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 six fields never change for a given rbd image */
110 u64 features; /* Might be changeable someday? */
112 /* The remaining fields need to be updated occasionally */
114 struct ceph_snap_context *snapc;
115 char *snap_names; /* format 1 only */
116 u64 *snap_sizes; /* format 1 only */
120 * An rbd image specification.
122 * The tuple (pool_id, image_id, snap_id) is sufficient to uniquely
123 * identify an image. Each rbd_dev structure includes a pointer to
124 * an rbd_spec structure that encapsulates this identity.
126 * Each of the id's in an rbd_spec has an associated name. For a
127 * user-mapped image, the names are supplied and the id's associated
128 * with them are looked up. For a layered image, a parent image is
129 * defined by the tuple, and the names are looked up.
131 * An rbd_dev structure contains a parent_spec pointer which is
132 * non-null if the image it represents is a child in a layered
133 * image. This pointer will refer to the rbd_spec structure used
134 * by the parent rbd_dev for its own identity (i.e., the structure
135 * is shared between the parent and child).
137 * Since these structures are populated once, during the discovery
138 * phase of image construction, they are effectively immutable so
139 * we make no effort to synchronize access to them.
141 * Note that code herein does not assume the image name is known (it
142 * could be a null pointer).
146 const char *pool_name;
148 const char *image_id;
149 const char *image_name;
152 const char *snap_name;
158 * an instance of the client. multiple devices may share an rbd client.
161 struct ceph_client *client;
163 struct list_head node;
166 struct rbd_img_request;
167 typedef void (*rbd_img_callback_t)(struct rbd_img_request *);
169 #define BAD_WHICH U32_MAX /* Good which or bad which, which? */
171 struct rbd_obj_request;
172 typedef void (*rbd_obj_callback_t)(struct rbd_obj_request *);
174 enum obj_request_type {
175 OBJ_REQUEST_NODATA, OBJ_REQUEST_BIO, OBJ_REQUEST_PAGES
179 OBJ_REQ_DONE, /* completion flag: not done = 0, done = 1 */
180 OBJ_REQ_IMG_DATA, /* object usage: standalone = 0, image = 1 */
181 OBJ_REQ_KNOWN, /* EXISTS flag valid: no = 0, yes = 1 */
182 OBJ_REQ_EXISTS, /* target exists: no = 0, yes = 1 */
185 struct rbd_obj_request {
186 const char *object_name;
187 u64 offset; /* object start byte */
188 u64 length; /* bytes from offset */
192 * An object request associated with an image will have its
193 * img_data flag set; a standalone object request will not.
195 * A standalone object request will have which == BAD_WHICH
196 * and a null obj_request pointer.
198 * An object request initiated in support of a layered image
199 * object (to check for its existence before a write) will
200 * have which == BAD_WHICH and a non-null obj_request pointer.
202 * Finally, an object request for rbd image data will have
203 * which != BAD_WHICH, and will have a non-null img_request
204 * pointer. The value of which will be in the range
205 * 0..(img_request->obj_request_count-1).
208 struct rbd_obj_request *obj_request; /* STAT op */
210 struct rbd_img_request *img_request;
212 /* links for img_request->obj_requests list */
213 struct list_head links;
216 u32 which; /* posn image request list */
218 enum obj_request_type type;
220 struct bio *bio_list;
226 struct page **copyup_pages;
228 struct ceph_osd_request *osd_req;
230 u64 xferred; /* bytes transferred */
233 rbd_obj_callback_t callback;
234 struct completion completion;
240 IMG_REQ_WRITE, /* I/O direction: read = 0, write = 1 */
241 IMG_REQ_CHILD, /* initiator: block = 0, child image = 1 */
242 IMG_REQ_LAYERED, /* ENOENT handling: normal = 0, layered = 1 */
245 struct rbd_img_request {
246 struct rbd_device *rbd_dev;
247 u64 offset; /* starting image byte offset */
248 u64 length; /* byte count from offset */
251 u64 snap_id; /* for reads */
252 struct ceph_snap_context *snapc; /* for writes */
255 struct request *rq; /* block request */
256 struct rbd_obj_request *obj_request; /* obj req initiator */
258 struct page **copyup_pages;
259 spinlock_t completion_lock;/* protects next_completion */
261 rbd_img_callback_t callback;
262 u64 xferred;/* aggregate bytes transferred */
263 int result; /* first nonzero obj_request result */
265 u32 obj_request_count;
266 struct list_head obj_requests; /* rbd_obj_request structs */
271 #define for_each_obj_request(ireq, oreq) \
272 list_for_each_entry(oreq, &(ireq)->obj_requests, links)
273 #define for_each_obj_request_from(ireq, oreq) \
274 list_for_each_entry_from(oreq, &(ireq)->obj_requests, links)
275 #define for_each_obj_request_safe(ireq, oreq, n) \
276 list_for_each_entry_safe_reverse(oreq, n, &(ireq)->obj_requests, links)
288 int dev_id; /* blkdev unique id */
290 int major; /* blkdev assigned major */
291 struct gendisk *disk; /* blkdev's gendisk and rq */
293 u32 image_format; /* Either 1 or 2 */
294 struct rbd_client *rbd_client;
296 char name[DEV_NAME_LEN]; /* blkdev name, e.g. rbd3 */
298 spinlock_t lock; /* queue, flags, open_count */
300 struct rbd_image_header header;
301 unsigned long flags; /* possibly lock protected */
302 struct rbd_spec *spec;
306 struct ceph_file_layout layout;
308 struct ceph_osd_event *watch_event;
309 struct rbd_obj_request *watch_request;
311 struct rbd_spec *parent_spec;
313 struct rbd_device *parent;
315 /* protects updating the header */
316 struct rw_semaphore header_rwsem;
318 struct rbd_mapping mapping;
320 struct list_head node;
324 unsigned long open_count; /* protected by lock */
328 * Flag bits for rbd_dev->flags. If atomicity is required,
329 * rbd_dev->lock is used to protect access.
331 * Currently, only the "removing" flag (which is coupled with the
332 * "open_count" field) requires atomic access.
335 RBD_DEV_FLAG_EXISTS, /* mapped snapshot has not been deleted */
336 RBD_DEV_FLAG_REMOVING, /* this mapping is being removed */
339 static DEFINE_MUTEX(ctl_mutex); /* Serialize open/close/setup/teardown */
341 static LIST_HEAD(rbd_dev_list); /* devices */
342 static DEFINE_SPINLOCK(rbd_dev_list_lock);
344 static LIST_HEAD(rbd_client_list); /* clients */
345 static DEFINE_SPINLOCK(rbd_client_list_lock);
347 /* Slab caches for frequently-allocated structures */
349 static struct kmem_cache *rbd_img_request_cache;
350 static struct kmem_cache *rbd_obj_request_cache;
351 static struct kmem_cache *rbd_segment_name_cache;
353 static int rbd_img_request_submit(struct rbd_img_request *img_request);
355 static void rbd_dev_device_release(struct device *dev);
357 static ssize_t rbd_add(struct bus_type *bus, const char *buf,
359 static ssize_t rbd_remove(struct bus_type *bus, const char *buf,
361 static int rbd_dev_image_probe(struct rbd_device *rbd_dev, bool read_only);
363 static struct bus_attribute rbd_bus_attrs[] = {
364 __ATTR(add, S_IWUSR, NULL, rbd_add),
365 __ATTR(remove, S_IWUSR, NULL, rbd_remove),
369 static struct bus_type rbd_bus_type = {
371 .bus_attrs = rbd_bus_attrs,
374 static void rbd_root_dev_release(struct device *dev)
378 static struct device rbd_root_dev = {
380 .release = rbd_root_dev_release,
383 static __printf(2, 3)
384 void rbd_warn(struct rbd_device *rbd_dev, const char *fmt, ...)
386 struct va_format vaf;
394 printk(KERN_WARNING "%s: %pV\n", RBD_DRV_NAME, &vaf);
395 else if (rbd_dev->disk)
396 printk(KERN_WARNING "%s: %s: %pV\n",
397 RBD_DRV_NAME, rbd_dev->disk->disk_name, &vaf);
398 else if (rbd_dev->spec && rbd_dev->spec->image_name)
399 printk(KERN_WARNING "%s: image %s: %pV\n",
400 RBD_DRV_NAME, rbd_dev->spec->image_name, &vaf);
401 else if (rbd_dev->spec && rbd_dev->spec->image_id)
402 printk(KERN_WARNING "%s: id %s: %pV\n",
403 RBD_DRV_NAME, rbd_dev->spec->image_id, &vaf);
405 printk(KERN_WARNING "%s: rbd_dev %p: %pV\n",
406 RBD_DRV_NAME, rbd_dev, &vaf);
411 #define rbd_assert(expr) \
412 if (unlikely(!(expr))) { \
413 printk(KERN_ERR "\nAssertion failure in %s() " \
415 "\trbd_assert(%s);\n\n", \
416 __func__, __LINE__, #expr); \
419 #else /* !RBD_DEBUG */
420 # define rbd_assert(expr) ((void) 0)
421 #endif /* !RBD_DEBUG */
423 static int rbd_img_obj_request_submit(struct rbd_obj_request *obj_request);
424 static void rbd_img_parent_read(struct rbd_obj_request *obj_request);
425 static void rbd_dev_remove_parent(struct rbd_device *rbd_dev);
427 static int rbd_dev_refresh(struct rbd_device *rbd_dev);
428 static int rbd_dev_v2_refresh(struct rbd_device *rbd_dev);
429 static const char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev,
431 static int _rbd_dev_v2_snap_size(struct rbd_device *rbd_dev, u64 snap_id,
432 u8 *order, u64 *snap_size);
433 static int _rbd_dev_v2_snap_features(struct rbd_device *rbd_dev, u64 snap_id,
435 static u64 rbd_snap_id_by_name(struct rbd_device *rbd_dev, const char *name);
437 static int rbd_open(struct block_device *bdev, fmode_t mode)
439 struct rbd_device *rbd_dev = bdev->bd_disk->private_data;
440 bool removing = false;
442 if ((mode & FMODE_WRITE) && rbd_dev->mapping.read_only)
445 spin_lock_irq(&rbd_dev->lock);
446 if (test_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags))
449 rbd_dev->open_count++;
450 spin_unlock_irq(&rbd_dev->lock);
454 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
455 (void) get_device(&rbd_dev->dev);
456 set_device_ro(bdev, rbd_dev->mapping.read_only);
457 mutex_unlock(&ctl_mutex);
462 static int rbd_release(struct gendisk *disk, fmode_t mode)
464 struct rbd_device *rbd_dev = disk->private_data;
465 unsigned long open_count_before;
467 spin_lock_irq(&rbd_dev->lock);
468 open_count_before = rbd_dev->open_count--;
469 spin_unlock_irq(&rbd_dev->lock);
470 rbd_assert(open_count_before > 0);
472 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
473 put_device(&rbd_dev->dev);
474 mutex_unlock(&ctl_mutex);
479 static const struct block_device_operations rbd_bd_ops = {
480 .owner = THIS_MODULE,
482 .release = rbd_release,
486 * Initialize an rbd client instance.
489 static struct rbd_client *rbd_client_create(struct ceph_options *ceph_opts)
491 struct rbd_client *rbdc;
494 dout("%s:\n", __func__);
495 rbdc = kmalloc(sizeof(struct rbd_client), GFP_KERNEL);
499 kref_init(&rbdc->kref);
500 INIT_LIST_HEAD(&rbdc->node);
502 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
504 rbdc->client = ceph_create_client(ceph_opts, rbdc, 0, 0);
505 if (IS_ERR(rbdc->client))
507 ceph_opts = NULL; /* Now rbdc->client is responsible for ceph_opts */
509 ret = ceph_open_session(rbdc->client);
513 spin_lock(&rbd_client_list_lock);
514 list_add_tail(&rbdc->node, &rbd_client_list);
515 spin_unlock(&rbd_client_list_lock);
517 mutex_unlock(&ctl_mutex);
518 dout("%s: rbdc %p\n", __func__, rbdc);
523 ceph_destroy_client(rbdc->client);
525 mutex_unlock(&ctl_mutex);
529 ceph_destroy_options(ceph_opts);
530 dout("%s: error %d\n", __func__, ret);
535 static struct rbd_client *__rbd_get_client(struct rbd_client *rbdc)
537 kref_get(&rbdc->kref);
543 * Find a ceph client with specific addr and configuration. If
544 * found, bump its reference count.
546 static struct rbd_client *rbd_client_find(struct ceph_options *ceph_opts)
548 struct rbd_client *client_node;
551 if (ceph_opts->flags & CEPH_OPT_NOSHARE)
554 spin_lock(&rbd_client_list_lock);
555 list_for_each_entry(client_node, &rbd_client_list, node) {
556 if (!ceph_compare_options(ceph_opts, client_node->client)) {
557 __rbd_get_client(client_node);
563 spin_unlock(&rbd_client_list_lock);
565 return found ? client_node : NULL;
575 /* string args above */
578 /* Boolean args above */
582 static match_table_t rbd_opts_tokens = {
584 /* string args above */
585 {Opt_read_only, "read_only"},
586 {Opt_read_only, "ro"}, /* Alternate spelling */
587 {Opt_read_write, "read_write"},
588 {Opt_read_write, "rw"}, /* Alternate spelling */
589 /* Boolean args above */
597 #define RBD_READ_ONLY_DEFAULT false
599 static int parse_rbd_opts_token(char *c, void *private)
601 struct rbd_options *rbd_opts = private;
602 substring_t argstr[MAX_OPT_ARGS];
603 int token, intval, ret;
605 token = match_token(c, rbd_opts_tokens, argstr);
609 if (token < Opt_last_int) {
610 ret = match_int(&argstr[0], &intval);
612 pr_err("bad mount option arg (not int) "
616 dout("got int token %d val %d\n", token, intval);
617 } else if (token > Opt_last_int && token < Opt_last_string) {
618 dout("got string token %d val %s\n", token,
620 } else if (token > Opt_last_string && token < Opt_last_bool) {
621 dout("got Boolean token %d\n", token);
623 dout("got token %d\n", token);
628 rbd_opts->read_only = true;
631 rbd_opts->read_only = false;
641 * Get a ceph client with specific addr and configuration, if one does
642 * not exist create it.
644 static struct rbd_client *rbd_get_client(struct ceph_options *ceph_opts)
646 struct rbd_client *rbdc;
648 rbdc = rbd_client_find(ceph_opts);
649 if (rbdc) /* using an existing client */
650 ceph_destroy_options(ceph_opts);
652 rbdc = rbd_client_create(ceph_opts);
658 * Destroy ceph client
660 * Caller must hold rbd_client_list_lock.
662 static void rbd_client_release(struct kref *kref)
664 struct rbd_client *rbdc = container_of(kref, struct rbd_client, kref);
666 dout("%s: rbdc %p\n", __func__, rbdc);
667 spin_lock(&rbd_client_list_lock);
668 list_del(&rbdc->node);
669 spin_unlock(&rbd_client_list_lock);
671 ceph_destroy_client(rbdc->client);
676 * Drop reference to ceph client node. If it's not referenced anymore, release
679 static void rbd_put_client(struct rbd_client *rbdc)
682 kref_put(&rbdc->kref, rbd_client_release);
685 static bool rbd_image_format_valid(u32 image_format)
687 return image_format == 1 || image_format == 2;
690 static bool rbd_dev_ondisk_valid(struct rbd_image_header_ondisk *ondisk)
695 /* The header has to start with the magic rbd header text */
696 if (memcmp(&ondisk->text, RBD_HEADER_TEXT, sizeof (RBD_HEADER_TEXT)))
699 /* The bio layer requires at least sector-sized I/O */
701 if (ondisk->options.order < SECTOR_SHIFT)
704 /* If we use u64 in a few spots we may be able to loosen this */
706 if (ondisk->options.order > 8 * sizeof (int) - 1)
710 * The size of a snapshot header has to fit in a size_t, and
711 * that limits the number of snapshots.
713 snap_count = le32_to_cpu(ondisk->snap_count);
714 size = SIZE_MAX - sizeof (struct ceph_snap_context);
715 if (snap_count > size / sizeof (__le64))
719 * Not only that, but the size of the entire the snapshot
720 * header must also be representable in a size_t.
722 size -= snap_count * sizeof (__le64);
723 if ((u64) size < le64_to_cpu(ondisk->snap_names_len))
730 * Create a new header structure, translate header format from the on-disk
733 static int rbd_header_from_disk(struct rbd_image_header *header,
734 struct rbd_image_header_ondisk *ondisk)
741 memset(header, 0, sizeof (*header));
743 snap_count = le32_to_cpu(ondisk->snap_count);
745 len = strnlen(ondisk->object_prefix, sizeof (ondisk->object_prefix));
746 header->object_prefix = kmalloc(len + 1, GFP_KERNEL);
747 if (!header->object_prefix)
749 memcpy(header->object_prefix, ondisk->object_prefix, len);
750 header->object_prefix[len] = '\0';
753 u64 snap_names_len = le64_to_cpu(ondisk->snap_names_len);
755 /* Save a copy of the snapshot names */
757 if (snap_names_len > (u64) SIZE_MAX)
759 header->snap_names = kmalloc(snap_names_len, GFP_KERNEL);
760 if (!header->snap_names)
763 * Note that rbd_dev_v1_header_read() guarantees
764 * the ondisk buffer we're working with has
765 * snap_names_len bytes beyond the end of the
766 * snapshot id array, this memcpy() is safe.
768 memcpy(header->snap_names, &ondisk->snaps[snap_count],
771 /* Record each snapshot's size */
773 size = snap_count * sizeof (*header->snap_sizes);
774 header->snap_sizes = kmalloc(size, GFP_KERNEL);
775 if (!header->snap_sizes)
777 for (i = 0; i < snap_count; i++)
778 header->snap_sizes[i] =
779 le64_to_cpu(ondisk->snaps[i].image_size);
781 header->snap_names = NULL;
782 header->snap_sizes = NULL;
785 header->features = 0; /* No features support in v1 images */
786 header->obj_order = ondisk->options.order;
787 header->crypt_type = ondisk->options.crypt_type;
788 header->comp_type = ondisk->options.comp_type;
790 /* Allocate and fill in the snapshot context */
792 header->image_size = le64_to_cpu(ondisk->image_size);
794 header->snapc = ceph_create_snap_context(snap_count, GFP_KERNEL);
797 header->snapc->seq = le64_to_cpu(ondisk->snap_seq);
798 for (i = 0; i < snap_count; i++)
799 header->snapc->snaps[i] = le64_to_cpu(ondisk->snaps[i].id);
804 kfree(header->snap_sizes);
805 header->snap_sizes = NULL;
806 kfree(header->snap_names);
807 header->snap_names = NULL;
808 kfree(header->object_prefix);
809 header->object_prefix = NULL;
814 static const char *_rbd_dev_v1_snap_name(struct rbd_device *rbd_dev, u32 which)
816 const char *snap_name;
818 rbd_assert(which < rbd_dev->header.snapc->num_snaps);
820 /* Skip over names until we find the one we are looking for */
822 snap_name = rbd_dev->header.snap_names;
824 snap_name += strlen(snap_name) + 1;
826 return kstrdup(snap_name, GFP_KERNEL);
830 * Snapshot id comparison function for use with qsort()/bsearch().
831 * Note that result is for snapshots in *descending* order.
833 static int snapid_compare_reverse(const void *s1, const void *s2)
835 u64 snap_id1 = *(u64 *)s1;
836 u64 snap_id2 = *(u64 *)s2;
838 if (snap_id1 < snap_id2)
840 return snap_id1 == snap_id2 ? 0 : -1;
844 * Search a snapshot context to see if the given snapshot id is
847 * Returns the position of the snapshot id in the array if it's found,
848 * or BAD_SNAP_INDEX otherwise.
850 * Note: The snapshot array is in kept sorted (by the osd) in
851 * reverse order, highest snapshot id first.
853 static u32 rbd_dev_snap_index(struct rbd_device *rbd_dev, u64 snap_id)
855 struct ceph_snap_context *snapc = rbd_dev->header.snapc;
858 found = bsearch(&snap_id, &snapc->snaps, snapc->num_snaps,
859 sizeof (snap_id), snapid_compare_reverse);
861 return found ? (u32)(found - &snapc->snaps[0]) : BAD_SNAP_INDEX;
864 static const char *rbd_dev_v1_snap_name(struct rbd_device *rbd_dev,
869 which = rbd_dev_snap_index(rbd_dev, snap_id);
870 if (which == BAD_SNAP_INDEX)
873 return _rbd_dev_v1_snap_name(rbd_dev, which);
876 static const char *rbd_snap_name(struct rbd_device *rbd_dev, u64 snap_id)
878 if (snap_id == CEPH_NOSNAP)
879 return RBD_SNAP_HEAD_NAME;
881 rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
882 if (rbd_dev->image_format == 1)
883 return rbd_dev_v1_snap_name(rbd_dev, snap_id);
885 return rbd_dev_v2_snap_name(rbd_dev, snap_id);
888 static int rbd_snap_size(struct rbd_device *rbd_dev, u64 snap_id,
891 rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
892 if (snap_id == CEPH_NOSNAP) {
893 *snap_size = rbd_dev->header.image_size;
894 } else if (rbd_dev->image_format == 1) {
897 which = rbd_dev_snap_index(rbd_dev, snap_id);
898 if (which == BAD_SNAP_INDEX)
901 *snap_size = rbd_dev->header.snap_sizes[which];
906 ret = _rbd_dev_v2_snap_size(rbd_dev, snap_id, NULL, &size);
915 static int rbd_snap_features(struct rbd_device *rbd_dev, u64 snap_id,
918 rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
919 if (snap_id == CEPH_NOSNAP) {
920 *snap_features = rbd_dev->header.features;
921 } else if (rbd_dev->image_format == 1) {
922 *snap_features = 0; /* No features for format 1 */
927 ret = _rbd_dev_v2_snap_features(rbd_dev, snap_id, &features);
931 *snap_features = features;
936 static int rbd_dev_mapping_set(struct rbd_device *rbd_dev)
938 u64 snap_id = rbd_dev->spec->snap_id;
943 ret = rbd_snap_size(rbd_dev, snap_id, &size);
946 ret = rbd_snap_features(rbd_dev, snap_id, &features);
950 rbd_dev->mapping.size = size;
951 rbd_dev->mapping.features = features;
956 static void rbd_dev_mapping_clear(struct rbd_device *rbd_dev)
958 rbd_dev->mapping.size = 0;
959 rbd_dev->mapping.features = 0;
962 static const char *rbd_segment_name(struct rbd_device *rbd_dev, u64 offset)
968 name = kmem_cache_alloc(rbd_segment_name_cache, GFP_NOIO);
971 segment = offset >> rbd_dev->header.obj_order;
972 ret = snprintf(name, MAX_OBJ_NAME_SIZE + 1, "%s.%012llx",
973 rbd_dev->header.object_prefix, segment);
974 if (ret < 0 || ret > MAX_OBJ_NAME_SIZE) {
975 pr_err("error formatting segment name for #%llu (%d)\n",
984 static void rbd_segment_name_free(const char *name)
986 /* The explicit cast here is needed to drop the const qualifier */
988 kmem_cache_free(rbd_segment_name_cache, (void *)name);
991 static u64 rbd_segment_offset(struct rbd_device *rbd_dev, u64 offset)
993 u64 segment_size = (u64) 1 << rbd_dev->header.obj_order;
995 return offset & (segment_size - 1);
998 static u64 rbd_segment_length(struct rbd_device *rbd_dev,
999 u64 offset, u64 length)
1001 u64 segment_size = (u64) 1 << rbd_dev->header.obj_order;
1003 offset &= segment_size - 1;
1005 rbd_assert(length <= U64_MAX - offset);
1006 if (offset + length > segment_size)
1007 length = segment_size - offset;
1013 * returns the size of an object in the image
1015 static u64 rbd_obj_bytes(struct rbd_image_header *header)
1017 return 1 << header->obj_order;
1024 static void bio_chain_put(struct bio *chain)
1030 chain = chain->bi_next;
1036 * zeros a bio chain, starting at specific offset
1038 static void zero_bio_chain(struct bio *chain, int start_ofs)
1041 unsigned long flags;
1047 bio_for_each_segment(bv, chain, i) {
1048 if (pos + bv->bv_len > start_ofs) {
1049 int remainder = max(start_ofs - pos, 0);
1050 buf = bvec_kmap_irq(bv, &flags);
1051 memset(buf + remainder, 0,
1052 bv->bv_len - remainder);
1053 bvec_kunmap_irq(buf, &flags);
1058 chain = chain->bi_next;
1063 * similar to zero_bio_chain(), zeros data defined by a page array,
1064 * starting at the given byte offset from the start of the array and
1065 * continuing up to the given end offset. The pages array is
1066 * assumed to be big enough to hold all bytes up to the end.
1068 static void zero_pages(struct page **pages, u64 offset, u64 end)
1070 struct page **page = &pages[offset >> PAGE_SHIFT];
1072 rbd_assert(end > offset);
1073 rbd_assert(end - offset <= (u64)SIZE_MAX);
1074 while (offset < end) {
1077 unsigned long flags;
1080 page_offset = (size_t)(offset & ~PAGE_MASK);
1081 length = min(PAGE_SIZE - page_offset, (size_t)(end - offset));
1082 local_irq_save(flags);
1083 kaddr = kmap_atomic(*page);
1084 memset(kaddr + page_offset, 0, length);
1085 kunmap_atomic(kaddr);
1086 local_irq_restore(flags);
1094 * Clone a portion of a bio, starting at the given byte offset
1095 * and continuing for the number of bytes indicated.
1097 static struct bio *bio_clone_range(struct bio *bio_src,
1098 unsigned int offset,
1106 unsigned short end_idx;
1107 unsigned short vcnt;
1110 /* Handle the easy case for the caller */
1112 if (!offset && len == bio_src->bi_size)
1113 return bio_clone(bio_src, gfpmask);
1115 if (WARN_ON_ONCE(!len))
1117 if (WARN_ON_ONCE(len > bio_src->bi_size))
1119 if (WARN_ON_ONCE(offset > bio_src->bi_size - len))
1122 /* Find first affected segment... */
1125 __bio_for_each_segment(bv, bio_src, idx, 0) {
1126 if (resid < bv->bv_len)
1128 resid -= bv->bv_len;
1132 /* ...and the last affected segment */
1135 __bio_for_each_segment(bv, bio_src, end_idx, idx) {
1136 if (resid <= bv->bv_len)
1138 resid -= bv->bv_len;
1140 vcnt = end_idx - idx + 1;
1142 /* Build the clone */
1144 bio = bio_alloc(gfpmask, (unsigned int) vcnt);
1146 return NULL; /* ENOMEM */
1148 bio->bi_bdev = bio_src->bi_bdev;
1149 bio->bi_sector = bio_src->bi_sector + (offset >> SECTOR_SHIFT);
1150 bio->bi_rw = bio_src->bi_rw;
1151 bio->bi_flags |= 1 << BIO_CLONED;
1154 * Copy over our part of the bio_vec, then update the first
1155 * and last (or only) entries.
1157 memcpy(&bio->bi_io_vec[0], &bio_src->bi_io_vec[idx],
1158 vcnt * sizeof (struct bio_vec));
1159 bio->bi_io_vec[0].bv_offset += voff;
1161 bio->bi_io_vec[0].bv_len -= voff;
1162 bio->bi_io_vec[vcnt - 1].bv_len = resid;
1164 bio->bi_io_vec[0].bv_len = len;
1167 bio->bi_vcnt = vcnt;
1175 * Clone a portion of a bio chain, starting at the given byte offset
1176 * into the first bio in the source chain and continuing for the
1177 * number of bytes indicated. The result is another bio chain of
1178 * exactly the given length, or a null pointer on error.
1180 * The bio_src and offset parameters are both in-out. On entry they
1181 * refer to the first source bio and the offset into that bio where
1182 * the start of data to be cloned is located.
1184 * On return, bio_src is updated to refer to the bio in the source
1185 * chain that contains first un-cloned byte, and *offset will
1186 * contain the offset of that byte within that bio.
1188 static struct bio *bio_chain_clone_range(struct bio **bio_src,
1189 unsigned int *offset,
1193 struct bio *bi = *bio_src;
1194 unsigned int off = *offset;
1195 struct bio *chain = NULL;
1198 /* Build up a chain of clone bios up to the limit */
1200 if (!bi || off >= bi->bi_size || !len)
1201 return NULL; /* Nothing to clone */
1205 unsigned int bi_size;
1209 rbd_warn(NULL, "bio_chain exhausted with %u left", len);
1210 goto out_err; /* EINVAL; ran out of bio's */
1212 bi_size = min_t(unsigned int, bi->bi_size - off, len);
1213 bio = bio_clone_range(bi, off, bi_size, gfpmask);
1215 goto out_err; /* ENOMEM */
1218 end = &bio->bi_next;
1221 if (off == bi->bi_size) {
1232 bio_chain_put(chain);
1238 * The default/initial value for all object request flags is 0. For
1239 * each flag, once its value is set to 1 it is never reset to 0
1242 static void obj_request_img_data_set(struct rbd_obj_request *obj_request)
1244 if (test_and_set_bit(OBJ_REQ_IMG_DATA, &obj_request->flags)) {
1245 struct rbd_device *rbd_dev;
1247 rbd_dev = obj_request->img_request->rbd_dev;
1248 rbd_warn(rbd_dev, "obj_request %p already marked img_data\n",
1253 static bool obj_request_img_data_test(struct rbd_obj_request *obj_request)
1256 return test_bit(OBJ_REQ_IMG_DATA, &obj_request->flags) != 0;
1259 static void obj_request_done_set(struct rbd_obj_request *obj_request)
1261 if (test_and_set_bit(OBJ_REQ_DONE, &obj_request->flags)) {
1262 struct rbd_device *rbd_dev = NULL;
1264 if (obj_request_img_data_test(obj_request))
1265 rbd_dev = obj_request->img_request->rbd_dev;
1266 rbd_warn(rbd_dev, "obj_request %p already marked done\n",
1271 static bool obj_request_done_test(struct rbd_obj_request *obj_request)
1274 return test_bit(OBJ_REQ_DONE, &obj_request->flags) != 0;
1278 * This sets the KNOWN flag after (possibly) setting the EXISTS
1279 * flag. The latter is set based on the "exists" value provided.
1281 * Note that for our purposes once an object exists it never goes
1282 * away again. It's possible that the response from two existence
1283 * checks are separated by the creation of the target object, and
1284 * the first ("doesn't exist") response arrives *after* the second
1285 * ("does exist"). In that case we ignore the second one.
1287 static void obj_request_existence_set(struct rbd_obj_request *obj_request,
1291 set_bit(OBJ_REQ_EXISTS, &obj_request->flags);
1292 set_bit(OBJ_REQ_KNOWN, &obj_request->flags);
1296 static bool obj_request_known_test(struct rbd_obj_request *obj_request)
1299 return test_bit(OBJ_REQ_KNOWN, &obj_request->flags) != 0;
1302 static bool obj_request_exists_test(struct rbd_obj_request *obj_request)
1305 return test_bit(OBJ_REQ_EXISTS, &obj_request->flags) != 0;
1308 static void rbd_obj_request_get(struct rbd_obj_request *obj_request)
1310 dout("%s: obj %p (was %d)\n", __func__, obj_request,
1311 atomic_read(&obj_request->kref.refcount));
1312 kref_get(&obj_request->kref);
1315 static void rbd_obj_request_destroy(struct kref *kref);
1316 static void rbd_obj_request_put(struct rbd_obj_request *obj_request)
1318 rbd_assert(obj_request != NULL);
1319 dout("%s: obj %p (was %d)\n", __func__, obj_request,
1320 atomic_read(&obj_request->kref.refcount));
1321 kref_put(&obj_request->kref, rbd_obj_request_destroy);
1324 static void rbd_img_request_get(struct rbd_img_request *img_request)
1326 dout("%s: img %p (was %d)\n", __func__, img_request,
1327 atomic_read(&img_request->kref.refcount));
1328 kref_get(&img_request->kref);
1331 static void rbd_img_request_destroy(struct kref *kref);
1332 static void rbd_img_request_put(struct rbd_img_request *img_request)
1334 rbd_assert(img_request != NULL);
1335 dout("%s: img %p (was %d)\n", __func__, img_request,
1336 atomic_read(&img_request->kref.refcount));
1337 kref_put(&img_request->kref, rbd_img_request_destroy);
1340 static inline void rbd_img_obj_request_add(struct rbd_img_request *img_request,
1341 struct rbd_obj_request *obj_request)
1343 rbd_assert(obj_request->img_request == NULL);
1345 /* Image request now owns object's original reference */
1346 obj_request->img_request = img_request;
1347 obj_request->which = img_request->obj_request_count;
1348 rbd_assert(!obj_request_img_data_test(obj_request));
1349 obj_request_img_data_set(obj_request);
1350 rbd_assert(obj_request->which != BAD_WHICH);
1351 img_request->obj_request_count++;
1352 list_add_tail(&obj_request->links, &img_request->obj_requests);
1353 dout("%s: img %p obj %p w=%u\n", __func__, img_request, obj_request,
1354 obj_request->which);
1357 static inline void rbd_img_obj_request_del(struct rbd_img_request *img_request,
1358 struct rbd_obj_request *obj_request)
1360 rbd_assert(obj_request->which != BAD_WHICH);
1362 dout("%s: img %p obj %p w=%u\n", __func__, img_request, obj_request,
1363 obj_request->which);
1364 list_del(&obj_request->links);
1365 rbd_assert(img_request->obj_request_count > 0);
1366 img_request->obj_request_count--;
1367 rbd_assert(obj_request->which == img_request->obj_request_count);
1368 obj_request->which = BAD_WHICH;
1369 rbd_assert(obj_request_img_data_test(obj_request));
1370 rbd_assert(obj_request->img_request == img_request);
1371 obj_request->img_request = NULL;
1372 obj_request->callback = NULL;
1373 rbd_obj_request_put(obj_request);
1376 static bool obj_request_type_valid(enum obj_request_type type)
1379 case OBJ_REQUEST_NODATA:
1380 case OBJ_REQUEST_BIO:
1381 case OBJ_REQUEST_PAGES:
1388 static int rbd_obj_request_submit(struct ceph_osd_client *osdc,
1389 struct rbd_obj_request *obj_request)
1391 dout("%s: osdc %p obj %p\n", __func__, osdc, obj_request);
1393 return ceph_osdc_start_request(osdc, obj_request->osd_req, false);
1396 static void rbd_img_request_complete(struct rbd_img_request *img_request)
1399 dout("%s: img %p\n", __func__, img_request);
1402 * If no error occurred, compute the aggregate transfer
1403 * count for the image request. We could instead use
1404 * atomic64_cmpxchg() to update it as each object request
1405 * completes; not clear which way is better off hand.
1407 if (!img_request->result) {
1408 struct rbd_obj_request *obj_request;
1411 for_each_obj_request(img_request, obj_request)
1412 xferred += obj_request->xferred;
1413 img_request->xferred = xferred;
1416 if (img_request->callback)
1417 img_request->callback(img_request);
1419 rbd_img_request_put(img_request);
1422 /* Caller is responsible for rbd_obj_request_destroy(obj_request) */
1424 static int rbd_obj_request_wait(struct rbd_obj_request *obj_request)
1426 dout("%s: obj %p\n", __func__, obj_request);
1428 return wait_for_completion_interruptible(&obj_request->completion);
1432 * The default/initial value for all image request flags is 0. Each
1433 * is conditionally set to 1 at image request initialization time
1434 * and currently never change thereafter.
1436 static void img_request_write_set(struct rbd_img_request *img_request)
1438 set_bit(IMG_REQ_WRITE, &img_request->flags);
1442 static bool img_request_write_test(struct rbd_img_request *img_request)
1445 return test_bit(IMG_REQ_WRITE, &img_request->flags) != 0;
1448 static void img_request_child_set(struct rbd_img_request *img_request)
1450 set_bit(IMG_REQ_CHILD, &img_request->flags);
1454 static bool img_request_child_test(struct rbd_img_request *img_request)
1457 return test_bit(IMG_REQ_CHILD, &img_request->flags) != 0;
1460 static void img_request_layered_set(struct rbd_img_request *img_request)
1462 set_bit(IMG_REQ_LAYERED, &img_request->flags);
1466 static bool img_request_layered_test(struct rbd_img_request *img_request)
1469 return test_bit(IMG_REQ_LAYERED, &img_request->flags) != 0;
1473 rbd_img_obj_request_read_callback(struct rbd_obj_request *obj_request)
1475 u64 xferred = obj_request->xferred;
1476 u64 length = obj_request->length;
1478 dout("%s: obj %p img %p result %d %llu/%llu\n", __func__,
1479 obj_request, obj_request->img_request, obj_request->result,
1482 * ENOENT means a hole in the image. We zero-fill the
1483 * entire length of the request. A short read also implies
1484 * zero-fill to the end of the request. Either way we
1485 * update the xferred count to indicate the whole request
1488 rbd_assert(obj_request->type != OBJ_REQUEST_NODATA);
1489 if (obj_request->result == -ENOENT) {
1490 if (obj_request->type == OBJ_REQUEST_BIO)
1491 zero_bio_chain(obj_request->bio_list, 0);
1493 zero_pages(obj_request->pages, 0, length);
1494 obj_request->result = 0;
1495 obj_request->xferred = length;
1496 } else if (xferred < length && !obj_request->result) {
1497 if (obj_request->type == OBJ_REQUEST_BIO)
1498 zero_bio_chain(obj_request->bio_list, xferred);
1500 zero_pages(obj_request->pages, xferred, length);
1501 obj_request->xferred = length;
1503 obj_request_done_set(obj_request);
1506 static void rbd_obj_request_complete(struct rbd_obj_request *obj_request)
1508 dout("%s: obj %p cb %p\n", __func__, obj_request,
1509 obj_request->callback);
1510 if (obj_request->callback)
1511 obj_request->callback(obj_request);
1513 complete_all(&obj_request->completion);
1516 static void rbd_osd_trivial_callback(struct rbd_obj_request *obj_request)
1518 dout("%s: obj %p\n", __func__, obj_request);
1519 obj_request_done_set(obj_request);
1522 static void rbd_osd_read_callback(struct rbd_obj_request *obj_request)
1524 struct rbd_img_request *img_request = NULL;
1525 struct rbd_device *rbd_dev = NULL;
1526 bool layered = false;
1528 if (obj_request_img_data_test(obj_request)) {
1529 img_request = obj_request->img_request;
1530 layered = img_request && img_request_layered_test(img_request);
1531 rbd_dev = img_request->rbd_dev;
1534 dout("%s: obj %p img %p result %d %llu/%llu\n", __func__,
1535 obj_request, img_request, obj_request->result,
1536 obj_request->xferred, obj_request->length);
1537 if (layered && obj_request->result == -ENOENT &&
1538 obj_request->img_offset < rbd_dev->parent_overlap)
1539 rbd_img_parent_read(obj_request);
1540 else if (img_request)
1541 rbd_img_obj_request_read_callback(obj_request);
1543 obj_request_done_set(obj_request);
1546 static void rbd_osd_write_callback(struct rbd_obj_request *obj_request)
1548 dout("%s: obj %p result %d %llu\n", __func__, obj_request,
1549 obj_request->result, obj_request->length);
1551 * There is no such thing as a successful short write. Set
1552 * it to our originally-requested length.
1554 obj_request->xferred = obj_request->length;
1555 obj_request_done_set(obj_request);
1559 * For a simple stat call there's nothing to do. We'll do more if
1560 * this is part of a write sequence for a layered image.
1562 static void rbd_osd_stat_callback(struct rbd_obj_request *obj_request)
1564 dout("%s: obj %p\n", __func__, obj_request);
1565 obj_request_done_set(obj_request);
1568 static void rbd_osd_req_callback(struct ceph_osd_request *osd_req,
1569 struct ceph_msg *msg)
1571 struct rbd_obj_request *obj_request = osd_req->r_priv;
1574 dout("%s: osd_req %p msg %p\n", __func__, osd_req, msg);
1575 rbd_assert(osd_req == obj_request->osd_req);
1576 if (obj_request_img_data_test(obj_request)) {
1577 rbd_assert(obj_request->img_request);
1578 rbd_assert(obj_request->which != BAD_WHICH);
1580 rbd_assert(obj_request->which == BAD_WHICH);
1583 if (osd_req->r_result < 0)
1584 obj_request->result = osd_req->r_result;
1586 BUG_ON(osd_req->r_num_ops > 2);
1589 * We support a 64-bit length, but ultimately it has to be
1590 * passed to blk_end_request(), which takes an unsigned int.
1592 obj_request->xferred = osd_req->r_reply_op_len[0];
1593 rbd_assert(obj_request->xferred < (u64)UINT_MAX);
1594 opcode = osd_req->r_ops[0].op;
1596 case CEPH_OSD_OP_READ:
1597 rbd_osd_read_callback(obj_request);
1599 case CEPH_OSD_OP_WRITE:
1600 rbd_osd_write_callback(obj_request);
1602 case CEPH_OSD_OP_STAT:
1603 rbd_osd_stat_callback(obj_request);
1605 case CEPH_OSD_OP_CALL:
1606 case CEPH_OSD_OP_NOTIFY_ACK:
1607 case CEPH_OSD_OP_WATCH:
1608 rbd_osd_trivial_callback(obj_request);
1611 rbd_warn(NULL, "%s: unsupported op %hu\n",
1612 obj_request->object_name, (unsigned short) opcode);
1616 if (obj_request_done_test(obj_request))
1617 rbd_obj_request_complete(obj_request);
1620 static void rbd_osd_req_format_read(struct rbd_obj_request *obj_request)
1622 struct rbd_img_request *img_request = obj_request->img_request;
1623 struct ceph_osd_request *osd_req = obj_request->osd_req;
1626 rbd_assert(osd_req != NULL);
1628 snap_id = img_request ? img_request->snap_id : CEPH_NOSNAP;
1629 ceph_osdc_build_request(osd_req, obj_request->offset,
1630 NULL, snap_id, NULL);
1633 static void rbd_osd_req_format_write(struct rbd_obj_request *obj_request)
1635 struct rbd_img_request *img_request = obj_request->img_request;
1636 struct ceph_osd_request *osd_req = obj_request->osd_req;
1637 struct ceph_snap_context *snapc;
1638 struct timespec mtime = CURRENT_TIME;
1640 rbd_assert(osd_req != NULL);
1642 snapc = img_request ? img_request->snapc : NULL;
1643 ceph_osdc_build_request(osd_req, obj_request->offset,
1644 snapc, CEPH_NOSNAP, &mtime);
1647 static struct ceph_osd_request *rbd_osd_req_create(
1648 struct rbd_device *rbd_dev,
1650 struct rbd_obj_request *obj_request)
1652 struct ceph_snap_context *snapc = NULL;
1653 struct ceph_osd_client *osdc;
1654 struct ceph_osd_request *osd_req;
1656 if (obj_request_img_data_test(obj_request)) {
1657 struct rbd_img_request *img_request = obj_request->img_request;
1659 rbd_assert(write_request ==
1660 img_request_write_test(img_request));
1662 snapc = img_request->snapc;
1665 /* Allocate and initialize the request, for the single op */
1667 osdc = &rbd_dev->rbd_client->client->osdc;
1668 osd_req = ceph_osdc_alloc_request(osdc, snapc, 1, false, GFP_ATOMIC);
1670 return NULL; /* ENOMEM */
1673 osd_req->r_flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK;
1675 osd_req->r_flags = CEPH_OSD_FLAG_READ;
1677 osd_req->r_callback = rbd_osd_req_callback;
1678 osd_req->r_priv = obj_request;
1680 osd_req->r_oid_len = strlen(obj_request->object_name);
1681 rbd_assert(osd_req->r_oid_len < sizeof (osd_req->r_oid));
1682 memcpy(osd_req->r_oid, obj_request->object_name, osd_req->r_oid_len);
1684 osd_req->r_file_layout = rbd_dev->layout; /* struct */
1690 * Create a copyup osd request based on the information in the
1691 * object request supplied. A copyup request has two osd ops,
1692 * a copyup method call, and a "normal" write request.
1694 static struct ceph_osd_request *
1695 rbd_osd_req_create_copyup(struct rbd_obj_request *obj_request)
1697 struct rbd_img_request *img_request;
1698 struct ceph_snap_context *snapc;
1699 struct rbd_device *rbd_dev;
1700 struct ceph_osd_client *osdc;
1701 struct ceph_osd_request *osd_req;
1703 rbd_assert(obj_request_img_data_test(obj_request));
1704 img_request = obj_request->img_request;
1705 rbd_assert(img_request);
1706 rbd_assert(img_request_write_test(img_request));
1708 /* Allocate and initialize the request, for the two ops */
1710 snapc = img_request->snapc;
1711 rbd_dev = img_request->rbd_dev;
1712 osdc = &rbd_dev->rbd_client->client->osdc;
1713 osd_req = ceph_osdc_alloc_request(osdc, snapc, 2, false, GFP_ATOMIC);
1715 return NULL; /* ENOMEM */
1717 osd_req->r_flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK;
1718 osd_req->r_callback = rbd_osd_req_callback;
1719 osd_req->r_priv = obj_request;
1721 osd_req->r_oid_len = strlen(obj_request->object_name);
1722 rbd_assert(osd_req->r_oid_len < sizeof (osd_req->r_oid));
1723 memcpy(osd_req->r_oid, obj_request->object_name, osd_req->r_oid_len);
1725 osd_req->r_file_layout = rbd_dev->layout; /* struct */
1731 static void rbd_osd_req_destroy(struct ceph_osd_request *osd_req)
1733 ceph_osdc_put_request(osd_req);
1736 /* object_name is assumed to be a non-null pointer and NUL-terminated */
1738 static struct rbd_obj_request *rbd_obj_request_create(const char *object_name,
1739 u64 offset, u64 length,
1740 enum obj_request_type type)
1742 struct rbd_obj_request *obj_request;
1746 rbd_assert(obj_request_type_valid(type));
1748 size = strlen(object_name) + 1;
1749 name = kmalloc(size, GFP_KERNEL);
1753 obj_request = kmem_cache_zalloc(rbd_obj_request_cache, GFP_KERNEL);
1759 obj_request->object_name = memcpy(name, object_name, size);
1760 obj_request->offset = offset;
1761 obj_request->length = length;
1762 obj_request->flags = 0;
1763 obj_request->which = BAD_WHICH;
1764 obj_request->type = type;
1765 INIT_LIST_HEAD(&obj_request->links);
1766 init_completion(&obj_request->completion);
1767 kref_init(&obj_request->kref);
1769 dout("%s: \"%s\" %llu/%llu %d -> obj %p\n", __func__, object_name,
1770 offset, length, (int)type, obj_request);
1775 static void rbd_obj_request_destroy(struct kref *kref)
1777 struct rbd_obj_request *obj_request;
1779 obj_request = container_of(kref, struct rbd_obj_request, kref);
1781 dout("%s: obj %p\n", __func__, obj_request);
1783 rbd_assert(obj_request->img_request == NULL);
1784 rbd_assert(obj_request->which == BAD_WHICH);
1786 if (obj_request->osd_req)
1787 rbd_osd_req_destroy(obj_request->osd_req);
1789 rbd_assert(obj_request_type_valid(obj_request->type));
1790 switch (obj_request->type) {
1791 case OBJ_REQUEST_NODATA:
1792 break; /* Nothing to do */
1793 case OBJ_REQUEST_BIO:
1794 if (obj_request->bio_list)
1795 bio_chain_put(obj_request->bio_list);
1797 case OBJ_REQUEST_PAGES:
1798 if (obj_request->pages)
1799 ceph_release_page_vector(obj_request->pages,
1800 obj_request->page_count);
1804 kfree(obj_request->object_name);
1805 obj_request->object_name = NULL;
1806 kmem_cache_free(rbd_obj_request_cache, obj_request);
1810 * Caller is responsible for filling in the list of object requests
1811 * that comprises the image request, and the Linux request pointer
1812 * (if there is one).
1814 static struct rbd_img_request *rbd_img_request_create(
1815 struct rbd_device *rbd_dev,
1816 u64 offset, u64 length,
1820 struct rbd_img_request *img_request;
1822 img_request = kmem_cache_alloc(rbd_img_request_cache, GFP_ATOMIC);
1826 if (write_request) {
1827 down_read(&rbd_dev->header_rwsem);
1828 ceph_get_snap_context(rbd_dev->header.snapc);
1829 up_read(&rbd_dev->header_rwsem);
1832 img_request->rq = NULL;
1833 img_request->rbd_dev = rbd_dev;
1834 img_request->offset = offset;
1835 img_request->length = length;
1836 img_request->flags = 0;
1837 if (write_request) {
1838 img_request_write_set(img_request);
1839 img_request->snapc = rbd_dev->header.snapc;
1841 img_request->snap_id = rbd_dev->spec->snap_id;
1844 img_request_child_set(img_request);
1845 if (rbd_dev->parent_spec)
1846 img_request_layered_set(img_request);
1847 spin_lock_init(&img_request->completion_lock);
1848 img_request->next_completion = 0;
1849 img_request->callback = NULL;
1850 img_request->result = 0;
1851 img_request->obj_request_count = 0;
1852 INIT_LIST_HEAD(&img_request->obj_requests);
1853 kref_init(&img_request->kref);
1855 rbd_img_request_get(img_request); /* Avoid a warning */
1856 rbd_img_request_put(img_request); /* TEMPORARY */
1858 dout("%s: rbd_dev %p %s %llu/%llu -> img %p\n", __func__, rbd_dev,
1859 write_request ? "write" : "read", offset, length,
1865 static void rbd_img_request_destroy(struct kref *kref)
1867 struct rbd_img_request *img_request;
1868 struct rbd_obj_request *obj_request;
1869 struct rbd_obj_request *next_obj_request;
1871 img_request = container_of(kref, struct rbd_img_request, kref);
1873 dout("%s: img %p\n", __func__, img_request);
1875 for_each_obj_request_safe(img_request, obj_request, next_obj_request)
1876 rbd_img_obj_request_del(img_request, obj_request);
1877 rbd_assert(img_request->obj_request_count == 0);
1879 if (img_request_write_test(img_request))
1880 ceph_put_snap_context(img_request->snapc);
1882 if (img_request_child_test(img_request))
1883 rbd_obj_request_put(img_request->obj_request);
1885 kmem_cache_free(rbd_img_request_cache, img_request);
1888 static bool rbd_img_obj_end_request(struct rbd_obj_request *obj_request)
1890 struct rbd_img_request *img_request;
1891 unsigned int xferred;
1895 rbd_assert(obj_request_img_data_test(obj_request));
1896 img_request = obj_request->img_request;
1898 rbd_assert(obj_request->xferred <= (u64)UINT_MAX);
1899 xferred = (unsigned int)obj_request->xferred;
1900 result = obj_request->result;
1902 struct rbd_device *rbd_dev = img_request->rbd_dev;
1904 rbd_warn(rbd_dev, "%s %llx at %llx (%llx)\n",
1905 img_request_write_test(img_request) ? "write" : "read",
1906 obj_request->length, obj_request->img_offset,
1907 obj_request->offset);
1908 rbd_warn(rbd_dev, " result %d xferred %x\n",
1910 if (!img_request->result)
1911 img_request->result = result;
1914 /* Image object requests don't own their page array */
1916 if (obj_request->type == OBJ_REQUEST_PAGES) {
1917 obj_request->pages = NULL;
1918 obj_request->page_count = 0;
1921 if (img_request_child_test(img_request)) {
1922 rbd_assert(img_request->obj_request != NULL);
1923 more = obj_request->which < img_request->obj_request_count - 1;
1925 rbd_assert(img_request->rq != NULL);
1926 more = blk_end_request(img_request->rq, result, xferred);
1932 static void rbd_img_obj_callback(struct rbd_obj_request *obj_request)
1934 struct rbd_img_request *img_request;
1935 u32 which = obj_request->which;
1938 rbd_assert(obj_request_img_data_test(obj_request));
1939 img_request = obj_request->img_request;
1941 dout("%s: img %p obj %p\n", __func__, img_request, obj_request);
1942 rbd_assert(img_request != NULL);
1943 rbd_assert(img_request->obj_request_count > 0);
1944 rbd_assert(which != BAD_WHICH);
1945 rbd_assert(which < img_request->obj_request_count);
1946 rbd_assert(which >= img_request->next_completion);
1948 spin_lock_irq(&img_request->completion_lock);
1949 if (which != img_request->next_completion)
1952 for_each_obj_request_from(img_request, obj_request) {
1954 rbd_assert(which < img_request->obj_request_count);
1956 if (!obj_request_done_test(obj_request))
1958 more = rbd_img_obj_end_request(obj_request);
1962 rbd_assert(more ^ (which == img_request->obj_request_count));
1963 img_request->next_completion = which;
1965 spin_unlock_irq(&img_request->completion_lock);
1968 rbd_img_request_complete(img_request);
1972 * Split up an image request into one or more object requests, each
1973 * to a different object. The "type" parameter indicates whether
1974 * "data_desc" is the pointer to the head of a list of bio
1975 * structures, or the base of a page array. In either case this
1976 * function assumes data_desc describes memory sufficient to hold
1977 * all data described by the image request.
1979 static int rbd_img_request_fill(struct rbd_img_request *img_request,
1980 enum obj_request_type type,
1983 struct rbd_device *rbd_dev = img_request->rbd_dev;
1984 struct rbd_obj_request *obj_request = NULL;
1985 struct rbd_obj_request *next_obj_request;
1986 bool write_request = img_request_write_test(img_request);
1987 struct bio *bio_list;
1988 unsigned int bio_offset = 0;
1989 struct page **pages;
1994 dout("%s: img %p type %d data_desc %p\n", __func__, img_request,
1995 (int)type, data_desc);
1997 opcode = write_request ? CEPH_OSD_OP_WRITE : CEPH_OSD_OP_READ;
1998 img_offset = img_request->offset;
1999 resid = img_request->length;
2000 rbd_assert(resid > 0);
2002 if (type == OBJ_REQUEST_BIO) {
2003 bio_list = data_desc;
2004 rbd_assert(img_offset == bio_list->bi_sector << SECTOR_SHIFT);
2006 rbd_assert(type == OBJ_REQUEST_PAGES);
2011 struct ceph_osd_request *osd_req;
2012 const char *object_name;
2016 object_name = rbd_segment_name(rbd_dev, img_offset);
2019 offset = rbd_segment_offset(rbd_dev, img_offset);
2020 length = rbd_segment_length(rbd_dev, img_offset, resid);
2021 obj_request = rbd_obj_request_create(object_name,
2022 offset, length, type);
2023 /* object request has its own copy of the object name */
2024 rbd_segment_name_free(object_name);
2028 if (type == OBJ_REQUEST_BIO) {
2029 unsigned int clone_size;
2031 rbd_assert(length <= (u64)UINT_MAX);
2032 clone_size = (unsigned int)length;
2033 obj_request->bio_list =
2034 bio_chain_clone_range(&bio_list,
2038 if (!obj_request->bio_list)
2041 unsigned int page_count;
2043 obj_request->pages = pages;
2044 page_count = (u32)calc_pages_for(offset, length);
2045 obj_request->page_count = page_count;
2046 if ((offset + length) & ~PAGE_MASK)
2047 page_count--; /* more on last page */
2048 pages += page_count;
2051 osd_req = rbd_osd_req_create(rbd_dev, write_request,
2055 obj_request->osd_req = osd_req;
2056 obj_request->callback = rbd_img_obj_callback;
2058 osd_req_op_extent_init(osd_req, 0, opcode, offset, length,
2060 if (type == OBJ_REQUEST_BIO)
2061 osd_req_op_extent_osd_data_bio(osd_req, 0,
2062 obj_request->bio_list, length);
2064 osd_req_op_extent_osd_data_pages(osd_req, 0,
2065 obj_request->pages, length,
2066 offset & ~PAGE_MASK, false, false);
2069 rbd_osd_req_format_write(obj_request);
2071 rbd_osd_req_format_read(obj_request);
2073 obj_request->img_offset = img_offset;
2074 rbd_img_obj_request_add(img_request, obj_request);
2076 img_offset += length;
2083 rbd_obj_request_put(obj_request);
2085 for_each_obj_request_safe(img_request, obj_request, next_obj_request)
2086 rbd_obj_request_put(obj_request);
2092 rbd_img_obj_copyup_callback(struct rbd_obj_request *obj_request)
2094 struct rbd_img_request *img_request;
2095 struct rbd_device *rbd_dev;
2099 rbd_assert(obj_request->type == OBJ_REQUEST_BIO);
2100 rbd_assert(obj_request_img_data_test(obj_request));
2101 img_request = obj_request->img_request;
2102 rbd_assert(img_request);
2104 rbd_dev = img_request->rbd_dev;
2105 rbd_assert(rbd_dev);
2106 length = (u64)1 << rbd_dev->header.obj_order;
2107 page_count = (u32)calc_pages_for(0, length);
2109 rbd_assert(obj_request->copyup_pages);
2110 ceph_release_page_vector(obj_request->copyup_pages, page_count);
2111 obj_request->copyup_pages = NULL;
2114 * We want the transfer count to reflect the size of the
2115 * original write request. There is no such thing as a
2116 * successful short write, so if the request was successful
2117 * we can just set it to the originally-requested length.
2119 if (!obj_request->result)
2120 obj_request->xferred = obj_request->length;
2122 /* Finish up with the normal image object callback */
2124 rbd_img_obj_callback(obj_request);
2128 rbd_img_obj_parent_read_full_callback(struct rbd_img_request *img_request)
2130 struct rbd_obj_request *orig_request;
2131 struct ceph_osd_request *osd_req;
2132 struct ceph_osd_client *osdc;
2133 struct rbd_device *rbd_dev;
2134 struct page **pages;
2139 rbd_assert(img_request_child_test(img_request));
2141 /* First get what we need from the image request */
2143 pages = img_request->copyup_pages;
2144 rbd_assert(pages != NULL);
2145 img_request->copyup_pages = NULL;
2147 orig_request = img_request->obj_request;
2148 rbd_assert(orig_request != NULL);
2149 rbd_assert(orig_request->type == OBJ_REQUEST_BIO);
2150 result = img_request->result;
2151 obj_size = img_request->length;
2152 xferred = img_request->xferred;
2154 rbd_dev = img_request->rbd_dev;
2155 rbd_assert(rbd_dev);
2156 rbd_assert(obj_size == (u64)1 << rbd_dev->header.obj_order);
2158 rbd_img_request_put(img_request);
2163 /* Allocate the new copyup osd request for the original request */
2166 rbd_assert(!orig_request->osd_req);
2167 osd_req = rbd_osd_req_create_copyup(orig_request);
2170 orig_request->osd_req = osd_req;
2171 orig_request->copyup_pages = pages;
2173 /* Initialize the copyup op */
2175 osd_req_op_cls_init(osd_req, 0, CEPH_OSD_OP_CALL, "rbd", "copyup");
2176 osd_req_op_cls_request_data_pages(osd_req, 0, pages, obj_size, 0,
2179 /* Then the original write request op */
2181 osd_req_op_extent_init(osd_req, 1, CEPH_OSD_OP_WRITE,
2182 orig_request->offset,
2183 orig_request->length, 0, 0);
2184 osd_req_op_extent_osd_data_bio(osd_req, 1, orig_request->bio_list,
2185 orig_request->length);
2187 rbd_osd_req_format_write(orig_request);
2189 /* All set, send it off. */
2191 orig_request->callback = rbd_img_obj_copyup_callback;
2192 osdc = &rbd_dev->rbd_client->client->osdc;
2193 result = rbd_obj_request_submit(osdc, orig_request);
2197 /* Record the error code and complete the request */
2199 orig_request->result = result;
2200 orig_request->xferred = 0;
2201 obj_request_done_set(orig_request);
2202 rbd_obj_request_complete(orig_request);
2206 * Read from the parent image the range of data that covers the
2207 * entire target of the given object request. This is used for
2208 * satisfying a layered image write request when the target of an
2209 * object request from the image request does not exist.
2211 * A page array big enough to hold the returned data is allocated
2212 * and supplied to rbd_img_request_fill() as the "data descriptor."
2213 * When the read completes, this page array will be transferred to
2214 * the original object request for the copyup operation.
2216 * If an error occurs, record it as the result of the original
2217 * object request and mark it done so it gets completed.
2219 static int rbd_img_obj_parent_read_full(struct rbd_obj_request *obj_request)
2221 struct rbd_img_request *img_request = NULL;
2222 struct rbd_img_request *parent_request = NULL;
2223 struct rbd_device *rbd_dev;
2226 struct page **pages = NULL;
2230 rbd_assert(obj_request_img_data_test(obj_request));
2231 rbd_assert(obj_request->type == OBJ_REQUEST_BIO);
2233 img_request = obj_request->img_request;
2234 rbd_assert(img_request != NULL);
2235 rbd_dev = img_request->rbd_dev;
2236 rbd_assert(rbd_dev->parent != NULL);
2239 * First things first. The original osd request is of no
2240 * use to use any more, we'll need a new one that can hold
2241 * the two ops in a copyup request. We'll get that later,
2242 * but for now we can release the old one.
2244 rbd_osd_req_destroy(obj_request->osd_req);
2245 obj_request->osd_req = NULL;
2248 * Determine the byte range covered by the object in the
2249 * child image to which the original request was to be sent.
2251 img_offset = obj_request->img_offset - obj_request->offset;
2252 length = (u64)1 << rbd_dev->header.obj_order;
2255 * There is no defined parent data beyond the parent
2256 * overlap, so limit what we read at that boundary if
2259 if (img_offset + length > rbd_dev->parent_overlap) {
2260 rbd_assert(img_offset < rbd_dev->parent_overlap);
2261 length = rbd_dev->parent_overlap - img_offset;
2265 * Allocate a page array big enough to receive the data read
2268 page_count = (u32)calc_pages_for(0, length);
2269 pages = ceph_alloc_page_vector(page_count, GFP_KERNEL);
2270 if (IS_ERR(pages)) {
2271 result = PTR_ERR(pages);
2277 parent_request = rbd_img_request_create(rbd_dev->parent,
2280 if (!parent_request)
2282 rbd_obj_request_get(obj_request);
2283 parent_request->obj_request = obj_request;
2285 result = rbd_img_request_fill(parent_request, OBJ_REQUEST_PAGES, pages);
2288 parent_request->copyup_pages = pages;
2290 parent_request->callback = rbd_img_obj_parent_read_full_callback;
2291 result = rbd_img_request_submit(parent_request);
2295 parent_request->copyup_pages = NULL;
2296 parent_request->obj_request = NULL;
2297 rbd_obj_request_put(obj_request);
2300 ceph_release_page_vector(pages, page_count);
2302 rbd_img_request_put(parent_request);
2303 obj_request->result = result;
2304 obj_request->xferred = 0;
2305 obj_request_done_set(obj_request);
2310 static void rbd_img_obj_exists_callback(struct rbd_obj_request *obj_request)
2312 struct rbd_obj_request *orig_request;
2315 rbd_assert(!obj_request_img_data_test(obj_request));
2318 * All we need from the object request is the original
2319 * request and the result of the STAT op. Grab those, then
2320 * we're done with the request.
2322 orig_request = obj_request->obj_request;
2323 obj_request->obj_request = NULL;
2324 rbd_assert(orig_request);
2325 rbd_assert(orig_request->img_request);
2327 result = obj_request->result;
2328 obj_request->result = 0;
2330 dout("%s: obj %p for obj %p result %d %llu/%llu\n", __func__,
2331 obj_request, orig_request, result,
2332 obj_request->xferred, obj_request->length);
2333 rbd_obj_request_put(obj_request);
2335 rbd_assert(orig_request);
2336 rbd_assert(orig_request->img_request);
2339 * Our only purpose here is to determine whether the object
2340 * exists, and we don't want to treat the non-existence as
2341 * an error. If something else comes back, transfer the
2342 * error to the original request and complete it now.
2345 obj_request_existence_set(orig_request, true);
2346 } else if (result == -ENOENT) {
2347 obj_request_existence_set(orig_request, false);
2348 } else if (result) {
2349 orig_request->result = result;
2354 * Resubmit the original request now that we have recorded
2355 * whether the target object exists.
2357 orig_request->result = rbd_img_obj_request_submit(orig_request);
2359 if (orig_request->result)
2360 rbd_obj_request_complete(orig_request);
2361 rbd_obj_request_put(orig_request);
2364 static int rbd_img_obj_exists_submit(struct rbd_obj_request *obj_request)
2366 struct rbd_obj_request *stat_request;
2367 struct rbd_device *rbd_dev;
2368 struct ceph_osd_client *osdc;
2369 struct page **pages = NULL;
2375 * The response data for a STAT call consists of:
2382 size = sizeof (__le64) + sizeof (__le32) + sizeof (__le32);
2383 page_count = (u32)calc_pages_for(0, size);
2384 pages = ceph_alloc_page_vector(page_count, GFP_KERNEL);
2386 return PTR_ERR(pages);
2389 stat_request = rbd_obj_request_create(obj_request->object_name, 0, 0,
2394 rbd_obj_request_get(obj_request);
2395 stat_request->obj_request = obj_request;
2396 stat_request->pages = pages;
2397 stat_request->page_count = page_count;
2399 rbd_assert(obj_request->img_request);
2400 rbd_dev = obj_request->img_request->rbd_dev;
2401 stat_request->osd_req = rbd_osd_req_create(rbd_dev, false,
2403 if (!stat_request->osd_req)
2405 stat_request->callback = rbd_img_obj_exists_callback;
2407 osd_req_op_init(stat_request->osd_req, 0, CEPH_OSD_OP_STAT);
2408 osd_req_op_raw_data_in_pages(stat_request->osd_req, 0, pages, size, 0,
2410 rbd_osd_req_format_read(stat_request);
2412 osdc = &rbd_dev->rbd_client->client->osdc;
2413 ret = rbd_obj_request_submit(osdc, stat_request);
2416 rbd_obj_request_put(obj_request);
2421 static int rbd_img_obj_request_submit(struct rbd_obj_request *obj_request)
2423 struct rbd_img_request *img_request;
2424 struct rbd_device *rbd_dev;
2427 rbd_assert(obj_request_img_data_test(obj_request));
2429 img_request = obj_request->img_request;
2430 rbd_assert(img_request);
2431 rbd_dev = img_request->rbd_dev;
2434 * Only writes to layered images need special handling.
2435 * Reads and non-layered writes are simple object requests.
2436 * Layered writes that start beyond the end of the overlap
2437 * with the parent have no parent data, so they too are
2438 * simple object requests. Finally, if the target object is
2439 * known to already exist, its parent data has already been
2440 * copied, so a write to the object can also be handled as a
2441 * simple object request.
2443 if (!img_request_write_test(img_request) ||
2444 !img_request_layered_test(img_request) ||
2445 rbd_dev->parent_overlap <= obj_request->img_offset ||
2446 ((known = obj_request_known_test(obj_request)) &&
2447 obj_request_exists_test(obj_request))) {
2449 struct rbd_device *rbd_dev;
2450 struct ceph_osd_client *osdc;
2452 rbd_dev = obj_request->img_request->rbd_dev;
2453 osdc = &rbd_dev->rbd_client->client->osdc;
2455 return rbd_obj_request_submit(osdc, obj_request);
2459 * It's a layered write. The target object might exist but
2460 * we may not know that yet. If we know it doesn't exist,
2461 * start by reading the data for the full target object from
2462 * the parent so we can use it for a copyup to the target.
2465 return rbd_img_obj_parent_read_full(obj_request);
2467 /* We don't know whether the target exists. Go find out. */
2469 return rbd_img_obj_exists_submit(obj_request);
2472 static int rbd_img_request_submit(struct rbd_img_request *img_request)
2474 struct rbd_obj_request *obj_request;
2475 struct rbd_obj_request *next_obj_request;
2477 dout("%s: img %p\n", __func__, img_request);
2478 for_each_obj_request_safe(img_request, obj_request, next_obj_request) {
2481 ret = rbd_img_obj_request_submit(obj_request);
2489 static void rbd_img_parent_read_callback(struct rbd_img_request *img_request)
2491 struct rbd_obj_request *obj_request;
2492 struct rbd_device *rbd_dev;
2495 rbd_assert(img_request_child_test(img_request));
2497 obj_request = img_request->obj_request;
2498 rbd_assert(obj_request);
2499 rbd_assert(obj_request->img_request);
2501 obj_request->result = img_request->result;
2502 if (obj_request->result)
2506 * We need to zero anything beyond the parent overlap
2507 * boundary. Since rbd_img_obj_request_read_callback()
2508 * will zero anything beyond the end of a short read, an
2509 * easy way to do this is to pretend the data from the
2510 * parent came up short--ending at the overlap boundary.
2512 rbd_assert(obj_request->img_offset < U64_MAX - obj_request->length);
2513 obj_end = obj_request->img_offset + obj_request->length;
2514 rbd_dev = obj_request->img_request->rbd_dev;
2515 if (obj_end > rbd_dev->parent_overlap) {
2518 if (obj_request->img_offset < rbd_dev->parent_overlap)
2519 xferred = rbd_dev->parent_overlap -
2520 obj_request->img_offset;
2522 obj_request->xferred = min(img_request->xferred, xferred);
2524 obj_request->xferred = img_request->xferred;
2527 rbd_img_request_put(img_request);
2528 rbd_img_obj_request_read_callback(obj_request);
2529 rbd_obj_request_complete(obj_request);
2532 static void rbd_img_parent_read(struct rbd_obj_request *obj_request)
2534 struct rbd_device *rbd_dev;
2535 struct rbd_img_request *img_request;
2538 rbd_assert(obj_request_img_data_test(obj_request));
2539 rbd_assert(obj_request->img_request != NULL);
2540 rbd_assert(obj_request->result == (s32) -ENOENT);
2541 rbd_assert(obj_request->type == OBJ_REQUEST_BIO);
2543 rbd_dev = obj_request->img_request->rbd_dev;
2544 rbd_assert(rbd_dev->parent != NULL);
2545 /* rbd_read_finish(obj_request, obj_request->length); */
2546 img_request = rbd_img_request_create(rbd_dev->parent,
2547 obj_request->img_offset,
2548 obj_request->length,
2554 rbd_obj_request_get(obj_request);
2555 img_request->obj_request = obj_request;
2557 result = rbd_img_request_fill(img_request, OBJ_REQUEST_BIO,
2558 obj_request->bio_list);
2562 img_request->callback = rbd_img_parent_read_callback;
2563 result = rbd_img_request_submit(img_request);
2570 rbd_img_request_put(img_request);
2571 obj_request->result = result;
2572 obj_request->xferred = 0;
2573 obj_request_done_set(obj_request);
2576 static int rbd_obj_notify_ack(struct rbd_device *rbd_dev, u64 notify_id)
2578 struct rbd_obj_request *obj_request;
2579 struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
2582 obj_request = rbd_obj_request_create(rbd_dev->header_name, 0, 0,
2583 OBJ_REQUEST_NODATA);
2588 obj_request->osd_req = rbd_osd_req_create(rbd_dev, false, obj_request);
2589 if (!obj_request->osd_req)
2591 obj_request->callback = rbd_obj_request_put;
2593 osd_req_op_watch_init(obj_request->osd_req, 0, CEPH_OSD_OP_NOTIFY_ACK,
2595 rbd_osd_req_format_read(obj_request);
2597 ret = rbd_obj_request_submit(osdc, obj_request);
2600 rbd_obj_request_put(obj_request);
2605 static void rbd_watch_cb(u64 ver, u64 notify_id, u8 opcode, void *data)
2607 struct rbd_device *rbd_dev = (struct rbd_device *)data;
2613 dout("%s: \"%s\" notify_id %llu opcode %u\n", __func__,
2614 rbd_dev->header_name, (unsigned long long)notify_id,
2615 (unsigned int)opcode);
2616 ret = rbd_dev_refresh(rbd_dev);
2618 rbd_warn(rbd_dev, ": header refresh error (%d)\n", ret);
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 if (offset + length > rbd_dev->mapping.size) {
2856 rbd_warn(rbd_dev, "beyond EOD (%llu~%llu > %llu)\n",
2857 offset, length, rbd_dev->mapping.size);
2862 img_request = rbd_img_request_create(rbd_dev, offset, length,
2863 write_request, false);
2867 img_request->rq = rq;
2869 result = rbd_img_request_fill(img_request, OBJ_REQUEST_BIO,
2872 result = rbd_img_request_submit(img_request);
2874 rbd_img_request_put(img_request);
2876 spin_lock_irq(q->queue_lock);
2878 rbd_warn(rbd_dev, "%s %llx at %llx result %d\n",
2879 write_request ? "write" : "read",
2880 length, offset, result);
2882 __blk_end_request_all(rq, result);
2888 * a queue callback. Makes sure that we don't create a bio that spans across
2889 * multiple osd objects. One exception would be with a single page bios,
2890 * which we handle later at bio_chain_clone_range()
2892 static int rbd_merge_bvec(struct request_queue *q, struct bvec_merge_data *bmd,
2893 struct bio_vec *bvec)
2895 struct rbd_device *rbd_dev = q->queuedata;
2896 sector_t sector_offset;
2897 sector_t sectors_per_obj;
2898 sector_t obj_sector_offset;
2902 * Find how far into its rbd object the partition-relative
2903 * bio start sector is to offset relative to the enclosing
2906 sector_offset = get_start_sect(bmd->bi_bdev) + bmd->bi_sector;
2907 sectors_per_obj = 1 << (rbd_dev->header.obj_order - SECTOR_SHIFT);
2908 obj_sector_offset = sector_offset & (sectors_per_obj - 1);
2911 * Compute the number of bytes from that offset to the end
2912 * of the object. Account for what's already used by the bio.
2914 ret = (int) (sectors_per_obj - obj_sector_offset) << SECTOR_SHIFT;
2915 if (ret > bmd->bi_size)
2916 ret -= bmd->bi_size;
2921 * Don't send back more than was asked for. And if the bio
2922 * was empty, let the whole thing through because: "Note
2923 * that a block device *must* allow a single page to be
2924 * added to an empty bio."
2926 rbd_assert(bvec->bv_len <= PAGE_SIZE);
2927 if (ret > (int) bvec->bv_len || !bmd->bi_size)
2928 ret = (int) bvec->bv_len;
2933 static void rbd_free_disk(struct rbd_device *rbd_dev)
2935 struct gendisk *disk = rbd_dev->disk;
2940 rbd_dev->disk = NULL;
2941 if (disk->flags & GENHD_FL_UP) {
2944 blk_cleanup_queue(disk->queue);
2949 static int rbd_obj_read_sync(struct rbd_device *rbd_dev,
2950 const char *object_name,
2951 u64 offset, u64 length, void *buf)
2954 struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
2955 struct rbd_obj_request *obj_request;
2956 struct page **pages = NULL;
2961 page_count = (u32) calc_pages_for(offset, length);
2962 pages = ceph_alloc_page_vector(page_count, GFP_KERNEL);
2964 ret = PTR_ERR(pages);
2967 obj_request = rbd_obj_request_create(object_name, offset, length,
2972 obj_request->pages = pages;
2973 obj_request->page_count = page_count;
2975 obj_request->osd_req = rbd_osd_req_create(rbd_dev, false, obj_request);
2976 if (!obj_request->osd_req)
2979 osd_req_op_extent_init(obj_request->osd_req, 0, CEPH_OSD_OP_READ,
2980 offset, length, 0, 0);
2981 osd_req_op_extent_osd_data_pages(obj_request->osd_req, 0,
2983 obj_request->length,
2984 obj_request->offset & ~PAGE_MASK,
2986 rbd_osd_req_format_read(obj_request);
2988 ret = rbd_obj_request_submit(osdc, obj_request);
2991 ret = rbd_obj_request_wait(obj_request);
2995 ret = obj_request->result;
2999 rbd_assert(obj_request->xferred <= (u64) SIZE_MAX);
3000 size = (size_t) obj_request->xferred;
3001 ceph_copy_from_page_vector(pages, buf, 0, size);
3002 rbd_assert(size <= (size_t)INT_MAX);
3006 rbd_obj_request_put(obj_request);
3008 ceph_release_page_vector(pages, page_count);
3014 * Read the complete header for the given rbd device.
3016 * Returns a pointer to a dynamically-allocated buffer containing
3017 * the complete and validated header. Caller can pass the address
3018 * of a variable that will be filled in with the version of the
3019 * header object at the time it was read.
3021 * Returns a pointer-coded errno if a failure occurs.
3023 static struct rbd_image_header_ondisk *
3024 rbd_dev_v1_header_read(struct rbd_device *rbd_dev)
3026 struct rbd_image_header_ondisk *ondisk = NULL;
3033 * The complete header will include an array of its 64-bit
3034 * snapshot ids, followed by the names of those snapshots as
3035 * a contiguous block of NUL-terminated strings. Note that
3036 * the number of snapshots could change by the time we read
3037 * it in, in which case we re-read it.
3044 size = sizeof (*ondisk);
3045 size += snap_count * sizeof (struct rbd_image_snap_ondisk);
3047 ondisk = kmalloc(size, GFP_KERNEL);
3049 return ERR_PTR(-ENOMEM);
3051 ret = rbd_obj_read_sync(rbd_dev, rbd_dev->header_name,
3055 if ((size_t)ret < size) {
3057 rbd_warn(rbd_dev, "short header read (want %zd got %d)",
3061 if (!rbd_dev_ondisk_valid(ondisk)) {
3063 rbd_warn(rbd_dev, "invalid header");
3067 names_size = le64_to_cpu(ondisk->snap_names_len);
3068 want_count = snap_count;
3069 snap_count = le32_to_cpu(ondisk->snap_count);
3070 } while (snap_count != want_count);
3077 return ERR_PTR(ret);
3081 * reload the ondisk the header
3083 static int rbd_read_header(struct rbd_device *rbd_dev,
3084 struct rbd_image_header *header)
3086 struct rbd_image_header_ondisk *ondisk;
3089 ondisk = rbd_dev_v1_header_read(rbd_dev);
3091 return PTR_ERR(ondisk);
3092 ret = rbd_header_from_disk(header, ondisk);
3099 * only read the first part of the ondisk header, without the snaps info
3101 static int rbd_dev_v1_refresh(struct rbd_device *rbd_dev)
3104 struct rbd_image_header h;
3106 ret = rbd_read_header(rbd_dev, &h);
3110 down_write(&rbd_dev->header_rwsem);
3112 /* Update image size, and check for resize of mapped image */
3113 rbd_dev->header.image_size = h.image_size;
3114 if (rbd_dev->spec->snap_id == CEPH_NOSNAP)
3115 if (rbd_dev->mapping.size != rbd_dev->header.image_size)
3116 rbd_dev->mapping.size = rbd_dev->header.image_size;
3118 /* rbd_dev->header.object_prefix shouldn't change */
3119 kfree(rbd_dev->header.snap_sizes);
3120 kfree(rbd_dev->header.snap_names);
3121 /* osd requests may still refer to snapc */
3122 ceph_put_snap_context(rbd_dev->header.snapc);
3124 rbd_dev->header.image_size = h.image_size;
3125 rbd_dev->header.snapc = h.snapc;
3126 rbd_dev->header.snap_names = h.snap_names;
3127 rbd_dev->header.snap_sizes = h.snap_sizes;
3128 /* Free the extra copy of the object prefix */
3129 if (strcmp(rbd_dev->header.object_prefix, h.object_prefix))
3130 rbd_warn(rbd_dev, "object prefix changed (ignoring)");
3131 kfree(h.object_prefix);
3133 up_write(&rbd_dev->header_rwsem);
3139 * Clear the rbd device's EXISTS flag if the snapshot it's mapped to
3140 * has disappeared from the (just updated) snapshot context.
3142 static void rbd_exists_validate(struct rbd_device *rbd_dev)
3146 if (!test_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags))
3149 snap_id = rbd_dev->spec->snap_id;
3150 if (snap_id == CEPH_NOSNAP)
3153 if (rbd_dev_snap_index(rbd_dev, snap_id) == BAD_SNAP_INDEX)
3154 clear_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags);
3157 static int rbd_dev_refresh(struct rbd_device *rbd_dev)
3162 rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
3163 mapping_size = rbd_dev->mapping.size;
3164 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
3165 if (rbd_dev->image_format == 1)
3166 ret = rbd_dev_v1_refresh(rbd_dev);
3168 ret = rbd_dev_v2_refresh(rbd_dev);
3170 /* If it's a mapped snapshot, validate its EXISTS flag */
3172 rbd_exists_validate(rbd_dev);
3173 mutex_unlock(&ctl_mutex);
3174 if (mapping_size != rbd_dev->mapping.size) {
3177 size = (sector_t)rbd_dev->mapping.size / SECTOR_SIZE;
3178 dout("setting size to %llu sectors", (unsigned long long)size);
3179 set_capacity(rbd_dev->disk, size);
3180 revalidate_disk(rbd_dev->disk);
3186 static int rbd_init_disk(struct rbd_device *rbd_dev)
3188 struct gendisk *disk;
3189 struct request_queue *q;
3192 /* create gendisk info */
3193 disk = alloc_disk(RBD_MINORS_PER_MAJOR);
3197 snprintf(disk->disk_name, sizeof(disk->disk_name), RBD_DRV_NAME "%d",
3199 disk->major = rbd_dev->major;
3200 disk->first_minor = 0;
3201 disk->fops = &rbd_bd_ops;
3202 disk->private_data = rbd_dev;
3204 q = blk_init_queue(rbd_request_fn, &rbd_dev->lock);
3208 /* We use the default size, but let's be explicit about it. */
3209 blk_queue_physical_block_size(q, SECTOR_SIZE);
3211 /* set io sizes to object size */
3212 segment_size = rbd_obj_bytes(&rbd_dev->header);
3213 blk_queue_max_hw_sectors(q, segment_size / SECTOR_SIZE);
3214 blk_queue_max_segment_size(q, segment_size);
3215 blk_queue_io_min(q, segment_size);
3216 blk_queue_io_opt(q, segment_size);
3218 blk_queue_merge_bvec(q, rbd_merge_bvec);
3221 q->queuedata = rbd_dev;
3223 rbd_dev->disk = disk;
3236 static struct rbd_device *dev_to_rbd_dev(struct device *dev)
3238 return container_of(dev, struct rbd_device, dev);
3241 static ssize_t rbd_size_show(struct device *dev,
3242 struct device_attribute *attr, char *buf)
3244 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3246 return sprintf(buf, "%llu\n",
3247 (unsigned long long)rbd_dev->mapping.size);
3251 * Note this shows the features for whatever's mapped, which is not
3252 * necessarily the base image.
3254 static ssize_t rbd_features_show(struct device *dev,
3255 struct device_attribute *attr, char *buf)
3257 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3259 return sprintf(buf, "0x%016llx\n",
3260 (unsigned long long)rbd_dev->mapping.features);
3263 static ssize_t rbd_major_show(struct device *dev,
3264 struct device_attribute *attr, char *buf)
3266 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3269 return sprintf(buf, "%d\n", rbd_dev->major);
3271 return sprintf(buf, "(none)\n");
3275 static ssize_t rbd_client_id_show(struct device *dev,
3276 struct device_attribute *attr, char *buf)
3278 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3280 return sprintf(buf, "client%lld\n",
3281 ceph_client_id(rbd_dev->rbd_client->client));
3284 static ssize_t rbd_pool_show(struct device *dev,
3285 struct device_attribute *attr, char *buf)
3287 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3289 return sprintf(buf, "%s\n", rbd_dev->spec->pool_name);
3292 static ssize_t rbd_pool_id_show(struct device *dev,
3293 struct device_attribute *attr, char *buf)
3295 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3297 return sprintf(buf, "%llu\n",
3298 (unsigned long long) rbd_dev->spec->pool_id);
3301 static ssize_t rbd_name_show(struct device *dev,
3302 struct device_attribute *attr, char *buf)
3304 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3306 if (rbd_dev->spec->image_name)
3307 return sprintf(buf, "%s\n", rbd_dev->spec->image_name);
3309 return sprintf(buf, "(unknown)\n");
3312 static ssize_t rbd_image_id_show(struct device *dev,
3313 struct device_attribute *attr, char *buf)
3315 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3317 return sprintf(buf, "%s\n", rbd_dev->spec->image_id);
3321 * Shows the name of the currently-mapped snapshot (or
3322 * RBD_SNAP_HEAD_NAME for the base image).
3324 static ssize_t rbd_snap_show(struct device *dev,
3325 struct device_attribute *attr,
3328 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3330 return sprintf(buf, "%s\n", rbd_dev->spec->snap_name);
3334 * For an rbd v2 image, shows the pool id, image id, and snapshot id
3335 * for the parent image. If there is no parent, simply shows
3336 * "(no parent image)".
3338 static ssize_t rbd_parent_show(struct device *dev,
3339 struct device_attribute *attr,
3342 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3343 struct rbd_spec *spec = rbd_dev->parent_spec;
3348 return sprintf(buf, "(no parent image)\n");
3350 count = sprintf(bufp, "pool_id %llu\npool_name %s\n",
3351 (unsigned long long) spec->pool_id, spec->pool_name);
3356 count = sprintf(bufp, "image_id %s\nimage_name %s\n", spec->image_id,
3357 spec->image_name ? spec->image_name : "(unknown)");
3362 count = sprintf(bufp, "snap_id %llu\nsnap_name %s\n",
3363 (unsigned long long) spec->snap_id, spec->snap_name);
3368 count = sprintf(bufp, "overlap %llu\n", rbd_dev->parent_overlap);
3373 return (ssize_t) (bufp - buf);
3376 static ssize_t rbd_image_refresh(struct device *dev,
3377 struct device_attribute *attr,
3381 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3384 ret = rbd_dev_refresh(rbd_dev);
3386 rbd_warn(rbd_dev, ": manual header refresh error (%d)\n", ret);
3388 return ret < 0 ? ret : size;
3391 static DEVICE_ATTR(size, S_IRUGO, rbd_size_show, NULL);
3392 static DEVICE_ATTR(features, S_IRUGO, rbd_features_show, NULL);
3393 static DEVICE_ATTR(major, S_IRUGO, rbd_major_show, NULL);
3394 static DEVICE_ATTR(client_id, S_IRUGO, rbd_client_id_show, NULL);
3395 static DEVICE_ATTR(pool, S_IRUGO, rbd_pool_show, NULL);
3396 static DEVICE_ATTR(pool_id, S_IRUGO, rbd_pool_id_show, NULL);
3397 static DEVICE_ATTR(name, S_IRUGO, rbd_name_show, NULL);
3398 static DEVICE_ATTR(image_id, S_IRUGO, rbd_image_id_show, NULL);
3399 static DEVICE_ATTR(refresh, S_IWUSR, NULL, rbd_image_refresh);
3400 static DEVICE_ATTR(current_snap, S_IRUGO, rbd_snap_show, NULL);
3401 static DEVICE_ATTR(parent, S_IRUGO, rbd_parent_show, NULL);
3403 static struct attribute *rbd_attrs[] = {
3404 &dev_attr_size.attr,
3405 &dev_attr_features.attr,
3406 &dev_attr_major.attr,
3407 &dev_attr_client_id.attr,
3408 &dev_attr_pool.attr,
3409 &dev_attr_pool_id.attr,
3410 &dev_attr_name.attr,
3411 &dev_attr_image_id.attr,
3412 &dev_attr_current_snap.attr,
3413 &dev_attr_parent.attr,
3414 &dev_attr_refresh.attr,
3418 static struct attribute_group rbd_attr_group = {
3422 static const struct attribute_group *rbd_attr_groups[] = {
3427 static void rbd_sysfs_dev_release(struct device *dev)
3431 static struct device_type rbd_device_type = {
3433 .groups = rbd_attr_groups,
3434 .release = rbd_sysfs_dev_release,
3437 static struct rbd_spec *rbd_spec_get(struct rbd_spec *spec)
3439 kref_get(&spec->kref);
3444 static void rbd_spec_free(struct kref *kref);
3445 static void rbd_spec_put(struct rbd_spec *spec)
3448 kref_put(&spec->kref, rbd_spec_free);
3451 static struct rbd_spec *rbd_spec_alloc(void)
3453 struct rbd_spec *spec;
3455 spec = kzalloc(sizeof (*spec), GFP_KERNEL);
3458 kref_init(&spec->kref);
3463 static void rbd_spec_free(struct kref *kref)
3465 struct rbd_spec *spec = container_of(kref, struct rbd_spec, kref);
3467 kfree(spec->pool_name);
3468 kfree(spec->image_id);
3469 kfree(spec->image_name);
3470 kfree(spec->snap_name);
3474 static struct rbd_device *rbd_dev_create(struct rbd_client *rbdc,
3475 struct rbd_spec *spec)
3477 struct rbd_device *rbd_dev;
3479 rbd_dev = kzalloc(sizeof (*rbd_dev), GFP_KERNEL);
3483 spin_lock_init(&rbd_dev->lock);
3485 INIT_LIST_HEAD(&rbd_dev->node);
3486 init_rwsem(&rbd_dev->header_rwsem);
3488 rbd_dev->spec = spec;
3489 rbd_dev->rbd_client = rbdc;
3491 /* Initialize the layout used for all rbd requests */
3493 rbd_dev->layout.fl_stripe_unit = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER);
3494 rbd_dev->layout.fl_stripe_count = cpu_to_le32(1);
3495 rbd_dev->layout.fl_object_size = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER);
3496 rbd_dev->layout.fl_pg_pool = cpu_to_le32((u32) spec->pool_id);
3501 static void rbd_dev_destroy(struct rbd_device *rbd_dev)
3503 rbd_put_client(rbd_dev->rbd_client);
3504 rbd_spec_put(rbd_dev->spec);
3509 * Get the size and object order for an image snapshot, or if
3510 * snap_id is CEPH_NOSNAP, gets this information for the base
3513 static int _rbd_dev_v2_snap_size(struct rbd_device *rbd_dev, u64 snap_id,
3514 u8 *order, u64 *snap_size)
3516 __le64 snapid = cpu_to_le64(snap_id);
3521 } __attribute__ ((packed)) size_buf = { 0 };
3523 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
3525 &snapid, sizeof (snapid),
3526 &size_buf, sizeof (size_buf));
3527 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
3530 if (ret < sizeof (size_buf))
3534 *order = size_buf.order;
3535 *snap_size = le64_to_cpu(size_buf.size);
3537 dout(" snap_id 0x%016llx order = %u, snap_size = %llu\n",
3538 (unsigned long long)snap_id, (unsigned int)*order,
3539 (unsigned long long)*snap_size);
3544 static int rbd_dev_v2_image_size(struct rbd_device *rbd_dev)
3546 return _rbd_dev_v2_snap_size(rbd_dev, CEPH_NOSNAP,
3547 &rbd_dev->header.obj_order,
3548 &rbd_dev->header.image_size);
3551 static int rbd_dev_v2_object_prefix(struct rbd_device *rbd_dev)
3557 reply_buf = kzalloc(RBD_OBJ_PREFIX_LEN_MAX, GFP_KERNEL);
3561 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
3562 "rbd", "get_object_prefix", NULL, 0,
3563 reply_buf, RBD_OBJ_PREFIX_LEN_MAX);
3564 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
3569 rbd_dev->header.object_prefix = ceph_extract_encoded_string(&p,
3570 p + ret, NULL, GFP_NOIO);
3573 if (IS_ERR(rbd_dev->header.object_prefix)) {
3574 ret = PTR_ERR(rbd_dev->header.object_prefix);
3575 rbd_dev->header.object_prefix = NULL;
3577 dout(" object_prefix = %s\n", rbd_dev->header.object_prefix);
3585 static int _rbd_dev_v2_snap_features(struct rbd_device *rbd_dev, u64 snap_id,
3588 __le64 snapid = cpu_to_le64(snap_id);
3592 } __attribute__ ((packed)) features_buf = { 0 };
3596 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
3597 "rbd", "get_features",
3598 &snapid, sizeof (snapid),
3599 &features_buf, sizeof (features_buf));
3600 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
3603 if (ret < sizeof (features_buf))
3606 incompat = le64_to_cpu(features_buf.incompat);
3607 if (incompat & ~RBD_FEATURES_SUPPORTED)
3610 *snap_features = le64_to_cpu(features_buf.features);
3612 dout(" snap_id 0x%016llx features = 0x%016llx incompat = 0x%016llx\n",
3613 (unsigned long long)snap_id,
3614 (unsigned long long)*snap_features,
3615 (unsigned long long)le64_to_cpu(features_buf.incompat));
3620 static int rbd_dev_v2_features(struct rbd_device *rbd_dev)
3622 return _rbd_dev_v2_snap_features(rbd_dev, CEPH_NOSNAP,
3623 &rbd_dev->header.features);
3626 static int rbd_dev_v2_parent_info(struct rbd_device *rbd_dev)
3628 struct rbd_spec *parent_spec;
3630 void *reply_buf = NULL;
3638 parent_spec = rbd_spec_alloc();
3642 size = sizeof (__le64) + /* pool_id */
3643 sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX + /* image_id */
3644 sizeof (__le64) + /* snap_id */
3645 sizeof (__le64); /* overlap */
3646 reply_buf = kmalloc(size, GFP_KERNEL);
3652 snapid = cpu_to_le64(CEPH_NOSNAP);
3653 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
3654 "rbd", "get_parent",
3655 &snapid, sizeof (snapid),
3657 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
3662 end = reply_buf + ret;
3664 ceph_decode_64_safe(&p, end, parent_spec->pool_id, out_err);
3665 if (parent_spec->pool_id == CEPH_NOPOOL)
3666 goto out; /* No parent? No problem. */
3668 /* The ceph file layout needs to fit pool id in 32 bits */
3671 if (parent_spec->pool_id > (u64)U32_MAX) {
3672 rbd_warn(NULL, "parent pool id too large (%llu > %u)\n",
3673 (unsigned long long)parent_spec->pool_id, U32_MAX);
3677 image_id = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL);
3678 if (IS_ERR(image_id)) {
3679 ret = PTR_ERR(image_id);
3682 parent_spec->image_id = image_id;
3683 ceph_decode_64_safe(&p, end, parent_spec->snap_id, out_err);
3684 ceph_decode_64_safe(&p, end, overlap, out_err);
3686 rbd_dev->parent_overlap = overlap;
3687 rbd_dev->parent_spec = parent_spec;
3688 parent_spec = NULL; /* rbd_dev now owns this */
3693 rbd_spec_put(parent_spec);
3698 static int rbd_dev_v2_striping_info(struct rbd_device *rbd_dev)
3702 __le64 stripe_count;
3703 } __attribute__ ((packed)) striping_info_buf = { 0 };
3704 size_t size = sizeof (striping_info_buf);
3711 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
3712 "rbd", "get_stripe_unit_count", NULL, 0,
3713 (char *)&striping_info_buf, size);
3714 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
3721 * We don't actually support the "fancy striping" feature
3722 * (STRIPINGV2) yet, but if the striping sizes are the
3723 * defaults the behavior is the same as before. So find
3724 * out, and only fail if the image has non-default values.
3727 obj_size = (u64)1 << rbd_dev->header.obj_order;
3728 p = &striping_info_buf;
3729 stripe_unit = ceph_decode_64(&p);
3730 if (stripe_unit != obj_size) {
3731 rbd_warn(rbd_dev, "unsupported stripe unit "
3732 "(got %llu want %llu)",
3733 stripe_unit, obj_size);
3736 stripe_count = ceph_decode_64(&p);
3737 if (stripe_count != 1) {
3738 rbd_warn(rbd_dev, "unsupported stripe count "
3739 "(got %llu want 1)", stripe_count);
3742 rbd_dev->header.stripe_unit = stripe_unit;
3743 rbd_dev->header.stripe_count = stripe_count;
3748 static char *rbd_dev_image_name(struct rbd_device *rbd_dev)
3750 size_t image_id_size;
3755 void *reply_buf = NULL;
3757 char *image_name = NULL;
3760 rbd_assert(!rbd_dev->spec->image_name);
3762 len = strlen(rbd_dev->spec->image_id);
3763 image_id_size = sizeof (__le32) + len;
3764 image_id = kmalloc(image_id_size, GFP_KERNEL);
3769 end = image_id + image_id_size;
3770 ceph_encode_string(&p, end, rbd_dev->spec->image_id, (u32)len);
3772 size = sizeof (__le32) + RBD_IMAGE_NAME_LEN_MAX;
3773 reply_buf = kmalloc(size, GFP_KERNEL);
3777 ret = rbd_obj_method_sync(rbd_dev, RBD_DIRECTORY,
3778 "rbd", "dir_get_name",
3779 image_id, image_id_size,
3784 end = reply_buf + ret;
3786 image_name = ceph_extract_encoded_string(&p, end, &len, GFP_KERNEL);
3787 if (IS_ERR(image_name))
3790 dout("%s: name is %s len is %zd\n", __func__, image_name, len);
3798 static u64 rbd_v1_snap_id_by_name(struct rbd_device *rbd_dev, const char *name)
3800 struct ceph_snap_context *snapc = rbd_dev->header.snapc;
3801 const char *snap_name;
3804 /* Skip over names until we find the one we are looking for */
3806 snap_name = rbd_dev->header.snap_names;
3807 while (which < snapc->num_snaps) {
3808 if (!strcmp(name, snap_name))
3809 return snapc->snaps[which];
3810 snap_name += strlen(snap_name) + 1;
3816 static u64 rbd_v2_snap_id_by_name(struct rbd_device *rbd_dev, const char *name)
3818 struct ceph_snap_context *snapc = rbd_dev->header.snapc;
3823 for (which = 0; !found && which < snapc->num_snaps; which++) {
3824 const char *snap_name;
3826 snap_id = snapc->snaps[which];
3827 snap_name = rbd_dev_v2_snap_name(rbd_dev, snap_id);
3828 if (IS_ERR(snap_name))
3830 found = !strcmp(name, snap_name);
3833 return found ? snap_id : CEPH_NOSNAP;
3837 * Assumes name is never RBD_SNAP_HEAD_NAME; returns CEPH_NOSNAP if
3838 * no snapshot by that name is found, or if an error occurs.
3840 static u64 rbd_snap_id_by_name(struct rbd_device *rbd_dev, const char *name)
3842 if (rbd_dev->image_format == 1)
3843 return rbd_v1_snap_id_by_name(rbd_dev, name);
3845 return rbd_v2_snap_id_by_name(rbd_dev, name);
3849 * When an rbd image has a parent image, it is identified by the
3850 * pool, image, and snapshot ids (not names). This function fills
3851 * in the names for those ids. (It's OK if we can't figure out the
3852 * name for an image id, but the pool and snapshot ids should always
3853 * exist and have names.) All names in an rbd spec are dynamically
3856 * When an image being mapped (not a parent) is probed, we have the
3857 * pool name and pool id, image name and image id, and the snapshot
3858 * name. The only thing we're missing is the snapshot id.
3860 static int rbd_dev_spec_update(struct rbd_device *rbd_dev)
3862 struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
3863 struct rbd_spec *spec = rbd_dev->spec;
3864 const char *pool_name;
3865 const char *image_name;
3866 const char *snap_name;
3870 * An image being mapped will have the pool name (etc.), but
3871 * we need to look up the snapshot id.
3873 if (spec->pool_name) {
3874 if (strcmp(spec->snap_name, RBD_SNAP_HEAD_NAME)) {
3877 snap_id = rbd_snap_id_by_name(rbd_dev, spec->snap_name);
3878 if (snap_id == CEPH_NOSNAP)
3880 spec->snap_id = snap_id;
3882 spec->snap_id = CEPH_NOSNAP;
3888 /* Get the pool name; we have to make our own copy of this */
3890 pool_name = ceph_pg_pool_name_by_id(osdc->osdmap, spec->pool_id);
3892 rbd_warn(rbd_dev, "no pool with id %llu", spec->pool_id);
3895 pool_name = kstrdup(pool_name, GFP_KERNEL);
3899 /* Fetch the image name; tolerate failure here */
3901 image_name = rbd_dev_image_name(rbd_dev);
3903 rbd_warn(rbd_dev, "unable to get image name");
3905 /* Look up the snapshot name, and make a copy */
3907 snap_name = rbd_snap_name(rbd_dev, spec->snap_id);
3913 spec->pool_name = pool_name;
3914 spec->image_name = image_name;
3915 spec->snap_name = snap_name;
3925 static int rbd_dev_v2_snap_context(struct rbd_device *rbd_dev)
3934 struct ceph_snap_context *snapc;
3938 * We'll need room for the seq value (maximum snapshot id),
3939 * snapshot count, and array of that many snapshot ids.
3940 * For now we have a fixed upper limit on the number we're
3941 * prepared to receive.
3943 size = sizeof (__le64) + sizeof (__le32) +
3944 RBD_MAX_SNAP_COUNT * sizeof (__le64);
3945 reply_buf = kzalloc(size, GFP_KERNEL);
3949 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
3950 "rbd", "get_snapcontext", NULL, 0,
3952 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
3957 end = reply_buf + ret;
3959 ceph_decode_64_safe(&p, end, seq, out);
3960 ceph_decode_32_safe(&p, end, snap_count, out);
3963 * Make sure the reported number of snapshot ids wouldn't go
3964 * beyond the end of our buffer. But before checking that,
3965 * make sure the computed size of the snapshot context we
3966 * allocate is representable in a size_t.
3968 if (snap_count > (SIZE_MAX - sizeof (struct ceph_snap_context))
3973 if (!ceph_has_room(&p, end, snap_count * sizeof (__le64)))
3977 snapc = ceph_create_snap_context(snap_count, GFP_KERNEL);
3983 for (i = 0; i < snap_count; i++)
3984 snapc->snaps[i] = ceph_decode_64(&p);
3986 ceph_put_snap_context(rbd_dev->header.snapc);
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 if (rbd_dev->spec->snap_id == CEPH_NOSNAP)
4048 if (rbd_dev->mapping.size != rbd_dev->header.image_size)
4049 rbd_dev->mapping.size = rbd_dev->header.image_size;
4051 ret = rbd_dev_v2_snap_context(rbd_dev);
4052 dout("rbd_dev_v2_snap_context returned %d\n", ret);
4056 up_write(&rbd_dev->header_rwsem);
4061 static int rbd_bus_add_dev(struct rbd_device *rbd_dev)
4066 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
4068 dev = &rbd_dev->dev;
4069 dev->bus = &rbd_bus_type;
4070 dev->type = &rbd_device_type;
4071 dev->parent = &rbd_root_dev;
4072 dev->release = rbd_dev_device_release;
4073 dev_set_name(dev, "%d", rbd_dev->dev_id);
4074 ret = device_register(dev);
4076 mutex_unlock(&ctl_mutex);
4081 static void rbd_bus_del_dev(struct rbd_device *rbd_dev)
4083 device_unregister(&rbd_dev->dev);
4086 static atomic64_t rbd_dev_id_max = ATOMIC64_INIT(0);
4089 * Get a unique rbd identifier for the given new rbd_dev, and add
4090 * the rbd_dev to the global list. The minimum rbd id is 1.
4092 static void rbd_dev_id_get(struct rbd_device *rbd_dev)
4094 rbd_dev->dev_id = atomic64_inc_return(&rbd_dev_id_max);
4096 spin_lock(&rbd_dev_list_lock);
4097 list_add_tail(&rbd_dev->node, &rbd_dev_list);
4098 spin_unlock(&rbd_dev_list_lock);
4099 dout("rbd_dev %p given dev id %llu\n", rbd_dev,
4100 (unsigned long long) rbd_dev->dev_id);
4104 * Remove an rbd_dev from the global list, and record that its
4105 * identifier is no longer in use.
4107 static void rbd_dev_id_put(struct rbd_device *rbd_dev)
4109 struct list_head *tmp;
4110 int rbd_id = rbd_dev->dev_id;
4113 rbd_assert(rbd_id > 0);
4115 dout("rbd_dev %p released dev id %llu\n", rbd_dev,
4116 (unsigned long long) rbd_dev->dev_id);
4117 spin_lock(&rbd_dev_list_lock);
4118 list_del_init(&rbd_dev->node);
4121 * If the id being "put" is not the current maximum, there
4122 * is nothing special we need to do.
4124 if (rbd_id != atomic64_read(&rbd_dev_id_max)) {
4125 spin_unlock(&rbd_dev_list_lock);
4130 * We need to update the current maximum id. Search the
4131 * list to find out what it is. We're more likely to find
4132 * the maximum at the end, so search the list backward.
4135 list_for_each_prev(tmp, &rbd_dev_list) {
4136 struct rbd_device *rbd_dev;
4138 rbd_dev = list_entry(tmp, struct rbd_device, node);
4139 if (rbd_dev->dev_id > max_id)
4140 max_id = rbd_dev->dev_id;
4142 spin_unlock(&rbd_dev_list_lock);
4145 * The max id could have been updated by rbd_dev_id_get(), in
4146 * which case it now accurately reflects the new maximum.
4147 * Be careful not to overwrite the maximum value in that
4150 atomic64_cmpxchg(&rbd_dev_id_max, rbd_id, max_id);
4151 dout(" max dev id has been reset\n");
4155 * Skips over white space at *buf, and updates *buf to point to the
4156 * first found non-space character (if any). Returns the length of
4157 * the token (string of non-white space characters) found. Note
4158 * that *buf must be terminated with '\0'.
4160 static inline size_t next_token(const char **buf)
4163 * These are the characters that produce nonzero for
4164 * isspace() in the "C" and "POSIX" locales.
4166 const char *spaces = " \f\n\r\t\v";
4168 *buf += strspn(*buf, spaces); /* Find start of token */
4170 return strcspn(*buf, spaces); /* Return token length */
4174 * Finds the next token in *buf, and if the provided token buffer is
4175 * big enough, copies the found token into it. The result, if
4176 * copied, is guaranteed to be terminated with '\0'. Note that *buf
4177 * must be terminated with '\0' on entry.
4179 * Returns the length of the token found (not including the '\0').
4180 * Return value will be 0 if no token is found, and it will be >=
4181 * token_size if the token would not fit.
4183 * The *buf pointer will be updated to point beyond the end of the
4184 * found token. Note that this occurs even if the token buffer is
4185 * too small to hold it.
4187 static inline size_t copy_token(const char **buf,
4193 len = next_token(buf);
4194 if (len < token_size) {
4195 memcpy(token, *buf, len);
4196 *(token + len) = '\0';
4204 * Finds the next token in *buf, dynamically allocates a buffer big
4205 * enough to hold a copy of it, and copies the token into the new
4206 * buffer. The copy is guaranteed to be terminated with '\0'. Note
4207 * that a duplicate buffer is created even for a zero-length token.
4209 * Returns a pointer to the newly-allocated duplicate, or a null
4210 * pointer if memory for the duplicate was not available. If
4211 * the lenp argument is a non-null pointer, the length of the token
4212 * (not including the '\0') is returned in *lenp.
4214 * If successful, the *buf pointer will be updated to point beyond
4215 * the end of the found token.
4217 * Note: uses GFP_KERNEL for allocation.
4219 static inline char *dup_token(const char **buf, size_t *lenp)
4224 len = next_token(buf);
4225 dup = kmemdup(*buf, len + 1, GFP_KERNEL);
4228 *(dup + len) = '\0';
4238 * Parse the options provided for an "rbd add" (i.e., rbd image
4239 * mapping) request. These arrive via a write to /sys/bus/rbd/add,
4240 * and the data written is passed here via a NUL-terminated buffer.
4241 * Returns 0 if successful or an error code otherwise.
4243 * The information extracted from these options is recorded in
4244 * the other parameters which return dynamically-allocated
4247 * The address of a pointer that will refer to a ceph options
4248 * structure. Caller must release the returned pointer using
4249 * ceph_destroy_options() when it is no longer needed.
4251 * Address of an rbd options pointer. Fully initialized by
4252 * this function; caller must release with kfree().
4254 * Address of an rbd image specification pointer. Fully
4255 * initialized by this function based on parsed options.
4256 * Caller must release with rbd_spec_put().
4258 * The options passed take this form:
4259 * <mon_addrs> <options> <pool_name> <image_name> [<snap_id>]
4262 * A comma-separated list of one or more monitor addresses.
4263 * A monitor address is an ip address, optionally followed
4264 * by a port number (separated by a colon).
4265 * I.e.: ip1[:port1][,ip2[:port2]...]
4267 * A comma-separated list of ceph and/or rbd options.
4269 * The name of the rados pool containing the rbd image.
4271 * The name of the image in that pool to map.
4273 * An optional snapshot id. If provided, the mapping will
4274 * present data from the image at the time that snapshot was
4275 * created. The image head is used if no snapshot id is
4276 * provided. Snapshot mappings are always read-only.
4278 static int rbd_add_parse_args(const char *buf,
4279 struct ceph_options **ceph_opts,
4280 struct rbd_options **opts,
4281 struct rbd_spec **rbd_spec)
4285 const char *mon_addrs;
4287 size_t mon_addrs_size;
4288 struct rbd_spec *spec = NULL;
4289 struct rbd_options *rbd_opts = NULL;
4290 struct ceph_options *copts;
4293 /* The first four tokens are required */
4295 len = next_token(&buf);
4297 rbd_warn(NULL, "no monitor address(es) provided");
4301 mon_addrs_size = len + 1;
4305 options = dup_token(&buf, NULL);
4309 rbd_warn(NULL, "no options provided");
4313 spec = rbd_spec_alloc();
4317 spec->pool_name = dup_token(&buf, NULL);
4318 if (!spec->pool_name)
4320 if (!*spec->pool_name) {
4321 rbd_warn(NULL, "no pool name provided");
4325 spec->image_name = dup_token(&buf, NULL);
4326 if (!spec->image_name)
4328 if (!*spec->image_name) {
4329 rbd_warn(NULL, "no image name provided");
4334 * Snapshot name is optional; default is to use "-"
4335 * (indicating the head/no snapshot).
4337 len = next_token(&buf);
4339 buf = RBD_SNAP_HEAD_NAME; /* No snapshot supplied */
4340 len = sizeof (RBD_SNAP_HEAD_NAME) - 1;
4341 } else if (len > RBD_MAX_SNAP_NAME_LEN) {
4342 ret = -ENAMETOOLONG;
4345 snap_name = kmemdup(buf, len + 1, GFP_KERNEL);
4348 *(snap_name + len) = '\0';
4349 spec->snap_name = snap_name;
4351 /* Initialize all rbd options to the defaults */
4353 rbd_opts = kzalloc(sizeof (*rbd_opts), GFP_KERNEL);
4357 rbd_opts->read_only = RBD_READ_ONLY_DEFAULT;
4359 copts = ceph_parse_options(options, mon_addrs,
4360 mon_addrs + mon_addrs_size - 1,
4361 parse_rbd_opts_token, rbd_opts);
4362 if (IS_ERR(copts)) {
4363 ret = PTR_ERR(copts);
4384 * An rbd format 2 image has a unique identifier, distinct from the
4385 * name given to it by the user. Internally, that identifier is
4386 * what's used to specify the names of objects related to the image.
4388 * A special "rbd id" object is used to map an rbd image name to its
4389 * id. If that object doesn't exist, then there is no v2 rbd image
4390 * with the supplied name.
4392 * This function will record the given rbd_dev's image_id field if
4393 * it can be determined, and in that case will return 0. If any
4394 * errors occur a negative errno will be returned and the rbd_dev's
4395 * image_id field will be unchanged (and should be NULL).
4397 static int rbd_dev_image_id(struct rbd_device *rbd_dev)
4406 * When probing a parent image, the image id is already
4407 * known (and the image name likely is not). There's no
4408 * need to fetch the image id again in this case. We
4409 * do still need to set the image format though.
4411 if (rbd_dev->spec->image_id) {
4412 rbd_dev->image_format = *rbd_dev->spec->image_id ? 2 : 1;
4418 * First, see if the format 2 image id file exists, and if
4419 * so, get the image's persistent id from it.
4421 size = sizeof (RBD_ID_PREFIX) + strlen(rbd_dev->spec->image_name);
4422 object_name = kmalloc(size, GFP_NOIO);
4425 sprintf(object_name, "%s%s", RBD_ID_PREFIX, rbd_dev->spec->image_name);
4426 dout("rbd id object name is %s\n", object_name);
4428 /* Response will be an encoded string, which includes a length */
4430 size = sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX;
4431 response = kzalloc(size, GFP_NOIO);
4437 /* If it doesn't exist we'll assume it's a format 1 image */
4439 ret = rbd_obj_method_sync(rbd_dev, object_name,
4440 "rbd", "get_id", NULL, 0,
4441 response, RBD_IMAGE_ID_LEN_MAX);
4442 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
4443 if (ret == -ENOENT) {
4444 image_id = kstrdup("", GFP_KERNEL);
4445 ret = image_id ? 0 : -ENOMEM;
4447 rbd_dev->image_format = 1;
4448 } else if (ret > sizeof (__le32)) {
4451 image_id = ceph_extract_encoded_string(&p, p + ret,
4453 ret = IS_ERR(image_id) ? PTR_ERR(image_id) : 0;
4455 rbd_dev->image_format = 2;
4461 rbd_dev->spec->image_id = image_id;
4462 dout("image_id is %s\n", image_id);
4471 /* Undo whatever state changes are made by v1 or v2 image probe */
4473 static void rbd_dev_unprobe(struct rbd_device *rbd_dev)
4475 struct rbd_image_header *header;
4477 rbd_dev_remove_parent(rbd_dev);
4478 rbd_spec_put(rbd_dev->parent_spec);
4479 rbd_dev->parent_spec = NULL;
4480 rbd_dev->parent_overlap = 0;
4482 /* Free dynamic fields from the header, then zero it out */
4484 header = &rbd_dev->header;
4485 ceph_put_snap_context(header->snapc);
4486 kfree(header->snap_sizes);
4487 kfree(header->snap_names);
4488 kfree(header->object_prefix);
4489 memset(header, 0, sizeof (*header));
4492 static int rbd_dev_v1_probe(struct rbd_device *rbd_dev)
4496 /* Populate rbd image metadata */
4498 ret = rbd_read_header(rbd_dev, &rbd_dev->header);
4502 /* Version 1 images have no parent (no layering) */
4504 rbd_dev->parent_spec = NULL;
4505 rbd_dev->parent_overlap = 0;
4507 dout("discovered version 1 image, header name is %s\n",
4508 rbd_dev->header_name);
4513 kfree(rbd_dev->header_name);
4514 rbd_dev->header_name = NULL;
4515 kfree(rbd_dev->spec->image_id);
4516 rbd_dev->spec->image_id = NULL;
4521 static int rbd_dev_v2_probe(struct rbd_device *rbd_dev)
4525 ret = rbd_dev_v2_image_size(rbd_dev);
4529 /* Get the object prefix (a.k.a. block_name) for the image */
4531 ret = rbd_dev_v2_object_prefix(rbd_dev);
4535 /* Get the and check features for the image */
4537 ret = rbd_dev_v2_features(rbd_dev);
4541 /* If the image supports layering, get the parent info */
4543 if (rbd_dev->header.features & RBD_FEATURE_LAYERING) {
4544 ret = rbd_dev_v2_parent_info(rbd_dev);
4548 * Print a warning if this image has a parent.
4549 * Don't print it if the image now being probed
4550 * is itself a parent. We can tell at this point
4551 * because we won't know its pool name yet (just its
4554 if (rbd_dev->parent_spec && rbd_dev->spec->pool_name)
4555 rbd_warn(rbd_dev, "WARNING: kernel layering "
4556 "is EXPERIMENTAL!");
4559 /* If the image supports fancy striping, get its parameters */
4561 if (rbd_dev->header.features & RBD_FEATURE_STRIPINGV2) {
4562 ret = rbd_dev_v2_striping_info(rbd_dev);
4567 /* crypto and compression type aren't (yet) supported for v2 images */
4569 rbd_dev->header.crypt_type = 0;
4570 rbd_dev->header.comp_type = 0;
4572 /* Get the snapshot context, plus the header version */
4574 ret = rbd_dev_v2_snap_context(rbd_dev);
4578 dout("discovered version 2 image, header name is %s\n",
4579 rbd_dev->header_name);
4583 rbd_dev->parent_overlap = 0;
4584 rbd_spec_put(rbd_dev->parent_spec);
4585 rbd_dev->parent_spec = NULL;
4586 kfree(rbd_dev->header_name);
4587 rbd_dev->header_name = NULL;
4588 kfree(rbd_dev->header.object_prefix);
4589 rbd_dev->header.object_prefix = NULL;
4594 static int rbd_dev_probe_parent(struct rbd_device *rbd_dev)
4596 struct rbd_device *parent = NULL;
4597 struct rbd_spec *parent_spec;
4598 struct rbd_client *rbdc;
4601 if (!rbd_dev->parent_spec)
4604 * We need to pass a reference to the client and the parent
4605 * spec when creating the parent rbd_dev. Images related by
4606 * parent/child relationships always share both.
4608 parent_spec = rbd_spec_get(rbd_dev->parent_spec);
4609 rbdc = __rbd_get_client(rbd_dev->rbd_client);
4612 parent = rbd_dev_create(rbdc, parent_spec);
4616 ret = rbd_dev_image_probe(parent, true);
4619 rbd_dev->parent = parent;
4624 rbd_spec_put(rbd_dev->parent_spec);
4625 kfree(rbd_dev->header_name);
4626 rbd_dev_destroy(parent);
4628 rbd_put_client(rbdc);
4629 rbd_spec_put(parent_spec);
4635 static int rbd_dev_device_setup(struct rbd_device *rbd_dev)
4639 /* generate unique id: find highest unique id, add one */
4640 rbd_dev_id_get(rbd_dev);
4642 /* Fill in the device name, now that we have its id. */
4643 BUILD_BUG_ON(DEV_NAME_LEN
4644 < sizeof (RBD_DRV_NAME) + MAX_INT_FORMAT_WIDTH);
4645 sprintf(rbd_dev->name, "%s%d", RBD_DRV_NAME, rbd_dev->dev_id);
4647 /* Get our block major device number. */
4649 ret = register_blkdev(0, rbd_dev->name);
4652 rbd_dev->major = ret;
4654 /* Set up the blkdev mapping. */
4656 ret = rbd_init_disk(rbd_dev);
4658 goto err_out_blkdev;
4660 ret = rbd_dev_mapping_set(rbd_dev);
4663 set_capacity(rbd_dev->disk, rbd_dev->mapping.size / SECTOR_SIZE);
4665 ret = rbd_bus_add_dev(rbd_dev);
4667 goto err_out_mapping;
4669 /* Everything's ready. Announce the disk to the world. */
4671 set_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags);
4672 add_disk(rbd_dev->disk);
4674 pr_info("%s: added with size 0x%llx\n", rbd_dev->disk->disk_name,
4675 (unsigned long long) rbd_dev->mapping.size);
4680 rbd_dev_mapping_clear(rbd_dev);
4682 rbd_free_disk(rbd_dev);
4684 unregister_blkdev(rbd_dev->major, rbd_dev->name);
4686 rbd_dev_id_put(rbd_dev);
4687 rbd_dev_mapping_clear(rbd_dev);
4692 static int rbd_dev_header_name(struct rbd_device *rbd_dev)
4694 struct rbd_spec *spec = rbd_dev->spec;
4697 /* Record the header object name for this rbd image. */
4699 rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
4701 if (rbd_dev->image_format == 1)
4702 size = strlen(spec->image_name) + sizeof (RBD_SUFFIX);
4704 size = sizeof (RBD_HEADER_PREFIX) + strlen(spec->image_id);
4706 rbd_dev->header_name = kmalloc(size, GFP_KERNEL);
4707 if (!rbd_dev->header_name)
4710 if (rbd_dev->image_format == 1)
4711 sprintf(rbd_dev->header_name, "%s%s",
4712 spec->image_name, RBD_SUFFIX);
4714 sprintf(rbd_dev->header_name, "%s%s",
4715 RBD_HEADER_PREFIX, spec->image_id);
4719 static void rbd_dev_image_release(struct rbd_device *rbd_dev)
4723 rbd_dev_unprobe(rbd_dev);
4724 ret = rbd_dev_header_watch_sync(rbd_dev, 0);
4726 rbd_warn(rbd_dev, "failed to cancel watch event (%d)\n", ret);
4727 kfree(rbd_dev->header_name);
4728 rbd_dev->header_name = NULL;
4729 rbd_dev->image_format = 0;
4730 kfree(rbd_dev->spec->image_id);
4731 rbd_dev->spec->image_id = NULL;
4733 rbd_dev_destroy(rbd_dev);
4737 * Probe for the existence of the header object for the given rbd
4738 * device. For format 2 images this includes determining the image
4741 static int rbd_dev_image_probe(struct rbd_device *rbd_dev, bool read_only)
4747 * Get the id from the image id object. If it's not a
4748 * format 2 image, we'll get ENOENT back, and we'll assume
4749 * it's a format 1 image.
4751 ret = rbd_dev_image_id(rbd_dev);
4754 rbd_assert(rbd_dev->spec->image_id);
4755 rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
4757 ret = rbd_dev_header_name(rbd_dev);
4759 goto err_out_format;
4761 ret = rbd_dev_header_watch_sync(rbd_dev, 1);
4763 goto out_header_name;
4765 if (rbd_dev->image_format == 1)
4766 ret = rbd_dev_v1_probe(rbd_dev);
4768 ret = rbd_dev_v2_probe(rbd_dev);
4772 ret = rbd_dev_spec_update(rbd_dev);
4776 /* If we are mapping a snapshot it must be marked read-only */
4778 if (rbd_dev->spec->snap_id != CEPH_NOSNAP)
4780 rbd_dev->mapping.read_only = read_only;
4782 ret = rbd_dev_probe_parent(rbd_dev);
4787 rbd_dev_unprobe(rbd_dev);
4789 tmp = rbd_dev_header_watch_sync(rbd_dev, 0);
4791 rbd_warn(rbd_dev, "unable to tear down watch request\n");
4793 kfree(rbd_dev->header_name);
4794 rbd_dev->header_name = NULL;
4796 rbd_dev->image_format = 0;
4797 kfree(rbd_dev->spec->image_id);
4798 rbd_dev->spec->image_id = NULL;
4800 dout("probe failed, returning %d\n", ret);
4805 static ssize_t rbd_add(struct bus_type *bus,
4809 struct rbd_device *rbd_dev = NULL;
4810 struct ceph_options *ceph_opts = NULL;
4811 struct rbd_options *rbd_opts = NULL;
4812 struct rbd_spec *spec = NULL;
4813 struct rbd_client *rbdc;
4814 struct ceph_osd_client *osdc;
4818 if (!try_module_get(THIS_MODULE))
4821 /* parse add command */
4822 rc = rbd_add_parse_args(buf, &ceph_opts, &rbd_opts, &spec);
4824 goto err_out_module;
4825 read_only = rbd_opts->read_only;
4827 rbd_opts = NULL; /* done with this */
4829 rbdc = rbd_get_client(ceph_opts);
4834 ceph_opts = NULL; /* rbd_dev client now owns this */
4837 osdc = &rbdc->client->osdc;
4838 rc = ceph_pg_poolid_by_name(osdc->osdmap, spec->pool_name);
4840 goto err_out_client;
4841 spec->pool_id = (u64)rc;
4843 /* The ceph file layout needs to fit pool id in 32 bits */
4845 if (spec->pool_id > (u64)U32_MAX) {
4846 rbd_warn(NULL, "pool id too large (%llu > %u)\n",
4847 (unsigned long long)spec->pool_id, U32_MAX);
4849 goto err_out_client;
4852 rbd_dev = rbd_dev_create(rbdc, spec);
4854 goto err_out_client;
4855 rbdc = NULL; /* rbd_dev now owns this */
4856 spec = NULL; /* rbd_dev now owns this */
4858 rc = rbd_dev_image_probe(rbd_dev, read_only);
4860 goto err_out_rbd_dev;
4862 rc = rbd_dev_device_setup(rbd_dev);
4866 rbd_dev_image_release(rbd_dev);
4868 rbd_dev_destroy(rbd_dev);
4870 rbd_put_client(rbdc);
4873 ceph_destroy_options(ceph_opts);
4877 module_put(THIS_MODULE);
4879 dout("Error adding device %s\n", buf);
4884 static struct rbd_device *__rbd_get_dev(unsigned long dev_id)
4886 struct list_head *tmp;
4887 struct rbd_device *rbd_dev;
4889 spin_lock(&rbd_dev_list_lock);
4890 list_for_each(tmp, &rbd_dev_list) {
4891 rbd_dev = list_entry(tmp, struct rbd_device, node);
4892 if (rbd_dev->dev_id == dev_id) {
4893 spin_unlock(&rbd_dev_list_lock);
4897 spin_unlock(&rbd_dev_list_lock);
4901 static void rbd_dev_device_release(struct device *dev)
4903 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
4905 rbd_free_disk(rbd_dev);
4906 clear_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags);
4907 rbd_dev_mapping_clear(rbd_dev);
4908 unregister_blkdev(rbd_dev->major, rbd_dev->name);
4910 rbd_dev_id_put(rbd_dev);
4911 rbd_dev_mapping_clear(rbd_dev);
4914 static void rbd_dev_remove_parent(struct rbd_device *rbd_dev)
4916 while (rbd_dev->parent) {
4917 struct rbd_device *first = rbd_dev;
4918 struct rbd_device *second = first->parent;
4919 struct rbd_device *third;
4922 * Follow to the parent with no grandparent and
4925 while (second && (third = second->parent)) {
4930 rbd_dev_image_release(second);
4931 first->parent = NULL;
4932 first->parent_overlap = 0;
4934 rbd_assert(first->parent_spec);
4935 rbd_spec_put(first->parent_spec);
4936 first->parent_spec = NULL;
4940 static ssize_t rbd_remove(struct bus_type *bus,
4944 struct rbd_device *rbd_dev = NULL;
4949 ret = strict_strtoul(buf, 10, &ul);
4953 /* convert to int; abort if we lost anything in the conversion */
4954 target_id = (int) ul;
4955 if (target_id != ul)
4958 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
4960 rbd_dev = __rbd_get_dev(target_id);
4966 spin_lock_irq(&rbd_dev->lock);
4967 if (rbd_dev->open_count)
4970 set_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags);
4971 spin_unlock_irq(&rbd_dev->lock);
4975 rbd_bus_del_dev(rbd_dev);
4976 rbd_dev_image_release(rbd_dev);
4977 module_put(THIS_MODULE);
4979 mutex_unlock(&ctl_mutex);
4985 * create control files in sysfs
4988 static int rbd_sysfs_init(void)
4992 ret = device_register(&rbd_root_dev);
4996 ret = bus_register(&rbd_bus_type);
4998 device_unregister(&rbd_root_dev);
5003 static void rbd_sysfs_cleanup(void)
5005 bus_unregister(&rbd_bus_type);
5006 device_unregister(&rbd_root_dev);
5009 static int rbd_slab_init(void)
5011 rbd_assert(!rbd_img_request_cache);
5012 rbd_img_request_cache = kmem_cache_create("rbd_img_request",
5013 sizeof (struct rbd_img_request),
5014 __alignof__(struct rbd_img_request),
5016 if (!rbd_img_request_cache)
5019 rbd_assert(!rbd_obj_request_cache);
5020 rbd_obj_request_cache = kmem_cache_create("rbd_obj_request",
5021 sizeof (struct rbd_obj_request),
5022 __alignof__(struct rbd_obj_request),
5024 if (!rbd_obj_request_cache)
5027 rbd_assert(!rbd_segment_name_cache);
5028 rbd_segment_name_cache = kmem_cache_create("rbd_segment_name",
5029 MAX_OBJ_NAME_SIZE + 1, 1, 0, NULL);
5030 if (rbd_segment_name_cache)
5033 if (rbd_obj_request_cache) {
5034 kmem_cache_destroy(rbd_obj_request_cache);
5035 rbd_obj_request_cache = NULL;
5038 kmem_cache_destroy(rbd_img_request_cache);
5039 rbd_img_request_cache = NULL;
5044 static void rbd_slab_exit(void)
5046 rbd_assert(rbd_segment_name_cache);
5047 kmem_cache_destroy(rbd_segment_name_cache);
5048 rbd_segment_name_cache = NULL;
5050 rbd_assert(rbd_obj_request_cache);
5051 kmem_cache_destroy(rbd_obj_request_cache);
5052 rbd_obj_request_cache = NULL;
5054 rbd_assert(rbd_img_request_cache);
5055 kmem_cache_destroy(rbd_img_request_cache);
5056 rbd_img_request_cache = NULL;
5059 static int __init rbd_init(void)
5063 if (!libceph_compatible(NULL)) {
5064 rbd_warn(NULL, "libceph incompatibility (quitting)");
5068 rc = rbd_slab_init();
5071 rc = rbd_sysfs_init();
5075 pr_info("loaded " RBD_DRV_NAME_LONG "\n");
5080 static void __exit rbd_exit(void)
5082 rbd_sysfs_cleanup();
5086 module_init(rbd_init);
5087 module_exit(rbd_exit);
5089 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
5090 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
5091 MODULE_DESCRIPTION("rados block device");
5093 /* following authorship retained from original osdblk.c */
5094 MODULE_AUTHOR("Jeff Garzik <jeff@garzik.org>");
5096 MODULE_LICENSE("GPL");