3 rbd.c -- Export ceph rados objects as a Linux block device
6 based on drivers/block/osdblk.c:
8 Copyright 2009 Red Hat, Inc.
10 This program is free software; you can redistribute it and/or modify
11 it under the terms of the GNU General Public License as published by
12 the Free Software Foundation.
14 This program is distributed in the hope that it will be useful,
15 but WITHOUT ANY WARRANTY; without even the implied warranty of
16 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
17 GNU General Public License for more details.
19 You should have received a copy of the GNU General Public License
20 along with this program; see the file COPYING. If not, write to
21 the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA.
25 For usage instructions, please refer to:
27 Documentation/ABI/testing/sysfs-bus-rbd
31 #include <linux/ceph/libceph.h>
32 #include <linux/ceph/osd_client.h>
33 #include <linux/ceph/mon_client.h>
34 #include <linux/ceph/decode.h>
35 #include <linux/parser.h>
36 #include <linux/bsearch.h>
38 #include <linux/kernel.h>
39 #include <linux/device.h>
40 #include <linux/module.h>
42 #include <linux/blkdev.h>
43 #include <linux/slab.h>
45 #include "rbd_types.h"
47 #define RBD_DEBUG /* Activate rbd_assert() calls */
50 * The basic unit of block I/O is a sector. It is interpreted in a
51 * number of contexts in Linux (blk, bio, genhd), but the default is
52 * universally 512 bytes. These symbols are just slightly more
53 * meaningful than the bare numbers they represent.
55 #define SECTOR_SHIFT 9
56 #define SECTOR_SIZE (1ULL << SECTOR_SHIFT)
58 #define RBD_DRV_NAME "rbd"
59 #define RBD_DRV_NAME_LONG "rbd (rados block device)"
61 #define RBD_MINORS_PER_MAJOR 256 /* max minors per blkdev */
63 #define RBD_SNAP_DEV_NAME_PREFIX "snap_"
64 #define RBD_MAX_SNAP_NAME_LEN \
65 (NAME_MAX - (sizeof (RBD_SNAP_DEV_NAME_PREFIX) - 1))
67 #define RBD_MAX_SNAP_COUNT 510 /* allows max snapc to fit in 4KB */
69 #define RBD_SNAP_HEAD_NAME "-"
71 #define BAD_SNAP_INDEX U32_MAX /* invalid index into snap array */
73 /* This allows a single page to hold an image name sent by OSD */
74 #define RBD_IMAGE_NAME_LEN_MAX (PAGE_SIZE - sizeof (__le32) - 1)
75 #define RBD_IMAGE_ID_LEN_MAX 64
77 #define RBD_OBJ_PREFIX_LEN_MAX 64
81 #define RBD_FEATURE_LAYERING (1<<0)
82 #define RBD_FEATURE_STRIPINGV2 (1<<1)
83 #define RBD_FEATURES_ALL \
84 (RBD_FEATURE_LAYERING | RBD_FEATURE_STRIPINGV2)
86 /* Features supported by this (client software) implementation. */
88 #define RBD_FEATURES_SUPPORTED (RBD_FEATURES_ALL)
91 * An RBD device name will be "rbd#", where the "rbd" comes from
92 * RBD_DRV_NAME above, and # is a unique integer identifier.
93 * MAX_INT_FORMAT_WIDTH is used in ensuring DEV_NAME_LEN is big
94 * enough to hold all possible device names.
96 #define DEV_NAME_LEN 32
97 #define MAX_INT_FORMAT_WIDTH ((5 * sizeof (int)) / 2 + 1)
100 * block device image metadata (in-memory version)
102 struct rbd_image_header {
103 /* These four fields never change for a given rbd image */
110 /* The remaining fields need to be updated occasionally */
112 struct ceph_snap_context *snapc;
121 * An rbd image specification.
123 * The tuple (pool_id, image_id, snap_id) is sufficient to uniquely
124 * identify an image. Each rbd_dev structure includes a pointer to
125 * an rbd_spec structure that encapsulates this identity.
127 * Each of the id's in an rbd_spec has an associated name. For a
128 * user-mapped image, the names are supplied and the id's associated
129 * with them are looked up. For a layered image, a parent image is
130 * defined by the tuple, and the names are looked up.
132 * An rbd_dev structure contains a parent_spec pointer which is
133 * non-null if the image it represents is a child in a layered
134 * image. This pointer will refer to the rbd_spec structure used
135 * by the parent rbd_dev for its own identity (i.e., the structure
136 * is shared between the parent and child).
138 * Since these structures are populated once, during the discovery
139 * phase of image construction, they are effectively immutable so
140 * we make no effort to synchronize access to them.
142 * Note that code herein does not assume the image name is known (it
143 * could be a null pointer).
147 const char *pool_name;
149 const char *image_id;
150 const char *image_name;
153 const char *snap_name;
159 * an instance of the client. multiple devices may share an rbd client.
162 struct ceph_client *client;
164 struct list_head node;
167 struct rbd_img_request;
168 typedef void (*rbd_img_callback_t)(struct rbd_img_request *);
170 #define BAD_WHICH U32_MAX /* Good which or bad which, which? */
172 struct rbd_obj_request;
173 typedef void (*rbd_obj_callback_t)(struct rbd_obj_request *);
175 enum obj_request_type {
176 OBJ_REQUEST_NODATA, OBJ_REQUEST_BIO, OBJ_REQUEST_PAGES
180 OBJ_REQ_DONE, /* completion flag: not done = 0, done = 1 */
181 OBJ_REQ_IMG_DATA, /* object usage: standalone = 0, image = 1 */
182 OBJ_REQ_KNOWN, /* EXISTS flag valid: no = 0, yes = 1 */
183 OBJ_REQ_EXISTS, /* target exists: no = 0, yes = 1 */
186 struct rbd_obj_request {
187 const char *object_name;
188 u64 offset; /* object start byte */
189 u64 length; /* bytes from offset */
193 * An object request associated with an image will have its
194 * img_data flag set; a standalone object request will not.
196 * A standalone object request will have which == BAD_WHICH
197 * and a null obj_request pointer.
199 * An object request initiated in support of a layered image
200 * object (to check for its existence before a write) will
201 * have which == BAD_WHICH and a non-null obj_request pointer.
203 * Finally, an object request for rbd image data will have
204 * which != BAD_WHICH, and will have a non-null img_request
205 * pointer. The value of which will be in the range
206 * 0..(img_request->obj_request_count-1).
209 struct rbd_obj_request *obj_request; /* STAT op */
211 struct rbd_img_request *img_request;
213 /* links for img_request->obj_requests list */
214 struct list_head links;
217 u32 which; /* posn image request list */
219 enum obj_request_type type;
221 struct bio *bio_list;
227 struct page **copyup_pages;
229 struct ceph_osd_request *osd_req;
231 u64 xferred; /* bytes transferred */
234 rbd_obj_callback_t callback;
235 struct completion completion;
241 IMG_REQ_WRITE, /* I/O direction: read = 0, write = 1 */
242 IMG_REQ_CHILD, /* initiator: block = 0, child image = 1 */
243 IMG_REQ_LAYERED, /* ENOENT handling: normal = 0, layered = 1 */
246 struct rbd_img_request {
247 struct rbd_device *rbd_dev;
248 u64 offset; /* starting image byte offset */
249 u64 length; /* byte count from offset */
252 u64 snap_id; /* for reads */
253 struct ceph_snap_context *snapc; /* for writes */
256 struct request *rq; /* block request */
257 struct rbd_obj_request *obj_request; /* obj req initiator */
259 struct page **copyup_pages;
260 spinlock_t completion_lock;/* protects next_completion */
262 rbd_img_callback_t callback;
263 u64 xferred;/* aggregate bytes transferred */
264 int result; /* first nonzero obj_request result */
266 u32 obj_request_count;
267 struct list_head obj_requests; /* rbd_obj_request structs */
272 #define for_each_obj_request(ireq, oreq) \
273 list_for_each_entry(oreq, &(ireq)->obj_requests, links)
274 #define for_each_obj_request_from(ireq, oreq) \
275 list_for_each_entry_from(oreq, &(ireq)->obj_requests, links)
276 #define for_each_obj_request_safe(ireq, oreq, n) \
277 list_for_each_entry_safe_reverse(oreq, n, &(ireq)->obj_requests, links)
289 int dev_id; /* blkdev unique id */
291 int major; /* blkdev assigned major */
292 struct gendisk *disk; /* blkdev's gendisk and rq */
294 u32 image_format; /* Either 1 or 2 */
295 struct rbd_client *rbd_client;
297 char name[DEV_NAME_LEN]; /* blkdev name, e.g. rbd3 */
299 spinlock_t lock; /* queue, flags, open_count */
301 struct rbd_image_header header;
302 unsigned long flags; /* possibly lock protected */
303 struct rbd_spec *spec;
307 struct ceph_file_layout layout;
309 struct ceph_osd_event *watch_event;
310 struct rbd_obj_request *watch_request;
312 struct rbd_spec *parent_spec;
314 struct rbd_device *parent;
316 /* protects updating the header */
317 struct rw_semaphore header_rwsem;
319 struct rbd_mapping mapping;
321 struct list_head node;
325 unsigned long open_count; /* protected by lock */
329 * Flag bits for rbd_dev->flags. If atomicity is required,
330 * rbd_dev->lock is used to protect access.
332 * Currently, only the "removing" flag (which is coupled with the
333 * "open_count" field) requires atomic access.
336 RBD_DEV_FLAG_EXISTS, /* mapped snapshot has not been deleted */
337 RBD_DEV_FLAG_REMOVING, /* this mapping is being removed */
340 static DEFINE_MUTEX(ctl_mutex); /* Serialize open/close/setup/teardown */
342 static LIST_HEAD(rbd_dev_list); /* devices */
343 static DEFINE_SPINLOCK(rbd_dev_list_lock);
345 static LIST_HEAD(rbd_client_list); /* clients */
346 static DEFINE_SPINLOCK(rbd_client_list_lock);
348 /* Slab caches for frequently-allocated structures */
350 static struct kmem_cache *rbd_img_request_cache;
351 static struct kmem_cache *rbd_obj_request_cache;
352 static struct kmem_cache *rbd_segment_name_cache;
354 static int rbd_img_request_submit(struct rbd_img_request *img_request);
356 static void rbd_dev_device_release(struct device *dev);
358 static ssize_t rbd_add(struct bus_type *bus, const char *buf,
360 static ssize_t rbd_remove(struct bus_type *bus, const char *buf,
362 static int rbd_dev_image_probe(struct rbd_device *rbd_dev);
364 static struct bus_attribute rbd_bus_attrs[] = {
365 __ATTR(add, S_IWUSR, NULL, rbd_add),
366 __ATTR(remove, S_IWUSR, NULL, rbd_remove),
370 static struct bus_type rbd_bus_type = {
372 .bus_attrs = rbd_bus_attrs,
375 static void rbd_root_dev_release(struct device *dev)
379 static struct device rbd_root_dev = {
381 .release = rbd_root_dev_release,
384 static __printf(2, 3)
385 void rbd_warn(struct rbd_device *rbd_dev, const char *fmt, ...)
387 struct va_format vaf;
395 printk(KERN_WARNING "%s: %pV\n", RBD_DRV_NAME, &vaf);
396 else if (rbd_dev->disk)
397 printk(KERN_WARNING "%s: %s: %pV\n",
398 RBD_DRV_NAME, rbd_dev->disk->disk_name, &vaf);
399 else if (rbd_dev->spec && rbd_dev->spec->image_name)
400 printk(KERN_WARNING "%s: image %s: %pV\n",
401 RBD_DRV_NAME, rbd_dev->spec->image_name, &vaf);
402 else if (rbd_dev->spec && rbd_dev->spec->image_id)
403 printk(KERN_WARNING "%s: id %s: %pV\n",
404 RBD_DRV_NAME, rbd_dev->spec->image_id, &vaf);
406 printk(KERN_WARNING "%s: rbd_dev %p: %pV\n",
407 RBD_DRV_NAME, rbd_dev, &vaf);
412 #define rbd_assert(expr) \
413 if (unlikely(!(expr))) { \
414 printk(KERN_ERR "\nAssertion failure in %s() " \
416 "\trbd_assert(%s);\n\n", \
417 __func__, __LINE__, #expr); \
420 #else /* !RBD_DEBUG */
421 # define rbd_assert(expr) ((void) 0)
422 #endif /* !RBD_DEBUG */
424 static int rbd_img_obj_request_submit(struct rbd_obj_request *obj_request);
425 static void rbd_img_parent_read(struct rbd_obj_request *obj_request);
426 static void rbd_dev_remove_parent(struct rbd_device *rbd_dev);
428 static int rbd_dev_refresh(struct rbd_device *rbd_dev);
429 static int rbd_dev_v2_refresh(struct rbd_device *rbd_dev);
430 static const char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev,
432 static int _rbd_dev_v2_snap_size(struct rbd_device *rbd_dev, u64 snap_id,
433 u8 *order, u64 *snap_size);
434 static int _rbd_dev_v2_snap_features(struct rbd_device *rbd_dev, u64 snap_id,
436 static u64 rbd_snap_id_by_name(struct rbd_device *rbd_dev, const char *name);
438 static int rbd_open(struct block_device *bdev, fmode_t mode)
440 struct rbd_device *rbd_dev = bdev->bd_disk->private_data;
441 bool removing = false;
443 if ((mode & FMODE_WRITE) && rbd_dev->mapping.read_only)
446 spin_lock_irq(&rbd_dev->lock);
447 if (test_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags))
450 rbd_dev->open_count++;
451 spin_unlock_irq(&rbd_dev->lock);
455 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
456 (void) get_device(&rbd_dev->dev);
457 set_device_ro(bdev, rbd_dev->mapping.read_only);
458 mutex_unlock(&ctl_mutex);
463 static int rbd_release(struct gendisk *disk, fmode_t mode)
465 struct rbd_device *rbd_dev = disk->private_data;
466 unsigned long open_count_before;
468 spin_lock_irq(&rbd_dev->lock);
469 open_count_before = rbd_dev->open_count--;
470 spin_unlock_irq(&rbd_dev->lock);
471 rbd_assert(open_count_before > 0);
473 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
474 put_device(&rbd_dev->dev);
475 mutex_unlock(&ctl_mutex);
480 static const struct block_device_operations rbd_bd_ops = {
481 .owner = THIS_MODULE,
483 .release = rbd_release,
487 * Initialize an rbd client instance.
490 static struct rbd_client *rbd_client_create(struct ceph_options *ceph_opts)
492 struct rbd_client *rbdc;
495 dout("%s:\n", __func__);
496 rbdc = kmalloc(sizeof(struct rbd_client), GFP_KERNEL);
500 kref_init(&rbdc->kref);
501 INIT_LIST_HEAD(&rbdc->node);
503 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
505 rbdc->client = ceph_create_client(ceph_opts, rbdc, 0, 0);
506 if (IS_ERR(rbdc->client))
508 ceph_opts = NULL; /* Now rbdc->client is responsible for ceph_opts */
510 ret = ceph_open_session(rbdc->client);
514 spin_lock(&rbd_client_list_lock);
515 list_add_tail(&rbdc->node, &rbd_client_list);
516 spin_unlock(&rbd_client_list_lock);
518 mutex_unlock(&ctl_mutex);
519 dout("%s: rbdc %p\n", __func__, rbdc);
524 ceph_destroy_client(rbdc->client);
526 mutex_unlock(&ctl_mutex);
530 ceph_destroy_options(ceph_opts);
531 dout("%s: error %d\n", __func__, ret);
536 static struct rbd_client *__rbd_get_client(struct rbd_client *rbdc)
538 kref_get(&rbdc->kref);
544 * Find a ceph client with specific addr and configuration. If
545 * found, bump its reference count.
547 static struct rbd_client *rbd_client_find(struct ceph_options *ceph_opts)
549 struct rbd_client *client_node;
552 if (ceph_opts->flags & CEPH_OPT_NOSHARE)
555 spin_lock(&rbd_client_list_lock);
556 list_for_each_entry(client_node, &rbd_client_list, node) {
557 if (!ceph_compare_options(ceph_opts, client_node->client)) {
558 __rbd_get_client(client_node);
564 spin_unlock(&rbd_client_list_lock);
566 return found ? client_node : NULL;
576 /* string args above */
579 /* Boolean args above */
583 static match_table_t rbd_opts_tokens = {
585 /* string args above */
586 {Opt_read_only, "read_only"},
587 {Opt_read_only, "ro"}, /* Alternate spelling */
588 {Opt_read_write, "read_write"},
589 {Opt_read_write, "rw"}, /* Alternate spelling */
590 /* Boolean args above */
598 #define RBD_READ_ONLY_DEFAULT false
600 static int parse_rbd_opts_token(char *c, void *private)
602 struct rbd_options *rbd_opts = private;
603 substring_t argstr[MAX_OPT_ARGS];
604 int token, intval, ret;
606 token = match_token(c, rbd_opts_tokens, argstr);
610 if (token < Opt_last_int) {
611 ret = match_int(&argstr[0], &intval);
613 pr_err("bad mount option arg (not int) "
617 dout("got int token %d val %d\n", token, intval);
618 } else if (token > Opt_last_int && token < Opt_last_string) {
619 dout("got string token %d val %s\n", token,
621 } else if (token > Opt_last_string && token < Opt_last_bool) {
622 dout("got Boolean token %d\n", token);
624 dout("got token %d\n", token);
629 rbd_opts->read_only = true;
632 rbd_opts->read_only = false;
642 * Get a ceph client with specific addr and configuration, if one does
643 * not exist create it.
645 static struct rbd_client *rbd_get_client(struct ceph_options *ceph_opts)
647 struct rbd_client *rbdc;
649 rbdc = rbd_client_find(ceph_opts);
650 if (rbdc) /* using an existing client */
651 ceph_destroy_options(ceph_opts);
653 rbdc = rbd_client_create(ceph_opts);
659 * Destroy ceph client
661 * Caller must hold rbd_client_list_lock.
663 static void rbd_client_release(struct kref *kref)
665 struct rbd_client *rbdc = container_of(kref, struct rbd_client, kref);
667 dout("%s: rbdc %p\n", __func__, rbdc);
668 spin_lock(&rbd_client_list_lock);
669 list_del(&rbdc->node);
670 spin_unlock(&rbd_client_list_lock);
672 ceph_destroy_client(rbdc->client);
677 * Drop reference to ceph client node. If it's not referenced anymore, release
680 static void rbd_put_client(struct rbd_client *rbdc)
683 kref_put(&rbdc->kref, rbd_client_release);
686 static bool rbd_image_format_valid(u32 image_format)
688 return image_format == 1 || image_format == 2;
691 static bool rbd_dev_ondisk_valid(struct rbd_image_header_ondisk *ondisk)
696 /* The header has to start with the magic rbd header text */
697 if (memcmp(&ondisk->text, RBD_HEADER_TEXT, sizeof (RBD_HEADER_TEXT)))
700 /* The bio layer requires at least sector-sized I/O */
702 if (ondisk->options.order < SECTOR_SHIFT)
705 /* If we use u64 in a few spots we may be able to loosen this */
707 if (ondisk->options.order > 8 * sizeof (int) - 1)
711 * The size of a snapshot header has to fit in a size_t, and
712 * that limits the number of snapshots.
714 snap_count = le32_to_cpu(ondisk->snap_count);
715 size = SIZE_MAX - sizeof (struct ceph_snap_context);
716 if (snap_count > size / sizeof (__le64))
720 * Not only that, but the size of the entire the snapshot
721 * header must also be representable in a size_t.
723 size -= snap_count * sizeof (__le64);
724 if ((u64) size < le64_to_cpu(ondisk->snap_names_len))
731 * Create a new header structure, translate header format from the on-disk
734 static int rbd_header_from_disk(struct rbd_image_header *header,
735 struct rbd_image_header_ondisk *ondisk)
742 memset(header, 0, sizeof (*header));
744 snap_count = le32_to_cpu(ondisk->snap_count);
746 len = strnlen(ondisk->object_prefix, sizeof (ondisk->object_prefix));
747 header->object_prefix = kmalloc(len + 1, GFP_KERNEL);
748 if (!header->object_prefix)
750 memcpy(header->object_prefix, ondisk->object_prefix, len);
751 header->object_prefix[len] = '\0';
754 u64 snap_names_len = le64_to_cpu(ondisk->snap_names_len);
756 /* Save a copy of the snapshot names */
758 if (snap_names_len > (u64) SIZE_MAX)
760 header->snap_names = kmalloc(snap_names_len, GFP_KERNEL);
761 if (!header->snap_names)
764 * Note that rbd_dev_v1_header_read() guarantees
765 * the ondisk buffer we're working with has
766 * snap_names_len bytes beyond the end of the
767 * snapshot id array, this memcpy() is safe.
769 memcpy(header->snap_names, &ondisk->snaps[snap_count],
772 /* Record each snapshot's size */
774 size = snap_count * sizeof (*header->snap_sizes);
775 header->snap_sizes = kmalloc(size, GFP_KERNEL);
776 if (!header->snap_sizes)
778 for (i = 0; i < snap_count; i++)
779 header->snap_sizes[i] =
780 le64_to_cpu(ondisk->snaps[i].image_size);
782 header->snap_names = NULL;
783 header->snap_sizes = NULL;
786 header->features = 0; /* No features support in v1 images */
787 header->obj_order = ondisk->options.order;
788 header->crypt_type = ondisk->options.crypt_type;
789 header->comp_type = ondisk->options.comp_type;
791 /* Allocate and fill in the snapshot context */
793 header->image_size = le64_to_cpu(ondisk->image_size);
795 header->snapc = ceph_create_snap_context(snap_count, GFP_KERNEL);
798 header->snapc->seq = le64_to_cpu(ondisk->snap_seq);
799 for (i = 0; i < snap_count; i++)
800 header->snapc->snaps[i] = le64_to_cpu(ondisk->snaps[i].id);
805 kfree(header->snap_sizes);
806 header->snap_sizes = NULL;
807 kfree(header->snap_names);
808 header->snap_names = NULL;
809 kfree(header->object_prefix);
810 header->object_prefix = NULL;
815 static const char *_rbd_dev_v1_snap_name(struct rbd_device *rbd_dev, u32 which)
817 const char *snap_name;
819 rbd_assert(which < rbd_dev->header.snapc->num_snaps);
821 /* Skip over names until we find the one we are looking for */
823 snap_name = rbd_dev->header.snap_names;
825 snap_name += strlen(snap_name) + 1;
827 return kstrdup(snap_name, GFP_KERNEL);
831 * Snapshot id comparison function for use with qsort()/bsearch().
832 * Note that result is for snapshots in *descending* order.
834 static int snapid_compare_reverse(const void *s1, const void *s2)
836 u64 snap_id1 = *(u64 *)s1;
837 u64 snap_id2 = *(u64 *)s2;
839 if (snap_id1 < snap_id2)
841 return snap_id1 == snap_id2 ? 0 : -1;
845 * Search a snapshot context to see if the given snapshot id is
848 * Returns the position of the snapshot id in the array if it's found,
849 * or BAD_SNAP_INDEX otherwise.
851 * Note: The snapshot array is in kept sorted (by the osd) in
852 * reverse order, highest snapshot id first.
854 static u32 rbd_dev_snap_index(struct rbd_device *rbd_dev, u64 snap_id)
856 struct ceph_snap_context *snapc = rbd_dev->header.snapc;
859 found = bsearch(&snap_id, &snapc->snaps, snapc->num_snaps,
860 sizeof (snap_id), snapid_compare_reverse);
862 return found ? (u32)(found - &snapc->snaps[0]) : BAD_SNAP_INDEX;
865 static const char *rbd_dev_v1_snap_name(struct rbd_device *rbd_dev,
870 which = rbd_dev_snap_index(rbd_dev, snap_id);
871 if (which == BAD_SNAP_INDEX)
874 return _rbd_dev_v1_snap_name(rbd_dev, which);
877 static const char *rbd_snap_name(struct rbd_device *rbd_dev, u64 snap_id)
879 if (snap_id == CEPH_NOSNAP)
880 return RBD_SNAP_HEAD_NAME;
882 rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
883 if (rbd_dev->image_format == 1)
884 return rbd_dev_v1_snap_name(rbd_dev, snap_id);
886 return rbd_dev_v2_snap_name(rbd_dev, snap_id);
889 static int rbd_snap_size(struct rbd_device *rbd_dev, u64 snap_id,
892 rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
893 if (snap_id == CEPH_NOSNAP) {
894 *snap_size = rbd_dev->header.image_size;
895 } else if (rbd_dev->image_format == 1) {
898 which = rbd_dev_snap_index(rbd_dev, snap_id);
899 if (which == BAD_SNAP_INDEX)
902 *snap_size = rbd_dev->header.snap_sizes[which];
907 ret = _rbd_dev_v2_snap_size(rbd_dev, snap_id, NULL, &size);
916 static int rbd_snap_features(struct rbd_device *rbd_dev, u64 snap_id,
919 rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
920 if (snap_id == CEPH_NOSNAP) {
921 *snap_features = rbd_dev->header.features;
922 } else if (rbd_dev->image_format == 1) {
923 *snap_features = 0; /* No features for format 1 */
928 ret = _rbd_dev_v2_snap_features(rbd_dev, snap_id, &features);
932 *snap_features = features;
937 static int rbd_dev_mapping_set(struct rbd_device *rbd_dev)
939 const char *snap_name = rbd_dev->spec->snap_name;
945 if (strcmp(snap_name, RBD_SNAP_HEAD_NAME)) {
946 snap_id = rbd_snap_id_by_name(rbd_dev, snap_name);
947 if (snap_id == CEPH_NOSNAP)
950 snap_id = CEPH_NOSNAP;
953 ret = rbd_snap_size(rbd_dev, snap_id, &size);
956 ret = rbd_snap_features(rbd_dev, snap_id, &features);
960 rbd_dev->mapping.size = size;
961 rbd_dev->mapping.features = features;
963 /* If we are mapping a snapshot it must be marked read-only */
965 if (snap_id != CEPH_NOSNAP)
966 rbd_dev->mapping.read_only = true;
971 static void rbd_dev_mapping_clear(struct rbd_device *rbd_dev)
973 rbd_dev->mapping.size = 0;
974 rbd_dev->mapping.features = 0;
975 rbd_dev->mapping.read_only = true;
978 static void rbd_dev_clear_mapping(struct rbd_device *rbd_dev)
980 rbd_dev->mapping.size = 0;
981 rbd_dev->mapping.features = 0;
982 rbd_dev->mapping.read_only = true;
985 static const char *rbd_segment_name(struct rbd_device *rbd_dev, u64 offset)
991 name = kmem_cache_alloc(rbd_segment_name_cache, GFP_NOIO);
994 segment = offset >> rbd_dev->header.obj_order;
995 ret = snprintf(name, MAX_OBJ_NAME_SIZE + 1, "%s.%012llx",
996 rbd_dev->header.object_prefix, segment);
997 if (ret < 0 || ret > MAX_OBJ_NAME_SIZE) {
998 pr_err("error formatting segment name for #%llu (%d)\n",
1007 static void rbd_segment_name_free(const char *name)
1009 /* The explicit cast here is needed to drop the const qualifier */
1011 kmem_cache_free(rbd_segment_name_cache, (void *)name);
1014 static u64 rbd_segment_offset(struct rbd_device *rbd_dev, u64 offset)
1016 u64 segment_size = (u64) 1 << rbd_dev->header.obj_order;
1018 return offset & (segment_size - 1);
1021 static u64 rbd_segment_length(struct rbd_device *rbd_dev,
1022 u64 offset, u64 length)
1024 u64 segment_size = (u64) 1 << rbd_dev->header.obj_order;
1026 offset &= segment_size - 1;
1028 rbd_assert(length <= U64_MAX - offset);
1029 if (offset + length > segment_size)
1030 length = segment_size - offset;
1036 * returns the size of an object in the image
1038 static u64 rbd_obj_bytes(struct rbd_image_header *header)
1040 return 1 << header->obj_order;
1047 static void bio_chain_put(struct bio *chain)
1053 chain = chain->bi_next;
1059 * zeros a bio chain, starting at specific offset
1061 static void zero_bio_chain(struct bio *chain, int start_ofs)
1064 unsigned long flags;
1070 bio_for_each_segment(bv, chain, i) {
1071 if (pos + bv->bv_len > start_ofs) {
1072 int remainder = max(start_ofs - pos, 0);
1073 buf = bvec_kmap_irq(bv, &flags);
1074 memset(buf + remainder, 0,
1075 bv->bv_len - remainder);
1076 bvec_kunmap_irq(buf, &flags);
1081 chain = chain->bi_next;
1086 * similar to zero_bio_chain(), zeros data defined by a page array,
1087 * starting at the given byte offset from the start of the array and
1088 * continuing up to the given end offset. The pages array is
1089 * assumed to be big enough to hold all bytes up to the end.
1091 static void zero_pages(struct page **pages, u64 offset, u64 end)
1093 struct page **page = &pages[offset >> PAGE_SHIFT];
1095 rbd_assert(end > offset);
1096 rbd_assert(end - offset <= (u64)SIZE_MAX);
1097 while (offset < end) {
1100 unsigned long flags;
1103 page_offset = (size_t)(offset & ~PAGE_MASK);
1104 length = min(PAGE_SIZE - page_offset, (size_t)(end - offset));
1105 local_irq_save(flags);
1106 kaddr = kmap_atomic(*page);
1107 memset(kaddr + page_offset, 0, length);
1108 kunmap_atomic(kaddr);
1109 local_irq_restore(flags);
1117 * Clone a portion of a bio, starting at the given byte offset
1118 * and continuing for the number of bytes indicated.
1120 static struct bio *bio_clone_range(struct bio *bio_src,
1121 unsigned int offset,
1129 unsigned short end_idx;
1130 unsigned short vcnt;
1133 /* Handle the easy case for the caller */
1135 if (!offset && len == bio_src->bi_size)
1136 return bio_clone(bio_src, gfpmask);
1138 if (WARN_ON_ONCE(!len))
1140 if (WARN_ON_ONCE(len > bio_src->bi_size))
1142 if (WARN_ON_ONCE(offset > bio_src->bi_size - len))
1145 /* Find first affected segment... */
1148 __bio_for_each_segment(bv, bio_src, idx, 0) {
1149 if (resid < bv->bv_len)
1151 resid -= bv->bv_len;
1155 /* ...and the last affected segment */
1158 __bio_for_each_segment(bv, bio_src, end_idx, idx) {
1159 if (resid <= bv->bv_len)
1161 resid -= bv->bv_len;
1163 vcnt = end_idx - idx + 1;
1165 /* Build the clone */
1167 bio = bio_alloc(gfpmask, (unsigned int) vcnt);
1169 return NULL; /* ENOMEM */
1171 bio->bi_bdev = bio_src->bi_bdev;
1172 bio->bi_sector = bio_src->bi_sector + (offset >> SECTOR_SHIFT);
1173 bio->bi_rw = bio_src->bi_rw;
1174 bio->bi_flags |= 1 << BIO_CLONED;
1177 * Copy over our part of the bio_vec, then update the first
1178 * and last (or only) entries.
1180 memcpy(&bio->bi_io_vec[0], &bio_src->bi_io_vec[idx],
1181 vcnt * sizeof (struct bio_vec));
1182 bio->bi_io_vec[0].bv_offset += voff;
1184 bio->bi_io_vec[0].bv_len -= voff;
1185 bio->bi_io_vec[vcnt - 1].bv_len = resid;
1187 bio->bi_io_vec[0].bv_len = len;
1190 bio->bi_vcnt = vcnt;
1198 * Clone a portion of a bio chain, starting at the given byte offset
1199 * into the first bio in the source chain and continuing for the
1200 * number of bytes indicated. The result is another bio chain of
1201 * exactly the given length, or a null pointer on error.
1203 * The bio_src and offset parameters are both in-out. On entry they
1204 * refer to the first source bio and the offset into that bio where
1205 * the start of data to be cloned is located.
1207 * On return, bio_src is updated to refer to the bio in the source
1208 * chain that contains first un-cloned byte, and *offset will
1209 * contain the offset of that byte within that bio.
1211 static struct bio *bio_chain_clone_range(struct bio **bio_src,
1212 unsigned int *offset,
1216 struct bio *bi = *bio_src;
1217 unsigned int off = *offset;
1218 struct bio *chain = NULL;
1221 /* Build up a chain of clone bios up to the limit */
1223 if (!bi || off >= bi->bi_size || !len)
1224 return NULL; /* Nothing to clone */
1228 unsigned int bi_size;
1232 rbd_warn(NULL, "bio_chain exhausted with %u left", len);
1233 goto out_err; /* EINVAL; ran out of bio's */
1235 bi_size = min_t(unsigned int, bi->bi_size - off, len);
1236 bio = bio_clone_range(bi, off, bi_size, gfpmask);
1238 goto out_err; /* ENOMEM */
1241 end = &bio->bi_next;
1244 if (off == bi->bi_size) {
1255 bio_chain_put(chain);
1261 * The default/initial value for all object request flags is 0. For
1262 * each flag, once its value is set to 1 it is never reset to 0
1265 static void obj_request_img_data_set(struct rbd_obj_request *obj_request)
1267 if (test_and_set_bit(OBJ_REQ_IMG_DATA, &obj_request->flags)) {
1268 struct rbd_device *rbd_dev;
1270 rbd_dev = obj_request->img_request->rbd_dev;
1271 rbd_warn(rbd_dev, "obj_request %p already marked img_data\n",
1276 static bool obj_request_img_data_test(struct rbd_obj_request *obj_request)
1279 return test_bit(OBJ_REQ_IMG_DATA, &obj_request->flags) != 0;
1282 static void obj_request_done_set(struct rbd_obj_request *obj_request)
1284 if (test_and_set_bit(OBJ_REQ_DONE, &obj_request->flags)) {
1285 struct rbd_device *rbd_dev = NULL;
1287 if (obj_request_img_data_test(obj_request))
1288 rbd_dev = obj_request->img_request->rbd_dev;
1289 rbd_warn(rbd_dev, "obj_request %p already marked done\n",
1294 static bool obj_request_done_test(struct rbd_obj_request *obj_request)
1297 return test_bit(OBJ_REQ_DONE, &obj_request->flags) != 0;
1301 * This sets the KNOWN flag after (possibly) setting the EXISTS
1302 * flag. The latter is set based on the "exists" value provided.
1304 * Note that for our purposes once an object exists it never goes
1305 * away again. It's possible that the response from two existence
1306 * checks are separated by the creation of the target object, and
1307 * the first ("doesn't exist") response arrives *after* the second
1308 * ("does exist"). In that case we ignore the second one.
1310 static void obj_request_existence_set(struct rbd_obj_request *obj_request,
1314 set_bit(OBJ_REQ_EXISTS, &obj_request->flags);
1315 set_bit(OBJ_REQ_KNOWN, &obj_request->flags);
1319 static bool obj_request_known_test(struct rbd_obj_request *obj_request)
1322 return test_bit(OBJ_REQ_KNOWN, &obj_request->flags) != 0;
1325 static bool obj_request_exists_test(struct rbd_obj_request *obj_request)
1328 return test_bit(OBJ_REQ_EXISTS, &obj_request->flags) != 0;
1331 static void rbd_obj_request_get(struct rbd_obj_request *obj_request)
1333 dout("%s: obj %p (was %d)\n", __func__, obj_request,
1334 atomic_read(&obj_request->kref.refcount));
1335 kref_get(&obj_request->kref);
1338 static void rbd_obj_request_destroy(struct kref *kref);
1339 static void rbd_obj_request_put(struct rbd_obj_request *obj_request)
1341 rbd_assert(obj_request != NULL);
1342 dout("%s: obj %p (was %d)\n", __func__, obj_request,
1343 atomic_read(&obj_request->kref.refcount));
1344 kref_put(&obj_request->kref, rbd_obj_request_destroy);
1347 static void rbd_img_request_get(struct rbd_img_request *img_request)
1349 dout("%s: img %p (was %d)\n", __func__, img_request,
1350 atomic_read(&img_request->kref.refcount));
1351 kref_get(&img_request->kref);
1354 static void rbd_img_request_destroy(struct kref *kref);
1355 static void rbd_img_request_put(struct rbd_img_request *img_request)
1357 rbd_assert(img_request != NULL);
1358 dout("%s: img %p (was %d)\n", __func__, img_request,
1359 atomic_read(&img_request->kref.refcount));
1360 kref_put(&img_request->kref, rbd_img_request_destroy);
1363 static inline void rbd_img_obj_request_add(struct rbd_img_request *img_request,
1364 struct rbd_obj_request *obj_request)
1366 rbd_assert(obj_request->img_request == NULL);
1368 /* Image request now owns object's original reference */
1369 obj_request->img_request = img_request;
1370 obj_request->which = img_request->obj_request_count;
1371 rbd_assert(!obj_request_img_data_test(obj_request));
1372 obj_request_img_data_set(obj_request);
1373 rbd_assert(obj_request->which != BAD_WHICH);
1374 img_request->obj_request_count++;
1375 list_add_tail(&obj_request->links, &img_request->obj_requests);
1376 dout("%s: img %p obj %p w=%u\n", __func__, img_request, obj_request,
1377 obj_request->which);
1380 static inline void rbd_img_obj_request_del(struct rbd_img_request *img_request,
1381 struct rbd_obj_request *obj_request)
1383 rbd_assert(obj_request->which != BAD_WHICH);
1385 dout("%s: img %p obj %p w=%u\n", __func__, img_request, obj_request,
1386 obj_request->which);
1387 list_del(&obj_request->links);
1388 rbd_assert(img_request->obj_request_count > 0);
1389 img_request->obj_request_count--;
1390 rbd_assert(obj_request->which == img_request->obj_request_count);
1391 obj_request->which = BAD_WHICH;
1392 rbd_assert(obj_request_img_data_test(obj_request));
1393 rbd_assert(obj_request->img_request == img_request);
1394 obj_request->img_request = NULL;
1395 obj_request->callback = NULL;
1396 rbd_obj_request_put(obj_request);
1399 static bool obj_request_type_valid(enum obj_request_type type)
1402 case OBJ_REQUEST_NODATA:
1403 case OBJ_REQUEST_BIO:
1404 case OBJ_REQUEST_PAGES:
1411 static int rbd_obj_request_submit(struct ceph_osd_client *osdc,
1412 struct rbd_obj_request *obj_request)
1414 dout("%s: osdc %p obj %p\n", __func__, osdc, obj_request);
1416 return ceph_osdc_start_request(osdc, obj_request->osd_req, false);
1419 static void rbd_img_request_complete(struct rbd_img_request *img_request)
1422 dout("%s: img %p\n", __func__, img_request);
1425 * If no error occurred, compute the aggregate transfer
1426 * count for the image request. We could instead use
1427 * atomic64_cmpxchg() to update it as each object request
1428 * completes; not clear which way is better off hand.
1430 if (!img_request->result) {
1431 struct rbd_obj_request *obj_request;
1434 for_each_obj_request(img_request, obj_request)
1435 xferred += obj_request->xferred;
1436 img_request->xferred = xferred;
1439 if (img_request->callback)
1440 img_request->callback(img_request);
1442 rbd_img_request_put(img_request);
1445 /* Caller is responsible for rbd_obj_request_destroy(obj_request) */
1447 static int rbd_obj_request_wait(struct rbd_obj_request *obj_request)
1449 dout("%s: obj %p\n", __func__, obj_request);
1451 return wait_for_completion_interruptible(&obj_request->completion);
1455 * The default/initial value for all image request flags is 0. Each
1456 * is conditionally set to 1 at image request initialization time
1457 * and currently never change thereafter.
1459 static void img_request_write_set(struct rbd_img_request *img_request)
1461 set_bit(IMG_REQ_WRITE, &img_request->flags);
1465 static bool img_request_write_test(struct rbd_img_request *img_request)
1468 return test_bit(IMG_REQ_WRITE, &img_request->flags) != 0;
1471 static void img_request_child_set(struct rbd_img_request *img_request)
1473 set_bit(IMG_REQ_CHILD, &img_request->flags);
1477 static bool img_request_child_test(struct rbd_img_request *img_request)
1480 return test_bit(IMG_REQ_CHILD, &img_request->flags) != 0;
1483 static void img_request_layered_set(struct rbd_img_request *img_request)
1485 set_bit(IMG_REQ_LAYERED, &img_request->flags);
1489 static bool img_request_layered_test(struct rbd_img_request *img_request)
1492 return test_bit(IMG_REQ_LAYERED, &img_request->flags) != 0;
1496 rbd_img_obj_request_read_callback(struct rbd_obj_request *obj_request)
1498 u64 xferred = obj_request->xferred;
1499 u64 length = obj_request->length;
1501 dout("%s: obj %p img %p result %d %llu/%llu\n", __func__,
1502 obj_request, obj_request->img_request, obj_request->result,
1505 * ENOENT means a hole in the image. We zero-fill the
1506 * entire length of the request. A short read also implies
1507 * zero-fill to the end of the request. Either way we
1508 * update the xferred count to indicate the whole request
1511 rbd_assert(obj_request->type != OBJ_REQUEST_NODATA);
1512 if (obj_request->result == -ENOENT) {
1513 if (obj_request->type == OBJ_REQUEST_BIO)
1514 zero_bio_chain(obj_request->bio_list, 0);
1516 zero_pages(obj_request->pages, 0, length);
1517 obj_request->result = 0;
1518 obj_request->xferred = length;
1519 } else if (xferred < length && !obj_request->result) {
1520 if (obj_request->type == OBJ_REQUEST_BIO)
1521 zero_bio_chain(obj_request->bio_list, xferred);
1523 zero_pages(obj_request->pages, xferred, length);
1524 obj_request->xferred = length;
1526 obj_request_done_set(obj_request);
1529 static void rbd_obj_request_complete(struct rbd_obj_request *obj_request)
1531 dout("%s: obj %p cb %p\n", __func__, obj_request,
1532 obj_request->callback);
1533 if (obj_request->callback)
1534 obj_request->callback(obj_request);
1536 complete_all(&obj_request->completion);
1539 static void rbd_osd_trivial_callback(struct rbd_obj_request *obj_request)
1541 dout("%s: obj %p\n", __func__, obj_request);
1542 obj_request_done_set(obj_request);
1545 static void rbd_osd_read_callback(struct rbd_obj_request *obj_request)
1547 struct rbd_img_request *img_request = NULL;
1548 struct rbd_device *rbd_dev = NULL;
1549 bool layered = false;
1551 if (obj_request_img_data_test(obj_request)) {
1552 img_request = obj_request->img_request;
1553 layered = img_request && img_request_layered_test(img_request);
1554 rbd_dev = img_request->rbd_dev;
1557 dout("%s: obj %p img %p result %d %llu/%llu\n", __func__,
1558 obj_request, img_request, obj_request->result,
1559 obj_request->xferred, obj_request->length);
1560 if (layered && obj_request->result == -ENOENT &&
1561 obj_request->img_offset < rbd_dev->parent_overlap)
1562 rbd_img_parent_read(obj_request);
1563 else if (img_request)
1564 rbd_img_obj_request_read_callback(obj_request);
1566 obj_request_done_set(obj_request);
1569 static void rbd_osd_write_callback(struct rbd_obj_request *obj_request)
1571 dout("%s: obj %p result %d %llu\n", __func__, obj_request,
1572 obj_request->result, obj_request->length);
1574 * There is no such thing as a successful short write. Set
1575 * it to our originally-requested length.
1577 obj_request->xferred = obj_request->length;
1578 obj_request_done_set(obj_request);
1582 * For a simple stat call there's nothing to do. We'll do more if
1583 * this is part of a write sequence for a layered image.
1585 static void rbd_osd_stat_callback(struct rbd_obj_request *obj_request)
1587 dout("%s: obj %p\n", __func__, obj_request);
1588 obj_request_done_set(obj_request);
1591 static void rbd_osd_req_callback(struct ceph_osd_request *osd_req,
1592 struct ceph_msg *msg)
1594 struct rbd_obj_request *obj_request = osd_req->r_priv;
1597 dout("%s: osd_req %p msg %p\n", __func__, osd_req, msg);
1598 rbd_assert(osd_req == obj_request->osd_req);
1599 if (obj_request_img_data_test(obj_request)) {
1600 rbd_assert(obj_request->img_request);
1601 rbd_assert(obj_request->which != BAD_WHICH);
1603 rbd_assert(obj_request->which == BAD_WHICH);
1606 if (osd_req->r_result < 0)
1607 obj_request->result = osd_req->r_result;
1609 BUG_ON(osd_req->r_num_ops > 2);
1612 * We support a 64-bit length, but ultimately it has to be
1613 * passed to blk_end_request(), which takes an unsigned int.
1615 obj_request->xferred = osd_req->r_reply_op_len[0];
1616 rbd_assert(obj_request->xferred < (u64)UINT_MAX);
1617 opcode = osd_req->r_ops[0].op;
1619 case CEPH_OSD_OP_READ:
1620 rbd_osd_read_callback(obj_request);
1622 case CEPH_OSD_OP_WRITE:
1623 rbd_osd_write_callback(obj_request);
1625 case CEPH_OSD_OP_STAT:
1626 rbd_osd_stat_callback(obj_request);
1628 case CEPH_OSD_OP_CALL:
1629 case CEPH_OSD_OP_NOTIFY_ACK:
1630 case CEPH_OSD_OP_WATCH:
1631 rbd_osd_trivial_callback(obj_request);
1634 rbd_warn(NULL, "%s: unsupported op %hu\n",
1635 obj_request->object_name, (unsigned short) opcode);
1639 if (obj_request_done_test(obj_request))
1640 rbd_obj_request_complete(obj_request);
1643 static void rbd_osd_req_format_read(struct rbd_obj_request *obj_request)
1645 struct rbd_img_request *img_request = obj_request->img_request;
1646 struct ceph_osd_request *osd_req = obj_request->osd_req;
1649 rbd_assert(osd_req != NULL);
1651 snap_id = img_request ? img_request->snap_id : CEPH_NOSNAP;
1652 ceph_osdc_build_request(osd_req, obj_request->offset,
1653 NULL, snap_id, NULL);
1656 static void rbd_osd_req_format_write(struct rbd_obj_request *obj_request)
1658 struct rbd_img_request *img_request = obj_request->img_request;
1659 struct ceph_osd_request *osd_req = obj_request->osd_req;
1660 struct ceph_snap_context *snapc;
1661 struct timespec mtime = CURRENT_TIME;
1663 rbd_assert(osd_req != NULL);
1665 snapc = img_request ? img_request->snapc : NULL;
1666 ceph_osdc_build_request(osd_req, obj_request->offset,
1667 snapc, CEPH_NOSNAP, &mtime);
1670 static struct ceph_osd_request *rbd_osd_req_create(
1671 struct rbd_device *rbd_dev,
1673 struct rbd_obj_request *obj_request)
1675 struct ceph_snap_context *snapc = NULL;
1676 struct ceph_osd_client *osdc;
1677 struct ceph_osd_request *osd_req;
1679 if (obj_request_img_data_test(obj_request)) {
1680 struct rbd_img_request *img_request = obj_request->img_request;
1682 rbd_assert(write_request ==
1683 img_request_write_test(img_request));
1685 snapc = img_request->snapc;
1688 /* Allocate and initialize the request, for the single op */
1690 osdc = &rbd_dev->rbd_client->client->osdc;
1691 osd_req = ceph_osdc_alloc_request(osdc, snapc, 1, false, GFP_ATOMIC);
1693 return NULL; /* ENOMEM */
1696 osd_req->r_flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK;
1698 osd_req->r_flags = CEPH_OSD_FLAG_READ;
1700 osd_req->r_callback = rbd_osd_req_callback;
1701 osd_req->r_priv = obj_request;
1703 osd_req->r_oid_len = strlen(obj_request->object_name);
1704 rbd_assert(osd_req->r_oid_len < sizeof (osd_req->r_oid));
1705 memcpy(osd_req->r_oid, obj_request->object_name, osd_req->r_oid_len);
1707 osd_req->r_file_layout = rbd_dev->layout; /* struct */
1713 * Create a copyup osd request based on the information in the
1714 * object request supplied. A copyup request has two osd ops,
1715 * a copyup method call, and a "normal" write request.
1717 static struct ceph_osd_request *
1718 rbd_osd_req_create_copyup(struct rbd_obj_request *obj_request)
1720 struct rbd_img_request *img_request;
1721 struct ceph_snap_context *snapc;
1722 struct rbd_device *rbd_dev;
1723 struct ceph_osd_client *osdc;
1724 struct ceph_osd_request *osd_req;
1726 rbd_assert(obj_request_img_data_test(obj_request));
1727 img_request = obj_request->img_request;
1728 rbd_assert(img_request);
1729 rbd_assert(img_request_write_test(img_request));
1731 /* Allocate and initialize the request, for the two ops */
1733 snapc = img_request->snapc;
1734 rbd_dev = img_request->rbd_dev;
1735 osdc = &rbd_dev->rbd_client->client->osdc;
1736 osd_req = ceph_osdc_alloc_request(osdc, snapc, 2, false, GFP_ATOMIC);
1738 return NULL; /* ENOMEM */
1740 osd_req->r_flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK;
1741 osd_req->r_callback = rbd_osd_req_callback;
1742 osd_req->r_priv = obj_request;
1744 osd_req->r_oid_len = strlen(obj_request->object_name);
1745 rbd_assert(osd_req->r_oid_len < sizeof (osd_req->r_oid));
1746 memcpy(osd_req->r_oid, obj_request->object_name, osd_req->r_oid_len);
1748 osd_req->r_file_layout = rbd_dev->layout; /* struct */
1754 static void rbd_osd_req_destroy(struct ceph_osd_request *osd_req)
1756 ceph_osdc_put_request(osd_req);
1759 /* object_name is assumed to be a non-null pointer and NUL-terminated */
1761 static struct rbd_obj_request *rbd_obj_request_create(const char *object_name,
1762 u64 offset, u64 length,
1763 enum obj_request_type type)
1765 struct rbd_obj_request *obj_request;
1769 rbd_assert(obj_request_type_valid(type));
1771 size = strlen(object_name) + 1;
1772 name = kmalloc(size, GFP_KERNEL);
1776 obj_request = kmem_cache_zalloc(rbd_obj_request_cache, GFP_KERNEL);
1782 obj_request->object_name = memcpy(name, object_name, size);
1783 obj_request->offset = offset;
1784 obj_request->length = length;
1785 obj_request->flags = 0;
1786 obj_request->which = BAD_WHICH;
1787 obj_request->type = type;
1788 INIT_LIST_HEAD(&obj_request->links);
1789 init_completion(&obj_request->completion);
1790 kref_init(&obj_request->kref);
1792 dout("%s: \"%s\" %llu/%llu %d -> obj %p\n", __func__, object_name,
1793 offset, length, (int)type, obj_request);
1798 static void rbd_obj_request_destroy(struct kref *kref)
1800 struct rbd_obj_request *obj_request;
1802 obj_request = container_of(kref, struct rbd_obj_request, kref);
1804 dout("%s: obj %p\n", __func__, obj_request);
1806 rbd_assert(obj_request->img_request == NULL);
1807 rbd_assert(obj_request->which == BAD_WHICH);
1809 if (obj_request->osd_req)
1810 rbd_osd_req_destroy(obj_request->osd_req);
1812 rbd_assert(obj_request_type_valid(obj_request->type));
1813 switch (obj_request->type) {
1814 case OBJ_REQUEST_NODATA:
1815 break; /* Nothing to do */
1816 case OBJ_REQUEST_BIO:
1817 if (obj_request->bio_list)
1818 bio_chain_put(obj_request->bio_list);
1820 case OBJ_REQUEST_PAGES:
1821 if (obj_request->pages)
1822 ceph_release_page_vector(obj_request->pages,
1823 obj_request->page_count);
1827 kfree(obj_request->object_name);
1828 obj_request->object_name = NULL;
1829 kmem_cache_free(rbd_obj_request_cache, obj_request);
1833 * Caller is responsible for filling in the list of object requests
1834 * that comprises the image request, and the Linux request pointer
1835 * (if there is one).
1837 static struct rbd_img_request *rbd_img_request_create(
1838 struct rbd_device *rbd_dev,
1839 u64 offset, u64 length,
1843 struct rbd_img_request *img_request;
1845 img_request = kmem_cache_alloc(rbd_img_request_cache, GFP_ATOMIC);
1849 if (write_request) {
1850 down_read(&rbd_dev->header_rwsem);
1851 ceph_get_snap_context(rbd_dev->header.snapc);
1852 up_read(&rbd_dev->header_rwsem);
1855 img_request->rq = NULL;
1856 img_request->rbd_dev = rbd_dev;
1857 img_request->offset = offset;
1858 img_request->length = length;
1859 img_request->flags = 0;
1860 if (write_request) {
1861 img_request_write_set(img_request);
1862 img_request->snapc = rbd_dev->header.snapc;
1864 img_request->snap_id = rbd_dev->spec->snap_id;
1867 img_request_child_set(img_request);
1868 if (rbd_dev->parent_spec)
1869 img_request_layered_set(img_request);
1870 spin_lock_init(&img_request->completion_lock);
1871 img_request->next_completion = 0;
1872 img_request->callback = NULL;
1873 img_request->result = 0;
1874 img_request->obj_request_count = 0;
1875 INIT_LIST_HEAD(&img_request->obj_requests);
1876 kref_init(&img_request->kref);
1878 rbd_img_request_get(img_request); /* Avoid a warning */
1879 rbd_img_request_put(img_request); /* TEMPORARY */
1881 dout("%s: rbd_dev %p %s %llu/%llu -> img %p\n", __func__, rbd_dev,
1882 write_request ? "write" : "read", offset, length,
1888 static void rbd_img_request_destroy(struct kref *kref)
1890 struct rbd_img_request *img_request;
1891 struct rbd_obj_request *obj_request;
1892 struct rbd_obj_request *next_obj_request;
1894 img_request = container_of(kref, struct rbd_img_request, kref);
1896 dout("%s: img %p\n", __func__, img_request);
1898 for_each_obj_request_safe(img_request, obj_request, next_obj_request)
1899 rbd_img_obj_request_del(img_request, obj_request);
1900 rbd_assert(img_request->obj_request_count == 0);
1902 if (img_request_write_test(img_request))
1903 ceph_put_snap_context(img_request->snapc);
1905 if (img_request_child_test(img_request))
1906 rbd_obj_request_put(img_request->obj_request);
1908 kmem_cache_free(rbd_img_request_cache, img_request);
1911 static bool rbd_img_obj_end_request(struct rbd_obj_request *obj_request)
1913 struct rbd_img_request *img_request;
1914 unsigned int xferred;
1918 rbd_assert(obj_request_img_data_test(obj_request));
1919 img_request = obj_request->img_request;
1921 rbd_assert(obj_request->xferred <= (u64)UINT_MAX);
1922 xferred = (unsigned int)obj_request->xferred;
1923 result = obj_request->result;
1925 struct rbd_device *rbd_dev = img_request->rbd_dev;
1927 rbd_warn(rbd_dev, "%s %llx at %llx (%llx)\n",
1928 img_request_write_test(img_request) ? "write" : "read",
1929 obj_request->length, obj_request->img_offset,
1930 obj_request->offset);
1931 rbd_warn(rbd_dev, " result %d xferred %x\n",
1933 if (!img_request->result)
1934 img_request->result = result;
1937 /* Image object requests don't own their page array */
1939 if (obj_request->type == OBJ_REQUEST_PAGES) {
1940 obj_request->pages = NULL;
1941 obj_request->page_count = 0;
1944 if (img_request_child_test(img_request)) {
1945 rbd_assert(img_request->obj_request != NULL);
1946 more = obj_request->which < img_request->obj_request_count - 1;
1948 rbd_assert(img_request->rq != NULL);
1949 more = blk_end_request(img_request->rq, result, xferred);
1955 static void rbd_img_obj_callback(struct rbd_obj_request *obj_request)
1957 struct rbd_img_request *img_request;
1958 u32 which = obj_request->which;
1961 rbd_assert(obj_request_img_data_test(obj_request));
1962 img_request = obj_request->img_request;
1964 dout("%s: img %p obj %p\n", __func__, img_request, obj_request);
1965 rbd_assert(img_request != NULL);
1966 rbd_assert(img_request->obj_request_count > 0);
1967 rbd_assert(which != BAD_WHICH);
1968 rbd_assert(which < img_request->obj_request_count);
1969 rbd_assert(which >= img_request->next_completion);
1971 spin_lock_irq(&img_request->completion_lock);
1972 if (which != img_request->next_completion)
1975 for_each_obj_request_from(img_request, obj_request) {
1977 rbd_assert(which < img_request->obj_request_count);
1979 if (!obj_request_done_test(obj_request))
1981 more = rbd_img_obj_end_request(obj_request);
1985 rbd_assert(more ^ (which == img_request->obj_request_count));
1986 img_request->next_completion = which;
1988 spin_unlock_irq(&img_request->completion_lock);
1991 rbd_img_request_complete(img_request);
1995 * Split up an image request into one or more object requests, each
1996 * to a different object. The "type" parameter indicates whether
1997 * "data_desc" is the pointer to the head of a list of bio
1998 * structures, or the base of a page array. In either case this
1999 * function assumes data_desc describes memory sufficient to hold
2000 * all data described by the image request.
2002 static int rbd_img_request_fill(struct rbd_img_request *img_request,
2003 enum obj_request_type type,
2006 struct rbd_device *rbd_dev = img_request->rbd_dev;
2007 struct rbd_obj_request *obj_request = NULL;
2008 struct rbd_obj_request *next_obj_request;
2009 bool write_request = img_request_write_test(img_request);
2010 struct bio *bio_list;
2011 unsigned int bio_offset = 0;
2012 struct page **pages;
2017 dout("%s: img %p type %d data_desc %p\n", __func__, img_request,
2018 (int)type, data_desc);
2020 opcode = write_request ? CEPH_OSD_OP_WRITE : CEPH_OSD_OP_READ;
2021 img_offset = img_request->offset;
2022 resid = img_request->length;
2023 rbd_assert(resid > 0);
2025 if (type == OBJ_REQUEST_BIO) {
2026 bio_list = data_desc;
2027 rbd_assert(img_offset == bio_list->bi_sector << SECTOR_SHIFT);
2029 rbd_assert(type == OBJ_REQUEST_PAGES);
2034 struct ceph_osd_request *osd_req;
2035 const char *object_name;
2039 object_name = rbd_segment_name(rbd_dev, img_offset);
2042 offset = rbd_segment_offset(rbd_dev, img_offset);
2043 length = rbd_segment_length(rbd_dev, img_offset, resid);
2044 obj_request = rbd_obj_request_create(object_name,
2045 offset, length, type);
2046 /* object request has its own copy of the object name */
2047 rbd_segment_name_free(object_name);
2051 if (type == OBJ_REQUEST_BIO) {
2052 unsigned int clone_size;
2054 rbd_assert(length <= (u64)UINT_MAX);
2055 clone_size = (unsigned int)length;
2056 obj_request->bio_list =
2057 bio_chain_clone_range(&bio_list,
2061 if (!obj_request->bio_list)
2064 unsigned int page_count;
2066 obj_request->pages = pages;
2067 page_count = (u32)calc_pages_for(offset, length);
2068 obj_request->page_count = page_count;
2069 if ((offset + length) & ~PAGE_MASK)
2070 page_count--; /* more on last page */
2071 pages += page_count;
2074 osd_req = rbd_osd_req_create(rbd_dev, write_request,
2078 obj_request->osd_req = osd_req;
2079 obj_request->callback = rbd_img_obj_callback;
2081 osd_req_op_extent_init(osd_req, 0, opcode, offset, length,
2083 if (type == OBJ_REQUEST_BIO)
2084 osd_req_op_extent_osd_data_bio(osd_req, 0,
2085 obj_request->bio_list, length);
2087 osd_req_op_extent_osd_data_pages(osd_req, 0,
2088 obj_request->pages, length,
2089 offset & ~PAGE_MASK, false, false);
2092 rbd_osd_req_format_write(obj_request);
2094 rbd_osd_req_format_read(obj_request);
2096 obj_request->img_offset = img_offset;
2097 rbd_img_obj_request_add(img_request, obj_request);
2099 img_offset += length;
2106 rbd_obj_request_put(obj_request);
2108 for_each_obj_request_safe(img_request, obj_request, next_obj_request)
2109 rbd_obj_request_put(obj_request);
2115 rbd_img_obj_copyup_callback(struct rbd_obj_request *obj_request)
2117 struct rbd_img_request *img_request;
2118 struct rbd_device *rbd_dev;
2122 rbd_assert(obj_request->type == OBJ_REQUEST_BIO);
2123 rbd_assert(obj_request_img_data_test(obj_request));
2124 img_request = obj_request->img_request;
2125 rbd_assert(img_request);
2127 rbd_dev = img_request->rbd_dev;
2128 rbd_assert(rbd_dev);
2129 length = (u64)1 << rbd_dev->header.obj_order;
2130 page_count = (u32)calc_pages_for(0, length);
2132 rbd_assert(obj_request->copyup_pages);
2133 ceph_release_page_vector(obj_request->copyup_pages, page_count);
2134 obj_request->copyup_pages = NULL;
2137 * We want the transfer count to reflect the size of the
2138 * original write request. There is no such thing as a
2139 * successful short write, so if the request was successful
2140 * we can just set it to the originally-requested length.
2142 if (!obj_request->result)
2143 obj_request->xferred = obj_request->length;
2145 /* Finish up with the normal image object callback */
2147 rbd_img_obj_callback(obj_request);
2151 rbd_img_obj_parent_read_full_callback(struct rbd_img_request *img_request)
2153 struct rbd_obj_request *orig_request;
2154 struct ceph_osd_request *osd_req;
2155 struct ceph_osd_client *osdc;
2156 struct rbd_device *rbd_dev;
2157 struct page **pages;
2162 rbd_assert(img_request_child_test(img_request));
2164 /* First get what we need from the image request */
2166 pages = img_request->copyup_pages;
2167 rbd_assert(pages != NULL);
2168 img_request->copyup_pages = NULL;
2170 orig_request = img_request->obj_request;
2171 rbd_assert(orig_request != NULL);
2172 rbd_assert(orig_request->type == OBJ_REQUEST_BIO);
2173 result = img_request->result;
2174 obj_size = img_request->length;
2175 xferred = img_request->xferred;
2177 rbd_dev = img_request->rbd_dev;
2178 rbd_assert(rbd_dev);
2179 rbd_assert(obj_size == (u64)1 << rbd_dev->header.obj_order);
2181 rbd_img_request_put(img_request);
2186 /* Allocate the new copyup osd request for the original request */
2189 rbd_assert(!orig_request->osd_req);
2190 osd_req = rbd_osd_req_create_copyup(orig_request);
2193 orig_request->osd_req = osd_req;
2194 orig_request->copyup_pages = pages;
2196 /* Initialize the copyup op */
2198 osd_req_op_cls_init(osd_req, 0, CEPH_OSD_OP_CALL, "rbd", "copyup");
2199 osd_req_op_cls_request_data_pages(osd_req, 0, pages, obj_size, 0,
2202 /* Then the original write request op */
2204 osd_req_op_extent_init(osd_req, 1, CEPH_OSD_OP_WRITE,
2205 orig_request->offset,
2206 orig_request->length, 0, 0);
2207 osd_req_op_extent_osd_data_bio(osd_req, 1, orig_request->bio_list,
2208 orig_request->length);
2210 rbd_osd_req_format_write(orig_request);
2212 /* All set, send it off. */
2214 orig_request->callback = rbd_img_obj_copyup_callback;
2215 osdc = &rbd_dev->rbd_client->client->osdc;
2216 result = rbd_obj_request_submit(osdc, orig_request);
2220 /* Record the error code and complete the request */
2222 orig_request->result = result;
2223 orig_request->xferred = 0;
2224 obj_request_done_set(orig_request);
2225 rbd_obj_request_complete(orig_request);
2229 * Read from the parent image the range of data that covers the
2230 * entire target of the given object request. This is used for
2231 * satisfying a layered image write request when the target of an
2232 * object request from the image request does not exist.
2234 * A page array big enough to hold the returned data is allocated
2235 * and supplied to rbd_img_request_fill() as the "data descriptor."
2236 * When the read completes, this page array will be transferred to
2237 * the original object request for the copyup operation.
2239 * If an error occurs, record it as the result of the original
2240 * object request and mark it done so it gets completed.
2242 static int rbd_img_obj_parent_read_full(struct rbd_obj_request *obj_request)
2244 struct rbd_img_request *img_request = NULL;
2245 struct rbd_img_request *parent_request = NULL;
2246 struct rbd_device *rbd_dev;
2249 struct page **pages = NULL;
2253 rbd_assert(obj_request_img_data_test(obj_request));
2254 rbd_assert(obj_request->type == OBJ_REQUEST_BIO);
2256 img_request = obj_request->img_request;
2257 rbd_assert(img_request != NULL);
2258 rbd_dev = img_request->rbd_dev;
2259 rbd_assert(rbd_dev->parent != NULL);
2262 * First things first. The original osd request is of no
2263 * use to use any more, we'll need a new one that can hold
2264 * the two ops in a copyup request. We'll get that later,
2265 * but for now we can release the old one.
2267 rbd_osd_req_destroy(obj_request->osd_req);
2268 obj_request->osd_req = NULL;
2271 * Determine the byte range covered by the object in the
2272 * child image to which the original request was to be sent.
2274 img_offset = obj_request->img_offset - obj_request->offset;
2275 length = (u64)1 << rbd_dev->header.obj_order;
2278 * There is no defined parent data beyond the parent
2279 * overlap, so limit what we read at that boundary if
2282 if (img_offset + length > rbd_dev->parent_overlap) {
2283 rbd_assert(img_offset < rbd_dev->parent_overlap);
2284 length = rbd_dev->parent_overlap - img_offset;
2288 * Allocate a page array big enough to receive the data read
2291 page_count = (u32)calc_pages_for(0, length);
2292 pages = ceph_alloc_page_vector(page_count, GFP_KERNEL);
2293 if (IS_ERR(pages)) {
2294 result = PTR_ERR(pages);
2300 parent_request = rbd_img_request_create(rbd_dev->parent,
2303 if (!parent_request)
2305 rbd_obj_request_get(obj_request);
2306 parent_request->obj_request = obj_request;
2308 result = rbd_img_request_fill(parent_request, OBJ_REQUEST_PAGES, pages);
2311 parent_request->copyup_pages = pages;
2313 parent_request->callback = rbd_img_obj_parent_read_full_callback;
2314 result = rbd_img_request_submit(parent_request);
2318 parent_request->copyup_pages = NULL;
2319 parent_request->obj_request = NULL;
2320 rbd_obj_request_put(obj_request);
2323 ceph_release_page_vector(pages, page_count);
2325 rbd_img_request_put(parent_request);
2326 obj_request->result = result;
2327 obj_request->xferred = 0;
2328 obj_request_done_set(obj_request);
2333 static void rbd_img_obj_exists_callback(struct rbd_obj_request *obj_request)
2335 struct rbd_obj_request *orig_request;
2338 rbd_assert(!obj_request_img_data_test(obj_request));
2341 * All we need from the object request is the original
2342 * request and the result of the STAT op. Grab those, then
2343 * we're done with the request.
2345 orig_request = obj_request->obj_request;
2346 obj_request->obj_request = NULL;
2347 rbd_assert(orig_request);
2348 rbd_assert(orig_request->img_request);
2350 result = obj_request->result;
2351 obj_request->result = 0;
2353 dout("%s: obj %p for obj %p result %d %llu/%llu\n", __func__,
2354 obj_request, orig_request, result,
2355 obj_request->xferred, obj_request->length);
2356 rbd_obj_request_put(obj_request);
2358 rbd_assert(orig_request);
2359 rbd_assert(orig_request->img_request);
2362 * Our only purpose here is to determine whether the object
2363 * exists, and we don't want to treat the non-existence as
2364 * an error. If something else comes back, transfer the
2365 * error to the original request and complete it now.
2368 obj_request_existence_set(orig_request, true);
2369 } else if (result == -ENOENT) {
2370 obj_request_existence_set(orig_request, false);
2371 } else if (result) {
2372 orig_request->result = result;
2377 * Resubmit the original request now that we have recorded
2378 * whether the target object exists.
2380 orig_request->result = rbd_img_obj_request_submit(orig_request);
2382 if (orig_request->result)
2383 rbd_obj_request_complete(orig_request);
2384 rbd_obj_request_put(orig_request);
2387 static int rbd_img_obj_exists_submit(struct rbd_obj_request *obj_request)
2389 struct rbd_obj_request *stat_request;
2390 struct rbd_device *rbd_dev;
2391 struct ceph_osd_client *osdc;
2392 struct page **pages = NULL;
2398 * The response data for a STAT call consists of:
2405 size = sizeof (__le64) + sizeof (__le32) + sizeof (__le32);
2406 page_count = (u32)calc_pages_for(0, size);
2407 pages = ceph_alloc_page_vector(page_count, GFP_KERNEL);
2409 return PTR_ERR(pages);
2412 stat_request = rbd_obj_request_create(obj_request->object_name, 0, 0,
2417 rbd_obj_request_get(obj_request);
2418 stat_request->obj_request = obj_request;
2419 stat_request->pages = pages;
2420 stat_request->page_count = page_count;
2422 rbd_assert(obj_request->img_request);
2423 rbd_dev = obj_request->img_request->rbd_dev;
2424 stat_request->osd_req = rbd_osd_req_create(rbd_dev, false,
2426 if (!stat_request->osd_req)
2428 stat_request->callback = rbd_img_obj_exists_callback;
2430 osd_req_op_init(stat_request->osd_req, 0, CEPH_OSD_OP_STAT);
2431 osd_req_op_raw_data_in_pages(stat_request->osd_req, 0, pages, size, 0,
2433 rbd_osd_req_format_read(stat_request);
2435 osdc = &rbd_dev->rbd_client->client->osdc;
2436 ret = rbd_obj_request_submit(osdc, stat_request);
2439 rbd_obj_request_put(obj_request);
2444 static int rbd_img_obj_request_submit(struct rbd_obj_request *obj_request)
2446 struct rbd_img_request *img_request;
2447 struct rbd_device *rbd_dev;
2450 rbd_assert(obj_request_img_data_test(obj_request));
2452 img_request = obj_request->img_request;
2453 rbd_assert(img_request);
2454 rbd_dev = img_request->rbd_dev;
2457 * Only writes to layered images need special handling.
2458 * Reads and non-layered writes are simple object requests.
2459 * Layered writes that start beyond the end of the overlap
2460 * with the parent have no parent data, so they too are
2461 * simple object requests. Finally, if the target object is
2462 * known to already exist, its parent data has already been
2463 * copied, so a write to the object can also be handled as a
2464 * simple object request.
2466 if (!img_request_write_test(img_request) ||
2467 !img_request_layered_test(img_request) ||
2468 rbd_dev->parent_overlap <= obj_request->img_offset ||
2469 ((known = obj_request_known_test(obj_request)) &&
2470 obj_request_exists_test(obj_request))) {
2472 struct rbd_device *rbd_dev;
2473 struct ceph_osd_client *osdc;
2475 rbd_dev = obj_request->img_request->rbd_dev;
2476 osdc = &rbd_dev->rbd_client->client->osdc;
2478 return rbd_obj_request_submit(osdc, obj_request);
2482 * It's a layered write. The target object might exist but
2483 * we may not know that yet. If we know it doesn't exist,
2484 * start by reading the data for the full target object from
2485 * the parent so we can use it for a copyup to the target.
2488 return rbd_img_obj_parent_read_full(obj_request);
2490 /* We don't know whether the target exists. Go find out. */
2492 return rbd_img_obj_exists_submit(obj_request);
2495 static int rbd_img_request_submit(struct rbd_img_request *img_request)
2497 struct rbd_obj_request *obj_request;
2498 struct rbd_obj_request *next_obj_request;
2500 dout("%s: img %p\n", __func__, img_request);
2501 for_each_obj_request_safe(img_request, obj_request, next_obj_request) {
2504 ret = rbd_img_obj_request_submit(obj_request);
2512 static void rbd_img_parent_read_callback(struct rbd_img_request *img_request)
2514 struct rbd_obj_request *obj_request;
2515 struct rbd_device *rbd_dev;
2518 rbd_assert(img_request_child_test(img_request));
2520 obj_request = img_request->obj_request;
2521 rbd_assert(obj_request);
2522 rbd_assert(obj_request->img_request);
2524 obj_request->result = img_request->result;
2525 if (obj_request->result)
2529 * We need to zero anything beyond the parent overlap
2530 * boundary. Since rbd_img_obj_request_read_callback()
2531 * will zero anything beyond the end of a short read, an
2532 * easy way to do this is to pretend the data from the
2533 * parent came up short--ending at the overlap boundary.
2535 rbd_assert(obj_request->img_offset < U64_MAX - obj_request->length);
2536 obj_end = obj_request->img_offset + obj_request->length;
2537 rbd_dev = obj_request->img_request->rbd_dev;
2538 if (obj_end > rbd_dev->parent_overlap) {
2541 if (obj_request->img_offset < rbd_dev->parent_overlap)
2542 xferred = rbd_dev->parent_overlap -
2543 obj_request->img_offset;
2545 obj_request->xferred = min(img_request->xferred, xferred);
2547 obj_request->xferred = img_request->xferred;
2550 rbd_img_request_put(img_request);
2551 rbd_img_obj_request_read_callback(obj_request);
2552 rbd_obj_request_complete(obj_request);
2555 static void rbd_img_parent_read(struct rbd_obj_request *obj_request)
2557 struct rbd_device *rbd_dev;
2558 struct rbd_img_request *img_request;
2561 rbd_assert(obj_request_img_data_test(obj_request));
2562 rbd_assert(obj_request->img_request != NULL);
2563 rbd_assert(obj_request->result == (s32) -ENOENT);
2564 rbd_assert(obj_request->type == OBJ_REQUEST_BIO);
2566 rbd_dev = obj_request->img_request->rbd_dev;
2567 rbd_assert(rbd_dev->parent != NULL);
2568 /* rbd_read_finish(obj_request, obj_request->length); */
2569 img_request = rbd_img_request_create(rbd_dev->parent,
2570 obj_request->img_offset,
2571 obj_request->length,
2577 rbd_obj_request_get(obj_request);
2578 img_request->obj_request = obj_request;
2580 result = rbd_img_request_fill(img_request, OBJ_REQUEST_BIO,
2581 obj_request->bio_list);
2585 img_request->callback = rbd_img_parent_read_callback;
2586 result = rbd_img_request_submit(img_request);
2593 rbd_img_request_put(img_request);
2594 obj_request->result = result;
2595 obj_request->xferred = 0;
2596 obj_request_done_set(obj_request);
2599 static int rbd_obj_notify_ack(struct rbd_device *rbd_dev, u64 notify_id)
2601 struct rbd_obj_request *obj_request;
2602 struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
2605 obj_request = rbd_obj_request_create(rbd_dev->header_name, 0, 0,
2606 OBJ_REQUEST_NODATA);
2611 obj_request->osd_req = rbd_osd_req_create(rbd_dev, false, obj_request);
2612 if (!obj_request->osd_req)
2614 obj_request->callback = rbd_obj_request_put;
2616 osd_req_op_watch_init(obj_request->osd_req, 0, CEPH_OSD_OP_NOTIFY_ACK,
2618 rbd_osd_req_format_read(obj_request);
2620 ret = rbd_obj_request_submit(osdc, obj_request);
2623 rbd_obj_request_put(obj_request);
2628 static void rbd_watch_cb(u64 ver, u64 notify_id, u8 opcode, void *data)
2630 struct rbd_device *rbd_dev = (struct rbd_device *)data;
2635 dout("%s: \"%s\" notify_id %llu opcode %u\n", __func__,
2636 rbd_dev->header_name, (unsigned long long)notify_id,
2637 (unsigned int)opcode);
2638 (void)rbd_dev_refresh(rbd_dev);
2640 rbd_obj_notify_ack(rbd_dev, notify_id);
2644 * Request sync osd watch/unwatch. The value of "start" determines
2645 * whether a watch request is being initiated or torn down.
2647 static int rbd_dev_header_watch_sync(struct rbd_device *rbd_dev, int start)
2649 struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
2650 struct rbd_obj_request *obj_request;
2653 rbd_assert(start ^ !!rbd_dev->watch_event);
2654 rbd_assert(start ^ !!rbd_dev->watch_request);
2657 ret = ceph_osdc_create_event(osdc, rbd_watch_cb, rbd_dev,
2658 &rbd_dev->watch_event);
2661 rbd_assert(rbd_dev->watch_event != NULL);
2665 obj_request = rbd_obj_request_create(rbd_dev->header_name, 0, 0,
2666 OBJ_REQUEST_NODATA);
2670 obj_request->osd_req = rbd_osd_req_create(rbd_dev, true, obj_request);
2671 if (!obj_request->osd_req)
2675 ceph_osdc_set_request_linger(osdc, obj_request->osd_req);
2677 ceph_osdc_unregister_linger_request(osdc,
2678 rbd_dev->watch_request->osd_req);
2680 osd_req_op_watch_init(obj_request->osd_req, 0, CEPH_OSD_OP_WATCH,
2681 rbd_dev->watch_event->cookie, 0, start);
2682 rbd_osd_req_format_write(obj_request);
2684 ret = rbd_obj_request_submit(osdc, obj_request);
2687 ret = rbd_obj_request_wait(obj_request);
2690 ret = obj_request->result;
2695 * A watch request is set to linger, so the underlying osd
2696 * request won't go away until we unregister it. We retain
2697 * a pointer to the object request during that time (in
2698 * rbd_dev->watch_request), so we'll keep a reference to
2699 * it. We'll drop that reference (below) after we've
2703 rbd_dev->watch_request = obj_request;
2708 /* We have successfully torn down the watch request */
2710 rbd_obj_request_put(rbd_dev->watch_request);
2711 rbd_dev->watch_request = NULL;
2713 /* Cancel the event if we're tearing down, or on error */
2714 ceph_osdc_cancel_event(rbd_dev->watch_event);
2715 rbd_dev->watch_event = NULL;
2717 rbd_obj_request_put(obj_request);
2723 * Synchronous osd object method call. Returns the number of bytes
2724 * returned in the outbound buffer, or a negative error code.
2726 static int rbd_obj_method_sync(struct rbd_device *rbd_dev,
2727 const char *object_name,
2728 const char *class_name,
2729 const char *method_name,
2730 const void *outbound,
2731 size_t outbound_size,
2733 size_t inbound_size)
2735 struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
2736 struct rbd_obj_request *obj_request;
2737 struct page **pages;
2742 * Method calls are ultimately read operations. The result
2743 * should placed into the inbound buffer provided. They
2744 * also supply outbound data--parameters for the object
2745 * method. Currently if this is present it will be a
2748 page_count = (u32)calc_pages_for(0, inbound_size);
2749 pages = ceph_alloc_page_vector(page_count, GFP_KERNEL);
2751 return PTR_ERR(pages);
2754 obj_request = rbd_obj_request_create(object_name, 0, inbound_size,
2759 obj_request->pages = pages;
2760 obj_request->page_count = page_count;
2762 obj_request->osd_req = rbd_osd_req_create(rbd_dev, false, obj_request);
2763 if (!obj_request->osd_req)
2766 osd_req_op_cls_init(obj_request->osd_req, 0, CEPH_OSD_OP_CALL,
2767 class_name, method_name);
2768 if (outbound_size) {
2769 struct ceph_pagelist *pagelist;
2771 pagelist = kmalloc(sizeof (*pagelist), GFP_NOFS);
2775 ceph_pagelist_init(pagelist);
2776 ceph_pagelist_append(pagelist, outbound, outbound_size);
2777 osd_req_op_cls_request_data_pagelist(obj_request->osd_req, 0,
2780 osd_req_op_cls_response_data_pages(obj_request->osd_req, 0,
2781 obj_request->pages, inbound_size,
2783 rbd_osd_req_format_read(obj_request);
2785 ret = rbd_obj_request_submit(osdc, obj_request);
2788 ret = rbd_obj_request_wait(obj_request);
2792 ret = obj_request->result;
2796 rbd_assert(obj_request->xferred < (u64)INT_MAX);
2797 ret = (int)obj_request->xferred;
2798 ceph_copy_from_page_vector(pages, inbound, 0, obj_request->xferred);
2801 rbd_obj_request_put(obj_request);
2803 ceph_release_page_vector(pages, page_count);
2808 static void rbd_request_fn(struct request_queue *q)
2809 __releases(q->queue_lock) __acquires(q->queue_lock)
2811 struct rbd_device *rbd_dev = q->queuedata;
2812 bool read_only = rbd_dev->mapping.read_only;
2816 while ((rq = blk_fetch_request(q))) {
2817 bool write_request = rq_data_dir(rq) == WRITE;
2818 struct rbd_img_request *img_request;
2822 /* Ignore any non-FS requests that filter through. */
2824 if (rq->cmd_type != REQ_TYPE_FS) {
2825 dout("%s: non-fs request type %d\n", __func__,
2826 (int) rq->cmd_type);
2827 __blk_end_request_all(rq, 0);
2831 /* Ignore/skip any zero-length requests */
2833 offset = (u64) blk_rq_pos(rq) << SECTOR_SHIFT;
2834 length = (u64) blk_rq_bytes(rq);
2837 dout("%s: zero-length request\n", __func__);
2838 __blk_end_request_all(rq, 0);
2842 spin_unlock_irq(q->queue_lock);
2844 /* Disallow writes to a read-only device */
2846 if (write_request) {
2850 rbd_assert(rbd_dev->spec->snap_id == CEPH_NOSNAP);
2854 * Quit early if the mapped snapshot no longer
2855 * exists. It's still possible the snapshot will
2856 * have disappeared by the time our request arrives
2857 * at the osd, but there's no sense in sending it if
2860 if (!test_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags)) {
2861 dout("request for non-existent snapshot");
2862 rbd_assert(rbd_dev->spec->snap_id != CEPH_NOSNAP);
2868 if (offset && length > U64_MAX - offset + 1) {
2869 rbd_warn(rbd_dev, "bad request range (%llu~%llu)\n",
2871 goto end_request; /* Shouldn't happen */
2875 img_request = rbd_img_request_create(rbd_dev, offset, length,
2876 write_request, false);
2880 img_request->rq = rq;
2882 result = rbd_img_request_fill(img_request, OBJ_REQUEST_BIO,
2885 result = rbd_img_request_submit(img_request);
2887 rbd_img_request_put(img_request);
2889 spin_lock_irq(q->queue_lock);
2891 rbd_warn(rbd_dev, "%s %llx at %llx result %d\n",
2892 write_request ? "write" : "read",
2893 length, offset, result);
2895 __blk_end_request_all(rq, result);
2901 * a queue callback. Makes sure that we don't create a bio that spans across
2902 * multiple osd objects. One exception would be with a single page bios,
2903 * which we handle later at bio_chain_clone_range()
2905 static int rbd_merge_bvec(struct request_queue *q, struct bvec_merge_data *bmd,
2906 struct bio_vec *bvec)
2908 struct rbd_device *rbd_dev = q->queuedata;
2909 sector_t sector_offset;
2910 sector_t sectors_per_obj;
2911 sector_t obj_sector_offset;
2915 * Find how far into its rbd object the partition-relative
2916 * bio start sector is to offset relative to the enclosing
2919 sector_offset = get_start_sect(bmd->bi_bdev) + bmd->bi_sector;
2920 sectors_per_obj = 1 << (rbd_dev->header.obj_order - SECTOR_SHIFT);
2921 obj_sector_offset = sector_offset & (sectors_per_obj - 1);
2924 * Compute the number of bytes from that offset to the end
2925 * of the object. Account for what's already used by the bio.
2927 ret = (int) (sectors_per_obj - obj_sector_offset) << SECTOR_SHIFT;
2928 if (ret > bmd->bi_size)
2929 ret -= bmd->bi_size;
2934 * Don't send back more than was asked for. And if the bio
2935 * was empty, let the whole thing through because: "Note
2936 * that a block device *must* allow a single page to be
2937 * added to an empty bio."
2939 rbd_assert(bvec->bv_len <= PAGE_SIZE);
2940 if (ret > (int) bvec->bv_len || !bmd->bi_size)
2941 ret = (int) bvec->bv_len;
2946 static void rbd_free_disk(struct rbd_device *rbd_dev)
2948 struct gendisk *disk = rbd_dev->disk;
2953 rbd_dev->disk = NULL;
2954 if (disk->flags & GENHD_FL_UP) {
2957 blk_cleanup_queue(disk->queue);
2962 static int rbd_obj_read_sync(struct rbd_device *rbd_dev,
2963 const char *object_name,
2964 u64 offset, u64 length, void *buf)
2967 struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
2968 struct rbd_obj_request *obj_request;
2969 struct page **pages = NULL;
2974 page_count = (u32) calc_pages_for(offset, length);
2975 pages = ceph_alloc_page_vector(page_count, GFP_KERNEL);
2977 ret = PTR_ERR(pages);
2980 obj_request = rbd_obj_request_create(object_name, offset, length,
2985 obj_request->pages = pages;
2986 obj_request->page_count = page_count;
2988 obj_request->osd_req = rbd_osd_req_create(rbd_dev, false, obj_request);
2989 if (!obj_request->osd_req)
2992 osd_req_op_extent_init(obj_request->osd_req, 0, CEPH_OSD_OP_READ,
2993 offset, length, 0, 0);
2994 osd_req_op_extent_osd_data_pages(obj_request->osd_req, 0,
2996 obj_request->length,
2997 obj_request->offset & ~PAGE_MASK,
2999 rbd_osd_req_format_read(obj_request);
3001 ret = rbd_obj_request_submit(osdc, obj_request);
3004 ret = rbd_obj_request_wait(obj_request);
3008 ret = obj_request->result;
3012 rbd_assert(obj_request->xferred <= (u64) SIZE_MAX);
3013 size = (size_t) obj_request->xferred;
3014 ceph_copy_from_page_vector(pages, buf, 0, size);
3015 rbd_assert(size <= (size_t)INT_MAX);
3019 rbd_obj_request_put(obj_request);
3021 ceph_release_page_vector(pages, page_count);
3027 * Read the complete header for the given rbd device.
3029 * Returns a pointer to a dynamically-allocated buffer containing
3030 * the complete and validated header. Caller can pass the address
3031 * of a variable that will be filled in with the version of the
3032 * header object at the time it was read.
3034 * Returns a pointer-coded errno if a failure occurs.
3036 static struct rbd_image_header_ondisk *
3037 rbd_dev_v1_header_read(struct rbd_device *rbd_dev)
3039 struct rbd_image_header_ondisk *ondisk = NULL;
3046 * The complete header will include an array of its 64-bit
3047 * snapshot ids, followed by the names of those snapshots as
3048 * a contiguous block of NUL-terminated strings. Note that
3049 * the number of snapshots could change by the time we read
3050 * it in, in which case we re-read it.
3057 size = sizeof (*ondisk);
3058 size += snap_count * sizeof (struct rbd_image_snap_ondisk);
3060 ondisk = kmalloc(size, GFP_KERNEL);
3062 return ERR_PTR(-ENOMEM);
3064 ret = rbd_obj_read_sync(rbd_dev, rbd_dev->header_name,
3068 if ((size_t)ret < size) {
3070 rbd_warn(rbd_dev, "short header read (want %zd got %d)",
3074 if (!rbd_dev_ondisk_valid(ondisk)) {
3076 rbd_warn(rbd_dev, "invalid header");
3080 names_size = le64_to_cpu(ondisk->snap_names_len);
3081 want_count = snap_count;
3082 snap_count = le32_to_cpu(ondisk->snap_count);
3083 } while (snap_count != want_count);
3090 return ERR_PTR(ret);
3094 * reload the ondisk the header
3096 static int rbd_read_header(struct rbd_device *rbd_dev,
3097 struct rbd_image_header *header)
3099 struct rbd_image_header_ondisk *ondisk;
3102 ondisk = rbd_dev_v1_header_read(rbd_dev);
3104 return PTR_ERR(ondisk);
3105 ret = rbd_header_from_disk(header, ondisk);
3111 static void rbd_update_mapping_size(struct rbd_device *rbd_dev)
3113 if (rbd_dev->spec->snap_id != CEPH_NOSNAP)
3116 if (rbd_dev->mapping.size != rbd_dev->header.image_size) {
3119 rbd_dev->mapping.size = rbd_dev->header.image_size;
3120 size = (sector_t)rbd_dev->mapping.size / SECTOR_SIZE;
3121 dout("setting size to %llu sectors", (unsigned long long)size);
3122 set_capacity(rbd_dev->disk, size);
3127 * only read the first part of the ondisk header, without the snaps info
3129 static int rbd_dev_v1_refresh(struct rbd_device *rbd_dev)
3132 struct rbd_image_header h;
3134 ret = rbd_read_header(rbd_dev, &h);
3138 down_write(&rbd_dev->header_rwsem);
3140 /* Update image size, and check for resize of mapped image */
3141 rbd_dev->header.image_size = h.image_size;
3142 rbd_update_mapping_size(rbd_dev);
3144 /* rbd_dev->header.object_prefix shouldn't change */
3145 kfree(rbd_dev->header.snap_sizes);
3146 kfree(rbd_dev->header.snap_names);
3147 /* osd requests may still refer to snapc */
3148 ceph_put_snap_context(rbd_dev->header.snapc);
3150 rbd_dev->header.image_size = h.image_size;
3151 rbd_dev->header.snapc = h.snapc;
3152 rbd_dev->header.snap_names = h.snap_names;
3153 rbd_dev->header.snap_sizes = h.snap_sizes;
3154 /* Free the extra copy of the object prefix */
3155 if (strcmp(rbd_dev->header.object_prefix, h.object_prefix))
3156 rbd_warn(rbd_dev, "object prefix changed (ignoring)");
3157 kfree(h.object_prefix);
3159 up_write(&rbd_dev->header_rwsem);
3165 * Clear the rbd device's EXISTS flag if the snapshot it's mapped to
3166 * has disappeared from the (just updated) snapshot context.
3168 static void rbd_exists_validate(struct rbd_device *rbd_dev)
3172 if (!test_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags))
3175 snap_id = rbd_dev->spec->snap_id;
3176 if (snap_id == CEPH_NOSNAP)
3179 if (rbd_dev_snap_index(rbd_dev, snap_id) == BAD_SNAP_INDEX)
3180 clear_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags);
3183 static int rbd_dev_refresh(struct rbd_device *rbd_dev)
3188 rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
3189 image_size = rbd_dev->header.image_size;
3190 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
3191 if (rbd_dev->image_format == 1)
3192 ret = rbd_dev_v1_refresh(rbd_dev);
3194 ret = rbd_dev_v2_refresh(rbd_dev);
3196 /* If it's a mapped snapshot, validate its EXISTS flag */
3198 rbd_exists_validate(rbd_dev);
3199 mutex_unlock(&ctl_mutex);
3201 rbd_warn(rbd_dev, "got notification but failed to "
3202 " update snaps: %d\n", ret);
3203 if (image_size != rbd_dev->header.image_size)
3204 revalidate_disk(rbd_dev->disk);
3209 static int rbd_init_disk(struct rbd_device *rbd_dev)
3211 struct gendisk *disk;
3212 struct request_queue *q;
3215 /* create gendisk info */
3216 disk = alloc_disk(RBD_MINORS_PER_MAJOR);
3220 snprintf(disk->disk_name, sizeof(disk->disk_name), RBD_DRV_NAME "%d",
3222 disk->major = rbd_dev->major;
3223 disk->first_minor = 0;
3224 disk->fops = &rbd_bd_ops;
3225 disk->private_data = rbd_dev;
3227 q = blk_init_queue(rbd_request_fn, &rbd_dev->lock);
3231 /* We use the default size, but let's be explicit about it. */
3232 blk_queue_physical_block_size(q, SECTOR_SIZE);
3234 /* set io sizes to object size */
3235 segment_size = rbd_obj_bytes(&rbd_dev->header);
3236 blk_queue_max_hw_sectors(q, segment_size / SECTOR_SIZE);
3237 blk_queue_max_segment_size(q, segment_size);
3238 blk_queue_io_min(q, segment_size);
3239 blk_queue_io_opt(q, segment_size);
3241 blk_queue_merge_bvec(q, rbd_merge_bvec);
3244 q->queuedata = rbd_dev;
3246 rbd_dev->disk = disk;
3259 static struct rbd_device *dev_to_rbd_dev(struct device *dev)
3261 return container_of(dev, struct rbd_device, dev);
3264 static ssize_t rbd_size_show(struct device *dev,
3265 struct device_attribute *attr, char *buf)
3267 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3269 return sprintf(buf, "%llu\n",
3270 (unsigned long long)rbd_dev->mapping.size);
3274 * Note this shows the features for whatever's mapped, which is not
3275 * necessarily the base image.
3277 static ssize_t rbd_features_show(struct device *dev,
3278 struct device_attribute *attr, char *buf)
3280 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3282 return sprintf(buf, "0x%016llx\n",
3283 (unsigned long long)rbd_dev->mapping.features);
3286 static ssize_t rbd_major_show(struct device *dev,
3287 struct device_attribute *attr, char *buf)
3289 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3292 return sprintf(buf, "%d\n", rbd_dev->major);
3294 return sprintf(buf, "(none)\n");
3298 static ssize_t rbd_client_id_show(struct device *dev,
3299 struct device_attribute *attr, char *buf)
3301 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3303 return sprintf(buf, "client%lld\n",
3304 ceph_client_id(rbd_dev->rbd_client->client));
3307 static ssize_t rbd_pool_show(struct device *dev,
3308 struct device_attribute *attr, char *buf)
3310 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3312 return sprintf(buf, "%s\n", rbd_dev->spec->pool_name);
3315 static ssize_t rbd_pool_id_show(struct device *dev,
3316 struct device_attribute *attr, char *buf)
3318 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3320 return sprintf(buf, "%llu\n",
3321 (unsigned long long) rbd_dev->spec->pool_id);
3324 static ssize_t rbd_name_show(struct device *dev,
3325 struct device_attribute *attr, char *buf)
3327 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3329 if (rbd_dev->spec->image_name)
3330 return sprintf(buf, "%s\n", rbd_dev->spec->image_name);
3332 return sprintf(buf, "(unknown)\n");
3335 static ssize_t rbd_image_id_show(struct device *dev,
3336 struct device_attribute *attr, char *buf)
3338 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3340 return sprintf(buf, "%s\n", rbd_dev->spec->image_id);
3344 * Shows the name of the currently-mapped snapshot (or
3345 * RBD_SNAP_HEAD_NAME for the base image).
3347 static ssize_t rbd_snap_show(struct device *dev,
3348 struct device_attribute *attr,
3351 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3353 return sprintf(buf, "%s\n", rbd_dev->spec->snap_name);
3357 * For an rbd v2 image, shows the pool id, image id, and snapshot id
3358 * for the parent image. If there is no parent, simply shows
3359 * "(no parent image)".
3361 static ssize_t rbd_parent_show(struct device *dev,
3362 struct device_attribute *attr,
3365 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3366 struct rbd_spec *spec = rbd_dev->parent_spec;
3371 return sprintf(buf, "(no parent image)\n");
3373 count = sprintf(bufp, "pool_id %llu\npool_name %s\n",
3374 (unsigned long long) spec->pool_id, spec->pool_name);
3379 count = sprintf(bufp, "image_id %s\nimage_name %s\n", spec->image_id,
3380 spec->image_name ? spec->image_name : "(unknown)");
3385 count = sprintf(bufp, "snap_id %llu\nsnap_name %s\n",
3386 (unsigned long long) spec->snap_id, spec->snap_name);
3391 count = sprintf(bufp, "overlap %llu\n", rbd_dev->parent_overlap);
3396 return (ssize_t) (bufp - buf);
3399 static ssize_t rbd_image_refresh(struct device *dev,
3400 struct device_attribute *attr,
3404 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3407 ret = rbd_dev_refresh(rbd_dev);
3409 return ret < 0 ? ret : size;
3412 static DEVICE_ATTR(size, S_IRUGO, rbd_size_show, NULL);
3413 static DEVICE_ATTR(features, S_IRUGO, rbd_features_show, NULL);
3414 static DEVICE_ATTR(major, S_IRUGO, rbd_major_show, NULL);
3415 static DEVICE_ATTR(client_id, S_IRUGO, rbd_client_id_show, NULL);
3416 static DEVICE_ATTR(pool, S_IRUGO, rbd_pool_show, NULL);
3417 static DEVICE_ATTR(pool_id, S_IRUGO, rbd_pool_id_show, NULL);
3418 static DEVICE_ATTR(name, S_IRUGO, rbd_name_show, NULL);
3419 static DEVICE_ATTR(image_id, S_IRUGO, rbd_image_id_show, NULL);
3420 static DEVICE_ATTR(refresh, S_IWUSR, NULL, rbd_image_refresh);
3421 static DEVICE_ATTR(current_snap, S_IRUGO, rbd_snap_show, NULL);
3422 static DEVICE_ATTR(parent, S_IRUGO, rbd_parent_show, NULL);
3424 static struct attribute *rbd_attrs[] = {
3425 &dev_attr_size.attr,
3426 &dev_attr_features.attr,
3427 &dev_attr_major.attr,
3428 &dev_attr_client_id.attr,
3429 &dev_attr_pool.attr,
3430 &dev_attr_pool_id.attr,
3431 &dev_attr_name.attr,
3432 &dev_attr_image_id.attr,
3433 &dev_attr_current_snap.attr,
3434 &dev_attr_parent.attr,
3435 &dev_attr_refresh.attr,
3439 static struct attribute_group rbd_attr_group = {
3443 static const struct attribute_group *rbd_attr_groups[] = {
3448 static void rbd_sysfs_dev_release(struct device *dev)
3452 static struct device_type rbd_device_type = {
3454 .groups = rbd_attr_groups,
3455 .release = rbd_sysfs_dev_release,
3458 static struct rbd_spec *rbd_spec_get(struct rbd_spec *spec)
3460 kref_get(&spec->kref);
3465 static void rbd_spec_free(struct kref *kref);
3466 static void rbd_spec_put(struct rbd_spec *spec)
3469 kref_put(&spec->kref, rbd_spec_free);
3472 static struct rbd_spec *rbd_spec_alloc(void)
3474 struct rbd_spec *spec;
3476 spec = kzalloc(sizeof (*spec), GFP_KERNEL);
3479 kref_init(&spec->kref);
3484 static void rbd_spec_free(struct kref *kref)
3486 struct rbd_spec *spec = container_of(kref, struct rbd_spec, kref);
3488 kfree(spec->pool_name);
3489 kfree(spec->image_id);
3490 kfree(spec->image_name);
3491 kfree(spec->snap_name);
3495 static struct rbd_device *rbd_dev_create(struct rbd_client *rbdc,
3496 struct rbd_spec *spec)
3498 struct rbd_device *rbd_dev;
3500 rbd_dev = kzalloc(sizeof (*rbd_dev), GFP_KERNEL);
3504 spin_lock_init(&rbd_dev->lock);
3506 INIT_LIST_HEAD(&rbd_dev->node);
3507 init_rwsem(&rbd_dev->header_rwsem);
3509 rbd_dev->spec = spec;
3510 rbd_dev->rbd_client = rbdc;
3512 /* Initialize the layout used for all rbd requests */
3514 rbd_dev->layout.fl_stripe_unit = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER);
3515 rbd_dev->layout.fl_stripe_count = cpu_to_le32(1);
3516 rbd_dev->layout.fl_object_size = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER);
3517 rbd_dev->layout.fl_pg_pool = cpu_to_le32((u32) spec->pool_id);
3522 static void rbd_dev_destroy(struct rbd_device *rbd_dev)
3524 rbd_put_client(rbd_dev->rbd_client);
3525 rbd_spec_put(rbd_dev->spec);
3530 * Get the size and object order for an image snapshot, or if
3531 * snap_id is CEPH_NOSNAP, gets this information for the base
3534 static int _rbd_dev_v2_snap_size(struct rbd_device *rbd_dev, u64 snap_id,
3535 u8 *order, u64 *snap_size)
3537 __le64 snapid = cpu_to_le64(snap_id);
3542 } __attribute__ ((packed)) size_buf = { 0 };
3544 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
3546 &snapid, sizeof (snapid),
3547 &size_buf, sizeof (size_buf));
3548 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
3551 if (ret < sizeof (size_buf))
3555 *order = size_buf.order;
3556 *snap_size = le64_to_cpu(size_buf.size);
3558 dout(" snap_id 0x%016llx order = %u, snap_size = %llu\n",
3559 (unsigned long long)snap_id, (unsigned int)*order,
3560 (unsigned long long)*snap_size);
3565 static int rbd_dev_v2_image_size(struct rbd_device *rbd_dev)
3567 return _rbd_dev_v2_snap_size(rbd_dev, CEPH_NOSNAP,
3568 &rbd_dev->header.obj_order,
3569 &rbd_dev->header.image_size);
3572 static int rbd_dev_v2_object_prefix(struct rbd_device *rbd_dev)
3578 reply_buf = kzalloc(RBD_OBJ_PREFIX_LEN_MAX, GFP_KERNEL);
3582 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
3583 "rbd", "get_object_prefix", NULL, 0,
3584 reply_buf, RBD_OBJ_PREFIX_LEN_MAX);
3585 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
3590 rbd_dev->header.object_prefix = ceph_extract_encoded_string(&p,
3591 p + ret, NULL, GFP_NOIO);
3594 if (IS_ERR(rbd_dev->header.object_prefix)) {
3595 ret = PTR_ERR(rbd_dev->header.object_prefix);
3596 rbd_dev->header.object_prefix = NULL;
3598 dout(" object_prefix = %s\n", rbd_dev->header.object_prefix);
3606 static int _rbd_dev_v2_snap_features(struct rbd_device *rbd_dev, u64 snap_id,
3609 __le64 snapid = cpu_to_le64(snap_id);
3613 } __attribute__ ((packed)) features_buf = { 0 };
3617 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
3618 "rbd", "get_features",
3619 &snapid, sizeof (snapid),
3620 &features_buf, sizeof (features_buf));
3621 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
3624 if (ret < sizeof (features_buf))
3627 incompat = le64_to_cpu(features_buf.incompat);
3628 if (incompat & ~RBD_FEATURES_SUPPORTED)
3631 *snap_features = le64_to_cpu(features_buf.features);
3633 dout(" snap_id 0x%016llx features = 0x%016llx incompat = 0x%016llx\n",
3634 (unsigned long long)snap_id,
3635 (unsigned long long)*snap_features,
3636 (unsigned long long)le64_to_cpu(features_buf.incompat));
3641 static int rbd_dev_v2_features(struct rbd_device *rbd_dev)
3643 return _rbd_dev_v2_snap_features(rbd_dev, CEPH_NOSNAP,
3644 &rbd_dev->header.features);
3647 static int rbd_dev_v2_parent_info(struct rbd_device *rbd_dev)
3649 struct rbd_spec *parent_spec;
3651 void *reply_buf = NULL;
3659 parent_spec = rbd_spec_alloc();
3663 size = sizeof (__le64) + /* pool_id */
3664 sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX + /* image_id */
3665 sizeof (__le64) + /* snap_id */
3666 sizeof (__le64); /* overlap */
3667 reply_buf = kmalloc(size, GFP_KERNEL);
3673 snapid = cpu_to_le64(CEPH_NOSNAP);
3674 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
3675 "rbd", "get_parent",
3676 &snapid, sizeof (snapid),
3678 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
3683 end = reply_buf + ret;
3685 ceph_decode_64_safe(&p, end, parent_spec->pool_id, out_err);
3686 if (parent_spec->pool_id == CEPH_NOPOOL)
3687 goto out; /* No parent? No problem. */
3689 /* The ceph file layout needs to fit pool id in 32 bits */
3692 if (parent_spec->pool_id > (u64)U32_MAX) {
3693 rbd_warn(NULL, "parent pool id too large (%llu > %u)\n",
3694 (unsigned long long)parent_spec->pool_id, U32_MAX);
3698 image_id = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL);
3699 if (IS_ERR(image_id)) {
3700 ret = PTR_ERR(image_id);
3703 parent_spec->image_id = image_id;
3704 ceph_decode_64_safe(&p, end, parent_spec->snap_id, out_err);
3705 ceph_decode_64_safe(&p, end, overlap, out_err);
3707 rbd_dev->parent_overlap = overlap;
3708 rbd_dev->parent_spec = parent_spec;
3709 parent_spec = NULL; /* rbd_dev now owns this */
3714 rbd_spec_put(parent_spec);
3719 static int rbd_dev_v2_striping_info(struct rbd_device *rbd_dev)
3723 __le64 stripe_count;
3724 } __attribute__ ((packed)) striping_info_buf = { 0 };
3725 size_t size = sizeof (striping_info_buf);
3732 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
3733 "rbd", "get_stripe_unit_count", NULL, 0,
3734 (char *)&striping_info_buf, size);
3735 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
3742 * We don't actually support the "fancy striping" feature
3743 * (STRIPINGV2) yet, but if the striping sizes are the
3744 * defaults the behavior is the same as before. So find
3745 * out, and only fail if the image has non-default values.
3748 obj_size = (u64)1 << rbd_dev->header.obj_order;
3749 p = &striping_info_buf;
3750 stripe_unit = ceph_decode_64(&p);
3751 if (stripe_unit != obj_size) {
3752 rbd_warn(rbd_dev, "unsupported stripe unit "
3753 "(got %llu want %llu)",
3754 stripe_unit, obj_size);
3757 stripe_count = ceph_decode_64(&p);
3758 if (stripe_count != 1) {
3759 rbd_warn(rbd_dev, "unsupported stripe count "
3760 "(got %llu want 1)", stripe_count);
3763 rbd_dev->header.stripe_unit = stripe_unit;
3764 rbd_dev->header.stripe_count = stripe_count;
3769 static char *rbd_dev_image_name(struct rbd_device *rbd_dev)
3771 size_t image_id_size;
3776 void *reply_buf = NULL;
3778 char *image_name = NULL;
3781 rbd_assert(!rbd_dev->spec->image_name);
3783 len = strlen(rbd_dev->spec->image_id);
3784 image_id_size = sizeof (__le32) + len;
3785 image_id = kmalloc(image_id_size, GFP_KERNEL);
3790 end = image_id + image_id_size;
3791 ceph_encode_string(&p, end, rbd_dev->spec->image_id, (u32)len);
3793 size = sizeof (__le32) + RBD_IMAGE_NAME_LEN_MAX;
3794 reply_buf = kmalloc(size, GFP_KERNEL);
3798 ret = rbd_obj_method_sync(rbd_dev, RBD_DIRECTORY,
3799 "rbd", "dir_get_name",
3800 image_id, image_id_size,
3805 end = reply_buf + ret;
3807 image_name = ceph_extract_encoded_string(&p, end, &len, GFP_KERNEL);
3808 if (IS_ERR(image_name))
3811 dout("%s: name is %s len is %zd\n", __func__, image_name, len);
3819 static u64 rbd_v1_snap_id_by_name(struct rbd_device *rbd_dev, const char *name)
3821 struct ceph_snap_context *snapc = rbd_dev->header.snapc;
3822 const char *snap_name;
3825 /* Skip over names until we find the one we are looking for */
3827 snap_name = rbd_dev->header.snap_names;
3828 while (which < snapc->num_snaps) {
3829 if (!strcmp(name, snap_name))
3830 return snapc->snaps[which];
3831 snap_name += strlen(snap_name) + 1;
3837 static u64 rbd_v2_snap_id_by_name(struct rbd_device *rbd_dev, const char *name)
3839 struct ceph_snap_context *snapc = rbd_dev->header.snapc;
3844 for (which = 0; !found && which < snapc->num_snaps; which++) {
3845 const char *snap_name;
3847 snap_id = snapc->snaps[which];
3848 snap_name = rbd_dev_v2_snap_name(rbd_dev, snap_id);
3849 if (IS_ERR(snap_name))
3851 found = !strcmp(name, snap_name);
3854 return found ? snap_id : CEPH_NOSNAP;
3858 * Assumes name is never RBD_SNAP_HEAD_NAME; returns CEPH_NOSNAP if
3859 * no snapshot by that name is found, or if an error occurs.
3861 static u64 rbd_snap_id_by_name(struct rbd_device *rbd_dev, const char *name)
3863 if (rbd_dev->image_format == 1)
3864 return rbd_v1_snap_id_by_name(rbd_dev, name);
3866 return rbd_v2_snap_id_by_name(rbd_dev, name);
3870 * When an rbd image has a parent image, it is identified by the
3871 * pool, image, and snapshot ids (not names). This function fills
3872 * in the names for those ids. (It's OK if we can't figure out the
3873 * name for an image id, but the pool and snapshot ids should always
3874 * exist and have names.) All names in an rbd spec are dynamically
3877 * When an image being mapped (not a parent) is probed, we have the
3878 * pool name and pool id, image name and image id, and the snapshot
3879 * name. The only thing we're missing is the snapshot id.
3881 static int rbd_dev_spec_update(struct rbd_device *rbd_dev)
3883 struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
3884 struct rbd_spec *spec = rbd_dev->spec;
3885 const char *pool_name;
3886 const char *image_name;
3887 const char *snap_name;
3891 * An image being mapped will have the pool name (etc.), but
3892 * we need to look up the snapshot id.
3894 if (spec->pool_name) {
3895 if (strcmp(spec->snap_name, RBD_SNAP_HEAD_NAME)) {
3898 snap_id = rbd_snap_id_by_name(rbd_dev, spec->snap_name);
3899 if (snap_id == CEPH_NOSNAP)
3901 spec->snap_id = snap_id;
3903 spec->snap_id = CEPH_NOSNAP;
3909 /* Get the pool name; we have to make our own copy of this */
3911 pool_name = ceph_pg_pool_name_by_id(osdc->osdmap, spec->pool_id);
3913 rbd_warn(rbd_dev, "no pool with id %llu", spec->pool_id);
3916 pool_name = kstrdup(pool_name, GFP_KERNEL);
3920 /* Fetch the image name; tolerate failure here */
3922 image_name = rbd_dev_image_name(rbd_dev);
3924 rbd_warn(rbd_dev, "unable to get image name");
3926 /* Look up the snapshot name, and make a copy */
3928 snap_name = rbd_snap_name(rbd_dev, spec->snap_id);
3934 spec->pool_name = pool_name;
3935 spec->image_name = image_name;
3936 spec->snap_name = snap_name;
3946 static int rbd_dev_v2_snap_context(struct rbd_device *rbd_dev)
3955 struct ceph_snap_context *snapc;
3959 * We'll need room for the seq value (maximum snapshot id),
3960 * snapshot count, and array of that many snapshot ids.
3961 * For now we have a fixed upper limit on the number we're
3962 * prepared to receive.
3964 size = sizeof (__le64) + sizeof (__le32) +
3965 RBD_MAX_SNAP_COUNT * sizeof (__le64);
3966 reply_buf = kzalloc(size, GFP_KERNEL);
3970 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
3971 "rbd", "get_snapcontext", NULL, 0,
3973 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
3978 end = reply_buf + ret;
3980 ceph_decode_64_safe(&p, end, seq, out);
3981 ceph_decode_32_safe(&p, end, snap_count, out);
3984 * Make sure the reported number of snapshot ids wouldn't go
3985 * beyond the end of our buffer. But before checking that,
3986 * make sure the computed size of the snapshot context we
3987 * allocate is representable in a size_t.
3989 if (snap_count > (SIZE_MAX - sizeof (struct ceph_snap_context))
3994 if (!ceph_has_room(&p, end, snap_count * sizeof (__le64)))
3998 snapc = ceph_create_snap_context(snap_count, GFP_KERNEL);
4004 for (i = 0; i < snap_count; i++)
4005 snapc->snaps[i] = ceph_decode_64(&p);
4007 rbd_dev->header.snapc = snapc;
4009 dout(" snap context seq = %llu, snap_count = %u\n",
4010 (unsigned long long)seq, (unsigned int)snap_count);
4017 static const char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev,
4028 size = sizeof (__le32) + RBD_MAX_SNAP_NAME_LEN;
4029 reply_buf = kmalloc(size, GFP_KERNEL);
4031 return ERR_PTR(-ENOMEM);
4033 snapid = cpu_to_le64(snap_id);
4034 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
4035 "rbd", "get_snapshot_name",
4036 &snapid, sizeof (snapid),
4038 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
4040 snap_name = ERR_PTR(ret);
4045 end = reply_buf + ret;
4046 snap_name = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL);
4047 if (IS_ERR(snap_name))
4050 dout(" snap_id 0x%016llx snap_name = %s\n",
4051 (unsigned long long)snap_id, snap_name);
4058 static int rbd_dev_v2_refresh(struct rbd_device *rbd_dev)
4062 down_write(&rbd_dev->header_rwsem);
4064 ret = rbd_dev_v2_image_size(rbd_dev);
4067 rbd_update_mapping_size(rbd_dev);
4069 ret = rbd_dev_v2_snap_context(rbd_dev);
4070 dout("rbd_dev_v2_snap_context returned %d\n", ret);
4074 up_write(&rbd_dev->header_rwsem);
4079 static int rbd_bus_add_dev(struct rbd_device *rbd_dev)
4084 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
4086 dev = &rbd_dev->dev;
4087 dev->bus = &rbd_bus_type;
4088 dev->type = &rbd_device_type;
4089 dev->parent = &rbd_root_dev;
4090 dev->release = rbd_dev_device_release;
4091 dev_set_name(dev, "%d", rbd_dev->dev_id);
4092 ret = device_register(dev);
4094 mutex_unlock(&ctl_mutex);
4099 static void rbd_bus_del_dev(struct rbd_device *rbd_dev)
4101 device_unregister(&rbd_dev->dev);
4104 static atomic64_t rbd_dev_id_max = ATOMIC64_INIT(0);
4107 * Get a unique rbd identifier for the given new rbd_dev, and add
4108 * the rbd_dev to the global list. The minimum rbd id is 1.
4110 static void rbd_dev_id_get(struct rbd_device *rbd_dev)
4112 rbd_dev->dev_id = atomic64_inc_return(&rbd_dev_id_max);
4114 spin_lock(&rbd_dev_list_lock);
4115 list_add_tail(&rbd_dev->node, &rbd_dev_list);
4116 spin_unlock(&rbd_dev_list_lock);
4117 dout("rbd_dev %p given dev id %llu\n", rbd_dev,
4118 (unsigned long long) rbd_dev->dev_id);
4122 * Remove an rbd_dev from the global list, and record that its
4123 * identifier is no longer in use.
4125 static void rbd_dev_id_put(struct rbd_device *rbd_dev)
4127 struct list_head *tmp;
4128 int rbd_id = rbd_dev->dev_id;
4131 rbd_assert(rbd_id > 0);
4133 dout("rbd_dev %p released dev id %llu\n", rbd_dev,
4134 (unsigned long long) rbd_dev->dev_id);
4135 spin_lock(&rbd_dev_list_lock);
4136 list_del_init(&rbd_dev->node);
4139 * If the id being "put" is not the current maximum, there
4140 * is nothing special we need to do.
4142 if (rbd_id != atomic64_read(&rbd_dev_id_max)) {
4143 spin_unlock(&rbd_dev_list_lock);
4148 * We need to update the current maximum id. Search the
4149 * list to find out what it is. We're more likely to find
4150 * the maximum at the end, so search the list backward.
4153 list_for_each_prev(tmp, &rbd_dev_list) {
4154 struct rbd_device *rbd_dev;
4156 rbd_dev = list_entry(tmp, struct rbd_device, node);
4157 if (rbd_dev->dev_id > max_id)
4158 max_id = rbd_dev->dev_id;
4160 spin_unlock(&rbd_dev_list_lock);
4163 * The max id could have been updated by rbd_dev_id_get(), in
4164 * which case it now accurately reflects the new maximum.
4165 * Be careful not to overwrite the maximum value in that
4168 atomic64_cmpxchg(&rbd_dev_id_max, rbd_id, max_id);
4169 dout(" max dev id has been reset\n");
4173 * Skips over white space at *buf, and updates *buf to point to the
4174 * first found non-space character (if any). Returns the length of
4175 * the token (string of non-white space characters) found. Note
4176 * that *buf must be terminated with '\0'.
4178 static inline size_t next_token(const char **buf)
4181 * These are the characters that produce nonzero for
4182 * isspace() in the "C" and "POSIX" locales.
4184 const char *spaces = " \f\n\r\t\v";
4186 *buf += strspn(*buf, spaces); /* Find start of token */
4188 return strcspn(*buf, spaces); /* Return token length */
4192 * Finds the next token in *buf, and if the provided token buffer is
4193 * big enough, copies the found token into it. The result, if
4194 * copied, is guaranteed to be terminated with '\0'. Note that *buf
4195 * must be terminated with '\0' on entry.
4197 * Returns the length of the token found (not including the '\0').
4198 * Return value will be 0 if no token is found, and it will be >=
4199 * token_size if the token would not fit.
4201 * The *buf pointer will be updated to point beyond the end of the
4202 * found token. Note that this occurs even if the token buffer is
4203 * too small to hold it.
4205 static inline size_t copy_token(const char **buf,
4211 len = next_token(buf);
4212 if (len < token_size) {
4213 memcpy(token, *buf, len);
4214 *(token + len) = '\0';
4222 * Finds the next token in *buf, dynamically allocates a buffer big
4223 * enough to hold a copy of it, and copies the token into the new
4224 * buffer. The copy is guaranteed to be terminated with '\0'. Note
4225 * that a duplicate buffer is created even for a zero-length token.
4227 * Returns a pointer to the newly-allocated duplicate, or a null
4228 * pointer if memory for the duplicate was not available. If
4229 * the lenp argument is a non-null pointer, the length of the token
4230 * (not including the '\0') is returned in *lenp.
4232 * If successful, the *buf pointer will be updated to point beyond
4233 * the end of the found token.
4235 * Note: uses GFP_KERNEL for allocation.
4237 static inline char *dup_token(const char **buf, size_t *lenp)
4242 len = next_token(buf);
4243 dup = kmemdup(*buf, len + 1, GFP_KERNEL);
4246 *(dup + len) = '\0';
4256 * Parse the options provided for an "rbd add" (i.e., rbd image
4257 * mapping) request. These arrive via a write to /sys/bus/rbd/add,
4258 * and the data written is passed here via a NUL-terminated buffer.
4259 * Returns 0 if successful or an error code otherwise.
4261 * The information extracted from these options is recorded in
4262 * the other parameters which return dynamically-allocated
4265 * The address of a pointer that will refer to a ceph options
4266 * structure. Caller must release the returned pointer using
4267 * ceph_destroy_options() when it is no longer needed.
4269 * Address of an rbd options pointer. Fully initialized by
4270 * this function; caller must release with kfree().
4272 * Address of an rbd image specification pointer. Fully
4273 * initialized by this function based on parsed options.
4274 * Caller must release with rbd_spec_put().
4276 * The options passed take this form:
4277 * <mon_addrs> <options> <pool_name> <image_name> [<snap_id>]
4280 * A comma-separated list of one or more monitor addresses.
4281 * A monitor address is an ip address, optionally followed
4282 * by a port number (separated by a colon).
4283 * I.e.: ip1[:port1][,ip2[:port2]...]
4285 * A comma-separated list of ceph and/or rbd options.
4287 * The name of the rados pool containing the rbd image.
4289 * The name of the image in that pool to map.
4291 * An optional snapshot id. If provided, the mapping will
4292 * present data from the image at the time that snapshot was
4293 * created. The image head is used if no snapshot id is
4294 * provided. Snapshot mappings are always read-only.
4296 static int rbd_add_parse_args(const char *buf,
4297 struct ceph_options **ceph_opts,
4298 struct rbd_options **opts,
4299 struct rbd_spec **rbd_spec)
4303 const char *mon_addrs;
4305 size_t mon_addrs_size;
4306 struct rbd_spec *spec = NULL;
4307 struct rbd_options *rbd_opts = NULL;
4308 struct ceph_options *copts;
4311 /* The first four tokens are required */
4313 len = next_token(&buf);
4315 rbd_warn(NULL, "no monitor address(es) provided");
4319 mon_addrs_size = len + 1;
4323 options = dup_token(&buf, NULL);
4327 rbd_warn(NULL, "no options provided");
4331 spec = rbd_spec_alloc();
4335 spec->pool_name = dup_token(&buf, NULL);
4336 if (!spec->pool_name)
4338 if (!*spec->pool_name) {
4339 rbd_warn(NULL, "no pool name provided");
4343 spec->image_name = dup_token(&buf, NULL);
4344 if (!spec->image_name)
4346 if (!*spec->image_name) {
4347 rbd_warn(NULL, "no image name provided");
4352 * Snapshot name is optional; default is to use "-"
4353 * (indicating the head/no snapshot).
4355 len = next_token(&buf);
4357 buf = RBD_SNAP_HEAD_NAME; /* No snapshot supplied */
4358 len = sizeof (RBD_SNAP_HEAD_NAME) - 1;
4359 } else if (len > RBD_MAX_SNAP_NAME_LEN) {
4360 ret = -ENAMETOOLONG;
4363 snap_name = kmemdup(buf, len + 1, GFP_KERNEL);
4366 *(snap_name + len) = '\0';
4367 spec->snap_name = snap_name;
4369 /* Initialize all rbd options to the defaults */
4371 rbd_opts = kzalloc(sizeof (*rbd_opts), GFP_KERNEL);
4375 rbd_opts->read_only = RBD_READ_ONLY_DEFAULT;
4377 copts = ceph_parse_options(options, mon_addrs,
4378 mon_addrs + mon_addrs_size - 1,
4379 parse_rbd_opts_token, rbd_opts);
4380 if (IS_ERR(copts)) {
4381 ret = PTR_ERR(copts);
4402 * An rbd format 2 image has a unique identifier, distinct from the
4403 * name given to it by the user. Internally, that identifier is
4404 * what's used to specify the names of objects related to the image.
4406 * A special "rbd id" object is used to map an rbd image name to its
4407 * id. If that object doesn't exist, then there is no v2 rbd image
4408 * with the supplied name.
4410 * This function will record the given rbd_dev's image_id field if
4411 * it can be determined, and in that case will return 0. If any
4412 * errors occur a negative errno will be returned and the rbd_dev's
4413 * image_id field will be unchanged (and should be NULL).
4415 static int rbd_dev_image_id(struct rbd_device *rbd_dev)
4424 * When probing a parent image, the image id is already
4425 * known (and the image name likely is not). There's no
4426 * need to fetch the image id again in this case. We
4427 * do still need to set the image format though.
4429 if (rbd_dev->spec->image_id) {
4430 rbd_dev->image_format = *rbd_dev->spec->image_id ? 2 : 1;
4436 * First, see if the format 2 image id file exists, and if
4437 * so, get the image's persistent id from it.
4439 size = sizeof (RBD_ID_PREFIX) + strlen(rbd_dev->spec->image_name);
4440 object_name = kmalloc(size, GFP_NOIO);
4443 sprintf(object_name, "%s%s", RBD_ID_PREFIX, rbd_dev->spec->image_name);
4444 dout("rbd id object name is %s\n", object_name);
4446 /* Response will be an encoded string, which includes a length */
4448 size = sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX;
4449 response = kzalloc(size, GFP_NOIO);
4455 /* If it doesn't exist we'll assume it's a format 1 image */
4457 ret = rbd_obj_method_sync(rbd_dev, object_name,
4458 "rbd", "get_id", NULL, 0,
4459 response, RBD_IMAGE_ID_LEN_MAX);
4460 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
4461 if (ret == -ENOENT) {
4462 image_id = kstrdup("", GFP_KERNEL);
4463 ret = image_id ? 0 : -ENOMEM;
4465 rbd_dev->image_format = 1;
4466 } else if (ret > sizeof (__le32)) {
4469 image_id = ceph_extract_encoded_string(&p, p + ret,
4471 ret = IS_ERR(image_id) ? PTR_ERR(image_id) : 0;
4473 rbd_dev->image_format = 2;
4479 rbd_dev->spec->image_id = image_id;
4480 dout("image_id is %s\n", image_id);
4489 /* Undo whatever state changes are made by v1 or v2 image probe */
4491 static void rbd_dev_unprobe(struct rbd_device *rbd_dev)
4493 struct rbd_image_header *header;
4495 rbd_dev_remove_parent(rbd_dev);
4496 rbd_spec_put(rbd_dev->parent_spec);
4497 rbd_dev->parent_spec = NULL;
4498 rbd_dev->parent_overlap = 0;
4500 /* Free dynamic fields from the header, then zero it out */
4502 header = &rbd_dev->header;
4503 ceph_put_snap_context(header->snapc);
4504 kfree(header->snap_sizes);
4505 kfree(header->snap_names);
4506 kfree(header->object_prefix);
4507 memset(header, 0, sizeof (*header));
4510 static int rbd_dev_v1_probe(struct rbd_device *rbd_dev)
4514 /* Populate rbd image metadata */
4516 ret = rbd_read_header(rbd_dev, &rbd_dev->header);
4520 /* Version 1 images have no parent (no layering) */
4522 rbd_dev->parent_spec = NULL;
4523 rbd_dev->parent_overlap = 0;
4525 dout("discovered version 1 image, header name is %s\n",
4526 rbd_dev->header_name);
4531 kfree(rbd_dev->header_name);
4532 rbd_dev->header_name = NULL;
4533 kfree(rbd_dev->spec->image_id);
4534 rbd_dev->spec->image_id = NULL;
4539 static int rbd_dev_v2_probe(struct rbd_device *rbd_dev)
4543 ret = rbd_dev_v2_image_size(rbd_dev);
4547 /* Get the object prefix (a.k.a. block_name) for the image */
4549 ret = rbd_dev_v2_object_prefix(rbd_dev);
4553 /* Get the and check features for the image */
4555 ret = rbd_dev_v2_features(rbd_dev);
4559 /* If the image supports layering, get the parent info */
4561 if (rbd_dev->header.features & RBD_FEATURE_LAYERING) {
4562 ret = rbd_dev_v2_parent_info(rbd_dev);
4567 * Don't print a warning for parent images. We can
4568 * tell this point because we won't know its pool
4569 * name yet (just its pool id).
4571 if (rbd_dev->spec->pool_name)
4572 rbd_warn(rbd_dev, "WARNING: kernel layering "
4573 "is EXPERIMENTAL!");
4576 /* If the image supports fancy striping, get its parameters */
4578 if (rbd_dev->header.features & RBD_FEATURE_STRIPINGV2) {
4579 ret = rbd_dev_v2_striping_info(rbd_dev);
4584 /* crypto and compression type aren't (yet) supported for v2 images */
4586 rbd_dev->header.crypt_type = 0;
4587 rbd_dev->header.comp_type = 0;
4589 /* Get the snapshot context, plus the header version */
4591 ret = rbd_dev_v2_snap_context(rbd_dev);
4595 dout("discovered version 2 image, header name is %s\n",
4596 rbd_dev->header_name);
4600 rbd_dev->parent_overlap = 0;
4601 rbd_spec_put(rbd_dev->parent_spec);
4602 rbd_dev->parent_spec = NULL;
4603 kfree(rbd_dev->header_name);
4604 rbd_dev->header_name = NULL;
4605 kfree(rbd_dev->header.object_prefix);
4606 rbd_dev->header.object_prefix = NULL;
4611 static int rbd_dev_probe_parent(struct rbd_device *rbd_dev)
4613 struct rbd_device *parent = NULL;
4614 struct rbd_spec *parent_spec;
4615 struct rbd_client *rbdc;
4618 if (!rbd_dev->parent_spec)
4621 * We need to pass a reference to the client and the parent
4622 * spec when creating the parent rbd_dev. Images related by
4623 * parent/child relationships always share both.
4625 parent_spec = rbd_spec_get(rbd_dev->parent_spec);
4626 rbdc = __rbd_get_client(rbd_dev->rbd_client);
4629 parent = rbd_dev_create(rbdc, parent_spec);
4633 ret = rbd_dev_image_probe(parent);
4636 rbd_dev->parent = parent;
4641 rbd_spec_put(rbd_dev->parent_spec);
4642 kfree(rbd_dev->header_name);
4643 rbd_dev_destroy(parent);
4645 rbd_put_client(rbdc);
4646 rbd_spec_put(parent_spec);
4652 static int rbd_dev_device_setup(struct rbd_device *rbd_dev)
4656 ret = rbd_dev_mapping_set(rbd_dev);
4660 /* generate unique id: find highest unique id, add one */
4661 rbd_dev_id_get(rbd_dev);
4663 /* Fill in the device name, now that we have its id. */
4664 BUILD_BUG_ON(DEV_NAME_LEN
4665 < sizeof (RBD_DRV_NAME) + MAX_INT_FORMAT_WIDTH);
4666 sprintf(rbd_dev->name, "%s%d", RBD_DRV_NAME, rbd_dev->dev_id);
4668 /* Get our block major device number. */
4670 ret = register_blkdev(0, rbd_dev->name);
4673 rbd_dev->major = ret;
4675 /* Set up the blkdev mapping. */
4677 ret = rbd_init_disk(rbd_dev);
4679 goto err_out_blkdev;
4681 ret = rbd_bus_add_dev(rbd_dev);
4685 /* Everything's ready. Announce the disk to the world. */
4687 set_capacity(rbd_dev->disk, rbd_dev->mapping.size / SECTOR_SIZE);
4688 set_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags);
4689 add_disk(rbd_dev->disk);
4691 pr_info("%s: added with size 0x%llx\n", rbd_dev->disk->disk_name,
4692 (unsigned long long) rbd_dev->mapping.size);
4697 rbd_free_disk(rbd_dev);
4699 unregister_blkdev(rbd_dev->major, rbd_dev->name);
4701 rbd_dev_id_put(rbd_dev);
4702 rbd_dev_mapping_clear(rbd_dev);
4707 static int rbd_dev_header_name(struct rbd_device *rbd_dev)
4709 struct rbd_spec *spec = rbd_dev->spec;
4712 /* Record the header object name for this rbd image. */
4714 rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
4716 if (rbd_dev->image_format == 1)
4717 size = strlen(spec->image_name) + sizeof (RBD_SUFFIX);
4719 size = sizeof (RBD_HEADER_PREFIX) + strlen(spec->image_id);
4721 rbd_dev->header_name = kmalloc(size, GFP_KERNEL);
4722 if (!rbd_dev->header_name)
4725 if (rbd_dev->image_format == 1)
4726 sprintf(rbd_dev->header_name, "%s%s",
4727 spec->image_name, RBD_SUFFIX);
4729 sprintf(rbd_dev->header_name, "%s%s",
4730 RBD_HEADER_PREFIX, spec->image_id);
4734 static void rbd_dev_image_release(struct rbd_device *rbd_dev)
4738 rbd_dev_unprobe(rbd_dev);
4739 ret = rbd_dev_header_watch_sync(rbd_dev, 0);
4741 rbd_warn(rbd_dev, "failed to cancel watch event (%d)\n", ret);
4742 kfree(rbd_dev->header_name);
4743 rbd_dev->header_name = NULL;
4744 rbd_dev->image_format = 0;
4745 kfree(rbd_dev->spec->image_id);
4746 rbd_dev->spec->image_id = NULL;
4748 rbd_dev_destroy(rbd_dev);
4752 * Probe for the existence of the header object for the given rbd
4753 * device. For format 2 images this includes determining the image
4756 static int rbd_dev_image_probe(struct rbd_device *rbd_dev)
4762 * Get the id from the image id object. If it's not a
4763 * format 2 image, we'll get ENOENT back, and we'll assume
4764 * it's a format 1 image.
4766 ret = rbd_dev_image_id(rbd_dev);
4769 rbd_assert(rbd_dev->spec->image_id);
4770 rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
4772 ret = rbd_dev_header_name(rbd_dev);
4774 goto err_out_format;
4776 ret = rbd_dev_header_watch_sync(rbd_dev, 1);
4778 goto out_header_name;
4780 if (rbd_dev->image_format == 1)
4781 ret = rbd_dev_v1_probe(rbd_dev);
4783 ret = rbd_dev_v2_probe(rbd_dev);
4787 ret = rbd_dev_spec_update(rbd_dev);
4791 ret = rbd_dev_probe_parent(rbd_dev);
4796 rbd_dev_unprobe(rbd_dev);
4798 tmp = rbd_dev_header_watch_sync(rbd_dev, 0);
4800 rbd_warn(rbd_dev, "unable to tear down watch request\n");
4802 kfree(rbd_dev->header_name);
4803 rbd_dev->header_name = NULL;
4805 rbd_dev->image_format = 0;
4806 kfree(rbd_dev->spec->image_id);
4807 rbd_dev->spec->image_id = NULL;
4809 dout("probe failed, returning %d\n", ret);
4814 static ssize_t rbd_add(struct bus_type *bus,
4818 struct rbd_device *rbd_dev = NULL;
4819 struct ceph_options *ceph_opts = NULL;
4820 struct rbd_options *rbd_opts = NULL;
4821 struct rbd_spec *spec = NULL;
4822 struct rbd_client *rbdc;
4823 struct ceph_osd_client *osdc;
4826 if (!try_module_get(THIS_MODULE))
4829 /* parse add command */
4830 rc = rbd_add_parse_args(buf, &ceph_opts, &rbd_opts, &spec);
4832 goto err_out_module;
4834 rbdc = rbd_get_client(ceph_opts);
4839 ceph_opts = NULL; /* rbd_dev client now owns this */
4842 osdc = &rbdc->client->osdc;
4843 rc = ceph_pg_poolid_by_name(osdc->osdmap, spec->pool_name);
4845 goto err_out_client;
4846 spec->pool_id = (u64)rc;
4848 /* The ceph file layout needs to fit pool id in 32 bits */
4850 if (spec->pool_id > (u64)U32_MAX) {
4851 rbd_warn(NULL, "pool id too large (%llu > %u)\n",
4852 (unsigned long long)spec->pool_id, U32_MAX);
4854 goto err_out_client;
4857 rbd_dev = rbd_dev_create(rbdc, spec);
4859 goto err_out_client;
4860 rbdc = NULL; /* rbd_dev now owns this */
4861 spec = NULL; /* rbd_dev now owns this */
4863 rbd_dev->mapping.read_only = rbd_opts->read_only;
4865 rbd_opts = NULL; /* done with this */
4867 rc = rbd_dev_image_probe(rbd_dev);
4869 goto err_out_rbd_dev;
4871 rc = rbd_dev_device_setup(rbd_dev);
4875 rbd_dev_image_release(rbd_dev);
4877 rbd_dev_destroy(rbd_dev);
4879 rbd_put_client(rbdc);
4882 ceph_destroy_options(ceph_opts);
4886 module_put(THIS_MODULE);
4888 dout("Error adding device %s\n", buf);
4893 static struct rbd_device *__rbd_get_dev(unsigned long dev_id)
4895 struct list_head *tmp;
4896 struct rbd_device *rbd_dev;
4898 spin_lock(&rbd_dev_list_lock);
4899 list_for_each(tmp, &rbd_dev_list) {
4900 rbd_dev = list_entry(tmp, struct rbd_device, node);
4901 if (rbd_dev->dev_id == dev_id) {
4902 spin_unlock(&rbd_dev_list_lock);
4906 spin_unlock(&rbd_dev_list_lock);
4910 static void rbd_dev_device_release(struct device *dev)
4912 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
4914 rbd_free_disk(rbd_dev);
4915 clear_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags);
4916 rbd_dev_clear_mapping(rbd_dev);
4917 unregister_blkdev(rbd_dev->major, rbd_dev->name);
4919 rbd_dev_id_put(rbd_dev);
4920 rbd_dev_mapping_clear(rbd_dev);
4923 static void rbd_dev_remove_parent(struct rbd_device *rbd_dev)
4925 while (rbd_dev->parent) {
4926 struct rbd_device *first = rbd_dev;
4927 struct rbd_device *second = first->parent;
4928 struct rbd_device *third;
4931 * Follow to the parent with no grandparent and
4934 while (second && (third = second->parent)) {
4939 rbd_dev_image_release(second);
4940 first->parent = NULL;
4941 first->parent_overlap = 0;
4943 rbd_assert(first->parent_spec);
4944 rbd_spec_put(first->parent_spec);
4945 first->parent_spec = NULL;
4949 static ssize_t rbd_remove(struct bus_type *bus,
4953 struct rbd_device *rbd_dev = NULL;
4958 ret = strict_strtoul(buf, 10, &ul);
4962 /* convert to int; abort if we lost anything in the conversion */
4963 target_id = (int) ul;
4964 if (target_id != ul)
4967 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
4969 rbd_dev = __rbd_get_dev(target_id);
4975 spin_lock_irq(&rbd_dev->lock);
4976 if (rbd_dev->open_count)
4979 set_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags);
4980 spin_unlock_irq(&rbd_dev->lock);
4984 rbd_bus_del_dev(rbd_dev);
4985 rbd_dev_image_release(rbd_dev);
4986 module_put(THIS_MODULE);
4988 mutex_unlock(&ctl_mutex);
4994 * create control files in sysfs
4997 static int rbd_sysfs_init(void)
5001 ret = device_register(&rbd_root_dev);
5005 ret = bus_register(&rbd_bus_type);
5007 device_unregister(&rbd_root_dev);
5012 static void rbd_sysfs_cleanup(void)
5014 bus_unregister(&rbd_bus_type);
5015 device_unregister(&rbd_root_dev);
5018 static int rbd_slab_init(void)
5020 rbd_assert(!rbd_img_request_cache);
5021 rbd_img_request_cache = kmem_cache_create("rbd_img_request",
5022 sizeof (struct rbd_img_request),
5023 __alignof__(struct rbd_img_request),
5025 if (!rbd_img_request_cache)
5028 rbd_assert(!rbd_obj_request_cache);
5029 rbd_obj_request_cache = kmem_cache_create("rbd_obj_request",
5030 sizeof (struct rbd_obj_request),
5031 __alignof__(struct rbd_obj_request),
5033 if (!rbd_obj_request_cache)
5036 rbd_assert(!rbd_segment_name_cache);
5037 rbd_segment_name_cache = kmem_cache_create("rbd_segment_name",
5038 MAX_OBJ_NAME_SIZE + 1, 1, 0, NULL);
5039 if (rbd_segment_name_cache)
5042 if (rbd_obj_request_cache) {
5043 kmem_cache_destroy(rbd_obj_request_cache);
5044 rbd_obj_request_cache = NULL;
5047 kmem_cache_destroy(rbd_img_request_cache);
5048 rbd_img_request_cache = NULL;
5053 static void rbd_slab_exit(void)
5055 rbd_assert(rbd_segment_name_cache);
5056 kmem_cache_destroy(rbd_segment_name_cache);
5057 rbd_segment_name_cache = NULL;
5059 rbd_assert(rbd_obj_request_cache);
5060 kmem_cache_destroy(rbd_obj_request_cache);
5061 rbd_obj_request_cache = NULL;
5063 rbd_assert(rbd_img_request_cache);
5064 kmem_cache_destroy(rbd_img_request_cache);
5065 rbd_img_request_cache = NULL;
5068 static int __init rbd_init(void)
5072 if (!libceph_compatible(NULL)) {
5073 rbd_warn(NULL, "libceph incompatibility (quitting)");
5077 rc = rbd_slab_init();
5080 rc = rbd_sysfs_init();
5084 pr_info("loaded " RBD_DRV_NAME_LONG "\n");
5089 static void __exit rbd_exit(void)
5091 rbd_sysfs_cleanup();
5095 module_init(rbd_init);
5096 module_exit(rbd_exit);
5098 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
5099 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
5100 MODULE_DESCRIPTION("rados block device");
5102 /* following authorship retained from original osdblk.c */
5103 MODULE_AUTHOR("Jeff Garzik <jeff@garzik.org>");
5105 MODULE_LICENSE("GPL");