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 static struct kmem_cache *rbd_img_request_cache;
350 static int rbd_img_request_submit(struct rbd_img_request *img_request);
352 static void rbd_dev_device_release(struct device *dev);
354 static ssize_t rbd_add(struct bus_type *bus, const char *buf,
356 static ssize_t rbd_remove(struct bus_type *bus, const char *buf,
358 static int rbd_dev_image_probe(struct rbd_device *rbd_dev);
360 static struct bus_attribute rbd_bus_attrs[] = {
361 __ATTR(add, S_IWUSR, NULL, rbd_add),
362 __ATTR(remove, S_IWUSR, NULL, rbd_remove),
366 static struct bus_type rbd_bus_type = {
368 .bus_attrs = rbd_bus_attrs,
371 static void rbd_root_dev_release(struct device *dev)
375 static struct device rbd_root_dev = {
377 .release = rbd_root_dev_release,
380 static __printf(2, 3)
381 void rbd_warn(struct rbd_device *rbd_dev, const char *fmt, ...)
383 struct va_format vaf;
391 printk(KERN_WARNING "%s: %pV\n", RBD_DRV_NAME, &vaf);
392 else if (rbd_dev->disk)
393 printk(KERN_WARNING "%s: %s: %pV\n",
394 RBD_DRV_NAME, rbd_dev->disk->disk_name, &vaf);
395 else if (rbd_dev->spec && rbd_dev->spec->image_name)
396 printk(KERN_WARNING "%s: image %s: %pV\n",
397 RBD_DRV_NAME, rbd_dev->spec->image_name, &vaf);
398 else if (rbd_dev->spec && rbd_dev->spec->image_id)
399 printk(KERN_WARNING "%s: id %s: %pV\n",
400 RBD_DRV_NAME, rbd_dev->spec->image_id, &vaf);
402 printk(KERN_WARNING "%s: rbd_dev %p: %pV\n",
403 RBD_DRV_NAME, rbd_dev, &vaf);
408 #define rbd_assert(expr) \
409 if (unlikely(!(expr))) { \
410 printk(KERN_ERR "\nAssertion failure in %s() " \
412 "\trbd_assert(%s);\n\n", \
413 __func__, __LINE__, #expr); \
416 #else /* !RBD_DEBUG */
417 # define rbd_assert(expr) ((void) 0)
418 #endif /* !RBD_DEBUG */
420 static int rbd_img_obj_request_submit(struct rbd_obj_request *obj_request);
421 static void rbd_img_parent_read(struct rbd_obj_request *obj_request);
422 static void rbd_dev_remove_parent(struct rbd_device *rbd_dev);
424 static int rbd_dev_refresh(struct rbd_device *rbd_dev);
425 static int rbd_dev_v2_refresh(struct rbd_device *rbd_dev);
426 static const char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev,
428 static int _rbd_dev_v2_snap_size(struct rbd_device *rbd_dev, u64 snap_id,
429 u8 *order, u64 *snap_size);
430 static int _rbd_dev_v2_snap_features(struct rbd_device *rbd_dev, u64 snap_id,
432 static u64 rbd_snap_id_by_name(struct rbd_device *rbd_dev, const char *name);
434 static int rbd_open(struct block_device *bdev, fmode_t mode)
436 struct rbd_device *rbd_dev = bdev->bd_disk->private_data;
437 bool removing = false;
439 if ((mode & FMODE_WRITE) && rbd_dev->mapping.read_only)
442 spin_lock_irq(&rbd_dev->lock);
443 if (test_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags))
446 rbd_dev->open_count++;
447 spin_unlock_irq(&rbd_dev->lock);
451 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
452 (void) get_device(&rbd_dev->dev);
453 set_device_ro(bdev, rbd_dev->mapping.read_only);
454 mutex_unlock(&ctl_mutex);
459 static int rbd_release(struct gendisk *disk, fmode_t mode)
461 struct rbd_device *rbd_dev = disk->private_data;
462 unsigned long open_count_before;
464 spin_lock_irq(&rbd_dev->lock);
465 open_count_before = rbd_dev->open_count--;
466 spin_unlock_irq(&rbd_dev->lock);
467 rbd_assert(open_count_before > 0);
469 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
470 put_device(&rbd_dev->dev);
471 mutex_unlock(&ctl_mutex);
476 static const struct block_device_operations rbd_bd_ops = {
477 .owner = THIS_MODULE,
479 .release = rbd_release,
483 * Initialize an rbd client instance.
486 static struct rbd_client *rbd_client_create(struct ceph_options *ceph_opts)
488 struct rbd_client *rbdc;
491 dout("%s:\n", __func__);
492 rbdc = kmalloc(sizeof(struct rbd_client), GFP_KERNEL);
496 kref_init(&rbdc->kref);
497 INIT_LIST_HEAD(&rbdc->node);
499 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
501 rbdc->client = ceph_create_client(ceph_opts, rbdc, 0, 0);
502 if (IS_ERR(rbdc->client))
504 ceph_opts = NULL; /* Now rbdc->client is responsible for ceph_opts */
506 ret = ceph_open_session(rbdc->client);
510 spin_lock(&rbd_client_list_lock);
511 list_add_tail(&rbdc->node, &rbd_client_list);
512 spin_unlock(&rbd_client_list_lock);
514 mutex_unlock(&ctl_mutex);
515 dout("%s: rbdc %p\n", __func__, rbdc);
520 ceph_destroy_client(rbdc->client);
522 mutex_unlock(&ctl_mutex);
526 ceph_destroy_options(ceph_opts);
527 dout("%s: error %d\n", __func__, ret);
532 static struct rbd_client *__rbd_get_client(struct rbd_client *rbdc)
534 kref_get(&rbdc->kref);
540 * Find a ceph client with specific addr and configuration. If
541 * found, bump its reference count.
543 static struct rbd_client *rbd_client_find(struct ceph_options *ceph_opts)
545 struct rbd_client *client_node;
548 if (ceph_opts->flags & CEPH_OPT_NOSHARE)
551 spin_lock(&rbd_client_list_lock);
552 list_for_each_entry(client_node, &rbd_client_list, node) {
553 if (!ceph_compare_options(ceph_opts, client_node->client)) {
554 __rbd_get_client(client_node);
560 spin_unlock(&rbd_client_list_lock);
562 return found ? client_node : NULL;
572 /* string args above */
575 /* Boolean args above */
579 static match_table_t rbd_opts_tokens = {
581 /* string args above */
582 {Opt_read_only, "read_only"},
583 {Opt_read_only, "ro"}, /* Alternate spelling */
584 {Opt_read_write, "read_write"},
585 {Opt_read_write, "rw"}, /* Alternate spelling */
586 /* Boolean args above */
594 #define RBD_READ_ONLY_DEFAULT false
596 static int parse_rbd_opts_token(char *c, void *private)
598 struct rbd_options *rbd_opts = private;
599 substring_t argstr[MAX_OPT_ARGS];
600 int token, intval, ret;
602 token = match_token(c, rbd_opts_tokens, argstr);
606 if (token < Opt_last_int) {
607 ret = match_int(&argstr[0], &intval);
609 pr_err("bad mount option arg (not int) "
613 dout("got int token %d val %d\n", token, intval);
614 } else if (token > Opt_last_int && token < Opt_last_string) {
615 dout("got string token %d val %s\n", token,
617 } else if (token > Opt_last_string && token < Opt_last_bool) {
618 dout("got Boolean token %d\n", token);
620 dout("got token %d\n", token);
625 rbd_opts->read_only = true;
628 rbd_opts->read_only = false;
638 * Get a ceph client with specific addr and configuration, if one does
639 * not exist create it.
641 static struct rbd_client *rbd_get_client(struct ceph_options *ceph_opts)
643 struct rbd_client *rbdc;
645 rbdc = rbd_client_find(ceph_opts);
646 if (rbdc) /* using an existing client */
647 ceph_destroy_options(ceph_opts);
649 rbdc = rbd_client_create(ceph_opts);
655 * Destroy ceph client
657 * Caller must hold rbd_client_list_lock.
659 static void rbd_client_release(struct kref *kref)
661 struct rbd_client *rbdc = container_of(kref, struct rbd_client, kref);
663 dout("%s: rbdc %p\n", __func__, rbdc);
664 spin_lock(&rbd_client_list_lock);
665 list_del(&rbdc->node);
666 spin_unlock(&rbd_client_list_lock);
668 ceph_destroy_client(rbdc->client);
673 * Drop reference to ceph client node. If it's not referenced anymore, release
676 static void rbd_put_client(struct rbd_client *rbdc)
679 kref_put(&rbdc->kref, rbd_client_release);
682 static bool rbd_image_format_valid(u32 image_format)
684 return image_format == 1 || image_format == 2;
687 static bool rbd_dev_ondisk_valid(struct rbd_image_header_ondisk *ondisk)
692 /* The header has to start with the magic rbd header text */
693 if (memcmp(&ondisk->text, RBD_HEADER_TEXT, sizeof (RBD_HEADER_TEXT)))
696 /* The bio layer requires at least sector-sized I/O */
698 if (ondisk->options.order < SECTOR_SHIFT)
701 /* If we use u64 in a few spots we may be able to loosen this */
703 if (ondisk->options.order > 8 * sizeof (int) - 1)
707 * The size of a snapshot header has to fit in a size_t, and
708 * that limits the number of snapshots.
710 snap_count = le32_to_cpu(ondisk->snap_count);
711 size = SIZE_MAX - sizeof (struct ceph_snap_context);
712 if (snap_count > size / sizeof (__le64))
716 * Not only that, but the size of the entire the snapshot
717 * header must also be representable in a size_t.
719 size -= snap_count * sizeof (__le64);
720 if ((u64) size < le64_to_cpu(ondisk->snap_names_len))
727 * Create a new header structure, translate header format from the on-disk
730 static int rbd_header_from_disk(struct rbd_image_header *header,
731 struct rbd_image_header_ondisk *ondisk)
738 memset(header, 0, sizeof (*header));
740 snap_count = le32_to_cpu(ondisk->snap_count);
742 len = strnlen(ondisk->object_prefix, sizeof (ondisk->object_prefix));
743 header->object_prefix = kmalloc(len + 1, GFP_KERNEL);
744 if (!header->object_prefix)
746 memcpy(header->object_prefix, ondisk->object_prefix, len);
747 header->object_prefix[len] = '\0';
750 u64 snap_names_len = le64_to_cpu(ondisk->snap_names_len);
752 /* Save a copy of the snapshot names */
754 if (snap_names_len > (u64) SIZE_MAX)
756 header->snap_names = kmalloc(snap_names_len, GFP_KERNEL);
757 if (!header->snap_names)
760 * Note that rbd_dev_v1_header_read() guarantees
761 * the ondisk buffer we're working with has
762 * snap_names_len bytes beyond the end of the
763 * snapshot id array, this memcpy() is safe.
765 memcpy(header->snap_names, &ondisk->snaps[snap_count],
768 /* Record each snapshot's size */
770 size = snap_count * sizeof (*header->snap_sizes);
771 header->snap_sizes = kmalloc(size, GFP_KERNEL);
772 if (!header->snap_sizes)
774 for (i = 0; i < snap_count; i++)
775 header->snap_sizes[i] =
776 le64_to_cpu(ondisk->snaps[i].image_size);
778 header->snap_names = NULL;
779 header->snap_sizes = NULL;
782 header->features = 0; /* No features support in v1 images */
783 header->obj_order = ondisk->options.order;
784 header->crypt_type = ondisk->options.crypt_type;
785 header->comp_type = ondisk->options.comp_type;
787 /* Allocate and fill in the snapshot context */
789 header->image_size = le64_to_cpu(ondisk->image_size);
791 header->snapc = ceph_create_snap_context(snap_count, GFP_KERNEL);
794 header->snapc->seq = le64_to_cpu(ondisk->snap_seq);
795 for (i = 0; i < snap_count; i++)
796 header->snapc->snaps[i] = le64_to_cpu(ondisk->snaps[i].id);
801 kfree(header->snap_sizes);
802 header->snap_sizes = NULL;
803 kfree(header->snap_names);
804 header->snap_names = NULL;
805 kfree(header->object_prefix);
806 header->object_prefix = NULL;
811 static const char *_rbd_dev_v1_snap_name(struct rbd_device *rbd_dev, u32 which)
813 const char *snap_name;
815 rbd_assert(which < rbd_dev->header.snapc->num_snaps);
817 /* Skip over names until we find the one we are looking for */
819 snap_name = rbd_dev->header.snap_names;
821 snap_name += strlen(snap_name) + 1;
823 return kstrdup(snap_name, GFP_KERNEL);
827 * Snapshot id comparison function for use with qsort()/bsearch().
828 * Note that result is for snapshots in *descending* order.
830 static int snapid_compare_reverse(const void *s1, const void *s2)
832 u64 snap_id1 = *(u64 *)s1;
833 u64 snap_id2 = *(u64 *)s2;
835 if (snap_id1 < snap_id2)
837 return snap_id1 == snap_id2 ? 0 : -1;
841 * Search a snapshot context to see if the given snapshot id is
844 * Returns the position of the snapshot id in the array if it's found,
845 * or BAD_SNAP_INDEX otherwise.
847 * Note: The snapshot array is in kept sorted (by the osd) in
848 * reverse order, highest snapshot id first.
850 static u32 rbd_dev_snap_index(struct rbd_device *rbd_dev, u64 snap_id)
852 struct ceph_snap_context *snapc = rbd_dev->header.snapc;
855 found = bsearch(&snap_id, &snapc->snaps, snapc->num_snaps,
856 sizeof (snap_id), snapid_compare_reverse);
858 return found ? (u32)(found - &snapc->snaps[0]) : BAD_SNAP_INDEX;
861 static const char *rbd_dev_v1_snap_name(struct rbd_device *rbd_dev,
866 which = rbd_dev_snap_index(rbd_dev, snap_id);
867 if (which == BAD_SNAP_INDEX)
870 return _rbd_dev_v1_snap_name(rbd_dev, which);
873 static const char *rbd_snap_name(struct rbd_device *rbd_dev, u64 snap_id)
875 if (snap_id == CEPH_NOSNAP)
876 return RBD_SNAP_HEAD_NAME;
878 rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
879 if (rbd_dev->image_format == 1)
880 return rbd_dev_v1_snap_name(rbd_dev, snap_id);
882 return rbd_dev_v2_snap_name(rbd_dev, snap_id);
885 static int rbd_snap_size(struct rbd_device *rbd_dev, u64 snap_id,
888 rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
889 if (snap_id == CEPH_NOSNAP) {
890 *snap_size = rbd_dev->header.image_size;
891 } else if (rbd_dev->image_format == 1) {
894 which = rbd_dev_snap_index(rbd_dev, snap_id);
895 if (which == BAD_SNAP_INDEX)
898 *snap_size = rbd_dev->header.snap_sizes[which];
903 ret = _rbd_dev_v2_snap_size(rbd_dev, snap_id, NULL, &size);
912 static int rbd_snap_features(struct rbd_device *rbd_dev, u64 snap_id,
915 rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
916 if (snap_id == CEPH_NOSNAP) {
917 *snap_features = rbd_dev->header.features;
918 } else if (rbd_dev->image_format == 1) {
919 *snap_features = 0; /* No features for format 1 */
924 ret = _rbd_dev_v2_snap_features(rbd_dev, snap_id, &features);
928 *snap_features = features;
933 static int rbd_dev_mapping_set(struct rbd_device *rbd_dev)
935 const char *snap_name = rbd_dev->spec->snap_name;
941 if (strcmp(snap_name, RBD_SNAP_HEAD_NAME)) {
942 snap_id = rbd_snap_id_by_name(rbd_dev, snap_name);
943 if (snap_id == CEPH_NOSNAP)
946 snap_id = CEPH_NOSNAP;
949 ret = rbd_snap_size(rbd_dev, snap_id, &size);
952 ret = rbd_snap_features(rbd_dev, snap_id, &features);
956 rbd_dev->mapping.size = size;
957 rbd_dev->mapping.features = features;
959 /* If we are mapping a snapshot it must be marked read-only */
961 if (snap_id != CEPH_NOSNAP)
962 rbd_dev->mapping.read_only = true;
967 static void rbd_dev_mapping_clear(struct rbd_device *rbd_dev)
969 rbd_dev->mapping.size = 0;
970 rbd_dev->mapping.features = 0;
971 rbd_dev->mapping.read_only = true;
974 static void rbd_dev_clear_mapping(struct rbd_device *rbd_dev)
976 rbd_dev->mapping.size = 0;
977 rbd_dev->mapping.features = 0;
978 rbd_dev->mapping.read_only = true;
981 static const char *rbd_segment_name(struct rbd_device *rbd_dev, u64 offset)
987 name = kmalloc(MAX_OBJ_NAME_SIZE + 1, GFP_NOIO);
990 segment = offset >> rbd_dev->header.obj_order;
991 ret = snprintf(name, MAX_OBJ_NAME_SIZE + 1, "%s.%012llx",
992 rbd_dev->header.object_prefix, segment);
993 if (ret < 0 || ret > MAX_OBJ_NAME_SIZE) {
994 pr_err("error formatting segment name for #%llu (%d)\n",
1003 static u64 rbd_segment_offset(struct rbd_device *rbd_dev, u64 offset)
1005 u64 segment_size = (u64) 1 << rbd_dev->header.obj_order;
1007 return offset & (segment_size - 1);
1010 static u64 rbd_segment_length(struct rbd_device *rbd_dev,
1011 u64 offset, u64 length)
1013 u64 segment_size = (u64) 1 << rbd_dev->header.obj_order;
1015 offset &= segment_size - 1;
1017 rbd_assert(length <= U64_MAX - offset);
1018 if (offset + length > segment_size)
1019 length = segment_size - offset;
1025 * returns the size of an object in the image
1027 static u64 rbd_obj_bytes(struct rbd_image_header *header)
1029 return 1 << header->obj_order;
1036 static void bio_chain_put(struct bio *chain)
1042 chain = chain->bi_next;
1048 * zeros a bio chain, starting at specific offset
1050 static void zero_bio_chain(struct bio *chain, int start_ofs)
1053 unsigned long flags;
1059 bio_for_each_segment(bv, chain, i) {
1060 if (pos + bv->bv_len > start_ofs) {
1061 int remainder = max(start_ofs - pos, 0);
1062 buf = bvec_kmap_irq(bv, &flags);
1063 memset(buf + remainder, 0,
1064 bv->bv_len - remainder);
1065 bvec_kunmap_irq(buf, &flags);
1070 chain = chain->bi_next;
1075 * similar to zero_bio_chain(), zeros data defined by a page array,
1076 * starting at the given byte offset from the start of the array and
1077 * continuing up to the given end offset. The pages array is
1078 * assumed to be big enough to hold all bytes up to the end.
1080 static void zero_pages(struct page **pages, u64 offset, u64 end)
1082 struct page **page = &pages[offset >> PAGE_SHIFT];
1084 rbd_assert(end > offset);
1085 rbd_assert(end - offset <= (u64)SIZE_MAX);
1086 while (offset < end) {
1089 unsigned long flags;
1092 page_offset = (size_t)(offset & ~PAGE_MASK);
1093 length = min(PAGE_SIZE - page_offset, (size_t)(end - offset));
1094 local_irq_save(flags);
1095 kaddr = kmap_atomic(*page);
1096 memset(kaddr + page_offset, 0, length);
1097 kunmap_atomic(kaddr);
1098 local_irq_restore(flags);
1106 * Clone a portion of a bio, starting at the given byte offset
1107 * and continuing for the number of bytes indicated.
1109 static struct bio *bio_clone_range(struct bio *bio_src,
1110 unsigned int offset,
1118 unsigned short end_idx;
1119 unsigned short vcnt;
1122 /* Handle the easy case for the caller */
1124 if (!offset && len == bio_src->bi_size)
1125 return bio_clone(bio_src, gfpmask);
1127 if (WARN_ON_ONCE(!len))
1129 if (WARN_ON_ONCE(len > bio_src->bi_size))
1131 if (WARN_ON_ONCE(offset > bio_src->bi_size - len))
1134 /* Find first affected segment... */
1137 __bio_for_each_segment(bv, bio_src, idx, 0) {
1138 if (resid < bv->bv_len)
1140 resid -= bv->bv_len;
1144 /* ...and the last affected segment */
1147 __bio_for_each_segment(bv, bio_src, end_idx, idx) {
1148 if (resid <= bv->bv_len)
1150 resid -= bv->bv_len;
1152 vcnt = end_idx - idx + 1;
1154 /* Build the clone */
1156 bio = bio_alloc(gfpmask, (unsigned int) vcnt);
1158 return NULL; /* ENOMEM */
1160 bio->bi_bdev = bio_src->bi_bdev;
1161 bio->bi_sector = bio_src->bi_sector + (offset >> SECTOR_SHIFT);
1162 bio->bi_rw = bio_src->bi_rw;
1163 bio->bi_flags |= 1 << BIO_CLONED;
1166 * Copy over our part of the bio_vec, then update the first
1167 * and last (or only) entries.
1169 memcpy(&bio->bi_io_vec[0], &bio_src->bi_io_vec[idx],
1170 vcnt * sizeof (struct bio_vec));
1171 bio->bi_io_vec[0].bv_offset += voff;
1173 bio->bi_io_vec[0].bv_len -= voff;
1174 bio->bi_io_vec[vcnt - 1].bv_len = resid;
1176 bio->bi_io_vec[0].bv_len = len;
1179 bio->bi_vcnt = vcnt;
1187 * Clone a portion of a bio chain, starting at the given byte offset
1188 * into the first bio in the source chain and continuing for the
1189 * number of bytes indicated. The result is another bio chain of
1190 * exactly the given length, or a null pointer on error.
1192 * The bio_src and offset parameters are both in-out. On entry they
1193 * refer to the first source bio and the offset into that bio where
1194 * the start of data to be cloned is located.
1196 * On return, bio_src is updated to refer to the bio in the source
1197 * chain that contains first un-cloned byte, and *offset will
1198 * contain the offset of that byte within that bio.
1200 static struct bio *bio_chain_clone_range(struct bio **bio_src,
1201 unsigned int *offset,
1205 struct bio *bi = *bio_src;
1206 unsigned int off = *offset;
1207 struct bio *chain = NULL;
1210 /* Build up a chain of clone bios up to the limit */
1212 if (!bi || off >= bi->bi_size || !len)
1213 return NULL; /* Nothing to clone */
1217 unsigned int bi_size;
1221 rbd_warn(NULL, "bio_chain exhausted with %u left", len);
1222 goto out_err; /* EINVAL; ran out of bio's */
1224 bi_size = min_t(unsigned int, bi->bi_size - off, len);
1225 bio = bio_clone_range(bi, off, bi_size, gfpmask);
1227 goto out_err; /* ENOMEM */
1230 end = &bio->bi_next;
1233 if (off == bi->bi_size) {
1244 bio_chain_put(chain);
1250 * The default/initial value for all object request flags is 0. For
1251 * each flag, once its value is set to 1 it is never reset to 0
1254 static void obj_request_img_data_set(struct rbd_obj_request *obj_request)
1256 if (test_and_set_bit(OBJ_REQ_IMG_DATA, &obj_request->flags)) {
1257 struct rbd_device *rbd_dev;
1259 rbd_dev = obj_request->img_request->rbd_dev;
1260 rbd_warn(rbd_dev, "obj_request %p already marked img_data\n",
1265 static bool obj_request_img_data_test(struct rbd_obj_request *obj_request)
1268 return test_bit(OBJ_REQ_IMG_DATA, &obj_request->flags) != 0;
1271 static void obj_request_done_set(struct rbd_obj_request *obj_request)
1273 if (test_and_set_bit(OBJ_REQ_DONE, &obj_request->flags)) {
1274 struct rbd_device *rbd_dev = NULL;
1276 if (obj_request_img_data_test(obj_request))
1277 rbd_dev = obj_request->img_request->rbd_dev;
1278 rbd_warn(rbd_dev, "obj_request %p already marked done\n",
1283 static bool obj_request_done_test(struct rbd_obj_request *obj_request)
1286 return test_bit(OBJ_REQ_DONE, &obj_request->flags) != 0;
1290 * This sets the KNOWN flag after (possibly) setting the EXISTS
1291 * flag. The latter is set based on the "exists" value provided.
1293 * Note that for our purposes once an object exists it never goes
1294 * away again. It's possible that the response from two existence
1295 * checks are separated by the creation of the target object, and
1296 * the first ("doesn't exist") response arrives *after* the second
1297 * ("does exist"). In that case we ignore the second one.
1299 static void obj_request_existence_set(struct rbd_obj_request *obj_request,
1303 set_bit(OBJ_REQ_EXISTS, &obj_request->flags);
1304 set_bit(OBJ_REQ_KNOWN, &obj_request->flags);
1308 static bool obj_request_known_test(struct rbd_obj_request *obj_request)
1311 return test_bit(OBJ_REQ_KNOWN, &obj_request->flags) != 0;
1314 static bool obj_request_exists_test(struct rbd_obj_request *obj_request)
1317 return test_bit(OBJ_REQ_EXISTS, &obj_request->flags) != 0;
1320 static void rbd_obj_request_get(struct rbd_obj_request *obj_request)
1322 dout("%s: obj %p (was %d)\n", __func__, obj_request,
1323 atomic_read(&obj_request->kref.refcount));
1324 kref_get(&obj_request->kref);
1327 static void rbd_obj_request_destroy(struct kref *kref);
1328 static void rbd_obj_request_put(struct rbd_obj_request *obj_request)
1330 rbd_assert(obj_request != NULL);
1331 dout("%s: obj %p (was %d)\n", __func__, obj_request,
1332 atomic_read(&obj_request->kref.refcount));
1333 kref_put(&obj_request->kref, rbd_obj_request_destroy);
1336 static void rbd_img_request_get(struct rbd_img_request *img_request)
1338 dout("%s: img %p (was %d)\n", __func__, img_request,
1339 atomic_read(&img_request->kref.refcount));
1340 kref_get(&img_request->kref);
1343 static void rbd_img_request_destroy(struct kref *kref);
1344 static void rbd_img_request_put(struct rbd_img_request *img_request)
1346 rbd_assert(img_request != NULL);
1347 dout("%s: img %p (was %d)\n", __func__, img_request,
1348 atomic_read(&img_request->kref.refcount));
1349 kref_put(&img_request->kref, rbd_img_request_destroy);
1352 static inline void rbd_img_obj_request_add(struct rbd_img_request *img_request,
1353 struct rbd_obj_request *obj_request)
1355 rbd_assert(obj_request->img_request == NULL);
1357 /* Image request now owns object's original reference */
1358 obj_request->img_request = img_request;
1359 obj_request->which = img_request->obj_request_count;
1360 rbd_assert(!obj_request_img_data_test(obj_request));
1361 obj_request_img_data_set(obj_request);
1362 rbd_assert(obj_request->which != BAD_WHICH);
1363 img_request->obj_request_count++;
1364 list_add_tail(&obj_request->links, &img_request->obj_requests);
1365 dout("%s: img %p obj %p w=%u\n", __func__, img_request, obj_request,
1366 obj_request->which);
1369 static inline void rbd_img_obj_request_del(struct rbd_img_request *img_request,
1370 struct rbd_obj_request *obj_request)
1372 rbd_assert(obj_request->which != BAD_WHICH);
1374 dout("%s: img %p obj %p w=%u\n", __func__, img_request, obj_request,
1375 obj_request->which);
1376 list_del(&obj_request->links);
1377 rbd_assert(img_request->obj_request_count > 0);
1378 img_request->obj_request_count--;
1379 rbd_assert(obj_request->which == img_request->obj_request_count);
1380 obj_request->which = BAD_WHICH;
1381 rbd_assert(obj_request_img_data_test(obj_request));
1382 rbd_assert(obj_request->img_request == img_request);
1383 obj_request->img_request = NULL;
1384 obj_request->callback = NULL;
1385 rbd_obj_request_put(obj_request);
1388 static bool obj_request_type_valid(enum obj_request_type type)
1391 case OBJ_REQUEST_NODATA:
1392 case OBJ_REQUEST_BIO:
1393 case OBJ_REQUEST_PAGES:
1400 static int rbd_obj_request_submit(struct ceph_osd_client *osdc,
1401 struct rbd_obj_request *obj_request)
1403 dout("%s: osdc %p obj %p\n", __func__, osdc, obj_request);
1405 return ceph_osdc_start_request(osdc, obj_request->osd_req, false);
1408 static void rbd_img_request_complete(struct rbd_img_request *img_request)
1411 dout("%s: img %p\n", __func__, img_request);
1414 * If no error occurred, compute the aggregate transfer
1415 * count for the image request. We could instead use
1416 * atomic64_cmpxchg() to update it as each object request
1417 * completes; not clear which way is better off hand.
1419 if (!img_request->result) {
1420 struct rbd_obj_request *obj_request;
1423 for_each_obj_request(img_request, obj_request)
1424 xferred += obj_request->xferred;
1425 img_request->xferred = xferred;
1428 if (img_request->callback)
1429 img_request->callback(img_request);
1431 rbd_img_request_put(img_request);
1434 /* Caller is responsible for rbd_obj_request_destroy(obj_request) */
1436 static int rbd_obj_request_wait(struct rbd_obj_request *obj_request)
1438 dout("%s: obj %p\n", __func__, obj_request);
1440 return wait_for_completion_interruptible(&obj_request->completion);
1444 * The default/initial value for all image request flags is 0. Each
1445 * is conditionally set to 1 at image request initialization time
1446 * and currently never change thereafter.
1448 static void img_request_write_set(struct rbd_img_request *img_request)
1450 set_bit(IMG_REQ_WRITE, &img_request->flags);
1454 static bool img_request_write_test(struct rbd_img_request *img_request)
1457 return test_bit(IMG_REQ_WRITE, &img_request->flags) != 0;
1460 static void img_request_child_set(struct rbd_img_request *img_request)
1462 set_bit(IMG_REQ_CHILD, &img_request->flags);
1466 static bool img_request_child_test(struct rbd_img_request *img_request)
1469 return test_bit(IMG_REQ_CHILD, &img_request->flags) != 0;
1472 static void img_request_layered_set(struct rbd_img_request *img_request)
1474 set_bit(IMG_REQ_LAYERED, &img_request->flags);
1478 static bool img_request_layered_test(struct rbd_img_request *img_request)
1481 return test_bit(IMG_REQ_LAYERED, &img_request->flags) != 0;
1485 rbd_img_obj_request_read_callback(struct rbd_obj_request *obj_request)
1487 u64 xferred = obj_request->xferred;
1488 u64 length = obj_request->length;
1490 dout("%s: obj %p img %p result %d %llu/%llu\n", __func__,
1491 obj_request, obj_request->img_request, obj_request->result,
1494 * ENOENT means a hole in the image. We zero-fill the
1495 * entire length of the request. A short read also implies
1496 * zero-fill to the end of the request. Either way we
1497 * update the xferred count to indicate the whole request
1500 rbd_assert(obj_request->type != OBJ_REQUEST_NODATA);
1501 if (obj_request->result == -ENOENT) {
1502 if (obj_request->type == OBJ_REQUEST_BIO)
1503 zero_bio_chain(obj_request->bio_list, 0);
1505 zero_pages(obj_request->pages, 0, length);
1506 obj_request->result = 0;
1507 obj_request->xferred = length;
1508 } else if (xferred < length && !obj_request->result) {
1509 if (obj_request->type == OBJ_REQUEST_BIO)
1510 zero_bio_chain(obj_request->bio_list, xferred);
1512 zero_pages(obj_request->pages, xferred, length);
1513 obj_request->xferred = length;
1515 obj_request_done_set(obj_request);
1518 static void rbd_obj_request_complete(struct rbd_obj_request *obj_request)
1520 dout("%s: obj %p cb %p\n", __func__, obj_request,
1521 obj_request->callback);
1522 if (obj_request->callback)
1523 obj_request->callback(obj_request);
1525 complete_all(&obj_request->completion);
1528 static void rbd_osd_trivial_callback(struct rbd_obj_request *obj_request)
1530 dout("%s: obj %p\n", __func__, obj_request);
1531 obj_request_done_set(obj_request);
1534 static void rbd_osd_read_callback(struct rbd_obj_request *obj_request)
1536 struct rbd_img_request *img_request = NULL;
1537 struct rbd_device *rbd_dev = NULL;
1538 bool layered = false;
1540 if (obj_request_img_data_test(obj_request)) {
1541 img_request = obj_request->img_request;
1542 layered = img_request && img_request_layered_test(img_request);
1543 rbd_dev = img_request->rbd_dev;
1546 dout("%s: obj %p img %p result %d %llu/%llu\n", __func__,
1547 obj_request, img_request, obj_request->result,
1548 obj_request->xferred, obj_request->length);
1549 if (layered && obj_request->result == -ENOENT &&
1550 obj_request->img_offset < rbd_dev->parent_overlap)
1551 rbd_img_parent_read(obj_request);
1552 else if (img_request)
1553 rbd_img_obj_request_read_callback(obj_request);
1555 obj_request_done_set(obj_request);
1558 static void rbd_osd_write_callback(struct rbd_obj_request *obj_request)
1560 dout("%s: obj %p result %d %llu\n", __func__, obj_request,
1561 obj_request->result, obj_request->length);
1563 * There is no such thing as a successful short write. Set
1564 * it to our originally-requested length.
1566 obj_request->xferred = obj_request->length;
1567 obj_request_done_set(obj_request);
1571 * For a simple stat call there's nothing to do. We'll do more if
1572 * this is part of a write sequence for a layered image.
1574 static void rbd_osd_stat_callback(struct rbd_obj_request *obj_request)
1576 dout("%s: obj %p\n", __func__, obj_request);
1577 obj_request_done_set(obj_request);
1580 static void rbd_osd_req_callback(struct ceph_osd_request *osd_req,
1581 struct ceph_msg *msg)
1583 struct rbd_obj_request *obj_request = osd_req->r_priv;
1586 dout("%s: osd_req %p msg %p\n", __func__, osd_req, msg);
1587 rbd_assert(osd_req == obj_request->osd_req);
1588 if (obj_request_img_data_test(obj_request)) {
1589 rbd_assert(obj_request->img_request);
1590 rbd_assert(obj_request->which != BAD_WHICH);
1592 rbd_assert(obj_request->which == BAD_WHICH);
1595 if (osd_req->r_result < 0)
1596 obj_request->result = osd_req->r_result;
1598 BUG_ON(osd_req->r_num_ops > 2);
1601 * We support a 64-bit length, but ultimately it has to be
1602 * passed to blk_end_request(), which takes an unsigned int.
1604 obj_request->xferred = osd_req->r_reply_op_len[0];
1605 rbd_assert(obj_request->xferred < (u64)UINT_MAX);
1606 opcode = osd_req->r_ops[0].op;
1608 case CEPH_OSD_OP_READ:
1609 rbd_osd_read_callback(obj_request);
1611 case CEPH_OSD_OP_WRITE:
1612 rbd_osd_write_callback(obj_request);
1614 case CEPH_OSD_OP_STAT:
1615 rbd_osd_stat_callback(obj_request);
1617 case CEPH_OSD_OP_CALL:
1618 case CEPH_OSD_OP_NOTIFY_ACK:
1619 case CEPH_OSD_OP_WATCH:
1620 rbd_osd_trivial_callback(obj_request);
1623 rbd_warn(NULL, "%s: unsupported op %hu\n",
1624 obj_request->object_name, (unsigned short) opcode);
1628 if (obj_request_done_test(obj_request))
1629 rbd_obj_request_complete(obj_request);
1632 static void rbd_osd_req_format_read(struct rbd_obj_request *obj_request)
1634 struct rbd_img_request *img_request = obj_request->img_request;
1635 struct ceph_osd_request *osd_req = obj_request->osd_req;
1638 rbd_assert(osd_req != NULL);
1640 snap_id = img_request ? img_request->snap_id : CEPH_NOSNAP;
1641 ceph_osdc_build_request(osd_req, obj_request->offset,
1642 NULL, snap_id, NULL);
1645 static void rbd_osd_req_format_write(struct rbd_obj_request *obj_request)
1647 struct rbd_img_request *img_request = obj_request->img_request;
1648 struct ceph_osd_request *osd_req = obj_request->osd_req;
1649 struct ceph_snap_context *snapc;
1650 struct timespec mtime = CURRENT_TIME;
1652 rbd_assert(osd_req != NULL);
1654 snapc = img_request ? img_request->snapc : NULL;
1655 ceph_osdc_build_request(osd_req, obj_request->offset,
1656 snapc, CEPH_NOSNAP, &mtime);
1659 static struct ceph_osd_request *rbd_osd_req_create(
1660 struct rbd_device *rbd_dev,
1662 struct rbd_obj_request *obj_request)
1664 struct ceph_snap_context *snapc = NULL;
1665 struct ceph_osd_client *osdc;
1666 struct ceph_osd_request *osd_req;
1668 if (obj_request_img_data_test(obj_request)) {
1669 struct rbd_img_request *img_request = obj_request->img_request;
1671 rbd_assert(write_request ==
1672 img_request_write_test(img_request));
1674 snapc = img_request->snapc;
1677 /* Allocate and initialize the request, for the single op */
1679 osdc = &rbd_dev->rbd_client->client->osdc;
1680 osd_req = ceph_osdc_alloc_request(osdc, snapc, 1, false, GFP_ATOMIC);
1682 return NULL; /* ENOMEM */
1685 osd_req->r_flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK;
1687 osd_req->r_flags = CEPH_OSD_FLAG_READ;
1689 osd_req->r_callback = rbd_osd_req_callback;
1690 osd_req->r_priv = obj_request;
1692 osd_req->r_oid_len = strlen(obj_request->object_name);
1693 rbd_assert(osd_req->r_oid_len < sizeof (osd_req->r_oid));
1694 memcpy(osd_req->r_oid, obj_request->object_name, osd_req->r_oid_len);
1696 osd_req->r_file_layout = rbd_dev->layout; /* struct */
1702 * Create a copyup osd request based on the information in the
1703 * object request supplied. A copyup request has two osd ops,
1704 * a copyup method call, and a "normal" write request.
1706 static struct ceph_osd_request *
1707 rbd_osd_req_create_copyup(struct rbd_obj_request *obj_request)
1709 struct rbd_img_request *img_request;
1710 struct ceph_snap_context *snapc;
1711 struct rbd_device *rbd_dev;
1712 struct ceph_osd_client *osdc;
1713 struct ceph_osd_request *osd_req;
1715 rbd_assert(obj_request_img_data_test(obj_request));
1716 img_request = obj_request->img_request;
1717 rbd_assert(img_request);
1718 rbd_assert(img_request_write_test(img_request));
1720 /* Allocate and initialize the request, for the two ops */
1722 snapc = img_request->snapc;
1723 rbd_dev = img_request->rbd_dev;
1724 osdc = &rbd_dev->rbd_client->client->osdc;
1725 osd_req = ceph_osdc_alloc_request(osdc, snapc, 2, false, GFP_ATOMIC);
1727 return NULL; /* ENOMEM */
1729 osd_req->r_flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK;
1730 osd_req->r_callback = rbd_osd_req_callback;
1731 osd_req->r_priv = obj_request;
1733 osd_req->r_oid_len = strlen(obj_request->object_name);
1734 rbd_assert(osd_req->r_oid_len < sizeof (osd_req->r_oid));
1735 memcpy(osd_req->r_oid, obj_request->object_name, osd_req->r_oid_len);
1737 osd_req->r_file_layout = rbd_dev->layout; /* struct */
1743 static void rbd_osd_req_destroy(struct ceph_osd_request *osd_req)
1745 ceph_osdc_put_request(osd_req);
1748 /* object_name is assumed to be a non-null pointer and NUL-terminated */
1750 static struct rbd_obj_request *rbd_obj_request_create(const char *object_name,
1751 u64 offset, u64 length,
1752 enum obj_request_type type)
1754 struct rbd_obj_request *obj_request;
1758 rbd_assert(obj_request_type_valid(type));
1760 size = strlen(object_name) + 1;
1761 obj_request = kzalloc(sizeof (*obj_request) + size, GFP_KERNEL);
1765 name = (char *)(obj_request + 1);
1766 obj_request->object_name = memcpy(name, object_name, size);
1767 obj_request->offset = offset;
1768 obj_request->length = length;
1769 obj_request->flags = 0;
1770 obj_request->which = BAD_WHICH;
1771 obj_request->type = type;
1772 INIT_LIST_HEAD(&obj_request->links);
1773 init_completion(&obj_request->completion);
1774 kref_init(&obj_request->kref);
1776 dout("%s: \"%s\" %llu/%llu %d -> obj %p\n", __func__, object_name,
1777 offset, length, (int)type, obj_request);
1782 static void rbd_obj_request_destroy(struct kref *kref)
1784 struct rbd_obj_request *obj_request;
1786 obj_request = container_of(kref, struct rbd_obj_request, kref);
1788 dout("%s: obj %p\n", __func__, obj_request);
1790 rbd_assert(obj_request->img_request == NULL);
1791 rbd_assert(obj_request->which == BAD_WHICH);
1793 if (obj_request->osd_req)
1794 rbd_osd_req_destroy(obj_request->osd_req);
1796 rbd_assert(obj_request_type_valid(obj_request->type));
1797 switch (obj_request->type) {
1798 case OBJ_REQUEST_NODATA:
1799 break; /* Nothing to do */
1800 case OBJ_REQUEST_BIO:
1801 if (obj_request->bio_list)
1802 bio_chain_put(obj_request->bio_list);
1804 case OBJ_REQUEST_PAGES:
1805 if (obj_request->pages)
1806 ceph_release_page_vector(obj_request->pages,
1807 obj_request->page_count);
1815 * Caller is responsible for filling in the list of object requests
1816 * that comprises the image request, and the Linux request pointer
1817 * (if there is one).
1819 static struct rbd_img_request *rbd_img_request_create(
1820 struct rbd_device *rbd_dev,
1821 u64 offset, u64 length,
1825 struct rbd_img_request *img_request;
1827 img_request = kmem_cache_alloc(rbd_img_request_cache, GFP_ATOMIC);
1831 if (write_request) {
1832 down_read(&rbd_dev->header_rwsem);
1833 ceph_get_snap_context(rbd_dev->header.snapc);
1834 up_read(&rbd_dev->header_rwsem);
1837 img_request->rq = NULL;
1838 img_request->rbd_dev = rbd_dev;
1839 img_request->offset = offset;
1840 img_request->length = length;
1841 img_request->flags = 0;
1842 if (write_request) {
1843 img_request_write_set(img_request);
1844 img_request->snapc = rbd_dev->header.snapc;
1846 img_request->snap_id = rbd_dev->spec->snap_id;
1849 img_request_child_set(img_request);
1850 if (rbd_dev->parent_spec)
1851 img_request_layered_set(img_request);
1852 spin_lock_init(&img_request->completion_lock);
1853 img_request->next_completion = 0;
1854 img_request->callback = NULL;
1855 img_request->result = 0;
1856 img_request->obj_request_count = 0;
1857 INIT_LIST_HEAD(&img_request->obj_requests);
1858 kref_init(&img_request->kref);
1860 rbd_img_request_get(img_request); /* Avoid a warning */
1861 rbd_img_request_put(img_request); /* TEMPORARY */
1863 dout("%s: rbd_dev %p %s %llu/%llu -> img %p\n", __func__, rbd_dev,
1864 write_request ? "write" : "read", offset, length,
1870 static void rbd_img_request_destroy(struct kref *kref)
1872 struct rbd_img_request *img_request;
1873 struct rbd_obj_request *obj_request;
1874 struct rbd_obj_request *next_obj_request;
1876 img_request = container_of(kref, struct rbd_img_request, kref);
1878 dout("%s: img %p\n", __func__, img_request);
1880 for_each_obj_request_safe(img_request, obj_request, next_obj_request)
1881 rbd_img_obj_request_del(img_request, obj_request);
1882 rbd_assert(img_request->obj_request_count == 0);
1884 if (img_request_write_test(img_request))
1885 ceph_put_snap_context(img_request->snapc);
1887 if (img_request_child_test(img_request))
1888 rbd_obj_request_put(img_request->obj_request);
1890 kmem_cache_free(rbd_img_request_cache, img_request);
1893 static bool rbd_img_obj_end_request(struct rbd_obj_request *obj_request)
1895 struct rbd_img_request *img_request;
1896 unsigned int xferred;
1900 rbd_assert(obj_request_img_data_test(obj_request));
1901 img_request = obj_request->img_request;
1903 rbd_assert(obj_request->xferred <= (u64)UINT_MAX);
1904 xferred = (unsigned int)obj_request->xferred;
1905 result = obj_request->result;
1907 struct rbd_device *rbd_dev = img_request->rbd_dev;
1909 rbd_warn(rbd_dev, "%s %llx at %llx (%llx)\n",
1910 img_request_write_test(img_request) ? "write" : "read",
1911 obj_request->length, obj_request->img_offset,
1912 obj_request->offset);
1913 rbd_warn(rbd_dev, " result %d xferred %x\n",
1915 if (!img_request->result)
1916 img_request->result = result;
1919 /* Image object requests don't own their page array */
1921 if (obj_request->type == OBJ_REQUEST_PAGES) {
1922 obj_request->pages = NULL;
1923 obj_request->page_count = 0;
1926 if (img_request_child_test(img_request)) {
1927 rbd_assert(img_request->obj_request != NULL);
1928 more = obj_request->which < img_request->obj_request_count - 1;
1930 rbd_assert(img_request->rq != NULL);
1931 more = blk_end_request(img_request->rq, result, xferred);
1937 static void rbd_img_obj_callback(struct rbd_obj_request *obj_request)
1939 struct rbd_img_request *img_request;
1940 u32 which = obj_request->which;
1943 rbd_assert(obj_request_img_data_test(obj_request));
1944 img_request = obj_request->img_request;
1946 dout("%s: img %p obj %p\n", __func__, img_request, obj_request);
1947 rbd_assert(img_request != NULL);
1948 rbd_assert(img_request->obj_request_count > 0);
1949 rbd_assert(which != BAD_WHICH);
1950 rbd_assert(which < img_request->obj_request_count);
1951 rbd_assert(which >= img_request->next_completion);
1953 spin_lock_irq(&img_request->completion_lock);
1954 if (which != img_request->next_completion)
1957 for_each_obj_request_from(img_request, obj_request) {
1959 rbd_assert(which < img_request->obj_request_count);
1961 if (!obj_request_done_test(obj_request))
1963 more = rbd_img_obj_end_request(obj_request);
1967 rbd_assert(more ^ (which == img_request->obj_request_count));
1968 img_request->next_completion = which;
1970 spin_unlock_irq(&img_request->completion_lock);
1973 rbd_img_request_complete(img_request);
1977 * Split up an image request into one or more object requests, each
1978 * to a different object. The "type" parameter indicates whether
1979 * "data_desc" is the pointer to the head of a list of bio
1980 * structures, or the base of a page array. In either case this
1981 * function assumes data_desc describes memory sufficient to hold
1982 * all data described by the image request.
1984 static int rbd_img_request_fill(struct rbd_img_request *img_request,
1985 enum obj_request_type type,
1988 struct rbd_device *rbd_dev = img_request->rbd_dev;
1989 struct rbd_obj_request *obj_request = NULL;
1990 struct rbd_obj_request *next_obj_request;
1991 bool write_request = img_request_write_test(img_request);
1992 struct bio *bio_list;
1993 unsigned int bio_offset = 0;
1994 struct page **pages;
1999 dout("%s: img %p type %d data_desc %p\n", __func__, img_request,
2000 (int)type, data_desc);
2002 opcode = write_request ? CEPH_OSD_OP_WRITE : CEPH_OSD_OP_READ;
2003 img_offset = img_request->offset;
2004 resid = img_request->length;
2005 rbd_assert(resid > 0);
2007 if (type == OBJ_REQUEST_BIO) {
2008 bio_list = data_desc;
2009 rbd_assert(img_offset == bio_list->bi_sector << SECTOR_SHIFT);
2011 rbd_assert(type == OBJ_REQUEST_PAGES);
2016 struct ceph_osd_request *osd_req;
2017 const char *object_name;
2021 object_name = rbd_segment_name(rbd_dev, img_offset);
2024 offset = rbd_segment_offset(rbd_dev, img_offset);
2025 length = rbd_segment_length(rbd_dev, img_offset, resid);
2026 obj_request = rbd_obj_request_create(object_name,
2027 offset, length, type);
2028 kfree(object_name); /* object request has its own copy */
2032 if (type == OBJ_REQUEST_BIO) {
2033 unsigned int clone_size;
2035 rbd_assert(length <= (u64)UINT_MAX);
2036 clone_size = (unsigned int)length;
2037 obj_request->bio_list =
2038 bio_chain_clone_range(&bio_list,
2042 if (!obj_request->bio_list)
2045 unsigned int page_count;
2047 obj_request->pages = pages;
2048 page_count = (u32)calc_pages_for(offset, length);
2049 obj_request->page_count = page_count;
2050 if ((offset + length) & ~PAGE_MASK)
2051 page_count--; /* more on last page */
2052 pages += page_count;
2055 osd_req = rbd_osd_req_create(rbd_dev, write_request,
2059 obj_request->osd_req = osd_req;
2060 obj_request->callback = rbd_img_obj_callback;
2062 osd_req_op_extent_init(osd_req, 0, opcode, offset, length,
2064 if (type == OBJ_REQUEST_BIO)
2065 osd_req_op_extent_osd_data_bio(osd_req, 0,
2066 obj_request->bio_list, length);
2068 osd_req_op_extent_osd_data_pages(osd_req, 0,
2069 obj_request->pages, length,
2070 offset & ~PAGE_MASK, false, false);
2073 rbd_osd_req_format_write(obj_request);
2075 rbd_osd_req_format_read(obj_request);
2077 obj_request->img_offset = img_offset;
2078 rbd_img_obj_request_add(img_request, obj_request);
2080 img_offset += length;
2087 rbd_obj_request_put(obj_request);
2089 for_each_obj_request_safe(img_request, obj_request, next_obj_request)
2090 rbd_obj_request_put(obj_request);
2096 rbd_img_obj_copyup_callback(struct rbd_obj_request *obj_request)
2098 struct rbd_img_request *img_request;
2099 struct rbd_device *rbd_dev;
2103 rbd_assert(obj_request->type == OBJ_REQUEST_BIO);
2104 rbd_assert(obj_request_img_data_test(obj_request));
2105 img_request = obj_request->img_request;
2106 rbd_assert(img_request);
2108 rbd_dev = img_request->rbd_dev;
2109 rbd_assert(rbd_dev);
2110 length = (u64)1 << rbd_dev->header.obj_order;
2111 page_count = (u32)calc_pages_for(0, length);
2113 rbd_assert(obj_request->copyup_pages);
2114 ceph_release_page_vector(obj_request->copyup_pages, page_count);
2115 obj_request->copyup_pages = NULL;
2118 * We want the transfer count to reflect the size of the
2119 * original write request. There is no such thing as a
2120 * successful short write, so if the request was successful
2121 * we can just set it to the originally-requested length.
2123 if (!obj_request->result)
2124 obj_request->xferred = obj_request->length;
2126 /* Finish up with the normal image object callback */
2128 rbd_img_obj_callback(obj_request);
2132 rbd_img_obj_parent_read_full_callback(struct rbd_img_request *img_request)
2134 struct rbd_obj_request *orig_request;
2135 struct ceph_osd_request *osd_req;
2136 struct ceph_osd_client *osdc;
2137 struct rbd_device *rbd_dev;
2138 struct page **pages;
2143 rbd_assert(img_request_child_test(img_request));
2145 /* First get what we need from the image request */
2147 pages = img_request->copyup_pages;
2148 rbd_assert(pages != NULL);
2149 img_request->copyup_pages = NULL;
2151 orig_request = img_request->obj_request;
2152 rbd_assert(orig_request != NULL);
2153 rbd_assert(orig_request->type == OBJ_REQUEST_BIO);
2154 result = img_request->result;
2155 obj_size = img_request->length;
2156 xferred = img_request->xferred;
2158 rbd_dev = img_request->rbd_dev;
2159 rbd_assert(rbd_dev);
2160 rbd_assert(obj_size == (u64)1 << rbd_dev->header.obj_order);
2162 rbd_img_request_put(img_request);
2167 /* Allocate the new copyup osd request for the original request */
2170 rbd_assert(!orig_request->osd_req);
2171 osd_req = rbd_osd_req_create_copyup(orig_request);
2174 orig_request->osd_req = osd_req;
2175 orig_request->copyup_pages = pages;
2177 /* Initialize the copyup op */
2179 osd_req_op_cls_init(osd_req, 0, CEPH_OSD_OP_CALL, "rbd", "copyup");
2180 osd_req_op_cls_request_data_pages(osd_req, 0, pages, obj_size, 0,
2183 /* Then the original write request op */
2185 osd_req_op_extent_init(osd_req, 1, CEPH_OSD_OP_WRITE,
2186 orig_request->offset,
2187 orig_request->length, 0, 0);
2188 osd_req_op_extent_osd_data_bio(osd_req, 1, orig_request->bio_list,
2189 orig_request->length);
2191 rbd_osd_req_format_write(orig_request);
2193 /* All set, send it off. */
2195 orig_request->callback = rbd_img_obj_copyup_callback;
2196 osdc = &rbd_dev->rbd_client->client->osdc;
2197 result = rbd_obj_request_submit(osdc, orig_request);
2201 /* Record the error code and complete the request */
2203 orig_request->result = result;
2204 orig_request->xferred = 0;
2205 obj_request_done_set(orig_request);
2206 rbd_obj_request_complete(orig_request);
2210 * Read from the parent image the range of data that covers the
2211 * entire target of the given object request. This is used for
2212 * satisfying a layered image write request when the target of an
2213 * object request from the image request does not exist.
2215 * A page array big enough to hold the returned data is allocated
2216 * and supplied to rbd_img_request_fill() as the "data descriptor."
2217 * When the read completes, this page array will be transferred to
2218 * the original object request for the copyup operation.
2220 * If an error occurs, record it as the result of the original
2221 * object request and mark it done so it gets completed.
2223 static int rbd_img_obj_parent_read_full(struct rbd_obj_request *obj_request)
2225 struct rbd_img_request *img_request = NULL;
2226 struct rbd_img_request *parent_request = NULL;
2227 struct rbd_device *rbd_dev;
2230 struct page **pages = NULL;
2234 rbd_assert(obj_request_img_data_test(obj_request));
2235 rbd_assert(obj_request->type == OBJ_REQUEST_BIO);
2237 img_request = obj_request->img_request;
2238 rbd_assert(img_request != NULL);
2239 rbd_dev = img_request->rbd_dev;
2240 rbd_assert(rbd_dev->parent != NULL);
2243 * First things first. The original osd request is of no
2244 * use to use any more, we'll need a new one that can hold
2245 * the two ops in a copyup request. We'll get that later,
2246 * but for now we can release the old one.
2248 rbd_osd_req_destroy(obj_request->osd_req);
2249 obj_request->osd_req = NULL;
2252 * Determine the byte range covered by the object in the
2253 * child image to which the original request was to be sent.
2255 img_offset = obj_request->img_offset - obj_request->offset;
2256 length = (u64)1 << rbd_dev->header.obj_order;
2259 * There is no defined parent data beyond the parent
2260 * overlap, so limit what we read at that boundary if
2263 if (img_offset + length > rbd_dev->parent_overlap) {
2264 rbd_assert(img_offset < rbd_dev->parent_overlap);
2265 length = rbd_dev->parent_overlap - img_offset;
2269 * Allocate a page array big enough to receive the data read
2272 page_count = (u32)calc_pages_for(0, length);
2273 pages = ceph_alloc_page_vector(page_count, GFP_KERNEL);
2274 if (IS_ERR(pages)) {
2275 result = PTR_ERR(pages);
2281 parent_request = rbd_img_request_create(rbd_dev->parent,
2284 if (!parent_request)
2286 rbd_obj_request_get(obj_request);
2287 parent_request->obj_request = obj_request;
2289 result = rbd_img_request_fill(parent_request, OBJ_REQUEST_PAGES, pages);
2292 parent_request->copyup_pages = pages;
2294 parent_request->callback = rbd_img_obj_parent_read_full_callback;
2295 result = rbd_img_request_submit(parent_request);
2299 parent_request->copyup_pages = NULL;
2300 parent_request->obj_request = NULL;
2301 rbd_obj_request_put(obj_request);
2304 ceph_release_page_vector(pages, page_count);
2306 rbd_img_request_put(parent_request);
2307 obj_request->result = result;
2308 obj_request->xferred = 0;
2309 obj_request_done_set(obj_request);
2314 static void rbd_img_obj_exists_callback(struct rbd_obj_request *obj_request)
2316 struct rbd_obj_request *orig_request;
2319 rbd_assert(!obj_request_img_data_test(obj_request));
2322 * All we need from the object request is the original
2323 * request and the result of the STAT op. Grab those, then
2324 * we're done with the request.
2326 orig_request = obj_request->obj_request;
2327 obj_request->obj_request = NULL;
2328 rbd_assert(orig_request);
2329 rbd_assert(orig_request->img_request);
2331 result = obj_request->result;
2332 obj_request->result = 0;
2334 dout("%s: obj %p for obj %p result %d %llu/%llu\n", __func__,
2335 obj_request, orig_request, result,
2336 obj_request->xferred, obj_request->length);
2337 rbd_obj_request_put(obj_request);
2339 rbd_assert(orig_request);
2340 rbd_assert(orig_request->img_request);
2343 * Our only purpose here is to determine whether the object
2344 * exists, and we don't want to treat the non-existence as
2345 * an error. If something else comes back, transfer the
2346 * error to the original request and complete it now.
2349 obj_request_existence_set(orig_request, true);
2350 } else if (result == -ENOENT) {
2351 obj_request_existence_set(orig_request, false);
2352 } else if (result) {
2353 orig_request->result = result;
2358 * Resubmit the original request now that we have recorded
2359 * whether the target object exists.
2361 orig_request->result = rbd_img_obj_request_submit(orig_request);
2363 if (orig_request->result)
2364 rbd_obj_request_complete(orig_request);
2365 rbd_obj_request_put(orig_request);
2368 static int rbd_img_obj_exists_submit(struct rbd_obj_request *obj_request)
2370 struct rbd_obj_request *stat_request;
2371 struct rbd_device *rbd_dev;
2372 struct ceph_osd_client *osdc;
2373 struct page **pages = NULL;
2379 * The response data for a STAT call consists of:
2386 size = sizeof (__le64) + sizeof (__le32) + sizeof (__le32);
2387 page_count = (u32)calc_pages_for(0, size);
2388 pages = ceph_alloc_page_vector(page_count, GFP_KERNEL);
2390 return PTR_ERR(pages);
2393 stat_request = rbd_obj_request_create(obj_request->object_name, 0, 0,
2398 rbd_obj_request_get(obj_request);
2399 stat_request->obj_request = obj_request;
2400 stat_request->pages = pages;
2401 stat_request->page_count = page_count;
2403 rbd_assert(obj_request->img_request);
2404 rbd_dev = obj_request->img_request->rbd_dev;
2405 stat_request->osd_req = rbd_osd_req_create(rbd_dev, false,
2407 if (!stat_request->osd_req)
2409 stat_request->callback = rbd_img_obj_exists_callback;
2411 osd_req_op_init(stat_request->osd_req, 0, CEPH_OSD_OP_STAT);
2412 osd_req_op_raw_data_in_pages(stat_request->osd_req, 0, pages, size, 0,
2414 rbd_osd_req_format_read(stat_request);
2416 osdc = &rbd_dev->rbd_client->client->osdc;
2417 ret = rbd_obj_request_submit(osdc, stat_request);
2420 rbd_obj_request_put(obj_request);
2425 static int rbd_img_obj_request_submit(struct rbd_obj_request *obj_request)
2427 struct rbd_img_request *img_request;
2428 struct rbd_device *rbd_dev;
2431 rbd_assert(obj_request_img_data_test(obj_request));
2433 img_request = obj_request->img_request;
2434 rbd_assert(img_request);
2435 rbd_dev = img_request->rbd_dev;
2438 * Only writes to layered images need special handling.
2439 * Reads and non-layered writes are simple object requests.
2440 * Layered writes that start beyond the end of the overlap
2441 * with the parent have no parent data, so they too are
2442 * simple object requests. Finally, if the target object is
2443 * known to already exist, its parent data has already been
2444 * copied, so a write to the object can also be handled as a
2445 * simple object request.
2447 if (!img_request_write_test(img_request) ||
2448 !img_request_layered_test(img_request) ||
2449 rbd_dev->parent_overlap <= obj_request->img_offset ||
2450 ((known = obj_request_known_test(obj_request)) &&
2451 obj_request_exists_test(obj_request))) {
2453 struct rbd_device *rbd_dev;
2454 struct ceph_osd_client *osdc;
2456 rbd_dev = obj_request->img_request->rbd_dev;
2457 osdc = &rbd_dev->rbd_client->client->osdc;
2459 return rbd_obj_request_submit(osdc, obj_request);
2463 * It's a layered write. The target object might exist but
2464 * we may not know that yet. If we know it doesn't exist,
2465 * start by reading the data for the full target object from
2466 * the parent so we can use it for a copyup to the target.
2469 return rbd_img_obj_parent_read_full(obj_request);
2471 /* We don't know whether the target exists. Go find out. */
2473 return rbd_img_obj_exists_submit(obj_request);
2476 static int rbd_img_request_submit(struct rbd_img_request *img_request)
2478 struct rbd_obj_request *obj_request;
2479 struct rbd_obj_request *next_obj_request;
2481 dout("%s: img %p\n", __func__, img_request);
2482 for_each_obj_request_safe(img_request, obj_request, next_obj_request) {
2485 ret = rbd_img_obj_request_submit(obj_request);
2493 static void rbd_img_parent_read_callback(struct rbd_img_request *img_request)
2495 struct rbd_obj_request *obj_request;
2496 struct rbd_device *rbd_dev;
2499 rbd_assert(img_request_child_test(img_request));
2501 obj_request = img_request->obj_request;
2502 rbd_assert(obj_request);
2503 rbd_assert(obj_request->img_request);
2505 obj_request->result = img_request->result;
2506 if (obj_request->result)
2510 * We need to zero anything beyond the parent overlap
2511 * boundary. Since rbd_img_obj_request_read_callback()
2512 * will zero anything beyond the end of a short read, an
2513 * easy way to do this is to pretend the data from the
2514 * parent came up short--ending at the overlap boundary.
2516 rbd_assert(obj_request->img_offset < U64_MAX - obj_request->length);
2517 obj_end = obj_request->img_offset + obj_request->length;
2518 rbd_dev = obj_request->img_request->rbd_dev;
2519 if (obj_end > rbd_dev->parent_overlap) {
2522 if (obj_request->img_offset < rbd_dev->parent_overlap)
2523 xferred = rbd_dev->parent_overlap -
2524 obj_request->img_offset;
2526 obj_request->xferred = min(img_request->xferred, xferred);
2528 obj_request->xferred = img_request->xferred;
2531 rbd_img_obj_request_read_callback(obj_request);
2532 rbd_obj_request_complete(obj_request);
2535 static void rbd_img_parent_read(struct rbd_obj_request *obj_request)
2537 struct rbd_device *rbd_dev;
2538 struct rbd_img_request *img_request;
2541 rbd_assert(obj_request_img_data_test(obj_request));
2542 rbd_assert(obj_request->img_request != NULL);
2543 rbd_assert(obj_request->result == (s32) -ENOENT);
2544 rbd_assert(obj_request->type == OBJ_REQUEST_BIO);
2546 rbd_dev = obj_request->img_request->rbd_dev;
2547 rbd_assert(rbd_dev->parent != NULL);
2548 /* rbd_read_finish(obj_request, obj_request->length); */
2549 img_request = rbd_img_request_create(rbd_dev->parent,
2550 obj_request->img_offset,
2551 obj_request->length,
2557 rbd_obj_request_get(obj_request);
2558 img_request->obj_request = obj_request;
2560 result = rbd_img_request_fill(img_request, OBJ_REQUEST_BIO,
2561 obj_request->bio_list);
2565 img_request->callback = rbd_img_parent_read_callback;
2566 result = rbd_img_request_submit(img_request);
2573 rbd_img_request_put(img_request);
2574 obj_request->result = result;
2575 obj_request->xferred = 0;
2576 obj_request_done_set(obj_request);
2579 static int rbd_obj_notify_ack(struct rbd_device *rbd_dev, u64 notify_id)
2581 struct rbd_obj_request *obj_request;
2582 struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
2585 obj_request = rbd_obj_request_create(rbd_dev->header_name, 0, 0,
2586 OBJ_REQUEST_NODATA);
2591 obj_request->osd_req = rbd_osd_req_create(rbd_dev, false, obj_request);
2592 if (!obj_request->osd_req)
2594 obj_request->callback = rbd_obj_request_put;
2596 osd_req_op_watch_init(obj_request->osd_req, 0, CEPH_OSD_OP_NOTIFY_ACK,
2598 rbd_osd_req_format_read(obj_request);
2600 ret = rbd_obj_request_submit(osdc, obj_request);
2603 rbd_obj_request_put(obj_request);
2608 static void rbd_watch_cb(u64 ver, u64 notify_id, u8 opcode, void *data)
2610 struct rbd_device *rbd_dev = (struct rbd_device *)data;
2615 dout("%s: \"%s\" notify_id %llu opcode %u\n", __func__,
2616 rbd_dev->header_name, (unsigned long long)notify_id,
2617 (unsigned int)opcode);
2618 (void)rbd_dev_refresh(rbd_dev);
2620 rbd_obj_notify_ack(rbd_dev, notify_id);
2624 * Request sync osd watch/unwatch. The value of "start" determines
2625 * whether a watch request is being initiated or torn down.
2627 static int rbd_dev_header_watch_sync(struct rbd_device *rbd_dev, int start)
2629 struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
2630 struct rbd_obj_request *obj_request;
2633 rbd_assert(start ^ !!rbd_dev->watch_event);
2634 rbd_assert(start ^ !!rbd_dev->watch_request);
2637 ret = ceph_osdc_create_event(osdc, rbd_watch_cb, rbd_dev,
2638 &rbd_dev->watch_event);
2641 rbd_assert(rbd_dev->watch_event != NULL);
2645 obj_request = rbd_obj_request_create(rbd_dev->header_name, 0, 0,
2646 OBJ_REQUEST_NODATA);
2650 obj_request->osd_req = rbd_osd_req_create(rbd_dev, true, obj_request);
2651 if (!obj_request->osd_req)
2655 ceph_osdc_set_request_linger(osdc, obj_request->osd_req);
2657 ceph_osdc_unregister_linger_request(osdc,
2658 rbd_dev->watch_request->osd_req);
2660 osd_req_op_watch_init(obj_request->osd_req, 0, CEPH_OSD_OP_WATCH,
2661 rbd_dev->watch_event->cookie, 0, start);
2662 rbd_osd_req_format_write(obj_request);
2664 ret = rbd_obj_request_submit(osdc, obj_request);
2667 ret = rbd_obj_request_wait(obj_request);
2670 ret = obj_request->result;
2675 * A watch request is set to linger, so the underlying osd
2676 * request won't go away until we unregister it. We retain
2677 * a pointer to the object request during that time (in
2678 * rbd_dev->watch_request), so we'll keep a reference to
2679 * it. We'll drop that reference (below) after we've
2683 rbd_dev->watch_request = obj_request;
2688 /* We have successfully torn down the watch request */
2690 rbd_obj_request_put(rbd_dev->watch_request);
2691 rbd_dev->watch_request = NULL;
2693 /* Cancel the event if we're tearing down, or on error */
2694 ceph_osdc_cancel_event(rbd_dev->watch_event);
2695 rbd_dev->watch_event = NULL;
2697 rbd_obj_request_put(obj_request);
2703 * Synchronous osd object method call. Returns the number of bytes
2704 * returned in the outbound buffer, or a negative error code.
2706 static int rbd_obj_method_sync(struct rbd_device *rbd_dev,
2707 const char *object_name,
2708 const char *class_name,
2709 const char *method_name,
2710 const void *outbound,
2711 size_t outbound_size,
2713 size_t inbound_size)
2715 struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
2716 struct rbd_obj_request *obj_request;
2717 struct page **pages;
2722 * Method calls are ultimately read operations. The result
2723 * should placed into the inbound buffer provided. They
2724 * also supply outbound data--parameters for the object
2725 * method. Currently if this is present it will be a
2728 page_count = (u32)calc_pages_for(0, inbound_size);
2729 pages = ceph_alloc_page_vector(page_count, GFP_KERNEL);
2731 return PTR_ERR(pages);
2734 obj_request = rbd_obj_request_create(object_name, 0, inbound_size,
2739 obj_request->pages = pages;
2740 obj_request->page_count = page_count;
2742 obj_request->osd_req = rbd_osd_req_create(rbd_dev, false, obj_request);
2743 if (!obj_request->osd_req)
2746 osd_req_op_cls_init(obj_request->osd_req, 0, CEPH_OSD_OP_CALL,
2747 class_name, method_name);
2748 if (outbound_size) {
2749 struct ceph_pagelist *pagelist;
2751 pagelist = kmalloc(sizeof (*pagelist), GFP_NOFS);
2755 ceph_pagelist_init(pagelist);
2756 ceph_pagelist_append(pagelist, outbound, outbound_size);
2757 osd_req_op_cls_request_data_pagelist(obj_request->osd_req, 0,
2760 osd_req_op_cls_response_data_pages(obj_request->osd_req, 0,
2761 obj_request->pages, inbound_size,
2763 rbd_osd_req_format_read(obj_request);
2765 ret = rbd_obj_request_submit(osdc, obj_request);
2768 ret = rbd_obj_request_wait(obj_request);
2772 ret = obj_request->result;
2776 rbd_assert(obj_request->xferred < (u64)INT_MAX);
2777 ret = (int)obj_request->xferred;
2778 ceph_copy_from_page_vector(pages, inbound, 0, obj_request->xferred);
2781 rbd_obj_request_put(obj_request);
2783 ceph_release_page_vector(pages, page_count);
2788 static void rbd_request_fn(struct request_queue *q)
2789 __releases(q->queue_lock) __acquires(q->queue_lock)
2791 struct rbd_device *rbd_dev = q->queuedata;
2792 bool read_only = rbd_dev->mapping.read_only;
2796 while ((rq = blk_fetch_request(q))) {
2797 bool write_request = rq_data_dir(rq) == WRITE;
2798 struct rbd_img_request *img_request;
2802 /* Ignore any non-FS requests that filter through. */
2804 if (rq->cmd_type != REQ_TYPE_FS) {
2805 dout("%s: non-fs request type %d\n", __func__,
2806 (int) rq->cmd_type);
2807 __blk_end_request_all(rq, 0);
2811 /* Ignore/skip any zero-length requests */
2813 offset = (u64) blk_rq_pos(rq) << SECTOR_SHIFT;
2814 length = (u64) blk_rq_bytes(rq);
2817 dout("%s: zero-length request\n", __func__);
2818 __blk_end_request_all(rq, 0);
2822 spin_unlock_irq(q->queue_lock);
2824 /* Disallow writes to a read-only device */
2826 if (write_request) {
2830 rbd_assert(rbd_dev->spec->snap_id == CEPH_NOSNAP);
2834 * Quit early if the mapped snapshot no longer
2835 * exists. It's still possible the snapshot will
2836 * have disappeared by the time our request arrives
2837 * at the osd, but there's no sense in sending it if
2840 if (!test_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags)) {
2841 dout("request for non-existent snapshot");
2842 rbd_assert(rbd_dev->spec->snap_id != CEPH_NOSNAP);
2848 if (offset && length > U64_MAX - offset + 1) {
2849 rbd_warn(rbd_dev, "bad request range (%llu~%llu)\n",
2851 goto end_request; /* Shouldn't happen */
2855 img_request = rbd_img_request_create(rbd_dev, offset, length,
2856 write_request, false);
2860 img_request->rq = rq;
2862 result = rbd_img_request_fill(img_request, OBJ_REQUEST_BIO,
2865 result = rbd_img_request_submit(img_request);
2867 rbd_img_request_put(img_request);
2869 spin_lock_irq(q->queue_lock);
2871 rbd_warn(rbd_dev, "%s %llx at %llx result %d\n",
2872 write_request ? "write" : "read",
2873 length, offset, result);
2875 __blk_end_request_all(rq, result);
2881 * a queue callback. Makes sure that we don't create a bio that spans across
2882 * multiple osd objects. One exception would be with a single page bios,
2883 * which we handle later at bio_chain_clone_range()
2885 static int rbd_merge_bvec(struct request_queue *q, struct bvec_merge_data *bmd,
2886 struct bio_vec *bvec)
2888 struct rbd_device *rbd_dev = q->queuedata;
2889 sector_t sector_offset;
2890 sector_t sectors_per_obj;
2891 sector_t obj_sector_offset;
2895 * Find how far into its rbd object the partition-relative
2896 * bio start sector is to offset relative to the enclosing
2899 sector_offset = get_start_sect(bmd->bi_bdev) + bmd->bi_sector;
2900 sectors_per_obj = 1 << (rbd_dev->header.obj_order - SECTOR_SHIFT);
2901 obj_sector_offset = sector_offset & (sectors_per_obj - 1);
2904 * Compute the number of bytes from that offset to the end
2905 * of the object. Account for what's already used by the bio.
2907 ret = (int) (sectors_per_obj - obj_sector_offset) << SECTOR_SHIFT;
2908 if (ret > bmd->bi_size)
2909 ret -= bmd->bi_size;
2914 * Don't send back more than was asked for. And if the bio
2915 * was empty, let the whole thing through because: "Note
2916 * that a block device *must* allow a single page to be
2917 * added to an empty bio."
2919 rbd_assert(bvec->bv_len <= PAGE_SIZE);
2920 if (ret > (int) bvec->bv_len || !bmd->bi_size)
2921 ret = (int) bvec->bv_len;
2926 static void rbd_free_disk(struct rbd_device *rbd_dev)
2928 struct gendisk *disk = rbd_dev->disk;
2933 rbd_dev->disk = NULL;
2934 if (disk->flags & GENHD_FL_UP) {
2937 blk_cleanup_queue(disk->queue);
2942 static int rbd_obj_read_sync(struct rbd_device *rbd_dev,
2943 const char *object_name,
2944 u64 offset, u64 length, void *buf)
2947 struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
2948 struct rbd_obj_request *obj_request;
2949 struct page **pages = NULL;
2954 page_count = (u32) calc_pages_for(offset, length);
2955 pages = ceph_alloc_page_vector(page_count, GFP_KERNEL);
2957 ret = PTR_ERR(pages);
2960 obj_request = rbd_obj_request_create(object_name, offset, length,
2965 obj_request->pages = pages;
2966 obj_request->page_count = page_count;
2968 obj_request->osd_req = rbd_osd_req_create(rbd_dev, false, obj_request);
2969 if (!obj_request->osd_req)
2972 osd_req_op_extent_init(obj_request->osd_req, 0, CEPH_OSD_OP_READ,
2973 offset, length, 0, 0);
2974 osd_req_op_extent_osd_data_pages(obj_request->osd_req, 0,
2976 obj_request->length,
2977 obj_request->offset & ~PAGE_MASK,
2979 rbd_osd_req_format_read(obj_request);
2981 ret = rbd_obj_request_submit(osdc, obj_request);
2984 ret = rbd_obj_request_wait(obj_request);
2988 ret = obj_request->result;
2992 rbd_assert(obj_request->xferred <= (u64) SIZE_MAX);
2993 size = (size_t) obj_request->xferred;
2994 ceph_copy_from_page_vector(pages, buf, 0, size);
2995 rbd_assert(size <= (size_t)INT_MAX);
2999 rbd_obj_request_put(obj_request);
3001 ceph_release_page_vector(pages, page_count);
3007 * Read the complete header for the given rbd device.
3009 * Returns a pointer to a dynamically-allocated buffer containing
3010 * the complete and validated header. Caller can pass the address
3011 * of a variable that will be filled in with the version of the
3012 * header object at the time it was read.
3014 * Returns a pointer-coded errno if a failure occurs.
3016 static struct rbd_image_header_ondisk *
3017 rbd_dev_v1_header_read(struct rbd_device *rbd_dev)
3019 struct rbd_image_header_ondisk *ondisk = NULL;
3026 * The complete header will include an array of its 64-bit
3027 * snapshot ids, followed by the names of those snapshots as
3028 * a contiguous block of NUL-terminated strings. Note that
3029 * the number of snapshots could change by the time we read
3030 * it in, in which case we re-read it.
3037 size = sizeof (*ondisk);
3038 size += snap_count * sizeof (struct rbd_image_snap_ondisk);
3040 ondisk = kmalloc(size, GFP_KERNEL);
3042 return ERR_PTR(-ENOMEM);
3044 ret = rbd_obj_read_sync(rbd_dev, rbd_dev->header_name,
3048 if ((size_t)ret < size) {
3050 rbd_warn(rbd_dev, "short header read (want %zd got %d)",
3054 if (!rbd_dev_ondisk_valid(ondisk)) {
3056 rbd_warn(rbd_dev, "invalid header");
3060 names_size = le64_to_cpu(ondisk->snap_names_len);
3061 want_count = snap_count;
3062 snap_count = le32_to_cpu(ondisk->snap_count);
3063 } while (snap_count != want_count);
3070 return ERR_PTR(ret);
3074 * reload the ondisk the header
3076 static int rbd_read_header(struct rbd_device *rbd_dev,
3077 struct rbd_image_header *header)
3079 struct rbd_image_header_ondisk *ondisk;
3082 ondisk = rbd_dev_v1_header_read(rbd_dev);
3084 return PTR_ERR(ondisk);
3085 ret = rbd_header_from_disk(header, ondisk);
3091 static void rbd_update_mapping_size(struct rbd_device *rbd_dev)
3093 if (rbd_dev->spec->snap_id != CEPH_NOSNAP)
3096 if (rbd_dev->mapping.size != rbd_dev->header.image_size) {
3099 rbd_dev->mapping.size = rbd_dev->header.image_size;
3100 size = (sector_t)rbd_dev->mapping.size / SECTOR_SIZE;
3101 dout("setting size to %llu sectors", (unsigned long long)size);
3102 set_capacity(rbd_dev->disk, size);
3107 * only read the first part of the ondisk header, without the snaps info
3109 static int rbd_dev_v1_refresh(struct rbd_device *rbd_dev)
3112 struct rbd_image_header h;
3114 ret = rbd_read_header(rbd_dev, &h);
3118 down_write(&rbd_dev->header_rwsem);
3120 /* Update image size, and check for resize of mapped image */
3121 rbd_dev->header.image_size = h.image_size;
3122 rbd_update_mapping_size(rbd_dev);
3124 /* rbd_dev->header.object_prefix shouldn't change */
3125 kfree(rbd_dev->header.snap_sizes);
3126 kfree(rbd_dev->header.snap_names);
3127 /* osd requests may still refer to snapc */
3128 ceph_put_snap_context(rbd_dev->header.snapc);
3130 rbd_dev->header.image_size = h.image_size;
3131 rbd_dev->header.snapc = h.snapc;
3132 rbd_dev->header.snap_names = h.snap_names;
3133 rbd_dev->header.snap_sizes = h.snap_sizes;
3134 /* Free the extra copy of the object prefix */
3135 if (strcmp(rbd_dev->header.object_prefix, h.object_prefix))
3136 rbd_warn(rbd_dev, "object prefix changed (ignoring)");
3137 kfree(h.object_prefix);
3139 up_write(&rbd_dev->header_rwsem);
3145 * Clear the rbd device's EXISTS flag if the snapshot it's mapped to
3146 * has disappeared from the (just updated) snapshot context.
3148 static void rbd_exists_validate(struct rbd_device *rbd_dev)
3152 if (!test_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags))
3155 snap_id = rbd_dev->spec->snap_id;
3156 if (snap_id == CEPH_NOSNAP)
3159 if (rbd_dev_snap_index(rbd_dev, snap_id) == BAD_SNAP_INDEX)
3160 clear_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags);
3163 static int rbd_dev_refresh(struct rbd_device *rbd_dev)
3168 rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
3169 image_size = rbd_dev->header.image_size;
3170 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
3171 if (rbd_dev->image_format == 1)
3172 ret = rbd_dev_v1_refresh(rbd_dev);
3174 ret = rbd_dev_v2_refresh(rbd_dev);
3176 /* If it's a mapped snapshot, validate its EXISTS flag */
3178 rbd_exists_validate(rbd_dev);
3179 mutex_unlock(&ctl_mutex);
3181 rbd_warn(rbd_dev, "got notification but failed to "
3182 " update snaps: %d\n", ret);
3183 if (image_size != rbd_dev->header.image_size)
3184 revalidate_disk(rbd_dev->disk);
3189 static int rbd_init_disk(struct rbd_device *rbd_dev)
3191 struct gendisk *disk;
3192 struct request_queue *q;
3195 /* create gendisk info */
3196 disk = alloc_disk(RBD_MINORS_PER_MAJOR);
3200 snprintf(disk->disk_name, sizeof(disk->disk_name), RBD_DRV_NAME "%d",
3202 disk->major = rbd_dev->major;
3203 disk->first_minor = 0;
3204 disk->fops = &rbd_bd_ops;
3205 disk->private_data = rbd_dev;
3207 q = blk_init_queue(rbd_request_fn, &rbd_dev->lock);
3211 /* We use the default size, but let's be explicit about it. */
3212 blk_queue_physical_block_size(q, SECTOR_SIZE);
3214 /* set io sizes to object size */
3215 segment_size = rbd_obj_bytes(&rbd_dev->header);
3216 blk_queue_max_hw_sectors(q, segment_size / SECTOR_SIZE);
3217 blk_queue_max_segment_size(q, segment_size);
3218 blk_queue_io_min(q, segment_size);
3219 blk_queue_io_opt(q, segment_size);
3221 blk_queue_merge_bvec(q, rbd_merge_bvec);
3224 q->queuedata = rbd_dev;
3226 rbd_dev->disk = disk;
3239 static struct rbd_device *dev_to_rbd_dev(struct device *dev)
3241 return container_of(dev, struct rbd_device, dev);
3244 static ssize_t rbd_size_show(struct device *dev,
3245 struct device_attribute *attr, char *buf)
3247 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3249 return sprintf(buf, "%llu\n",
3250 (unsigned long long)rbd_dev->mapping.size);
3254 * Note this shows the features for whatever's mapped, which is not
3255 * necessarily the base image.
3257 static ssize_t rbd_features_show(struct device *dev,
3258 struct device_attribute *attr, char *buf)
3260 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3262 return sprintf(buf, "0x%016llx\n",
3263 (unsigned long long)rbd_dev->mapping.features);
3266 static ssize_t rbd_major_show(struct device *dev,
3267 struct device_attribute *attr, char *buf)
3269 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3272 return sprintf(buf, "%d\n", rbd_dev->major);
3274 return sprintf(buf, "(none)\n");
3278 static ssize_t rbd_client_id_show(struct device *dev,
3279 struct device_attribute *attr, char *buf)
3281 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3283 return sprintf(buf, "client%lld\n",
3284 ceph_client_id(rbd_dev->rbd_client->client));
3287 static ssize_t rbd_pool_show(struct device *dev,
3288 struct device_attribute *attr, char *buf)
3290 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3292 return sprintf(buf, "%s\n", rbd_dev->spec->pool_name);
3295 static ssize_t rbd_pool_id_show(struct device *dev,
3296 struct device_attribute *attr, char *buf)
3298 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3300 return sprintf(buf, "%llu\n",
3301 (unsigned long long) rbd_dev->spec->pool_id);
3304 static ssize_t rbd_name_show(struct device *dev,
3305 struct device_attribute *attr, char *buf)
3307 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3309 if (rbd_dev->spec->image_name)
3310 return sprintf(buf, "%s\n", rbd_dev->spec->image_name);
3312 return sprintf(buf, "(unknown)\n");
3315 static ssize_t rbd_image_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, "%s\n", rbd_dev->spec->image_id);
3324 * Shows the name of the currently-mapped snapshot (or
3325 * RBD_SNAP_HEAD_NAME for the base image).
3327 static ssize_t rbd_snap_show(struct device *dev,
3328 struct device_attribute *attr,
3331 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3333 return sprintf(buf, "%s\n", rbd_dev->spec->snap_name);
3337 * For an rbd v2 image, shows the pool id, image id, and snapshot id
3338 * for the parent image. If there is no parent, simply shows
3339 * "(no parent image)".
3341 static ssize_t rbd_parent_show(struct device *dev,
3342 struct device_attribute *attr,
3345 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3346 struct rbd_spec *spec = rbd_dev->parent_spec;
3351 return sprintf(buf, "(no parent image)\n");
3353 count = sprintf(bufp, "pool_id %llu\npool_name %s\n",
3354 (unsigned long long) spec->pool_id, spec->pool_name);
3359 count = sprintf(bufp, "image_id %s\nimage_name %s\n", spec->image_id,
3360 spec->image_name ? spec->image_name : "(unknown)");
3365 count = sprintf(bufp, "snap_id %llu\nsnap_name %s\n",
3366 (unsigned long long) spec->snap_id, spec->snap_name);
3371 count = sprintf(bufp, "overlap %llu\n", rbd_dev->parent_overlap);
3376 return (ssize_t) (bufp - buf);
3379 static ssize_t rbd_image_refresh(struct device *dev,
3380 struct device_attribute *attr,
3384 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3387 ret = rbd_dev_refresh(rbd_dev);
3389 return ret < 0 ? ret : size;
3392 static DEVICE_ATTR(size, S_IRUGO, rbd_size_show, NULL);
3393 static DEVICE_ATTR(features, S_IRUGO, rbd_features_show, NULL);
3394 static DEVICE_ATTR(major, S_IRUGO, rbd_major_show, NULL);
3395 static DEVICE_ATTR(client_id, S_IRUGO, rbd_client_id_show, NULL);
3396 static DEVICE_ATTR(pool, S_IRUGO, rbd_pool_show, NULL);
3397 static DEVICE_ATTR(pool_id, S_IRUGO, rbd_pool_id_show, NULL);
3398 static DEVICE_ATTR(name, S_IRUGO, rbd_name_show, NULL);
3399 static DEVICE_ATTR(image_id, S_IRUGO, rbd_image_id_show, NULL);
3400 static DEVICE_ATTR(refresh, S_IWUSR, NULL, rbd_image_refresh);
3401 static DEVICE_ATTR(current_snap, S_IRUGO, rbd_snap_show, NULL);
3402 static DEVICE_ATTR(parent, S_IRUGO, rbd_parent_show, NULL);
3404 static struct attribute *rbd_attrs[] = {
3405 &dev_attr_size.attr,
3406 &dev_attr_features.attr,
3407 &dev_attr_major.attr,
3408 &dev_attr_client_id.attr,
3409 &dev_attr_pool.attr,
3410 &dev_attr_pool_id.attr,
3411 &dev_attr_name.attr,
3412 &dev_attr_image_id.attr,
3413 &dev_attr_current_snap.attr,
3414 &dev_attr_parent.attr,
3415 &dev_attr_refresh.attr,
3419 static struct attribute_group rbd_attr_group = {
3423 static const struct attribute_group *rbd_attr_groups[] = {
3428 static void rbd_sysfs_dev_release(struct device *dev)
3432 static struct device_type rbd_device_type = {
3434 .groups = rbd_attr_groups,
3435 .release = rbd_sysfs_dev_release,
3438 static struct rbd_spec *rbd_spec_get(struct rbd_spec *spec)
3440 kref_get(&spec->kref);
3445 static void rbd_spec_free(struct kref *kref);
3446 static void rbd_spec_put(struct rbd_spec *spec)
3449 kref_put(&spec->kref, rbd_spec_free);
3452 static struct rbd_spec *rbd_spec_alloc(void)
3454 struct rbd_spec *spec;
3456 spec = kzalloc(sizeof (*spec), GFP_KERNEL);
3459 kref_init(&spec->kref);
3464 static void rbd_spec_free(struct kref *kref)
3466 struct rbd_spec *spec = container_of(kref, struct rbd_spec, kref);
3468 kfree(spec->pool_name);
3469 kfree(spec->image_id);
3470 kfree(spec->image_name);
3471 kfree(spec->snap_name);
3475 static struct rbd_device *rbd_dev_create(struct rbd_client *rbdc,
3476 struct rbd_spec *spec)
3478 struct rbd_device *rbd_dev;
3480 rbd_dev = kzalloc(sizeof (*rbd_dev), GFP_KERNEL);
3484 spin_lock_init(&rbd_dev->lock);
3486 INIT_LIST_HEAD(&rbd_dev->node);
3487 init_rwsem(&rbd_dev->header_rwsem);
3489 rbd_dev->spec = spec;
3490 rbd_dev->rbd_client = rbdc;
3492 /* Initialize the layout used for all rbd requests */
3494 rbd_dev->layout.fl_stripe_unit = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER);
3495 rbd_dev->layout.fl_stripe_count = cpu_to_le32(1);
3496 rbd_dev->layout.fl_object_size = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER);
3497 rbd_dev->layout.fl_pg_pool = cpu_to_le32((u32) spec->pool_id);
3502 static void rbd_dev_destroy(struct rbd_device *rbd_dev)
3504 rbd_put_client(rbd_dev->rbd_client);
3505 rbd_spec_put(rbd_dev->spec);
3510 * Get the size and object order for an image snapshot, or if
3511 * snap_id is CEPH_NOSNAP, gets this information for the base
3514 static int _rbd_dev_v2_snap_size(struct rbd_device *rbd_dev, u64 snap_id,
3515 u8 *order, u64 *snap_size)
3517 __le64 snapid = cpu_to_le64(snap_id);
3522 } __attribute__ ((packed)) size_buf = { 0 };
3524 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
3526 &snapid, sizeof (snapid),
3527 &size_buf, sizeof (size_buf));
3528 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
3531 if (ret < sizeof (size_buf))
3535 *order = size_buf.order;
3536 *snap_size = le64_to_cpu(size_buf.size);
3538 dout(" snap_id 0x%016llx order = %u, snap_size = %llu\n",
3539 (unsigned long long)snap_id, (unsigned int)*order,
3540 (unsigned long long)*snap_size);
3545 static int rbd_dev_v2_image_size(struct rbd_device *rbd_dev)
3547 return _rbd_dev_v2_snap_size(rbd_dev, CEPH_NOSNAP,
3548 &rbd_dev->header.obj_order,
3549 &rbd_dev->header.image_size);
3552 static int rbd_dev_v2_object_prefix(struct rbd_device *rbd_dev)
3558 reply_buf = kzalloc(RBD_OBJ_PREFIX_LEN_MAX, GFP_KERNEL);
3562 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
3563 "rbd", "get_object_prefix", NULL, 0,
3564 reply_buf, RBD_OBJ_PREFIX_LEN_MAX);
3565 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
3570 rbd_dev->header.object_prefix = ceph_extract_encoded_string(&p,
3571 p + ret, NULL, GFP_NOIO);
3574 if (IS_ERR(rbd_dev->header.object_prefix)) {
3575 ret = PTR_ERR(rbd_dev->header.object_prefix);
3576 rbd_dev->header.object_prefix = NULL;
3578 dout(" object_prefix = %s\n", rbd_dev->header.object_prefix);
3586 static int _rbd_dev_v2_snap_features(struct rbd_device *rbd_dev, u64 snap_id,
3589 __le64 snapid = cpu_to_le64(snap_id);
3593 } __attribute__ ((packed)) features_buf = { 0 };
3597 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
3598 "rbd", "get_features",
3599 &snapid, sizeof (snapid),
3600 &features_buf, sizeof (features_buf));
3601 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
3604 if (ret < sizeof (features_buf))
3607 incompat = le64_to_cpu(features_buf.incompat);
3608 if (incompat & ~RBD_FEATURES_SUPPORTED)
3611 *snap_features = le64_to_cpu(features_buf.features);
3613 dout(" snap_id 0x%016llx features = 0x%016llx incompat = 0x%016llx\n",
3614 (unsigned long long)snap_id,
3615 (unsigned long long)*snap_features,
3616 (unsigned long long)le64_to_cpu(features_buf.incompat));
3621 static int rbd_dev_v2_features(struct rbd_device *rbd_dev)
3623 return _rbd_dev_v2_snap_features(rbd_dev, CEPH_NOSNAP,
3624 &rbd_dev->header.features);
3627 static int rbd_dev_v2_parent_info(struct rbd_device *rbd_dev)
3629 struct rbd_spec *parent_spec;
3631 void *reply_buf = NULL;
3639 parent_spec = rbd_spec_alloc();
3643 size = sizeof (__le64) + /* pool_id */
3644 sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX + /* image_id */
3645 sizeof (__le64) + /* snap_id */
3646 sizeof (__le64); /* overlap */
3647 reply_buf = kmalloc(size, GFP_KERNEL);
3653 snapid = cpu_to_le64(CEPH_NOSNAP);
3654 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
3655 "rbd", "get_parent",
3656 &snapid, sizeof (snapid),
3658 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
3663 end = reply_buf + ret;
3665 ceph_decode_64_safe(&p, end, parent_spec->pool_id, out_err);
3666 if (parent_spec->pool_id == CEPH_NOPOOL)
3667 goto out; /* No parent? No problem. */
3669 /* The ceph file layout needs to fit pool id in 32 bits */
3672 if (parent_spec->pool_id > (u64)U32_MAX) {
3673 rbd_warn(NULL, "parent pool id too large (%llu > %u)\n",
3674 (unsigned long long)parent_spec->pool_id, U32_MAX);
3678 image_id = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL);
3679 if (IS_ERR(image_id)) {
3680 ret = PTR_ERR(image_id);
3683 parent_spec->image_id = image_id;
3684 ceph_decode_64_safe(&p, end, parent_spec->snap_id, out_err);
3685 ceph_decode_64_safe(&p, end, overlap, out_err);
3687 rbd_dev->parent_overlap = overlap;
3688 rbd_dev->parent_spec = parent_spec;
3689 parent_spec = NULL; /* rbd_dev now owns this */
3694 rbd_spec_put(parent_spec);
3699 static int rbd_dev_v2_striping_info(struct rbd_device *rbd_dev)
3703 __le64 stripe_count;
3704 } __attribute__ ((packed)) striping_info_buf = { 0 };
3705 size_t size = sizeof (striping_info_buf);
3712 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
3713 "rbd", "get_stripe_unit_count", NULL, 0,
3714 (char *)&striping_info_buf, size);
3715 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
3722 * We don't actually support the "fancy striping" feature
3723 * (STRIPINGV2) yet, but if the striping sizes are the
3724 * defaults the behavior is the same as before. So find
3725 * out, and only fail if the image has non-default values.
3728 obj_size = (u64)1 << rbd_dev->header.obj_order;
3729 p = &striping_info_buf;
3730 stripe_unit = ceph_decode_64(&p);
3731 if (stripe_unit != obj_size) {
3732 rbd_warn(rbd_dev, "unsupported stripe unit "
3733 "(got %llu want %llu)",
3734 stripe_unit, obj_size);
3737 stripe_count = ceph_decode_64(&p);
3738 if (stripe_count != 1) {
3739 rbd_warn(rbd_dev, "unsupported stripe count "
3740 "(got %llu want 1)", stripe_count);
3743 rbd_dev->header.stripe_unit = stripe_unit;
3744 rbd_dev->header.stripe_count = stripe_count;
3749 static char *rbd_dev_image_name(struct rbd_device *rbd_dev)
3751 size_t image_id_size;
3756 void *reply_buf = NULL;
3758 char *image_name = NULL;
3761 rbd_assert(!rbd_dev->spec->image_name);
3763 len = strlen(rbd_dev->spec->image_id);
3764 image_id_size = sizeof (__le32) + len;
3765 image_id = kmalloc(image_id_size, GFP_KERNEL);
3770 end = image_id + image_id_size;
3771 ceph_encode_string(&p, end, rbd_dev->spec->image_id, (u32)len);
3773 size = sizeof (__le32) + RBD_IMAGE_NAME_LEN_MAX;
3774 reply_buf = kmalloc(size, GFP_KERNEL);
3778 ret = rbd_obj_method_sync(rbd_dev, RBD_DIRECTORY,
3779 "rbd", "dir_get_name",
3780 image_id, image_id_size,
3785 end = reply_buf + ret;
3787 image_name = ceph_extract_encoded_string(&p, end, &len, GFP_KERNEL);
3788 if (IS_ERR(image_name))
3791 dout("%s: name is %s len is %zd\n", __func__, image_name, len);
3799 static u64 rbd_v1_snap_id_by_name(struct rbd_device *rbd_dev, const char *name)
3801 struct ceph_snap_context *snapc = rbd_dev->header.snapc;
3802 const char *snap_name;
3805 /* Skip over names until we find the one we are looking for */
3807 snap_name = rbd_dev->header.snap_names;
3808 while (which < snapc->num_snaps) {
3809 if (!strcmp(name, snap_name))
3810 return snapc->snaps[which];
3811 snap_name += strlen(snap_name) + 1;
3817 static u64 rbd_v2_snap_id_by_name(struct rbd_device *rbd_dev, const char *name)
3819 struct ceph_snap_context *snapc = rbd_dev->header.snapc;
3824 for (which = 0; !found && which < snapc->num_snaps; which++) {
3825 const char *snap_name;
3827 snap_id = snapc->snaps[which];
3828 snap_name = rbd_dev_v2_snap_name(rbd_dev, snap_id);
3829 if (IS_ERR(snap_name))
3831 found = !strcmp(name, snap_name);
3834 return found ? snap_id : CEPH_NOSNAP;
3838 * Assumes name is never RBD_SNAP_HEAD_NAME; returns CEPH_NOSNAP if
3839 * no snapshot by that name is found, or if an error occurs.
3841 static u64 rbd_snap_id_by_name(struct rbd_device *rbd_dev, const char *name)
3843 if (rbd_dev->image_format == 1)
3844 return rbd_v1_snap_id_by_name(rbd_dev, name);
3846 return rbd_v2_snap_id_by_name(rbd_dev, name);
3850 * When an rbd image has a parent image, it is identified by the
3851 * pool, image, and snapshot ids (not names). This function fills
3852 * in the names for those ids. (It's OK if we can't figure out the
3853 * name for an image id, but the pool and snapshot ids should always
3854 * exist and have names.) All names in an rbd spec are dynamically
3857 * When an image being mapped (not a parent) is probed, we have the
3858 * pool name and pool id, image name and image id, and the snapshot
3859 * name. The only thing we're missing is the snapshot id.
3861 static int rbd_dev_spec_update(struct rbd_device *rbd_dev)
3863 struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
3864 struct rbd_spec *spec = rbd_dev->spec;
3865 const char *pool_name;
3866 const char *image_name;
3867 const char *snap_name;
3871 * An image being mapped will have the pool name (etc.), but
3872 * we need to look up the snapshot id.
3874 if (spec->pool_name) {
3875 if (strcmp(spec->snap_name, RBD_SNAP_HEAD_NAME)) {
3878 snap_id = rbd_snap_id_by_name(rbd_dev, spec->snap_name);
3879 if (snap_id == CEPH_NOSNAP)
3881 spec->snap_id = snap_id;
3883 spec->snap_id = CEPH_NOSNAP;
3889 /* Get the pool name; we have to make our own copy of this */
3891 pool_name = ceph_pg_pool_name_by_id(osdc->osdmap, spec->pool_id);
3893 rbd_warn(rbd_dev, "no pool with id %llu", spec->pool_id);
3896 pool_name = kstrdup(pool_name, GFP_KERNEL);
3900 /* Fetch the image name; tolerate failure here */
3902 image_name = rbd_dev_image_name(rbd_dev);
3904 rbd_warn(rbd_dev, "unable to get image name");
3906 /* Look up the snapshot name, and make a copy */
3908 snap_name = rbd_snap_name(rbd_dev, spec->snap_id);
3914 spec->pool_name = pool_name;
3915 spec->image_name = image_name;
3916 spec->snap_name = snap_name;
3926 static int rbd_dev_v2_snap_context(struct rbd_device *rbd_dev)
3935 struct ceph_snap_context *snapc;
3939 * We'll need room for the seq value (maximum snapshot id),
3940 * snapshot count, and array of that many snapshot ids.
3941 * For now we have a fixed upper limit on the number we're
3942 * prepared to receive.
3944 size = sizeof (__le64) + sizeof (__le32) +
3945 RBD_MAX_SNAP_COUNT * sizeof (__le64);
3946 reply_buf = kzalloc(size, GFP_KERNEL);
3950 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
3951 "rbd", "get_snapcontext", NULL, 0,
3953 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
3958 end = reply_buf + ret;
3960 ceph_decode_64_safe(&p, end, seq, out);
3961 ceph_decode_32_safe(&p, end, snap_count, out);
3964 * Make sure the reported number of snapshot ids wouldn't go
3965 * beyond the end of our buffer. But before checking that,
3966 * make sure the computed size of the snapshot context we
3967 * allocate is representable in a size_t.
3969 if (snap_count > (SIZE_MAX - sizeof (struct ceph_snap_context))
3974 if (!ceph_has_room(&p, end, snap_count * sizeof (__le64)))
3978 snapc = ceph_create_snap_context(snap_count, GFP_KERNEL);
3984 for (i = 0; i < snap_count; i++)
3985 snapc->snaps[i] = ceph_decode_64(&p);
3987 rbd_dev->header.snapc = snapc;
3989 dout(" snap context seq = %llu, snap_count = %u\n",
3990 (unsigned long long)seq, (unsigned int)snap_count);
3997 static const char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev,
4008 size = sizeof (__le32) + RBD_MAX_SNAP_NAME_LEN;
4009 reply_buf = kmalloc(size, GFP_KERNEL);
4011 return ERR_PTR(-ENOMEM);
4013 snapid = cpu_to_le64(snap_id);
4014 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
4015 "rbd", "get_snapshot_name",
4016 &snapid, sizeof (snapid),
4018 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
4020 snap_name = ERR_PTR(ret);
4025 end = reply_buf + ret;
4026 snap_name = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL);
4027 if (IS_ERR(snap_name))
4030 dout(" snap_id 0x%016llx snap_name = %s\n",
4031 (unsigned long long)snap_id, snap_name);
4038 static int rbd_dev_v2_refresh(struct rbd_device *rbd_dev)
4042 down_write(&rbd_dev->header_rwsem);
4044 ret = rbd_dev_v2_image_size(rbd_dev);
4047 rbd_update_mapping_size(rbd_dev);
4049 ret = rbd_dev_v2_snap_context(rbd_dev);
4050 dout("rbd_dev_v2_snap_context returned %d\n", ret);
4054 up_write(&rbd_dev->header_rwsem);
4059 static int rbd_bus_add_dev(struct rbd_device *rbd_dev)
4064 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
4066 dev = &rbd_dev->dev;
4067 dev->bus = &rbd_bus_type;
4068 dev->type = &rbd_device_type;
4069 dev->parent = &rbd_root_dev;
4070 dev->release = rbd_dev_device_release;
4071 dev_set_name(dev, "%d", rbd_dev->dev_id);
4072 ret = device_register(dev);
4074 mutex_unlock(&ctl_mutex);
4079 static void rbd_bus_del_dev(struct rbd_device *rbd_dev)
4081 device_unregister(&rbd_dev->dev);
4084 static atomic64_t rbd_dev_id_max = ATOMIC64_INIT(0);
4087 * Get a unique rbd identifier for the given new rbd_dev, and add
4088 * the rbd_dev to the global list. The minimum rbd id is 1.
4090 static void rbd_dev_id_get(struct rbd_device *rbd_dev)
4092 rbd_dev->dev_id = atomic64_inc_return(&rbd_dev_id_max);
4094 spin_lock(&rbd_dev_list_lock);
4095 list_add_tail(&rbd_dev->node, &rbd_dev_list);
4096 spin_unlock(&rbd_dev_list_lock);
4097 dout("rbd_dev %p given dev id %llu\n", rbd_dev,
4098 (unsigned long long) rbd_dev->dev_id);
4102 * Remove an rbd_dev from the global list, and record that its
4103 * identifier is no longer in use.
4105 static void rbd_dev_id_put(struct rbd_device *rbd_dev)
4107 struct list_head *tmp;
4108 int rbd_id = rbd_dev->dev_id;
4111 rbd_assert(rbd_id > 0);
4113 dout("rbd_dev %p released dev id %llu\n", rbd_dev,
4114 (unsigned long long) rbd_dev->dev_id);
4115 spin_lock(&rbd_dev_list_lock);
4116 list_del_init(&rbd_dev->node);
4119 * If the id being "put" is not the current maximum, there
4120 * is nothing special we need to do.
4122 if (rbd_id != atomic64_read(&rbd_dev_id_max)) {
4123 spin_unlock(&rbd_dev_list_lock);
4128 * We need to update the current maximum id. Search the
4129 * list to find out what it is. We're more likely to find
4130 * the maximum at the end, so search the list backward.
4133 list_for_each_prev(tmp, &rbd_dev_list) {
4134 struct rbd_device *rbd_dev;
4136 rbd_dev = list_entry(tmp, struct rbd_device, node);
4137 if (rbd_dev->dev_id > max_id)
4138 max_id = rbd_dev->dev_id;
4140 spin_unlock(&rbd_dev_list_lock);
4143 * The max id could have been updated by rbd_dev_id_get(), in
4144 * which case it now accurately reflects the new maximum.
4145 * Be careful not to overwrite the maximum value in that
4148 atomic64_cmpxchg(&rbd_dev_id_max, rbd_id, max_id);
4149 dout(" max dev id has been reset\n");
4153 * Skips over white space at *buf, and updates *buf to point to the
4154 * first found non-space character (if any). Returns the length of
4155 * the token (string of non-white space characters) found. Note
4156 * that *buf must be terminated with '\0'.
4158 static inline size_t next_token(const char **buf)
4161 * These are the characters that produce nonzero for
4162 * isspace() in the "C" and "POSIX" locales.
4164 const char *spaces = " \f\n\r\t\v";
4166 *buf += strspn(*buf, spaces); /* Find start of token */
4168 return strcspn(*buf, spaces); /* Return token length */
4172 * Finds the next token in *buf, and if the provided token buffer is
4173 * big enough, copies the found token into it. The result, if
4174 * copied, is guaranteed to be terminated with '\0'. Note that *buf
4175 * must be terminated with '\0' on entry.
4177 * Returns the length of the token found (not including the '\0').
4178 * Return value will be 0 if no token is found, and it will be >=
4179 * token_size if the token would not fit.
4181 * The *buf pointer will be updated to point beyond the end of the
4182 * found token. Note that this occurs even if the token buffer is
4183 * too small to hold it.
4185 static inline size_t copy_token(const char **buf,
4191 len = next_token(buf);
4192 if (len < token_size) {
4193 memcpy(token, *buf, len);
4194 *(token + len) = '\0';
4202 * Finds the next token in *buf, dynamically allocates a buffer big
4203 * enough to hold a copy of it, and copies the token into the new
4204 * buffer. The copy is guaranteed to be terminated with '\0'. Note
4205 * that a duplicate buffer is created even for a zero-length token.
4207 * Returns a pointer to the newly-allocated duplicate, or a null
4208 * pointer if memory for the duplicate was not available. If
4209 * the lenp argument is a non-null pointer, the length of the token
4210 * (not including the '\0') is returned in *lenp.
4212 * If successful, the *buf pointer will be updated to point beyond
4213 * the end of the found token.
4215 * Note: uses GFP_KERNEL for allocation.
4217 static inline char *dup_token(const char **buf, size_t *lenp)
4222 len = next_token(buf);
4223 dup = kmemdup(*buf, len + 1, GFP_KERNEL);
4226 *(dup + len) = '\0';
4236 * Parse the options provided for an "rbd add" (i.e., rbd image
4237 * mapping) request. These arrive via a write to /sys/bus/rbd/add,
4238 * and the data written is passed here via a NUL-terminated buffer.
4239 * Returns 0 if successful or an error code otherwise.
4241 * The information extracted from these options is recorded in
4242 * the other parameters which return dynamically-allocated
4245 * The address of a pointer that will refer to a ceph options
4246 * structure. Caller must release the returned pointer using
4247 * ceph_destroy_options() when it is no longer needed.
4249 * Address of an rbd options pointer. Fully initialized by
4250 * this function; caller must release with kfree().
4252 * Address of an rbd image specification pointer. Fully
4253 * initialized by this function based on parsed options.
4254 * Caller must release with rbd_spec_put().
4256 * The options passed take this form:
4257 * <mon_addrs> <options> <pool_name> <image_name> [<snap_id>]
4260 * A comma-separated list of one or more monitor addresses.
4261 * A monitor address is an ip address, optionally followed
4262 * by a port number (separated by a colon).
4263 * I.e.: ip1[:port1][,ip2[:port2]...]
4265 * A comma-separated list of ceph and/or rbd options.
4267 * The name of the rados pool containing the rbd image.
4269 * The name of the image in that pool to map.
4271 * An optional snapshot id. If provided, the mapping will
4272 * present data from the image at the time that snapshot was
4273 * created. The image head is used if no snapshot id is
4274 * provided. Snapshot mappings are always read-only.
4276 static int rbd_add_parse_args(const char *buf,
4277 struct ceph_options **ceph_opts,
4278 struct rbd_options **opts,
4279 struct rbd_spec **rbd_spec)
4283 const char *mon_addrs;
4285 size_t mon_addrs_size;
4286 struct rbd_spec *spec = NULL;
4287 struct rbd_options *rbd_opts = NULL;
4288 struct ceph_options *copts;
4291 /* The first four tokens are required */
4293 len = next_token(&buf);
4295 rbd_warn(NULL, "no monitor address(es) provided");
4299 mon_addrs_size = len + 1;
4303 options = dup_token(&buf, NULL);
4307 rbd_warn(NULL, "no options provided");
4311 spec = rbd_spec_alloc();
4315 spec->pool_name = dup_token(&buf, NULL);
4316 if (!spec->pool_name)
4318 if (!*spec->pool_name) {
4319 rbd_warn(NULL, "no pool name provided");
4323 spec->image_name = dup_token(&buf, NULL);
4324 if (!spec->image_name)
4326 if (!*spec->image_name) {
4327 rbd_warn(NULL, "no image name provided");
4332 * Snapshot name is optional; default is to use "-"
4333 * (indicating the head/no snapshot).
4335 len = next_token(&buf);
4337 buf = RBD_SNAP_HEAD_NAME; /* No snapshot supplied */
4338 len = sizeof (RBD_SNAP_HEAD_NAME) - 1;
4339 } else if (len > RBD_MAX_SNAP_NAME_LEN) {
4340 ret = -ENAMETOOLONG;
4343 snap_name = kmemdup(buf, len + 1, GFP_KERNEL);
4346 *(snap_name + len) = '\0';
4347 spec->snap_name = snap_name;
4349 /* Initialize all rbd options to the defaults */
4351 rbd_opts = kzalloc(sizeof (*rbd_opts), GFP_KERNEL);
4355 rbd_opts->read_only = RBD_READ_ONLY_DEFAULT;
4357 copts = ceph_parse_options(options, mon_addrs,
4358 mon_addrs + mon_addrs_size - 1,
4359 parse_rbd_opts_token, rbd_opts);
4360 if (IS_ERR(copts)) {
4361 ret = PTR_ERR(copts);
4382 * An rbd format 2 image has a unique identifier, distinct from the
4383 * name given to it by the user. Internally, that identifier is
4384 * what's used to specify the names of objects related to the image.
4386 * A special "rbd id" object is used to map an rbd image name to its
4387 * id. If that object doesn't exist, then there is no v2 rbd image
4388 * with the supplied name.
4390 * This function will record the given rbd_dev's image_id field if
4391 * it can be determined, and in that case will return 0. If any
4392 * errors occur a negative errno will be returned and the rbd_dev's
4393 * image_id field will be unchanged (and should be NULL).
4395 static int rbd_dev_image_id(struct rbd_device *rbd_dev)
4404 * When probing a parent image, the image id is already
4405 * known (and the image name likely is not). There's no
4406 * need to fetch the image id again in this case. We
4407 * do still need to set the image format though.
4409 if (rbd_dev->spec->image_id) {
4410 rbd_dev->image_format = *rbd_dev->spec->image_id ? 2 : 1;
4416 * First, see if the format 2 image id file exists, and if
4417 * so, get the image's persistent id from it.
4419 size = sizeof (RBD_ID_PREFIX) + strlen(rbd_dev->spec->image_name);
4420 object_name = kmalloc(size, GFP_NOIO);
4423 sprintf(object_name, "%s%s", RBD_ID_PREFIX, rbd_dev->spec->image_name);
4424 dout("rbd id object name is %s\n", object_name);
4426 /* Response will be an encoded string, which includes a length */
4428 size = sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX;
4429 response = kzalloc(size, GFP_NOIO);
4435 /* If it doesn't exist we'll assume it's a format 1 image */
4437 ret = rbd_obj_method_sync(rbd_dev, object_name,
4438 "rbd", "get_id", NULL, 0,
4439 response, RBD_IMAGE_ID_LEN_MAX);
4440 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
4441 if (ret == -ENOENT) {
4442 image_id = kstrdup("", GFP_KERNEL);
4443 ret = image_id ? 0 : -ENOMEM;
4445 rbd_dev->image_format = 1;
4446 } else if (ret > sizeof (__le32)) {
4449 image_id = ceph_extract_encoded_string(&p, p + ret,
4451 ret = IS_ERR(image_id) ? PTR_ERR(image_id) : 0;
4453 rbd_dev->image_format = 2;
4459 rbd_dev->spec->image_id = image_id;
4460 dout("image_id is %s\n", image_id);
4469 /* Undo whatever state changes are made by v1 or v2 image probe */
4471 static void rbd_dev_unprobe(struct rbd_device *rbd_dev)
4473 struct rbd_image_header *header;
4475 rbd_dev_remove_parent(rbd_dev);
4476 rbd_spec_put(rbd_dev->parent_spec);
4477 rbd_dev->parent_spec = NULL;
4478 rbd_dev->parent_overlap = 0;
4480 /* Free dynamic fields from the header, then zero it out */
4482 header = &rbd_dev->header;
4483 ceph_put_snap_context(header->snapc);
4484 kfree(header->snap_sizes);
4485 kfree(header->snap_names);
4486 kfree(header->object_prefix);
4487 memset(header, 0, sizeof (*header));
4490 static int rbd_dev_v1_probe(struct rbd_device *rbd_dev)
4494 /* Populate rbd image metadata */
4496 ret = rbd_read_header(rbd_dev, &rbd_dev->header);
4500 /* Version 1 images have no parent (no layering) */
4502 rbd_dev->parent_spec = NULL;
4503 rbd_dev->parent_overlap = 0;
4505 dout("discovered version 1 image, header name is %s\n",
4506 rbd_dev->header_name);
4511 kfree(rbd_dev->header_name);
4512 rbd_dev->header_name = NULL;
4513 kfree(rbd_dev->spec->image_id);
4514 rbd_dev->spec->image_id = NULL;
4519 static int rbd_dev_v2_probe(struct rbd_device *rbd_dev)
4523 ret = rbd_dev_v2_image_size(rbd_dev);
4527 /* Get the object prefix (a.k.a. block_name) for the image */
4529 ret = rbd_dev_v2_object_prefix(rbd_dev);
4533 /* Get the and check features for the image */
4535 ret = rbd_dev_v2_features(rbd_dev);
4539 /* If the image supports layering, get the parent info */
4541 if (rbd_dev->header.features & RBD_FEATURE_LAYERING) {
4542 ret = rbd_dev_v2_parent_info(rbd_dev);
4547 * Don't print a warning for parent images. We can
4548 * tell this point because we won't know its pool
4549 * name yet (just its pool id).
4551 if (rbd_dev->spec->pool_name)
4552 rbd_warn(rbd_dev, "WARNING: kernel layering "
4553 "is EXPERIMENTAL!");
4556 /* If the image supports fancy striping, get its parameters */
4558 if (rbd_dev->header.features & RBD_FEATURE_STRIPINGV2) {
4559 ret = rbd_dev_v2_striping_info(rbd_dev);
4564 /* crypto and compression type aren't (yet) supported for v2 images */
4566 rbd_dev->header.crypt_type = 0;
4567 rbd_dev->header.comp_type = 0;
4569 /* Get the snapshot context, plus the header version */
4571 ret = rbd_dev_v2_snap_context(rbd_dev);
4575 dout("discovered version 2 image, header name is %s\n",
4576 rbd_dev->header_name);
4580 rbd_dev->parent_overlap = 0;
4581 rbd_spec_put(rbd_dev->parent_spec);
4582 rbd_dev->parent_spec = NULL;
4583 kfree(rbd_dev->header_name);
4584 rbd_dev->header_name = NULL;
4585 kfree(rbd_dev->header.object_prefix);
4586 rbd_dev->header.object_prefix = NULL;
4591 static int rbd_dev_probe_parent(struct rbd_device *rbd_dev)
4593 struct rbd_device *parent = NULL;
4594 struct rbd_spec *parent_spec;
4595 struct rbd_client *rbdc;
4598 if (!rbd_dev->parent_spec)
4601 * We need to pass a reference to the client and the parent
4602 * spec when creating the parent rbd_dev. Images related by
4603 * parent/child relationships always share both.
4605 parent_spec = rbd_spec_get(rbd_dev->parent_spec);
4606 rbdc = __rbd_get_client(rbd_dev->rbd_client);
4609 parent = rbd_dev_create(rbdc, parent_spec);
4613 ret = rbd_dev_image_probe(parent);
4616 rbd_dev->parent = parent;
4621 rbd_spec_put(rbd_dev->parent_spec);
4622 kfree(rbd_dev->header_name);
4623 rbd_dev_destroy(parent);
4625 rbd_put_client(rbdc);
4626 rbd_spec_put(parent_spec);
4632 static int rbd_dev_device_setup(struct rbd_device *rbd_dev)
4636 ret = rbd_dev_mapping_set(rbd_dev);
4640 /* generate unique id: find highest unique id, add one */
4641 rbd_dev_id_get(rbd_dev);
4643 /* Fill in the device name, now that we have its id. */
4644 BUILD_BUG_ON(DEV_NAME_LEN
4645 < sizeof (RBD_DRV_NAME) + MAX_INT_FORMAT_WIDTH);
4646 sprintf(rbd_dev->name, "%s%d", RBD_DRV_NAME, rbd_dev->dev_id);
4648 /* Get our block major device number. */
4650 ret = register_blkdev(0, rbd_dev->name);
4653 rbd_dev->major = ret;
4655 /* Set up the blkdev mapping. */
4657 ret = rbd_init_disk(rbd_dev);
4659 goto err_out_blkdev;
4661 ret = rbd_bus_add_dev(rbd_dev);
4665 /* Everything's ready. Announce the disk to the world. */
4667 set_capacity(rbd_dev->disk, rbd_dev->mapping.size / SECTOR_SIZE);
4668 set_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags);
4669 add_disk(rbd_dev->disk);
4671 pr_info("%s: added with size 0x%llx\n", rbd_dev->disk->disk_name,
4672 (unsigned long long) rbd_dev->mapping.size);
4677 rbd_free_disk(rbd_dev);
4679 unregister_blkdev(rbd_dev->major, rbd_dev->name);
4681 rbd_dev_id_put(rbd_dev);
4682 rbd_dev_mapping_clear(rbd_dev);
4687 static int rbd_dev_header_name(struct rbd_device *rbd_dev)
4689 struct rbd_spec *spec = rbd_dev->spec;
4692 /* Record the header object name for this rbd image. */
4694 rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
4696 if (rbd_dev->image_format == 1)
4697 size = strlen(spec->image_name) + sizeof (RBD_SUFFIX);
4699 size = sizeof (RBD_HEADER_PREFIX) + strlen(spec->image_id);
4701 rbd_dev->header_name = kmalloc(size, GFP_KERNEL);
4702 if (!rbd_dev->header_name)
4705 if (rbd_dev->image_format == 1)
4706 sprintf(rbd_dev->header_name, "%s%s",
4707 spec->image_name, RBD_SUFFIX);
4709 sprintf(rbd_dev->header_name, "%s%s",
4710 RBD_HEADER_PREFIX, spec->image_id);
4714 static void rbd_dev_image_release(struct rbd_device *rbd_dev)
4718 rbd_dev_unprobe(rbd_dev);
4719 ret = rbd_dev_header_watch_sync(rbd_dev, 0);
4721 rbd_warn(rbd_dev, "failed to cancel watch event (%d)\n", ret);
4722 kfree(rbd_dev->header_name);
4723 rbd_dev->header_name = NULL;
4724 rbd_dev->image_format = 0;
4725 kfree(rbd_dev->spec->image_id);
4726 rbd_dev->spec->image_id = NULL;
4728 rbd_dev_destroy(rbd_dev);
4732 * Probe for the existence of the header object for the given rbd
4733 * device. For format 2 images this includes determining the image
4736 static int rbd_dev_image_probe(struct rbd_device *rbd_dev)
4742 * Get the id from the image id object. If it's not a
4743 * format 2 image, we'll get ENOENT back, and we'll assume
4744 * it's a format 1 image.
4746 ret = rbd_dev_image_id(rbd_dev);
4749 rbd_assert(rbd_dev->spec->image_id);
4750 rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
4752 ret = rbd_dev_header_name(rbd_dev);
4754 goto err_out_format;
4756 ret = rbd_dev_header_watch_sync(rbd_dev, 1);
4758 goto out_header_name;
4760 if (rbd_dev->image_format == 1)
4761 ret = rbd_dev_v1_probe(rbd_dev);
4763 ret = rbd_dev_v2_probe(rbd_dev);
4767 ret = rbd_dev_spec_update(rbd_dev);
4771 ret = rbd_dev_probe_parent(rbd_dev);
4776 rbd_dev_unprobe(rbd_dev);
4778 tmp = rbd_dev_header_watch_sync(rbd_dev, 0);
4780 rbd_warn(rbd_dev, "unable to tear down watch request\n");
4782 kfree(rbd_dev->header_name);
4783 rbd_dev->header_name = NULL;
4785 rbd_dev->image_format = 0;
4786 kfree(rbd_dev->spec->image_id);
4787 rbd_dev->spec->image_id = NULL;
4789 dout("probe failed, returning %d\n", ret);
4794 static ssize_t rbd_add(struct bus_type *bus,
4798 struct rbd_device *rbd_dev = NULL;
4799 struct ceph_options *ceph_opts = NULL;
4800 struct rbd_options *rbd_opts = NULL;
4801 struct rbd_spec *spec = NULL;
4802 struct rbd_client *rbdc;
4803 struct ceph_osd_client *osdc;
4806 if (!try_module_get(THIS_MODULE))
4809 /* parse add command */
4810 rc = rbd_add_parse_args(buf, &ceph_opts, &rbd_opts, &spec);
4812 goto err_out_module;
4814 rbdc = rbd_get_client(ceph_opts);
4819 ceph_opts = NULL; /* rbd_dev client now owns this */
4822 osdc = &rbdc->client->osdc;
4823 rc = ceph_pg_poolid_by_name(osdc->osdmap, spec->pool_name);
4825 goto err_out_client;
4826 spec->pool_id = (u64)rc;
4828 /* The ceph file layout needs to fit pool id in 32 bits */
4830 if (spec->pool_id > (u64)U32_MAX) {
4831 rbd_warn(NULL, "pool id too large (%llu > %u)\n",
4832 (unsigned long long)spec->pool_id, U32_MAX);
4834 goto err_out_client;
4837 rbd_dev = rbd_dev_create(rbdc, spec);
4839 goto err_out_client;
4840 rbdc = NULL; /* rbd_dev now owns this */
4841 spec = NULL; /* rbd_dev now owns this */
4843 rbd_dev->mapping.read_only = rbd_opts->read_only;
4845 rbd_opts = NULL; /* done with this */
4847 rc = rbd_dev_image_probe(rbd_dev);
4849 goto err_out_rbd_dev;
4851 rc = rbd_dev_device_setup(rbd_dev);
4855 rbd_dev_image_release(rbd_dev);
4857 rbd_dev_destroy(rbd_dev);
4859 rbd_put_client(rbdc);
4862 ceph_destroy_options(ceph_opts);
4866 module_put(THIS_MODULE);
4868 dout("Error adding device %s\n", buf);
4873 static struct rbd_device *__rbd_get_dev(unsigned long dev_id)
4875 struct list_head *tmp;
4876 struct rbd_device *rbd_dev;
4878 spin_lock(&rbd_dev_list_lock);
4879 list_for_each(tmp, &rbd_dev_list) {
4880 rbd_dev = list_entry(tmp, struct rbd_device, node);
4881 if (rbd_dev->dev_id == dev_id) {
4882 spin_unlock(&rbd_dev_list_lock);
4886 spin_unlock(&rbd_dev_list_lock);
4890 static void rbd_dev_device_release(struct device *dev)
4892 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
4894 rbd_free_disk(rbd_dev);
4895 clear_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags);
4896 rbd_dev_clear_mapping(rbd_dev);
4897 unregister_blkdev(rbd_dev->major, rbd_dev->name);
4899 rbd_dev_id_put(rbd_dev);
4900 rbd_dev_mapping_clear(rbd_dev);
4903 static void rbd_dev_remove_parent(struct rbd_device *rbd_dev)
4905 while (rbd_dev->parent) {
4906 struct rbd_device *first = rbd_dev;
4907 struct rbd_device *second = first->parent;
4908 struct rbd_device *third;
4911 * Follow to the parent with no grandparent and
4914 while (second && (third = second->parent)) {
4919 rbd_dev_image_release(second);
4920 first->parent = NULL;
4921 first->parent_overlap = 0;
4923 rbd_assert(first->parent_spec);
4924 rbd_spec_put(first->parent_spec);
4925 first->parent_spec = NULL;
4929 static ssize_t rbd_remove(struct bus_type *bus,
4933 struct rbd_device *rbd_dev = NULL;
4938 ret = strict_strtoul(buf, 10, &ul);
4942 /* convert to int; abort if we lost anything in the conversion */
4943 target_id = (int) ul;
4944 if (target_id != ul)
4947 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
4949 rbd_dev = __rbd_get_dev(target_id);
4955 spin_lock_irq(&rbd_dev->lock);
4956 if (rbd_dev->open_count)
4959 set_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags);
4960 spin_unlock_irq(&rbd_dev->lock);
4964 rbd_bus_del_dev(rbd_dev);
4965 rbd_dev_image_release(rbd_dev);
4966 module_put(THIS_MODULE);
4968 mutex_unlock(&ctl_mutex);
4974 * create control files in sysfs
4977 static int rbd_sysfs_init(void)
4981 ret = device_register(&rbd_root_dev);
4985 ret = bus_register(&rbd_bus_type);
4987 device_unregister(&rbd_root_dev);
4992 static void rbd_sysfs_cleanup(void)
4994 bus_unregister(&rbd_bus_type);
4995 device_unregister(&rbd_root_dev);
4998 static int rbd_slab_init(void)
5000 rbd_assert(!rbd_img_request_cache);
5001 rbd_img_request_cache = kmem_cache_create("rbd_img_request",
5002 sizeof (struct rbd_img_request),
5003 __alignof__(struct rbd_img_request),
5005 if (rbd_img_request_cache)
5011 static void rbd_slab_exit(void)
5013 rbd_assert(rbd_img_request_cache);
5014 kmem_cache_destroy(rbd_img_request_cache);
5015 rbd_img_request_cache = NULL;
5018 static int __init rbd_init(void)
5022 if (!libceph_compatible(NULL)) {
5023 rbd_warn(NULL, "libceph incompatibility (quitting)");
5027 rc = rbd_slab_init();
5030 rc = rbd_sysfs_init();
5034 pr_info("loaded " RBD_DRV_NAME_LONG "\n");
5039 static void __exit rbd_exit(void)
5041 rbd_sysfs_cleanup();
5045 module_init(rbd_init);
5046 module_exit(rbd_exit);
5048 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
5049 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
5050 MODULE_DESCRIPTION("rados block device");
5052 /* following authorship retained from original osdblk.c */
5053 MODULE_AUTHOR("Jeff Garzik <jeff@garzik.org>");
5055 MODULE_LICENSE("GPL");