we know it doesn't exist,
2376 * start by reading the data for the full target object from
2377 * the parent so we can use it for a copyup to the target.
2380 return rbd_img_obj_parent_read_full(obj_request);
2382 /* We don't know whether the target exists. Go find out. */
2384 return rbd_img_obj_exists_submit(obj_request);
2387 static int rbd_img_request_submit(struct rbd_img_request *img_request)
2389 struct rbd_obj_request *obj_request;
2390 struct rbd_obj_request *next_obj_request;
2392 dout("%s: img %p\n", __func__, img_request);
2393 for_each_obj_request_safe(img_request, obj_request, next_obj_request) {
2396 ret = rbd_img_obj_request_submit(obj_request);
2404 static void rbd_img_parent_read_callback(struct rbd_img_request *img_request)
2406 struct rbd_obj_request *obj_request;
2407 struct rbd_device *rbd_dev;
2410 rbd_assert(img_request_child_test(img_request));
2412 obj_request = img_request->obj_request;
2413 rbd_assert(obj_request);
2414 rbd_assert(obj_request->img_request);
2416 obj_request->result = img_request->result;
2417 if (obj_request->result)
2421 * We need to zero anything beyond the parent overlap
2422 * boundary. Since rbd_img_obj_request_read_callback()
2423 * will zero anything beyond the end of a short read, an
2424 * easy way to do this is to pretend the data from the
2425 * parent came up short--ending at the overlap boundary.
2427 rbd_assert(obj_request->img_offset < U64_MAX - obj_request->length);
2428 obj_end = obj_request->img_offset + obj_request->length;
2429 rbd_dev = obj_request->img_request->rbd_dev;
2430 if (obj_end > rbd_dev->parent_overlap) {
2433 if (obj_request->img_offset < rbd_dev->parent_overlap)
2434 xferred = rbd_dev->parent_overlap -
2435 obj_request->img_offset;
2437 obj_request->xferred = min(img_request->xferred, xferred);
2439 obj_request->xferred = img_request->xferred;
2442 rbd_img_obj_request_read_callback(obj_request);
2443 rbd_obj_request_complete(obj_request);
2446 static void rbd_img_parent_read(struct rbd_obj_request *obj_request)
2448 struct rbd_device *rbd_dev;
2449 struct rbd_img_request *img_request;
2452 rbd_assert(obj_request_img_data_test(obj_request));
2453 rbd_assert(obj_request->img_request != NULL);
2454 rbd_assert(obj_request->result == (s32) -ENOENT);
2455 rbd_assert(obj_request->type == OBJ_REQUEST_BIO);
2457 rbd_dev = obj_request->img_request->rbd_dev;
2458 rbd_assert(rbd_dev->parent != NULL);
2459 /* rbd_read_finish(obj_request, obj_request->length); */
2460 img_request = rbd_img_request_create(rbd_dev->parent,
2461 obj_request->img_offset,
2462 obj_request->length,
2468 rbd_obj_request_get(obj_request);
2469 img_request->obj_request = obj_request;
2471 result = rbd_img_request_fill(img_request, OBJ_REQUEST_BIO,
2472 obj_request->bio_list);
2476 img_request->callback = rbd_img_parent_read_callback;
2477 result = rbd_img_request_submit(img_request);
2484 rbd_img_request_put(img_request);
2485 obj_request->result = result;
2486 obj_request->xferred = 0;
2487 obj_request_done_set(obj_request);
2490 static int rbd_obj_notify_ack(struct rbd_device *rbd_dev,
2491 u64 ver, u64 notify_id)
2493 struct rbd_obj_request *obj_request;
2494 struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
2497 obj_request = rbd_obj_request_create(rbd_dev->header_name, 0, 0,
2498 OBJ_REQUEST_NODATA);
2503 obj_request->osd_req = rbd_osd_req_create(rbd_dev, false, obj_request);
2504 if (!obj_request->osd_req)
2506 obj_request->callback = rbd_obj_request_put;
2508 osd_req_op_watch_init(obj_request->osd_req, 0, CEPH_OSD_OP_NOTIFY_ACK,
2510 rbd_osd_req_format_read(obj_request);
2512 ret = rbd_obj_request_submit(osdc, obj_request);
2515 rbd_obj_request_put(obj_request);
2520 static void rbd_watch_cb(u64 ver, u64 notify_id, u8 opcode, void *data)
2522 struct rbd_device *rbd_dev = (struct rbd_device *)data;
2528 dout("%s: \"%s\" notify_id %llu opcode %u\n", __func__,
2529 rbd_dev->header_name, (unsigned long long) notify_id,
2530 (unsigned int) opcode);
2531 (void)rbd_dev_refresh(rbd_dev, &hver);
2533 rbd_obj_notify_ack(rbd_dev, hver, notify_id);
2537 * Request sync osd watch/unwatch. The value of "start" determines
2538 * whether a watch request is being initiated or torn down.
2540 static int rbd_dev_header_watch_sync(struct rbd_device *rbd_dev, int start)
2542 struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
2543 struct rbd_obj_request *obj_request;
2546 rbd_assert(start ^ !!rbd_dev->watch_event);
2547 rbd_assert(start ^ !!rbd_dev->watch_request);
2550 ret = ceph_osdc_create_event(osdc, rbd_watch_cb, rbd_dev,
2551 &rbd_dev->watch_event);
2554 rbd_assert(rbd_dev->watch_event != NULL);
2558 obj_request = rbd_obj_request_create(rbd_dev->header_name, 0, 0,
2559 OBJ_REQUEST_NODATA);
2563 obj_request->osd_req = rbd_osd_req_create(rbd_dev, true, obj_request);
2564 if (!obj_request->osd_req)
2568 ceph_osdc_set_request_linger(osdc, obj_request->osd_req);
2570 ceph_osdc_unregister_linger_request(osdc,
2571 rbd_dev->watch_request->osd_req);
2573 osd_req_op_watch_init(obj_request->osd_req, 0, CEPH_OSD_OP_WATCH,
2574 rbd_dev->watch_event->cookie,
2575 rbd_dev->header.obj_version, start);
2576 rbd_osd_req_format_write(obj_request);
2578 ret = rbd_obj_request_submit(osdc, obj_request);
2581 ret = rbd_obj_request_wait(obj_request);
2584 ret = obj_request->result;
2589 * A watch request is set to linger, so the underlying osd
2590 * request won't go away until we unregister it. We retain
2591 * a pointer to the object request during that time (in
2592 * rbd_dev->watch_request), so we'll keep a reference to
2593 * it. We'll drop that reference (below) after we've
2597 rbd_dev->watch_request = obj_request;
2602 /* We have successfully torn down the watch request */
2604 rbd_obj_request_put(rbd_dev->watch_request);
2605 rbd_dev->watch_request = NULL;
2607 /* Cancel the event if we're tearing down, or on error */
2608 ceph_osdc_cancel_event(rbd_dev->watch_event);
2609 rbd_dev->watch_event = NULL;
2611 rbd_obj_request_put(obj_request);
2617 * Synchronous osd object method call. Returns the number of bytes
2618 * returned in the outbound buffer, or a negative error code.
2620 static int rbd_obj_method_sync(struct rbd_device *rbd_dev,
2621 const char *object_name,
2622 const char *class_name,
2623 const char *method_name,
2624 const void *outbound,
2625 size_t outbound_size,
2627 size_t inbound_size,
2630 struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
2631 struct rbd_obj_request *obj_request;
2632 struct page **pages;
2637 * Method calls are ultimately read operations. The result
2638 * should placed into the inbound buffer provided. They
2639 * also supply outbound data--parameters for the object
2640 * method. Currently if this is present it will be a
2643 page_count = (u32)calc_pages_for(0, inbound_size);
2644 pages = ceph_alloc_page_vector(page_count, GFP_KERNEL);
2646 return PTR_ERR(pages);
2649 obj_request = rbd_obj_request_create(object_name, 0, inbound_size,
2654 obj_request->pages = pages;
2655 obj_request->page_count = page_count;
2657 obj_request->osd_req = rbd_osd_req_create(rbd_dev, false, obj_request);
2658 if (!obj_request->osd_req)
2661 osd_req_op_cls_init(obj_request->osd_req, 0, CEPH_OSD_OP_CALL,
2662 class_name, method_name);
2663 if (outbound_size) {
2664 struct ceph_pagelist *pagelist;
2666 pagelist = kmalloc(sizeof (*pagelist), GFP_NOFS);
2670 ceph_pagelist_init(pagelist);
2671 ceph_pagelist_append(pagelist, outbound, outbound_size);
2672 osd_req_op_cls_request_data_pagelist(obj_request->osd_req, 0,
2675 osd_req_op_cls_response_data_pages(obj_request->osd_req, 0,
2676 obj_request->pages, inbound_size,
2678 rbd_osd_req_format_read(obj_request);
2680 ret = rbd_obj_request_submit(osdc, obj_request);
2683 ret = rbd_obj_request_wait(obj_request);
2687 ret = obj_request->result;
2691 rbd_assert(obj_request->xferred < (u64)INT_MAX);
2692 ret = (int)obj_request->xferred;
2693 ceph_copy_from_page_vector(pages, inbound, 0, obj_request->xferred);
2695 *version = obj_request->version;
2698 rbd_obj_request_put(obj_request);
2700 ceph_release_page_vector(pages, page_count);
2705 static void rbd_request_fn(struct request_queue *q)
2706 __releases(q->queue_lock) __acquires(q->queue_lock)
2708 struct rbd_device *rbd_dev = q->queuedata;
2709 bool read_only = rbd_dev->mapping.read_only;
2713 while ((rq = blk_fetch_request(q))) {
2714 bool write_request = rq_data_dir(rq) == WRITE;
2715 struct rbd_img_request *img_request;
2719 /* Ignore any non-FS requests that filter through. */
2721 if (rq->cmd_type != REQ_TYPE_FS) {
2722 dout("%s: non-fs request type %d\n", __func__,
2723 (int) rq->cmd_type);
2724 __blk_end_request_all(rq, 0);
2728 /* Ignore/skip any zero-length requests */
2730 offset = (u64) blk_rq_pos(rq) << SECTOR_SHIFT;
2731 length = (u64) blk_rq_bytes(rq);
2734 dout("%s: zero-length request\n", __func__);
2735 __blk_end_request_all(rq, 0);
2739 spin_unlock_irq(q->queue_lock);
2741 /* Disallow writes to a read-only device */
2743 if (write_request) {
2747 rbd_assert(rbd_dev->spec->snap_id == CEPH_NOSNAP);
2751 * Quit early if the mapped snapshot no longer
2752 * exists. It's still possible the snapshot will
2753 * have disappeared by the time our request arrives
2754 * at the osd, but there's no sense in sending it if
2757 if (!test_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags)) {
2758 dout("request for non-existent snapshot");
2759 rbd_assert(rbd_dev->spec->snap_id != CEPH_NOSNAP);
2765 if (WARN_ON(offset && length > U64_MAX - offset + 1))
2766 goto end_request; /* Shouldn't happen */
2769 img_request = rbd_img_request_create(rbd_dev, offset, length,
2770 write_request, false);
2774 img_request->rq = rq;
2776 result = rbd_img_request_fill(img_request, OBJ_REQUEST_BIO,
2779 result = rbd_img_request_submit(img_request);
2781 rbd_img_request_put(img_request);
2783 spin_lock_irq(q->queue_lock);
2785 rbd_warn(rbd_dev, "%s %llx at %llx result %d\n",
2786 write_request ? "write" : "read",
2787 length, offset, result);
2789 __blk_end_request_all(rq, result);
2795 * a queue callback. Makes sure that we don't create a bio that spans across
2796 * multiple osd objects. One exception would be with a single page bios,
2797 * which we handle later at bio_chain_clone_range()
2799 static int rbd_merge_bvec(struct request_queue *q, struct bvec_merge_data *bmd,
2800 struct bio_vec *bvec)
2802 struct rbd_device *rbd_dev = q->queuedata;
2803 sector_t sector_offset;
2804 sector_t sectors_per_obj;
2805 sector_t obj_sector_offset;
2809 * Find how far into its rbd object the partition-relative
2810 * bio start sector is to offset relative to the enclosing
2813 sector_offset = get_start_sect(bmd->bi_bdev) + bmd->bi_sector;
2814 sectors_per_obj = 1 << (rbd_dev->header.obj_order - SECTOR_SHIFT);
2815 obj_sector_offset = sector_offset & (sectors_per_obj - 1);
2818 * Compute the number of bytes from that offset to the end
2819 * of the object. Account for what's already used by the bio.
2821 ret = (int) (sectors_per_obj - obj_sector_offset) << SECTOR_SHIFT;
2822 if (ret > bmd->bi_size)
2823 ret -= bmd->bi_size;
2828 * Don't send back more than was asked for. And if the bio
2829 * was empty, let the whole thing through because: "Note
2830 * that a block device *must* allow a single page to be
2831 * added to an empty bio."
2833 rbd_assert(bvec->bv_len <= PAGE_SIZE);
2834 if (ret > (int) bvec->bv_len || !bmd->bi_size)
2835 ret = (int) bvec->bv_len;
2840 static void rbd_free_disk(struct rbd_device *rbd_dev)
2842 struct gendisk *disk = rbd_dev->disk;
2847 rbd_dev->disk = NULL;
2848 if (disk->flags & GENHD_FL_UP) {
2851 blk_cleanup_queue(disk->queue);
2856 static int rbd_obj_read_sync(struct rbd_device *rbd_dev,
2857 const char *object_name,
2858 u64 offset, u64 length,
2859 void *buf, u64 *version)
2862 struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
2863 struct rbd_obj_request *obj_request;
2864 struct page **pages = NULL;
2869 page_count = (u32) calc_pages_for(offset, length);
2870 pages = ceph_alloc_page_vector(page_count, GFP_KERNEL);
2872 ret = PTR_ERR(pages);
2875 obj_request = rbd_obj_request_create(object_name, offset, length,
2880 obj_request->pages = pages;
2881 obj_request->page_count = page_count;
2883 obj_request->osd_req = rbd_osd_req_create(rbd_dev, false, obj_request);
2884 if (!obj_request->osd_req)
2887 osd_req_op_extent_init(obj_request->osd_req, 0, CEPH_OSD_OP_READ,
2888 offset, length, 0, 0);
2889 osd_req_op_extent_osd_data_pages(obj_request->osd_req, 0,
2891 obj_request->length,
2892 obj_request->offset & ~PAGE_MASK,
2894 rbd_osd_req_format_read(obj_request);
2896 ret = rbd_obj_request_submit(osdc, obj_request);
2899 ret = rbd_obj_request_wait(obj_request);
2903 ret = obj_request->result;
2907 rbd_assert(obj_request->xferred <= (u64) SIZE_MAX);
2908 size = (size_t) obj_request->xferred;
2909 ceph_copy_from_page_vector(pages, buf, 0, size);
2910 rbd_assert(size <= (size_t) INT_MAX);
2913 *version = obj_request->version;
2916 rbd_obj_request_put(obj_request);
2918 ceph_release_page_vector(pages, page_count);
2924 * Read the complete header for the given rbd device.
2926 * Returns a pointer to a dynamically-allocated buffer containing
2927 * the complete and validated header. Caller can pass the address
2928 * of a variable that will be filled in with the version of the
2929 * header object at the time it was read.
2931 * Returns a pointer-coded errno if a failure occurs.
2933 static struct rbd_image_header_ondisk *
2934 rbd_dev_v1_header_read(struct rbd_device *rbd_dev, u64 *version)
2936 struct rbd_image_header_ondisk *ondisk = NULL;
2943 * The complete header will include an array of its 64-bit
2944 * snapshot ids, followed by the names of those snapshots as
2945 * a contiguous block of NUL-terminated strings. Note that
2946 * the number of snapshots could change by the time we read
2947 * it in, in which case we re-read it.
2954 size = sizeof (*ondisk);
2955 size += snap_count * sizeof (struct rbd_image_snap_ondisk);
2957 ondisk = kmalloc(size, GFP_KERNEL);
2959 return ERR_PTR(-ENOMEM);
2961 ret = rbd_obj_read_sync(rbd_dev, rbd_dev->header_name,
2962 0, size, ondisk, version);
2965 if (WARN_ON((size_t) ret < size)) {
2967 rbd_warn(rbd_dev, "short header read (want %zd got %d)",
2971 if (!rbd_dev_ondisk_valid(ondisk)) {
2973 rbd_warn(rbd_dev, "invalid header");
2977 names_size = le64_to_cpu(ondisk->snap_names_len);
2978 want_count = snap_count;
2979 snap_count = le32_to_cpu(ondisk->snap_count);
2980 } while (snap_count != want_count);
2987 return ERR_PTR(ret);
2991 * reload the ondisk the header
2993 static int rbd_read_header(struct rbd_device *rbd_dev,
2994 struct rbd_image_header *header)
2996 struct rbd_image_header_ondisk *ondisk;
3000 ondisk = rbd_dev_v1_header_read(rbd_dev, &ver);
3002 return PTR_ERR(ondisk);
3003 ret = rbd_header_from_disk(header, ondisk);
3005 header->obj_version = ver;
3011 static void rbd_remove_all_snaps(struct rbd_device *rbd_dev)
3013 struct rbd_snap *snap;
3014 struct rbd_snap *next;
3016 list_for_each_entry_safe(snap, next, &rbd_dev->snaps, node) {
3017 list_del(&snap->node);
3018 rbd_snap_destroy(snap);
3022 static void rbd_update_mapping_size(struct rbd_device *rbd_dev)
3026 if (rbd_dev->spec->snap_id != CEPH_NOSNAP)
3029 size = (sector_t) rbd_dev->header.image_size / SECTOR_SIZE;
3030 dout("setting size to %llu sectors", (unsigned long long) size);
3031 rbd_dev->mapping.size = (u64) size;
3032 set_capacity(rbd_dev->disk, size);
3036 * only read the first part of the ondisk header, without the snaps info
3038 static int rbd_dev_v1_refresh(struct rbd_device *rbd_dev, u64 *hver)
3041 struct rbd_image_header h;
3043 ret = rbd_read_header(rbd_dev, &h);
3047 down_write(&rbd_dev->header_rwsem);
3049 /* Update image size, and check for resize of mapped image */
3050 rbd_dev->header.image_size = h.image_size;
3051 rbd_update_mapping_size(rbd_dev);
3053 /* rbd_dev->header.object_prefix shouldn't change */
3054 kfree(rbd_dev->header.snap_sizes);
3055 kfree(rbd_dev->header.snap_names);
3056 /* osd requests may still refer to snapc */
3057 ceph_put_snap_context(rbd_dev->header.snapc);
3060 *hver = h.obj_version;
3061 rbd_dev->header.obj_version = h.obj_version;
3062 rbd_dev->header.image_size = h.image_size;
3063 rbd_dev->header.snapc = h.snapc;
3064 rbd_dev->header.snap_names = h.snap_names;
3065 rbd_dev->header.snap_sizes = h.snap_sizes;
3066 /* Free the extra copy of the object prefix */
3067 WARN_ON(strcmp(rbd_dev->header.object_prefix, h.object_prefix));
3068 kfree(h.object_prefix);
3070 ret = rbd_dev_snaps_update(rbd_dev);
3072 up_write(&rbd_dev->header_rwsem);
3077 static int rbd_dev_refresh(struct rbd_device *rbd_dev, u64 *hver)
3081 rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
3082 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
3083 if (rbd_dev->image_format == 1)
3084 ret = rbd_dev_v1_refresh(rbd_dev, hver);
3086 ret = rbd_dev_v2_refresh(rbd_dev, hver);
3087 mutex_unlock(&ctl_mutex);
3088 revalidate_disk(rbd_dev->disk);
3090 rbd_warn(rbd_dev, "got notification but failed to "
3091 " update snaps: %d\n", ret);
3096 static int rbd_init_disk(struct rbd_device *rbd_dev)
3098 struct gendisk *disk;
3099 struct request_queue *q;
3102 /* create gendisk info */
3103 disk = alloc_disk(RBD_MINORS_PER_MAJOR);
3107 snprintf(disk->disk_name, sizeof(disk->disk_name), RBD_DRV_NAME "%d",
3109 disk->major = rbd_dev->major;
3110 disk->first_minor = 0;
3111 disk->fops = &rbd_bd_ops;
3112 disk->private_data = rbd_dev;
3114 q = blk_init_queue(rbd_request_fn, &rbd_dev->lock);
3118 /* We use the default size, but let's be explicit about it. */
3119 blk_queue_physical_block_size(q, SECTOR_SIZE);
3121 /* set io sizes to object size */
3122 segment_size = rbd_obj_bytes(&rbd_dev->header);
3123 blk_queue_max_hw_sectors(q, segment_size / SECTOR_SIZE);
3124 blk_queue_max_segment_size(q, segment_size);
3125 blk_queue_io_min(q, segment_size);
3126 blk_queue_io_opt(q, segment_size);
3128 blk_queue_merge_bvec(q, rbd_merge_bvec);
3131 q->queuedata = rbd_dev;
3133 rbd_dev->disk = disk;
3135 set_capacity(rbd_dev->disk, rbd_dev->mapping.size / SECTOR_SIZE);
3148 static struct rbd_device *dev_to_rbd_dev(struct device *dev)
3150 return container_of(dev, struct rbd_device, dev);
3153 static ssize_t rbd_size_show(struct device *dev,
3154 struct device_attribute *attr, char *buf)
3156 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3159 down_read(&rbd_dev->header_rwsem);
3160 size = get_capacity(rbd_dev->disk);
3161 up_read(&rbd_dev->header_rwsem);
3163 return sprintf(buf, "%llu\n", (unsigned long long) size * SECTOR_SIZE);
3167 * Note this shows the features for whatever's mapped, which is not
3168 * necessarily the base image.
3170 static ssize_t rbd_features_show(struct device *dev,
3171 struct device_attribute *attr, char *buf)
3173 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3175 return sprintf(buf, "0x%016llx\n",
3176 (unsigned long long) rbd_dev->mapping.features);
3179 static ssize_t rbd_major_show(struct device *dev,
3180 struct device_attribute *attr, char *buf)
3182 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3184 return sprintf(buf, "%d\n", rbd_dev->major);
3187 static ssize_t rbd_client_id_show(struct device *dev,
3188 struct device_attribute *attr, char *buf)
3190 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3192 return sprintf(buf, "client%lld\n",
3193 ceph_client_id(rbd_dev->rbd_client->client));
3196 static ssize_t rbd_pool_show(struct device *dev,
3197 struct device_attribute *attr, char *buf)
3199 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3201 return sprintf(buf, "%s\n", rbd_dev->spec->pool_name);
3204 static ssize_t rbd_pool_id_show(struct device *dev,
3205 struct device_attribute *attr, char *buf)
3207 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3209 return sprintf(buf, "%llu\n",
3210 (unsigned long long) rbd_dev->spec->pool_id);
3213 static ssize_t rbd_name_show(struct device *dev,
3214 struct device_attribute *attr, char *buf)
3216 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3218 if (rbd_dev->spec->image_name)
3219 return sprintf(buf, "%s\n", rbd_dev->spec->image_name);
3221 return sprintf(buf, "(unknown)\n");
3224 static ssize_t rbd_image_id_show(struct device *dev,
3225 struct device_attribute *attr, char *buf)
3227 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3229 return sprintf(buf, "%s\n", rbd_dev->spec->image_id);
3233 * Shows the name of the currently-mapped snapshot (or
3234 * RBD_SNAP_HEAD_NAME for the base image).
3236 static ssize_t rbd_snap_show(struct device *dev,
3237 struct device_attribute *attr,
3240 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3242 return sprintf(buf, "%s\n", rbd_dev->spec->snap_name);
3246 * For an rbd v2 image, shows the pool id, image id, and snapshot id
3247 * for the parent image. If there is no parent, simply shows
3248 * "(no parent image)".
3250 static ssize_t rbd_parent_show(struct device *dev,
3251 struct device_attribute *attr,
3254 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3255 struct rbd_spec *spec = rbd_dev->parent_spec;
3260 return sprintf(buf, "(no parent image)\n");
3262 count = sprintf(bufp, "pool_id %llu\npool_name %s\n",
3263 (unsigned long long) spec->pool_id, spec->pool_name);
3268 count = sprintf(bufp, "image_id %s\nimage_name %s\n", spec->image_id,
3269 spec->image_name ? spec->image_name : "(unknown)");
3274 count = sprintf(bufp, "snap_id %llu\nsnap_name %s\n",
3275 (unsigned long long) spec->snap_id, spec->snap_name);
3280 count = sprintf(bufp, "overlap %llu\n", rbd_dev->parent_overlap);
3285 return (ssize_t) (bufp - buf);
3288 static ssize_t rbd_image_refresh(struct device *dev,
3289 struct device_attribute *attr,
3293 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3296 ret = rbd_dev_refresh(rbd_dev, NULL);
3298 return ret < 0 ? ret : size;
3301 static DEVICE_ATTR(size, S_IRUGO, rbd_size_show, NULL);
3302 static DEVICE_ATTR(features, S_IRUGO, rbd_features_show, NULL);
3303 static DEVICE_ATTR(major, S_IRUGO, rbd_major_show, NULL);
3304 static DEVICE_ATTR(client_id, S_IRUGO, rbd_client_id_show, NULL);
3305 static DEVICE_ATTR(pool, S_IRUGO, rbd_pool_show, NULL);
3306 static DEVICE_ATTR(pool_id, S_IRUGO, rbd_pool_id_show, NULL);
3307 static DEVICE_ATTR(name, S_IRUGO, rbd_name_show, NULL);
3308 static DEVICE_ATTR(image_id, S_IRUGO, rbd_image_id_show, NULL);
3309 static DEVICE_ATTR(refresh, S_IWUSR, NULL, rbd_image_refresh);
3310 static DEVICE_ATTR(current_snap, S_IRUGO, rbd_snap_show, NULL);
3311 static DEVICE_ATTR(parent, S_IRUGO, rbd_parent_show, NULL);
3313 static struct attribute *rbd_attrs[] = {
3314 &dev_attr_size.attr,
3315 &dev_attr_features.attr,
3316 &dev_attr_major.attr,
3317 &dev_attr_client_id.attr,
3318 &dev_attr_pool.attr,
3319 &dev_attr_pool_id.attr,
3320 &dev_attr_name.attr,
3321 &dev_attr_image_id.attr,
3322 &dev_attr_current_snap.attr,
3323 &dev_attr_parent.attr,
3324 &dev_attr_refresh.attr,
3328 static struct attribute_group rbd_attr_group = {
3332 static const struct attribute_group *rbd_attr_groups[] = {
3337 static void rbd_sysfs_dev_release(struct device *dev)
3341 static struct device_type rbd_device_type = {
3343 .groups = rbd_attr_groups,
3344 .release = rbd_sysfs_dev_release,
3347 static struct rbd_spec *rbd_spec_get(struct rbd_spec *spec)
3349 kref_get(&spec->kref);
3354 static void rbd_spec_free(struct kref *kref);
3355 static void rbd_spec_put(struct rbd_spec *spec)
3358 kref_put(&spec->kref, rbd_spec_free);
3361 static struct rbd_spec *rbd_spec_alloc(void)
3363 struct rbd_spec *spec;
3365 spec = kzalloc(sizeof (*spec), GFP_KERNEL);
3368 kref_init(&spec->kref);
3373 static void rbd_spec_free(struct kref *kref)
3375 struct rbd_spec *spec = container_of(kref, struct rbd_spec, kref);
3377 kfree(spec->pool_name);
3378 kfree(spec->image_id);
3379 kfree(spec->image_name);
3380 kfree(spec->snap_name);
3384 static struct rbd_device *rbd_dev_create(struct rbd_client *rbdc,
3385 struct rbd_spec *spec)
3387 struct rbd_device *rbd_dev;
3389 rbd_dev = kzalloc(sizeof (*rbd_dev), GFP_KERNEL);
3393 spin_lock_init(&rbd_dev->lock);
3395 INIT_LIST_HEAD(&rbd_dev->node);
3396 INIT_LIST_HEAD(&rbd_dev->snaps);
3397 init_rwsem(&rbd_dev->header_rwsem);
3399 rbd_dev->spec = spec;
3400 rbd_dev->rbd_client = rbdc;
3402 /* Initialize the layout used for all rbd requests */
3404 rbd_dev->layout.fl_stripe_unit = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER);
3405 rbd_dev->layout.fl_stripe_count = cpu_to_le32(1);
3406 rbd_dev->layout.fl_object_size = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER);
3407 rbd_dev->layout.fl_pg_pool = cpu_to_le32((u32) spec->pool_id);
3412 static void rbd_dev_destroy(struct rbd_device *rbd_dev)
3414 rbd_spec_put(rbd_dev->parent_spec);
3415 kfree(rbd_dev->header_name);
3416 rbd_put_client(rbd_dev->rbd_client);
3417 rbd_spec_put(rbd_dev->spec);
3421 static void rbd_snap_destroy(struct rbd_snap *snap)
3427 static struct rbd_snap *rbd_snap_create(struct rbd_device *rbd_dev,
3428 const char *snap_name,
3429 u64 snap_id, u64 snap_size,
3432 struct rbd_snap *snap;
3434 snap = kzalloc(sizeof (*snap), GFP_KERNEL);
3436 return ERR_PTR(-ENOMEM);
3438 snap->name = snap_name;
3440 snap->size = snap_size;
3441 snap->features = snap_features;
3447 * Returns a dynamically-allocated snapshot name if successful, or a
3448 * pointer-coded error otherwise.
3450 static char *rbd_dev_v1_snap_info(struct rbd_device *rbd_dev, u32 which,
3451 u64 *snap_size, u64 *snap_features)
3456 rbd_assert(which < rbd_dev->header.snapc->num_snaps);
3458 /* Skip over names until we find the one we are looking for */
3460 snap_name = rbd_dev->header.snap_names;
3461 for (i = 0; i < which; i++)
3462 snap_name += strlen(snap_name) + 1;
3464 snap_name = kstrdup(snap_name, GFP_KERNEL);
3466 return ERR_PTR(-ENOMEM);
3468 *snap_size = rbd_dev->header.snap_sizes[which];
3469 *snap_features = 0; /* No features for v1 */
3475 * Get the size and object order for an image snapshot, or if
3476 * snap_id is CEPH_NOSNAP, gets this information for the base
3479 static int _rbd_dev_v2_snap_size(struct rbd_device *rbd_dev, u64 snap_id,
3480 u8 *order, u64 *snap_size)
3482 __le64 snapid = cpu_to_le64(snap_id);
3487 } __attribute__ ((packed)) size_buf = { 0 };
3489 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
3491 &snapid, sizeof (snapid),
3492 &size_buf, sizeof (size_buf), NULL);
3493 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
3496 if (ret < sizeof (size_buf))
3500 *order = size_buf.order;
3501 *snap_size = le64_to_cpu(size_buf.size);
3503 dout(" snap_id 0x%016llx order = %u, snap_size = %llu\n",
3504 (unsigned long long)snap_id, (unsigned int)*order,
3505 (unsigned long long)*snap_size);
3510 static int rbd_dev_v2_image_size(struct rbd_device *rbd_dev)
3512 return _rbd_dev_v2_snap_size(rbd_dev, CEPH_NOSNAP,
3513 &rbd_dev->header.obj_order,
3514 &rbd_dev->header.image_size);
3517 static int rbd_dev_v2_object_prefix(struct rbd_device *rbd_dev)
3523 reply_buf = kzalloc(RBD_OBJ_PREFIX_LEN_MAX, GFP_KERNEL);
3527 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
3528 "rbd", "get_object_prefix", NULL, 0,
3529 reply_buf, RBD_OBJ_PREFIX_LEN_MAX, NULL);
3530 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
3535 rbd_dev->header.object_prefix = ceph_extract_encoded_string(&p,
3536 p + ret, NULL, GFP_NOIO);
3539 if (IS_ERR(rbd_dev->header.object_prefix)) {
3540 ret = PTR_ERR(rbd_dev->header.object_prefix);
3541 rbd_dev->header.object_prefix = NULL;
3543 dout(" object_prefix = %s\n", rbd_dev->header.object_prefix);
3551 static int _rbd_dev_v2_snap_features(struct rbd_device *rbd_dev, u64 snap_id,
3554 __le64 snapid = cpu_to_le64(snap_id);
3558 } __attribute__ ((packed)) features_buf = { 0 };
3562 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
3563 "rbd", "get_features",
3564 &snapid, sizeof (snapid),
3565 &features_buf, sizeof (features_buf), NULL);
3566 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
3569 if (ret < sizeof (features_buf))
3572 incompat = le64_to_cpu(features_buf.incompat);
3573 if (incompat & ~RBD_FEATURES_SUPPORTED)
3576 *snap_features = le64_to_cpu(features_buf.features);
3578 dout(" snap_id 0x%016llx features = 0x%016llx incompat = 0x%016llx\n",
3579 (unsigned long long)snap_id,
3580 (unsigned long long)*snap_features,
3581 (unsigned long long)le64_to_cpu(features_buf.incompat));
3586 static int rbd_dev_v2_features(struct rbd_device *rbd_dev)
3588 return _rbd_dev_v2_snap_features(rbd_dev, CEPH_NOSNAP,
3589 &rbd_dev->header.features);
3592 static int rbd_dev_v2_parent_info(struct rbd_device *rbd_dev)
3594 struct rbd_spec *parent_spec;
3596 void *reply_buf = NULL;
3604 parent_spec = rbd_spec_alloc();
3608 size = sizeof (__le64) + /* pool_id */
3609 sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX + /* image_id */
3610 sizeof (__le64) + /* snap_id */
3611 sizeof (__le64); /* overlap */
3612 reply_buf = kmalloc(size, GFP_KERNEL);
3618 snapid = cpu_to_le64(CEPH_NOSNAP);
3619 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
3620 "rbd", "get_parent",
3621 &snapid, sizeof (snapid),
3622 reply_buf, size, NULL);
3623 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
3628 end = reply_buf + ret;
3630 ceph_decode_64_safe(&p, end, parent_spec->pool_id, out_err);
3631 if (parent_spec->pool_id == CEPH_NOPOOL)
3632 goto out; /* No parent? No problem. */
3634 /* The ceph file layout needs to fit pool id in 32 bits */
3637 if (WARN_ON(parent_spec->pool_id > (u64)U32_MAX))
3640 image_id = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL);
3641 if (IS_ERR(image_id)) {
3642 ret = PTR_ERR(image_id);
3645 parent_spec->image_id = image_id;
3646 ceph_decode_64_safe(&p, end, parent_spec->snap_id, out_err);
3647 ceph_decode_64_safe(&p, end, overlap, out_err);
3649 rbd_dev->parent_overlap = overlap;
3650 rbd_dev->parent_spec = parent_spec;
3651 parent_spec = NULL; /* rbd_dev now owns this */
3656 rbd_spec_put(parent_spec);
3661 static int rbd_dev_v2_striping_info(struct rbd_device *rbd_dev)
3665 __le64 stripe_count;
3666 } __attribute__ ((packed)) striping_info_buf = { 0 };
3667 size_t size = sizeof (striping_info_buf);
3674 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
3675 "rbd", "get_stripe_unit_count", NULL, 0,
3676 (char *)&striping_info_buf, size, NULL);
3677 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
3684 * We don't actually support the "fancy striping" feature
3685 * (STRIPINGV2) yet, but if the striping sizes are the
3686 * defaults the behavior is the same as before. So find
3687 * out, and only fail if the image has non-default values.
3690 obj_size = (u64)1 << rbd_dev->header.obj_order;
3691 p = &striping_info_buf;
3692 stripe_unit = ceph_decode_64(&p);
3693 if (stripe_unit != obj_size) {
3694 rbd_warn(rbd_dev, "unsupported stripe unit "
3695 "(got %llu want %llu)",
3696 stripe_unit, obj_size);
3699 stripe_count = ceph_decode_64(&p);
3700 if (stripe_count != 1) {
3701 rbd_warn(rbd_dev, "unsupported stripe count "
3702 "(got %llu want 1)", stripe_count);
3705 rbd_dev->stripe_unit = stripe_unit;
3706 rbd_dev->stripe_count = stripe_count;
3711 static char *rbd_dev_image_name(struct rbd_device *rbd_dev)
3713 size_t image_id_size;
3718 void *reply_buf = NULL;
3720 char *image_name = NULL;
3723 rbd_assert(!rbd_dev->spec->image_name);
3725 len = strlen(rbd_dev->spec->image_id);
3726 image_id_size = sizeof (__le32) + len;
3727 image_id = kmalloc(image_id_size, GFP_KERNEL);
3732 end = image_id + image_id_size;
3733 ceph_encode_string(&p, end, rbd_dev->spec->image_id, (u32)len);
3735 size = sizeof (__le32) + RBD_IMAGE_NAME_LEN_MAX;
3736 reply_buf = kmalloc(size, GFP_KERNEL);
3740 ret = rbd_obj_method_sync(rbd_dev, RBD_DIRECTORY,
3741 "rbd", "dir_get_name",
3742 image_id, image_id_size,
3743 reply_buf, size, NULL);
3747 end = reply_buf + ret;
3749 image_name = ceph_extract_encoded_string(&p, end, &len, GFP_KERNEL);
3750 if (IS_ERR(image_name))
3753 dout("%s: name is %s len is %zd\n", __func__, image_name, len);
3762 * When a parent image gets probed, we only have the pool, image,
3763 * and snapshot ids but not the names of any of them. This call
3764 * is made later to fill in those names. It has to be done after
3765 * rbd_dev_snaps_update() has completed because some of the
3766 * information (in particular, snapshot name) is not available
3769 static int rbd_dev_probe_update_spec(struct rbd_device *rbd_dev)
3771 struct ceph_osd_client *osdc;
3773 void *reply_buf = NULL;
3776 if (rbd_dev->spec->pool_name)
3777 return 0; /* Already have the names */
3779 /* Look up the pool name */
3781 osdc = &rbd_dev->rbd_client->client->osdc;
3782 name = ceph_pg_pool_name_by_id(osdc->osdmap, rbd_dev->spec->pool_id);
3784 rbd_warn(rbd_dev, "there is no pool with id %llu",
3785 rbd_dev->spec->pool_id); /* Really a BUG() */
3789 rbd_dev->spec->pool_name = kstrdup(name, GFP_KERNEL);
3790 if (!rbd_dev->spec->pool_name)
3793 /* Fetch the image name; tolerate failure here */
3795 name = rbd_dev_image_name(rbd_dev);
3797 rbd_dev->spec->image_name = (char *)name;
3799 rbd_warn(rbd_dev, "unable to get image name");
3801 /* Look up the snapshot name. */
3803 name = rbd_snap_name(rbd_dev, rbd_dev->spec->snap_id);
3805 rbd_warn(rbd_dev, "no snapshot with id %llu",
3806 rbd_dev->spec->snap_id); /* Really a BUG() */
3810 rbd_dev->spec->snap_name = kstrdup(name, GFP_KERNEL);
3811 if(!rbd_dev->spec->snap_name)
3817 kfree(rbd_dev->spec->pool_name);
3818 rbd_dev->spec->pool_name = NULL;
3823 static int rbd_dev_v2_snap_context(struct rbd_device *rbd_dev, u64 *ver)
3832 struct ceph_snap_context *snapc;
3836 * We'll need room for the seq value (maximum snapshot id),
3837 * snapshot count, and array of that many snapshot ids.
3838 * For now we have a fixed upper limit on the number we're
3839 * prepared to receive.
3841 size = sizeof (__le64) + sizeof (__le32) +
3842 RBD_MAX_SNAP_COUNT * sizeof (__le64);
3843 reply_buf = kzalloc(size, GFP_KERNEL);
3847 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
3848 "rbd", "get_snapcontext", NULL, 0,
3849 reply_buf, size, ver);
3850 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
3855 end = reply_buf + ret;
3857 ceph_decode_64_safe(&p, end, seq, out);
3858 ceph_decode_32_safe(&p, end, snap_count, out);
3861 * Make sure the reported number of snapshot ids wouldn't go
3862 * beyond the end of our buffer. But before checking that,
3863 * make sure the computed size of the snapshot context we
3864 * allocate is representable in a size_t.
3866 if (snap_count > (SIZE_MAX - sizeof (struct ceph_snap_context))
3871 if (!ceph_has_room(&p, end, snap_count * sizeof (__le64)))
3874 size = sizeof (struct ceph_snap_context) +
3875 snap_count * sizeof (snapc->snaps[0]);
3876 snapc = kmalloc(size, GFP_KERNEL);
3883 atomic_set(&snapc->nref, 1);
3885 snapc->num_snaps = snap_count;
3886 for (i = 0; i < snap_count; i++)
3887 snapc->snaps[i] = ceph_decode_64(&p);
3889 rbd_dev->header.snapc = snapc;
3891 dout(" snap context seq = %llu, snap_count = %u\n",
3892 (unsigned long long)seq, (unsigned int)snap_count);
3899 static char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev, u32 which)
3909 size = sizeof (__le32) + RBD_MAX_SNAP_NAME_LEN;
3910 reply_buf = kmalloc(size, GFP_KERNEL);
3912 return ERR_PTR(-ENOMEM);
3914 rbd_assert(which < rbd_dev->header.snapc->num_snaps);
3915 snap_id = cpu_to_le64(rbd_dev->header.snapc->snaps[which]);
3916 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
3917 "rbd", "get_snapshot_name",
3918 &snap_id, sizeof (snap_id),
3919 reply_buf, size, NULL);
3920 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
3922 snap_name = ERR_PTR(ret);
3927 end = reply_buf + ret;
3928 snap_name = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL);
3929 if (IS_ERR(snap_name))
3932 dout(" snap_id 0x%016llx snap_name = %s\n",
3933 (unsigned long long)le64_to_cpu(snap_id), snap_name);
3940 static char *rbd_dev_v2_snap_info(struct rbd_device *rbd_dev, u32 which,
3941 u64 *snap_size, u64 *snap_features)
3949 rbd_assert(which < rbd_dev->header.snapc->num_snaps);
3950 snap_id = rbd_dev->header.snapc->snaps[which];
3951 ret = _rbd_dev_v2_snap_size(rbd_dev, snap_id, NULL, &size);
3955 ret = _rbd_dev_v2_snap_features(rbd_dev, snap_id, &features);
3959 snap_name = rbd_dev_v2_snap_name(rbd_dev, which);
3960 if (!IS_ERR(snap_name)) {
3962 *snap_features = features;
3967 return ERR_PTR(ret);
3970 static char *rbd_dev_snap_info(struct rbd_device *rbd_dev, u32 which,
3971 u64 *snap_size, u64 *snap_features)
3973 if (rbd_dev->image_format == 1)
3974 return rbd_dev_v1_snap_info(rbd_dev, which,
3975 snap_size, snap_features);
3976 if (rbd_dev->image_format == 2)
3977 return rbd_dev_v2_snap_info(rbd_dev, which,
3978 snap_size, snap_features);
3979 return ERR_PTR(-EINVAL);
3982 static int rbd_dev_v2_refresh(struct rbd_device *rbd_dev, u64 *hver)
3987 down_write(&rbd_dev->header_rwsem);
3989 /* Grab old order first, to see if it changes */
3991 obj_order = rbd_dev->header.obj_order,
3992 ret = rbd_dev_v2_image_size(rbd_dev);
3995 if (rbd_dev->header.obj_order != obj_order) {
3999 rbd_update_mapping_size(rbd_dev);
4001 ret = rbd_dev_v2_snap_context(rbd_dev, hver);
4002 dout("rbd_dev_v2_snap_context returned %d\n", ret);
4005 ret = rbd_dev_snaps_update(rbd_dev);
4006 dout("rbd_dev_snaps_update returned %d\n", ret);
4010 up_write(&rbd_dev->header_rwsem);
4016 * Scan the rbd device's current snapshot list and compare it to the
4017 * newly-received snapshot context. Remove any existing snapshots
4018 * not present in the new snapshot context. Add a new snapshot for
4019 * any snaphots in the snapshot context not in the current list.
4020 * And verify there are no changes to snapshots we already know
4023 * Assumes the snapshots in the snapshot context are sorted by
4024 * snapshot id, highest id first. (Snapshots in the rbd_dev's list
4025 * are also maintained in that order.)
4027 * Note that any error occurs while updating the snapshot list
4028 * aborts the update, and the entire list is cleared. The snapshot
4029 * list becomes inconsistent at that point anyway, so it might as
4032 static int rbd_dev_snaps_update(struct rbd_device *rbd_dev)
4034 struct ceph_snap_context *snapc = rbd_dev->header.snapc;
4035 const u32 snap_count = snapc->num_snaps;
4036 struct list_head *head = &rbd_dev->snaps;
4037 struct list_head *links = head->next;
4041 dout("%s: snap count is %u\n", __func__, (unsigned int)snap_count);
4042 while (index < snap_count || links != head) {
4044 struct rbd_snap *snap;
4047 u64 snap_features = 0;
4049 snap_id = index < snap_count ? snapc->snaps[index]
4051 snap = links != head ? list_entry(links, struct rbd_snap, node)
4053 rbd_assert(!snap || snap->id != CEPH_NOSNAP);
4055 if (snap_id == CEPH_NOSNAP || (snap && snap->id > snap_id)) {
4056 struct list_head *next = links->next;
4059 * A previously-existing snapshot is not in
4060 * the new snap context.
4062 * If the now-missing snapshot is the one
4063 * the image represents, clear its existence
4064 * flag so we can avoid sending any more
4067 if (rbd_dev->spec->snap_id == snap->id)
4068 clear_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags);
4069 dout("removing %ssnap id %llu\n",
4070 rbd_dev->spec->snap_id == snap->id ?
4072 (unsigned long long)snap->id);
4074 list_del(&snap->node);
4075 rbd_snap_destroy(snap);
4077 /* Done with this list entry; advance */
4083 snap_name = rbd_dev_snap_info(rbd_dev, index,
4084 &snap_size, &snap_features);
4085 if (IS_ERR(snap_name)) {
4086 ret = PTR_ERR(snap_name);
4087 dout("failed to get snap info, error %d\n", ret);
4091 dout("entry %u: snap_id = %llu\n", (unsigned int)snap_count,
4092 (unsigned long long)snap_id);
4093 if (!snap || (snap_id != CEPH_NOSNAP && snap->id < snap_id)) {
4094 struct rbd_snap *new_snap;
4096 /* We haven't seen this snapshot before */
4098 new_snap = rbd_snap_create(rbd_dev, snap_name,
4099 snap_id, snap_size, snap_features);
4100 if (IS_ERR(new_snap)) {
4101 ret = PTR_ERR(new_snap);
4102 dout(" failed to add dev, error %d\n", ret);
4106 /* New goes before existing, or at end of list */
4108 dout(" added dev%s\n", snap ? "" : " at end\n");
4110 list_add_tail(&new_snap->node, &snap->node);
4112 list_add_tail(&new_snap->node, head);
4114 /* Already have this one */
4116 dout(" already present\n");
4118 rbd_assert(snap->size == snap_size);
4119 rbd_assert(!strcmp(snap->name, snap_name));
4120 rbd_assert(snap->features == snap_features);
4122 /* Done with this list entry; advance */
4124 links = links->next;
4127 /* Advance to the next entry in the snapshot context */
4131 dout("%s: done\n", __func__);
4135 rbd_remove_all_snaps(rbd_dev);
4140 static int rbd_bus_add_dev(struct rbd_device *rbd_dev)
4145 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
4147 dev = &rbd_dev->dev;
4148 dev->bus = &rbd_bus_type;
4149 dev->type = &rbd_device_type;
4150 dev->parent = &rbd_root_dev;
4151 dev->release = rbd_dev_release;
4152 dev_set_name(dev, "%d", rbd_dev->dev_id);
4153 ret = device_register(dev);
4155 mutex_unlock(&ctl_mutex);
4160 static void rbd_bus_del_dev(struct rbd_device *rbd_dev)
4162 device_unregister(&rbd_dev->dev);
4165 static atomic64_t rbd_dev_id_max = ATOMIC64_INIT(0);
4168 * Get a unique rbd identifier for the given new rbd_dev, and add
4169 * the rbd_dev to the global list. The minimum rbd id is 1.
4171 static void rbd_dev_id_get(struct rbd_device *rbd_dev)
4173 rbd_dev->dev_id = atomic64_inc_return(&rbd_dev_id_max);
4175 spin_lock(&rbd_dev_list_lock);
4176 list_add_tail(&rbd_dev->node, &rbd_dev_list);
4177 spin_unlock(&rbd_dev_list_lock);
4178 dout("rbd_dev %p given dev id %llu\n", rbd_dev,
4179 (unsigned long long) rbd_dev->dev_id);
4183 * Remove an rbd_dev from the global list, and record that its
4184 * identifier is no longer in use.
4186 static void rbd_dev_id_put(struct rbd_device *rbd_dev)
4188 struct list_head *tmp;
4189 int rbd_id = rbd_dev->dev_id;
4192 rbd_assert(rbd_id > 0);
4194 dout("rbd_dev %p released dev id %llu\n", rbd_dev,
4195 (unsigned long long) rbd_dev->dev_id);
4196 spin_lock(&rbd_dev_list_lock);
4197 list_del_init(&rbd_dev->node);
4200 * If the id being "put" is not the current maximum, there
4201 * is nothing special we need to do.
4203 if (rbd_id != atomic64_read(&rbd_dev_id_max)) {
4204 spin_unlock(&rbd_dev_list_lock);
4209 * We need to update the current maximum id. Search the
4210 * list to find out what it is. We're more likely to find
4211 * the maximum at the end, so search the list backward.
4214 list_for_each_prev(tmp, &rbd_dev_list) {
4215 struct rbd_device *rbd_dev;
4217 rbd_dev = list_entry(tmp, struct rbd_device, node);
4218 if (rbd_dev->dev_id > max_id)
4219 max_id = rbd_dev->dev_id;
4221 spin_unlock(&rbd_dev_list_lock);
4224 * The max id could have been updated by rbd_dev_id_get(), in
4225 * which case it now accurately reflects the new maximum.
4226 * Be careful not to overwrite the maximum value in that
4229 atomic64_cmpxchg(&rbd_dev_id_max, rbd_id, max_id);
4230 dout(" max dev id has been reset\n");
4234 * Skips over white space at *buf, and updates *buf to point to the
4235 * first found non-space character (if any). Returns the length of
4236 * the token (string of non-white space characters) found. Note
4237 * that *buf must be terminated with '\0'.
4239 static inline size_t next_token(const char **buf)
4242 * These are the characters that produce nonzero for
4243 * isspace() in the "C" and "POSIX" locales.
4245 const char *spaces = " \f\n\r\t\v";
4247 *buf += strspn(*buf, spaces); /* Find start of token */
4249 return strcspn(*buf, spaces); /* Return token length */
4253 * Finds the next token in *buf, and if the provided token buffer is
4254 * big enough, copies the found token into it. The result, if
4255 * copied, is guaranteed to be terminated with '\0'. Note that *buf
4256 * must be terminated with '\0' on entry.
4258 * Returns the length of the token found (not including the '\0').
4259 * Return value will be 0 if no token is found, and it will be >=
4260 * token_size if the token would not fit.
4262 * The *buf pointer will be updated to point beyond the end of the
4263 * found token. Note that this occurs even if the token buffer is
4264 * too small to hold it.
4266 static inline size_t copy_token(const char **buf,
4272 len = next_token(buf);
4273 if (len < token_size) {
4274 memcpy(token, *buf, len);
4275 *(token + len) = '\0';
4283 * Finds the next token in *buf, dynamically allocates a buffer big
4284 * enough to hold a copy of it, and copies the token into the new
4285 * buffer. The copy is guaranteed to be terminated with '\0'. Note
4286 * that a duplicate buffer is created even for a zero-length token.
4288 * Returns a pointer to the newly-allocated duplicate, or a null
4289 * pointer if memory for the duplicate was not available. If
4290 * the lenp argument is a non-null pointer, the length of the token
4291 * (not including the '\0') is returned in *lenp.
4293 * If successful, the *buf pointer will be updated to point beyond
4294 * the end of the found token.
4296 * Note: uses GFP_KERNEL for allocation.
4298 static inline char *dup_token(const char **buf, size_t *lenp)
4303 len = next_token(buf);
4304 dup = kmemdup(*buf, len + 1, GFP_KERNEL);
4307 *(dup + len) = '\0';
4317 * Parse the options provided for an "rbd add" (i.e., rbd image
4318 * mapping) request. These arrive via a write to /sys/bus/rbd/add,
4319 * and the data written is passed here via a NUL-terminated buffer.
4320 * Returns 0 if successful or an error code otherwise.
4322 * The information extracted from these options is recorded in
4323 * the other parameters which return dynamically-allocated
4326 * The address of a pointer that will refer to a ceph options
4327 * structure. Caller must release the returned pointer using
4328 * ceph_destroy_options() when it is no longer needed.
4330 * Address of an rbd options pointer. Fully initialized by
4331 * this function; caller must release with kfree().
4333 * Address of an rbd image specification pointer. Fully
4334 * initialized by this function based on parsed options.
4335 * Caller must release with rbd_spec_put().
4337 * The options passed take this form:
4338 * <mon_addrs> <options> <pool_name> <image_name> [<snap_id>]
4341 * A comma-separated list of one or more monitor addresses.
4342 * A monitor address is an ip address, optionally followed
4343 * by a port number (separated by a colon).
4344 * I.e.: ip1[:port1][,ip2[:port2]...]
4346 * A comma-separated list of ceph and/or rbd options.
4348 * The name of the rados pool containing the rbd image.
4350 * The name of the image in that pool to map.
4352 * An optional snapshot id. If provided, the mapping will
4353 * present data from the image at the time that snapshot was
4354 * created. The image head is used if no snapshot id is
4355 * provided. Snapshot mappings are always read-only.
4357 static int rbd_add_parse_args(const char *buf,
4358 struct ceph_options **ceph_opts,
4359 struct rbd_options **opts,
4360 struct rbd_spec **rbd_spec)
4364 const char *mon_addrs;
4365 size_t mon_addrs_size;
4366 struct rbd_spec *spec = NULL;
4367 struct rbd_options *rbd_opts = NULL;
4368 struct ceph_options *copts;
4371 /* The first four tokens are required */
4373 len = next_token(&buf);
4375 rbd_warn(NULL, "no monitor address(es) provided");
4379 mon_addrs_size = len + 1;
4383 options = dup_token(&buf, NULL);
4387 rbd_warn(NULL, "no options provided");
4391 spec = rbd_spec_alloc();
4395 spec->pool_name = dup_token(&buf, NULL);
4396 if (!spec->pool_name)
4398 if (!*spec->pool_name) {
4399 rbd_warn(NULL, "no pool name provided");
4403 spec->image_name = dup_token(&buf, NULL);
4404 if (!spec->image_name)
4406 if (!*spec->image_name) {
4407 rbd_warn(NULL, "no image name provided");
4412 * Snapshot name is optional; default is to use "-"
4413 * (indicating the head/no snapshot).
4415 len = next_token(&buf);
4417 buf = RBD_SNAP_HEAD_NAME; /* No snapshot supplied */
4418 len = sizeof (RBD_SNAP_HEAD_NAME) - 1;
4419 } else if (len > RBD_MAX_SNAP_NAME_LEN) {
4420 ret = -ENAMETOOLONG;
4423 spec->snap_name = kmemdup(buf, len + 1, GFP_KERNEL);
4424 if (!spec->snap_name)
4426 *(spec->snap_name + len) = '\0';
4428 /* Initialize all rbd options to the defaults */
4430 rbd_opts = kzalloc(sizeof (*rbd_opts), GFP_KERNEL);
4434 rbd_opts->read_only = RBD_READ_ONLY_DEFAULT;
4436 copts = ceph_parse_options(options, mon_addrs,
4437 mon_addrs + mon_addrs_size - 1,
4438 parse_rbd_opts_token, rbd_opts);
4439 if (IS_ERR(copts)) {
4440 ret = PTR_ERR(copts);
4461 * An rbd format 2 image has a unique identifier, distinct from the
4462 * name given to it by the user. Internally, that identifier is
4463 * what's used to specify the names of objects related to the image.
4465 * A special "rbd id" object is used to map an rbd image name to its
4466 * id. If that object doesn't exist, then there is no v2 rbd image
4467 * with the supplied name.
4469 * This function will record the given rbd_dev's image_id field if
4470 * it can be determined, and in that case will return 0. If any
4471 * errors occur a negative errno will be returned and the rbd_dev's
4472 * image_id field will be unchanged (and should be NULL).
4474 static int rbd_dev_image_id(struct rbd_device *rbd_dev)
4483 * When probing a parent image, the image id is already
4484 * known (and the image name likely is not). There's no
4485 * need to fetch the image id again in this case. We
4486 * do still need to set the image format though.
4488 if (rbd_dev->spec->image_id) {
4489 rbd_dev->image_format = *rbd_dev->spec->image_id ? 2 : 1;
4495 * First, see if the format 2 image id file exists, and if
4496 * so, get the image's persistent id from it.
4498 size = sizeof (RBD_ID_PREFIX) + strlen(rbd_dev->spec->image_name);
4499 object_name = kmalloc(size, GFP_NOIO);
4502 sprintf(object_name, "%s%s", RBD_ID_PREFIX, rbd_dev->spec->image_name);
4503 dout("rbd id object name is %s\n", object_name);
4505 /* Response will be an encoded string, which includes a length */
4507 size = sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX;
4508 response = kzalloc(size, GFP_NOIO);
4514 /* If it doesn't exist we'll assume it's a format 1 image */
4516 ret = rbd_obj_method_sync(rbd_dev, object_name,
4517 "rbd", "get_id", NULL, 0,
4518 response, RBD_IMAGE_ID_LEN_MAX, NULL);
4519 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
4520 if (ret == -ENOENT) {
4521 image_id = kstrdup("", GFP_KERNEL);
4522 ret = image_id ? 0 : -ENOMEM;
4524 rbd_dev->image_format = 1;
4525 } else if (ret > sizeof (__le32)) {
4528 image_id = ceph_extract_encoded_string(&p, p + ret,
4530 ret = IS_ERR(image_id) ? PTR_ERR(image_id) : 0;
4532 rbd_dev->image_format = 2;
4538 rbd_dev->spec->image_id = image_id;
4539 dout("image_id is %s\n", image_id);
4548 static int rbd_dev_v1_probe(struct rbd_device *rbd_dev)
4553 /* Record the header object name for this rbd image. */
4555 size = strlen(rbd_dev->spec->image_name) + sizeof (RBD_SUFFIX);
4556 rbd_dev->header_name = kmalloc(size, GFP_KERNEL);
4557 if (!rbd_dev->header_name) {
4561 sprintf(rbd_dev->header_name, "%s%s",
4562 rbd_dev->spec->image_name, RBD_SUFFIX);
4564 /* Populate rbd image metadata */
4566 ret = rbd_read_header(rbd_dev, &rbd_dev->header);
4570 /* Version 1 images have no parent (no layering) */
4572 rbd_dev->parent_spec = NULL;
4573 rbd_dev->parent_overlap = 0;
4575 dout("discovered version 1 image, header name is %s\n",
4576 rbd_dev->header_name);
4581 kfree(rbd_dev->header_name);
4582 rbd_dev->header_name = NULL;
4583 kfree(rbd_dev->spec->image_id);
4584 rbd_dev->spec->image_id = NULL;
4589 static int rbd_dev_v2_probe(struct rbd_device *rbd_dev)
4596 * Image id was filled in by the caller. Record the header
4597 * object name for this rbd image.
4599 size = sizeof (RBD_HEADER_PREFIX) + strlen(rbd_dev->spec->image_id);
4600 rbd_dev->header_name = kmalloc(size, GFP_KERNEL);
4601 if (!rbd_dev->header_name)
4603 sprintf(rbd_dev->header_name, "%s%s",
4604 RBD_HEADER_PREFIX, rbd_dev->spec->image_id);
4606 /* Get the size and object order for the image */
4607 ret = rbd_dev_v2_image_size(rbd_dev);
4611 /* Get the object prefix (a.k.a. block_name) for the image */
4613 ret = rbd_dev_v2_object_prefix(rbd_dev);
4617 /* Get the and check features for the image */
4619 ret = rbd_dev_v2_features(rbd_dev);
4623 /* If the image supports layering, get the parent info */
4625 if (rbd_dev->header.features & RBD_FEATURE_LAYERING) {
4626 ret = rbd_dev_v2_parent_info(rbd_dev);
4629 rbd_warn(rbd_dev, "WARNING: kernel support for "
4630 "layered rbd images is EXPERIMENTAL!");
4633 /* If the image supports fancy striping, get its parameters */
4635 if (rbd_dev->header.features & RBD_FEATURE_STRIPINGV2) {
4636 ret = rbd_dev_v2_striping_info(rbd_dev);
4641 /* crypto and compression type aren't (yet) supported for v2 images */
4643 rbd_dev->header.crypt_type = 0;
4644 rbd_dev->header.comp_type = 0;
4646 /* Get the snapshot context, plus the header version */
4648 ret = rbd_dev_v2_snap_context(rbd_dev, &ver);
4651 rbd_dev->header.obj_version = ver;
4653 dout("discovered version 2 image, header name is %s\n",
4654 rbd_dev->header_name);
4658 rbd_dev->parent_overlap = 0;
4659 rbd_spec_put(rbd_dev->parent_spec);
4660 rbd_dev->parent_spec = NULL;
4661 kfree(rbd_dev->header_name);
4662 rbd_dev->header_name = NULL;
4663 kfree(rbd_dev->header.object_prefix);
4664 rbd_dev->header.object_prefix = NULL;
4669 static int rbd_dev_probe_finish(struct rbd_device *rbd_dev)
4671 struct rbd_device *parent = NULL;
4672 struct rbd_spec *parent_spec = NULL;
4673 struct rbd_client *rbdc = NULL;
4676 /* no need to lock here, as rbd_dev is not registered yet */
4677 ret = rbd_dev_snaps_update(rbd_dev);
4681 ret = rbd_dev_probe_update_spec(rbd_dev);
4685 ret = rbd_dev_set_mapping(rbd_dev);
4689 /* generate unique id: find highest unique id, add one */
4690 rbd_dev_id_get(rbd_dev);
4692 /* Fill in the device name, now that we have its id. */
4693 BUILD_BUG_ON(DEV_NAME_LEN
4694 < sizeof (RBD_DRV_NAME) + MAX_INT_FORMAT_WIDTH);
4695 sprintf(rbd_dev->name, "%s%d", RBD_DRV_NAME, rbd_dev->dev_id);
4697 /* Get our block major device number. */
4699 ret = register_blkdev(0, rbd_dev->name);
4702 rbd_dev->major = ret;
4704 /* Set up the blkdev mapping. */
4706 ret = rbd_init_disk(rbd_dev);
4708 goto err_out_blkdev;
4710 ret = rbd_bus_add_dev(rbd_dev);
4715 * At this point cleanup in the event of an error is the job
4716 * of the sysfs code (initiated by rbd_bus_del_dev()).
4718 /* Probe the parent if there is one */
4720 if (rbd_dev->parent_spec) {
4722 * We need to pass a reference to the client and the
4723 * parent spec when creating the parent rbd_dev.
4724 * Images related by parent/child relationships
4725 * always share both.
4727 parent_spec = rbd_spec_get(rbd_dev->parent_spec);
4728 rbdc = __rbd_get_client(rbd_dev->rbd_client);
4730 parent = rbd_dev_create(rbdc, parent_spec);
4735 rbdc = NULL; /* parent now owns reference */
4736 parent_spec = NULL; /* parent now owns reference */
4737 ret = rbd_dev_probe(parent);
4739 goto err_out_parent;
4740 rbd_dev->parent = parent;
4743 ret = rbd_dev_header_watch_sync(rbd_dev, 1);
4747 /* Everything's ready. Announce the disk to the world. */
4749 add_disk(rbd_dev->disk);
4751 pr_info("%s: added with size 0x%llx\n", rbd_dev->disk->disk_name,
4752 (unsigned long long) rbd_dev->mapping.size);
4757 rbd_dev_destroy(parent);
4759 rbd_spec_put(parent_spec);
4760 rbd_put_client(rbdc);
4762 /* this will also clean up rest of rbd_dev stuff */
4764 rbd_bus_del_dev(rbd_dev);
4768 rbd_free_disk(rbd_dev);
4770 unregister_blkdev(rbd_dev->major, rbd_dev->name);
4772 rbd_dev_id_put(rbd_dev);
4774 rbd_remove_all_snaps(rbd_dev);
4780 * Probe for the existence of the header object for the given rbd
4781 * device. For format 2 images this includes determining the image
4784 static int rbd_dev_probe(struct rbd_device *rbd_dev)
4789 * Get the id from the image id object. If it's not a
4790 * format 2 image, we'll get ENOENT back, and we'll assume
4791 * it's a format 1 image.
4793 ret = rbd_dev_image_id(rbd_dev);
4796 rbd_assert(rbd_dev->spec->image_id);
4797 rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
4799 if (rbd_dev->image_format == 1)
4800 ret = rbd_dev_v1_probe(rbd_dev);
4802 ret = rbd_dev_v2_probe(rbd_dev);
4806 ret = rbd_dev_probe_finish(rbd_dev);
4808 rbd_header_free(&rbd_dev->header);
4812 kfree(rbd_dev->spec->image_id);
4813 rbd_dev->spec->image_id = NULL;
4815 dout("probe failed, returning %d\n", ret);
4820 static ssize_t rbd_add(struct bus_type *bus,
4824 struct rbd_device *rbd_dev = NULL;
4825 struct ceph_options *ceph_opts = NULL;
4826 struct rbd_options *rbd_opts = NULL;
4827 struct rbd_spec *spec = NULL;
4828 struct rbd_client *rbdc;
4829 struct ceph_osd_client *osdc;
4832 if (!try_module_get(THIS_MODULE))
4835 /* parse add command */
4836 rc = rbd_add_parse_args(buf, &ceph_opts, &rbd_opts, &spec);
4838 goto err_out_module;
4840 rbdc = rbd_get_client(ceph_opts);
4845 ceph_opts = NULL; /* rbd_dev client now owns this */
4848 osdc = &rbdc->client->osdc;
4849 rc = ceph_pg_poolid_by_name(osdc->osdmap, spec->pool_name);
4851 goto err_out_client;
4852 spec->pool_id = (u64) rc;
4854 /* The ceph file layout needs to fit pool id in 32 bits */
4856 if (WARN_ON(spec->pool_id > (u64) U32_MAX)) {
4858 goto err_out_client;
4861 rbd_dev = rbd_dev_create(rbdc, spec);
4863 goto err_out_client;
4864 rbdc = NULL; /* rbd_dev now owns this */
4865 spec = NULL; /* rbd_dev now owns this */
4867 rbd_dev->mapping.read_only = rbd_opts->read_only;
4869 rbd_opts = NULL; /* done with this */
4871 rc = rbd_dev_probe(rbd_dev);
4873 goto err_out_rbd_dev;
4877 rbd_dev_destroy(rbd_dev);
4879 rbd_put_client(rbdc);
4882 ceph_destroy_options(ceph_opts);
4886 module_put(THIS_MODULE);
4888 dout("Error adding device %s\n", buf);
4890 return (ssize_t) rc;
4893 static struct rbd_device *__rbd_get_dev(unsigned long dev_id)
4895 struct list_head *tmp;
4896 struct rbd_device *rbd_dev;
4898 spin_lock(&rbd_dev_list_lock);
4899 list_for_each(tmp, &rbd_dev_list) {
4900 rbd_dev = list_entry(tmp, struct rbd_device, node);
4901 if (rbd_dev->dev_id == dev_id) {
4902 spin_unlock(&rbd_dev_list_lock);
4906 spin_unlock(&rbd_dev_list_lock);
4910 static void rbd_dev_release(struct device *dev)
4912 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
4914 if (rbd_dev->watch_event)
4915 rbd_dev_header_watch_sync(rbd_dev, 0);
4917 /* clean up and free blkdev */
4918 rbd_free_disk(rbd_dev);
4919 unregister_blkdev(rbd_dev->major, rbd_dev->name);
4921 /* release allocated disk header fields */
4922 rbd_header_free(&rbd_dev->header);
4924 /* done with the id, and with the rbd_dev */
4925 rbd_dev_id_put(rbd_dev);
4926 rbd_assert(rbd_dev->rbd_client != NULL);
4927 rbd_dev_destroy(rbd_dev);
4929 /* release module ref */
4930 module_put(THIS_MODULE);
4933 static void __rbd_remove(struct rbd_device *rbd_dev)
4935 rbd_remove_all_snaps(rbd_dev);
4936 rbd_bus_del_dev(rbd_dev);
4939 static ssize_t rbd_remove(struct bus_type *bus,
4943 struct rbd_device *rbd_dev = NULL;
4948 rc = strict_strtoul(buf, 10, &ul);
4952 /* convert to int; abort if we lost anything in the conversion */
4953 target_id = (int) ul;
4954 if (target_id != ul)
4957 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
4959 rbd_dev = __rbd_get_dev(target_id);
4965 spin_lock_irq(&rbd_dev->lock);
4966 if (rbd_dev->open_count)
4969 set_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags);
4970 spin_unlock_irq(&rbd_dev->lock);
4974 while (rbd_dev->parent_spec) {
4975 struct rbd_device *first = rbd_dev;
4976 struct rbd_device *second = first->parent;
4977 struct rbd_device *third;
4980 * Follow to the parent with no grandparent and
4983 while (second && (third = second->parent)) {
4987 __rbd_remove(second);
4988 rbd_spec_put(first->parent_spec);
4989 first->parent_spec = NULL;
4990 first->parent_overlap = 0;
4991 first->parent = NULL;
4993 __rbd_remove(rbd_dev);
4996 mutex_unlock(&ctl_mutex);
5002 * create control files in sysfs
5005 static int rbd_sysfs_init(void)
5009 ret = device_register(&rbd_root_dev);
5013 ret = bus_register(&rbd_bus_type);
5015 device_unregister(&rbd_root_dev);
5020 static void rbd_sysfs_cleanup(void)
5022 bus_unregister(&rbd_bus_type);
5023 device_unregister(&rbd_root_dev);
5026 static int __init rbd_init(void)
5030 if (!libceph_compatible(NULL)) {
5031 rbd_warn(NULL, "libceph incompatibility (quitting)");
5035 rc = rbd_sysfs_init();
5038 pr_info("loaded " RBD_DRV_NAME_LONG "\n");
5042 static void __exit rbd_exit(void)
5044 rbd_sysfs_cleanup();
5047 module_init(rbd_init);
5048 module_exit(rbd_exit);
5050 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
5051 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
5052 MODULE_DESCRIPTION("rados block device");
5054 /* following authorship retained from original osdblk.c */
5055 MODULE_AUTHOR("Jeff Garzik <jeff@garzik.org>");
5057 MODULE_LICENSE("GPL");