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;
118 * An rbd image specification.
120 * The tuple (pool_id, image_id, snap_id) is sufficient to uniquely
121 * identify an image. Each rbd_dev structure includes a pointer to
122 * an rbd_spec structure that encapsulates this identity.
124 * Each of the id's in an rbd_spec has an associated name. For a
125 * user-mapped image, the names are supplied and the id's associated
126 * with them are looked up. For a layered image, a parent image is
127 * defined by the tuple, and the names are looked up.
129 * An rbd_dev structure contains a parent_spec pointer which is
130 * non-null if the image it represents is a child in a layered
131 * image. This pointer will refer to the rbd_spec structure used
132 * by the parent rbd_dev for its own identity (i.e., the structure
133 * is shared between the parent and child).
135 * Since these structures are populated once, during the discovery
136 * phase of image construction, they are effectively immutable so
137 * we make no effort to synchronize access to them.
139 * Note that code herein does not assume the image name is known (it
140 * could be a null pointer).
144 const char *pool_name;
146 const char *image_id;
147 const char *image_name;
150 const char *snap_name;
156 * an instance of the client. multiple devices may share an rbd client.
159 struct ceph_client *client;
161 struct list_head node;
164 struct rbd_img_request;
165 typedef void (*rbd_img_callback_t)(struct rbd_img_request *);
167 #define BAD_WHICH U32_MAX /* Good which or bad which, which? */
169 struct rbd_obj_request;
170 typedef void (*rbd_obj_callback_t)(struct rbd_obj_request *);
172 enum obj_request_type {
173 OBJ_REQUEST_NODATA, OBJ_REQUEST_BIO, OBJ_REQUEST_PAGES
177 OBJ_REQ_DONE, /* completion flag: not done = 0, done = 1 */
178 OBJ_REQ_IMG_DATA, /* object usage: standalone = 0, image = 1 */
179 OBJ_REQ_KNOWN, /* EXISTS flag valid: no = 0, yes = 1 */
180 OBJ_REQ_EXISTS, /* target exists: no = 0, yes = 1 */
183 struct rbd_obj_request {
184 const char *object_name;
185 u64 offset; /* object start byte */
186 u64 length; /* bytes from offset */
190 * An object request associated with an image will have its
191 * img_data flag set; a standalone object request will not.
193 * A standalone object request will have which == BAD_WHICH
194 * and a null obj_request pointer.
196 * An object request initiated in support of a layered image
197 * object (to check for its existence before a write) will
198 * have which == BAD_WHICH and a non-null obj_request pointer.
200 * Finally, an object request for rbd image data will have
201 * which != BAD_WHICH, and will have a non-null img_request
202 * pointer. The value of which will be in the range
203 * 0..(img_request->obj_request_count-1).
206 struct rbd_obj_request *obj_request; /* STAT op */
208 struct rbd_img_request *img_request;
210 /* links for img_request->obj_requests list */
211 struct list_head links;
214 u32 which; /* posn image request list */
216 enum obj_request_type type;
218 struct bio *bio_list;
224 struct page **copyup_pages;
226 struct ceph_osd_request *osd_req;
228 u64 xferred; /* bytes transferred */
232 rbd_obj_callback_t callback;
233 struct completion completion;
239 IMG_REQ_WRITE, /* I/O direction: read = 0, write = 1 */
240 IMG_REQ_CHILD, /* initiator: block = 0, child image = 1 */
241 IMG_REQ_LAYERED, /* ENOENT handling: normal = 0, layered = 1 */
244 struct rbd_img_request {
245 struct rbd_device *rbd_dev;
246 u64 offset; /* starting image byte offset */
247 u64 length; /* byte count from offset */
250 u64 snap_id; /* for reads */
251 struct ceph_snap_context *snapc; /* for writes */
254 struct request *rq; /* block request */
255 struct rbd_obj_request *obj_request; /* obj req initiator */
257 struct page **copyup_pages;
258 spinlock_t completion_lock;/* protects next_completion */
260 rbd_img_callback_t callback;
261 u64 xferred;/* aggregate bytes transferred */
262 int result; /* first nonzero obj_request result */
264 u32 obj_request_count;
265 struct list_head obj_requests; /* rbd_obj_request structs */
270 #define for_each_obj_request(ireq, oreq) \
271 list_for_each_entry(oreq, &(ireq)->obj_requests, links)
272 #define for_each_obj_request_from(ireq, oreq) \
273 list_for_each_entry_from(oreq, &(ireq)->obj_requests, links)
274 #define for_each_obj_request_safe(ireq, oreq, n) \
275 list_for_each_entry_safe_reverse(oreq, n, &(ireq)->obj_requests, links)
280 struct list_head node;
295 int dev_id; /* blkdev unique id */
297 int major; /* blkdev assigned major */
298 struct gendisk *disk; /* blkdev's gendisk and rq */
300 u32 image_format; /* Either 1 or 2 */
301 struct rbd_client *rbd_client;
303 char name[DEV_NAME_LEN]; /* blkdev name, e.g. rbd3 */
305 spinlock_t lock; /* queue, flags, open_count */
307 struct rbd_image_header header;
308 unsigned long flags; /* possibly lock protected */
309 struct rbd_spec *spec;
313 struct ceph_file_layout layout;
315 struct ceph_osd_event *watch_event;
316 struct rbd_obj_request *watch_request;
318 struct rbd_spec *parent_spec;
320 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_image_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);
674 /* Caller has to fill in snapc->seq and snapc->snaps[0..snap_count-1] */
676 static struct ceph_snap_context *rbd_snap_context_create(u32 snap_count)
678 struct ceph_snap_context *snapc;
681 size = sizeof (struct ceph_snap_context);
682 size += snap_count * sizeof (snapc->snaps[0]);
683 snapc = kzalloc(size, GFP_KERNEL);
687 atomic_set(&snapc->nref, 1);
688 snapc->num_snaps = snap_count;
693 static inline void rbd_snap_context_get(struct ceph_snap_context *snapc)
695 (void)ceph_get_snap_context(snapc);
698 static inline void rbd_snap_context_put(struct ceph_snap_context *snapc)
700 ceph_put_snap_context(snapc);
704 * Drop reference to ceph client node. If it's not referenced anymore, release
707 static void rbd_put_client(struct rbd_client *rbdc)
710 kref_put(&rbdc->kref, rbd_client_release);
713 static bool rbd_image_format_valid(u32 image_format)
715 return image_format == 1 || image_format == 2;
718 static bool rbd_dev_ondisk_valid(struct rbd_image_header_ondisk *ondisk)
723 /* The header has to start with the magic rbd header text */
724 if (memcmp(&ondisk->text, RBD_HEADER_TEXT, sizeof (RBD_HEADER_TEXT)))
727 /* The bio layer requires at least sector-sized I/O */
729 if (ondisk->options.order < SECTOR_SHIFT)
732 /* If we use u64 in a few spots we may be able to loosen this */
734 if (ondisk->options.order > 8 * sizeof (int) - 1)
738 * The size of a snapshot header has to fit in a size_t, and
739 * that limits the number of snapshots.
741 snap_count = le32_to_cpu(ondisk->snap_count);
742 size = SIZE_MAX - sizeof (struct ceph_snap_context);
743 if (snap_count > size / sizeof (__le64))
747 * Not only that, but the size of the entire the snapshot
748 * header must also be representable in a size_t.
750 size -= snap_count * sizeof (__le64);
751 if ((u64) size < le64_to_cpu(ondisk->snap_names_len))
758 * Create a new header structure, translate header format from the on-disk
761 static int rbd_header_from_disk(struct rbd_image_header *header,
762 struct rbd_image_header_ondisk *ondisk)
769 memset(header, 0, sizeof (*header));
771 snap_count = le32_to_cpu(ondisk->snap_count);
773 len = strnlen(ondisk->object_prefix, sizeof (ondisk->object_prefix));
774 header->object_prefix = kmalloc(len + 1, GFP_KERNEL);
775 if (!header->object_prefix)
777 memcpy(header->object_prefix, ondisk->object_prefix, len);
778 header->object_prefix[len] = '\0';
781 u64 snap_names_len = le64_to_cpu(ondisk->snap_names_len);
783 /* Save a copy of the snapshot names */
785 if (snap_names_len > (u64) SIZE_MAX)
787 header->snap_names = kmalloc(snap_names_len, GFP_KERNEL);
788 if (!header->snap_names)
791 * Note that rbd_dev_v1_header_read() guarantees
792 * the ondisk buffer we're working with has
793 * snap_names_len bytes beyond the end of the
794 * snapshot id array, this memcpy() is safe.
796 memcpy(header->snap_names, &ondisk->snaps[snap_count],
799 /* Record each snapshot's size */
801 size = snap_count * sizeof (*header->snap_sizes);
802 header->snap_sizes = kmalloc(size, GFP_KERNEL);
803 if (!header->snap_sizes)
805 for (i = 0; i < snap_count; i++)
806 header->snap_sizes[i] =
807 le64_to_cpu(ondisk->snaps[i].image_size);
809 header->snap_names = NULL;
810 header->snap_sizes = NULL;
813 header->features = 0; /* No features support in v1 images */
814 header->obj_order = ondisk->options.order;
815 header->crypt_type = ondisk->options.crypt_type;
816 header->comp_type = ondisk->options.comp_type;
818 /* Allocate and fill in the snapshot context */
820 header->image_size = le64_to_cpu(ondisk->image_size);
822 header->snapc = rbd_snap_context_create(snap_count);
825 header->snapc->seq = le64_to_cpu(ondisk->snap_seq);
826 for (i = 0; i < snap_count; i++)
827 header->snapc->snaps[i] = le64_to_cpu(ondisk->snaps[i].id);
832 kfree(header->snap_sizes);
833 header->snap_sizes = NULL;
834 kfree(header->snap_names);
835 header->snap_names = NULL;
836 kfree(header->object_prefix);
837 header->object_prefix = NULL;
842 static const char *rbd_snap_name(struct rbd_device *rbd_dev, u64 snap_id)
844 struct rbd_snap *snap;
846 if (snap_id == CEPH_NOSNAP)
847 return RBD_SNAP_HEAD_NAME;
849 list_for_each_entry(snap, &rbd_dev->snaps, node)
850 if (snap_id == snap->id)
856 static struct rbd_snap *snap_by_name(struct rbd_device *rbd_dev,
857 const char *snap_name)
859 struct rbd_snap *snap;
861 list_for_each_entry(snap, &rbd_dev->snaps, node)
862 if (!strcmp(snap_name, snap->name))
868 static int rbd_dev_set_mapping(struct rbd_device *rbd_dev)
870 if (!memcmp(rbd_dev->spec->snap_name, RBD_SNAP_HEAD_NAME,
871 sizeof (RBD_SNAP_HEAD_NAME))) {
872 rbd_dev->mapping.size = rbd_dev->header.image_size;
873 rbd_dev->mapping.features = rbd_dev->header.features;
875 struct rbd_snap *snap;
877 snap = snap_by_name(rbd_dev, rbd_dev->spec->snap_name);
880 rbd_dev->mapping.size = snap->size;
881 rbd_dev->mapping.features = snap->features;
882 rbd_dev->mapping.read_only = true;
884 set_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags);
889 static void rbd_header_free(struct rbd_image_header *header)
891 kfree(header->object_prefix);
892 header->object_prefix = NULL;
893 kfree(header->snap_sizes);
894 header->snap_sizes = NULL;
895 kfree(header->snap_names);
896 header->snap_names = NULL;
897 rbd_snap_context_put(header->snapc);
898 header->snapc = NULL;
901 static const char *rbd_segment_name(struct rbd_device *rbd_dev, u64 offset)
907 name = kmalloc(MAX_OBJ_NAME_SIZE + 1, GFP_NOIO);
910 segment = offset >> rbd_dev->header.obj_order;
911 ret = snprintf(name, MAX_OBJ_NAME_SIZE + 1, "%s.%012llx",
912 rbd_dev->header.object_prefix, segment);
913 if (ret < 0 || ret > MAX_OBJ_NAME_SIZE) {
914 pr_err("error formatting segment name for #%llu (%d)\n",
923 static u64 rbd_segment_offset(struct rbd_device *rbd_dev, u64 offset)
925 u64 segment_size = (u64) 1 << rbd_dev->header.obj_order;
927 return offset & (segment_size - 1);
930 static u64 rbd_segment_length(struct rbd_device *rbd_dev,
931 u64 offset, u64 length)
933 u64 segment_size = (u64) 1 << rbd_dev->header.obj_order;
935 offset &= segment_size - 1;
937 rbd_assert(length <= U64_MAX - offset);
938 if (offset + length > segment_size)
939 length = segment_size - offset;
945 * returns the size of an object in the image
947 static u64 rbd_obj_bytes(struct rbd_image_header *header)
949 return 1 << header->obj_order;
956 static void bio_chain_put(struct bio *chain)
962 chain = chain->bi_next;
968 * zeros a bio chain, starting at specific offset
970 static void zero_bio_chain(struct bio *chain, int start_ofs)
979 bio_for_each_segment(bv, chain, i) {
980 if (pos + bv->bv_len > start_ofs) {
981 int remainder = max(start_ofs - pos, 0);
982 buf = bvec_kmap_irq(bv, &flags);
983 memset(buf + remainder, 0,
984 bv->bv_len - remainder);
985 bvec_kunmap_irq(buf, &flags);
990 chain = chain->bi_next;
995 * similar to zero_bio_chain(), zeros data defined by a page array,
996 * starting at the given byte offset from the start of the array and
997 * continuing up to the given end offset. The pages array is
998 * assumed to be big enough to hold all bytes up to the end.
1000 static void zero_pages(struct page **pages, u64 offset, u64 end)
1002 struct page **page = &pages[offset >> PAGE_SHIFT];
1004 rbd_assert(end > offset);
1005 rbd_assert(end - offset <= (u64)SIZE_MAX);
1006 while (offset < end) {
1009 unsigned long flags;
1012 page_offset = (size_t)(offset & ~PAGE_MASK);
1013 length = min(PAGE_SIZE - page_offset, (size_t)(end - offset));
1014 local_irq_save(flags);
1015 kaddr = kmap_atomic(*page);
1016 memset(kaddr + page_offset, 0, length);
1017 kunmap_atomic(kaddr);
1018 local_irq_restore(flags);
1026 * Clone a portion of a bio, starting at the given byte offset
1027 * and continuing for the number of bytes indicated.
1029 static struct bio *bio_clone_range(struct bio *bio_src,
1030 unsigned int offset,
1038 unsigned short end_idx;
1039 unsigned short vcnt;
1042 /* Handle the easy case for the caller */
1044 if (!offset && len == bio_src->bi_size)
1045 return bio_clone(bio_src, gfpmask);
1047 if (WARN_ON_ONCE(!len))
1049 if (WARN_ON_ONCE(len > bio_src->bi_size))
1051 if (WARN_ON_ONCE(offset > bio_src->bi_size - len))
1054 /* Find first affected segment... */
1057 __bio_for_each_segment(bv, bio_src, idx, 0) {
1058 if (resid < bv->bv_len)
1060 resid -= bv->bv_len;
1064 /* ...and the last affected segment */
1067 __bio_for_each_segment(bv, bio_src, end_idx, idx) {
1068 if (resid <= bv->bv_len)
1070 resid -= bv->bv_len;
1072 vcnt = end_idx - idx + 1;
1074 /* Build the clone */
1076 bio = bio_alloc(gfpmask, (unsigned int) vcnt);
1078 return NULL; /* ENOMEM */
1080 bio->bi_bdev = bio_src->bi_bdev;
1081 bio->bi_sector = bio_src->bi_sector + (offset >> SECTOR_SHIFT);
1082 bio->bi_rw = bio_src->bi_rw;
1083 bio->bi_flags |= 1 << BIO_CLONED;
1086 * Copy over our part of the bio_vec, then update the first
1087 * and last (or only) entries.
1089 memcpy(&bio->bi_io_vec[0], &bio_src->bi_io_vec[idx],
1090 vcnt * sizeof (struct bio_vec));
1091 bio->bi_io_vec[0].bv_offset += voff;
1093 bio->bi_io_vec[0].bv_len -= voff;
1094 bio->bi_io_vec[vcnt - 1].bv_len = resid;
1096 bio->bi_io_vec[0].bv_len = len;
1099 bio->bi_vcnt = vcnt;
1107 * Clone a portion of a bio chain, starting at the given byte offset
1108 * into the first bio in the source chain and continuing for the
1109 * number of bytes indicated. The result is another bio chain of
1110 * exactly the given length, or a null pointer on error.
1112 * The bio_src and offset parameters are both in-out. On entry they
1113 * refer to the first source bio and the offset into that bio where
1114 * the start of data to be cloned is located.
1116 * On return, bio_src is updated to refer to the bio in the source
1117 * chain that contains first un-cloned byte, and *offset will
1118 * contain the offset of that byte within that bio.
1120 static struct bio *bio_chain_clone_range(struct bio **bio_src,
1121 unsigned int *offset,
1125 struct bio *bi = *bio_src;
1126 unsigned int off = *offset;
1127 struct bio *chain = NULL;
1130 /* Build up a chain of clone bios up to the limit */
1132 if (!bi || off >= bi->bi_size || !len)
1133 return NULL; /* Nothing to clone */
1137 unsigned int bi_size;
1141 rbd_warn(NULL, "bio_chain exhausted with %u left", len);
1142 goto out_err; /* EINVAL; ran out of bio's */
1144 bi_size = min_t(unsigned int, bi->bi_size - off, len);
1145 bio = bio_clone_range(bi, off, bi_size, gfpmask);
1147 goto out_err; /* ENOMEM */
1150 end = &bio->bi_next;
1153 if (off == bi->bi_size) {
1164 bio_chain_put(chain);
1170 * The default/initial value for all object request flags is 0. For
1171 * each flag, once its value is set to 1 it is never reset to 0
1174 static void obj_request_img_data_set(struct rbd_obj_request *obj_request)
1176 if (test_and_set_bit(OBJ_REQ_IMG_DATA, &obj_request->flags)) {
1177 struct rbd_device *rbd_dev;
1179 rbd_dev = obj_request->img_request->rbd_dev;
1180 rbd_warn(rbd_dev, "obj_request %p already marked img_data\n",
1185 static bool obj_request_img_data_test(struct rbd_obj_request *obj_request)
1188 return test_bit(OBJ_REQ_IMG_DATA, &obj_request->flags) != 0;
1191 static void obj_request_done_set(struct rbd_obj_request *obj_request)
1193 if (test_and_set_bit(OBJ_REQ_DONE, &obj_request->flags)) {
1194 struct rbd_device *rbd_dev = NULL;
1196 if (obj_request_img_data_test(obj_request))
1197 rbd_dev = obj_request->img_request->rbd_dev;
1198 rbd_warn(rbd_dev, "obj_request %p already marked done\n",
1203 static bool obj_request_done_test(struct rbd_obj_request *obj_request)
1206 return test_bit(OBJ_REQ_DONE, &obj_request->flags) != 0;
1210 * This sets the KNOWN flag after (possibly) setting the EXISTS
1211 * flag. The latter is set based on the "exists" value provided.
1213 * Note that for our purposes once an object exists it never goes
1214 * away again. It's possible that the response from two existence
1215 * checks are separated by the creation of the target object, and
1216 * the first ("doesn't exist") response arrives *after* the second
1217 * ("does exist"). In that case we ignore the second one.
1219 static void obj_request_existence_set(struct rbd_obj_request *obj_request,
1223 set_bit(OBJ_REQ_EXISTS, &obj_request->flags);
1224 set_bit(OBJ_REQ_KNOWN, &obj_request->flags);
1228 static bool obj_request_known_test(struct rbd_obj_request *obj_request)
1231 return test_bit(OBJ_REQ_KNOWN, &obj_request->flags) != 0;
1234 static bool obj_request_exists_test(struct rbd_obj_request *obj_request)
1237 return test_bit(OBJ_REQ_EXISTS, &obj_request->flags) != 0;
1240 static void rbd_obj_request_get(struct rbd_obj_request *obj_request)
1242 dout("%s: obj %p (was %d)\n", __func__, obj_request,
1243 atomic_read(&obj_request->kref.refcount));
1244 kref_get(&obj_request->kref);
1247 static void rbd_obj_request_destroy(struct kref *kref);
1248 static void rbd_obj_request_put(struct rbd_obj_request *obj_request)
1250 rbd_assert(obj_request != NULL);
1251 dout("%s: obj %p (was %d)\n", __func__, obj_request,
1252 atomic_read(&obj_request->kref.refcount));
1253 kref_put(&obj_request->kref, rbd_obj_request_destroy);
1256 static void rbd_img_request_get(struct rbd_img_request *img_request)
1258 dout("%s: img %p (was %d)\n", __func__, img_request,
1259 atomic_read(&img_request->kref.refcount));
1260 kref_get(&img_request->kref);
1263 static void rbd_img_request_destroy(struct kref *kref);
1264 static void rbd_img_request_put(struct rbd_img_request *img_request)
1266 rbd_assert(img_request != NULL);
1267 dout("%s: img %p (was %d)\n", __func__, img_request,
1268 atomic_read(&img_request->kref.refcount));
1269 kref_put(&img_request->kref, rbd_img_request_destroy);
1272 static inline void rbd_img_obj_request_add(struct rbd_img_request *img_request,
1273 struct rbd_obj_request *obj_request)
1275 rbd_assert(obj_request->img_request == NULL);
1277 /* Image request now owns object's original reference */
1278 obj_request->img_request = img_request;
1279 obj_request->which = img_request->obj_request_count;
1280 rbd_assert(!obj_request_img_data_test(obj_request));
1281 obj_request_img_data_set(obj_request);
1282 rbd_assert(obj_request->which != BAD_WHICH);
1283 img_request->obj_request_count++;
1284 list_add_tail(&obj_request->links, &img_request->obj_requests);
1285 dout("%s: img %p obj %p w=%u\n", __func__, img_request, obj_request,
1286 obj_request->which);
1289 static inline void rbd_img_obj_request_del(struct rbd_img_request *img_request,
1290 struct rbd_obj_request *obj_request)
1292 rbd_assert(obj_request->which != BAD_WHICH);
1294 dout("%s: img %p obj %p w=%u\n", __func__, img_request, obj_request,
1295 obj_request->which);
1296 list_del(&obj_request->links);
1297 rbd_assert(img_request->obj_request_count > 0);
1298 img_request->obj_request_count--;
1299 rbd_assert(obj_request->which == img_request->obj_request_count);
1300 obj_request->which = BAD_WHICH;
1301 rbd_assert(obj_request_img_data_test(obj_request));
1302 rbd_assert(obj_request->img_request == img_request);
1303 obj_request->img_request = NULL;
1304 obj_request->callback = NULL;
1305 rbd_obj_request_put(obj_request);
1308 static bool obj_request_type_valid(enum obj_request_type type)
1311 case OBJ_REQUEST_NODATA:
1312 case OBJ_REQUEST_BIO:
1313 case OBJ_REQUEST_PAGES:
1320 static int rbd_obj_request_submit(struct ceph_osd_client *osdc,
1321 struct rbd_obj_request *obj_request)
1323 dout("%s: osdc %p obj %p\n", __func__, osdc, obj_request);
1325 return ceph_osdc_start_request(osdc, obj_request->osd_req, false);
1328 static void rbd_img_request_complete(struct rbd_img_request *img_request)
1331 dout("%s: img %p\n", __func__, img_request);
1334 * If no error occurred, compute the aggregate transfer
1335 * count for the image request. We could instead use
1336 * atomic64_cmpxchg() to update it as each object request
1337 * completes; not clear which way is better off hand.
1339 if (!img_request->result) {
1340 struct rbd_obj_request *obj_request;
1343 for_each_obj_request(img_request, obj_request)
1344 xferred += obj_request->xferred;
1345 img_request->xferred = xferred;
1348 if (img_request->callback)
1349 img_request->callback(img_request);
1351 rbd_img_request_put(img_request);
1354 /* Caller is responsible for rbd_obj_request_destroy(obj_request) */
1356 static int rbd_obj_request_wait(struct rbd_obj_request *obj_request)
1358 dout("%s: obj %p\n", __func__, obj_request);
1360 return wait_for_completion_interruptible(&obj_request->completion);
1364 * The default/initial value for all image request flags is 0. Each
1365 * is conditionally set to 1 at image request initialization time
1366 * and currently never change thereafter.
1368 static void img_request_write_set(struct rbd_img_request *img_request)
1370 set_bit(IMG_REQ_WRITE, &img_request->flags);
1374 static bool img_request_write_test(struct rbd_img_request *img_request)
1377 return test_bit(IMG_REQ_WRITE, &img_request->flags) != 0;
1380 static void img_request_child_set(struct rbd_img_request *img_request)
1382 set_bit(IMG_REQ_CHILD, &img_request->flags);
1386 static bool img_request_child_test(struct rbd_img_request *img_request)
1389 return test_bit(IMG_REQ_CHILD, &img_request->flags) != 0;
1392 static void img_request_layered_set(struct rbd_img_request *img_request)
1394 set_bit(IMG_REQ_LAYERED, &img_request->flags);
1398 static bool img_request_layered_test(struct rbd_img_request *img_request)
1401 return test_bit(IMG_REQ_LAYERED, &img_request->flags) != 0;
1405 rbd_img_obj_request_read_callback(struct rbd_obj_request *obj_request)
1407 u64 xferred = obj_request->xferred;
1408 u64 length = obj_request->length;
1410 dout("%s: obj %p img %p result %d %llu/%llu\n", __func__,
1411 obj_request, obj_request->img_request, obj_request->result,
1414 * ENOENT means a hole in the image. We zero-fill the
1415 * entire length of the request. A short read also implies
1416 * zero-fill to the end of the request. Either way we
1417 * update the xferred count to indicate the whole request
1420 rbd_assert(obj_request->type != OBJ_REQUEST_NODATA);
1421 if (obj_request->result == -ENOENT) {
1422 if (obj_request->type == OBJ_REQUEST_BIO)
1423 zero_bio_chain(obj_request->bio_list, 0);
1425 zero_pages(obj_request->pages, 0, length);
1426 obj_request->result = 0;
1427 obj_request->xferred = length;
1428 } else if (xferred < length && !obj_request->result) {
1429 if (obj_request->type == OBJ_REQUEST_BIO)
1430 zero_bio_chain(obj_request->bio_list, xferred);
1432 zero_pages(obj_request->pages, xferred, length);
1433 obj_request->xferred = length;
1435 obj_request_done_set(obj_request);
1438 static void rbd_obj_request_complete(struct rbd_obj_request *obj_request)
1440 dout("%s: obj %p cb %p\n", __func__, obj_request,
1441 obj_request->callback);
1442 if (obj_request->callback)
1443 obj_request->callback(obj_request);
1445 complete_all(&obj_request->completion);
1448 static void rbd_osd_trivial_callback(struct rbd_obj_request *obj_request)
1450 dout("%s: obj %p\n", __func__, obj_request);
1451 obj_request_done_set(obj_request);
1454 static void rbd_osd_read_callback(struct rbd_obj_request *obj_request)
1456 struct rbd_img_request *img_request = NULL;
1457 struct rbd_device *rbd_dev = NULL;
1458 bool layered = false;
1460 if (obj_request_img_data_test(obj_request)) {
1461 img_request = obj_request->img_request;
1462 layered = img_request && img_request_layered_test(img_request);
1463 rbd_dev = img_request->rbd_dev;
1466 dout("%s: obj %p img %p result %d %llu/%llu\n", __func__,
1467 obj_request, img_request, obj_request->result,
1468 obj_request->xferred, obj_request->length);
1469 if (layered && obj_request->result == -ENOENT &&
1470 obj_request->img_offset < rbd_dev->parent_overlap)
1471 rbd_img_parent_read(obj_request);
1472 else if (img_request)
1473 rbd_img_obj_request_read_callback(obj_request);
1475 obj_request_done_set(obj_request);
1478 static void rbd_osd_write_callback(struct rbd_obj_request *obj_request)
1480 dout("%s: obj %p result %d %llu\n", __func__, obj_request,
1481 obj_request->result, obj_request->length);
1483 * There is no such thing as a successful short write. Set
1484 * it to our originally-requested length.
1486 obj_request->xferred = obj_request->length;
1487 obj_request_done_set(obj_request);
1491 * For a simple stat call there's nothing to do. We'll do more if
1492 * this is part of a write sequence for a layered image.
1494 static void rbd_osd_stat_callback(struct rbd_obj_request *obj_request)
1496 dout("%s: obj %p\n", __func__, obj_request);
1497 obj_request_done_set(obj_request);
1500 static void rbd_osd_req_callback(struct ceph_osd_request *osd_req,
1501 struct ceph_msg *msg)
1503 struct rbd_obj_request *obj_request = osd_req->r_priv;
1506 dout("%s: osd_req %p msg %p\n", __func__, osd_req, msg);
1507 rbd_assert(osd_req == obj_request->osd_req);
1508 if (obj_request_img_data_test(obj_request)) {
1509 rbd_assert(obj_request->img_request);
1510 rbd_assert(obj_request->which != BAD_WHICH);
1512 rbd_assert(obj_request->which == BAD_WHICH);
1515 if (osd_req->r_result < 0)
1516 obj_request->result = osd_req->r_result;
1517 obj_request->version = le64_to_cpu(osd_req->r_reassert_version.version);
1519 BUG_ON(osd_req->r_num_ops > 2);
1522 * We support a 64-bit length, but ultimately it has to be
1523 * passed to blk_end_request(), which takes an unsigned int.
1525 obj_request->xferred = osd_req->r_reply_op_len[0];
1526 rbd_assert(obj_request->xferred < (u64)UINT_MAX);
1527 opcode = osd_req->r_ops[0].op;
1529 case CEPH_OSD_OP_READ:
1530 rbd_osd_read_callback(obj_request);
1532 case CEPH_OSD_OP_WRITE:
1533 rbd_osd_write_callback(obj_request);
1535 case CEPH_OSD_OP_STAT:
1536 rbd_osd_stat_callback(obj_request);
1538 case CEPH_OSD_OP_CALL:
1539 case CEPH_OSD_OP_NOTIFY_ACK:
1540 case CEPH_OSD_OP_WATCH:
1541 rbd_osd_trivial_callback(obj_request);
1544 rbd_warn(NULL, "%s: unsupported op %hu\n",
1545 obj_request->object_name, (unsigned short) opcode);
1549 if (obj_request_done_test(obj_request))
1550 rbd_obj_request_complete(obj_request);
1553 static void rbd_osd_req_format_read(struct rbd_obj_request *obj_request)
1555 struct rbd_img_request *img_request = obj_request->img_request;
1556 struct ceph_osd_request *osd_req = obj_request->osd_req;
1559 rbd_assert(osd_req != NULL);
1561 snap_id = img_request ? img_request->snap_id : CEPH_NOSNAP;
1562 ceph_osdc_build_request(osd_req, obj_request->offset,
1563 NULL, snap_id, NULL);
1566 static void rbd_osd_req_format_write(struct rbd_obj_request *obj_request)
1568 struct rbd_img_request *img_request = obj_request->img_request;
1569 struct ceph_osd_request *osd_req = obj_request->osd_req;
1570 struct ceph_snap_context *snapc;
1571 struct timespec mtime = CURRENT_TIME;
1573 rbd_assert(osd_req != NULL);
1575 snapc = img_request ? img_request->snapc : NULL;
1576 ceph_osdc_build_request(osd_req, obj_request->offset,
1577 snapc, CEPH_NOSNAP, &mtime);
1580 static struct ceph_osd_request *rbd_osd_req_create(
1581 struct rbd_device *rbd_dev,
1583 struct rbd_obj_request *obj_request)
1585 struct ceph_snap_context *snapc = NULL;
1586 struct ceph_osd_client *osdc;
1587 struct ceph_osd_request *osd_req;
1589 if (obj_request_img_data_test(obj_request)) {
1590 struct rbd_img_request *img_request = obj_request->img_request;
1592 rbd_assert(write_request ==
1593 img_request_write_test(img_request));
1595 snapc = img_request->snapc;
1598 /* Allocate and initialize the request, for the single op */
1600 osdc = &rbd_dev->rbd_client->client->osdc;
1601 osd_req = ceph_osdc_alloc_request(osdc, snapc, 1, false, GFP_ATOMIC);
1603 return NULL; /* ENOMEM */
1606 osd_req->r_flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK;
1608 osd_req->r_flags = CEPH_OSD_FLAG_READ;
1610 osd_req->r_callback = rbd_osd_req_callback;
1611 osd_req->r_priv = obj_request;
1613 osd_req->r_oid_len = strlen(obj_request->object_name);
1614 rbd_assert(osd_req->r_oid_len < sizeof (osd_req->r_oid));
1615 memcpy(osd_req->r_oid, obj_request->object_name, osd_req->r_oid_len);
1617 osd_req->r_file_layout = rbd_dev->layout; /* struct */
1623 * Create a copyup osd request based on the information in the
1624 * object request supplied. A copyup request has two osd ops,
1625 * a copyup method call, and a "normal" write request.
1627 static struct ceph_osd_request *
1628 rbd_osd_req_create_copyup(struct rbd_obj_request *obj_request)
1630 struct rbd_img_request *img_request;
1631 struct ceph_snap_context *snapc;
1632 struct rbd_device *rbd_dev;
1633 struct ceph_osd_client *osdc;
1634 struct ceph_osd_request *osd_req;
1636 rbd_assert(obj_request_img_data_test(obj_request));
1637 img_request = obj_request->img_request;
1638 rbd_assert(img_request);
1639 rbd_assert(img_request_write_test(img_request));
1641 /* Allocate and initialize the request, for the two ops */
1643 snapc = img_request->snapc;
1644 rbd_dev = img_request->rbd_dev;
1645 osdc = &rbd_dev->rbd_client->client->osdc;
1646 osd_req = ceph_osdc_alloc_request(osdc, snapc, 2, false, GFP_ATOMIC);
1648 return NULL; /* ENOMEM */
1650 osd_req->r_flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK;
1651 osd_req->r_callback = rbd_osd_req_callback;
1652 osd_req->r_priv = obj_request;
1654 osd_req->r_oid_len = strlen(obj_request->object_name);
1655 rbd_assert(osd_req->r_oid_len < sizeof (osd_req->r_oid));
1656 memcpy(osd_req->r_oid, obj_request->object_name, osd_req->r_oid_len);
1658 osd_req->r_file_layout = rbd_dev->layout; /* struct */
1664 static void rbd_osd_req_destroy(struct ceph_osd_request *osd_req)
1666 ceph_osdc_put_request(osd_req);
1669 /* object_name is assumed to be a non-null pointer and NUL-terminated */
1671 static struct rbd_obj_request *rbd_obj_request_create(const char *object_name,
1672 u64 offset, u64 length,
1673 enum obj_request_type type)
1675 struct rbd_obj_request *obj_request;
1679 rbd_assert(obj_request_type_valid(type));
1681 size = strlen(object_name) + 1;
1682 obj_request = kzalloc(sizeof (*obj_request) + size, GFP_KERNEL);
1686 name = (char *)(obj_request + 1);
1687 obj_request->object_name = memcpy(name, object_name, size);
1688 obj_request->offset = offset;
1689 obj_request->length = length;
1690 obj_request->flags = 0;
1691 obj_request->which = BAD_WHICH;
1692 obj_request->type = type;
1693 INIT_LIST_HEAD(&obj_request->links);
1694 init_completion(&obj_request->completion);
1695 kref_init(&obj_request->kref);
1697 dout("%s: \"%s\" %llu/%llu %d -> obj %p\n", __func__, object_name,
1698 offset, length, (int)type, obj_request);
1703 static void rbd_obj_request_destroy(struct kref *kref)
1705 struct rbd_obj_request *obj_request;
1707 obj_request = container_of(kref, struct rbd_obj_request, kref);
1709 dout("%s: obj %p\n", __func__, obj_request);
1711 rbd_assert(obj_request->img_request == NULL);
1712 rbd_assert(obj_request->which == BAD_WHICH);
1714 if (obj_request->osd_req)
1715 rbd_osd_req_destroy(obj_request->osd_req);
1717 rbd_assert(obj_request_type_valid(obj_request->type));
1718 switch (obj_request->type) {
1719 case OBJ_REQUEST_NODATA:
1720 break; /* Nothing to do */
1721 case OBJ_REQUEST_BIO:
1722 if (obj_request->bio_list)
1723 bio_chain_put(obj_request->bio_list);
1725 case OBJ_REQUEST_PAGES:
1726 if (obj_request->pages)
1727 ceph_release_page_vector(obj_request->pages,
1728 obj_request->page_count);
1736 * Caller is responsible for filling in the list of object requests
1737 * that comprises the image request, and the Linux request pointer
1738 * (if there is one).
1740 static struct rbd_img_request *rbd_img_request_create(
1741 struct rbd_device *rbd_dev,
1742 u64 offset, u64 length,
1746 struct rbd_img_request *img_request;
1748 img_request = kmalloc(sizeof (*img_request), GFP_ATOMIC);
1752 if (write_request) {
1753 down_read(&rbd_dev->header_rwsem);
1754 rbd_snap_context_get(rbd_dev->header.snapc);
1755 up_read(&rbd_dev->header_rwsem);
1758 img_request->rq = NULL;
1759 img_request->rbd_dev = rbd_dev;
1760 img_request->offset = offset;
1761 img_request->length = length;
1762 img_request->flags = 0;
1763 if (write_request) {
1764 img_request_write_set(img_request);
1765 img_request->snapc = rbd_dev->header.snapc;
1767 img_request->snap_id = rbd_dev->spec->snap_id;
1770 img_request_child_set(img_request);
1771 if (rbd_dev->parent_spec)
1772 img_request_layered_set(img_request);
1773 spin_lock_init(&img_request->completion_lock);
1774 img_request->next_completion = 0;
1775 img_request->callback = NULL;
1776 img_request->result = 0;
1777 img_request->obj_request_count = 0;
1778 INIT_LIST_HEAD(&img_request->obj_requests);
1779 kref_init(&img_request->kref);
1781 rbd_img_request_get(img_request); /* Avoid a warning */
1782 rbd_img_request_put(img_request); /* TEMPORARY */
1784 dout("%s: rbd_dev %p %s %llu/%llu -> img %p\n", __func__, rbd_dev,
1785 write_request ? "write" : "read", offset, length,
1791 static void rbd_img_request_destroy(struct kref *kref)
1793 struct rbd_img_request *img_request;
1794 struct rbd_obj_request *obj_request;
1795 struct rbd_obj_request *next_obj_request;
1797 img_request = container_of(kref, struct rbd_img_request, kref);
1799 dout("%s: img %p\n", __func__, img_request);
1801 for_each_obj_request_safe(img_request, obj_request, next_obj_request)
1802 rbd_img_obj_request_del(img_request, obj_request);
1803 rbd_assert(img_request->obj_request_count == 0);
1805 if (img_request_write_test(img_request))
1806 rbd_snap_context_put(img_request->snapc);
1808 if (img_request_child_test(img_request))
1809 rbd_obj_request_put(img_request->obj_request);
1814 static bool rbd_img_obj_end_request(struct rbd_obj_request *obj_request)
1816 struct rbd_img_request *img_request;
1817 unsigned int xferred;
1821 rbd_assert(obj_request_img_data_test(obj_request));
1822 img_request = obj_request->img_request;
1824 rbd_assert(obj_request->xferred <= (u64)UINT_MAX);
1825 xferred = (unsigned int)obj_request->xferred;
1826 result = obj_request->result;
1828 struct rbd_device *rbd_dev = img_request->rbd_dev;
1830 rbd_warn(rbd_dev, "%s %llx at %llx (%llx)\n",
1831 img_request_write_test(img_request) ? "write" : "read",
1832 obj_request->length, obj_request->img_offset,
1833 obj_request->offset);
1834 rbd_warn(rbd_dev, " result %d xferred %x\n",
1836 if (!img_request->result)
1837 img_request->result = result;
1840 /* Image object requests don't own their page array */
1842 if (obj_request->type == OBJ_REQUEST_PAGES) {
1843 obj_request->pages = NULL;
1844 obj_request->page_count = 0;
1847 if (img_request_child_test(img_request)) {
1848 rbd_assert(img_request->obj_request != NULL);
1849 more = obj_request->which < img_request->obj_request_count - 1;
1851 rbd_assert(img_request->rq != NULL);
1852 more = blk_end_request(img_request->rq, result, xferred);
1858 static void rbd_img_obj_callback(struct rbd_obj_request *obj_request)
1860 struct rbd_img_request *img_request;
1861 u32 which = obj_request->which;
1864 rbd_assert(obj_request_img_data_test(obj_request));
1865 img_request = obj_request->img_request;
1867 dout("%s: img %p obj %p\n", __func__, img_request, obj_request);
1868 rbd_assert(img_request != NULL);
1869 rbd_assert(img_request->obj_request_count > 0);
1870 rbd_assert(which != BAD_WHICH);
1871 rbd_assert(which < img_request->obj_request_count);
1872 rbd_assert(which >= img_request->next_completion);
1874 spin_lock_irq(&img_request->completion_lock);
1875 if (which != img_request->next_completion)
1878 for_each_obj_request_from(img_request, obj_request) {
1880 rbd_assert(which < img_request->obj_request_count);
1882 if (!obj_request_done_test(obj_request))
1884 more = rbd_img_obj_end_request(obj_request);
1888 rbd_assert(more ^ (which == img_request->obj_request_count));
1889 img_request->next_completion = which;
1891 spin_unlock_irq(&img_request->completion_lock);
1894 rbd_img_request_complete(img_request);
1898 * Split up an image request into one or more object requests, each
1899 * to a different object. The "type" parameter indicates whether
1900 * "data_desc" is the pointer to the head of a list of bio
1901 * structures, or the base of a page array. In either case this
1902 * function assumes data_desc describes memory sufficient to hold
1903 * all data described by the image request.
1905 static int rbd_img_request_fill(struct rbd_img_request *img_request,
1906 enum obj_request_type type,
1909 struct rbd_device *rbd_dev = img_request->rbd_dev;
1910 struct rbd_obj_request *obj_request = NULL;
1911 struct rbd_obj_request *next_obj_request;
1912 bool write_request = img_request_write_test(img_request);
1913 struct bio *bio_list;
1914 unsigned int bio_offset = 0;
1915 struct page **pages;
1920 dout("%s: img %p type %d data_desc %p\n", __func__, img_request,
1921 (int)type, data_desc);
1923 opcode = write_request ? CEPH_OSD_OP_WRITE : CEPH_OSD_OP_READ;
1924 img_offset = img_request->offset;
1925 resid = img_request->length;
1926 rbd_assert(resid > 0);
1928 if (type == OBJ_REQUEST_BIO) {
1929 bio_list = data_desc;
1930 rbd_assert(img_offset == bio_list->bi_sector << SECTOR_SHIFT);
1932 rbd_assert(type == OBJ_REQUEST_PAGES);
1937 struct ceph_osd_request *osd_req;
1938 const char *object_name;
1942 object_name = rbd_segment_name(rbd_dev, img_offset);
1945 offset = rbd_segment_offset(rbd_dev, img_offset);
1946 length = rbd_segment_length(rbd_dev, img_offset, resid);
1947 obj_request = rbd_obj_request_create(object_name,
1948 offset, length, type);
1949 kfree(object_name); /* object request has its own copy */
1953 if (type == OBJ_REQUEST_BIO) {
1954 unsigned int clone_size;
1956 rbd_assert(length <= (u64)UINT_MAX);
1957 clone_size = (unsigned int)length;
1958 obj_request->bio_list =
1959 bio_chain_clone_range(&bio_list,
1963 if (!obj_request->bio_list)
1966 unsigned int page_count;
1968 obj_request->pages = pages;
1969 page_count = (u32)calc_pages_for(offset, length);
1970 obj_request->page_count = page_count;
1971 if ((offset + length) & ~PAGE_MASK)
1972 page_count--; /* more on last page */
1973 pages += page_count;
1976 osd_req = rbd_osd_req_create(rbd_dev, write_request,
1980 obj_request->osd_req = osd_req;
1981 obj_request->callback = rbd_img_obj_callback;
1983 osd_req_op_extent_init(osd_req, 0, opcode, offset, length,
1985 if (type == OBJ_REQUEST_BIO)
1986 osd_req_op_extent_osd_data_bio(osd_req, 0,
1987 obj_request->bio_list, length);
1989 osd_req_op_extent_osd_data_pages(osd_req, 0,
1990 obj_request->pages, length,
1991 offset & ~PAGE_MASK, false, false);
1994 rbd_osd_req_format_write(obj_request);
1996 rbd_osd_req_format_read(obj_request);
1998 obj_request->img_offset = img_offset;
1999 rbd_img_obj_request_add(img_request, obj_request);
2001 img_offset += length;
2008 rbd_obj_request_put(obj_request);
2010 for_each_obj_request_safe(img_request, obj_request, next_obj_request)
2011 rbd_obj_request_put(obj_request);
2017 rbd_img_obj_copyup_callback(struct rbd_obj_request *obj_request)
2019 struct rbd_img_request *img_request;
2020 struct rbd_device *rbd_dev;
2024 rbd_assert(obj_request->type == OBJ_REQUEST_BIO);
2025 rbd_assert(obj_request_img_data_test(obj_request));
2026 img_request = obj_request->img_request;
2027 rbd_assert(img_request);
2029 rbd_dev = img_request->rbd_dev;
2030 rbd_assert(rbd_dev);
2031 length = (u64)1 << rbd_dev->header.obj_order;
2032 page_count = (u32)calc_pages_for(0, length);
2034 rbd_assert(obj_request->copyup_pages);
2035 ceph_release_page_vector(obj_request->copyup_pages, page_count);
2036 obj_request->copyup_pages = NULL;
2039 * We want the transfer count to reflect the size of the
2040 * original write request. There is no such thing as a
2041 * successful short write, so if the request was successful
2042 * we can just set it to the originally-requested length.
2044 if (!obj_request->result)
2045 obj_request->xferred = obj_request->length;
2047 /* Finish up with the normal image object callback */
2049 rbd_img_obj_callback(obj_request);
2053 rbd_img_obj_parent_read_full_callback(struct rbd_img_request *img_request)
2055 struct rbd_obj_request *orig_request;
2056 struct ceph_osd_request *osd_req;
2057 struct ceph_osd_client *osdc;
2058 struct rbd_device *rbd_dev;
2059 struct page **pages;
2064 rbd_assert(img_request_child_test(img_request));
2066 /* First get what we need from the image request */
2068 pages = img_request->copyup_pages;
2069 rbd_assert(pages != NULL);
2070 img_request->copyup_pages = NULL;
2072 orig_request = img_request->obj_request;
2073 rbd_assert(orig_request != NULL);
2074 rbd_assert(orig_request->type == OBJ_REQUEST_BIO);
2075 result = img_request->result;
2076 obj_size = img_request->length;
2077 xferred = img_request->xferred;
2079 rbd_dev = img_request->rbd_dev;
2080 rbd_assert(rbd_dev);
2081 rbd_assert(obj_size == (u64)1 << rbd_dev->header.obj_order);
2083 rbd_img_request_put(img_request);
2088 /* Allocate the new copyup osd request for the original request */
2091 rbd_assert(!orig_request->osd_req);
2092 osd_req = rbd_osd_req_create_copyup(orig_request);
2095 orig_request->osd_req = osd_req;
2096 orig_request->copyup_pages = pages;
2098 /* Initialize the copyup op */
2100 osd_req_op_cls_init(osd_req, 0, CEPH_OSD_OP_CALL, "rbd", "copyup");
2101 osd_req_op_cls_request_data_pages(osd_req, 0, pages, obj_size, 0,
2104 /* Then the original write request op */
2106 osd_req_op_extent_init(osd_req, 1, CEPH_OSD_OP_WRITE,
2107 orig_request->offset,
2108 orig_request->length, 0, 0);
2109 osd_req_op_extent_osd_data_bio(osd_req, 1, orig_request->bio_list,
2110 orig_request->length);
2112 rbd_osd_req_format_write(orig_request);
2114 /* All set, send it off. */
2116 orig_request->callback = rbd_img_obj_copyup_callback;
2117 osdc = &rbd_dev->rbd_client->client->osdc;
2118 result = rbd_obj_request_submit(osdc, orig_request);
2122 /* Record the error code and complete the request */
2124 orig_request->result = result;
2125 orig_request->xferred = 0;
2126 obj_request_done_set(orig_request);
2127 rbd_obj_request_complete(orig_request);
2131 * Read from the parent image the range of data that covers the
2132 * entire target of the given object request. This is used for
2133 * satisfying a layered image write request when the target of an
2134 * object request from the image request does not exist.
2136 * A page array big enough to hold the returned data is allocated
2137 * and supplied to rbd_img_request_fill() as the "data descriptor."
2138 * When the read completes, this page array will be transferred to
2139 * the original object request for the copyup operation.
2141 * If an error occurs, record it as the result of the original
2142 * object request and mark it done so it gets completed.
2144 static int rbd_img_obj_parent_read_full(struct rbd_obj_request *obj_request)
2146 struct rbd_img_request *img_request = NULL;
2147 struct rbd_img_request *parent_request = NULL;
2148 struct rbd_device *rbd_dev;
2151 struct page **pages = NULL;
2155 rbd_assert(obj_request_img_data_test(obj_request));
2156 rbd_assert(obj_request->type == OBJ_REQUEST_BIO);
2158 img_request = obj_request->img_request;
2159 rbd_assert(img_request != NULL);
2160 rbd_dev = img_request->rbd_dev;
2161 rbd_assert(rbd_dev->parent != NULL);
2164 * First things first. The original osd request is of no
2165 * use to use any more, we'll need a new one that can hold
2166 * the two ops in a copyup request. We'll get that later,
2167 * but for now we can release the old one.
2169 rbd_osd_req_destroy(obj_request->osd_req);
2170 obj_request->osd_req = NULL;
2173 * Determine the byte range covered by the object in the
2174 * child image to which the original request was to be sent.
2176 img_offset = obj_request->img_offset - obj_request->offset;
2177 length = (u64)1 << rbd_dev->header.obj_order;
2180 * There is no defined parent data beyond the parent
2181 * overlap, so limit what we read at that boundary if
2184 if (img_offset + length > rbd_dev->parent_overlap) {
2185 rbd_assert(img_offset < rbd_dev->parent_overlap);
2186 length = rbd_dev->parent_overlap - img_offset;
2190 * Allocate a page array big enough to receive the data read
2193 page_count = (u32)calc_pages_for(0, length);
2194 pages = ceph_alloc_page_vector(page_count, GFP_KERNEL);
2195 if (IS_ERR(pages)) {
2196 result = PTR_ERR(pages);
2202 parent_request = rbd_img_request_create(rbd_dev->parent,
2205 if (!parent_request)
2207 rbd_obj_request_get(obj_request);
2208 parent_request->obj_request = obj_request;
2210 result = rbd_img_request_fill(parent_request, OBJ_REQUEST_PAGES, pages);
2213 parent_request->copyup_pages = pages;
2215 parent_request->callback = rbd_img_obj_parent_read_full_callback;
2216 result = rbd_img_request_submit(parent_request);
2220 parent_request->copyup_pages = NULL;
2221 parent_request->obj_request = NULL;
2222 rbd_obj_request_put(obj_request);
2225 ceph_release_page_vector(pages, page_count);
2227 rbd_img_request_put(parent_request);
2228 obj_request->result = result;
2229 obj_request->xferred = 0;
2230 obj_request_done_set(obj_request);
2235 static void rbd_img_obj_exists_callback(struct rbd_obj_request *obj_request)
2237 struct rbd_obj_request *orig_request;
2240 rbd_assert(!obj_request_img_data_test(obj_request));
2243 * All we need from the object request is the original
2244 * request and the result of the STAT op. Grab those, then
2245 * we're done with the request.
2247 orig_request = obj_request->obj_request;
2248 obj_request->obj_request = NULL;
2249 rbd_assert(orig_request);
2250 rbd_assert(orig_request->img_request);
2252 result = obj_request->result;
2253 obj_request->result = 0;
2255 dout("%s: obj %p for obj %p result %d %llu/%llu\n", __func__,
2256 obj_request, orig_request, result,
2257 obj_request->xferred, obj_request->length);
2258 rbd_obj_request_put(obj_request);
2260 rbd_assert(orig_request);
2261 rbd_assert(orig_request->img_request);
2264 * Our only purpose here is to determine whether the object
2265 * exists, and we don't want to treat the non-existence as
2266 * an error. If something else comes back, transfer the
2267 * error to the original request and complete it now.
2270 obj_request_existence_set(orig_request, true);
2271 } else if (result == -ENOENT) {
2272 obj_request_existence_set(orig_request, false);
2273 } else if (result) {
2274 orig_request->result = result;
2279 * Resubmit the original request now that we have recorded
2280 * whether the target object exists.
2282 orig_request->result = rbd_img_obj_request_submit(orig_request);
2284 if (orig_request->result)
2285 rbd_obj_request_complete(orig_request);
2286 rbd_obj_request_put(orig_request);
2289 static int rbd_img_obj_exists_submit(struct rbd_obj_request *obj_request)
2291 struct rbd_obj_request *stat_request;
2292 struct rbd_device *rbd_dev;
2293 struct ceph_osd_client *osdc;
2294 struct page **pages = NULL;
2300 * The response data for a STAT call consists of:
2307 size = sizeof (__le64) + sizeof (__le32) + sizeof (__le32);
2308 page_count = (u32)calc_pages_for(0, size);
2309 pages = ceph_alloc_page_vector(page_count, GFP_KERNEL);
2311 return PTR_ERR(pages);
2314 stat_request = rbd_obj_request_create(obj_request->object_name, 0, 0,
2319 rbd_obj_request_get(obj_request);
2320 stat_request->obj_request = obj_request;
2321 stat_request->pages = pages;
2322 stat_request->page_count = page_count;
2324 rbd_assert(obj_request->img_request);
2325 rbd_dev = obj_request->img_request->rbd_dev;
2326 stat_request->osd_req = rbd_osd_req_create(rbd_dev, false,
2328 if (!stat_request->osd_req)
2330 stat_request->callback = rbd_img_obj_exists_callback;
2332 osd_req_op_init(stat_request->osd_req, 0, CEPH_OSD_OP_STAT);
2333 osd_req_op_raw_data_in_pages(stat_request->osd_req, 0, pages, size, 0,
2335 rbd_osd_req_format_read(stat_request);
2337 osdc = &rbd_dev->rbd_client->client->osdc;
2338 ret = rbd_obj_request_submit(osdc, stat_request);
2341 rbd_obj_request_put(obj_request);
2346 static int rbd_img_obj_request_submit(struct rbd_obj_request *obj_request)
2348 struct rbd_img_request *img_request;
2349 struct rbd_device *rbd_dev;
2352 rbd_assert(obj_request_img_data_test(obj_request));
2354 img_request = obj_request->img_request;
2355 rbd_assert(img_request);
2356 rbd_dev = img_request->rbd_dev;
2359 * Only writes to layered images need special handling.
2360 * Reads and non-layered writes are simple object requests.
2361 * Layered writes that start beyond the end of the overlap
2362 * with the parent have no parent data, so they too are
2363 * simple object requests. Finally, if the target object is
2364 * known to already exist, its parent data has already been
2365 * copied, so a write to the object can also be handled as a
2366 * simple object request.
2368 if (!img_request_write_test(img_request) ||
2369 !img_request_layered_test(img_request) ||
2370 rbd_dev->parent_overlap <= obj_request->img_offset ||
2371 ((known = obj_request_known_test(obj_request)) &&
2372 obj_request_exists_test(obj_request))) {
2374 struct rbd_device *rbd_dev;
2375 struct ceph_osd_client *osdc;
2377 rbd_dev = obj_request->img_request->rbd_dev;
2378 osdc = &rbd_dev->rbd_client->client->osdc;
2380 return rbd_obj_request_submit(osdc, obj_request);
2384 * It's a layered write. The target object might exist but
2385 * we may not know that yet. If we know it doesn't exist,
2386 * start by reading the data for the full target object from
2387 * the parent so we can use it for a copyup to the target.
2390 return rbd_img_obj_parent_read_full(obj_request);
2392 /* We don't know whether the target exists. Go find out. */
2394 return rbd_img_obj_exists_submit(obj_request);
2397 static int rbd_img_request_submit(struct rbd_img_request *img_request)
2399 struct rbd_obj_request *obj_request;
2400 struct rbd_obj_request *next_obj_request;
2402 dout("%s: img %p\n", __func__, img_request);
2403 for_each_obj_request_safe(img_request, obj_request, next_obj_request) {
2406 ret = rbd_img_obj_request_submit(obj_request);
2414 static void rbd_img_parent_read_callback(struct rbd_img_request *img_request)
2416 struct rbd_obj_request *obj_request;
2417 struct rbd_device *rbd_dev;
2420 rbd_assert(img_request_child_test(img_request));
2422 obj_request = img_request->obj_request;
2423 rbd_assert(obj_request);
2424 rbd_assert(obj_request->img_request);
2426 obj_request->result = img_request->result;
2427 if (obj_request->result)
2431 * We need to zero anything beyond the parent overlap
2432 * boundary. Since rbd_img_obj_request_read_callback()
2433 * will zero anything beyond the end of a short read, an
2434 * easy way to do this is to pretend the data from the
2435 * parent came up short--ending at the overlap boundary.
2437 rbd_assert(obj_request->img_offset < U64_MAX - obj_request->length);
2438 obj_end = obj_request->img_offset + obj_request->length;
2439 rbd_dev = obj_request->img_request->rbd_dev;
2440 if (obj_end > rbd_dev->parent_overlap) {
2443 if (obj_request->img_offset < rbd_dev->parent_overlap)
2444 xferred = rbd_dev->parent_overlap -
2445 obj_request->img_offset;
2447 obj_request->xferred = min(img_request->xferred, xferred);
2449 obj_request->xferred = img_request->xferred;
2452 rbd_img_obj_request_read_callback(obj_request);
2453 rbd_obj_request_complete(obj_request);
2456 static void rbd_img_parent_read(struct rbd_obj_request *obj_request)
2458 struct rbd_device *rbd_dev;
2459 struct rbd_img_request *img_request;
2462 rbd_assert(obj_request_img_data_test(obj_request));
2463 rbd_assert(obj_request->img_request != NULL);
2464 rbd_assert(obj_request->result == (s32) -ENOENT);
2465 rbd_assert(obj_request->type == OBJ_REQUEST_BIO);
2467 rbd_dev = obj_request->img_request->rbd_dev;
2468 rbd_assert(rbd_dev->parent != NULL);
2469 /* rbd_read_finish(obj_request, obj_request->length); */
2470 img_request = rbd_img_request_create(rbd_dev->parent,
2471 obj_request->img_offset,
2472 obj_request->length,
2478 rbd_obj_request_get(obj_request);
2479 img_request->obj_request = obj_request;
2481 result = rbd_img_request_fill(img_request, OBJ_REQUEST_BIO,
2482 obj_request->bio_list);
2486 img_request->callback = rbd_img_parent_read_callback;
2487 result = rbd_img_request_submit(img_request);
2494 rbd_img_request_put(img_request);
2495 obj_request->result = result;
2496 obj_request->xferred = 0;
2497 obj_request_done_set(obj_request);
2500 static int rbd_obj_notify_ack(struct rbd_device *rbd_dev,
2501 u64 ver, u64 notify_id)
2503 struct rbd_obj_request *obj_request;
2504 struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
2507 obj_request = rbd_obj_request_create(rbd_dev->header_name, 0, 0,
2508 OBJ_REQUEST_NODATA);
2513 obj_request->osd_req = rbd_osd_req_create(rbd_dev, false, obj_request);
2514 if (!obj_request->osd_req)
2516 obj_request->callback = rbd_obj_request_put;
2518 osd_req_op_watch_init(obj_request->osd_req, 0, CEPH_OSD_OP_NOTIFY_ACK,
2520 rbd_osd_req_format_read(obj_request);
2522 ret = rbd_obj_request_submit(osdc, obj_request);
2525 rbd_obj_request_put(obj_request);
2530 static void rbd_watch_cb(u64 ver, u64 notify_id, u8 opcode, void *data)
2532 struct rbd_device *rbd_dev = (struct rbd_device *)data;
2538 dout("%s: \"%s\" notify_id %llu opcode %u\n", __func__,
2539 rbd_dev->header_name, (unsigned long long) notify_id,
2540 (unsigned int) opcode);
2541 (void)rbd_dev_refresh(rbd_dev, &hver);
2543 rbd_obj_notify_ack(rbd_dev, hver, notify_id);
2547 * Request sync osd watch/unwatch. The value of "start" determines
2548 * whether a watch request is being initiated or torn down.
2550 static int rbd_dev_header_watch_sync(struct rbd_device *rbd_dev, int start)
2552 struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
2553 struct rbd_obj_request *obj_request;
2556 rbd_assert(start ^ !!rbd_dev->watch_event);
2557 rbd_assert(start ^ !!rbd_dev->watch_request);
2560 ret = ceph_osdc_create_event(osdc, rbd_watch_cb, rbd_dev,
2561 &rbd_dev->watch_event);
2564 rbd_assert(rbd_dev->watch_event != NULL);
2568 obj_request = rbd_obj_request_create(rbd_dev->header_name, 0, 0,
2569 OBJ_REQUEST_NODATA);
2573 obj_request->osd_req = rbd_osd_req_create(rbd_dev, true, obj_request);
2574 if (!obj_request->osd_req)
2578 ceph_osdc_set_request_linger(osdc, obj_request->osd_req);
2580 ceph_osdc_unregister_linger_request(osdc,
2581 rbd_dev->watch_request->osd_req);
2583 osd_req_op_watch_init(obj_request->osd_req, 0, CEPH_OSD_OP_WATCH,
2584 rbd_dev->watch_event->cookie,
2585 rbd_dev->header.obj_version, start);
2586 rbd_osd_req_format_write(obj_request);
2588 ret = rbd_obj_request_submit(osdc, obj_request);
2591 ret = rbd_obj_request_wait(obj_request);
2594 ret = obj_request->result;
2599 * A watch request is set to linger, so the underlying osd
2600 * request won't go away until we unregister it. We retain
2601 * a pointer to the object request during that time (in
2602 * rbd_dev->watch_request), so we'll keep a reference to
2603 * it. We'll drop that reference (below) after we've
2607 rbd_dev->watch_request = obj_request;
2612 /* We have successfully torn down the watch request */
2614 rbd_obj_request_put(rbd_dev->watch_request);
2615 rbd_dev->watch_request = NULL;
2617 /* Cancel the event if we're tearing down, or on error */
2618 ceph_osdc_cancel_event(rbd_dev->watch_event);
2619 rbd_dev->watch_event = NULL;
2621 rbd_obj_request_put(obj_request);
2627 * Synchronous osd object method call. Returns the number of bytes
2628 * returned in the outbound buffer, or a negative error code.
2630 static int rbd_obj_method_sync(struct rbd_device *rbd_dev,
2631 const char *object_name,
2632 const char *class_name,
2633 const char *method_name,
2634 const void *outbound,
2635 size_t outbound_size,
2637 size_t inbound_size,
2640 struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
2641 struct rbd_obj_request *obj_request;
2642 struct page **pages;
2647 * Method calls are ultimately read operations. The result
2648 * should placed into the inbound buffer provided. They
2649 * also supply outbound data--parameters for the object
2650 * method. Currently if this is present it will be a
2653 page_count = (u32)calc_pages_for(0, inbound_size);
2654 pages = ceph_alloc_page_vector(page_count, GFP_KERNEL);
2656 return PTR_ERR(pages);
2659 obj_request = rbd_obj_request_create(object_name, 0, inbound_size,
2664 obj_request->pages = pages;
2665 obj_request->page_count = page_count;
2667 obj_request->osd_req = rbd_osd_req_create(rbd_dev, false, obj_request);
2668 if (!obj_request->osd_req)
2671 osd_req_op_cls_init(obj_request->osd_req, 0, CEPH_OSD_OP_CALL,
2672 class_name, method_name);
2673 if (outbound_size) {
2674 struct ceph_pagelist *pagelist;
2676 pagelist = kmalloc(sizeof (*pagelist), GFP_NOFS);
2680 ceph_pagelist_init(pagelist);
2681 ceph_pagelist_append(pagelist, outbound, outbound_size);
2682 osd_req_op_cls_request_data_pagelist(obj_request->osd_req, 0,
2685 osd_req_op_cls_response_data_pages(obj_request->osd_req, 0,
2686 obj_request->pages, inbound_size,
2688 rbd_osd_req_format_read(obj_request);
2690 ret = rbd_obj_request_submit(osdc, obj_request);
2693 ret = rbd_obj_request_wait(obj_request);
2697 ret = obj_request->result;
2701 rbd_assert(obj_request->xferred < (u64)INT_MAX);
2702 ret = (int)obj_request->xferred;
2703 ceph_copy_from_page_vector(pages, inbound, 0, obj_request->xferred);
2705 *version = obj_request->version;
2708 rbd_obj_request_put(obj_request);
2710 ceph_release_page_vector(pages, page_count);
2715 static void rbd_request_fn(struct request_queue *q)
2716 __releases(q->queue_lock) __acquires(q->queue_lock)
2718 struct rbd_device *rbd_dev = q->queuedata;
2719 bool read_only = rbd_dev->mapping.read_only;
2723 while ((rq = blk_fetch_request(q))) {
2724 bool write_request = rq_data_dir(rq) == WRITE;
2725 struct rbd_img_request *img_request;
2729 /* Ignore any non-FS requests that filter through. */
2731 if (rq->cmd_type != REQ_TYPE_FS) {
2732 dout("%s: non-fs request type %d\n", __func__,
2733 (int) rq->cmd_type);
2734 __blk_end_request_all(rq, 0);
2738 /* Ignore/skip any zero-length requests */
2740 offset = (u64) blk_rq_pos(rq) << SECTOR_SHIFT;
2741 length = (u64) blk_rq_bytes(rq);
2744 dout("%s: zero-length request\n", __func__);
2745 __blk_end_request_all(rq, 0);
2749 spin_unlock_irq(q->queue_lock);
2751 /* Disallow writes to a read-only device */
2753 if (write_request) {
2757 rbd_assert(rbd_dev->spec->snap_id == CEPH_NOSNAP);
2761 * Quit early if the mapped snapshot no longer
2762 * exists. It's still possible the snapshot will
2763 * have disappeared by the time our request arrives
2764 * at the osd, but there's no sense in sending it if
2767 if (!test_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags)) {
2768 dout("request for non-existent snapshot");
2769 rbd_assert(rbd_dev->spec->snap_id != CEPH_NOSNAP);
2775 if (offset && length > U64_MAX - offset + 1) {
2776 rbd_warn(rbd_dev, "bad request range (%llu~%llu)\n",
2778 goto end_request; /* Shouldn't happen */
2782 img_request = rbd_img_request_create(rbd_dev, offset, length,
2783 write_request, false);
2787 img_request->rq = rq;
2789 result = rbd_img_request_fill(img_request, OBJ_REQUEST_BIO,
2792 result = rbd_img_request_submit(img_request);
2794 rbd_img_request_put(img_request);
2796 spin_lock_irq(q->queue_lock);
2798 rbd_warn(rbd_dev, "%s %llx at %llx result %d\n",
2799 write_request ? "write" : "read",
2800 length, offset, result);
2802 __blk_end_request_all(rq, result);
2808 * a queue callback. Makes sure that we don't create a bio that spans across
2809 * multiple osd objects. One exception would be with a single page bios,
2810 * which we handle later at bio_chain_clone_range()
2812 static int rbd_merge_bvec(struct request_queue *q, struct bvec_merge_data *bmd,
2813 struct bio_vec *bvec)
2815 struct rbd_device *rbd_dev = q->queuedata;
2816 sector_t sector_offset;
2817 sector_t sectors_per_obj;
2818 sector_t obj_sector_offset;
2822 * Find how far into its rbd object the partition-relative
2823 * bio start sector is to offset relative to the enclosing
2826 sector_offset = get_start_sect(bmd->bi_bdev) + bmd->bi_sector;
2827 sectors_per_obj = 1 << (rbd_dev->header.obj_order - SECTOR_SHIFT);
2828 obj_sector_offset = sector_offset & (sectors_per_obj - 1);
2831 * Compute the number of bytes from that offset to the end
2832 * of the object. Account for what's already used by the bio.
2834 ret = (int) (sectors_per_obj - obj_sector_offset) << SECTOR_SHIFT;
2835 if (ret > bmd->bi_size)
2836 ret -= bmd->bi_size;
2841 * Don't send back more than was asked for. And if the bio
2842 * was empty, let the whole thing through because: "Note
2843 * that a block device *must* allow a single page to be
2844 * added to an empty bio."
2846 rbd_assert(bvec->bv_len <= PAGE_SIZE);
2847 if (ret > (int) bvec->bv_len || !bmd->bi_size)
2848 ret = (int) bvec->bv_len;
2853 static void rbd_free_disk(struct rbd_device *rbd_dev)
2855 struct gendisk *disk = rbd_dev->disk;
2860 rbd_dev->disk = NULL;
2861 if (disk->flags & GENHD_FL_UP) {
2864 blk_cleanup_queue(disk->queue);
2869 static int rbd_obj_read_sync(struct rbd_device *rbd_dev,
2870 const char *object_name,
2871 u64 offset, u64 length,
2872 void *buf, u64 *version)
2875 struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
2876 struct rbd_obj_request *obj_request;
2877 struct page **pages = NULL;
2882 page_count = (u32) calc_pages_for(offset, length);
2883 pages = ceph_alloc_page_vector(page_count, GFP_KERNEL);
2885 ret = PTR_ERR(pages);
2888 obj_request = rbd_obj_request_create(object_name, offset, length,
2893 obj_request->pages = pages;
2894 obj_request->page_count = page_count;
2896 obj_request->osd_req = rbd_osd_req_create(rbd_dev, false, obj_request);
2897 if (!obj_request->osd_req)
2900 osd_req_op_extent_init(obj_request->osd_req, 0, CEPH_OSD_OP_READ,
2901 offset, length, 0, 0);
2902 osd_req_op_extent_osd_data_pages(obj_request->osd_req, 0,
2904 obj_request->length,
2905 obj_request->offset & ~PAGE_MASK,
2907 rbd_osd_req_format_read(obj_request);
2909 ret = rbd_obj_request_submit(osdc, obj_request);
2912 ret = rbd_obj_request_wait(obj_request);
2916 ret = obj_request->result;
2920 rbd_assert(obj_request->xferred <= (u64) SIZE_MAX);
2921 size = (size_t) obj_request->xferred;
2922 ceph_copy_from_page_vector(pages, buf, 0, size);
2923 rbd_assert(size <= (size_t) INT_MAX);
2926 *version = obj_request->version;
2929 rbd_obj_request_put(obj_request);
2931 ceph_release_page_vector(pages, page_count);
2937 * Read the complete header for the given rbd device.
2939 * Returns a pointer to a dynamically-allocated buffer containing
2940 * the complete and validated header. Caller can pass the address
2941 * of a variable that will be filled in with the version of the
2942 * header object at the time it was read.
2944 * Returns a pointer-coded errno if a failure occurs.
2946 static struct rbd_image_header_ondisk *
2947 rbd_dev_v1_header_read(struct rbd_device *rbd_dev, u64 *version)
2949 struct rbd_image_header_ondisk *ondisk = NULL;
2956 * The complete header will include an array of its 64-bit
2957 * snapshot ids, followed by the names of those snapshots as
2958 * a contiguous block of NUL-terminated strings. Note that
2959 * the number of snapshots could change by the time we read
2960 * it in, in which case we re-read it.
2967 size = sizeof (*ondisk);
2968 size += snap_count * sizeof (struct rbd_image_snap_ondisk);
2970 ondisk = kmalloc(size, GFP_KERNEL);
2972 return ERR_PTR(-ENOMEM);
2974 ret = rbd_obj_read_sync(rbd_dev, rbd_dev->header_name,
2975 0, size, ondisk, version);
2978 if ((size_t)ret < size) {
2980 rbd_warn(rbd_dev, "short header read (want %zd got %d)",
2984 if (!rbd_dev_ondisk_valid(ondisk)) {
2986 rbd_warn(rbd_dev, "invalid header");
2990 names_size = le64_to_cpu(ondisk->snap_names_len);
2991 want_count = snap_count;
2992 snap_count = le32_to_cpu(ondisk->snap_count);
2993 } while (snap_count != want_count);
3000 return ERR_PTR(ret);
3004 * reload the ondisk the header
3006 static int rbd_read_header(struct rbd_device *rbd_dev,
3007 struct rbd_image_header *header)
3009 struct rbd_image_header_ondisk *ondisk;
3013 ondisk = rbd_dev_v1_header_read(rbd_dev, &ver);
3015 return PTR_ERR(ondisk);
3016 ret = rbd_header_from_disk(header, ondisk);
3018 header->obj_version = ver;
3024 static void rbd_remove_all_snaps(struct rbd_device *rbd_dev)
3026 struct rbd_snap *snap;
3027 struct rbd_snap *next;
3029 list_for_each_entry_safe(snap, next, &rbd_dev->snaps, node) {
3030 list_del(&snap->node);
3031 rbd_snap_destroy(snap);
3035 static void rbd_update_mapping_size(struct rbd_device *rbd_dev)
3039 if (rbd_dev->spec->snap_id != CEPH_NOSNAP)
3042 size = (sector_t) rbd_dev->header.image_size / SECTOR_SIZE;
3043 dout("setting size to %llu sectors", (unsigned long long) size);
3044 rbd_dev->mapping.size = (u64) size;
3045 set_capacity(rbd_dev->disk, size);
3049 * only read the first part of the ondisk header, without the snaps info
3051 static int rbd_dev_v1_refresh(struct rbd_device *rbd_dev, u64 *hver)
3054 struct rbd_image_header h;
3056 ret = rbd_read_header(rbd_dev, &h);
3060 down_write(&rbd_dev->header_rwsem);
3062 /* Update image size, and check for resize of mapped image */
3063 rbd_dev->header.image_size = h.image_size;
3064 rbd_update_mapping_size(rbd_dev);
3066 /* rbd_dev->header.object_prefix shouldn't change */
3067 kfree(rbd_dev->header.snap_sizes);
3068 kfree(rbd_dev->header.snap_names);
3069 /* osd requests may still refer to snapc */
3070 rbd_snap_context_put(rbd_dev->header.snapc);
3073 *hver = h.obj_version;
3074 rbd_dev->header.obj_version = h.obj_version;
3075 rbd_dev->header.image_size = h.image_size;
3076 rbd_dev->header.snapc = h.snapc;
3077 rbd_dev->header.snap_names = h.snap_names;
3078 rbd_dev->header.snap_sizes = h.snap_sizes;
3079 /* Free the extra copy of the object prefix */
3080 if (strcmp(rbd_dev->header.object_prefix, h.object_prefix))
3081 rbd_warn(rbd_dev, "object prefix changed (ignoring)");
3082 kfree(h.object_prefix);
3084 ret = rbd_dev_snaps_update(rbd_dev);
3086 up_write(&rbd_dev->header_rwsem);
3091 static int rbd_dev_refresh(struct rbd_device *rbd_dev, u64 *hver)
3095 rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
3096 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
3097 if (rbd_dev->image_format == 1)
3098 ret = rbd_dev_v1_refresh(rbd_dev, hver);
3100 ret = rbd_dev_v2_refresh(rbd_dev, hver);
3101 mutex_unlock(&ctl_mutex);
3102 revalidate_disk(rbd_dev->disk);
3104 rbd_warn(rbd_dev, "got notification but failed to "
3105 " update snaps: %d\n", ret);
3110 static int rbd_init_disk(struct rbd_device *rbd_dev)
3112 struct gendisk *disk;
3113 struct request_queue *q;
3116 /* create gendisk info */
3117 disk = alloc_disk(RBD_MINORS_PER_MAJOR);
3121 snprintf(disk->disk_name, sizeof(disk->disk_name), RBD_DRV_NAME "%d",
3123 disk->major = rbd_dev->major;
3124 disk->first_minor = 0;
3125 disk->fops = &rbd_bd_ops;
3126 disk->private_data = rbd_dev;
3128 q = blk_init_queue(rbd_request_fn, &rbd_dev->lock);
3132 /* We use the default size, but let's be explicit about it. */
3133 blk_queue_physical_block_size(q, SECTOR_SIZE);
3135 /* set io sizes to object size */
3136 segment_size = rbd_obj_bytes(&rbd_dev->header);
3137 blk_queue_max_hw_sectors(q, segment_size / SECTOR_SIZE);
3138 blk_queue_max_segment_size(q, segment_size);
3139 blk_queue_io_min(q, segment_size);
3140 blk_queue_io_opt(q, segment_size);
3142 blk_queue_merge_bvec(q, rbd_merge_bvec);
3145 q->queuedata = rbd_dev;
3147 rbd_dev->disk = disk;
3149 set_capacity(rbd_dev->disk, rbd_dev->mapping.size / SECTOR_SIZE);
3162 static struct rbd_device *dev_to_rbd_dev(struct device *dev)
3164 return container_of(dev, struct rbd_device, dev);
3167 static ssize_t rbd_size_show(struct device *dev,
3168 struct device_attribute *attr, char *buf)
3170 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3173 down_read(&rbd_dev->header_rwsem);
3174 size = get_capacity(rbd_dev->disk);
3175 up_read(&rbd_dev->header_rwsem);
3177 return sprintf(buf, "%llu\n", (unsigned long long) size * SECTOR_SIZE);
3181 * Note this shows the features for whatever's mapped, which is not
3182 * necessarily the base image.
3184 static ssize_t rbd_features_show(struct device *dev,
3185 struct device_attribute *attr, char *buf)
3187 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3189 return sprintf(buf, "0x%016llx\n",
3190 (unsigned long long) rbd_dev->mapping.features);
3193 static ssize_t rbd_major_show(struct device *dev,
3194 struct device_attribute *attr, char *buf)
3196 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3198 return sprintf(buf, "%d\n", rbd_dev->major);
3201 static ssize_t rbd_client_id_show(struct device *dev,
3202 struct device_attribute *attr, char *buf)
3204 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3206 return sprintf(buf, "client%lld\n",
3207 ceph_client_id(rbd_dev->rbd_client->client));
3210 static ssize_t rbd_pool_show(struct device *dev,
3211 struct device_attribute *attr, char *buf)
3213 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3215 return sprintf(buf, "%s\n", rbd_dev->spec->pool_name);
3218 static ssize_t rbd_pool_id_show(struct device *dev,
3219 struct device_attribute *attr, char *buf)
3221 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3223 return sprintf(buf, "%llu\n",
3224 (unsigned long long) rbd_dev->spec->pool_id);
3227 static ssize_t rbd_name_show(struct device *dev,
3228 struct device_attribute *attr, char *buf)
3230 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3232 if (rbd_dev->spec->image_name)
3233 return sprintf(buf, "%s\n", rbd_dev->spec->image_name);
3235 return sprintf(buf, "(unknown)\n");
3238 static ssize_t rbd_image_id_show(struct device *dev,
3239 struct device_attribute *attr, char *buf)
3241 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3243 return sprintf(buf, "%s\n", rbd_dev->spec->image_id);
3247 * Shows the name of the currently-mapped snapshot (or
3248 * RBD_SNAP_HEAD_NAME for the base image).
3250 static ssize_t rbd_snap_show(struct device *dev,
3251 struct device_attribute *attr,
3254 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3256 return sprintf(buf, "%s\n", rbd_dev->spec->snap_name);
3260 * For an rbd v2 image, shows the pool id, image id, and snapshot id
3261 * for the parent image. If there is no parent, simply shows
3262 * "(no parent image)".
3264 static ssize_t rbd_parent_show(struct device *dev,
3265 struct device_attribute *attr,
3268 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3269 struct rbd_spec *spec = rbd_dev->parent_spec;
3274 return sprintf(buf, "(no parent image)\n");
3276 count = sprintf(bufp, "pool_id %llu\npool_name %s\n",
3277 (unsigned long long) spec->pool_id, spec->pool_name);
3282 count = sprintf(bufp, "image_id %s\nimage_name %s\n", spec->image_id,
3283 spec->image_name ? spec->image_name : "(unknown)");
3288 count = sprintf(bufp, "snap_id %llu\nsnap_name %s\n",
3289 (unsigned long long) spec->snap_id, spec->snap_name);
3294 count = sprintf(bufp, "overlap %llu\n", rbd_dev->parent_overlap);
3299 return (ssize_t) (bufp - buf);
3302 static ssize_t rbd_image_refresh(struct device *dev,
3303 struct device_attribute *attr,
3307 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3310 ret = rbd_dev_refresh(rbd_dev, NULL);
3312 return ret < 0 ? ret : size;
3315 static DEVICE_ATTR(size, S_IRUGO, rbd_size_show, NULL);
3316 static DEVICE_ATTR(features, S_IRUGO, rbd_features_show, NULL);
3317 static DEVICE_ATTR(major, S_IRUGO, rbd_major_show, NULL);
3318 static DEVICE_ATTR(client_id, S_IRUGO, rbd_client_id_show, NULL);
3319 static DEVICE_ATTR(pool, S_IRUGO, rbd_pool_show, NULL);
3320 static DEVICE_ATTR(pool_id, S_IRUGO, rbd_pool_id_show, NULL);
3321 static DEVICE_ATTR(name, S_IRUGO, rbd_name_show, NULL);
3322 static DEVICE_ATTR(image_id, S_IRUGO, rbd_image_id_show, NULL);
3323 static DEVICE_ATTR(refresh, S_IWUSR, NULL, rbd_image_refresh);
3324 static DEVICE_ATTR(current_snap, S_IRUGO, rbd_snap_show, NULL);
3325 static DEVICE_ATTR(parent, S_IRUGO, rbd_parent_show, NULL);
3327 static struct attribute *rbd_attrs[] = {
3328 &dev_attr_size.attr,
3329 &dev_attr_features.attr,
3330 &dev_attr_major.attr,
3331 &dev_attr_client_id.attr,
3332 &dev_attr_pool.attr,
3333 &dev_attr_pool_id.attr,
3334 &dev_attr_name.attr,
3335 &dev_attr_image_id.attr,
3336 &dev_attr_current_snap.attr,
3337 &dev_attr_parent.attr,
3338 &dev_attr_refresh.attr,
3342 static struct attribute_group rbd_attr_group = {
3346 static const struct attribute_group *rbd_attr_groups[] = {
3351 static void rbd_sysfs_dev_release(struct device *dev)
3355 static struct device_type rbd_device_type = {
3357 .groups = rbd_attr_groups,
3358 .release = rbd_sysfs_dev_release,
3361 static struct rbd_spec *rbd_spec_get(struct rbd_spec *spec)
3363 kref_get(&spec->kref);
3368 static void rbd_spec_free(struct kref *kref);
3369 static void rbd_spec_put(struct rbd_spec *spec)
3372 kref_put(&spec->kref, rbd_spec_free);
3375 static struct rbd_spec *rbd_spec_alloc(void)
3377 struct rbd_spec *spec;
3379 spec = kzalloc(sizeof (*spec), GFP_KERNEL);
3382 kref_init(&spec->kref);
3387 static void rbd_spec_free(struct kref *kref)
3389 struct rbd_spec *spec = container_of(kref, struct rbd_spec, kref);
3391 kfree(spec->pool_name);
3392 kfree(spec->image_id);
3393 kfree(spec->image_name);
3394 kfree(spec->snap_name);
3398 static struct rbd_device *rbd_dev_create(struct rbd_client *rbdc,
3399 struct rbd_spec *spec)
3401 struct rbd_device *rbd_dev;
3403 rbd_dev = kzalloc(sizeof (*rbd_dev), GFP_KERNEL);
3407 spin_lock_init(&rbd_dev->lock);
3409 INIT_LIST_HEAD(&rbd_dev->node);
3410 INIT_LIST_HEAD(&rbd_dev->snaps);
3411 init_rwsem(&rbd_dev->header_rwsem);
3413 rbd_dev->spec = spec;
3414 rbd_dev->rbd_client = rbdc;
3416 /* Initialize the layout used for all rbd requests */
3418 rbd_dev->layout.fl_stripe_unit = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER);
3419 rbd_dev->layout.fl_stripe_count = cpu_to_le32(1);
3420 rbd_dev->layout.fl_object_size = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER);
3421 rbd_dev->layout.fl_pg_pool = cpu_to_le32((u32) spec->pool_id);
3426 static void rbd_dev_destroy(struct rbd_device *rbd_dev)
3428 rbd_put_client(rbd_dev->rbd_client);
3429 rbd_spec_put(rbd_dev->spec);
3433 static void rbd_snap_destroy(struct rbd_snap *snap)
3439 static struct rbd_snap *rbd_snap_create(struct rbd_device *rbd_dev,
3440 const char *snap_name,
3441 u64 snap_id, u64 snap_size,
3444 struct rbd_snap *snap;
3446 snap = kzalloc(sizeof (*snap), GFP_KERNEL);
3448 return ERR_PTR(-ENOMEM);
3450 snap->name = snap_name;
3452 snap->size = snap_size;
3453 snap->features = snap_features;
3459 * Returns a dynamically-allocated snapshot name if successful, or a
3460 * pointer-coded error otherwise.
3462 static char *rbd_dev_v1_snap_info(struct rbd_device *rbd_dev, u32 which,
3463 u64 *snap_size, u64 *snap_features)
3468 rbd_assert(which < rbd_dev->header.snapc->num_snaps);
3470 /* Skip over names until we find the one we are looking for */
3472 snap_name = rbd_dev->header.snap_names;
3473 for (i = 0; i < which; i++)
3474 snap_name += strlen(snap_name) + 1;
3476 snap_name = kstrdup(snap_name, GFP_KERNEL);
3478 return ERR_PTR(-ENOMEM);
3480 *snap_size = rbd_dev->header.snap_sizes[which];
3481 *snap_features = 0; /* No features for v1 */
3487 * Get the size and object order for an image snapshot, or if
3488 * snap_id is CEPH_NOSNAP, gets this information for the base
3491 static int _rbd_dev_v2_snap_size(struct rbd_device *rbd_dev, u64 snap_id,
3492 u8 *order, u64 *snap_size)
3494 __le64 snapid = cpu_to_le64(snap_id);
3499 } __attribute__ ((packed)) size_buf = { 0 };
3501 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
3503 &snapid, sizeof (snapid),
3504 &size_buf, sizeof (size_buf), NULL);
3505 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
3508 if (ret < sizeof (size_buf))
3512 *order = size_buf.order;
3513 *snap_size = le64_to_cpu(size_buf.size);
3515 dout(" snap_id 0x%016llx order = %u, snap_size = %llu\n",
3516 (unsigned long long)snap_id, (unsigned int)*order,
3517 (unsigned long long)*snap_size);
3522 static int rbd_dev_v2_image_size(struct rbd_device *rbd_dev)
3524 return _rbd_dev_v2_snap_size(rbd_dev, CEPH_NOSNAP,
3525 &rbd_dev->header.obj_order,
3526 &rbd_dev->header.image_size);
3529 static int rbd_dev_v2_object_prefix(struct rbd_device *rbd_dev)
3535 reply_buf = kzalloc(RBD_OBJ_PREFIX_LEN_MAX, GFP_KERNEL);
3539 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
3540 "rbd", "get_object_prefix", NULL, 0,
3541 reply_buf, RBD_OBJ_PREFIX_LEN_MAX, NULL);
3542 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
3547 rbd_dev->header.object_prefix = ceph_extract_encoded_string(&p,
3548 p + ret, NULL, GFP_NOIO);
3551 if (IS_ERR(rbd_dev->header.object_prefix)) {
3552 ret = PTR_ERR(rbd_dev->header.object_prefix);
3553 rbd_dev->header.object_prefix = NULL;
3555 dout(" object_prefix = %s\n", rbd_dev->header.object_prefix);
3563 static int _rbd_dev_v2_snap_features(struct rbd_device *rbd_dev, u64 snap_id,
3566 __le64 snapid = cpu_to_le64(snap_id);
3570 } __attribute__ ((packed)) features_buf = { 0 };
3574 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
3575 "rbd", "get_features",
3576 &snapid, sizeof (snapid),
3577 &features_buf, sizeof (features_buf), NULL);
3578 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
3581 if (ret < sizeof (features_buf))
3584 incompat = le64_to_cpu(features_buf.incompat);
3585 if (incompat & ~RBD_FEATURES_SUPPORTED)
3588 *snap_features = le64_to_cpu(features_buf.features);
3590 dout(" snap_id 0x%016llx features = 0x%016llx incompat = 0x%016llx\n",
3591 (unsigned long long)snap_id,
3592 (unsigned long long)*snap_features,
3593 (unsigned long long)le64_to_cpu(features_buf.incompat));
3598 static int rbd_dev_v2_features(struct rbd_device *rbd_dev)
3600 return _rbd_dev_v2_snap_features(rbd_dev, CEPH_NOSNAP,
3601 &rbd_dev->header.features);
3604 static int rbd_dev_v2_parent_info(struct rbd_device *rbd_dev)
3606 struct rbd_spec *parent_spec;
3608 void *reply_buf = NULL;
3616 parent_spec = rbd_spec_alloc();
3620 size = sizeof (__le64) + /* pool_id */
3621 sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX + /* image_id */
3622 sizeof (__le64) + /* snap_id */
3623 sizeof (__le64); /* overlap */
3624 reply_buf = kmalloc(size, GFP_KERNEL);
3630 snapid = cpu_to_le64(CEPH_NOSNAP);
3631 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
3632 "rbd", "get_parent",
3633 &snapid, sizeof (snapid),
3634 reply_buf, size, NULL);
3635 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
3640 end = reply_buf + ret;
3642 ceph_decode_64_safe(&p, end, parent_spec->pool_id, out_err);
3643 if (parent_spec->pool_id == CEPH_NOPOOL)
3644 goto out; /* No parent? No problem. */
3646 /* The ceph file layout needs to fit pool id in 32 bits */
3649 if (parent_spec->pool_id > (u64)U32_MAX) {
3650 rbd_warn(NULL, "parent pool id too large (%llu > %u)\n",
3651 (unsigned long long)parent_spec->pool_id, U32_MAX);
3655 image_id = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL);
3656 if (IS_ERR(image_id)) {
3657 ret = PTR_ERR(image_id);
3660 parent_spec->image_id = image_id;
3661 ceph_decode_64_safe(&p, end, parent_spec->snap_id, out_err);
3662 ceph_decode_64_safe(&p, end, overlap, out_err);
3664 rbd_dev->parent_overlap = overlap;
3665 rbd_dev->parent_spec = parent_spec;
3666 parent_spec = NULL; /* rbd_dev now owns this */
3671 rbd_spec_put(parent_spec);
3676 static int rbd_dev_v2_striping_info(struct rbd_device *rbd_dev)
3680 __le64 stripe_count;
3681 } __attribute__ ((packed)) striping_info_buf = { 0 };
3682 size_t size = sizeof (striping_info_buf);
3689 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
3690 "rbd", "get_stripe_unit_count", NULL, 0,
3691 (char *)&striping_info_buf, size, NULL);
3692 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
3699 * We don't actually support the "fancy striping" feature
3700 * (STRIPINGV2) yet, but if the striping sizes are the
3701 * defaults the behavior is the same as before. So find
3702 * out, and only fail if the image has non-default values.
3705 obj_size = (u64)1 << rbd_dev->header.obj_order;
3706 p = &striping_info_buf;
3707 stripe_unit = ceph_decode_64(&p);
3708 if (stripe_unit != obj_size) {
3709 rbd_warn(rbd_dev, "unsupported stripe unit "
3710 "(got %llu want %llu)",
3711 stripe_unit, obj_size);
3714 stripe_count = ceph_decode_64(&p);
3715 if (stripe_count != 1) {
3716 rbd_warn(rbd_dev, "unsupported stripe count "
3717 "(got %llu want 1)", stripe_count);
3720 rbd_dev->header.stripe_unit = stripe_unit;
3721 rbd_dev->header.stripe_count = stripe_count;
3726 static char *rbd_dev_image_name(struct rbd_device *rbd_dev)
3728 size_t image_id_size;
3733 void *reply_buf = NULL;
3735 char *image_name = NULL;
3738 rbd_assert(!rbd_dev->spec->image_name);
3740 len = strlen(rbd_dev->spec->image_id);
3741 image_id_size = sizeof (__le32) + len;
3742 image_id = kmalloc(image_id_size, GFP_KERNEL);
3747 end = image_id + image_id_size;
3748 ceph_encode_string(&p, end, rbd_dev->spec->image_id, (u32)len);
3750 size = sizeof (__le32) + RBD_IMAGE_NAME_LEN_MAX;
3751 reply_buf = kmalloc(size, GFP_KERNEL);
3755 ret = rbd_obj_method_sync(rbd_dev, RBD_DIRECTORY,
3756 "rbd", "dir_get_name",
3757 image_id, image_id_size,
3758 reply_buf, size, NULL);
3762 end = reply_buf + ret;
3764 image_name = ceph_extract_encoded_string(&p, end, &len, GFP_KERNEL);
3765 if (IS_ERR(image_name))
3768 dout("%s: name is %s len is %zd\n", __func__, image_name, len);
3777 * When a parent image gets probed, we only have the pool, image,
3778 * and snapshot ids but not the names of any of them. This call
3779 * is made later to fill in those names. It has to be done after
3780 * rbd_dev_snaps_update() has completed because some of the
3781 * information (in particular, snapshot name) is not available
3784 * When an image being mapped (not a parent) is probed, we have the
3785 * pool name and pool id, image name and image id, and the snapshot
3786 * name. The only thing we're missing is the snapshot id.
3788 static int rbd_dev_probe_update_spec(struct rbd_device *rbd_dev)
3790 struct ceph_osd_client *osdc;
3792 void *reply_buf = NULL;
3796 * An image being mapped will have the pool name (etc.), but
3797 * we need to look up the snapshot id.
3799 if (rbd_dev->spec->pool_name) {
3800 if (strcmp(rbd_dev->spec->snap_name, RBD_SNAP_HEAD_NAME)) {
3801 struct rbd_snap *snap;
3803 snap = snap_by_name(rbd_dev, rbd_dev->spec->snap_name);
3806 rbd_dev->spec->snap_id = snap->id;
3808 rbd_dev->spec->snap_id = CEPH_NOSNAP;
3814 /* Look up the pool name */
3816 osdc = &rbd_dev->rbd_client->client->osdc;
3817 name = ceph_pg_pool_name_by_id(osdc->osdmap, rbd_dev->spec->pool_id);
3819 rbd_warn(rbd_dev, "there is no pool with id %llu",
3820 rbd_dev->spec->pool_id); /* Really a BUG() */
3824 rbd_dev->spec->pool_name = kstrdup(name, GFP_KERNEL);
3825 if (!rbd_dev->spec->pool_name)
3828 /* Fetch the image name; tolerate failure here */
3830 name = rbd_dev_image_name(rbd_dev);
3832 rbd_dev->spec->image_name = (char *)name;
3834 rbd_warn(rbd_dev, "unable to get image name");
3836 /* Look up the snapshot name. */
3838 name = rbd_snap_name(rbd_dev, rbd_dev->spec->snap_id);
3840 rbd_warn(rbd_dev, "no snapshot with id %llu",
3841 rbd_dev->spec->snap_id); /* Really a BUG() */
3845 rbd_dev->spec->snap_name = kstrdup(name, GFP_KERNEL);
3846 if(!rbd_dev->spec->snap_name)
3852 kfree(rbd_dev->spec->pool_name);
3853 rbd_dev->spec->pool_name = NULL;
3858 static int rbd_dev_v2_snap_context(struct rbd_device *rbd_dev, u64 *ver)
3867 struct ceph_snap_context *snapc;
3871 * We'll need room for the seq value (maximum snapshot id),
3872 * snapshot count, and array of that many snapshot ids.
3873 * For now we have a fixed upper limit on the number we're
3874 * prepared to receive.
3876 size = sizeof (__le64) + sizeof (__le32) +
3877 RBD_MAX_SNAP_COUNT * sizeof (__le64);
3878 reply_buf = kzalloc(size, GFP_KERNEL);
3882 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
3883 "rbd", "get_snapcontext", NULL, 0,
3884 reply_buf, size, ver);
3885 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
3890 end = reply_buf + ret;
3892 ceph_decode_64_safe(&p, end, seq, out);
3893 ceph_decode_32_safe(&p, end, snap_count, out);
3896 * Make sure the reported number of snapshot ids wouldn't go
3897 * beyond the end of our buffer. But before checking that,
3898 * make sure the computed size of the snapshot context we
3899 * allocate is representable in a size_t.
3901 if (snap_count > (SIZE_MAX - sizeof (struct ceph_snap_context))
3906 if (!ceph_has_room(&p, end, snap_count * sizeof (__le64)))
3910 snapc = rbd_snap_context_create(snap_count);
3916 for (i = 0; i < snap_count; i++)
3917 snapc->snaps[i] = ceph_decode_64(&p);
3919 rbd_dev->header.snapc = snapc;
3921 dout(" snap context seq = %llu, snap_count = %u\n",
3922 (unsigned long long)seq, (unsigned int)snap_count);
3929 static char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev, u32 which)
3939 size = sizeof (__le32) + RBD_MAX_SNAP_NAME_LEN;
3940 reply_buf = kmalloc(size, GFP_KERNEL);
3942 return ERR_PTR(-ENOMEM);
3944 rbd_assert(which < rbd_dev->header.snapc->num_snaps);
3945 snap_id = cpu_to_le64(rbd_dev->header.snapc->snaps[which]);
3946 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
3947 "rbd", "get_snapshot_name",
3948 &snap_id, sizeof (snap_id),
3949 reply_buf, size, NULL);
3950 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
3952 snap_name = ERR_PTR(ret);
3957 end = reply_buf + ret;
3958 snap_name = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL);
3959 if (IS_ERR(snap_name))
3962 dout(" snap_id 0x%016llx snap_name = %s\n",
3963 (unsigned long long)le64_to_cpu(snap_id), snap_name);
3970 static char *rbd_dev_v2_snap_info(struct rbd_device *rbd_dev, u32 which,
3971 u64 *snap_size, u64 *snap_features)
3979 rbd_assert(which < rbd_dev->header.snapc->num_snaps);
3980 snap_id = rbd_dev->header.snapc->snaps[which];
3981 ret = _rbd_dev_v2_snap_size(rbd_dev, snap_id, NULL, &size);
3985 ret = _rbd_dev_v2_snap_features(rbd_dev, snap_id, &features);
3989 snap_name = rbd_dev_v2_snap_name(rbd_dev, which);
3990 if (!IS_ERR(snap_name)) {
3992 *snap_features = features;
3997 return ERR_PTR(ret);
4000 static char *rbd_dev_snap_info(struct rbd_device *rbd_dev, u32 which,
4001 u64 *snap_size, u64 *snap_features)
4003 if (rbd_dev->image_format == 1)
4004 return rbd_dev_v1_snap_info(rbd_dev, which,
4005 snap_size, snap_features);
4006 if (rbd_dev->image_format == 2)
4007 return rbd_dev_v2_snap_info(rbd_dev, which,
4008 snap_size, snap_features);
4009 return ERR_PTR(-EINVAL);
4012 static int rbd_dev_v2_refresh(struct rbd_device *rbd_dev, u64 *hver)
4017 down_write(&rbd_dev->header_rwsem);
4019 /* Grab old order first, to see if it changes */
4021 obj_order = rbd_dev->header.obj_order,
4022 ret = rbd_dev_v2_image_size(rbd_dev);
4025 if (rbd_dev->header.obj_order != obj_order) {
4029 rbd_update_mapping_size(rbd_dev);
4031 ret = rbd_dev_v2_snap_context(rbd_dev, hver);
4032 dout("rbd_dev_v2_snap_context returned %d\n", ret);
4035 ret = rbd_dev_snaps_update(rbd_dev);
4036 dout("rbd_dev_snaps_update returned %d\n", ret);
4040 up_write(&rbd_dev->header_rwsem);
4046 * Scan the rbd device's current snapshot list and compare it to the
4047 * newly-received snapshot context. Remove any existing snapshots
4048 * not present in the new snapshot context. Add a new snapshot for
4049 * any snaphots in the snapshot context not in the current list.
4050 * And verify there are no changes to snapshots we already know
4053 * Assumes the snapshots in the snapshot context are sorted by
4054 * snapshot id, highest id first. (Snapshots in the rbd_dev's list
4055 * are also maintained in that order.)
4057 * Note that any error occurs while updating the snapshot list
4058 * aborts the update, and the entire list is cleared. The snapshot
4059 * list becomes inconsistent at that point anyway, so it might as
4062 static int rbd_dev_snaps_update(struct rbd_device *rbd_dev)
4064 struct ceph_snap_context *snapc = rbd_dev->header.snapc;
4065 const u32 snap_count = snapc->num_snaps;
4066 struct list_head *head = &rbd_dev->snaps;
4067 struct list_head *links = head->next;
4071 dout("%s: snap count is %u\n", __func__, (unsigned int)snap_count);
4072 while (index < snap_count || links != head) {
4074 struct rbd_snap *snap;
4077 u64 snap_features = 0;
4079 snap_id = index < snap_count ? snapc->snaps[index]
4081 snap = links != head ? list_entry(links, struct rbd_snap, node)
4083 rbd_assert(!snap || snap->id != CEPH_NOSNAP);
4085 if (snap_id == CEPH_NOSNAP || (snap && snap->id > snap_id)) {
4086 struct list_head *next = links->next;
4089 * A previously-existing snapshot is not in
4090 * the new snap context.
4092 * If the now-missing snapshot is the one
4093 * the image represents, clear its existence
4094 * flag so we can avoid sending any more
4097 if (rbd_dev->spec->snap_id == snap->id)
4098 clear_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags);
4099 dout("removing %ssnap id %llu\n",
4100 rbd_dev->spec->snap_id == snap->id ?
4102 (unsigned long long)snap->id);
4104 list_del(&snap->node);
4105 rbd_snap_destroy(snap);
4107 /* Done with this list entry; advance */
4113 snap_name = rbd_dev_snap_info(rbd_dev, index,
4114 &snap_size, &snap_features);
4115 if (IS_ERR(snap_name)) {
4116 ret = PTR_ERR(snap_name);
4117 dout("failed to get snap info, error %d\n", ret);
4121 dout("entry %u: snap_id = %llu\n", (unsigned int)snap_count,
4122 (unsigned long long)snap_id);
4123 if (!snap || (snap_id != CEPH_NOSNAP && snap->id < snap_id)) {
4124 struct rbd_snap *new_snap;
4126 /* We haven't seen this snapshot before */
4128 new_snap = rbd_snap_create(rbd_dev, snap_name,
4129 snap_id, snap_size, snap_features);
4130 if (IS_ERR(new_snap)) {
4131 ret = PTR_ERR(new_snap);
4132 dout(" failed to add dev, error %d\n", ret);
4136 /* New goes before existing, or at end of list */
4138 dout(" added dev%s\n", snap ? "" : " at end\n");
4140 list_add_tail(&new_snap->node, &snap->node);
4142 list_add_tail(&new_snap->node, head);
4144 /* Already have this one */
4146 dout(" already present\n");
4148 rbd_assert(snap->size == snap_size);
4149 rbd_assert(!strcmp(snap->name, snap_name));
4150 rbd_assert(snap->features == snap_features);
4152 /* Done with this list entry; advance */
4154 links = links->next;
4157 /* Advance to the next entry in the snapshot context */
4161 dout("%s: done\n", __func__);
4165 rbd_remove_all_snaps(rbd_dev);
4170 static int rbd_bus_add_dev(struct rbd_device *rbd_dev)
4175 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
4177 dev = &rbd_dev->dev;
4178 dev->bus = &rbd_bus_type;
4179 dev->type = &rbd_device_type;
4180 dev->parent = &rbd_root_dev;
4181 dev->release = rbd_dev_release;
4182 dev_set_name(dev, "%d", rbd_dev->dev_id);
4183 ret = device_register(dev);
4185 mutex_unlock(&ctl_mutex);
4190 static void rbd_bus_del_dev(struct rbd_device *rbd_dev)
4192 device_unregister(&rbd_dev->dev);
4195 static atomic64_t rbd_dev_id_max = ATOMIC64_INIT(0);
4198 * Get a unique rbd identifier for the given new rbd_dev, and add
4199 * the rbd_dev to the global list. The minimum rbd id is 1.
4201 static void rbd_dev_id_get(struct rbd_device *rbd_dev)
4203 rbd_dev->dev_id = atomic64_inc_return(&rbd_dev_id_max);
4205 spin_lock(&rbd_dev_list_lock);
4206 list_add_tail(&rbd_dev->node, &rbd_dev_list);
4207 spin_unlock(&rbd_dev_list_lock);
4208 dout("rbd_dev %p given dev id %llu\n", rbd_dev,
4209 (unsigned long long) rbd_dev->dev_id);
4213 * Remove an rbd_dev from the global list, and record that its
4214 * identifier is no longer in use.
4216 static void rbd_dev_id_put(struct rbd_device *rbd_dev)
4218 struct list_head *tmp;
4219 int rbd_id = rbd_dev->dev_id;
4222 rbd_assert(rbd_id > 0);
4224 dout("rbd_dev %p released dev id %llu\n", rbd_dev,
4225 (unsigned long long) rbd_dev->dev_id);
4226 spin_lock(&rbd_dev_list_lock);
4227 list_del_init(&rbd_dev->node);
4230 * If the id being "put" is not the current maximum, there
4231 * is nothing special we need to do.
4233 if (rbd_id != atomic64_read(&rbd_dev_id_max)) {
4234 spin_unlock(&rbd_dev_list_lock);
4239 * We need to update the current maximum id. Search the
4240 * list to find out what it is. We're more likely to find
4241 * the maximum at the end, so search the list backward.
4244 list_for_each_prev(tmp, &rbd_dev_list) {
4245 struct rbd_device *rbd_dev;
4247 rbd_dev = list_entry(tmp, struct rbd_device, node);
4248 if (rbd_dev->dev_id > max_id)
4249 max_id = rbd_dev->dev_id;
4251 spin_unlock(&rbd_dev_list_lock);
4254 * The max id could have been updated by rbd_dev_id_get(), in
4255 * which case it now accurately reflects the new maximum.
4256 * Be careful not to overwrite the maximum value in that
4259 atomic64_cmpxchg(&rbd_dev_id_max, rbd_id, max_id);
4260 dout(" max dev id has been reset\n");
4264 * Skips over white space at *buf, and updates *buf to point to the
4265 * first found non-space character (if any). Returns the length of
4266 * the token (string of non-white space characters) found. Note
4267 * that *buf must be terminated with '\0'.
4269 static inline size_t next_token(const char **buf)
4272 * These are the characters that produce nonzero for
4273 * isspace() in the "C" and "POSIX" locales.
4275 const char *spaces = " \f\n\r\t\v";
4277 *buf += strspn(*buf, spaces); /* Find start of token */
4279 return strcspn(*buf, spaces); /* Return token length */
4283 * Finds the next token in *buf, and if the provided token buffer is
4284 * big enough, copies the found token into it. The result, if
4285 * copied, is guaranteed to be terminated with '\0'. Note that *buf
4286 * must be terminated with '\0' on entry.
4288 * Returns the length of the token found (not including the '\0').
4289 * Return value will be 0 if no token is found, and it will be >=
4290 * token_size if the token would not fit.
4292 * The *buf pointer will be updated to point beyond the end of the
4293 * found token. Note that this occurs even if the token buffer is
4294 * too small to hold it.
4296 static inline size_t copy_token(const char **buf,
4302 len = next_token(buf);
4303 if (len < token_size) {
4304 memcpy(token, *buf, len);
4305 *(token + len) = '\0';
4313 * Finds the next token in *buf, dynamically allocates a buffer big
4314 * enough to hold a copy of it, and copies the token into the new
4315 * buffer. The copy is guaranteed to be terminated with '\0'. Note
4316 * that a duplicate buffer is created even for a zero-length token.
4318 * Returns a pointer to the newly-allocated duplicate, or a null
4319 * pointer if memory for the duplicate was not available. If
4320 * the lenp argument is a non-null pointer, the length of the token
4321 * (not including the '\0') is returned in *lenp.
4323 * If successful, the *buf pointer will be updated to point beyond
4324 * the end of the found token.
4326 * Note: uses GFP_KERNEL for allocation.
4328 static inline char *dup_token(const char **buf, size_t *lenp)
4333 len = next_token(buf);
4334 dup = kmemdup(*buf, len + 1, GFP_KERNEL);
4337 *(dup + len) = '\0';
4347 * Parse the options provided for an "rbd add" (i.e., rbd image
4348 * mapping) request. These arrive via a write to /sys/bus/rbd/add,
4349 * and the data written is passed here via a NUL-terminated buffer.
4350 * Returns 0 if successful or an error code otherwise.
4352 * The information extracted from these options is recorded in
4353 * the other parameters which return dynamically-allocated
4356 * The address of a pointer that will refer to a ceph options
4357 * structure. Caller must release the returned pointer using
4358 * ceph_destroy_options() when it is no longer needed.
4360 * Address of an rbd options pointer. Fully initialized by
4361 * this function; caller must release with kfree().
4363 * Address of an rbd image specification pointer. Fully
4364 * initialized by this function based on parsed options.
4365 * Caller must release with rbd_spec_put().
4367 * The options passed take this form:
4368 * <mon_addrs> <options> <pool_name> <image_name> [<snap_id>]
4371 * A comma-separated list of one or more monitor addresses.
4372 * A monitor address is an ip address, optionally followed
4373 * by a port number (separated by a colon).
4374 * I.e.: ip1[:port1][,ip2[:port2]...]
4376 * A comma-separated list of ceph and/or rbd options.
4378 * The name of the rados pool containing the rbd image.
4380 * The name of the image in that pool to map.
4382 * An optional snapshot id. If provided, the mapping will
4383 * present data from the image at the time that snapshot was
4384 * created. The image head is used if no snapshot id is
4385 * provided. Snapshot mappings are always read-only.
4387 static int rbd_add_parse_args(const char *buf,
4388 struct ceph_options **ceph_opts,
4389 struct rbd_options **opts,
4390 struct rbd_spec **rbd_spec)
4394 const char *mon_addrs;
4396 size_t mon_addrs_size;
4397 struct rbd_spec *spec = NULL;
4398 struct rbd_options *rbd_opts = NULL;
4399 struct ceph_options *copts;
4402 /* The first four tokens are required */
4404 len = next_token(&buf);
4406 rbd_warn(NULL, "no monitor address(es) provided");
4410 mon_addrs_size = len + 1;
4414 options = dup_token(&buf, NULL);
4418 rbd_warn(NULL, "no options provided");
4422 spec = rbd_spec_alloc();
4426 spec->pool_name = dup_token(&buf, NULL);
4427 if (!spec->pool_name)
4429 if (!*spec->pool_name) {
4430 rbd_warn(NULL, "no pool name provided");
4434 spec->image_name = dup_token(&buf, NULL);
4435 if (!spec->image_name)
4437 if (!*spec->image_name) {
4438 rbd_warn(NULL, "no image name provided");
4443 * Snapshot name is optional; default is to use "-"
4444 * (indicating the head/no snapshot).
4446 len = next_token(&buf);
4448 buf = RBD_SNAP_HEAD_NAME; /* No snapshot supplied */
4449 len = sizeof (RBD_SNAP_HEAD_NAME) - 1;
4450 } else if (len > RBD_MAX_SNAP_NAME_LEN) {
4451 ret = -ENAMETOOLONG;
4454 snap_name = kmemdup(buf, len + 1, GFP_KERNEL);
4457 *(snap_name + len) = '\0';
4458 spec->snap_name = snap_name;
4460 /* Initialize all rbd options to the defaults */
4462 rbd_opts = kzalloc(sizeof (*rbd_opts), GFP_KERNEL);
4466 rbd_opts->read_only = RBD_READ_ONLY_DEFAULT;
4468 copts = ceph_parse_options(options, mon_addrs,
4469 mon_addrs + mon_addrs_size - 1,
4470 parse_rbd_opts_token, rbd_opts);
4471 if (IS_ERR(copts)) {
4472 ret = PTR_ERR(copts);
4493 * An rbd format 2 image has a unique identifier, distinct from the
4494 * name given to it by the user. Internally, that identifier is
4495 * what's used to specify the names of objects related to the image.
4497 * A special "rbd id" object is used to map an rbd image name to its
4498 * id. If that object doesn't exist, then there is no v2 rbd image
4499 * with the supplied name.
4501 * This function will record the given rbd_dev's image_id field if
4502 * it can be determined, and in that case will return 0. If any
4503 * errors occur a negative errno will be returned and the rbd_dev's
4504 * image_id field will be unchanged (and should be NULL).
4506 static int rbd_dev_image_id(struct rbd_device *rbd_dev)
4515 * When probing a parent image, the image id is already
4516 * known (and the image name likely is not). There's no
4517 * need to fetch the image id again in this case. We
4518 * do still need to set the image format though.
4520 if (rbd_dev->spec->image_id) {
4521 rbd_dev->image_format = *rbd_dev->spec->image_id ? 2 : 1;
4527 * First, see if the format 2 image id file exists, and if
4528 * so, get the image's persistent id from it.
4530 size = sizeof (RBD_ID_PREFIX) + strlen(rbd_dev->spec->image_name);
4531 object_name = kmalloc(size, GFP_NOIO);
4534 sprintf(object_name, "%s%s", RBD_ID_PREFIX, rbd_dev->spec->image_name);
4535 dout("rbd id object name is %s\n", object_name);
4537 /* Response will be an encoded string, which includes a length */
4539 size = sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX;
4540 response = kzalloc(size, GFP_NOIO);
4546 /* If it doesn't exist we'll assume it's a format 1 image */
4548 ret = rbd_obj_method_sync(rbd_dev, object_name,
4549 "rbd", "get_id", NULL, 0,
4550 response, RBD_IMAGE_ID_LEN_MAX, NULL);
4551 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
4552 if (ret == -ENOENT) {
4553 image_id = kstrdup("", GFP_KERNEL);
4554 ret = image_id ? 0 : -ENOMEM;
4556 rbd_dev->image_format = 1;
4557 } else if (ret > sizeof (__le32)) {
4560 image_id = ceph_extract_encoded_string(&p, p + ret,
4562 ret = IS_ERR(image_id) ? PTR_ERR(image_id) : 0;
4564 rbd_dev->image_format = 2;
4570 rbd_dev->spec->image_id = image_id;
4571 dout("image_id is %s\n", image_id);
4580 static int rbd_dev_v1_probe(struct rbd_device *rbd_dev)
4585 /* Record the header object name for this rbd image. */
4587 size = strlen(rbd_dev->spec->image_name) + sizeof (RBD_SUFFIX);
4588 rbd_dev->header_name = kmalloc(size, GFP_KERNEL);
4589 if (!rbd_dev->header_name) {
4593 sprintf(rbd_dev->header_name, "%s%s",
4594 rbd_dev->spec->image_name, RBD_SUFFIX);
4596 /* Populate rbd image metadata */
4598 ret = rbd_read_header(rbd_dev, &rbd_dev->header);
4602 /* Version 1 images have no parent (no layering) */
4604 rbd_dev->parent_spec = NULL;
4605 rbd_dev->parent_overlap = 0;
4607 dout("discovered version 1 image, header name is %s\n",
4608 rbd_dev->header_name);
4613 kfree(rbd_dev->header_name);
4614 rbd_dev->header_name = NULL;
4615 kfree(rbd_dev->spec->image_id);
4616 rbd_dev->spec->image_id = NULL;
4621 static int rbd_dev_v2_probe(struct rbd_device *rbd_dev)
4628 * Image id was filled in by the caller. Record the header
4629 * object name for this rbd image.
4631 size = sizeof (RBD_HEADER_PREFIX) + strlen(rbd_dev->spec->image_id);
4632 rbd_dev->header_name = kmalloc(size, GFP_KERNEL);
4633 if (!rbd_dev->header_name)
4635 sprintf(rbd_dev->header_name, "%s%s",
4636 RBD_HEADER_PREFIX, rbd_dev->spec->image_id);
4638 /* Get the size and object order for the image */
4639 ret = rbd_dev_v2_image_size(rbd_dev);
4643 /* Get the object prefix (a.k.a. block_name) for the image */
4645 ret = rbd_dev_v2_object_prefix(rbd_dev);
4649 /* Get the and check features for the image */
4651 ret = rbd_dev_v2_features(rbd_dev);
4655 /* If the image supports layering, get the parent info */
4657 if (rbd_dev->header.features & RBD_FEATURE_LAYERING) {
4658 ret = rbd_dev_v2_parent_info(rbd_dev);
4661 rbd_warn(rbd_dev, "WARNING: kernel support for "
4662 "layered rbd images is EXPERIMENTAL!");
4665 /* If the image supports fancy striping, get its parameters */
4667 if (rbd_dev->header.features & RBD_FEATURE_STRIPINGV2) {
4668 ret = rbd_dev_v2_striping_info(rbd_dev);
4673 /* crypto and compression type aren't (yet) supported for v2 images */
4675 rbd_dev->header.crypt_type = 0;
4676 rbd_dev->header.comp_type = 0;
4678 /* Get the snapshot context, plus the header version */
4680 ret = rbd_dev_v2_snap_context(rbd_dev, &ver);
4683 rbd_dev->header.obj_version = ver;
4685 dout("discovered version 2 image, header name is %s\n",
4686 rbd_dev->header_name);
4690 rbd_dev->parent_overlap = 0;
4691 rbd_spec_put(rbd_dev->parent_spec);
4692 rbd_dev->parent_spec = NULL;
4693 kfree(rbd_dev->header_name);
4694 rbd_dev->header_name = NULL;
4695 kfree(rbd_dev->header.object_prefix);
4696 rbd_dev->header.object_prefix = NULL;
4701 static int rbd_dev_probe_finish(struct rbd_device *rbd_dev)
4703 struct rbd_device *parent = NULL;
4704 struct rbd_spec *parent_spec = NULL;
4705 struct rbd_client *rbdc = NULL;
4708 /* no need to lock here, as rbd_dev is not registered yet */
4709 ret = rbd_dev_snaps_update(rbd_dev);
4713 ret = rbd_dev_probe_update_spec(rbd_dev);
4717 ret = rbd_dev_set_mapping(rbd_dev);
4721 /* generate unique id: find highest unique id, add one */
4722 rbd_dev_id_get(rbd_dev);
4724 /* Fill in the device name, now that we have its id. */
4725 BUILD_BUG_ON(DEV_NAME_LEN
4726 < sizeof (RBD_DRV_NAME) + MAX_INT_FORMAT_WIDTH);
4727 sprintf(rbd_dev->name, "%s%d", RBD_DRV_NAME, rbd_dev->dev_id);
4729 /* Get our block major device number. */
4731 ret = register_blkdev(0, rbd_dev->name);
4734 rbd_dev->major = ret;
4736 /* Set up the blkdev mapping. */
4738 ret = rbd_init_disk(rbd_dev);
4740 goto err_out_blkdev;
4742 ret = rbd_bus_add_dev(rbd_dev);
4747 * At this point cleanup in the event of an error is the job
4748 * of the sysfs code (initiated by rbd_bus_del_dev()).
4750 /* Probe the parent if there is one */
4752 if (rbd_dev->parent_spec) {
4754 * We need to pass a reference to the client and the
4755 * parent spec when creating the parent rbd_dev.
4756 * Images related by parent/child relationships
4757 * always share both.
4759 parent_spec = rbd_spec_get(rbd_dev->parent_spec);
4760 rbdc = __rbd_get_client(rbd_dev->rbd_client);
4762 parent = rbd_dev_create(rbdc, parent_spec);
4767 rbdc = NULL; /* parent now owns reference */
4768 parent_spec = NULL; /* parent now owns reference */
4769 ret = rbd_dev_image_probe(parent);
4771 goto err_out_parent;
4772 rbd_dev->parent = parent;
4775 ret = rbd_dev_header_watch_sync(rbd_dev, 1);
4779 /* Everything's ready. Announce the disk to the world. */
4781 add_disk(rbd_dev->disk);
4783 pr_info("%s: added with size 0x%llx\n", rbd_dev->disk->disk_name,
4784 (unsigned long long) rbd_dev->mapping.size);
4789 rbd_spec_put(rbd_dev->parent_spec);
4790 kfree(rbd_dev->header_name);
4791 rbd_dev_destroy(parent);
4793 rbd_spec_put(parent_spec);
4794 rbd_put_client(rbdc);
4796 /* this will also clean up rest of rbd_dev stuff */
4798 rbd_bus_del_dev(rbd_dev);
4802 rbd_free_disk(rbd_dev);
4804 unregister_blkdev(rbd_dev->major, rbd_dev->name);
4806 rbd_dev_id_put(rbd_dev);
4808 rbd_remove_all_snaps(rbd_dev);
4814 * Probe for the existence of the header object for the given rbd
4815 * device. For format 2 images this includes determining the image
4818 static int rbd_dev_image_probe(struct rbd_device *rbd_dev)
4823 * Get the id from the image id object. If it's not a
4824 * format 2 image, we'll get ENOENT back, and we'll assume
4825 * it's a format 1 image.
4827 ret = rbd_dev_image_id(rbd_dev);
4830 rbd_assert(rbd_dev->spec->image_id);
4831 rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
4833 if (rbd_dev->image_format == 1)
4834 ret = rbd_dev_v1_probe(rbd_dev);
4836 ret = rbd_dev_v2_probe(rbd_dev);
4840 ret = rbd_dev_probe_finish(rbd_dev);
4842 rbd_header_free(&rbd_dev->header);
4846 kfree(rbd_dev->spec->image_id);
4847 rbd_dev->spec->image_id = NULL;
4849 dout("probe failed, returning %d\n", ret);
4854 static ssize_t rbd_add(struct bus_type *bus,
4858 struct rbd_device *rbd_dev = NULL;
4859 struct ceph_options *ceph_opts = NULL;
4860 struct rbd_options *rbd_opts = NULL;
4861 struct rbd_spec *spec = NULL;
4862 struct rbd_client *rbdc;
4863 struct ceph_osd_client *osdc;
4866 if (!try_module_get(THIS_MODULE))
4869 /* parse add command */
4870 rc = rbd_add_parse_args(buf, &ceph_opts, &rbd_opts, &spec);
4872 goto err_out_module;
4874 rbdc = rbd_get_client(ceph_opts);
4879 ceph_opts = NULL; /* rbd_dev client now owns this */
4882 osdc = &rbdc->client->osdc;
4883 rc = ceph_pg_poolid_by_name(osdc->osdmap, spec->pool_name);
4885 goto err_out_client;
4886 spec->pool_id = (u64)rc;
4888 /* The ceph file layout needs to fit pool id in 32 bits */
4890 if (spec->pool_id > (u64)U32_MAX) {
4891 rbd_warn(NULL, "pool id too large (%llu > %u)\n",
4892 (unsigned long long)spec->pool_id, U32_MAX);
4894 goto err_out_client;
4897 rbd_dev = rbd_dev_create(rbdc, spec);
4899 goto err_out_client;
4900 rbdc = NULL; /* rbd_dev now owns this */
4901 spec = NULL; /* rbd_dev now owns this */
4903 rbd_dev->mapping.read_only = rbd_opts->read_only;
4905 rbd_opts = NULL; /* done with this */
4907 rc = rbd_dev_image_probe(rbd_dev);
4909 goto err_out_rbd_dev;
4913 rbd_spec_put(rbd_dev->parent_spec);
4914 kfree(rbd_dev->header_name);
4915 rbd_dev_destroy(rbd_dev);
4917 rbd_put_client(rbdc);
4920 ceph_destroy_options(ceph_opts);
4924 module_put(THIS_MODULE);
4926 dout("Error adding device %s\n", buf);
4931 static struct rbd_device *__rbd_get_dev(unsigned long dev_id)
4933 struct list_head *tmp;
4934 struct rbd_device *rbd_dev;
4936 spin_lock(&rbd_dev_list_lock);
4937 list_for_each(tmp, &rbd_dev_list) {
4938 rbd_dev = list_entry(tmp, struct rbd_device, node);
4939 if (rbd_dev->dev_id == dev_id) {
4940 spin_unlock(&rbd_dev_list_lock);
4944 spin_unlock(&rbd_dev_list_lock);
4948 static void rbd_dev_release(struct device *dev)
4950 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
4952 if (rbd_dev->watch_event)
4953 rbd_dev_header_watch_sync(rbd_dev, 0);
4955 /* clean up and free blkdev */
4956 rbd_free_disk(rbd_dev);
4957 unregister_blkdev(rbd_dev->major, rbd_dev->name);
4959 /* release allocated disk header fields */
4960 rbd_header_free(&rbd_dev->header);
4962 /* done with the id, and with the rbd_dev */
4963 rbd_dev_id_put(rbd_dev);
4964 rbd_assert(rbd_dev->rbd_client != NULL);
4965 rbd_spec_put(rbd_dev->parent_spec);
4966 kfree(rbd_dev->header_name);
4967 rbd_dev_destroy(rbd_dev);
4969 /* release module ref */
4970 module_put(THIS_MODULE);
4973 static void __rbd_remove(struct rbd_device *rbd_dev)
4975 rbd_remove_all_snaps(rbd_dev);
4976 rbd_bus_del_dev(rbd_dev);
4979 static ssize_t rbd_remove(struct bus_type *bus,
4983 struct rbd_device *rbd_dev = NULL;
4988 rc = strict_strtoul(buf, 10, &ul);
4992 /* convert to int; abort if we lost anything in the conversion */
4993 target_id = (int) ul;
4994 if (target_id != ul)
4997 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
4999 rbd_dev = __rbd_get_dev(target_id);
5005 spin_lock_irq(&rbd_dev->lock);
5006 if (rbd_dev->open_count)
5009 set_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags);
5010 spin_unlock_irq(&rbd_dev->lock);
5014 while (rbd_dev->parent_spec) {
5015 struct rbd_device *first = rbd_dev;
5016 struct rbd_device *second = first->parent;
5017 struct rbd_device *third;
5020 * Follow to the parent with no grandparent and
5023 while (second && (third = second->parent)) {
5027 __rbd_remove(second);
5028 rbd_spec_put(first->parent_spec);
5029 first->parent_spec = NULL;
5030 first->parent_overlap = 0;
5031 first->parent = NULL;
5033 __rbd_remove(rbd_dev);
5036 mutex_unlock(&ctl_mutex);
5042 * create control files in sysfs
5045 static int rbd_sysfs_init(void)
5049 ret = device_register(&rbd_root_dev);
5053 ret = bus_register(&rbd_bus_type);
5055 device_unregister(&rbd_root_dev);
5060 static void rbd_sysfs_cleanup(void)
5062 bus_unregister(&rbd_bus_type);
5063 device_unregister(&rbd_root_dev);
5066 static int __init rbd_init(void)
5070 if (!libceph_compatible(NULL)) {
5071 rbd_warn(NULL, "libceph incompatibility (quitting)");
5075 rc = rbd_sysfs_init();
5078 pr_info("loaded " RBD_DRV_NAME_LONG "\n");
5082 static void __exit rbd_exit(void)
5084 rbd_sysfs_cleanup();
5087 module_init(rbd_init);
5088 module_exit(rbd_exit);
5090 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
5091 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
5092 MODULE_DESCRIPTION("rados block device");
5094 /* following authorship retained from original osdblk.c */
5095 MODULE_AUTHOR("Jeff Garzik <jeff@garzik.org>");
5097 MODULE_LICENSE("GPL");