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>
44 #include "rbd_types.h"
46 #define RBD_DEBUG /* Activate rbd_assert() calls */
49 * The basic unit of block I/O is a sector. It is interpreted in a
50 * number of contexts in Linux (blk, bio, genhd), but the default is
51 * universally 512 bytes. These symbols are just slightly more
52 * meaningful than the bare numbers they represent.
54 #define SECTOR_SHIFT 9
55 #define SECTOR_SIZE (1ULL << SECTOR_SHIFT)
57 #define RBD_DRV_NAME "rbd"
58 #define RBD_DRV_NAME_LONG "rbd (rados block device)"
60 #define RBD_MINORS_PER_MAJOR 256 /* max minors per blkdev */
62 #define RBD_SNAP_DEV_NAME_PREFIX "snap_"
63 #define RBD_MAX_SNAP_NAME_LEN \
64 (NAME_MAX - (sizeof (RBD_SNAP_DEV_NAME_PREFIX) - 1))
66 #define RBD_MAX_SNAP_COUNT 510 /* allows max snapc to fit in 4KB */
68 #define RBD_SNAP_HEAD_NAME "-"
70 #define BAD_SNAP_INDEX U32_MAX /* invalid index into snap array */
72 /* This allows a single page to hold an image name sent by OSD */
73 #define RBD_IMAGE_NAME_LEN_MAX (PAGE_SIZE - sizeof (__le32) - 1)
74 #define RBD_IMAGE_ID_LEN_MAX 64
76 #define RBD_OBJ_PREFIX_LEN_MAX 64
80 #define RBD_FEATURE_LAYERING (1<<0)
81 #define RBD_FEATURE_STRIPINGV2 (1<<1)
82 #define RBD_FEATURES_ALL \
83 (RBD_FEATURE_LAYERING | RBD_FEATURE_STRIPINGV2)
85 /* Features supported by this (client software) implementation. */
87 #define RBD_FEATURES_SUPPORTED (RBD_FEATURES_ALL)
90 * An RBD device name will be "rbd#", where the "rbd" comes from
91 * RBD_DRV_NAME above, and # is a unique integer identifier.
92 * MAX_INT_FORMAT_WIDTH is used in ensuring DEV_NAME_LEN is big
93 * enough to hold all possible device names.
95 #define DEV_NAME_LEN 32
96 #define MAX_INT_FORMAT_WIDTH ((5 * sizeof (int)) / 2 + 1)
99 * block device image metadata (in-memory version)
101 struct rbd_image_header {
102 /* These four fields never change for a given rbd image */
109 /* The remaining fields need to be updated occasionally */
111 struct ceph_snap_context *snapc;
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 static int rbd_img_request_submit(struct rbd_img_request *img_request);
349 static void rbd_dev_device_release(struct device *dev);
351 static ssize_t rbd_add(struct bus_type *bus, const char *buf,
353 static ssize_t rbd_remove(struct bus_type *bus, const char *buf,
355 static int rbd_dev_image_probe(struct rbd_device *rbd_dev);
357 static struct bus_attribute rbd_bus_attrs[] = {
358 __ATTR(add, S_IWUSR, NULL, rbd_add),
359 __ATTR(remove, S_IWUSR, NULL, rbd_remove),
363 static struct bus_type rbd_bus_type = {
365 .bus_attrs = rbd_bus_attrs,
368 static void rbd_root_dev_release(struct device *dev)
372 static struct device rbd_root_dev = {
374 .release = rbd_root_dev_release,
377 static __printf(2, 3)
378 void rbd_warn(struct rbd_device *rbd_dev, const char *fmt, ...)
380 struct va_format vaf;
388 printk(KERN_WARNING "%s: %pV\n", RBD_DRV_NAME, &vaf);
389 else if (rbd_dev->disk)
390 printk(KERN_WARNING "%s: %s: %pV\n",
391 RBD_DRV_NAME, rbd_dev->disk->disk_name, &vaf);
392 else if (rbd_dev->spec && rbd_dev->spec->image_name)
393 printk(KERN_WARNING "%s: image %s: %pV\n",
394 RBD_DRV_NAME, rbd_dev->spec->image_name, &vaf);
395 else if (rbd_dev->spec && rbd_dev->spec->image_id)
396 printk(KERN_WARNING "%s: id %s: %pV\n",
397 RBD_DRV_NAME, rbd_dev->spec->image_id, &vaf);
399 printk(KERN_WARNING "%s: rbd_dev %p: %pV\n",
400 RBD_DRV_NAME, rbd_dev, &vaf);
405 #define rbd_assert(expr) \
406 if (unlikely(!(expr))) { \
407 printk(KERN_ERR "\nAssertion failure in %s() " \
409 "\trbd_assert(%s);\n\n", \
410 __func__, __LINE__, #expr); \
413 #else /* !RBD_DEBUG */
414 # define rbd_assert(expr) ((void) 0)
415 #endif /* !RBD_DEBUG */
417 static int rbd_img_obj_request_submit(struct rbd_obj_request *obj_request);
418 static void rbd_img_parent_read(struct rbd_obj_request *obj_request);
419 static void rbd_dev_remove_parent(struct rbd_device *rbd_dev);
421 static int rbd_dev_refresh(struct rbd_device *rbd_dev);
422 static int rbd_dev_v2_refresh(struct rbd_device *rbd_dev);
423 static const char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev,
425 static int _rbd_dev_v2_snap_size(struct rbd_device *rbd_dev, u64 snap_id,
426 u8 *order, u64 *snap_size);
427 static int _rbd_dev_v2_snap_features(struct rbd_device *rbd_dev, u64 snap_id,
429 static u64 rbd_snap_id_by_name(struct rbd_device *rbd_dev, const char *name);
431 static int rbd_open(struct block_device *bdev, fmode_t mode)
433 struct rbd_device *rbd_dev = bdev->bd_disk->private_data;
434 bool removing = false;
436 if ((mode & FMODE_WRITE) && rbd_dev->mapping.read_only)
439 spin_lock_irq(&rbd_dev->lock);
440 if (test_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags))
443 rbd_dev->open_count++;
444 spin_unlock_irq(&rbd_dev->lock);
448 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
449 (void) get_device(&rbd_dev->dev);
450 set_device_ro(bdev, rbd_dev->mapping.read_only);
451 mutex_unlock(&ctl_mutex);
456 static int rbd_release(struct gendisk *disk, fmode_t mode)
458 struct rbd_device *rbd_dev = disk->private_data;
459 unsigned long open_count_before;
461 spin_lock_irq(&rbd_dev->lock);
462 open_count_before = rbd_dev->open_count--;
463 spin_unlock_irq(&rbd_dev->lock);
464 rbd_assert(open_count_before > 0);
466 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
467 put_device(&rbd_dev->dev);
468 mutex_unlock(&ctl_mutex);
473 static const struct block_device_operations rbd_bd_ops = {
474 .owner = THIS_MODULE,
476 .release = rbd_release,
480 * Initialize an rbd client instance.
483 static struct rbd_client *rbd_client_create(struct ceph_options *ceph_opts)
485 struct rbd_client *rbdc;
488 dout("%s:\n", __func__);
489 rbdc = kmalloc(sizeof(struct rbd_client), GFP_KERNEL);
493 kref_init(&rbdc->kref);
494 INIT_LIST_HEAD(&rbdc->node);
496 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
498 rbdc->client = ceph_create_client(ceph_opts, rbdc, 0, 0);
499 if (IS_ERR(rbdc->client))
501 ceph_opts = NULL; /* Now rbdc->client is responsible for ceph_opts */
503 ret = ceph_open_session(rbdc->client);
507 spin_lock(&rbd_client_list_lock);
508 list_add_tail(&rbdc->node, &rbd_client_list);
509 spin_unlock(&rbd_client_list_lock);
511 mutex_unlock(&ctl_mutex);
512 dout("%s: rbdc %p\n", __func__, rbdc);
517 ceph_destroy_client(rbdc->client);
519 mutex_unlock(&ctl_mutex);
523 ceph_destroy_options(ceph_opts);
524 dout("%s: error %d\n", __func__, ret);
529 static struct rbd_client *__rbd_get_client(struct rbd_client *rbdc)
531 kref_get(&rbdc->kref);
537 * Find a ceph client with specific addr and configuration. If
538 * found, bump its reference count.
540 static struct rbd_client *rbd_client_find(struct ceph_options *ceph_opts)
542 struct rbd_client *client_node;
545 if (ceph_opts->flags & CEPH_OPT_NOSHARE)
548 spin_lock(&rbd_client_list_lock);
549 list_for_each_entry(client_node, &rbd_client_list, node) {
550 if (!ceph_compare_options(ceph_opts, client_node->client)) {
551 __rbd_get_client(client_node);
557 spin_unlock(&rbd_client_list_lock);
559 return found ? client_node : NULL;
569 /* string args above */
572 /* Boolean args above */
576 static match_table_t rbd_opts_tokens = {
578 /* string args above */
579 {Opt_read_only, "read_only"},
580 {Opt_read_only, "ro"}, /* Alternate spelling */
581 {Opt_read_write, "read_write"},
582 {Opt_read_write, "rw"}, /* Alternate spelling */
583 /* Boolean args above */
591 #define RBD_READ_ONLY_DEFAULT false
593 static int parse_rbd_opts_token(char *c, void *private)
595 struct rbd_options *rbd_opts = private;
596 substring_t argstr[MAX_OPT_ARGS];
597 int token, intval, ret;
599 token = match_token(c, rbd_opts_tokens, argstr);
603 if (token < Opt_last_int) {
604 ret = match_int(&argstr[0], &intval);
606 pr_err("bad mount option arg (not int) "
610 dout("got int token %d val %d\n", token, intval);
611 } else if (token > Opt_last_int && token < Opt_last_string) {
612 dout("got string token %d val %s\n", token,
614 } else if (token > Opt_last_string && token < Opt_last_bool) {
615 dout("got Boolean token %d\n", token);
617 dout("got token %d\n", token);
622 rbd_opts->read_only = true;
625 rbd_opts->read_only = false;
635 * Get a ceph client with specific addr and configuration, if one does
636 * not exist create it.
638 static struct rbd_client *rbd_get_client(struct ceph_options *ceph_opts)
640 struct rbd_client *rbdc;
642 rbdc = rbd_client_find(ceph_opts);
643 if (rbdc) /* using an existing client */
644 ceph_destroy_options(ceph_opts);
646 rbdc = rbd_client_create(ceph_opts);
652 * Destroy ceph client
654 * Caller must hold rbd_client_list_lock.
656 static void rbd_client_release(struct kref *kref)
658 struct rbd_client *rbdc = container_of(kref, struct rbd_client, kref);
660 dout("%s: rbdc %p\n", __func__, rbdc);
661 spin_lock(&rbd_client_list_lock);
662 list_del(&rbdc->node);
663 spin_unlock(&rbd_client_list_lock);
665 ceph_destroy_client(rbdc->client);
670 * Drop reference to ceph client node. If it's not referenced anymore, release
673 static void rbd_put_client(struct rbd_client *rbdc)
676 kref_put(&rbdc->kref, rbd_client_release);
679 static bool rbd_image_format_valid(u32 image_format)
681 return image_format == 1 || image_format == 2;
684 static bool rbd_dev_ondisk_valid(struct rbd_image_header_ondisk *ondisk)
689 /* The header has to start with the magic rbd header text */
690 if (memcmp(&ondisk->text, RBD_HEADER_TEXT, sizeof (RBD_HEADER_TEXT)))
693 /* The bio layer requires at least sector-sized I/O */
695 if (ondisk->options.order < SECTOR_SHIFT)
698 /* If we use u64 in a few spots we may be able to loosen this */
700 if (ondisk->options.order > 8 * sizeof (int) - 1)
704 * The size of a snapshot header has to fit in a size_t, and
705 * that limits the number of snapshots.
707 snap_count = le32_to_cpu(ondisk->snap_count);
708 size = SIZE_MAX - sizeof (struct ceph_snap_context);
709 if (snap_count > size / sizeof (__le64))
713 * Not only that, but the size of the entire the snapshot
714 * header must also be representable in a size_t.
716 size -= snap_count * sizeof (__le64);
717 if ((u64) size < le64_to_cpu(ondisk->snap_names_len))
724 * Create a new header structure, translate header format from the on-disk
727 static int rbd_header_from_disk(struct rbd_image_header *header,
728 struct rbd_image_header_ondisk *ondisk)
735 memset(header, 0, sizeof (*header));
737 snap_count = le32_to_cpu(ondisk->snap_count);
739 len = strnlen(ondisk->object_prefix, sizeof (ondisk->object_prefix));
740 header->object_prefix = kmalloc(len + 1, GFP_KERNEL);
741 if (!header->object_prefix)
743 memcpy(header->object_prefix, ondisk->object_prefix, len);
744 header->object_prefix[len] = '\0';
747 u64 snap_names_len = le64_to_cpu(ondisk->snap_names_len);
749 /* Save a copy of the snapshot names */
751 if (snap_names_len > (u64) SIZE_MAX)
753 header->snap_names = kmalloc(snap_names_len, GFP_KERNEL);
754 if (!header->snap_names)
757 * Note that rbd_dev_v1_header_read() guarantees
758 * the ondisk buffer we're working with has
759 * snap_names_len bytes beyond the end of the
760 * snapshot id array, this memcpy() is safe.
762 memcpy(header->snap_names, &ondisk->snaps[snap_count],
765 /* Record each snapshot's size */
767 size = snap_count * sizeof (*header->snap_sizes);
768 header->snap_sizes = kmalloc(size, GFP_KERNEL);
769 if (!header->snap_sizes)
771 for (i = 0; i < snap_count; i++)
772 header->snap_sizes[i] =
773 le64_to_cpu(ondisk->snaps[i].image_size);
775 header->snap_names = NULL;
776 header->snap_sizes = NULL;
779 header->features = 0; /* No features support in v1 images */
780 header->obj_order = ondisk->options.order;
781 header->crypt_type = ondisk->options.crypt_type;
782 header->comp_type = ondisk->options.comp_type;
784 /* Allocate and fill in the snapshot context */
786 header->image_size = le64_to_cpu(ondisk->image_size);
788 header->snapc = ceph_create_snap_context(snap_count, GFP_KERNEL);
791 header->snapc->seq = le64_to_cpu(ondisk->snap_seq);
792 for (i = 0; i < snap_count; i++)
793 header->snapc->snaps[i] = le64_to_cpu(ondisk->snaps[i].id);
798 kfree(header->snap_sizes);
799 header->snap_sizes = NULL;
800 kfree(header->snap_names);
801 header->snap_names = NULL;
802 kfree(header->object_prefix);
803 header->object_prefix = NULL;
808 static const char *_rbd_dev_v1_snap_name(struct rbd_device *rbd_dev, u32 which)
810 const char *snap_name;
812 rbd_assert(which < rbd_dev->header.snapc->num_snaps);
814 /* Skip over names until we find the one we are looking for */
816 snap_name = rbd_dev->header.snap_names;
818 snap_name += strlen(snap_name) + 1;
820 return kstrdup(snap_name, GFP_KERNEL);
824 * Snapshot id comparison function for use with qsort()/bsearch().
825 * Note that result is for snapshots in *descending* order.
827 static int snapid_compare_reverse(const void *s1, const void *s2)
829 u64 snap_id1 = *(u64 *)s1;
830 u64 snap_id2 = *(u64 *)s2;
832 if (snap_id1 < snap_id2)
834 return snap_id1 == snap_id2 ? 0 : -1;
838 * Search a snapshot context to see if the given snapshot id is
841 * Returns the position of the snapshot id in the array if it's found,
842 * or BAD_SNAP_INDEX otherwise.
844 * Note: The snapshot array is in kept sorted (by the osd) in
845 * reverse order, highest snapshot id first.
847 static u32 rbd_dev_snap_index(struct rbd_device *rbd_dev, u64 snap_id)
849 struct ceph_snap_context *snapc = rbd_dev->header.snapc;
852 found = bsearch(&snap_id, &snapc->snaps, snapc->num_snaps,
853 sizeof (snap_id), snapid_compare_reverse);
855 return found ? (u32)(found - &snapc->snaps[0]) : BAD_SNAP_INDEX;
858 static const char *rbd_dev_v1_snap_name(struct rbd_device *rbd_dev,
863 which = rbd_dev_snap_index(rbd_dev, snap_id);
864 if (which == BAD_SNAP_INDEX)
867 return _rbd_dev_v1_snap_name(rbd_dev, which);
870 static const char *rbd_snap_name(struct rbd_device *rbd_dev, u64 snap_id)
872 if (snap_id == CEPH_NOSNAP)
873 return RBD_SNAP_HEAD_NAME;
875 rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
876 if (rbd_dev->image_format == 1)
877 return rbd_dev_v1_snap_name(rbd_dev, snap_id);
879 return rbd_dev_v2_snap_name(rbd_dev, snap_id);
882 static int rbd_snap_size(struct rbd_device *rbd_dev, u64 snap_id,
885 rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
886 if (snap_id == CEPH_NOSNAP) {
887 *snap_size = rbd_dev->header.image_size;
888 } else if (rbd_dev->image_format == 1) {
891 which = rbd_dev_snap_index(rbd_dev, snap_id);
892 if (which == BAD_SNAP_INDEX)
895 *snap_size = rbd_dev->header.snap_sizes[which];
900 ret = _rbd_dev_v2_snap_size(rbd_dev, snap_id, NULL, &size);
909 static int rbd_snap_features(struct rbd_device *rbd_dev, u64 snap_id,
912 rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
913 if (snap_id == CEPH_NOSNAP) {
914 *snap_features = rbd_dev->header.features;
915 } else if (rbd_dev->image_format == 1) {
916 *snap_features = 0; /* No features for format 1 */
921 ret = _rbd_dev_v2_snap_features(rbd_dev, snap_id, &features);
925 *snap_features = features;
930 static int rbd_dev_mapping_set(struct rbd_device *rbd_dev)
932 const char *snap_name = rbd_dev->spec->snap_name;
938 if (strcmp(snap_name, RBD_SNAP_HEAD_NAME)) {
939 snap_id = rbd_snap_id_by_name(rbd_dev, snap_name);
940 if (snap_id == CEPH_NOSNAP)
943 snap_id = CEPH_NOSNAP;
946 ret = rbd_snap_size(rbd_dev, snap_id, &size);
949 ret = rbd_snap_features(rbd_dev, snap_id, &features);
953 rbd_dev->mapping.size = size;
954 rbd_dev->mapping.features = features;
956 /* If we are mapping a snapshot it must be marked read-only */
958 if (snap_id != CEPH_NOSNAP)
959 rbd_dev->mapping.read_only = true;
964 static void rbd_dev_mapping_clear(struct rbd_device *rbd_dev)
966 rbd_dev->mapping.size = 0;
967 rbd_dev->mapping.features = 0;
968 rbd_dev->mapping.read_only = true;
971 static void rbd_dev_clear_mapping(struct rbd_device *rbd_dev)
973 rbd_dev->mapping.size = 0;
974 rbd_dev->mapping.features = 0;
975 rbd_dev->mapping.read_only = true;
978 static const char *rbd_segment_name(struct rbd_device *rbd_dev, u64 offset)
984 name = kmalloc(MAX_OBJ_NAME_SIZE + 1, GFP_NOIO);
987 segment = offset >> rbd_dev->header.obj_order;
988 ret = snprintf(name, MAX_OBJ_NAME_SIZE + 1, "%s.%012llx",
989 rbd_dev->header.object_prefix, segment);
990 if (ret < 0 || ret > MAX_OBJ_NAME_SIZE) {
991 pr_err("error formatting segment name for #%llu (%d)\n",
1000 static u64 rbd_segment_offset(struct rbd_device *rbd_dev, u64 offset)
1002 u64 segment_size = (u64) 1 << rbd_dev->header.obj_order;
1004 return offset & (segment_size - 1);
1007 static u64 rbd_segment_length(struct rbd_device *rbd_dev,
1008 u64 offset, u64 length)
1010 u64 segment_size = (u64) 1 << rbd_dev->header.obj_order;
1012 offset &= segment_size - 1;
1014 rbd_assert(length <= U64_MAX - offset);
1015 if (offset + length > segment_size)
1016 length = segment_size - offset;
1022 * returns the size of an object in the image
1024 static u64 rbd_obj_bytes(struct rbd_image_header *header)
1026 return 1 << header->obj_order;
1033 static void bio_chain_put(struct bio *chain)
1039 chain = chain->bi_next;
1045 * zeros a bio chain, starting at specific offset
1047 static void zero_bio_chain(struct bio *chain, int start_ofs)
1050 unsigned long flags;
1056 bio_for_each_segment(bv, chain, i) {
1057 if (pos + bv->bv_len > start_ofs) {
1058 int remainder = max(start_ofs - pos, 0);
1059 buf = bvec_kmap_irq(bv, &flags);
1060 memset(buf + remainder, 0,
1061 bv->bv_len - remainder);
1062 bvec_kunmap_irq(buf, &flags);
1067 chain = chain->bi_next;
1072 * similar to zero_bio_chain(), zeros data defined by a page array,
1073 * starting at the given byte offset from the start of the array and
1074 * continuing up to the given end offset. The pages array is
1075 * assumed to be big enough to hold all bytes up to the end.
1077 static void zero_pages(struct page **pages, u64 offset, u64 end)
1079 struct page **page = &pages[offset >> PAGE_SHIFT];
1081 rbd_assert(end > offset);
1082 rbd_assert(end - offset <= (u64)SIZE_MAX);
1083 while (offset < end) {
1086 unsigned long flags;
1089 page_offset = (size_t)(offset & ~PAGE_MASK);
1090 length = min(PAGE_SIZE - page_offset, (size_t)(end - offset));
1091 local_irq_save(flags);
1092 kaddr = kmap_atomic(*page);
1093 memset(kaddr + page_offset, 0, length);
1094 kunmap_atomic(kaddr);
1095 local_irq_restore(flags);
1103 * Clone a portion of a bio, starting at the given byte offset
1104 * and continuing for the number of bytes indicated.
1106 static struct bio *bio_clone_range(struct bio *bio_src,
1107 unsigned int offset,
1115 unsigned short end_idx;
1116 unsigned short vcnt;
1119 /* Handle the easy case for the caller */
1121 if (!offset && len == bio_src->bi_size)
1122 return bio_clone(bio_src, gfpmask);
1124 if (WARN_ON_ONCE(!len))
1126 if (WARN_ON_ONCE(len > bio_src->bi_size))
1128 if (WARN_ON_ONCE(offset > bio_src->bi_size - len))
1131 /* Find first affected segment... */
1134 __bio_for_each_segment(bv, bio_src, idx, 0) {
1135 if (resid < bv->bv_len)
1137 resid -= bv->bv_len;
1141 /* ...and the last affected segment */
1144 __bio_for_each_segment(bv, bio_src, end_idx, idx) {
1145 if (resid <= bv->bv_len)
1147 resid -= bv->bv_len;
1149 vcnt = end_idx - idx + 1;
1151 /* Build the clone */
1153 bio = bio_alloc(gfpmask, (unsigned int) vcnt);
1155 return NULL; /* ENOMEM */
1157 bio->bi_bdev = bio_src->bi_bdev;
1158 bio->bi_sector = bio_src->bi_sector + (offset >> SECTOR_SHIFT);
1159 bio->bi_rw = bio_src->bi_rw;
1160 bio->bi_flags |= 1 << BIO_CLONED;
1163 * Copy over our part of the bio_vec, then update the first
1164 * and last (or only) entries.
1166 memcpy(&bio->bi_io_vec[0], &bio_src->bi_io_vec[idx],
1167 vcnt * sizeof (struct bio_vec));
1168 bio->bi_io_vec[0].bv_offset += voff;
1170 bio->bi_io_vec[0].bv_len -= voff;
1171 bio->bi_io_vec[vcnt - 1].bv_len = resid;
1173 bio->bi_io_vec[0].bv_len = len;
1176 bio->bi_vcnt = vcnt;
1184 * Clone a portion of a bio chain, starting at the given byte offset
1185 * into the first bio in the source chain and continuing for the
1186 * number of bytes indicated. The result is another bio chain of
1187 * exactly the given length, or a null pointer on error.
1189 * The bio_src and offset parameters are both in-out. On entry they
1190 * refer to the first source bio and the offset into that bio where
1191 * the start of data to be cloned is located.
1193 * On return, bio_src is updated to refer to the bio in the source
1194 * chain that contains first un-cloned byte, and *offset will
1195 * contain the offset of that byte within that bio.
1197 static struct bio *bio_chain_clone_range(struct bio **bio_src,
1198 unsigned int *offset,
1202 struct bio *bi = *bio_src;
1203 unsigned int off = *offset;
1204 struct bio *chain = NULL;
1207 /* Build up a chain of clone bios up to the limit */
1209 if (!bi || off >= bi->bi_size || !len)
1210 return NULL; /* Nothing to clone */
1214 unsigned int bi_size;
1218 rbd_warn(NULL, "bio_chain exhausted with %u left", len);
1219 goto out_err; /* EINVAL; ran out of bio's */
1221 bi_size = min_t(unsigned int, bi->bi_size - off, len);
1222 bio = bio_clone_range(bi, off, bi_size, gfpmask);
1224 goto out_err; /* ENOMEM */
1227 end = &bio->bi_next;
1230 if (off == bi->bi_size) {
1241 bio_chain_put(chain);
1247 * The default/initial value for all object request flags is 0. For
1248 * each flag, once its value is set to 1 it is never reset to 0
1251 static void obj_request_img_data_set(struct rbd_obj_request *obj_request)
1253 if (test_and_set_bit(OBJ_REQ_IMG_DATA, &obj_request->flags)) {
1254 struct rbd_device *rbd_dev;
1256 rbd_dev = obj_request->img_request->rbd_dev;
1257 rbd_warn(rbd_dev, "obj_request %p already marked img_data\n",
1262 static bool obj_request_img_data_test(struct rbd_obj_request *obj_request)
1265 return test_bit(OBJ_REQ_IMG_DATA, &obj_request->flags) != 0;
1268 static void obj_request_done_set(struct rbd_obj_request *obj_request)
1270 if (test_and_set_bit(OBJ_REQ_DONE, &obj_request->flags)) {
1271 struct rbd_device *rbd_dev = NULL;
1273 if (obj_request_img_data_test(obj_request))
1274 rbd_dev = obj_request->img_request->rbd_dev;
1275 rbd_warn(rbd_dev, "obj_request %p already marked done\n",
1280 static bool obj_request_done_test(struct rbd_obj_request *obj_request)
1283 return test_bit(OBJ_REQ_DONE, &obj_request->flags) != 0;
1287 * This sets the KNOWN flag after (possibly) setting the EXISTS
1288 * flag. The latter is set based on the "exists" value provided.
1290 * Note that for our purposes once an object exists it never goes
1291 * away again. It's possible that the response from two existence
1292 * checks are separated by the creation of the target object, and
1293 * the first ("doesn't exist") response arrives *after* the second
1294 * ("does exist"). In that case we ignore the second one.
1296 static void obj_request_existence_set(struct rbd_obj_request *obj_request,
1300 set_bit(OBJ_REQ_EXISTS, &obj_request->flags);
1301 set_bit(OBJ_REQ_KNOWN, &obj_request->flags);
1305 static bool obj_request_known_test(struct rbd_obj_request *obj_request)
1308 return test_bit(OBJ_REQ_KNOWN, &obj_request->flags) != 0;
1311 static bool obj_request_exists_test(struct rbd_obj_request *obj_request)
1314 return test_bit(OBJ_REQ_EXISTS, &obj_request->flags) != 0;
1317 static void rbd_obj_request_get(struct rbd_obj_request *obj_request)
1319 dout("%s: obj %p (was %d)\n", __func__, obj_request,
1320 atomic_read(&obj_request->kref.refcount));
1321 kref_get(&obj_request->kref);
1324 static void rbd_obj_request_destroy(struct kref *kref);
1325 static void rbd_obj_request_put(struct rbd_obj_request *obj_request)
1327 rbd_assert(obj_request != NULL);
1328 dout("%s: obj %p (was %d)\n", __func__, obj_request,
1329 atomic_read(&obj_request->kref.refcount));
1330 kref_put(&obj_request->kref, rbd_obj_request_destroy);
1333 static void rbd_img_request_get(struct rbd_img_request *img_request)
1335 dout("%s: img %p (was %d)\n", __func__, img_request,
1336 atomic_read(&img_request->kref.refcount));
1337 kref_get(&img_request->kref);
1340 static void rbd_img_request_destroy(struct kref *kref);
1341 static void rbd_img_request_put(struct rbd_img_request *img_request)
1343 rbd_assert(img_request != NULL);
1344 dout("%s: img %p (was %d)\n", __func__, img_request,
1345 atomic_read(&img_request->kref.refcount));
1346 kref_put(&img_request->kref, rbd_img_request_destroy);
1349 static inline void rbd_img_obj_request_add(struct rbd_img_request *img_request,
1350 struct rbd_obj_request *obj_request)
1352 rbd_assert(obj_request->img_request == NULL);
1354 /* Image request now owns object's original reference */
1355 obj_request->img_request = img_request;
1356 obj_request->which = img_request->obj_request_count;
1357 rbd_assert(!obj_request_img_data_test(obj_request));
1358 obj_request_img_data_set(obj_request);
1359 rbd_assert(obj_request->which != BAD_WHICH);
1360 img_request->obj_request_count++;
1361 list_add_tail(&obj_request->links, &img_request->obj_requests);
1362 dout("%s: img %p obj %p w=%u\n", __func__, img_request, obj_request,
1363 obj_request->which);
1366 static inline void rbd_img_obj_request_del(struct rbd_img_request *img_request,
1367 struct rbd_obj_request *obj_request)
1369 rbd_assert(obj_request->which != BAD_WHICH);
1371 dout("%s: img %p obj %p w=%u\n", __func__, img_request, obj_request,
1372 obj_request->which);
1373 list_del(&obj_request->links);
1374 rbd_assert(img_request->obj_request_count > 0);
1375 img_request->obj_request_count--;
1376 rbd_assert(obj_request->which == img_request->obj_request_count);
1377 obj_request->which = BAD_WHICH;
1378 rbd_assert(obj_request_img_data_test(obj_request));
1379 rbd_assert(obj_request->img_request == img_request);
1380 obj_request->img_request = NULL;
1381 obj_request->callback = NULL;
1382 rbd_obj_request_put(obj_request);
1385 static bool obj_request_type_valid(enum obj_request_type type)
1388 case OBJ_REQUEST_NODATA:
1389 case OBJ_REQUEST_BIO:
1390 case OBJ_REQUEST_PAGES:
1397 static int rbd_obj_request_submit(struct ceph_osd_client *osdc,
1398 struct rbd_obj_request *obj_request)
1400 dout("%s: osdc %p obj %p\n", __func__, osdc, obj_request);
1402 return ceph_osdc_start_request(osdc, obj_request->osd_req, false);
1405 static void rbd_img_request_complete(struct rbd_img_request *img_request)
1408 dout("%s: img %p\n", __func__, img_request);
1411 * If no error occurred, compute the aggregate transfer
1412 * count for the image request. We could instead use
1413 * atomic64_cmpxchg() to update it as each object request
1414 * completes; not clear which way is better off hand.
1416 if (!img_request->result) {
1417 struct rbd_obj_request *obj_request;
1420 for_each_obj_request(img_request, obj_request)
1421 xferred += obj_request->xferred;
1422 img_request->xferred = xferred;
1425 if (img_request->callback)
1426 img_request->callback(img_request);
1428 rbd_img_request_put(img_request);
1431 /* Caller is responsible for rbd_obj_request_destroy(obj_request) */
1433 static int rbd_obj_request_wait(struct rbd_obj_request *obj_request)
1435 dout("%s: obj %p\n", __func__, obj_request);
1437 return wait_for_completion_interruptible(&obj_request->completion);
1441 * The default/initial value for all image request flags is 0. Each
1442 * is conditionally set to 1 at image request initialization time
1443 * and currently never change thereafter.
1445 static void img_request_write_set(struct rbd_img_request *img_request)
1447 set_bit(IMG_REQ_WRITE, &img_request->flags);
1451 static bool img_request_write_test(struct rbd_img_request *img_request)
1454 return test_bit(IMG_REQ_WRITE, &img_request->flags) != 0;
1457 static void img_request_child_set(struct rbd_img_request *img_request)
1459 set_bit(IMG_REQ_CHILD, &img_request->flags);
1463 static bool img_request_child_test(struct rbd_img_request *img_request)
1466 return test_bit(IMG_REQ_CHILD, &img_request->flags) != 0;
1469 static void img_request_layered_set(struct rbd_img_request *img_request)
1471 set_bit(IMG_REQ_LAYERED, &img_request->flags);
1475 static bool img_request_layered_test(struct rbd_img_request *img_request)
1478 return test_bit(IMG_REQ_LAYERED, &img_request->flags) != 0;
1482 rbd_img_obj_request_read_callback(struct rbd_obj_request *obj_request)
1484 u64 xferred = obj_request->xferred;
1485 u64 length = obj_request->length;
1487 dout("%s: obj %p img %p result %d %llu/%llu\n", __func__,
1488 obj_request, obj_request->img_request, obj_request->result,
1491 * ENOENT means a hole in the image. We zero-fill the
1492 * entire length of the request. A short read also implies
1493 * zero-fill to the end of the request. Either way we
1494 * update the xferred count to indicate the whole request
1497 rbd_assert(obj_request->type != OBJ_REQUEST_NODATA);
1498 if (obj_request->result == -ENOENT) {
1499 if (obj_request->type == OBJ_REQUEST_BIO)
1500 zero_bio_chain(obj_request->bio_list, 0);
1502 zero_pages(obj_request->pages, 0, length);
1503 obj_request->result = 0;
1504 obj_request->xferred = length;
1505 } else if (xferred < length && !obj_request->result) {
1506 if (obj_request->type == OBJ_REQUEST_BIO)
1507 zero_bio_chain(obj_request->bio_list, xferred);
1509 zero_pages(obj_request->pages, xferred, length);
1510 obj_request->xferred = length;
1512 obj_request_done_set(obj_request);
1515 static void rbd_obj_request_complete(struct rbd_obj_request *obj_request)
1517 dout("%s: obj %p cb %p\n", __func__, obj_request,
1518 obj_request->callback);
1519 if (obj_request->callback)
1520 obj_request->callback(obj_request);
1522 complete_all(&obj_request->completion);
1525 static void rbd_osd_trivial_callback(struct rbd_obj_request *obj_request)
1527 dout("%s: obj %p\n", __func__, obj_request);
1528 obj_request_done_set(obj_request);
1531 static void rbd_osd_read_callback(struct rbd_obj_request *obj_request)
1533 struct rbd_img_request *img_request = NULL;
1534 struct rbd_device *rbd_dev = NULL;
1535 bool layered = false;
1537 if (obj_request_img_data_test(obj_request)) {
1538 img_request = obj_request->img_request;
1539 layered = img_request && img_request_layered_test(img_request);
1540 rbd_dev = img_request->rbd_dev;
1543 dout("%s: obj %p img %p result %d %llu/%llu\n", __func__,
1544 obj_request, img_request, obj_request->result,
1545 obj_request->xferred, obj_request->length);
1546 if (layered && obj_request->result == -ENOENT &&
1547 obj_request->img_offset < rbd_dev->parent_overlap)
1548 rbd_img_parent_read(obj_request);
1549 else if (img_request)
1550 rbd_img_obj_request_read_callback(obj_request);
1552 obj_request_done_set(obj_request);
1555 static void rbd_osd_write_callback(struct rbd_obj_request *obj_request)
1557 dout("%s: obj %p result %d %llu\n", __func__, obj_request,
1558 obj_request->result, obj_request->length);
1560 * There is no such thing as a successful short write. Set
1561 * it to our originally-requested length.
1563 obj_request->xferred = obj_request->length;
1564 obj_request_done_set(obj_request);
1568 * For a simple stat call there's nothing to do. We'll do more if
1569 * this is part of a write sequence for a layered image.
1571 static void rbd_osd_stat_callback(struct rbd_obj_request *obj_request)
1573 dout("%s: obj %p\n", __func__, obj_request);
1574 obj_request_done_set(obj_request);
1577 static void rbd_osd_req_callback(struct ceph_osd_request *osd_req,
1578 struct ceph_msg *msg)
1580 struct rbd_obj_request *obj_request = osd_req->r_priv;
1583 dout("%s: osd_req %p msg %p\n", __func__, osd_req, msg);
1584 rbd_assert(osd_req == obj_request->osd_req);
1585 if (obj_request_img_data_test(obj_request)) {
1586 rbd_assert(obj_request->img_request);
1587 rbd_assert(obj_request->which != BAD_WHICH);
1589 rbd_assert(obj_request->which == BAD_WHICH);
1592 if (osd_req->r_result < 0)
1593 obj_request->result = osd_req->r_result;
1595 BUG_ON(osd_req->r_num_ops > 2);
1598 * We support a 64-bit length, but ultimately it has to be
1599 * passed to blk_end_request(), which takes an unsigned int.
1601 obj_request->xferred = osd_req->r_reply_op_len[0];
1602 rbd_assert(obj_request->xferred < (u64)UINT_MAX);
1603 opcode = osd_req->r_ops[0].op;
1605 case CEPH_OSD_OP_READ:
1606 rbd_osd_read_callback(obj_request);
1608 case CEPH_OSD_OP_WRITE:
1609 rbd_osd_write_callback(obj_request);
1611 case CEPH_OSD_OP_STAT:
1612 rbd_osd_stat_callback(obj_request);
1614 case CEPH_OSD_OP_CALL:
1615 case CEPH_OSD_OP_NOTIFY_ACK:
1616 case CEPH_OSD_OP_WATCH:
1617 rbd_osd_trivial_callback(obj_request);
1620 rbd_warn(NULL, "%s: unsupported op %hu\n",
1621 obj_request->object_name, (unsigned short) opcode);
1625 if (obj_request_done_test(obj_request))
1626 rbd_obj_request_complete(obj_request);
1629 static void rbd_osd_req_format_read(struct rbd_obj_request *obj_request)
1631 struct rbd_img_request *img_request = obj_request->img_request;
1632 struct ceph_osd_request *osd_req = obj_request->osd_req;
1635 rbd_assert(osd_req != NULL);
1637 snap_id = img_request ? img_request->snap_id : CEPH_NOSNAP;
1638 ceph_osdc_build_request(osd_req, obj_request->offset,
1639 NULL, snap_id, NULL);
1642 static void rbd_osd_req_format_write(struct rbd_obj_request *obj_request)
1644 struct rbd_img_request *img_request = obj_request->img_request;
1645 struct ceph_osd_request *osd_req = obj_request->osd_req;
1646 struct ceph_snap_context *snapc;
1647 struct timespec mtime = CURRENT_TIME;
1649 rbd_assert(osd_req != NULL);
1651 snapc = img_request ? img_request->snapc : NULL;
1652 ceph_osdc_build_request(osd_req, obj_request->offset,
1653 snapc, CEPH_NOSNAP, &mtime);
1656 static struct ceph_osd_request *rbd_osd_req_create(
1657 struct rbd_device *rbd_dev,
1659 struct rbd_obj_request *obj_request)
1661 struct ceph_snap_context *snapc = NULL;
1662 struct ceph_osd_client *osdc;
1663 struct ceph_osd_request *osd_req;
1665 if (obj_request_img_data_test(obj_request)) {
1666 struct rbd_img_request *img_request = obj_request->img_request;
1668 rbd_assert(write_request ==
1669 img_request_write_test(img_request));
1671 snapc = img_request->snapc;
1674 /* Allocate and initialize the request, for the single op */
1676 osdc = &rbd_dev->rbd_client->client->osdc;
1677 osd_req = ceph_osdc_alloc_request(osdc, snapc, 1, false, GFP_ATOMIC);
1679 return NULL; /* ENOMEM */
1682 osd_req->r_flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK;
1684 osd_req->r_flags = CEPH_OSD_FLAG_READ;
1686 osd_req->r_callback = rbd_osd_req_callback;
1687 osd_req->r_priv = obj_request;
1689 osd_req->r_oid_len = strlen(obj_request->object_name);
1690 rbd_assert(osd_req->r_oid_len < sizeof (osd_req->r_oid));
1691 memcpy(osd_req->r_oid, obj_request->object_name, osd_req->r_oid_len);
1693 osd_req->r_file_layout = rbd_dev->layout; /* struct */
1699 * Create a copyup osd request based on the information in the
1700 * object request supplied. A copyup request has two osd ops,
1701 * a copyup method call, and a "normal" write request.
1703 static struct ceph_osd_request *
1704 rbd_osd_req_create_copyup(struct rbd_obj_request *obj_request)
1706 struct rbd_img_request *img_request;
1707 struct ceph_snap_context *snapc;
1708 struct rbd_device *rbd_dev;
1709 struct ceph_osd_client *osdc;
1710 struct ceph_osd_request *osd_req;
1712 rbd_assert(obj_request_img_data_test(obj_request));
1713 img_request = obj_request->img_request;
1714 rbd_assert(img_request);
1715 rbd_assert(img_request_write_test(img_request));
1717 /* Allocate and initialize the request, for the two ops */
1719 snapc = img_request->snapc;
1720 rbd_dev = img_request->rbd_dev;
1721 osdc = &rbd_dev->rbd_client->client->osdc;
1722 osd_req = ceph_osdc_alloc_request(osdc, snapc, 2, false, GFP_ATOMIC);
1724 return NULL; /* ENOMEM */
1726 osd_req->r_flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK;
1727 osd_req->r_callback = rbd_osd_req_callback;
1728 osd_req->r_priv = obj_request;
1730 osd_req->r_oid_len = strlen(obj_request->object_name);
1731 rbd_assert(osd_req->r_oid_len < sizeof (osd_req->r_oid));
1732 memcpy(osd_req->r_oid, obj_request->object_name, osd_req->r_oid_len);
1734 osd_req->r_file_layout = rbd_dev->layout; /* struct */
1740 static void rbd_osd_req_destroy(struct ceph_osd_request *osd_req)
1742 ceph_osdc_put_request(osd_req);
1745 /* object_name is assumed to be a non-null pointer and NUL-terminated */
1747 static struct rbd_obj_request *rbd_obj_request_create(const char *object_name,
1748 u64 offset, u64 length,
1749 enum obj_request_type type)
1751 struct rbd_obj_request *obj_request;
1755 rbd_assert(obj_request_type_valid(type));
1757 size = strlen(object_name) + 1;
1758 obj_request = kzalloc(sizeof (*obj_request) + size, GFP_KERNEL);
1762 name = (char *)(obj_request + 1);
1763 obj_request->object_name = memcpy(name, object_name, size);
1764 obj_request->offset = offset;
1765 obj_request->length = length;
1766 obj_request->flags = 0;
1767 obj_request->which = BAD_WHICH;
1768 obj_request->type = type;
1769 INIT_LIST_HEAD(&obj_request->links);
1770 init_completion(&obj_request->completion);
1771 kref_init(&obj_request->kref);
1773 dout("%s: \"%s\" %llu/%llu %d -> obj %p\n", __func__, object_name,
1774 offset, length, (int)type, obj_request);
1779 static void rbd_obj_request_destroy(struct kref *kref)
1781 struct rbd_obj_request *obj_request;
1783 obj_request = container_of(kref, struct rbd_obj_request, kref);
1785 dout("%s: obj %p\n", __func__, obj_request);
1787 rbd_assert(obj_request->img_request == NULL);
1788 rbd_assert(obj_request->which == BAD_WHICH);
1790 if (obj_request->osd_req)
1791 rbd_osd_req_destroy(obj_request->osd_req);
1793 rbd_assert(obj_request_type_valid(obj_request->type));
1794 switch (obj_request->type) {
1795 case OBJ_REQUEST_NODATA:
1796 break; /* Nothing to do */
1797 case OBJ_REQUEST_BIO:
1798 if (obj_request->bio_list)
1799 bio_chain_put(obj_request->bio_list);
1801 case OBJ_REQUEST_PAGES:
1802 if (obj_request->pages)
1803 ceph_release_page_vector(obj_request->pages,
1804 obj_request->page_count);
1812 * Caller is responsible for filling in the list of object requests
1813 * that comprises the image request, and the Linux request pointer
1814 * (if there is one).
1816 static struct rbd_img_request *rbd_img_request_create(
1817 struct rbd_device *rbd_dev,
1818 u64 offset, u64 length,
1822 struct rbd_img_request *img_request;
1824 img_request = kmalloc(sizeof (*img_request), GFP_ATOMIC);
1828 if (write_request) {
1829 down_read(&rbd_dev->header_rwsem);
1830 ceph_get_snap_context(rbd_dev->header.snapc);
1831 up_read(&rbd_dev->header_rwsem);
1834 img_request->rq = NULL;
1835 img_request->rbd_dev = rbd_dev;
1836 img_request->offset = offset;
1837 img_request->length = length;
1838 img_request->flags = 0;
1839 if (write_request) {
1840 img_request_write_set(img_request);
1841 img_request->snapc = rbd_dev->header.snapc;
1843 img_request->snap_id = rbd_dev->spec->snap_id;
1846 img_request_child_set(img_request);
1847 if (rbd_dev->parent_spec)
1848 img_request_layered_set(img_request);
1849 spin_lock_init(&img_request->completion_lock);
1850 img_request->next_completion = 0;
1851 img_request->callback = NULL;
1852 img_request->result = 0;
1853 img_request->obj_request_count = 0;
1854 INIT_LIST_HEAD(&img_request->obj_requests);
1855 kref_init(&img_request->kref);
1857 rbd_img_request_get(img_request); /* Avoid a warning */
1858 rbd_img_request_put(img_request); /* TEMPORARY */
1860 dout("%s: rbd_dev %p %s %llu/%llu -> img %p\n", __func__, rbd_dev,
1861 write_request ? "write" : "read", offset, length,
1867 static void rbd_img_request_destroy(struct kref *kref)
1869 struct rbd_img_request *img_request;
1870 struct rbd_obj_request *obj_request;
1871 struct rbd_obj_request *next_obj_request;
1873 img_request = container_of(kref, struct rbd_img_request, kref);
1875 dout("%s: img %p\n", __func__, img_request);
1877 for_each_obj_request_safe(img_request, obj_request, next_obj_request)
1878 rbd_img_obj_request_del(img_request, obj_request);
1879 rbd_assert(img_request->obj_request_count == 0);
1881 if (img_request_write_test(img_request))
1882 ceph_put_snap_context(img_request->snapc);
1884 if (img_request_child_test(img_request))
1885 rbd_obj_request_put(img_request->obj_request);
1890 static bool rbd_img_obj_end_request(struct rbd_obj_request *obj_request)
1892 struct rbd_img_request *img_request;
1893 unsigned int xferred;
1897 rbd_assert(obj_request_img_data_test(obj_request));
1898 img_request = obj_request->img_request;
1900 rbd_assert(obj_request->xferred <= (u64)UINT_MAX);
1901 xferred = (unsigned int)obj_request->xferred;
1902 result = obj_request->result;
1904 struct rbd_device *rbd_dev = img_request->rbd_dev;
1906 rbd_warn(rbd_dev, "%s %llx at %llx (%llx)\n",
1907 img_request_write_test(img_request) ? "write" : "read",
1908 obj_request->length, obj_request->img_offset,
1909 obj_request->offset);
1910 rbd_warn(rbd_dev, " result %d xferred %x\n",
1912 if (!img_request->result)
1913 img_request->result = result;
1916 /* Image object requests don't own their page array */
1918 if (obj_request->type == OBJ_REQUEST_PAGES) {
1919 obj_request->pages = NULL;
1920 obj_request->page_count = 0;
1923 if (img_request_child_test(img_request)) {
1924 rbd_assert(img_request->obj_request != NULL);
1925 more = obj_request->which < img_request->obj_request_count - 1;
1927 rbd_assert(img_request->rq != NULL);
1928 more = blk_end_request(img_request->rq, result, xferred);
1934 static void rbd_img_obj_callback(struct rbd_obj_request *obj_request)
1936 struct rbd_img_request *img_request;
1937 u32 which = obj_request->which;
1940 rbd_assert(obj_request_img_data_test(obj_request));
1941 img_request = obj_request->img_request;
1943 dout("%s: img %p obj %p\n", __func__, img_request, obj_request);
1944 rbd_assert(img_request != NULL);
1945 rbd_assert(img_request->obj_request_count > 0);
1946 rbd_assert(which != BAD_WHICH);
1947 rbd_assert(which < img_request->obj_request_count);
1948 rbd_assert(which >= img_request->next_completion);
1950 spin_lock_irq(&img_request->completion_lock);
1951 if (which != img_request->next_completion)
1954 for_each_obj_request_from(img_request, obj_request) {
1956 rbd_assert(which < img_request->obj_request_count);
1958 if (!obj_request_done_test(obj_request))
1960 more = rbd_img_obj_end_request(obj_request);
1964 rbd_assert(more ^ (which == img_request->obj_request_count));
1965 img_request->next_completion = which;
1967 spin_unlock_irq(&img_request->completion_lock);
1970 rbd_img_request_complete(img_request);
1974 * Split up an image request into one or more object requests, each
1975 * to a different object. The "type" parameter indicates whether
1976 * "data_desc" is the pointer to the head of a list of bio
1977 * structures, or the base of a page array. In either case this
1978 * function assumes data_desc describes memory sufficient to hold
1979 * all data described by the image request.
1981 static int rbd_img_request_fill(struct rbd_img_request *img_request,
1982 enum obj_request_type type,
1985 struct rbd_device *rbd_dev = img_request->rbd_dev;
1986 struct rbd_obj_request *obj_request = NULL;
1987 struct rbd_obj_request *next_obj_request;
1988 bool write_request = img_request_write_test(img_request);
1989 struct bio *bio_list;
1990 unsigned int bio_offset = 0;
1991 struct page **pages;
1996 dout("%s: img %p type %d data_desc %p\n", __func__, img_request,
1997 (int)type, data_desc);
1999 opcode = write_request ? CEPH_OSD_OP_WRITE : CEPH_OSD_OP_READ;
2000 img_offset = img_request->offset;
2001 resid = img_request->length;
2002 rbd_assert(resid > 0);
2004 if (type == OBJ_REQUEST_BIO) {
2005 bio_list = data_desc;
2006 rbd_assert(img_offset == bio_list->bi_sector << SECTOR_SHIFT);
2008 rbd_assert(type == OBJ_REQUEST_PAGES);
2013 struct ceph_osd_request *osd_req;
2014 const char *object_name;
2018 object_name = rbd_segment_name(rbd_dev, img_offset);
2021 offset = rbd_segment_offset(rbd_dev, img_offset);
2022 length = rbd_segment_length(rbd_dev, img_offset, resid);
2023 obj_request = rbd_obj_request_create(object_name,
2024 offset, length, type);
2025 kfree(object_name); /* object request has its own copy */
2029 if (type == OBJ_REQUEST_BIO) {
2030 unsigned int clone_size;
2032 rbd_assert(length <= (u64)UINT_MAX);
2033 clone_size = (unsigned int)length;
2034 obj_request->bio_list =
2035 bio_chain_clone_range(&bio_list,
2039 if (!obj_request->bio_list)
2042 unsigned int page_count;
2044 obj_request->pages = pages;
2045 page_count = (u32)calc_pages_for(offset, length);
2046 obj_request->page_count = page_count;
2047 if ((offset + length) & ~PAGE_MASK)
2048 page_count--; /* more on last page */
2049 pages += page_count;
2052 osd_req = rbd_osd_req_create(rbd_dev, write_request,
2056 obj_request->osd_req = osd_req;
2057 obj_request->callback = rbd_img_obj_callback;
2059 osd_req_op_extent_init(osd_req, 0, opcode, offset, length,
2061 if (type == OBJ_REQUEST_BIO)
2062 osd_req_op_extent_osd_data_bio(osd_req, 0,
2063 obj_request->bio_list, length);
2065 osd_req_op_extent_osd_data_pages(osd_req, 0,
2066 obj_request->pages, length,
2067 offset & ~PAGE_MASK, false, false);
2070 rbd_osd_req_format_write(obj_request);
2072 rbd_osd_req_format_read(obj_request);
2074 obj_request->img_offset = img_offset;
2075 rbd_img_obj_request_add(img_request, obj_request);
2077 img_offset += length;
2084 rbd_obj_request_put(obj_request);
2086 for_each_obj_request_safe(img_request, obj_request, next_obj_request)
2087 rbd_obj_request_put(obj_request);
2093 rbd_img_obj_copyup_callback(struct rbd_obj_request *obj_request)
2095 struct rbd_img_request *img_request;
2096 struct rbd_device *rbd_dev;
2100 rbd_assert(obj_request->type == OBJ_REQUEST_BIO);
2101 rbd_assert(obj_request_img_data_test(obj_request));
2102 img_request = obj_request->img_request;
2103 rbd_assert(img_request);
2105 rbd_dev = img_request->rbd_dev;
2106 rbd_assert(rbd_dev);
2107 length = (u64)1 << rbd_dev->header.obj_order;
2108 page_count = (u32)calc_pages_for(0, length);
2110 rbd_assert(obj_request->copyup_pages);
2111 ceph_release_page_vector(obj_request->copyup_pages, page_count);
2112 obj_request->copyup_pages = NULL;
2115 * We want the transfer count to reflect the size of the
2116 * original write request. There is no such thing as a
2117 * successful short write, so if the request was successful
2118 * we can just set it to the originally-requested length.
2120 if (!obj_request->result)
2121 obj_request->xferred = obj_request->length;
2123 /* Finish up with the normal image object callback */
2125 rbd_img_obj_callback(obj_request);
2129 rbd_img_obj_parent_read_full_callback(struct rbd_img_request *img_request)
2131 struct rbd_obj_request *orig_request;
2132 struct ceph_osd_request *osd_req;
2133 struct ceph_osd_client *osdc;
2134 struct rbd_device *rbd_dev;
2135 struct page **pages;
2140 rbd_assert(img_request_child_test(img_request));
2142 /* First get what we need from the image request */
2144 pages = img_request->copyup_pages;
2145 rbd_assert(pages != NULL);
2146 img_request->copyup_pages = NULL;
2148 orig_request = img_request->obj_request;
2149 rbd_assert(orig_request != NULL);
2150 rbd_assert(orig_request->type == OBJ_REQUEST_BIO);
2151 result = img_request->result;
2152 obj_size = img_request->length;
2153 xferred = img_request->xferred;
2155 rbd_dev = img_request->rbd_dev;
2156 rbd_assert(rbd_dev);
2157 rbd_assert(obj_size == (u64)1 << rbd_dev->header.obj_order);
2159 rbd_img_request_put(img_request);
2164 /* Allocate the new copyup osd request for the original request */
2167 rbd_assert(!orig_request->osd_req);
2168 osd_req = rbd_osd_req_create_copyup(orig_request);
2171 orig_request->osd_req = osd_req;
2172 orig_request->copyup_pages = pages;
2174 /* Initialize the copyup op */
2176 osd_req_op_cls_init(osd_req, 0, CEPH_OSD_OP_CALL, "rbd", "copyup");
2177 osd_req_op_cls_request_data_pages(osd_req, 0, pages, obj_size, 0,
2180 /* Then the original write request op */
2182 osd_req_op_extent_init(osd_req, 1, CEPH_OSD_OP_WRITE,
2183 orig_request->offset,
2184 orig_request->length, 0, 0);
2185 osd_req_op_extent_osd_data_bio(osd_req, 1, orig_request->bio_list,
2186 orig_request->length);
2188 rbd_osd_req_format_write(orig_request);
2190 /* All set, send it off. */
2192 orig_request->callback = rbd_img_obj_copyup_callback;
2193 osdc = &rbd_dev->rbd_client->client->osdc;
2194 result = rbd_obj_request_submit(osdc, orig_request);
2198 /* Record the error code and complete the request */
2200 orig_request->result = result;
2201 orig_request->xferred = 0;
2202 obj_request_done_set(orig_request);
2203 rbd_obj_request_complete(orig_request);
2207 * Read from the parent image the range of data that covers the
2208 * entire target of the given object request. This is used for
2209 * satisfying a layered image write request when the target of an
2210 * object request from the image request does not exist.
2212 * A page array big enough to hold the returned data is allocated
2213 * and supplied to rbd_img_request_fill() as the "data descriptor."
2214 * When the read completes, this page array will be transferred to
2215 * the original object request for the copyup operation.
2217 * If an error occurs, record it as the result of the original
2218 * object request and mark it done so it gets completed.
2220 static int rbd_img_obj_parent_read_full(struct rbd_obj_request *obj_request)
2222 struct rbd_img_request *img_request = NULL;
2223 struct rbd_img_request *parent_request = NULL;
2224 struct rbd_device *rbd_dev;
2227 struct page **pages = NULL;
2231 rbd_assert(obj_request_img_data_test(obj_request));
2232 rbd_assert(obj_request->type == OBJ_REQUEST_BIO);
2234 img_request = obj_request->img_request;
2235 rbd_assert(img_request != NULL);
2236 rbd_dev = img_request->rbd_dev;
2237 rbd_assert(rbd_dev->parent != NULL);
2240 * First things first. The original osd request is of no
2241 * use to use any more, we'll need a new one that can hold
2242 * the two ops in a copyup request. We'll get that later,
2243 * but for now we can release the old one.
2245 rbd_osd_req_destroy(obj_request->osd_req);
2246 obj_request->osd_req = NULL;
2249 * Determine the byte range covered by the object in the
2250 * child image to which the original request was to be sent.
2252 img_offset = obj_request->img_offset - obj_request->offset;
2253 length = (u64)1 << rbd_dev->header.obj_order;
2256 * There is no defined parent data beyond the parent
2257 * overlap, so limit what we read at that boundary if
2260 if (img_offset + length > rbd_dev->parent_overlap) {
2261 rbd_assert(img_offset < rbd_dev->parent_overlap);
2262 length = rbd_dev->parent_overlap - img_offset;
2266 * Allocate a page array big enough to receive the data read
2269 page_count = (u32)calc_pages_for(0, length);
2270 pages = ceph_alloc_page_vector(page_count, GFP_KERNEL);
2271 if (IS_ERR(pages)) {
2272 result = PTR_ERR(pages);
2278 parent_request = rbd_img_request_create(rbd_dev->parent,
2281 if (!parent_request)
2283 rbd_obj_request_get(obj_request);
2284 parent_request->obj_request = obj_request;
2286 result = rbd_img_request_fill(parent_request, OBJ_REQUEST_PAGES, pages);
2289 parent_request->copyup_pages = pages;
2291 parent_request->callback = rbd_img_obj_parent_read_full_callback;
2292 result = rbd_img_request_submit(parent_request);
2296 parent_request->copyup_pages = NULL;
2297 parent_request->obj_request = NULL;
2298 rbd_obj_request_put(obj_request);
2301 ceph_release_page_vector(pages, page_count);
2303 rbd_img_request_put(parent_request);
2304 obj_request->result = result;
2305 obj_request->xferred = 0;
2306 obj_request_done_set(obj_request);
2311 static void rbd_img_obj_exists_callback(struct rbd_obj_request *obj_request)
2313 struct rbd_obj_request *orig_request;
2316 rbd_assert(!obj_request_img_data_test(obj_request));
2319 * All we need from the object request is the original
2320 * request and the result of the STAT op. Grab those, then
2321 * we're done with the request.
2323 orig_request = obj_request->obj_request;
2324 obj_request->obj_request = NULL;
2325 rbd_assert(orig_request);
2326 rbd_assert(orig_request->img_request);
2328 result = obj_request->result;
2329 obj_request->result = 0;
2331 dout("%s: obj %p for obj %p result %d %llu/%llu\n", __func__,
2332 obj_request, orig_request, result,
2333 obj_request->xferred, obj_request->length);
2334 rbd_obj_request_put(obj_request);
2336 rbd_assert(orig_request);
2337 rbd_assert(orig_request->img_request);
2340 * Our only purpose here is to determine whether the object
2341 * exists, and we don't want to treat the non-existence as
2342 * an error. If something else comes back, transfer the
2343 * error to the original request and complete it now.
2346 obj_request_existence_set(orig_request, true);
2347 } else if (result == -ENOENT) {
2348 obj_request_existence_set(orig_request, false);
2349 } else if (result) {
2350 orig_request->result = result;
2355 * Resubmit the original request now that we have recorded
2356 * whether the target object exists.
2358 orig_request->result = rbd_img_obj_request_submit(orig_request);
2360 if (orig_request->result)
2361 rbd_obj_request_complete(orig_request);
2362 rbd_obj_request_put(orig_request);
2365 static int rbd_img_obj_exists_submit(struct rbd_obj_request *obj_request)
2367 struct rbd_obj_request *stat_request;
2368 struct rbd_device *rbd_dev;
2369 struct ceph_osd_client *osdc;
2370 struct page **pages = NULL;
2376 * The response data for a STAT call consists of:
2383 size = sizeof (__le64) + sizeof (__le32) + sizeof (__le32);
2384 page_count = (u32)calc_pages_for(0, size);
2385 pages = ceph_alloc_page_vector(page_count, GFP_KERNEL);
2387 return PTR_ERR(pages);
2390 stat_request = rbd_obj_request_create(obj_request->object_name, 0, 0,
2395 rbd_obj_request_get(obj_request);
2396 stat_request->obj_request = obj_request;
2397 stat_request->pages = pages;
2398 stat_request->page_count = page_count;
2400 rbd_assert(obj_request->img_request);
2401 rbd_dev = obj_request->img_request->rbd_dev;
2402 stat_request->osd_req = rbd_osd_req_create(rbd_dev, false,
2404 if (!stat_request->osd_req)
2406 stat_request->callback = rbd_img_obj_exists_callback;
2408 osd_req_op_init(stat_request->osd_req, 0, CEPH_OSD_OP_STAT);
2409 osd_req_op_raw_data_in_pages(stat_request->osd_req, 0, pages, size, 0,
2411 rbd_osd_req_format_read(stat_request);
2413 osdc = &rbd_dev->rbd_client->client->osdc;
2414 ret = rbd_obj_request_submit(osdc, stat_request);
2417 rbd_obj_request_put(obj_request);
2422 static int rbd_img_obj_request_submit(struct rbd_obj_request *obj_request)
2424 struct rbd_img_request *img_request;
2425 struct rbd_device *rbd_dev;
2428 rbd_assert(obj_request_img_data_test(obj_request));
2430 img_request = obj_request->img_request;
2431 rbd_assert(img_request);
2432 rbd_dev = img_request->rbd_dev;
2435 * Only writes to layered images need special handling.
2436 * Reads and non-layered writes are simple object requests.
2437 * Layered writes that start beyond the end of the overlap
2438 * with the parent have no parent data, so they too are
2439 * simple object requests. Finally, if the target object is
2440 * known to already exist, its parent data has already been
2441 * copied, so a write to the object can also be handled as a
2442 * simple object request.
2444 if (!img_request_write_test(img_request) ||
2445 !img_request_layered_test(img_request) ||
2446 rbd_dev->parent_overlap <= obj_request->img_offset ||
2447 ((known = obj_request_known_test(obj_request)) &&
2448 obj_request_exists_test(obj_request))) {
2450 struct rbd_device *rbd_dev;
2451 struct ceph_osd_client *osdc;
2453 rbd_dev = obj_request->img_request->rbd_dev;
2454 osdc = &rbd_dev->rbd_client->client->osdc;
2456 return rbd_obj_request_submit(osdc, obj_request);
2460 * It's a layered write. The target object might exist but
2461 * we may not know that yet. If we know it doesn't exist,
2462 * start by reading the data for the full target object from
2463 * the parent so we can use it for a copyup to the target.
2466 return rbd_img_obj_parent_read_full(obj_request);
2468 /* We don't know whether the target exists. Go find out. */
2470 return rbd_img_obj_exists_submit(obj_request);
2473 static int rbd_img_request_submit(struct rbd_img_request *img_request)
2475 struct rbd_obj_request *obj_request;
2476 struct rbd_obj_request *next_obj_request;
2478 dout("%s: img %p\n", __func__, img_request);
2479 for_each_obj_request_safe(img_request, obj_request, next_obj_request) {
2482 ret = rbd_img_obj_request_submit(obj_request);
2490 static void rbd_img_parent_read_callback(struct rbd_img_request *img_request)
2492 struct rbd_obj_request *obj_request;
2493 struct rbd_device *rbd_dev;
2496 rbd_assert(img_request_child_test(img_request));
2498 obj_request = img_request->obj_request;
2499 rbd_assert(obj_request);
2500 rbd_assert(obj_request->img_request);
2502 obj_request->result = img_request->result;
2503 if (obj_request->result)
2507 * We need to zero anything beyond the parent overlap
2508 * boundary. Since rbd_img_obj_request_read_callback()
2509 * will zero anything beyond the end of a short read, an
2510 * easy way to do this is to pretend the data from the
2511 * parent came up short--ending at the overlap boundary.
2513 rbd_assert(obj_request->img_offset < U64_MAX - obj_request->length);
2514 obj_end = obj_request->img_offset + obj_request->length;
2515 rbd_dev = obj_request->img_request->rbd_dev;
2516 if (obj_end > rbd_dev->parent_overlap) {
2519 if (obj_request->img_offset < rbd_dev->parent_overlap)
2520 xferred = rbd_dev->parent_overlap -
2521 obj_request->img_offset;
2523 obj_request->xferred = min(img_request->xferred, xferred);
2525 obj_request->xferred = img_request->xferred;
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;
2612 dout("%s: \"%s\" notify_id %llu opcode %u\n", __func__,
2613 rbd_dev->header_name, (unsigned long long)notify_id,
2614 (unsigned int)opcode);
2615 (void)rbd_dev_refresh(rbd_dev);
2617 rbd_obj_notify_ack(rbd_dev, notify_id);
2621 * Request sync osd watch/unwatch. The value of "start" determines
2622 * whether a watch request is being initiated or torn down.
2624 static int rbd_dev_header_watch_sync(struct rbd_device *rbd_dev, int start)
2626 struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
2627 struct rbd_obj_request *obj_request;
2630 rbd_assert(start ^ !!rbd_dev->watch_event);
2631 rbd_assert(start ^ !!rbd_dev->watch_request);
2634 ret = ceph_osdc_create_event(osdc, rbd_watch_cb, rbd_dev,
2635 &rbd_dev->watch_event);
2638 rbd_assert(rbd_dev->watch_event != NULL);
2642 obj_request = rbd_obj_request_create(rbd_dev->header_name, 0, 0,
2643 OBJ_REQUEST_NODATA);
2647 obj_request->osd_req = rbd_osd_req_create(rbd_dev, true, obj_request);
2648 if (!obj_request->osd_req)
2652 ceph_osdc_set_request_linger(osdc, obj_request->osd_req);
2654 ceph_osdc_unregister_linger_request(osdc,
2655 rbd_dev->watch_request->osd_req);
2657 osd_req_op_watch_init(obj_request->osd_req, 0, CEPH_OSD_OP_WATCH,
2658 rbd_dev->watch_event->cookie, 0, start);
2659 rbd_osd_req_format_write(obj_request);
2661 ret = rbd_obj_request_submit(osdc, obj_request);
2664 ret = rbd_obj_request_wait(obj_request);
2667 ret = obj_request->result;
2672 * A watch request is set to linger, so the underlying osd
2673 * request won't go away until we unregister it. We retain
2674 * a pointer to the object request during that time (in
2675 * rbd_dev->watch_request), so we'll keep a reference to
2676 * it. We'll drop that reference (below) after we've
2680 rbd_dev->watch_request = obj_request;
2685 /* We have successfully torn down the watch request */
2687 rbd_obj_request_put(rbd_dev->watch_request);
2688 rbd_dev->watch_request = NULL;
2690 /* Cancel the event if we're tearing down, or on error */
2691 ceph_osdc_cancel_event(rbd_dev->watch_event);
2692 rbd_dev->watch_event = NULL;
2694 rbd_obj_request_put(obj_request);
2700 * Synchronous osd object method call. Returns the number of bytes
2701 * returned in the outbound buffer, or a negative error code.
2703 static int rbd_obj_method_sync(struct rbd_device *rbd_dev,
2704 const char *object_name,
2705 const char *class_name,
2706 const char *method_name,
2707 const void *outbound,
2708 size_t outbound_size,
2710 size_t inbound_size)
2712 struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
2713 struct rbd_obj_request *obj_request;
2714 struct page **pages;
2719 * Method calls are ultimately read operations. The result
2720 * should placed into the inbound buffer provided. They
2721 * also supply outbound data--parameters for the object
2722 * method. Currently if this is present it will be a
2725 page_count = (u32)calc_pages_for(0, inbound_size);
2726 pages = ceph_alloc_page_vector(page_count, GFP_KERNEL);
2728 return PTR_ERR(pages);
2731 obj_request = rbd_obj_request_create(object_name, 0, inbound_size,
2736 obj_request->pages = pages;
2737 obj_request->page_count = page_count;
2739 obj_request->osd_req = rbd_osd_req_create(rbd_dev, false, obj_request);
2740 if (!obj_request->osd_req)
2743 osd_req_op_cls_init(obj_request->osd_req, 0, CEPH_OSD_OP_CALL,
2744 class_name, method_name);
2745 if (outbound_size) {
2746 struct ceph_pagelist *pagelist;
2748 pagelist = kmalloc(sizeof (*pagelist), GFP_NOFS);
2752 ceph_pagelist_init(pagelist);
2753 ceph_pagelist_append(pagelist, outbound, outbound_size);
2754 osd_req_op_cls_request_data_pagelist(obj_request->osd_req, 0,
2757 osd_req_op_cls_response_data_pages(obj_request->osd_req, 0,
2758 obj_request->pages, inbound_size,
2760 rbd_osd_req_format_read(obj_request);
2762 ret = rbd_obj_request_submit(osdc, obj_request);
2765 ret = rbd_obj_request_wait(obj_request);
2769 ret = obj_request->result;
2773 rbd_assert(obj_request->xferred < (u64)INT_MAX);
2774 ret = (int)obj_request->xferred;
2775 ceph_copy_from_page_vector(pages, inbound, 0, obj_request->xferred);
2778 rbd_obj_request_put(obj_request);
2780 ceph_release_page_vector(pages, page_count);
2785 static void rbd_request_fn(struct request_queue *q)
2786 __releases(q->queue_lock) __acquires(q->queue_lock)
2788 struct rbd_device *rbd_dev = q->queuedata;
2789 bool read_only = rbd_dev->mapping.read_only;
2793 while ((rq = blk_fetch_request(q))) {
2794 bool write_request = rq_data_dir(rq) == WRITE;
2795 struct rbd_img_request *img_request;
2799 /* Ignore any non-FS requests that filter through. */
2801 if (rq->cmd_type != REQ_TYPE_FS) {
2802 dout("%s: non-fs request type %d\n", __func__,
2803 (int) rq->cmd_type);
2804 __blk_end_request_all(rq, 0);
2808 /* Ignore/skip any zero-length requests */
2810 offset = (u64) blk_rq_pos(rq) << SECTOR_SHIFT;
2811 length = (u64) blk_rq_bytes(rq);
2814 dout("%s: zero-length request\n", __func__);
2815 __blk_end_request_all(rq, 0);
2819 spin_unlock_irq(q->queue_lock);
2821 /* Disallow writes to a read-only device */
2823 if (write_request) {
2827 rbd_assert(rbd_dev->spec->snap_id == CEPH_NOSNAP);
2831 * Quit early if the mapped snapshot no longer
2832 * exists. It's still possible the snapshot will
2833 * have disappeared by the time our request arrives
2834 * at the osd, but there's no sense in sending it if
2837 if (!test_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags)) {
2838 dout("request for non-existent snapshot");
2839 rbd_assert(rbd_dev->spec->snap_id != CEPH_NOSNAP);
2845 if (offset && length > U64_MAX - offset + 1) {
2846 rbd_warn(rbd_dev, "bad request range (%llu~%llu)\n",
2848 goto end_request; /* Shouldn't happen */
2852 img_request = rbd_img_request_create(rbd_dev, offset, length,
2853 write_request, false);
2857 img_request->rq = rq;
2859 result = rbd_img_request_fill(img_request, OBJ_REQUEST_BIO,
2862 result = rbd_img_request_submit(img_request);
2864 rbd_img_request_put(img_request);
2866 spin_lock_irq(q->queue_lock);
2868 rbd_warn(rbd_dev, "%s %llx at %llx result %d\n",
2869 write_request ? "write" : "read",
2870 length, offset, result);
2872 __blk_end_request_all(rq, result);
2878 * a queue callback. Makes sure that we don't create a bio that spans across
2879 * multiple osd objects. One exception would be with a single page bios,
2880 * which we handle later at bio_chain_clone_range()
2882 static int rbd_merge_bvec(struct request_queue *q, struct bvec_merge_data *bmd,
2883 struct bio_vec *bvec)
2885 struct rbd_device *rbd_dev = q->queuedata;
2886 sector_t sector_offset;
2887 sector_t sectors_per_obj;
2888 sector_t obj_sector_offset;
2892 * Find how far into its rbd object the partition-relative
2893 * bio start sector is to offset relative to the enclosing
2896 sector_offset = get_start_sect(bmd->bi_bdev) + bmd->bi_sector;
2897 sectors_per_obj = 1 << (rbd_dev->header.obj_order - SECTOR_SHIFT);
2898 obj_sector_offset = sector_offset & (sectors_per_obj - 1);
2901 * Compute the number of bytes from that offset to the end
2902 * of the object. Account for what's already used by the bio.
2904 ret = (int) (sectors_per_obj - obj_sector_offset) << SECTOR_SHIFT;
2905 if (ret > bmd->bi_size)
2906 ret -= bmd->bi_size;
2911 * Don't send back more than was asked for. And if the bio
2912 * was empty, let the whole thing through because: "Note
2913 * that a block device *must* allow a single page to be
2914 * added to an empty bio."
2916 rbd_assert(bvec->bv_len <= PAGE_SIZE);
2917 if (ret > (int) bvec->bv_len || !bmd->bi_size)
2918 ret = (int) bvec->bv_len;
2923 static void rbd_free_disk(struct rbd_device *rbd_dev)
2925 struct gendisk *disk = rbd_dev->disk;
2930 rbd_dev->disk = NULL;
2931 if (disk->flags & GENHD_FL_UP) {
2934 blk_cleanup_queue(disk->queue);
2939 static int rbd_obj_read_sync(struct rbd_device *rbd_dev,
2940 const char *object_name,
2941 u64 offset, u64 length, void *buf)
2944 struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
2945 struct rbd_obj_request *obj_request;
2946 struct page **pages = NULL;
2951 page_count = (u32) calc_pages_for(offset, length);
2952 pages = ceph_alloc_page_vector(page_count, GFP_KERNEL);
2954 ret = PTR_ERR(pages);
2957 obj_request = rbd_obj_request_create(object_name, offset, length,
2962 obj_request->pages = pages;
2963 obj_request->page_count = page_count;
2965 obj_request->osd_req = rbd_osd_req_create(rbd_dev, false, obj_request);
2966 if (!obj_request->osd_req)
2969 osd_req_op_extent_init(obj_request->osd_req, 0, CEPH_OSD_OP_READ,
2970 offset, length, 0, 0);
2971 osd_req_op_extent_osd_data_pages(obj_request->osd_req, 0,
2973 obj_request->length,
2974 obj_request->offset & ~PAGE_MASK,
2976 rbd_osd_req_format_read(obj_request);
2978 ret = rbd_obj_request_submit(osdc, obj_request);
2981 ret = rbd_obj_request_wait(obj_request);
2985 ret = obj_request->result;
2989 rbd_assert(obj_request->xferred <= (u64) SIZE_MAX);
2990 size = (size_t) obj_request->xferred;
2991 ceph_copy_from_page_vector(pages, buf, 0, size);
2992 rbd_assert(size <= (size_t)INT_MAX);
2996 rbd_obj_request_put(obj_request);
2998 ceph_release_page_vector(pages, page_count);
3004 * Read the complete header for the given rbd device.
3006 * Returns a pointer to a dynamically-allocated buffer containing
3007 * the complete and validated header. Caller can pass the address
3008 * of a variable that will be filled in with the version of the
3009 * header object at the time it was read.
3011 * Returns a pointer-coded errno if a failure occurs.
3013 static struct rbd_image_header_ondisk *
3014 rbd_dev_v1_header_read(struct rbd_device *rbd_dev)
3016 struct rbd_image_header_ondisk *ondisk = NULL;
3023 * The complete header will include an array of its 64-bit
3024 * snapshot ids, followed by the names of those snapshots as
3025 * a contiguous block of NUL-terminated strings. Note that
3026 * the number of snapshots could change by the time we read
3027 * it in, in which case we re-read it.
3034 size = sizeof (*ondisk);
3035 size += snap_count * sizeof (struct rbd_image_snap_ondisk);
3037 ondisk = kmalloc(size, GFP_KERNEL);
3039 return ERR_PTR(-ENOMEM);
3041 ret = rbd_obj_read_sync(rbd_dev, rbd_dev->header_name,
3045 if ((size_t)ret < size) {
3047 rbd_warn(rbd_dev, "short header read (want %zd got %d)",
3051 if (!rbd_dev_ondisk_valid(ondisk)) {
3053 rbd_warn(rbd_dev, "invalid header");
3057 names_size = le64_to_cpu(ondisk->snap_names_len);
3058 want_count = snap_count;
3059 snap_count = le32_to_cpu(ondisk->snap_count);
3060 } while (snap_count != want_count);
3067 return ERR_PTR(ret);
3071 * reload the ondisk the header
3073 static int rbd_read_header(struct rbd_device *rbd_dev,
3074 struct rbd_image_header *header)
3076 struct rbd_image_header_ondisk *ondisk;
3079 ondisk = rbd_dev_v1_header_read(rbd_dev);
3081 return PTR_ERR(ondisk);
3082 ret = rbd_header_from_disk(header, ondisk);
3088 static void rbd_update_mapping_size(struct rbd_device *rbd_dev)
3090 if (rbd_dev->spec->snap_id != CEPH_NOSNAP)
3093 if (rbd_dev->mapping.size != rbd_dev->header.image_size) {
3096 rbd_dev->mapping.size = rbd_dev->header.image_size;
3097 size = (sector_t)rbd_dev->mapping.size / SECTOR_SIZE;
3098 dout("setting size to %llu sectors", (unsigned long long)size);
3099 set_capacity(rbd_dev->disk, size);
3104 * only read the first part of the ondisk header, without the snaps info
3106 static int rbd_dev_v1_refresh(struct rbd_device *rbd_dev)
3109 struct rbd_image_header h;
3111 ret = rbd_read_header(rbd_dev, &h);
3115 down_write(&rbd_dev->header_rwsem);
3117 /* Update image size, and check for resize of mapped image */
3118 rbd_dev->header.image_size = h.image_size;
3119 rbd_update_mapping_size(rbd_dev);
3121 /* rbd_dev->header.object_prefix shouldn't change */
3122 kfree(rbd_dev->header.snap_sizes);
3123 kfree(rbd_dev->header.snap_names);
3124 /* osd requests may still refer to snapc */
3125 ceph_put_snap_context(rbd_dev->header.snapc);
3127 rbd_dev->header.image_size = h.image_size;
3128 rbd_dev->header.snapc = h.snapc;
3129 rbd_dev->header.snap_names = h.snap_names;
3130 rbd_dev->header.snap_sizes = h.snap_sizes;
3131 /* Free the extra copy of the object prefix */
3132 if (strcmp(rbd_dev->header.object_prefix, h.object_prefix))
3133 rbd_warn(rbd_dev, "object prefix changed (ignoring)");
3134 kfree(h.object_prefix);
3136 up_write(&rbd_dev->header_rwsem);
3142 * Clear the rbd device's EXISTS flag if the snapshot it's mapped to
3143 * has disappeared from the (just updated) snapshot context.
3145 static void rbd_exists_validate(struct rbd_device *rbd_dev)
3149 if (!test_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags))
3152 snap_id = rbd_dev->spec->snap_id;
3153 if (snap_id == CEPH_NOSNAP)
3156 if (rbd_dev_snap_index(rbd_dev, snap_id) == BAD_SNAP_INDEX)
3157 clear_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags);
3160 static int rbd_dev_refresh(struct rbd_device *rbd_dev)
3165 rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
3166 image_size = rbd_dev->header.image_size;
3167 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
3168 if (rbd_dev->image_format == 1)
3169 ret = rbd_dev_v1_refresh(rbd_dev);
3171 ret = rbd_dev_v2_refresh(rbd_dev);
3173 /* If it's a mapped snapshot, validate its EXISTS flag */
3175 rbd_exists_validate(rbd_dev);
3176 mutex_unlock(&ctl_mutex);
3178 rbd_warn(rbd_dev, "got notification but failed to "
3179 " update snaps: %d\n", ret);
3180 if (image_size != rbd_dev->header.image_size)
3181 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 return ret < 0 ? ret : size;
3389 static DEVICE_ATTR(size, S_IRUGO, rbd_size_show, NULL);
3390 static DEVICE_ATTR(features, S_IRUGO, rbd_features_show, NULL);
3391 static DEVICE_ATTR(major, S_IRUGO, rbd_major_show, NULL);
3392 static DEVICE_ATTR(client_id, S_IRUGO, rbd_client_id_show, NULL);
3393 static DEVICE_ATTR(pool, S_IRUGO, rbd_pool_show, NULL);
3394 static DEVICE_ATTR(pool_id, S_IRUGO, rbd_pool_id_show, NULL);
3395 static DEVICE_ATTR(name, S_IRUGO, rbd_name_show, NULL);
3396 static DEVICE_ATTR(image_id, S_IRUGO, rbd_image_id_show, NULL);
3397 static DEVICE_ATTR(refresh, S_IWUSR, NULL, rbd_image_refresh);
3398 static DEVICE_ATTR(current_snap, S_IRUGO, rbd_snap_show, NULL);
3399 static DEVICE_ATTR(parent, S_IRUGO, rbd_parent_show, NULL);
3401 static struct attribute *rbd_attrs[] = {
3402 &dev_attr_size.attr,
3403 &dev_attr_features.attr,
3404 &dev_attr_major.attr,
3405 &dev_attr_client_id.attr,
3406 &dev_attr_pool.attr,
3407 &dev_attr_pool_id.attr,
3408 &dev_attr_name.attr,
3409 &dev_attr_image_id.attr,
3410 &dev_attr_current_snap.attr,
3411 &dev_attr_parent.attr,
3412 &dev_attr_refresh.attr,
3416 static struct attribute_group rbd_attr_group = {
3420 static const struct attribute_group *rbd_attr_groups[] = {
3425 static void rbd_sysfs_dev_release(struct device *dev)
3429 static struct device_type rbd_device_type = {
3431 .groups = rbd_attr_groups,
3432 .release = rbd_sysfs_dev_release,
3435 static struct rbd_spec *rbd_spec_get(struct rbd_spec *spec)
3437 kref_get(&spec->kref);
3442 static void rbd_spec_free(struct kref *kref);
3443 static void rbd_spec_put(struct rbd_spec *spec)
3446 kref_put(&spec->kref, rbd_spec_free);
3449 static struct rbd_spec *rbd_spec_alloc(void)
3451 struct rbd_spec *spec;
3453 spec = kzalloc(sizeof (*spec), GFP_KERNEL);
3456 kref_init(&spec->kref);
3461 static void rbd_spec_free(struct kref *kref)
3463 struct rbd_spec *spec = container_of(kref, struct rbd_spec, kref);
3465 kfree(spec->pool_name);
3466 kfree(spec->image_id);
3467 kfree(spec->image_name);
3468 kfree(spec->snap_name);
3472 static struct rbd_device *rbd_dev_create(struct rbd_client *rbdc,
3473 struct rbd_spec *spec)
3475 struct rbd_device *rbd_dev;
3477 rbd_dev = kzalloc(sizeof (*rbd_dev), GFP_KERNEL);
3481 spin_lock_init(&rbd_dev->lock);
3483 INIT_LIST_HEAD(&rbd_dev->node);
3484 init_rwsem(&rbd_dev->header_rwsem);
3486 rbd_dev->spec = spec;
3487 rbd_dev->rbd_client = rbdc;
3489 /* Initialize the layout used for all rbd requests */
3491 rbd_dev->layout.fl_stripe_unit = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER);
3492 rbd_dev->layout.fl_stripe_count = cpu_to_le32(1);
3493 rbd_dev->layout.fl_object_size = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER);
3494 rbd_dev->layout.fl_pg_pool = cpu_to_le32((u32) spec->pool_id);
3499 static void rbd_dev_destroy(struct rbd_device *rbd_dev)
3501 rbd_put_client(rbd_dev->rbd_client);
3502 rbd_spec_put(rbd_dev->spec);
3507 * Get the size and object order for an image snapshot, or if
3508 * snap_id is CEPH_NOSNAP, gets this information for the base
3511 static int _rbd_dev_v2_snap_size(struct rbd_device *rbd_dev, u64 snap_id,
3512 u8 *order, u64 *snap_size)
3514 __le64 snapid = cpu_to_le64(snap_id);
3519 } __attribute__ ((packed)) size_buf = { 0 };
3521 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
3523 &snapid, sizeof (snapid),
3524 &size_buf, sizeof (size_buf));
3525 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
3528 if (ret < sizeof (size_buf))
3532 *order = size_buf.order;
3533 *snap_size = le64_to_cpu(size_buf.size);
3535 dout(" snap_id 0x%016llx order = %u, snap_size = %llu\n",
3536 (unsigned long long)snap_id, (unsigned int)*order,
3537 (unsigned long long)*snap_size);
3542 static int rbd_dev_v2_image_size(struct rbd_device *rbd_dev)
3544 return _rbd_dev_v2_snap_size(rbd_dev, CEPH_NOSNAP,
3545 &rbd_dev->header.obj_order,
3546 &rbd_dev->header.image_size);
3549 static int rbd_dev_v2_object_prefix(struct rbd_device *rbd_dev)
3555 reply_buf = kzalloc(RBD_OBJ_PREFIX_LEN_MAX, GFP_KERNEL);
3559 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
3560 "rbd", "get_object_prefix", NULL, 0,
3561 reply_buf, RBD_OBJ_PREFIX_LEN_MAX);
3562 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
3567 rbd_dev->header.object_prefix = ceph_extract_encoded_string(&p,
3568 p + ret, NULL, GFP_NOIO);
3571 if (IS_ERR(rbd_dev->header.object_prefix)) {
3572 ret = PTR_ERR(rbd_dev->header.object_prefix);
3573 rbd_dev->header.object_prefix = NULL;
3575 dout(" object_prefix = %s\n", rbd_dev->header.object_prefix);
3583 static int _rbd_dev_v2_snap_features(struct rbd_device *rbd_dev, u64 snap_id,
3586 __le64 snapid = cpu_to_le64(snap_id);
3590 } __attribute__ ((packed)) features_buf = { 0 };
3594 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
3595 "rbd", "get_features",
3596 &snapid, sizeof (snapid),
3597 &features_buf, sizeof (features_buf));
3598 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
3601 if (ret < sizeof (features_buf))
3604 incompat = le64_to_cpu(features_buf.incompat);
3605 if (incompat & ~RBD_FEATURES_SUPPORTED)
3608 *snap_features = le64_to_cpu(features_buf.features);
3610 dout(" snap_id 0x%016llx features = 0x%016llx incompat = 0x%016llx\n",
3611 (unsigned long long)snap_id,
3612 (unsigned long long)*snap_features,
3613 (unsigned long long)le64_to_cpu(features_buf.incompat));
3618 static int rbd_dev_v2_features(struct rbd_device *rbd_dev)
3620 return _rbd_dev_v2_snap_features(rbd_dev, CEPH_NOSNAP,
3621 &rbd_dev->header.features);
3624 static int rbd_dev_v2_parent_info(struct rbd_device *rbd_dev)
3626 struct rbd_spec *parent_spec;
3628 void *reply_buf = NULL;
3636 parent_spec = rbd_spec_alloc();
3640 size = sizeof (__le64) + /* pool_id */
3641 sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX + /* image_id */
3642 sizeof (__le64) + /* snap_id */
3643 sizeof (__le64); /* overlap */
3644 reply_buf = kmalloc(size, GFP_KERNEL);
3650 snapid = cpu_to_le64(CEPH_NOSNAP);
3651 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
3652 "rbd", "get_parent",
3653 &snapid, sizeof (snapid),
3655 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
3660 end = reply_buf + ret;
3662 ceph_decode_64_safe(&p, end, parent_spec->pool_id, out_err);
3663 if (parent_spec->pool_id == CEPH_NOPOOL)
3664 goto out; /* No parent? No problem. */
3666 /* The ceph file layout needs to fit pool id in 32 bits */
3669 if (parent_spec->pool_id > (u64)U32_MAX) {
3670 rbd_warn(NULL, "parent pool id too large (%llu > %u)\n",
3671 (unsigned long long)parent_spec->pool_id, U32_MAX);
3675 image_id = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL);
3676 if (IS_ERR(image_id)) {
3677 ret = PTR_ERR(image_id);
3680 parent_spec->image_id = image_id;
3681 ceph_decode_64_safe(&p, end, parent_spec->snap_id, out_err);
3682 ceph_decode_64_safe(&p, end, overlap, out_err);
3684 rbd_dev->parent_overlap = overlap;
3685 rbd_dev->parent_spec = parent_spec;
3686 parent_spec = NULL; /* rbd_dev now owns this */
3691 rbd_spec_put(parent_spec);
3696 static int rbd_dev_v2_striping_info(struct rbd_device *rbd_dev)
3700 __le64 stripe_count;
3701 } __attribute__ ((packed)) striping_info_buf = { 0 };
3702 size_t size = sizeof (striping_info_buf);
3709 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
3710 "rbd", "get_stripe_unit_count", NULL, 0,
3711 (char *)&striping_info_buf, size);
3712 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
3719 * We don't actually support the "fancy striping" feature
3720 * (STRIPINGV2) yet, but if the striping sizes are the
3721 * defaults the behavior is the same as before. So find
3722 * out, and only fail if the image has non-default values.
3725 obj_size = (u64)1 << rbd_dev->header.obj_order;
3726 p = &striping_info_buf;
3727 stripe_unit = ceph_decode_64(&p);
3728 if (stripe_unit != obj_size) {
3729 rbd_warn(rbd_dev, "unsupported stripe unit "
3730 "(got %llu want %llu)",
3731 stripe_unit, obj_size);
3734 stripe_count = ceph_decode_64(&p);
3735 if (stripe_count != 1) {
3736 rbd_warn(rbd_dev, "unsupported stripe count "
3737 "(got %llu want 1)", stripe_count);
3740 rbd_dev->header.stripe_unit = stripe_unit;
3741 rbd_dev->header.stripe_count = stripe_count;
3746 static char *rbd_dev_image_name(struct rbd_device *rbd_dev)
3748 size_t image_id_size;
3753 void *reply_buf = NULL;
3755 char *image_name = NULL;
3758 rbd_assert(!rbd_dev->spec->image_name);
3760 len = strlen(rbd_dev->spec->image_id);
3761 image_id_size = sizeof (__le32) + len;
3762 image_id = kmalloc(image_id_size, GFP_KERNEL);
3767 end = image_id + image_id_size;
3768 ceph_encode_string(&p, end, rbd_dev->spec->image_id, (u32)len);
3770 size = sizeof (__le32) + RBD_IMAGE_NAME_LEN_MAX;
3771 reply_buf = kmalloc(size, GFP_KERNEL);
3775 ret = rbd_obj_method_sync(rbd_dev, RBD_DIRECTORY,
3776 "rbd", "dir_get_name",
3777 image_id, image_id_size,
3782 end = reply_buf + ret;
3784 image_name = ceph_extract_encoded_string(&p, end, &len, GFP_KERNEL);
3785 if (IS_ERR(image_name))
3788 dout("%s: name is %s len is %zd\n", __func__, image_name, len);
3796 static u64 rbd_v1_snap_id_by_name(struct rbd_device *rbd_dev, const char *name)
3798 struct ceph_snap_context *snapc = rbd_dev->header.snapc;
3799 const char *snap_name;
3802 /* Skip over names until we find the one we are looking for */
3804 snap_name = rbd_dev->header.snap_names;
3805 while (which < snapc->num_snaps) {
3806 if (!strcmp(name, snap_name))
3807 return snapc->snaps[which];
3808 snap_name += strlen(snap_name) + 1;
3814 static u64 rbd_v2_snap_id_by_name(struct rbd_device *rbd_dev, const char *name)
3816 struct ceph_snap_context *snapc = rbd_dev->header.snapc;
3821 for (which = 0; !found && which < snapc->num_snaps; which++) {
3822 const char *snap_name;
3824 snap_id = snapc->snaps[which];
3825 snap_name = rbd_dev_v2_snap_name(rbd_dev, snap_id);
3826 if (IS_ERR(snap_name))
3828 found = !strcmp(name, snap_name);
3831 return found ? snap_id : CEPH_NOSNAP;
3835 * Assumes name is never RBD_SNAP_HEAD_NAME; returns CEPH_NOSNAP if
3836 * no snapshot by that name is found, or if an error occurs.
3838 static u64 rbd_snap_id_by_name(struct rbd_device *rbd_dev, const char *name)
3840 if (rbd_dev->image_format == 1)
3841 return rbd_v1_snap_id_by_name(rbd_dev, name);
3843 return rbd_v2_snap_id_by_name(rbd_dev, name);
3847 * When an rbd image has a parent image, it is identified by the
3848 * pool, image, and snapshot ids (not names). This function fills
3849 * in the names for those ids. (It's OK if we can't figure out the
3850 * name for an image id, but the pool and snapshot ids should always
3851 * exist and have names.) All names in an rbd spec are dynamically
3854 * When an image being mapped (not a parent) is probed, we have the
3855 * pool name and pool id, image name and image id, and the snapshot
3856 * name. The only thing we're missing is the snapshot id.
3858 static int rbd_dev_spec_update(struct rbd_device *rbd_dev)
3860 struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
3861 struct rbd_spec *spec = rbd_dev->spec;
3862 const char *pool_name;
3863 const char *image_name;
3864 const char *snap_name;
3868 * An image being mapped will have the pool name (etc.), but
3869 * we need to look up the snapshot id.
3871 if (spec->pool_name) {
3872 if (strcmp(spec->snap_name, RBD_SNAP_HEAD_NAME)) {
3875 snap_id = rbd_snap_id_by_name(rbd_dev, spec->snap_name);
3876 if (snap_id == CEPH_NOSNAP)
3878 spec->snap_id = snap_id;
3880 spec->snap_id = CEPH_NOSNAP;
3886 /* Get the pool name; we have to make our own copy of this */
3888 pool_name = ceph_pg_pool_name_by_id(osdc->osdmap, spec->pool_id);
3890 rbd_warn(rbd_dev, "no pool with id %llu", spec->pool_id);
3893 pool_name = kstrdup(pool_name, GFP_KERNEL);
3897 /* Fetch the image name; tolerate failure here */
3899 image_name = rbd_dev_image_name(rbd_dev);
3901 rbd_warn(rbd_dev, "unable to get image name");
3903 /* Look up the snapshot name, and make a copy */
3905 snap_name = rbd_snap_name(rbd_dev, spec->snap_id);
3911 spec->pool_name = pool_name;
3912 spec->image_name = image_name;
3913 spec->snap_name = snap_name;
3923 static int rbd_dev_v2_snap_context(struct rbd_device *rbd_dev)
3932 struct ceph_snap_context *snapc;
3936 * We'll need room for the seq value (maximum snapshot id),
3937 * snapshot count, and array of that many snapshot ids.
3938 * For now we have a fixed upper limit on the number we're
3939 * prepared to receive.
3941 size = sizeof (__le64) + sizeof (__le32) +
3942 RBD_MAX_SNAP_COUNT * sizeof (__le64);
3943 reply_buf = kzalloc(size, GFP_KERNEL);
3947 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
3948 "rbd", "get_snapcontext", NULL, 0,
3950 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
3955 end = reply_buf + ret;
3957 ceph_decode_64_safe(&p, end, seq, out);
3958 ceph_decode_32_safe(&p, end, snap_count, out);
3961 * Make sure the reported number of snapshot ids wouldn't go
3962 * beyond the end of our buffer. But before checking that,
3963 * make sure the computed size of the snapshot context we
3964 * allocate is representable in a size_t.
3966 if (snap_count > (SIZE_MAX - sizeof (struct ceph_snap_context))
3971 if (!ceph_has_room(&p, end, snap_count * sizeof (__le64)))
3975 snapc = ceph_create_snap_context(snap_count, GFP_KERNEL);
3981 for (i = 0; i < snap_count; i++)
3982 snapc->snaps[i] = ceph_decode_64(&p);
3984 rbd_dev->header.snapc = snapc;
3986 dout(" snap context seq = %llu, snap_count = %u\n",
3987 (unsigned long long)seq, (unsigned int)snap_count);
3994 static const char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev,
4005 size = sizeof (__le32) + RBD_MAX_SNAP_NAME_LEN;
4006 reply_buf = kmalloc(size, GFP_KERNEL);
4008 return ERR_PTR(-ENOMEM);
4010 snapid = cpu_to_le64(snap_id);
4011 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
4012 "rbd", "get_snapshot_name",
4013 &snapid, sizeof (snapid),
4015 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
4017 snap_name = ERR_PTR(ret);
4022 end = reply_buf + ret;
4023 snap_name = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL);
4024 if (IS_ERR(snap_name))
4027 dout(" snap_id 0x%016llx snap_name = %s\n",
4028 (unsigned long long)snap_id, snap_name);
4035 static int rbd_dev_v2_refresh(struct rbd_device *rbd_dev)
4039 down_write(&rbd_dev->header_rwsem);
4041 ret = rbd_dev_v2_image_size(rbd_dev);
4044 rbd_update_mapping_size(rbd_dev);
4046 ret = rbd_dev_v2_snap_context(rbd_dev);
4047 dout("rbd_dev_v2_snap_context returned %d\n", ret);
4051 up_write(&rbd_dev->header_rwsem);
4056 static int rbd_bus_add_dev(struct rbd_device *rbd_dev)
4061 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
4063 dev = &rbd_dev->dev;
4064 dev->bus = &rbd_bus_type;
4065 dev->type = &rbd_device_type;
4066 dev->parent = &rbd_root_dev;
4067 dev->release = rbd_dev_device_release;
4068 dev_set_name(dev, "%d", rbd_dev->dev_id);
4069 ret = device_register(dev);
4071 mutex_unlock(&ctl_mutex);
4076 static void rbd_bus_del_dev(struct rbd_device *rbd_dev)
4078 device_unregister(&rbd_dev->dev);
4081 static atomic64_t rbd_dev_id_max = ATOMIC64_INIT(0);
4084 * Get a unique rbd identifier for the given new rbd_dev, and add
4085 * the rbd_dev to the global list. The minimum rbd id is 1.
4087 static void rbd_dev_id_get(struct rbd_device *rbd_dev)
4089 rbd_dev->dev_id = atomic64_inc_return(&rbd_dev_id_max);
4091 spin_lock(&rbd_dev_list_lock);
4092 list_add_tail(&rbd_dev->node, &rbd_dev_list);
4093 spin_unlock(&rbd_dev_list_lock);
4094 dout("rbd_dev %p given dev id %llu\n", rbd_dev,
4095 (unsigned long long) rbd_dev->dev_id);
4099 * Remove an rbd_dev from the global list, and record that its
4100 * identifier is no longer in use.
4102 static void rbd_dev_id_put(struct rbd_device *rbd_dev)
4104 struct list_head *tmp;
4105 int rbd_id = rbd_dev->dev_id;
4108 rbd_assert(rbd_id > 0);
4110 dout("rbd_dev %p released dev id %llu\n", rbd_dev,
4111 (unsigned long long) rbd_dev->dev_id);
4112 spin_lock(&rbd_dev_list_lock);
4113 list_del_init(&rbd_dev->node);
4116 * If the id being "put" is not the current maximum, there
4117 * is nothing special we need to do.
4119 if (rbd_id != atomic64_read(&rbd_dev_id_max)) {
4120 spin_unlock(&rbd_dev_list_lock);
4125 * We need to update the current maximum id. Search the
4126 * list to find out what it is. We're more likely to find
4127 * the maximum at the end, so search the list backward.
4130 list_for_each_prev(tmp, &rbd_dev_list) {
4131 struct rbd_device *rbd_dev;
4133 rbd_dev = list_entry(tmp, struct rbd_device, node);
4134 if (rbd_dev->dev_id > max_id)
4135 max_id = rbd_dev->dev_id;
4137 spin_unlock(&rbd_dev_list_lock);
4140 * The max id could have been updated by rbd_dev_id_get(), in
4141 * which case it now accurately reflects the new maximum.
4142 * Be careful not to overwrite the maximum value in that
4145 atomic64_cmpxchg(&rbd_dev_id_max, rbd_id, max_id);
4146 dout(" max dev id has been reset\n");
4150 * Skips over white space at *buf, and updates *buf to point to the
4151 * first found non-space character (if any). Returns the length of
4152 * the token (string of non-white space characters) found. Note
4153 * that *buf must be terminated with '\0'.
4155 static inline size_t next_token(const char **buf)
4158 * These are the characters that produce nonzero for
4159 * isspace() in the "C" and "POSIX" locales.
4161 const char *spaces = " \f\n\r\t\v";
4163 *buf += strspn(*buf, spaces); /* Find start of token */
4165 return strcspn(*buf, spaces); /* Return token length */
4169 * Finds the next token in *buf, and if the provided token buffer is
4170 * big enough, copies the found token into it. The result, if
4171 * copied, is guaranteed to be terminated with '\0'. Note that *buf
4172 * must be terminated with '\0' on entry.
4174 * Returns the length of the token found (not including the '\0').
4175 * Return value will be 0 if no token is found, and it will be >=
4176 * token_size if the token would not fit.
4178 * The *buf pointer will be updated to point beyond the end of the
4179 * found token. Note that this occurs even if the token buffer is
4180 * too small to hold it.
4182 static inline size_t copy_token(const char **buf,
4188 len = next_token(buf);
4189 if (len < token_size) {
4190 memcpy(token, *buf, len);
4191 *(token + len) = '\0';
4199 * Finds the next token in *buf, dynamically allocates a buffer big
4200 * enough to hold a copy of it, and copies the token into the new
4201 * buffer. The copy is guaranteed to be terminated with '\0'. Note
4202 * that a duplicate buffer is created even for a zero-length token.
4204 * Returns a pointer to the newly-allocated duplicate, or a null
4205 * pointer if memory for the duplicate was not available. If
4206 * the lenp argument is a non-null pointer, the length of the token
4207 * (not including the '\0') is returned in *lenp.
4209 * If successful, the *buf pointer will be updated to point beyond
4210 * the end of the found token.
4212 * Note: uses GFP_KERNEL for allocation.
4214 static inline char *dup_token(const char **buf, size_t *lenp)
4219 len = next_token(buf);
4220 dup = kmemdup(*buf, len + 1, GFP_KERNEL);
4223 *(dup + len) = '\0';
4233 * Parse the options provided for an "rbd add" (i.e., rbd image
4234 * mapping) request. These arrive via a write to /sys/bus/rbd/add,
4235 * and the data written is passed here via a NUL-terminated buffer.
4236 * Returns 0 if successful or an error code otherwise.
4238 * The information extracted from these options is recorded in
4239 * the other parameters which return dynamically-allocated
4242 * The address of a pointer that will refer to a ceph options
4243 * structure. Caller must release the returned pointer using
4244 * ceph_destroy_options() when it is no longer needed.
4246 * Address of an rbd options pointer. Fully initialized by
4247 * this function; caller must release with kfree().
4249 * Address of an rbd image specification pointer. Fully
4250 * initialized by this function based on parsed options.
4251 * Caller must release with rbd_spec_put().
4253 * The options passed take this form:
4254 * <mon_addrs> <options> <pool_name> <image_name> [<snap_id>]
4257 * A comma-separated list of one or more monitor addresses.
4258 * A monitor address is an ip address, optionally followed
4259 * by a port number (separated by a colon).
4260 * I.e.: ip1[:port1][,ip2[:port2]...]
4262 * A comma-separated list of ceph and/or rbd options.
4264 * The name of the rados pool containing the rbd image.
4266 * The name of the image in that pool to map.
4268 * An optional snapshot id. If provided, the mapping will
4269 * present data from the image at the time that snapshot was
4270 * created. The image head is used if no snapshot id is
4271 * provided. Snapshot mappings are always read-only.
4273 static int rbd_add_parse_args(const char *buf,
4274 struct ceph_options **ceph_opts,
4275 struct rbd_options **opts,
4276 struct rbd_spec **rbd_spec)
4280 const char *mon_addrs;
4282 size_t mon_addrs_size;
4283 struct rbd_spec *spec = NULL;
4284 struct rbd_options *rbd_opts = NULL;
4285 struct ceph_options *copts;
4288 /* The first four tokens are required */
4290 len = next_token(&buf);
4292 rbd_warn(NULL, "no monitor address(es) provided");
4296 mon_addrs_size = len + 1;
4300 options = dup_token(&buf, NULL);
4304 rbd_warn(NULL, "no options provided");
4308 spec = rbd_spec_alloc();
4312 spec->pool_name = dup_token(&buf, NULL);
4313 if (!spec->pool_name)
4315 if (!*spec->pool_name) {
4316 rbd_warn(NULL, "no pool name provided");
4320 spec->image_name = dup_token(&buf, NULL);
4321 if (!spec->image_name)
4323 if (!*spec->image_name) {
4324 rbd_warn(NULL, "no image name provided");
4329 * Snapshot name is optional; default is to use "-"
4330 * (indicating the head/no snapshot).
4332 len = next_token(&buf);
4334 buf = RBD_SNAP_HEAD_NAME; /* No snapshot supplied */
4335 len = sizeof (RBD_SNAP_HEAD_NAME) - 1;
4336 } else if (len > RBD_MAX_SNAP_NAME_LEN) {
4337 ret = -ENAMETOOLONG;
4340 snap_name = kmemdup(buf, len + 1, GFP_KERNEL);
4343 *(snap_name + len) = '\0';
4344 spec->snap_name = snap_name;
4346 /* Initialize all rbd options to the defaults */
4348 rbd_opts = kzalloc(sizeof (*rbd_opts), GFP_KERNEL);
4352 rbd_opts->read_only = RBD_READ_ONLY_DEFAULT;
4354 copts = ceph_parse_options(options, mon_addrs,
4355 mon_addrs + mon_addrs_size - 1,
4356 parse_rbd_opts_token, rbd_opts);
4357 if (IS_ERR(copts)) {
4358 ret = PTR_ERR(copts);
4379 * An rbd format 2 image has a unique identifier, distinct from the
4380 * name given to it by the user. Internally, that identifier is
4381 * what's used to specify the names of objects related to the image.
4383 * A special "rbd id" object is used to map an rbd image name to its
4384 * id. If that object doesn't exist, then there is no v2 rbd image
4385 * with the supplied name.
4387 * This function will record the given rbd_dev's image_id field if
4388 * it can be determined, and in that case will return 0. If any
4389 * errors occur a negative errno will be returned and the rbd_dev's
4390 * image_id field will be unchanged (and should be NULL).
4392 static int rbd_dev_image_id(struct rbd_device *rbd_dev)
4401 * When probing a parent image, the image id is already
4402 * known (and the image name likely is not). There's no
4403 * need to fetch the image id again in this case. We
4404 * do still need to set the image format though.
4406 if (rbd_dev->spec->image_id) {
4407 rbd_dev->image_format = *rbd_dev->spec->image_id ? 2 : 1;
4413 * First, see if the format 2 image id file exists, and if
4414 * so, get the image's persistent id from it.
4416 size = sizeof (RBD_ID_PREFIX) + strlen(rbd_dev->spec->image_name);
4417 object_name = kmalloc(size, GFP_NOIO);
4420 sprintf(object_name, "%s%s", RBD_ID_PREFIX, rbd_dev->spec->image_name);
4421 dout("rbd id object name is %s\n", object_name);
4423 /* Response will be an encoded string, which includes a length */
4425 size = sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX;
4426 response = kzalloc(size, GFP_NOIO);
4432 /* If it doesn't exist we'll assume it's a format 1 image */
4434 ret = rbd_obj_method_sync(rbd_dev, object_name,
4435 "rbd", "get_id", NULL, 0,
4436 response, RBD_IMAGE_ID_LEN_MAX);
4437 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
4438 if (ret == -ENOENT) {
4439 image_id = kstrdup("", GFP_KERNEL);
4440 ret = image_id ? 0 : -ENOMEM;
4442 rbd_dev->image_format = 1;
4443 } else if (ret > sizeof (__le32)) {
4446 image_id = ceph_extract_encoded_string(&p, p + ret,
4448 ret = IS_ERR(image_id) ? PTR_ERR(image_id) : 0;
4450 rbd_dev->image_format = 2;
4456 rbd_dev->spec->image_id = image_id;
4457 dout("image_id is %s\n", image_id);
4466 /* Undo whatever state changes are made by v1 or v2 image probe */
4468 static void rbd_dev_unprobe(struct rbd_device *rbd_dev)
4470 struct rbd_image_header *header;
4472 rbd_dev_remove_parent(rbd_dev);
4473 rbd_spec_put(rbd_dev->parent_spec);
4474 rbd_dev->parent_spec = NULL;
4475 rbd_dev->parent_overlap = 0;
4477 /* Free dynamic fields from the header, then zero it out */
4479 header = &rbd_dev->header;
4480 ceph_put_snap_context(header->snapc);
4481 kfree(header->snap_sizes);
4482 kfree(header->snap_names);
4483 kfree(header->object_prefix);
4484 memset(header, 0, sizeof (*header));
4487 static int rbd_dev_v1_probe(struct rbd_device *rbd_dev)
4491 /* Populate rbd image metadata */
4493 ret = rbd_read_header(rbd_dev, &rbd_dev->header);
4497 /* Version 1 images have no parent (no layering) */
4499 rbd_dev->parent_spec = NULL;
4500 rbd_dev->parent_overlap = 0;
4502 dout("discovered version 1 image, header name is %s\n",
4503 rbd_dev->header_name);
4508 kfree(rbd_dev->header_name);
4509 rbd_dev->header_name = NULL;
4510 kfree(rbd_dev->spec->image_id);
4511 rbd_dev->spec->image_id = NULL;
4516 static int rbd_dev_v2_probe(struct rbd_device *rbd_dev)
4520 ret = rbd_dev_v2_image_size(rbd_dev);
4524 /* Get the object prefix (a.k.a. block_name) for the image */
4526 ret = rbd_dev_v2_object_prefix(rbd_dev);
4530 /* Get the and check features for the image */
4532 ret = rbd_dev_v2_features(rbd_dev);
4536 /* If the image supports layering, get the parent info */
4538 if (rbd_dev->header.features & RBD_FEATURE_LAYERING) {
4539 ret = rbd_dev_v2_parent_info(rbd_dev);
4544 * Don't print a warning for parent images. We can
4545 * tell this point because we won't know its pool
4546 * name yet (just its pool id).
4548 if (rbd_dev->spec->pool_name)
4549 rbd_warn(rbd_dev, "WARNING: kernel layering "
4550 "is EXPERIMENTAL!");
4553 /* If the image supports fancy striping, get its parameters */
4555 if (rbd_dev->header.features & RBD_FEATURE_STRIPINGV2) {
4556 ret = rbd_dev_v2_striping_info(rbd_dev);
4561 /* crypto and compression type aren't (yet) supported for v2 images */
4563 rbd_dev->header.crypt_type = 0;
4564 rbd_dev->header.comp_type = 0;
4566 /* Get the snapshot context, plus the header version */
4568 ret = rbd_dev_v2_snap_context(rbd_dev);
4572 dout("discovered version 2 image, header name is %s\n",
4573 rbd_dev->header_name);
4577 rbd_dev->parent_overlap = 0;
4578 rbd_spec_put(rbd_dev->parent_spec);
4579 rbd_dev->parent_spec = NULL;
4580 kfree(rbd_dev->header_name);
4581 rbd_dev->header_name = NULL;
4582 kfree(rbd_dev->header.object_prefix);
4583 rbd_dev->header.object_prefix = NULL;
4588 static int rbd_dev_probe_parent(struct rbd_device *rbd_dev)
4590 struct rbd_device *parent = NULL;
4591 struct rbd_spec *parent_spec;
4592 struct rbd_client *rbdc;
4595 if (!rbd_dev->parent_spec)
4598 * We need to pass a reference to the client and the parent
4599 * spec when creating the parent rbd_dev. Images related by
4600 * parent/child relationships always share both.
4602 parent_spec = rbd_spec_get(rbd_dev->parent_spec);
4603 rbdc = __rbd_get_client(rbd_dev->rbd_client);
4606 parent = rbd_dev_create(rbdc, parent_spec);
4610 ret = rbd_dev_image_probe(parent);
4613 rbd_dev->parent = parent;
4618 rbd_spec_put(rbd_dev->parent_spec);
4619 kfree(rbd_dev->header_name);
4620 rbd_dev_destroy(parent);
4622 rbd_put_client(rbdc);
4623 rbd_spec_put(parent_spec);
4629 static int rbd_dev_device_setup(struct rbd_device *rbd_dev)
4633 ret = rbd_dev_mapping_set(rbd_dev);
4637 /* generate unique id: find highest unique id, add one */
4638 rbd_dev_id_get(rbd_dev);
4640 /* Fill in the device name, now that we have its id. */
4641 BUILD_BUG_ON(DEV_NAME_LEN
4642 < sizeof (RBD_DRV_NAME) + MAX_INT_FORMAT_WIDTH);
4643 sprintf(rbd_dev->name, "%s%d", RBD_DRV_NAME, rbd_dev->dev_id);
4645 /* Get our block major device number. */
4647 ret = register_blkdev(0, rbd_dev->name);
4650 rbd_dev->major = ret;
4652 /* Set up the blkdev mapping. */
4654 ret = rbd_init_disk(rbd_dev);
4656 goto err_out_blkdev;
4658 ret = rbd_bus_add_dev(rbd_dev);
4662 /* Everything's ready. Announce the disk to the world. */
4664 set_capacity(rbd_dev->disk, rbd_dev->mapping.size / SECTOR_SIZE);
4665 set_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags);
4666 add_disk(rbd_dev->disk);
4668 pr_info("%s: added with size 0x%llx\n", rbd_dev->disk->disk_name,
4669 (unsigned long long) rbd_dev->mapping.size);
4674 rbd_free_disk(rbd_dev);
4676 unregister_blkdev(rbd_dev->major, rbd_dev->name);
4678 rbd_dev_id_put(rbd_dev);
4679 rbd_dev_mapping_clear(rbd_dev);
4684 static int rbd_dev_header_name(struct rbd_device *rbd_dev)
4686 struct rbd_spec *spec = rbd_dev->spec;
4689 /* Record the header object name for this rbd image. */
4691 rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
4693 if (rbd_dev->image_format == 1)
4694 size = strlen(spec->image_name) + sizeof (RBD_SUFFIX);
4696 size = sizeof (RBD_HEADER_PREFIX) + strlen(spec->image_id);
4698 rbd_dev->header_name = kmalloc(size, GFP_KERNEL);
4699 if (!rbd_dev->header_name)
4702 if (rbd_dev->image_format == 1)
4703 sprintf(rbd_dev->header_name, "%s%s",
4704 spec->image_name, RBD_SUFFIX);
4706 sprintf(rbd_dev->header_name, "%s%s",
4707 RBD_HEADER_PREFIX, spec->image_id);
4711 static void rbd_dev_image_release(struct rbd_device *rbd_dev)
4715 rbd_dev_unprobe(rbd_dev);
4716 ret = rbd_dev_header_watch_sync(rbd_dev, 0);
4718 rbd_warn(rbd_dev, "failed to cancel watch event (%d)\n", ret);
4719 kfree(rbd_dev->header_name);
4720 rbd_dev->header_name = NULL;
4721 rbd_dev->image_format = 0;
4722 kfree(rbd_dev->spec->image_id);
4723 rbd_dev->spec->image_id = NULL;
4725 rbd_dev_destroy(rbd_dev);
4729 * Probe for the existence of the header object for the given rbd
4730 * device. For format 2 images this includes determining the image
4733 static int rbd_dev_image_probe(struct rbd_device *rbd_dev)
4739 * Get the id from the image id object. If it's not a
4740 * format 2 image, we'll get ENOENT back, and we'll assume
4741 * it's a format 1 image.
4743 ret = rbd_dev_image_id(rbd_dev);
4746 rbd_assert(rbd_dev->spec->image_id);
4747 rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
4749 ret = rbd_dev_header_name(rbd_dev);
4751 goto err_out_format;
4753 ret = rbd_dev_header_watch_sync(rbd_dev, 1);
4755 goto out_header_name;
4757 if (rbd_dev->image_format == 1)
4758 ret = rbd_dev_v1_probe(rbd_dev);
4760 ret = rbd_dev_v2_probe(rbd_dev);
4764 ret = rbd_dev_spec_update(rbd_dev);
4768 ret = rbd_dev_probe_parent(rbd_dev);
4773 rbd_dev_unprobe(rbd_dev);
4775 tmp = rbd_dev_header_watch_sync(rbd_dev, 0);
4777 rbd_warn(rbd_dev, "unable to tear down watch request\n");
4779 kfree(rbd_dev->header_name);
4780 rbd_dev->header_name = NULL;
4782 rbd_dev->image_format = 0;
4783 kfree(rbd_dev->spec->image_id);
4784 rbd_dev->spec->image_id = NULL;
4786 dout("probe failed, returning %d\n", ret);
4791 static ssize_t rbd_add(struct bus_type *bus,
4795 struct rbd_device *rbd_dev = NULL;
4796 struct ceph_options *ceph_opts = NULL;
4797 struct rbd_options *rbd_opts = NULL;
4798 struct rbd_spec *spec = NULL;
4799 struct rbd_client *rbdc;
4800 struct ceph_osd_client *osdc;
4803 if (!try_module_get(THIS_MODULE))
4806 /* parse add command */
4807 rc = rbd_add_parse_args(buf, &ceph_opts, &rbd_opts, &spec);
4809 goto err_out_module;
4811 rbdc = rbd_get_client(ceph_opts);
4816 ceph_opts = NULL; /* rbd_dev client now owns this */
4819 osdc = &rbdc->client->osdc;
4820 rc = ceph_pg_poolid_by_name(osdc->osdmap, spec->pool_name);
4822 goto err_out_client;
4823 spec->pool_id = (u64)rc;
4825 /* The ceph file layout needs to fit pool id in 32 bits */
4827 if (spec->pool_id > (u64)U32_MAX) {
4828 rbd_warn(NULL, "pool id too large (%llu > %u)\n",
4829 (unsigned long long)spec->pool_id, U32_MAX);
4831 goto err_out_client;
4834 rbd_dev = rbd_dev_create(rbdc, spec);
4836 goto err_out_client;
4837 rbdc = NULL; /* rbd_dev now owns this */
4838 spec = NULL; /* rbd_dev now owns this */
4840 rbd_dev->mapping.read_only = rbd_opts->read_only;
4842 rbd_opts = NULL; /* done with this */
4844 rc = rbd_dev_image_probe(rbd_dev);
4846 goto err_out_rbd_dev;
4848 rc = rbd_dev_device_setup(rbd_dev);
4852 rbd_dev_image_release(rbd_dev);
4854 rbd_dev_destroy(rbd_dev);
4856 rbd_put_client(rbdc);
4859 ceph_destroy_options(ceph_opts);
4863 module_put(THIS_MODULE);
4865 dout("Error adding device %s\n", buf);
4870 static struct rbd_device *__rbd_get_dev(unsigned long dev_id)
4872 struct list_head *tmp;
4873 struct rbd_device *rbd_dev;
4875 spin_lock(&rbd_dev_list_lock);
4876 list_for_each(tmp, &rbd_dev_list) {
4877 rbd_dev = list_entry(tmp, struct rbd_device, node);
4878 if (rbd_dev->dev_id == dev_id) {
4879 spin_unlock(&rbd_dev_list_lock);
4883 spin_unlock(&rbd_dev_list_lock);
4887 static void rbd_dev_device_release(struct device *dev)
4889 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
4891 rbd_free_disk(rbd_dev);
4892 clear_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags);
4893 rbd_dev_clear_mapping(rbd_dev);
4894 unregister_blkdev(rbd_dev->major, rbd_dev->name);
4896 rbd_dev_id_put(rbd_dev);
4897 rbd_dev_mapping_clear(rbd_dev);
4900 static void rbd_dev_remove_parent(struct rbd_device *rbd_dev)
4902 while (rbd_dev->parent) {
4903 struct rbd_device *first = rbd_dev;
4904 struct rbd_device *second = first->parent;
4905 struct rbd_device *third;
4908 * Follow to the parent with no grandparent and
4911 while (second && (third = second->parent)) {
4916 rbd_dev_image_release(second);
4917 first->parent = NULL;
4918 first->parent_overlap = 0;
4920 rbd_assert(first->parent_spec);
4921 rbd_spec_put(first->parent_spec);
4922 first->parent_spec = NULL;
4926 static ssize_t rbd_remove(struct bus_type *bus,
4930 struct rbd_device *rbd_dev = NULL;
4935 ret = strict_strtoul(buf, 10, &ul);
4939 /* convert to int; abort if we lost anything in the conversion */
4940 target_id = (int) ul;
4941 if (target_id != ul)
4944 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
4946 rbd_dev = __rbd_get_dev(target_id);
4952 spin_lock_irq(&rbd_dev->lock);
4953 if (rbd_dev->open_count)
4956 set_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags);
4957 spin_unlock_irq(&rbd_dev->lock);
4961 rbd_bus_del_dev(rbd_dev);
4962 rbd_dev_image_release(rbd_dev);
4963 module_put(THIS_MODULE);
4965 mutex_unlock(&ctl_mutex);
4971 * create control files in sysfs
4974 static int rbd_sysfs_init(void)
4978 ret = device_register(&rbd_root_dev);
4982 ret = bus_register(&rbd_bus_type);
4984 device_unregister(&rbd_root_dev);
4989 static void rbd_sysfs_cleanup(void)
4991 bus_unregister(&rbd_bus_type);
4992 device_unregister(&rbd_root_dev);
4995 static int __init rbd_init(void)
4999 if (!libceph_compatible(NULL)) {
5000 rbd_warn(NULL, "libceph incompatibility (quitting)");
5004 rc = rbd_sysfs_init();
5007 pr_info("loaded " RBD_DRV_NAME_LONG "\n");
5011 static void __exit rbd_exit(void)
5013 rbd_sysfs_cleanup();
5016 module_init(rbd_init);
5017 module_exit(rbd_exit);
5019 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
5020 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
5021 MODULE_DESCRIPTION("rados block device");
5023 /* following authorship retained from original osdblk.c */
5024 MODULE_AUTHOR("Jeff Garzik <jeff@garzik.org>");
5026 MODULE_LICENSE("GPL");