2 rbd.c -- Export ceph rados objects as a Linux block device
5 based on drivers/block/osdblk.c:
7 Copyright 2009 Red Hat, Inc.
9 This program is free software; you can redistribute it and/or modify
10 it under the terms of the GNU General Public License as published by
11 the Free Software Foundation.
13 This program is distributed in the hope that it will be useful,
14 but WITHOUT ANY WARRANTY; without even the implied warranty of
15 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 GNU General Public License for more details.
18 You should have received a copy of the GNU General Public License
19 along with this program; see the file COPYING. If not, write to
20 the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA.
24 For usage instructions, please refer to:
26 Documentation/ABI/testing/sysfs-bus-rbd
30 #include <linux/ceph/libceph.h>
31 #include <linux/ceph/osd_client.h>
32 #include <linux/ceph/mon_client.h>
33 #include <linux/ceph/decode.h>
34 #include <linux/parser.h>
36 #include <linux/kernel.h>
37 #include <linux/device.h>
38 #include <linux/module.h>
40 #include <linux/blkdev.h>
42 #include "rbd_types.h"
44 #define RBD_DEBUG /* Activate rbd_assert() calls */
47 * The basic unit of block I/O is a sector. It is interpreted in a
48 * number of contexts in Linux (blk, bio, genhd), but the default is
49 * universally 512 bytes. These symbols are just slightly more
50 * meaningful than the bare numbers they represent.
52 #define SECTOR_SHIFT 9
53 #define SECTOR_SIZE (1ULL << SECTOR_SHIFT)
55 #define RBD_DRV_NAME "rbd"
56 #define RBD_DRV_NAME_LONG "rbd (rados block device)"
58 #define RBD_MINORS_PER_MAJOR 256 /* max minors per blkdev */
60 #define RBD_SNAP_DEV_NAME_PREFIX "snap_"
61 #define RBD_MAX_SNAP_NAME_LEN \
62 (NAME_MAX - (sizeof (RBD_SNAP_DEV_NAME_PREFIX) - 1))
64 #define RBD_MAX_SNAP_COUNT 510 /* allows max snapc to fit in 4KB */
66 #define RBD_SNAP_HEAD_NAME "-"
68 /* This allows a single page to hold an image name sent by OSD */
69 #define RBD_IMAGE_NAME_LEN_MAX (PAGE_SIZE - sizeof (__le32) - 1)
70 #define RBD_IMAGE_ID_LEN_MAX 64
72 #define RBD_OBJ_PREFIX_LEN_MAX 64
76 #define RBD_FEATURE_LAYERING (1<<0)
77 #define RBD_FEATURE_STRIPINGV2 (1<<1)
78 #define RBD_FEATURES_ALL \
79 (RBD_FEATURE_LAYERING | RBD_FEATURE_STRIPINGV2)
81 /* Features supported by this (client software) implementation. */
83 #define RBD_FEATURES_SUPPORTED (0)
86 * An RBD device name will be "rbd#", where the "rbd" comes from
87 * RBD_DRV_NAME above, and # is a unique integer identifier.
88 * MAX_INT_FORMAT_WIDTH is used in ensuring DEV_NAME_LEN is big
89 * enough to hold all possible device names.
91 #define DEV_NAME_LEN 32
92 #define MAX_INT_FORMAT_WIDTH ((5 * sizeof (int)) / 2 + 1)
95 * block device image metadata (in-memory version)
97 struct rbd_image_header {
98 /* These four fields never change for a given rbd image */
105 /* The remaining fields need to be updated occasionally */
107 struct ceph_snap_context *snapc;
115 * An rbd image specification.
117 * The tuple (pool_id, image_id, snap_id) is sufficient to uniquely
118 * identify an image. Each rbd_dev structure includes a pointer to
119 * an rbd_spec structure that encapsulates this identity.
121 * Each of the id's in an rbd_spec has an associated name. For a
122 * user-mapped image, the names are supplied and the id's associated
123 * with them are looked up. For a layered image, a parent image is
124 * defined by the tuple, and the names are looked up.
126 * An rbd_dev structure contains a parent_spec pointer which is
127 * non-null if the image it represents is a child in a layered
128 * image. This pointer will refer to the rbd_spec structure used
129 * by the parent rbd_dev for its own identity (i.e., the structure
130 * is shared between the parent and child).
132 * Since these structures are populated once, during the discovery
133 * phase of image construction, they are effectively immutable so
134 * we make no effort to synchronize access to them.
136 * Note that code herein does not assume the image name is known (it
137 * could be a null pointer).
153 * an instance of the client. multiple devices may share an rbd client.
156 struct ceph_client *client;
158 struct list_head node;
161 struct rbd_img_request;
162 typedef void (*rbd_img_callback_t)(struct rbd_img_request *);
164 #define BAD_WHICH U32_MAX /* Good which or bad which, which? */
166 struct rbd_obj_request;
167 typedef void (*rbd_obj_callback_t)(struct rbd_obj_request *);
169 enum obj_request_type {
170 OBJ_REQUEST_NODATA, OBJ_REQUEST_BIO, OBJ_REQUEST_PAGES
174 OBJ_REQ_DONE, /* completion flag: not done = 0, done = 1 */
175 OBJ_REQ_IMG_DATA, /* object usage: standalone = 0, image = 1 */
176 OBJ_REQ_KNOWN, /* EXISTS flag valid: no = 0, yes = 1 */
177 OBJ_REQ_EXISTS, /* target exists: no = 0, yes = 1 */
180 struct rbd_obj_request {
181 const char *object_name;
182 u64 offset; /* object start byte */
183 u64 length; /* bytes from offset */
187 * An object request associated with an image will have its
188 * img_data flag set; a standalone object request will not.
190 * A standalone object request will have which == BAD_WHICH
191 * and a null obj_request pointer.
193 * An object request initiated in support of a layered image
194 * object (to check for its existence before a write) will
195 * have which == BAD_WHICH and a non-null obj_request pointer.
197 * Finally, an object request for rbd image data will have
198 * which != BAD_WHICH, and will have a non-null img_request
199 * pointer. The value of which will be in the range
200 * 0..(img_request->obj_request_count-1).
203 struct rbd_obj_request *obj_request; /* STAT op */
205 struct rbd_img_request *img_request;
207 /* links for img_request->obj_requests list */
208 struct list_head links;
211 u32 which; /* posn image request list */
213 enum obj_request_type type;
215 struct bio *bio_list;
221 struct page **copyup_pages;
223 struct ceph_osd_request *osd_req;
225 u64 xferred; /* bytes transferred */
229 rbd_obj_callback_t callback;
230 struct completion completion;
236 IMG_REQ_WRITE, /* I/O direction: read = 0, write = 1 */
237 IMG_REQ_CHILD, /* initiator: block = 0, child image = 1 */
238 IMG_REQ_LAYERED, /* ENOENT handling: normal = 0, layered = 1 */
241 struct rbd_img_request {
242 struct rbd_device *rbd_dev;
243 u64 offset; /* starting image byte offset */
244 u64 length; /* byte count from offset */
247 u64 snap_id; /* for reads */
248 struct ceph_snap_context *snapc; /* for writes */
251 struct request *rq; /* block request */
252 struct rbd_obj_request *obj_request; /* obj req initiator */
254 struct page **copyup_pages;
255 spinlock_t completion_lock;/* protects next_completion */
257 rbd_img_callback_t callback;
258 u64 xferred;/* aggregate bytes transferred */
259 int result; /* first nonzero obj_request result */
261 u32 obj_request_count;
262 struct list_head obj_requests; /* rbd_obj_request structs */
267 #define for_each_obj_request(ireq, oreq) \
268 list_for_each_entry(oreq, &(ireq)->obj_requests, links)
269 #define for_each_obj_request_from(ireq, oreq) \
270 list_for_each_entry_from(oreq, &(ireq)->obj_requests, links)
271 #define for_each_obj_request_safe(ireq, oreq, n) \
272 list_for_each_entry_safe_reverse(oreq, n, &(ireq)->obj_requests, links)
278 struct list_head node;
293 int dev_id; /* blkdev unique id */
295 int major; /* blkdev assigned major */
296 struct gendisk *disk; /* blkdev's gendisk and rq */
298 u32 image_format; /* Either 1 or 2 */
299 struct rbd_client *rbd_client;
301 char name[DEV_NAME_LEN]; /* blkdev name, e.g. rbd3 */
303 spinlock_t lock; /* queue, flags, open_count */
305 struct rbd_image_header header;
306 unsigned long flags; /* possibly lock protected */
307 struct rbd_spec *spec;
311 struct ceph_file_layout layout;
313 struct ceph_osd_event *watch_event;
314 struct rbd_obj_request *watch_request;
316 struct rbd_spec *parent_spec;
318 struct rbd_device *parent;
320 /* protects updating the header */
321 struct rw_semaphore header_rwsem;
323 struct rbd_mapping mapping;
325 struct list_head node;
327 /* list of snapshots */
328 struct list_head snaps;
332 unsigned long open_count; /* protected by lock */
336 * Flag bits for rbd_dev->flags. If atomicity is required,
337 * rbd_dev->lock is used to protect access.
339 * Currently, only the "removing" flag (which is coupled with the
340 * "open_count" field) requires atomic access.
343 RBD_DEV_FLAG_EXISTS, /* mapped snapshot has not been deleted */
344 RBD_DEV_FLAG_REMOVING, /* this mapping is being removed */
347 static DEFINE_MUTEX(ctl_mutex); /* Serialize open/close/setup/teardown */
349 static LIST_HEAD(rbd_dev_list); /* devices */
350 static DEFINE_SPINLOCK(rbd_dev_list_lock);
352 static LIST_HEAD(rbd_client_list); /* clients */
353 static DEFINE_SPINLOCK(rbd_client_list_lock);
355 static int rbd_img_request_submit(struct rbd_img_request *img_request);
357 static int rbd_dev_snaps_update(struct rbd_device *rbd_dev);
358 static int rbd_dev_snaps_register(struct rbd_device *rbd_dev);
360 static void rbd_dev_release(struct device *dev);
361 static void rbd_remove_snap_dev(struct rbd_snap *snap);
363 static ssize_t rbd_add(struct bus_type *bus, const char *buf,
365 static ssize_t rbd_remove(struct bus_type *bus, const char *buf,
367 static int rbd_dev_probe(struct rbd_device *rbd_dev);
369 static struct bus_attribute rbd_bus_attrs[] = {
370 __ATTR(add, S_IWUSR, NULL, rbd_add),
371 __ATTR(remove, S_IWUSR, NULL, rbd_remove),
375 static struct bus_type rbd_bus_type = {
377 .bus_attrs = rbd_bus_attrs,
380 static void rbd_root_dev_release(struct device *dev)
384 static struct device rbd_root_dev = {
386 .release = rbd_root_dev_release,
389 static __printf(2, 3)
390 void rbd_warn(struct rbd_device *rbd_dev, const char *fmt, ...)
392 struct va_format vaf;
400 printk(KERN_WARNING "%s: %pV\n", RBD_DRV_NAME, &vaf);
401 else if (rbd_dev->disk)
402 printk(KERN_WARNING "%s: %s: %pV\n",
403 RBD_DRV_NAME, rbd_dev->disk->disk_name, &vaf);
404 else if (rbd_dev->spec && rbd_dev->spec->image_name)
405 printk(KERN_WARNING "%s: image %s: %pV\n",
406 RBD_DRV_NAME, rbd_dev->spec->image_name, &vaf);
407 else if (rbd_dev->spec && rbd_dev->spec->image_id)
408 printk(KERN_WARNING "%s: id %s: %pV\n",
409 RBD_DRV_NAME, rbd_dev->spec->image_id, &vaf);
411 printk(KERN_WARNING "%s: rbd_dev %p: %pV\n",
412 RBD_DRV_NAME, rbd_dev, &vaf);
417 #define rbd_assert(expr) \
418 if (unlikely(!(expr))) { \
419 printk(KERN_ERR "\nAssertion failure in %s() " \
421 "\trbd_assert(%s);\n\n", \
422 __func__, __LINE__, #expr); \
425 #else /* !RBD_DEBUG */
426 # define rbd_assert(expr) ((void) 0)
427 #endif /* !RBD_DEBUG */
429 static void rbd_img_parent_read(struct rbd_obj_request *obj_request);
430 static int rbd_img_obj_request_submit(struct rbd_obj_request *obj_request);
432 static int rbd_dev_refresh(struct rbd_device *rbd_dev, u64 *hver);
433 static int rbd_dev_v2_refresh(struct rbd_device *rbd_dev, u64 *hver);
435 static int rbd_open(struct block_device *bdev, fmode_t mode)
437 struct rbd_device *rbd_dev = bdev->bd_disk->private_data;
438 bool removing = false;
440 if ((mode & FMODE_WRITE) && rbd_dev->mapping.read_only)
443 spin_lock_irq(&rbd_dev->lock);
444 if (test_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags))
447 rbd_dev->open_count++;
448 spin_unlock_irq(&rbd_dev->lock);
452 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
453 (void) get_device(&rbd_dev->dev);
454 set_device_ro(bdev, rbd_dev->mapping.read_only);
455 mutex_unlock(&ctl_mutex);
460 static int rbd_release(struct gendisk *disk, fmode_t mode)
462 struct rbd_device *rbd_dev = disk->private_data;
463 unsigned long open_count_before;
465 spin_lock_irq(&rbd_dev->lock);
466 open_count_before = rbd_dev->open_count--;
467 spin_unlock_irq(&rbd_dev->lock);
468 rbd_assert(open_count_before > 0);
470 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
471 put_device(&rbd_dev->dev);
472 mutex_unlock(&ctl_mutex);
477 static const struct block_device_operations rbd_bd_ops = {
478 .owner = THIS_MODULE,
480 .release = rbd_release,
484 * Initialize an rbd client instance.
487 static struct rbd_client *rbd_client_create(struct ceph_options *ceph_opts)
489 struct rbd_client *rbdc;
492 dout("%s:\n", __func__);
493 rbdc = kmalloc(sizeof(struct rbd_client), GFP_KERNEL);
497 kref_init(&rbdc->kref);
498 INIT_LIST_HEAD(&rbdc->node);
500 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
502 rbdc->client = ceph_create_client(ceph_opts, rbdc, 0, 0);
503 if (IS_ERR(rbdc->client))
505 ceph_opts = NULL; /* Now rbdc->client is responsible for ceph_opts */
507 ret = ceph_open_session(rbdc->client);
511 spin_lock(&rbd_client_list_lock);
512 list_add_tail(&rbdc->node, &rbd_client_list);
513 spin_unlock(&rbd_client_list_lock);
515 mutex_unlock(&ctl_mutex);
516 dout("%s: rbdc %p\n", __func__, rbdc);
521 ceph_destroy_client(rbdc->client);
523 mutex_unlock(&ctl_mutex);
527 ceph_destroy_options(ceph_opts);
528 dout("%s: error %d\n", __func__, ret);
533 static struct rbd_client *__rbd_get_client(struct rbd_client *rbdc)
535 kref_get(&rbdc->kref);
541 * Find a ceph client with specific addr and configuration. If
542 * found, bump its reference count.
544 static struct rbd_client *rbd_client_find(struct ceph_options *ceph_opts)
546 struct rbd_client *client_node;
549 if (ceph_opts->flags & CEPH_OPT_NOSHARE)
552 spin_lock(&rbd_client_list_lock);
553 list_for_each_entry(client_node, &rbd_client_list, node) {
554 if (!ceph_compare_options(ceph_opts, client_node->client)) {
555 __rbd_get_client(client_node);
561 spin_unlock(&rbd_client_list_lock);
563 return found ? client_node : NULL;
573 /* string args above */
576 /* Boolean args above */
580 static match_table_t rbd_opts_tokens = {
582 /* string args above */
583 {Opt_read_only, "read_only"},
584 {Opt_read_only, "ro"}, /* Alternate spelling */
585 {Opt_read_write, "read_write"},
586 {Opt_read_write, "rw"}, /* Alternate spelling */
587 /* Boolean args above */
595 #define RBD_READ_ONLY_DEFAULT false
597 static int parse_rbd_opts_token(char *c, void *private)
599 struct rbd_options *rbd_opts = private;
600 substring_t argstr[MAX_OPT_ARGS];
601 int token, intval, ret;
603 token = match_token(c, rbd_opts_tokens, argstr);
607 if (token < Opt_last_int) {
608 ret = match_int(&argstr[0], &intval);
610 pr_err("bad mount option arg (not int) "
614 dout("got int token %d val %d\n", token, intval);
615 } else if (token > Opt_last_int && token < Opt_last_string) {
616 dout("got string token %d val %s\n", token,
618 } else if (token > Opt_last_string && token < Opt_last_bool) {
619 dout("got Boolean token %d\n", token);
621 dout("got token %d\n", token);
626 rbd_opts->read_only = true;
629 rbd_opts->read_only = false;
639 * Get a ceph client with specific addr and configuration, if one does
640 * not exist create it.
642 static struct rbd_client *rbd_get_client(struct ceph_options *ceph_opts)
644 struct rbd_client *rbdc;
646 rbdc = rbd_client_find(ceph_opts);
647 if (rbdc) /* using an existing client */
648 ceph_destroy_options(ceph_opts);
650 rbdc = rbd_client_create(ceph_opts);
656 * Destroy ceph client
658 * Caller must hold rbd_client_list_lock.
660 static void rbd_client_release(struct kref *kref)
662 struct rbd_client *rbdc = container_of(kref, struct rbd_client, kref);
664 dout("%s: rbdc %p\n", __func__, rbdc);
665 spin_lock(&rbd_client_list_lock);
666 list_del(&rbdc->node);
667 spin_unlock(&rbd_client_list_lock);
669 ceph_destroy_client(rbdc->client);
674 * Drop reference to ceph client node. If it's not referenced anymore, release
677 static void rbd_put_client(struct rbd_client *rbdc)
680 kref_put(&rbdc->kref, rbd_client_release);
683 static bool rbd_image_format_valid(u32 image_format)
685 return image_format == 1 || image_format == 2;
688 static bool rbd_dev_ondisk_valid(struct rbd_image_header_ondisk *ondisk)
693 /* The header has to start with the magic rbd header text */
694 if (memcmp(&ondisk->text, RBD_HEADER_TEXT, sizeof (RBD_HEADER_TEXT)))
697 /* The bio layer requires at least sector-sized I/O */
699 if (ondisk->options.order < SECTOR_SHIFT)
702 /* If we use u64 in a few spots we may be able to loosen this */
704 if (ondisk->options.order > 8 * sizeof (int) - 1)
708 * The size of a snapshot header has to fit in a size_t, and
709 * that limits the number of snapshots.
711 snap_count = le32_to_cpu(ondisk->snap_count);
712 size = SIZE_MAX - sizeof (struct ceph_snap_context);
713 if (snap_count > size / sizeof (__le64))
717 * Not only that, but the size of the entire the snapshot
718 * header must also be representable in a size_t.
720 size -= snap_count * sizeof (__le64);
721 if ((u64) size < le64_to_cpu(ondisk->snap_names_len))
728 * Create a new header structure, translate header format from the on-disk
731 static int rbd_header_from_disk(struct rbd_image_header *header,
732 struct rbd_image_header_ondisk *ondisk)
739 memset(header, 0, sizeof (*header));
741 snap_count = le32_to_cpu(ondisk->snap_count);
743 len = strnlen(ondisk->object_prefix, sizeof (ondisk->object_prefix));
744 header->object_prefix = kmalloc(len + 1, GFP_KERNEL);
745 if (!header->object_prefix)
747 memcpy(header->object_prefix, ondisk->object_prefix, len);
748 header->object_prefix[len] = '\0';
751 u64 snap_names_len = le64_to_cpu(ondisk->snap_names_len);
753 /* Save a copy of the snapshot names */
755 if (snap_names_len > (u64) SIZE_MAX)
757 header->snap_names = kmalloc(snap_names_len, GFP_KERNEL);
758 if (!header->snap_names)
761 * Note that rbd_dev_v1_header_read() guarantees
762 * the ondisk buffer we're working with has
763 * snap_names_len bytes beyond the end of the
764 * snapshot id array, this memcpy() is safe.
766 memcpy(header->snap_names, &ondisk->snaps[snap_count],
769 /* Record each snapshot's size */
771 size = snap_count * sizeof (*header->snap_sizes);
772 header->snap_sizes = kmalloc(size, GFP_KERNEL);
773 if (!header->snap_sizes)
775 for (i = 0; i < snap_count; i++)
776 header->snap_sizes[i] =
777 le64_to_cpu(ondisk->snaps[i].image_size);
779 WARN_ON(ondisk->snap_names_len);
780 header->snap_names = NULL;
781 header->snap_sizes = NULL;
784 header->features = 0; /* No features support in v1 images */
785 header->obj_order = ondisk->options.order;
786 header->crypt_type = ondisk->options.crypt_type;
787 header->comp_type = ondisk->options.comp_type;
789 /* Allocate and fill in the snapshot context */
791 header->image_size = le64_to_cpu(ondisk->image_size);
792 size = sizeof (struct ceph_snap_context);
793 size += snap_count * sizeof (header->snapc->snaps[0]);
794 header->snapc = kzalloc(size, GFP_KERNEL);
798 atomic_set(&header->snapc->nref, 1);
799 header->snapc->seq = le64_to_cpu(ondisk->snap_seq);
800 header->snapc->num_snaps = snap_count;
801 for (i = 0; i < snap_count; i++)
802 header->snapc->snaps[i] =
803 le64_to_cpu(ondisk->snaps[i].id);
808 kfree(header->snap_sizes);
809 header->snap_sizes = NULL;
810 kfree(header->snap_names);
811 header->snap_names = NULL;
812 kfree(header->object_prefix);
813 header->object_prefix = NULL;
818 static const char *rbd_snap_name(struct rbd_device *rbd_dev, u64 snap_id)
820 struct rbd_snap *snap;
822 if (snap_id == CEPH_NOSNAP)
823 return RBD_SNAP_HEAD_NAME;
825 list_for_each_entry(snap, &rbd_dev->snaps, node)
826 if (snap_id == snap->id)
832 static int snap_by_name(struct rbd_device *rbd_dev, const char *snap_name)
835 struct rbd_snap *snap;
837 list_for_each_entry(snap, &rbd_dev->snaps, node) {
838 if (!strcmp(snap_name, snap->name)) {
839 rbd_dev->spec->snap_id = snap->id;
840 rbd_dev->mapping.size = snap->size;
841 rbd_dev->mapping.features = snap->features;
850 static int rbd_dev_set_mapping(struct rbd_device *rbd_dev)
854 if (!memcmp(rbd_dev->spec->snap_name, RBD_SNAP_HEAD_NAME,
855 sizeof (RBD_SNAP_HEAD_NAME))) {
856 rbd_dev->spec->snap_id = CEPH_NOSNAP;
857 rbd_dev->mapping.size = rbd_dev->header.image_size;
858 rbd_dev->mapping.features = rbd_dev->header.features;
861 ret = snap_by_name(rbd_dev, rbd_dev->spec->snap_name);
864 rbd_dev->mapping.read_only = true;
866 set_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags);
872 static void rbd_header_free(struct rbd_image_header *header)
874 kfree(header->object_prefix);
875 header->object_prefix = NULL;
876 kfree(header->snap_sizes);
877 header->snap_sizes = NULL;
878 kfree(header->snap_names);
879 header->snap_names = NULL;
880 ceph_put_snap_context(header->snapc);
881 header->snapc = NULL;
884 static const char *rbd_segment_name(struct rbd_device *rbd_dev, u64 offset)
890 name = kmalloc(MAX_OBJ_NAME_SIZE + 1, GFP_NOIO);
893 segment = offset >> rbd_dev->header.obj_order;
894 ret = snprintf(name, MAX_OBJ_NAME_SIZE + 1, "%s.%012llx",
895 rbd_dev->header.object_prefix, segment);
896 if (ret < 0 || ret > MAX_OBJ_NAME_SIZE) {
897 pr_err("error formatting segment name for #%llu (%d)\n",
906 static u64 rbd_segment_offset(struct rbd_device *rbd_dev, u64 offset)
908 u64 segment_size = (u64) 1 << rbd_dev->header.obj_order;
910 return offset & (segment_size - 1);
913 static u64 rbd_segment_length(struct rbd_device *rbd_dev,
914 u64 offset, u64 length)
916 u64 segment_size = (u64) 1 << rbd_dev->header.obj_order;
918 offset &= segment_size - 1;
920 rbd_assert(length <= U64_MAX - offset);
921 if (offset + length > segment_size)
922 length = segment_size - offset;
928 * returns the size of an object in the image
930 static u64 rbd_obj_bytes(struct rbd_image_header *header)
932 return 1 << header->obj_order;
939 static void bio_chain_put(struct bio *chain)
945 chain = chain->bi_next;
951 * zeros a bio chain, starting at specific offset
953 static void zero_bio_chain(struct bio *chain, int start_ofs)
962 bio_for_each_segment(bv, chain, i) {
963 if (pos + bv->bv_len > start_ofs) {
964 int remainder = max(start_ofs - pos, 0);
965 buf = bvec_kmap_irq(bv, &flags);
966 memset(buf + remainder, 0,
967 bv->bv_len - remainder);
968 bvec_kunmap_irq(buf, &flags);
973 chain = chain->bi_next;
978 * similar to zero_bio_chain(), zeros data defined by a page array,
979 * starting at the given byte offset from the start of the array and
980 * continuing up to the given end offset. The pages array is
981 * assumed to be big enough to hold all bytes up to the end.
983 static void zero_pages(struct page **pages, u64 offset, u64 end)
985 struct page **page = &pages[offset >> PAGE_SHIFT];
987 rbd_assert(end > offset);
988 rbd_assert(end - offset <= (u64)SIZE_MAX);
989 while (offset < end) {
995 page_offset = (size_t)(offset & ~PAGE_MASK);
996 length = min(PAGE_SIZE - page_offset, (size_t)(end - offset));
997 local_irq_save(flags);
998 kaddr = kmap_atomic(*page);
999 memset(kaddr + page_offset, 0, length);
1000 kunmap_atomic(kaddr);
1001 local_irq_restore(flags);
1009 * Clone a portion of a bio, starting at the given byte offset
1010 * and continuing for the number of bytes indicated.
1012 static struct bio *bio_clone_range(struct bio *bio_src,
1013 unsigned int offset,
1021 unsigned short end_idx;
1022 unsigned short vcnt;
1025 /* Handle the easy case for the caller */
1027 if (!offset && len == bio_src->bi_size)
1028 return bio_clone(bio_src, gfpmask);
1030 if (WARN_ON_ONCE(!len))
1032 if (WARN_ON_ONCE(len > bio_src->bi_size))
1034 if (WARN_ON_ONCE(offset > bio_src->bi_size - len))
1037 /* Find first affected segment... */
1040 __bio_for_each_segment(bv, bio_src, idx, 0) {
1041 if (resid < bv->bv_len)
1043 resid -= bv->bv_len;
1047 /* ...and the last affected segment */
1050 __bio_for_each_segment(bv, bio_src, end_idx, idx) {
1051 if (resid <= bv->bv_len)
1053 resid -= bv->bv_len;
1055 vcnt = end_idx - idx + 1;
1057 /* Build the clone */
1059 bio = bio_alloc(gfpmask, (unsigned int) vcnt);
1061 return NULL; /* ENOMEM */
1063 bio->bi_bdev = bio_src->bi_bdev;
1064 bio->bi_sector = bio_src->bi_sector + (offset >> SECTOR_SHIFT);
1065 bio->bi_rw = bio_src->bi_rw;
1066 bio->bi_flags |= 1 << BIO_CLONED;
1069 * Copy over our part of the bio_vec, then update the first
1070 * and last (or only) entries.
1072 memcpy(&bio->bi_io_vec[0], &bio_src->bi_io_vec[idx],
1073 vcnt * sizeof (struct bio_vec));
1074 bio->bi_io_vec[0].bv_offset += voff;
1076 bio->bi_io_vec[0].bv_len -= voff;
1077 bio->bi_io_vec[vcnt - 1].bv_len = resid;
1079 bio->bi_io_vec[0].bv_len = len;
1082 bio->bi_vcnt = vcnt;
1090 * Clone a portion of a bio chain, starting at the given byte offset
1091 * into the first bio in the source chain and continuing for the
1092 * number of bytes indicated. The result is another bio chain of
1093 * exactly the given length, or a null pointer on error.
1095 * The bio_src and offset parameters are both in-out. On entry they
1096 * refer to the first source bio and the offset into that bio where
1097 * the start of data to be cloned is located.
1099 * On return, bio_src is updated to refer to the bio in the source
1100 * chain that contains first un-cloned byte, and *offset will
1101 * contain the offset of that byte within that bio.
1103 static struct bio *bio_chain_clone_range(struct bio **bio_src,
1104 unsigned int *offset,
1108 struct bio *bi = *bio_src;
1109 unsigned int off = *offset;
1110 struct bio *chain = NULL;
1113 /* Build up a chain of clone bios up to the limit */
1115 if (!bi || off >= bi->bi_size || !len)
1116 return NULL; /* Nothing to clone */
1120 unsigned int bi_size;
1124 rbd_warn(NULL, "bio_chain exhausted with %u left", len);
1125 goto out_err; /* EINVAL; ran out of bio's */
1127 bi_size = min_t(unsigned int, bi->bi_size - off, len);
1128 bio = bio_clone_range(bi, off, bi_size, gfpmask);
1130 goto out_err; /* ENOMEM */
1133 end = &bio->bi_next;
1136 if (off == bi->bi_size) {
1147 bio_chain_put(chain);
1153 * The default/initial value for all object request flags is 0. For
1154 * each flag, once its value is set to 1 it is never reset to 0
1157 static void obj_request_img_data_set(struct rbd_obj_request *obj_request)
1159 if (test_and_set_bit(OBJ_REQ_IMG_DATA, &obj_request->flags)) {
1160 struct rbd_device *rbd_dev;
1162 rbd_dev = obj_request->img_request->rbd_dev;
1163 rbd_warn(rbd_dev, "obj_request %p already marked img_data\n",
1168 static bool obj_request_img_data_test(struct rbd_obj_request *obj_request)
1171 return test_bit(OBJ_REQ_IMG_DATA, &obj_request->flags) != 0;
1174 static void obj_request_done_set(struct rbd_obj_request *obj_request)
1176 if (test_and_set_bit(OBJ_REQ_DONE, &obj_request->flags)) {
1177 struct rbd_device *rbd_dev = NULL;
1179 if (obj_request_img_data_test(obj_request))
1180 rbd_dev = obj_request->img_request->rbd_dev;
1181 rbd_warn(rbd_dev, "obj_request %p already marked done\n",
1186 static bool obj_request_done_test(struct rbd_obj_request *obj_request)
1189 return test_bit(OBJ_REQ_DONE, &obj_request->flags) != 0;
1193 * This sets the KNOWN flag after (possibly) setting the EXISTS
1194 * flag. The latter is set based on the "exists" value provided.
1196 * Note that for our purposes once an object exists it never goes
1197 * away again. It's possible that the response from two existence
1198 * checks are separated by the creation of the target object, and
1199 * the first ("doesn't exist") response arrives *after* the second
1200 * ("does exist"). In that case we ignore the second one.
1202 static void obj_request_existence_set(struct rbd_obj_request *obj_request,
1206 set_bit(OBJ_REQ_EXISTS, &obj_request->flags);
1207 set_bit(OBJ_REQ_KNOWN, &obj_request->flags);
1211 static bool obj_request_known_test(struct rbd_obj_request *obj_request)
1214 return test_bit(OBJ_REQ_KNOWN, &obj_request->flags) != 0;
1217 static bool obj_request_exists_test(struct rbd_obj_request *obj_request)
1220 return test_bit(OBJ_REQ_EXISTS, &obj_request->flags) != 0;
1223 static void rbd_obj_request_get(struct rbd_obj_request *obj_request)
1225 dout("%s: obj %p (was %d)\n", __func__, obj_request,
1226 atomic_read(&obj_request->kref.refcount));
1227 kref_get(&obj_request->kref);
1230 static void rbd_obj_request_destroy(struct kref *kref);
1231 static void rbd_obj_request_put(struct rbd_obj_request *obj_request)
1233 rbd_assert(obj_request != NULL);
1234 dout("%s: obj %p (was %d)\n", __func__, obj_request,
1235 atomic_read(&obj_request->kref.refcount));
1236 kref_put(&obj_request->kref, rbd_obj_request_destroy);
1239 static void rbd_img_request_get(struct rbd_img_request *img_request)
1241 dout("%s: img %p (was %d)\n", __func__, img_request,
1242 atomic_read(&img_request->kref.refcount));
1243 kref_get(&img_request->kref);
1246 static void rbd_img_request_destroy(struct kref *kref);
1247 static void rbd_img_request_put(struct rbd_img_request *img_request)
1249 rbd_assert(img_request != NULL);
1250 dout("%s: img %p (was %d)\n", __func__, img_request,
1251 atomic_read(&img_request->kref.refcount));
1252 kref_put(&img_request->kref, rbd_img_request_destroy);
1255 static inline void rbd_img_obj_request_add(struct rbd_img_request *img_request,
1256 struct rbd_obj_request *obj_request)
1258 rbd_assert(obj_request->img_request == NULL);
1260 /* Image request now owns object's original reference */
1261 obj_request->img_request = img_request;
1262 obj_request->which = img_request->obj_request_count;
1263 rbd_assert(!obj_request_img_data_test(obj_request));
1264 obj_request_img_data_set(obj_request);
1265 rbd_assert(obj_request->which != BAD_WHICH);
1266 img_request->obj_request_count++;
1267 list_add_tail(&obj_request->links, &img_request->obj_requests);
1268 dout("%s: img %p obj %p w=%u\n", __func__, img_request, obj_request,
1269 obj_request->which);
1272 static inline void rbd_img_obj_request_del(struct rbd_img_request *img_request,
1273 struct rbd_obj_request *obj_request)
1275 rbd_assert(obj_request->which != BAD_WHICH);
1277 dout("%s: img %p obj %p w=%u\n", __func__, img_request, obj_request,
1278 obj_request->which);
1279 list_del(&obj_request->links);
1280 rbd_assert(img_request->obj_request_count > 0);
1281 img_request->obj_request_count--;
1282 rbd_assert(obj_request->which == img_request->obj_request_count);
1283 obj_request->which = BAD_WHICH;
1284 rbd_assert(obj_request_img_data_test(obj_request));
1285 rbd_assert(obj_request->img_request == img_request);
1286 obj_request->img_request = NULL;
1287 obj_request->callback = NULL;
1288 rbd_obj_request_put(obj_request);
1291 static bool obj_request_type_valid(enum obj_request_type type)
1294 case OBJ_REQUEST_NODATA:
1295 case OBJ_REQUEST_BIO:
1296 case OBJ_REQUEST_PAGES:
1303 static int rbd_obj_request_submit(struct ceph_osd_client *osdc,
1304 struct rbd_obj_request *obj_request)
1306 dout("%s: osdc %p obj %p\n", __func__, osdc, obj_request);
1308 return ceph_osdc_start_request(osdc, obj_request->osd_req, false);
1311 static void rbd_img_request_complete(struct rbd_img_request *img_request)
1314 dout("%s: img %p\n", __func__, img_request);
1317 * If no error occurred, compute the aggregate transfer
1318 * count for the image request. We could instead use
1319 * atomic64_cmpxchg() to update it as each object request
1320 * completes; not clear which way is better off hand.
1322 if (!img_request->result) {
1323 struct rbd_obj_request *obj_request;
1326 for_each_obj_request(img_request, obj_request)
1327 xferred += obj_request->xferred;
1328 img_request->xferred = xferred;
1331 if (img_request->callback)
1332 img_request->callback(img_request);
1334 rbd_img_request_put(img_request);
1337 /* Caller is responsible for rbd_obj_request_destroy(obj_request) */
1339 static int rbd_obj_request_wait(struct rbd_obj_request *obj_request)
1341 dout("%s: obj %p\n", __func__, obj_request);
1343 return wait_for_completion_interruptible(&obj_request->completion);
1347 * The default/initial value for all image request flags is 0. Each
1348 * is conditionally set to 1 at image request initialization time
1349 * and currently never change thereafter.
1351 static void img_request_write_set(struct rbd_img_request *img_request)
1353 set_bit(IMG_REQ_WRITE, &img_request->flags);
1357 static bool img_request_write_test(struct rbd_img_request *img_request)
1360 return test_bit(IMG_REQ_WRITE, &img_request->flags) != 0;
1363 static void img_request_child_set(struct rbd_img_request *img_request)
1365 set_bit(IMG_REQ_CHILD, &img_request->flags);
1369 static bool img_request_child_test(struct rbd_img_request *img_request)
1372 return test_bit(IMG_REQ_CHILD, &img_request->flags) != 0;
1375 static void img_request_layered_set(struct rbd_img_request *img_request)
1377 set_bit(IMG_REQ_LAYERED, &img_request->flags);
1381 static bool img_request_layered_test(struct rbd_img_request *img_request)
1384 return test_bit(IMG_REQ_LAYERED, &img_request->flags) != 0;
1388 rbd_img_obj_request_read_callback(struct rbd_obj_request *obj_request)
1390 u64 xferred = obj_request->xferred;
1391 u64 length = obj_request->length;
1393 dout("%s: obj %p img %p result %d %llu/%llu\n", __func__,
1394 obj_request, obj_request->img_request, obj_request->result,
1397 * ENOENT means a hole in the image. We zero-fill the
1398 * entire length of the request. A short read also implies
1399 * zero-fill to the end of the request. Either way we
1400 * update the xferred count to indicate the whole request
1403 rbd_assert(obj_request->type != OBJ_REQUEST_NODATA);
1404 if (obj_request->result == -ENOENT) {
1405 if (obj_request->type == OBJ_REQUEST_BIO)
1406 zero_bio_chain(obj_request->bio_list, 0);
1408 zero_pages(obj_request->pages, 0, length);
1409 obj_request->result = 0;
1410 obj_request->xferred = length;
1411 } else if (xferred < length && !obj_request->result) {
1412 if (obj_request->type == OBJ_REQUEST_BIO)
1413 zero_bio_chain(obj_request->bio_list, xferred);
1415 zero_pages(obj_request->pages, xferred, length);
1416 obj_request->xferred = length;
1418 obj_request_done_set(obj_request);
1421 static void rbd_obj_request_complete(struct rbd_obj_request *obj_request)
1423 dout("%s: obj %p cb %p\n", __func__, obj_request,
1424 obj_request->callback);
1425 if (obj_request->callback)
1426 obj_request->callback(obj_request);
1428 complete_all(&obj_request->completion);
1431 static void rbd_osd_trivial_callback(struct rbd_obj_request *obj_request)
1433 dout("%s: obj %p\n", __func__, obj_request);
1434 obj_request_done_set(obj_request);
1437 static void rbd_osd_read_callback(struct rbd_obj_request *obj_request)
1439 struct rbd_img_request *img_request = NULL;
1440 struct rbd_device *rbd_dev = NULL;
1441 bool layered = false;
1443 if (obj_request_img_data_test(obj_request)) {
1444 img_request = obj_request->img_request;
1445 layered = img_request && img_request_layered_test(img_request);
1446 rbd_dev = img_request->rbd_dev;
1449 dout("%s: obj %p img %p result %d %llu/%llu\n", __func__,
1450 obj_request, img_request, obj_request->result,
1451 obj_request->xferred, obj_request->length);
1452 if (layered && obj_request->result == -ENOENT &&
1453 obj_request->img_offset < rbd_dev->parent_overlap)
1454 rbd_img_parent_read(obj_request);
1455 else if (img_request)
1456 rbd_img_obj_request_read_callback(obj_request);
1458 obj_request_done_set(obj_request);
1461 static void rbd_osd_write_callback(struct rbd_obj_request *obj_request)
1463 dout("%s: obj %p result %d %llu\n", __func__, obj_request,
1464 obj_request->result, obj_request->length);
1466 * There is no such thing as a successful short write. Set
1467 * it to our originally-requested length.
1469 obj_request->xferred = obj_request->length;
1470 obj_request_done_set(obj_request);
1474 * For a simple stat call there's nothing to do. We'll do more if
1475 * this is part of a write sequence for a layered image.
1477 static void rbd_osd_stat_callback(struct rbd_obj_request *obj_request)
1479 dout("%s: obj %p\n", __func__, obj_request);
1480 obj_request_done_set(obj_request);
1483 static void rbd_osd_req_callback(struct ceph_osd_request *osd_req,
1484 struct ceph_msg *msg)
1486 struct rbd_obj_request *obj_request = osd_req->r_priv;
1489 dout("%s: osd_req %p msg %p\n", __func__, osd_req, msg);
1490 rbd_assert(osd_req == obj_request->osd_req);
1491 if (obj_request_img_data_test(obj_request)) {
1492 rbd_assert(obj_request->img_request);
1493 rbd_assert(obj_request->which != BAD_WHICH);
1495 rbd_assert(obj_request->which == BAD_WHICH);
1498 if (osd_req->r_result < 0)
1499 obj_request->result = osd_req->r_result;
1500 obj_request->version = le64_to_cpu(osd_req->r_reassert_version.version);
1502 BUG_ON(osd_req->r_num_ops > 2);
1505 * We support a 64-bit length, but ultimately it has to be
1506 * passed to blk_end_request(), which takes an unsigned int.
1508 obj_request->xferred = osd_req->r_reply_op_len[0];
1509 rbd_assert(obj_request->xferred < (u64)UINT_MAX);
1510 opcode = osd_req->r_ops[0].op;
1512 case CEPH_OSD_OP_READ:
1513 rbd_osd_read_callback(obj_request);
1515 case CEPH_OSD_OP_WRITE:
1516 rbd_osd_write_callback(obj_request);
1518 case CEPH_OSD_OP_STAT:
1519 rbd_osd_stat_callback(obj_request);
1521 case CEPH_OSD_OP_CALL:
1522 case CEPH_OSD_OP_NOTIFY_ACK:
1523 case CEPH_OSD_OP_WATCH:
1524 rbd_osd_trivial_callback(obj_request);
1527 rbd_warn(NULL, "%s: unsupported op %hu\n",
1528 obj_request->object_name, (unsigned short) opcode);
1532 if (obj_request_done_test(obj_request))
1533 rbd_obj_request_complete(obj_request);
1536 static void rbd_osd_req_format_read(struct rbd_obj_request *obj_request)
1538 struct rbd_img_request *img_request = obj_request->img_request;
1539 struct ceph_osd_request *osd_req = obj_request->osd_req;
1542 rbd_assert(osd_req != NULL);
1544 snap_id = img_request ? img_request->snap_id : CEPH_NOSNAP;
1545 ceph_osdc_build_request(osd_req, obj_request->offset,
1546 NULL, snap_id, NULL);
1549 static void rbd_osd_req_format_write(struct rbd_obj_request *obj_request)
1551 struct rbd_img_request *img_request = obj_request->img_request;
1552 struct ceph_osd_request *osd_req = obj_request->osd_req;
1553 struct ceph_snap_context *snapc;
1554 struct timespec mtime = CURRENT_TIME;
1556 rbd_assert(osd_req != NULL);
1558 snapc = img_request ? img_request->snapc : NULL;
1559 ceph_osdc_build_request(osd_req, obj_request->offset,
1560 snapc, CEPH_NOSNAP, &mtime);
1563 static struct ceph_osd_request *rbd_osd_req_create(
1564 struct rbd_device *rbd_dev,
1566 struct rbd_obj_request *obj_request)
1568 struct ceph_snap_context *snapc = NULL;
1569 struct ceph_osd_client *osdc;
1570 struct ceph_osd_request *osd_req;
1572 if (obj_request_img_data_test(obj_request)) {
1573 struct rbd_img_request *img_request = obj_request->img_request;
1575 rbd_assert(write_request ==
1576 img_request_write_test(img_request));
1578 snapc = img_request->snapc;
1581 /* Allocate and initialize the request, for the single op */
1583 osdc = &rbd_dev->rbd_client->client->osdc;
1584 osd_req = ceph_osdc_alloc_request(osdc, snapc, 1, false, GFP_ATOMIC);
1586 return NULL; /* ENOMEM */
1589 osd_req->r_flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK;
1591 osd_req->r_flags = CEPH_OSD_FLAG_READ;
1593 osd_req->r_callback = rbd_osd_req_callback;
1594 osd_req->r_priv = obj_request;
1596 osd_req->r_oid_len = strlen(obj_request->object_name);
1597 rbd_assert(osd_req->r_oid_len < sizeof (osd_req->r_oid));
1598 memcpy(osd_req->r_oid, obj_request->object_name, osd_req->r_oid_len);
1600 osd_req->r_file_layout = rbd_dev->layout; /* struct */
1606 * Create a copyup osd request based on the information in the
1607 * object request supplied. A copyup request has two osd ops,
1608 * a copyup method call, and a "normal" write request.
1610 static struct ceph_osd_request *
1611 rbd_osd_req_create_copyup(struct rbd_obj_request *obj_request)
1613 struct rbd_img_request *img_request;
1614 struct ceph_snap_context *snapc;
1615 struct rbd_device *rbd_dev;
1616 struct ceph_osd_client *osdc;
1617 struct ceph_osd_request *osd_req;
1619 rbd_assert(obj_request_img_data_test(obj_request));
1620 img_request = obj_request->img_request;
1621 rbd_assert(img_request);
1622 rbd_assert(img_request_write_test(img_request));
1624 /* Allocate and initialize the request, for the two ops */
1626 snapc = img_request->snapc;
1627 rbd_dev = img_request->rbd_dev;
1628 osdc = &rbd_dev->rbd_client->client->osdc;
1629 osd_req = ceph_osdc_alloc_request(osdc, snapc, 2, false, GFP_ATOMIC);
1631 return NULL; /* ENOMEM */
1633 osd_req->r_flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK;
1634 osd_req->r_callback = rbd_osd_req_callback;
1635 osd_req->r_priv = obj_request;
1637 osd_req->r_oid_len = strlen(obj_request->object_name);
1638 rbd_assert(osd_req->r_oid_len < sizeof (osd_req->r_oid));
1639 memcpy(osd_req->r_oid, obj_request->object_name, osd_req->r_oid_len);
1641 osd_req->r_file_layout = rbd_dev->layout; /* struct */
1647 static void rbd_osd_req_destroy(struct ceph_osd_request *osd_req)
1649 ceph_osdc_put_request(osd_req);
1652 /* object_name is assumed to be a non-null pointer and NUL-terminated */
1654 static struct rbd_obj_request *rbd_obj_request_create(const char *object_name,
1655 u64 offset, u64 length,
1656 enum obj_request_type type)
1658 struct rbd_obj_request *obj_request;
1662 rbd_assert(obj_request_type_valid(type));
1664 size = strlen(object_name) + 1;
1665 obj_request = kzalloc(sizeof (*obj_request) + size, GFP_KERNEL);
1669 name = (char *)(obj_request + 1);
1670 obj_request->object_name = memcpy(name, object_name, size);
1671 obj_request->offset = offset;
1672 obj_request->length = length;
1673 obj_request->flags = 0;
1674 obj_request->which = BAD_WHICH;
1675 obj_request->type = type;
1676 INIT_LIST_HEAD(&obj_request->links);
1677 init_completion(&obj_request->completion);
1678 kref_init(&obj_request->kref);
1680 dout("%s: \"%s\" %llu/%llu %d -> obj %p\n", __func__, object_name,
1681 offset, length, (int)type, obj_request);
1686 static void rbd_obj_request_destroy(struct kref *kref)
1688 struct rbd_obj_request *obj_request;
1690 obj_request = container_of(kref, struct rbd_obj_request, kref);
1692 dout("%s: obj %p\n", __func__, obj_request);
1694 rbd_assert(obj_request->img_request == NULL);
1695 rbd_assert(obj_request->which == BAD_WHICH);
1697 if (obj_request->osd_req)
1698 rbd_osd_req_destroy(obj_request->osd_req);
1700 rbd_assert(obj_request_type_valid(obj_request->type));
1701 switch (obj_request->type) {
1702 case OBJ_REQUEST_NODATA:
1703 break; /* Nothing to do */
1704 case OBJ_REQUEST_BIO:
1705 if (obj_request->bio_list)
1706 bio_chain_put(obj_request->bio_list);
1708 case OBJ_REQUEST_PAGES:
1709 if (obj_request->pages)
1710 ceph_release_page_vector(obj_request->pages,
1711 obj_request->page_count);
1719 * Caller is responsible for filling in the list of object requests
1720 * that comprises the image request, and the Linux request pointer
1721 * (if there is one).
1723 static struct rbd_img_request *rbd_img_request_create(
1724 struct rbd_device *rbd_dev,
1725 u64 offset, u64 length,
1729 struct rbd_img_request *img_request;
1730 struct ceph_snap_context *snapc = NULL;
1732 img_request = kmalloc(sizeof (*img_request), GFP_ATOMIC);
1736 if (write_request) {
1737 down_read(&rbd_dev->header_rwsem);
1738 snapc = ceph_get_snap_context(rbd_dev->header.snapc);
1739 up_read(&rbd_dev->header_rwsem);
1740 if (WARN_ON(!snapc)) {
1742 return NULL; /* Shouldn't happen */
1747 img_request->rq = NULL;
1748 img_request->rbd_dev = rbd_dev;
1749 img_request->offset = offset;
1750 img_request->length = length;
1751 img_request->flags = 0;
1752 if (write_request) {
1753 img_request_write_set(img_request);
1754 img_request->snapc = snapc;
1756 img_request->snap_id = rbd_dev->spec->snap_id;
1759 img_request_child_set(img_request);
1760 if (rbd_dev->parent_spec)
1761 img_request_layered_set(img_request);
1762 spin_lock_init(&img_request->completion_lock);
1763 img_request->next_completion = 0;
1764 img_request->callback = NULL;
1765 img_request->result = 0;
1766 img_request->obj_request_count = 0;
1767 INIT_LIST_HEAD(&img_request->obj_requests);
1768 kref_init(&img_request->kref);
1770 rbd_img_request_get(img_request); /* Avoid a warning */
1771 rbd_img_request_put(img_request); /* TEMPORARY */
1773 dout("%s: rbd_dev %p %s %llu/%llu -> img %p\n", __func__, rbd_dev,
1774 write_request ? "write" : "read", offset, length,
1780 static void rbd_img_request_destroy(struct kref *kref)
1782 struct rbd_img_request *img_request;
1783 struct rbd_obj_request *obj_request;
1784 struct rbd_obj_request *next_obj_request;
1786 img_request = container_of(kref, struct rbd_img_request, kref);
1788 dout("%s: img %p\n", __func__, img_request);
1790 for_each_obj_request_safe(img_request, obj_request, next_obj_request)
1791 rbd_img_obj_request_del(img_request, obj_request);
1792 rbd_assert(img_request->obj_request_count == 0);
1794 if (img_request_write_test(img_request))
1795 ceph_put_snap_context(img_request->snapc);
1797 if (img_request_child_test(img_request))
1798 rbd_obj_request_put(img_request->obj_request);
1803 static bool rbd_img_obj_end_request(struct rbd_obj_request *obj_request)
1805 struct rbd_img_request *img_request;
1806 unsigned int xferred;
1810 rbd_assert(obj_request_img_data_test(obj_request));
1811 img_request = obj_request->img_request;
1813 rbd_assert(obj_request->xferred <= (u64)UINT_MAX);
1814 xferred = (unsigned int)obj_request->xferred;
1815 result = obj_request->result;
1817 struct rbd_device *rbd_dev = img_request->rbd_dev;
1819 rbd_warn(rbd_dev, "%s %llx at %llx (%llx)\n",
1820 img_request_write_test(img_request) ? "write" : "read",
1821 obj_request->length, obj_request->img_offset,
1822 obj_request->offset);
1823 rbd_warn(rbd_dev, " result %d xferred %x\n",
1825 if (!img_request->result)
1826 img_request->result = result;
1829 /* Image object requests don't own their page array */
1831 if (obj_request->type == OBJ_REQUEST_PAGES) {
1832 obj_request->pages = NULL;
1833 obj_request->page_count = 0;
1836 if (img_request_child_test(img_request)) {
1837 rbd_assert(img_request->obj_request != NULL);
1838 more = obj_request->which < img_request->obj_request_count - 1;
1840 rbd_assert(img_request->rq != NULL);
1841 more = blk_end_request(img_request->rq, result, xferred);
1847 static void rbd_img_obj_callback(struct rbd_obj_request *obj_request)
1849 struct rbd_img_request *img_request;
1850 u32 which = obj_request->which;
1853 rbd_assert(obj_request_img_data_test(obj_request));
1854 img_request = obj_request->img_request;
1856 dout("%s: img %p obj %p\n", __func__, img_request, obj_request);
1857 rbd_assert(img_request != NULL);
1858 rbd_assert(img_request->obj_request_count > 0);
1859 rbd_assert(which != BAD_WHICH);
1860 rbd_assert(which < img_request->obj_request_count);
1861 rbd_assert(which >= img_request->next_completion);
1863 spin_lock_irq(&img_request->completion_lock);
1864 if (which != img_request->next_completion)
1867 for_each_obj_request_from(img_request, obj_request) {
1869 rbd_assert(which < img_request->obj_request_count);
1871 if (!obj_request_done_test(obj_request))
1873 more = rbd_img_obj_end_request(obj_request);
1877 rbd_assert(more ^ (which == img_request->obj_request_count));
1878 img_request->next_completion = which;
1880 spin_unlock_irq(&img_request->completion_lock);
1883 rbd_img_request_complete(img_request);
1887 * Split up an image request into one or more object requests, each
1888 * to a different object. The "type" parameter indicates whether
1889 * "data_desc" is the pointer to the head of a list of bio
1890 * structures, or the base of a page array. In either case this
1891 * function assumes data_desc describes memory sufficient to hold
1892 * all data described by the image request.
1894 static int rbd_img_request_fill(struct rbd_img_request *img_request,
1895 enum obj_request_type type,
1898 struct rbd_device *rbd_dev = img_request->rbd_dev;
1899 struct rbd_obj_request *obj_request = NULL;
1900 struct rbd_obj_request *next_obj_request;
1901 bool write_request = img_request_write_test(img_request);
1902 struct bio *bio_list;
1903 unsigned int bio_offset = 0;
1904 struct page **pages;
1909 dout("%s: img %p type %d data_desc %p\n", __func__, img_request,
1910 (int)type, data_desc);
1912 opcode = write_request ? CEPH_OSD_OP_WRITE : CEPH_OSD_OP_READ;
1913 img_offset = img_request->offset;
1914 resid = img_request->length;
1915 rbd_assert(resid > 0);
1917 if (type == OBJ_REQUEST_BIO) {
1918 bio_list = data_desc;
1919 rbd_assert(img_offset == bio_list->bi_sector << SECTOR_SHIFT);
1921 rbd_assert(type == OBJ_REQUEST_PAGES);
1926 struct ceph_osd_request *osd_req;
1927 const char *object_name;
1931 object_name = rbd_segment_name(rbd_dev, img_offset);
1934 offset = rbd_segment_offset(rbd_dev, img_offset);
1935 length = rbd_segment_length(rbd_dev, img_offset, resid);
1936 obj_request = rbd_obj_request_create(object_name,
1937 offset, length, type);
1938 kfree(object_name); /* object request has its own copy */
1942 if (type == OBJ_REQUEST_BIO) {
1943 unsigned int clone_size;
1945 rbd_assert(length <= (u64)UINT_MAX);
1946 clone_size = (unsigned int)length;
1947 obj_request->bio_list =
1948 bio_chain_clone_range(&bio_list,
1952 if (!obj_request->bio_list)
1955 unsigned int page_count;
1957 obj_request->pages = pages;
1958 page_count = (u32)calc_pages_for(offset, length);
1959 obj_request->page_count = page_count;
1960 if ((offset + length) & ~PAGE_MASK)
1961 page_count--; /* more on last page */
1962 pages += page_count;
1965 osd_req = rbd_osd_req_create(rbd_dev, write_request,
1969 obj_request->osd_req = osd_req;
1970 obj_request->callback = rbd_img_obj_callback;
1972 osd_req_op_extent_init(osd_req, 0, opcode, offset, length,
1974 if (type == OBJ_REQUEST_BIO)
1975 osd_req_op_extent_osd_data_bio(osd_req, 0,
1976 obj_request->bio_list, length);
1978 osd_req_op_extent_osd_data_pages(osd_req, 0,
1979 obj_request->pages, length,
1980 offset & ~PAGE_MASK, false, false);
1983 rbd_osd_req_format_write(obj_request);
1985 rbd_osd_req_format_read(obj_request);
1987 obj_request->img_offset = img_offset;
1988 rbd_img_obj_request_add(img_request, obj_request);
1990 img_offset += length;
1997 rbd_obj_request_put(obj_request);
1999 for_each_obj_request_safe(img_request, obj_request, next_obj_request)
2000 rbd_obj_request_put(obj_request);
2006 rbd_img_obj_copyup_callback(struct rbd_obj_request *obj_request)
2008 struct rbd_img_request *img_request;
2009 struct rbd_device *rbd_dev;
2013 rbd_assert(obj_request->type == OBJ_REQUEST_BIO);
2014 rbd_assert(obj_request_img_data_test(obj_request));
2015 img_request = obj_request->img_request;
2016 rbd_assert(img_request);
2018 rbd_dev = img_request->rbd_dev;
2019 rbd_assert(rbd_dev);
2020 length = (u64)1 << rbd_dev->header.obj_order;
2021 page_count = (u32)calc_pages_for(0, length);
2023 rbd_assert(obj_request->copyup_pages);
2024 ceph_release_page_vector(obj_request->copyup_pages, page_count);
2025 obj_request->copyup_pages = NULL;
2028 * We want the transfer count to reflect the size of the
2029 * original write request. There is no such thing as a
2030 * successful short write, so if the request was successful
2031 * we can just set it to the originally-requested length.
2033 if (!obj_request->result)
2034 obj_request->xferred = obj_request->length;
2036 /* Finish up with the normal image object callback */
2038 rbd_img_obj_callback(obj_request);
2042 rbd_img_obj_parent_read_full_callback(struct rbd_img_request *img_request)
2044 struct rbd_obj_request *orig_request;
2045 struct ceph_osd_request *osd_req;
2046 struct ceph_osd_client *osdc;
2047 struct rbd_device *rbd_dev;
2048 struct page **pages;
2053 rbd_assert(img_request_child_test(img_request));
2055 /* First get what we need from the image request */
2057 pages = img_request->copyup_pages;
2058 rbd_assert(pages != NULL);
2059 img_request->copyup_pages = NULL;
2061 orig_request = img_request->obj_request;
2062 rbd_assert(orig_request != NULL);
2063 rbd_assert(orig_request->type == OBJ_REQUEST_BIO);
2064 result = img_request->result;
2065 obj_size = img_request->length;
2066 xferred = img_request->xferred;
2068 rbd_dev = img_request->rbd_dev;
2069 rbd_assert(rbd_dev);
2070 rbd_assert(obj_size == (u64)1 << rbd_dev->header.obj_order);
2072 rbd_img_request_put(img_request);
2077 /* Allocate the new copyup osd request for the original request */
2080 rbd_assert(!orig_request->osd_req);
2081 osd_req = rbd_osd_req_create_copyup(orig_request);
2084 orig_request->osd_req = osd_req;
2085 orig_request->copyup_pages = pages;
2087 /* Initialize the copyup op */
2089 osd_req_op_cls_init(osd_req, 0, CEPH_OSD_OP_CALL, "rbd", "copyup");
2090 osd_req_op_cls_request_data_pages(osd_req, 0, pages, obj_size, 0,
2093 /* Then the original write request op */
2095 osd_req_op_extent_init(osd_req, 1, CEPH_OSD_OP_WRITE,
2096 orig_request->offset,
2097 orig_request->length, 0, 0);
2098 osd_req_op_extent_osd_data_bio(osd_req, 1, orig_request->bio_list,
2099 orig_request->length);
2101 rbd_osd_req_format_write(orig_request);
2103 /* All set, send it off. */
2105 orig_request->callback = rbd_img_obj_copyup_callback;
2106 osdc = &rbd_dev->rbd_client->client->osdc;
2107 result = rbd_obj_request_submit(osdc, orig_request);
2111 /* Record the error code and complete the request */
2113 orig_request->result = result;
2114 orig_request->xferred = 0;
2115 obj_request_done_set(orig_request);
2116 rbd_obj_request_complete(orig_request);
2120 * Read from the parent image the range of data that covers the
2121 * entire target of the given object request. This is used for
2122 * satisfying a layered image write request when the target of an
2123 * object request from the image request does not exist.
2125 * A page array big enough to hold the returned data is allocated
2126 * and supplied to rbd_img_request_fill() as the "data descriptor."
2127 * When the read completes, this page array will be transferred to
2128 * the original object request for the copyup operation.
2130 * If an error occurs, record it as the result of the original
2131 * object request and mark it done so it gets completed.
2133 static int rbd_img_obj_parent_read_full(struct rbd_obj_request *obj_request)
2135 struct rbd_img_request *img_request = NULL;
2136 struct rbd_img_request *parent_request = NULL;
2137 struct rbd_device *rbd_dev;
2140 struct page **pages = NULL;
2144 rbd_assert(obj_request_img_data_test(obj_request));
2145 rbd_assert(obj_request->type == OBJ_REQUEST_BIO);
2147 img_request = obj_request->img_request;
2148 rbd_assert(img_request != NULL);
2149 rbd_dev = img_request->rbd_dev;
2150 rbd_assert(rbd_dev->parent != NULL);
2153 * First things first. The original osd request is of no
2154 * use to use any more, we'll need a new one that can hold
2155 * the two ops in a copyup request. We'll get that later,
2156 * but for now we can release the old one.
2158 rbd_osd_req_destroy(obj_request->osd_req);
2159 obj_request->osd_req = NULL;
2162 * Determine the byte range covered by the object in the
2163 * child image to which the original request was to be sent.
2165 img_offset = obj_request->img_offset - obj_request->offset;
2166 length = (u64)1 << rbd_dev->header.obj_order;
2169 * There is no defined parent data beyond the parent
2170 * overlap, so limit what we read at that boundary if
2173 if (img_offset + length > rbd_dev->parent_overlap) {
2174 rbd_assert(img_offset < rbd_dev->parent_overlap);
2175 length = rbd_dev->parent_overlap - img_offset;
2179 * Allocate a page array big enough to receive the data read
2182 page_count = (u32)calc_pages_for(0, length);
2183 pages = ceph_alloc_page_vector(page_count, GFP_KERNEL);
2184 if (IS_ERR(pages)) {
2185 result = PTR_ERR(pages);
2191 parent_request = rbd_img_request_create(rbd_dev->parent,
2194 if (!parent_request)
2196 rbd_obj_request_get(obj_request);
2197 parent_request->obj_request = obj_request;
2199 result = rbd_img_request_fill(parent_request, OBJ_REQUEST_PAGES, pages);
2202 parent_request->copyup_pages = pages;
2204 parent_request->callback = rbd_img_obj_parent_read_full_callback;
2205 result = rbd_img_request_submit(parent_request);
2209 parent_request->copyup_pages = NULL;
2210 parent_request->obj_request = NULL;
2211 rbd_obj_request_put(obj_request);
2214 ceph_release_page_vector(pages, page_count);
2216 rbd_img_request_put(parent_request);
2217 obj_request->result = result;
2218 obj_request->xferred = 0;
2219 obj_request_done_set(obj_request);
2224 static void rbd_img_obj_exists_callback(struct rbd_obj_request *obj_request)
2226 struct rbd_obj_request *orig_request;
2229 rbd_assert(!obj_request_img_data_test(obj_request));
2232 * All we need from the object request is the original
2233 * request and the result of the STAT op. Grab those, then
2234 * we're done with the request.
2236 orig_request = obj_request->obj_request;
2237 obj_request->obj_request = NULL;
2238 rbd_assert(orig_request);
2239 rbd_assert(orig_request->img_request);
2241 result = obj_request->result;
2242 obj_request->result = 0;
2244 dout("%s: obj %p for obj %p result %d %llu/%llu\n", __func__,
2245 obj_request, orig_request, result,
2246 obj_request->xferred, obj_request->length);
2247 rbd_obj_request_put(obj_request);
2249 rbd_assert(orig_request);
2250 rbd_assert(orig_request->img_request);
2253 * Our only purpose here is to determine whether the object
2254 * exists, and we don't want to treat the non-existence as
2255 * an error. If something else comes back, transfer the
2256 * error to the original request and complete it now.
2259 obj_request_existence_set(orig_request, true);
2260 } else if (result == -ENOENT) {
2261 obj_request_existence_set(orig_request, false);
2262 } else if (result) {
2263 orig_request->result = result;
2268 * Resubmit the original request now that we have recorded
2269 * whether the target object exists.
2271 orig_request->result = rbd_img_obj_request_submit(orig_request);
2273 if (orig_request->result)
2274 rbd_obj_request_complete(orig_request);
2275 rbd_obj_request_put(orig_request);
2278 static int rbd_img_obj_exists_submit(struct rbd_obj_request *obj_request)
2280 struct rbd_obj_request *stat_request;
2281 struct rbd_device *rbd_dev;
2282 struct ceph_osd_client *osdc;
2283 struct page **pages = NULL;
2289 * The response data for a STAT call consists of:
2296 size = sizeof (__le64) + sizeof (__le32) + sizeof (__le32);
2297 page_count = (u32)calc_pages_for(0, size);
2298 pages = ceph_alloc_page_vector(page_count, GFP_KERNEL);
2300 return PTR_ERR(pages);
2303 stat_request = rbd_obj_request_create(obj_request->object_name, 0, 0,
2308 rbd_obj_request_get(obj_request);
2309 stat_request->obj_request = obj_request;
2310 stat_request->pages = pages;
2311 stat_request->page_count = page_count;
2313 rbd_assert(obj_request->img_request);
2314 rbd_dev = obj_request->img_request->rbd_dev;
2315 stat_request->osd_req = rbd_osd_req_create(rbd_dev, false,
2317 if (!stat_request->osd_req)
2319 stat_request->callback = rbd_img_obj_exists_callback;
2321 osd_req_op_init(stat_request->osd_req, 0, CEPH_OSD_OP_STAT);
2322 osd_req_op_raw_data_in_pages(stat_request->osd_req, 0, pages, size, 0,
2324 rbd_osd_req_format_read(stat_request);
2326 osdc = &rbd_dev->rbd_client->client->osdc;
2327 ret = rbd_obj_request_submit(osdc, stat_request);
2330 rbd_obj_request_put(obj_request);
2335 static int rbd_img_obj_request_submit(struct rbd_obj_request *obj_request)
2337 struct rbd_img_request *img_request;
2338 struct rbd_device *rbd_dev;
2341 rbd_assert(obj_request_img_data_test(obj_request));
2343 img_request = obj_request->img_request;
2344 rbd_assert(img_request);
2345 rbd_dev = img_request->rbd_dev;
2348 * Only writes to layered images need special handling.
2349 * Reads and non-layered writes are simple object requests.
2350 * Layered writes that start beyond the end of the overlap
2351 * with the parent have no parent data, so they too are
2352 * simple object requests. Finally, if the target object is
2353 * known to already exist, its parent data has already been
2354 * copied, so a write to the object can also be handled as a
2355 * simple object request.
2357 if (!img_request_write_test(img_request) ||
2358 !img_request_layered_test(img_request) ||
2359 rbd_dev->parent_overlap <= obj_request->img_offset ||
2360 ((known = obj_request_known_test(obj_request)) &&
2361 obj_request_exists_test(obj_request))) {
2363 struct rbd_device *rbd_dev;
2364 struct ceph_osd_client *osdc;
2366 rbd_dev = obj_request->img_request->rbd_dev;
2367 osdc = &rbd_dev->rbd_client->client->osdc;
2369 return rbd_obj_request_submit(osdc, obj_request);
2373 * It's a layered write. The target object might exist but
2374 * we may not know that yet. If we know it doesn't exist,
2375 * start by reading the data for the full target object from
2376 * the parent so we can use it for a copyup to the target.
2379 return rbd_img_obj_parent_read_full(obj_request);
2381 /* We don't know whether the target exists. Go find out. */
2383 return rbd_img_obj_exists_submit(obj_request);
2386 static int rbd_img_request_submit(struct rbd_img_request *img_request)
2388 struct rbd_obj_request *obj_request;
2389 struct rbd_obj_request *next_obj_request;
2391 dout("%s: img %p\n", __func__, img_request);
2392 for_each_obj_request_safe(img_request, obj_request, next_obj_request) {
2395 ret = rbd_img_obj_request_submit(obj_request);
2403 static void rbd_img_parent_read_callback(struct rbd_img_request *img_request)
2405 struct rbd_obj_request *obj_request;
2406 struct rbd_device *rbd_dev;
2409 rbd_assert(img_request_child_test(img_request));
2411 obj_request = img_request->obj_request;
2412 rbd_assert(obj_request);
2413 rbd_assert(obj_request->img_request);
2415 obj_request->result = img_request->result;
2416 if (obj_request->result)
2420 * We need to zero anything beyond the parent overlap
2421 * boundary. Since rbd_img_obj_request_read_callback()
2422 * will zero anything beyond the end of a short read, an
2423 * easy way to do this is to pretend the data from the
2424 * parent came up short--ending at the overlap boundary.
2426 rbd_assert(obj_request->img_offset < U64_MAX - obj_request->length);
2427 obj_end = obj_request->img_offset + obj_request->length;
2428 rbd_dev = obj_request->img_request->rbd_dev;
2429 if (obj_end > rbd_dev->parent_overlap) {
2432 if (obj_request->img_offset < rbd_dev->parent_overlap)
2433 xferred = rbd_dev->parent_overlap -
2434 obj_request->img_offset;
2436 obj_request->xferred = min(img_request->xferred, xferred);
2438 obj_request->xferred = img_request->xferred;
2441 rbd_img_obj_request_read_callback(obj_request);
2442 rbd_obj_request_complete(obj_request);
2445 static void rbd_img_parent_read(struct rbd_obj_request *obj_request)
2447 struct rbd_device *rbd_dev;
2448 struct rbd_img_request *img_request;
2451 rbd_assert(obj_request_img_data_test(obj_request));
2452 rbd_assert(obj_request->img_request != NULL);
2453 rbd_assert(obj_request->result == (s32) -ENOENT);
2454 rbd_assert(obj_request->type == OBJ_REQUEST_BIO);
2456 rbd_dev = obj_request->img_request->rbd_dev;
2457 rbd_assert(rbd_dev->parent != NULL);
2458 /* rbd_read_finish(obj_request, obj_request->length); */
2459 img_request = rbd_img_request_create(rbd_dev->parent,
2460 obj_request->img_offset,
2461 obj_request->length,
2467 rbd_obj_request_get(obj_request);
2468 img_request->obj_request = obj_request;
2470 result = rbd_img_request_fill(img_request, OBJ_REQUEST_BIO,
2471 obj_request->bio_list);
2475 img_request->callback = rbd_img_parent_read_callback;
2476 result = rbd_img_request_submit(img_request);
2483 rbd_img_request_put(img_request);
2484 obj_request->result = result;
2485 obj_request->xferred = 0;
2486 obj_request_done_set(obj_request);
2489 static int rbd_obj_notify_ack(struct rbd_device *rbd_dev,
2490 u64 ver, u64 notify_id)
2492 struct rbd_obj_request *obj_request;
2493 struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
2496 obj_request = rbd_obj_request_create(rbd_dev->header_name, 0, 0,
2497 OBJ_REQUEST_NODATA);
2502 obj_request->osd_req = rbd_osd_req_create(rbd_dev, false, obj_request);
2503 if (!obj_request->osd_req)
2505 obj_request->callback = rbd_obj_request_put;
2507 osd_req_op_watch_init(obj_request->osd_req, 0, CEPH_OSD_OP_NOTIFY_ACK,
2509 rbd_osd_req_format_read(obj_request);
2511 ret = rbd_obj_request_submit(osdc, obj_request);
2514 rbd_obj_request_put(obj_request);
2519 static void rbd_watch_cb(u64 ver, u64 notify_id, u8 opcode, void *data)
2521 struct rbd_device *rbd_dev = (struct rbd_device *)data;
2528 dout("%s: \"%s\" notify_id %llu opcode %u\n", __func__,
2529 rbd_dev->header_name, (unsigned long long) notify_id,
2530 (unsigned int) opcode);
2531 rc = rbd_dev_refresh(rbd_dev, &hver);
2533 rbd_warn(rbd_dev, "got notification but failed to "
2534 " update snaps: %d\n", rc);
2536 rbd_obj_notify_ack(rbd_dev, hver, notify_id);
2540 * Request sync osd watch/unwatch. The value of "start" determines
2541 * whether a watch request is being initiated or torn down.
2543 static int rbd_dev_header_watch_sync(struct rbd_device *rbd_dev, int start)
2545 struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
2546 struct rbd_obj_request *obj_request;
2549 rbd_assert(start ^ !!rbd_dev->watch_event);
2550 rbd_assert(start ^ !!rbd_dev->watch_request);
2553 ret = ceph_osdc_create_event(osdc, rbd_watch_cb, rbd_dev,
2554 &rbd_dev->watch_event);
2557 rbd_assert(rbd_dev->watch_event != NULL);
2561 obj_request = rbd_obj_request_create(rbd_dev->header_name, 0, 0,
2562 OBJ_REQUEST_NODATA);
2566 obj_request->osd_req = rbd_osd_req_create(rbd_dev, true, obj_request);
2567 if (!obj_request->osd_req)
2571 ceph_osdc_set_request_linger(osdc, obj_request->osd_req);
2573 ceph_osdc_unregister_linger_request(osdc,
2574 rbd_dev->watch_request->osd_req);
2576 osd_req_op_watch_init(obj_request->osd_req, 0, CEPH_OSD_OP_WATCH,
2577 rbd_dev->watch_event->cookie,
2578 rbd_dev->header.obj_version, start);
2579 rbd_osd_req_format_write(obj_request);
2581 ret = rbd_obj_request_submit(osdc, obj_request);
2584 ret = rbd_obj_request_wait(obj_request);
2587 ret = obj_request->result;
2592 * A watch request is set to linger, so the underlying osd
2593 * request won't go away until we unregister it. We retain
2594 * a pointer to the object request during that time (in
2595 * rbd_dev->watch_request), so we'll keep a reference to
2596 * it. We'll drop that reference (below) after we've
2600 rbd_dev->watch_request = obj_request;
2605 /* We have successfully torn down the watch request */
2607 rbd_obj_request_put(rbd_dev->watch_request);
2608 rbd_dev->watch_request = NULL;
2610 /* Cancel the event if we're tearing down, or on error */
2611 ceph_osdc_cancel_event(rbd_dev->watch_event);
2612 rbd_dev->watch_event = NULL;
2614 rbd_obj_request_put(obj_request);
2620 * Synchronous osd object method call
2622 static int rbd_obj_method_sync(struct rbd_device *rbd_dev,
2623 const char *object_name,
2624 const char *class_name,
2625 const char *method_name,
2626 const void *outbound,
2627 size_t outbound_size,
2629 size_t inbound_size,
2632 struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
2633 struct rbd_obj_request *obj_request;
2634 struct page **pages;
2639 * Method calls are ultimately read operations. The result
2640 * should placed into the inbound buffer provided. They
2641 * also supply outbound data--parameters for the object
2642 * method. Currently if this is present it will be a
2645 page_count = (u32) calc_pages_for(0, inbound_size);
2646 pages = ceph_alloc_page_vector(page_count, GFP_KERNEL);
2648 return PTR_ERR(pages);
2651 obj_request = rbd_obj_request_create(object_name, 0, inbound_size,
2656 obj_request->pages = pages;
2657 obj_request->page_count = page_count;
2659 obj_request->osd_req = rbd_osd_req_create(rbd_dev, false, obj_request);
2660 if (!obj_request->osd_req)
2663 osd_req_op_cls_init(obj_request->osd_req, 0, CEPH_OSD_OP_CALL,
2664 class_name, method_name);
2665 if (outbound_size) {
2666 struct ceph_pagelist *pagelist;
2668 pagelist = kmalloc(sizeof (*pagelist), GFP_NOFS);
2672 ceph_pagelist_init(pagelist);
2673 ceph_pagelist_append(pagelist, outbound, outbound_size);
2674 osd_req_op_cls_request_data_pagelist(obj_request->osd_req, 0,
2677 osd_req_op_cls_response_data_pages(obj_request->osd_req, 0,
2678 obj_request->pages, inbound_size,
2680 rbd_osd_req_format_read(obj_request);
2682 ret = rbd_obj_request_submit(osdc, obj_request);
2685 ret = rbd_obj_request_wait(obj_request);
2689 ret = obj_request->result;
2693 ceph_copy_from_page_vector(pages, inbound, 0, obj_request->xferred);
2695 *version = obj_request->version;
2698 rbd_obj_request_put(obj_request);
2700 ceph_release_page_vector(pages, page_count);
2705 static void rbd_request_fn(struct request_queue *q)
2706 __releases(q->queue_lock) __acquires(q->queue_lock)
2708 struct rbd_device *rbd_dev = q->queuedata;
2709 bool read_only = rbd_dev->mapping.read_only;
2713 while ((rq = blk_fetch_request(q))) {
2714 bool write_request = rq_data_dir(rq) == WRITE;
2715 struct rbd_img_request *img_request;
2719 /* Ignore any non-FS requests that filter through. */
2721 if (rq->cmd_type != REQ_TYPE_FS) {
2722 dout("%s: non-fs request type %d\n", __func__,
2723 (int) rq->cmd_type);
2724 __blk_end_request_all(rq, 0);
2728 /* Ignore/skip any zero-length requests */
2730 offset = (u64) blk_rq_pos(rq) << SECTOR_SHIFT;
2731 length = (u64) blk_rq_bytes(rq);
2734 dout("%s: zero-length request\n", __func__);
2735 __blk_end_request_all(rq, 0);
2739 spin_unlock_irq(q->queue_lock);
2741 /* Disallow writes to a read-only device */
2743 if (write_request) {
2747 rbd_assert(rbd_dev->spec->snap_id == CEPH_NOSNAP);
2751 * Quit early if the mapped snapshot no longer
2752 * exists. It's still possible the snapshot will
2753 * have disappeared by the time our request arrives
2754 * at the osd, but there's no sense in sending it if
2757 if (!test_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags)) {
2758 dout("request for non-existent snapshot");
2759 rbd_assert(rbd_dev->spec->snap_id != CEPH_NOSNAP);
2765 if (WARN_ON(offset && length > U64_MAX - offset + 1))
2766 goto end_request; /* Shouldn't happen */
2769 img_request = rbd_img_request_create(rbd_dev, offset, length,
2770 write_request, false);
2774 img_request->rq = rq;
2776 result = rbd_img_request_fill(img_request, OBJ_REQUEST_BIO,
2779 result = rbd_img_request_submit(img_request);
2781 rbd_img_request_put(img_request);
2783 spin_lock_irq(q->queue_lock);
2785 rbd_warn(rbd_dev, "%s %llx at %llx result %d\n",
2786 write_request ? "write" : "read",
2787 length, offset, result);
2789 __blk_end_request_all(rq, result);
2795 * a queue callback. Makes sure that we don't create a bio that spans across
2796 * multiple osd objects. One exception would be with a single page bios,
2797 * which we handle later at bio_chain_clone_range()
2799 static int rbd_merge_bvec(struct request_queue *q, struct bvec_merge_data *bmd,
2800 struct bio_vec *bvec)
2802 struct rbd_device *rbd_dev = q->queuedata;
2803 sector_t sector_offset;
2804 sector_t sectors_per_obj;
2805 sector_t obj_sector_offset;
2809 * Find how far into its rbd object the partition-relative
2810 * bio start sector is to offset relative to the enclosing
2813 sector_offset = get_start_sect(bmd->bi_bdev) + bmd->bi_sector;
2814 sectors_per_obj = 1 << (rbd_dev->header.obj_order - SECTOR_SHIFT);
2815 obj_sector_offset = sector_offset & (sectors_per_obj - 1);
2818 * Compute the number of bytes from that offset to the end
2819 * of the object. Account for what's already used by the bio.
2821 ret = (int) (sectors_per_obj - obj_sector_offset) << SECTOR_SHIFT;
2822 if (ret > bmd->bi_size)
2823 ret -= bmd->bi_size;
2828 * Don't send back more than was asked for. And if the bio
2829 * was empty, let the whole thing through because: "Note
2830 * that a block device *must* allow a single page to be
2831 * added to an empty bio."
2833 rbd_assert(bvec->bv_len <= PAGE_SIZE);
2834 if (ret > (int) bvec->bv_len || !bmd->bi_size)
2835 ret = (int) bvec->bv_len;
2840 static void rbd_free_disk(struct rbd_device *rbd_dev)
2842 struct gendisk *disk = rbd_dev->disk;
2847 if (disk->flags & GENHD_FL_UP)
2850 blk_cleanup_queue(disk->queue);
2854 static int rbd_obj_read_sync(struct rbd_device *rbd_dev,
2855 const char *object_name,
2856 u64 offset, u64 length,
2857 void *buf, u64 *version)
2860 struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
2861 struct rbd_obj_request *obj_request;
2862 struct page **pages = NULL;
2867 page_count = (u32) calc_pages_for(offset, length);
2868 pages = ceph_alloc_page_vector(page_count, GFP_KERNEL);
2870 ret = PTR_ERR(pages);
2873 obj_request = rbd_obj_request_create(object_name, offset, length,
2878 obj_request->pages = pages;
2879 obj_request->page_count = page_count;
2881 obj_request->osd_req = rbd_osd_req_create(rbd_dev, false, obj_request);
2882 if (!obj_request->osd_req)
2885 osd_req_op_extent_init(obj_request->osd_req, 0, CEPH_OSD_OP_READ,
2886 offset, length, 0, 0);
2887 osd_req_op_extent_osd_data_pages(obj_request->osd_req, 0,
2889 obj_request->length,
2890 obj_request->offset & ~PAGE_MASK,
2892 rbd_osd_req_format_read(obj_request);
2894 ret = rbd_obj_request_submit(osdc, obj_request);
2897 ret = rbd_obj_request_wait(obj_request);
2901 ret = obj_request->result;
2905 rbd_assert(obj_request->xferred <= (u64) SIZE_MAX);
2906 size = (size_t) obj_request->xferred;
2907 ceph_copy_from_page_vector(pages, buf, 0, size);
2908 rbd_assert(size <= (size_t) INT_MAX);
2911 *version = obj_request->version;
2914 rbd_obj_request_put(obj_request);
2916 ceph_release_page_vector(pages, page_count);
2922 * Read the complete header for the given rbd device.
2924 * Returns a pointer to a dynamically-allocated buffer containing
2925 * the complete and validated header. Caller can pass the address
2926 * of a variable that will be filled in with the version of the
2927 * header object at the time it was read.
2929 * Returns a pointer-coded errno if a failure occurs.
2931 static struct rbd_image_header_ondisk *
2932 rbd_dev_v1_header_read(struct rbd_device *rbd_dev, u64 *version)
2934 struct rbd_image_header_ondisk *ondisk = NULL;
2941 * The complete header will include an array of its 64-bit
2942 * snapshot ids, followed by the names of those snapshots as
2943 * a contiguous block of NUL-terminated strings. Note that
2944 * the number of snapshots could change by the time we read
2945 * it in, in which case we re-read it.
2952 size = sizeof (*ondisk);
2953 size += snap_count * sizeof (struct rbd_image_snap_ondisk);
2955 ondisk = kmalloc(size, GFP_KERNEL);
2957 return ERR_PTR(-ENOMEM);
2959 ret = rbd_obj_read_sync(rbd_dev, rbd_dev->header_name,
2960 0, size, ondisk, version);
2963 if (WARN_ON((size_t) ret < size)) {
2965 rbd_warn(rbd_dev, "short header read (want %zd got %d)",
2969 if (!rbd_dev_ondisk_valid(ondisk)) {
2971 rbd_warn(rbd_dev, "invalid header");
2975 names_size = le64_to_cpu(ondisk->snap_names_len);
2976 want_count = snap_count;
2977 snap_count = le32_to_cpu(ondisk->snap_count);
2978 } while (snap_count != want_count);
2985 return ERR_PTR(ret);
2989 * reload the ondisk the header
2991 static int rbd_read_header(struct rbd_device *rbd_dev,
2992 struct rbd_image_header *header)
2994 struct rbd_image_header_ondisk *ondisk;
2998 ondisk = rbd_dev_v1_header_read(rbd_dev, &ver);
3000 return PTR_ERR(ondisk);
3001 ret = rbd_header_from_disk(header, ondisk);
3003 header->obj_version = ver;
3009 static void rbd_remove_all_snaps(struct rbd_device *rbd_dev)
3011 struct rbd_snap *snap;
3012 struct rbd_snap *next;
3014 list_for_each_entry_safe(snap, next, &rbd_dev->snaps, node)
3015 rbd_remove_snap_dev(snap);
3018 static void rbd_update_mapping_size(struct rbd_device *rbd_dev)
3022 if (rbd_dev->spec->snap_id != CEPH_NOSNAP)
3025 size = (sector_t) rbd_dev->header.image_size / SECTOR_SIZE;
3026 dout("setting size to %llu sectors", (unsigned long long) size);
3027 rbd_dev->mapping.size = (u64) size;
3028 set_capacity(rbd_dev->disk, size);
3032 * only read the first part of the ondisk header, without the snaps info
3034 static int rbd_dev_v1_refresh(struct rbd_device *rbd_dev, u64 *hver)
3037 struct rbd_image_header h;
3039 ret = rbd_read_header(rbd_dev, &h);
3043 down_write(&rbd_dev->header_rwsem);
3045 /* Update image size, and check for resize of mapped image */
3046 rbd_dev->header.image_size = h.image_size;
3047 rbd_update_mapping_size(rbd_dev);
3049 /* rbd_dev->header.object_prefix shouldn't change */
3050 kfree(rbd_dev->header.snap_sizes);
3051 kfree(rbd_dev->header.snap_names);
3052 /* osd requests may still refer to snapc */
3053 ceph_put_snap_context(rbd_dev->header.snapc);
3056 *hver = h.obj_version;
3057 rbd_dev->header.obj_version = h.obj_version;
3058 rbd_dev->header.image_size = h.image_size;
3059 rbd_dev->header.snapc = h.snapc;
3060 rbd_dev->header.snap_names = h.snap_names;
3061 rbd_dev->header.snap_sizes = h.snap_sizes;
3062 /* Free the extra copy of the object prefix */
3063 WARN_ON(strcmp(rbd_dev->header.object_prefix, h.object_prefix));
3064 kfree(h.object_prefix);
3066 ret = rbd_dev_snaps_update(rbd_dev);
3068 ret = rbd_dev_snaps_register(rbd_dev);
3070 up_write(&rbd_dev->header_rwsem);
3075 static int rbd_dev_refresh(struct rbd_device *rbd_dev, u64 *hver)
3079 rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
3080 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
3081 if (rbd_dev->image_format == 1)
3082 ret = rbd_dev_v1_refresh(rbd_dev, hver);
3084 ret = rbd_dev_v2_refresh(rbd_dev, hver);
3085 mutex_unlock(&ctl_mutex);
3086 revalidate_disk(rbd_dev->disk);
3091 static int rbd_init_disk(struct rbd_device *rbd_dev)
3093 struct gendisk *disk;
3094 struct request_queue *q;
3097 /* create gendisk info */
3098 disk = alloc_disk(RBD_MINORS_PER_MAJOR);
3102 snprintf(disk->disk_name, sizeof(disk->disk_name), RBD_DRV_NAME "%d",
3104 disk->major = rbd_dev->major;
3105 disk->first_minor = 0;
3106 disk->fops = &rbd_bd_ops;
3107 disk->private_data = rbd_dev;
3109 q = blk_init_queue(rbd_request_fn, &rbd_dev->lock);
3113 /* We use the default size, but let's be explicit about it. */
3114 blk_queue_physical_block_size(q, SECTOR_SIZE);
3116 /* set io sizes to object size */
3117 segment_size = rbd_obj_bytes(&rbd_dev->header);
3118 blk_queue_max_hw_sectors(q, segment_size / SECTOR_SIZE);
3119 blk_queue_max_segment_size(q, segment_size);
3120 blk_queue_io_min(q, segment_size);
3121 blk_queue_io_opt(q, segment_size);
3123 blk_queue_merge_bvec(q, rbd_merge_bvec);
3126 q->queuedata = rbd_dev;
3128 rbd_dev->disk = disk;
3130 set_capacity(rbd_dev->disk, rbd_dev->mapping.size / SECTOR_SIZE);
3143 static struct rbd_device *dev_to_rbd_dev(struct device *dev)
3145 return container_of(dev, struct rbd_device, dev);
3148 static ssize_t rbd_size_show(struct device *dev,
3149 struct device_attribute *attr, char *buf)
3151 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3154 down_read(&rbd_dev->header_rwsem);
3155 size = get_capacity(rbd_dev->disk);
3156 up_read(&rbd_dev->header_rwsem);
3158 return sprintf(buf, "%llu\n", (unsigned long long) size * SECTOR_SIZE);
3162 * Note this shows the features for whatever's mapped, which is not
3163 * necessarily the base image.
3165 static ssize_t rbd_features_show(struct device *dev,
3166 struct device_attribute *attr, char *buf)
3168 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3170 return sprintf(buf, "0x%016llx\n",
3171 (unsigned long long) rbd_dev->mapping.features);
3174 static ssize_t rbd_major_show(struct device *dev,
3175 struct device_attribute *attr, char *buf)
3177 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3179 return sprintf(buf, "%d\n", rbd_dev->major);
3182 static ssize_t rbd_client_id_show(struct device *dev,
3183 struct device_attribute *attr, char *buf)
3185 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3187 return sprintf(buf, "client%lld\n",
3188 ceph_client_id(rbd_dev->rbd_client->client));
3191 static ssize_t rbd_pool_show(struct device *dev,
3192 struct device_attribute *attr, char *buf)
3194 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3196 return sprintf(buf, "%s\n", rbd_dev->spec->pool_name);
3199 static ssize_t rbd_pool_id_show(struct device *dev,
3200 struct device_attribute *attr, char *buf)
3202 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3204 return sprintf(buf, "%llu\n",
3205 (unsigned long long) rbd_dev->spec->pool_id);
3208 static ssize_t rbd_name_show(struct device *dev,
3209 struct device_attribute *attr, char *buf)
3211 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3213 if (rbd_dev->spec->image_name)
3214 return sprintf(buf, "%s\n", rbd_dev->spec->image_name);
3216 return sprintf(buf, "(unknown)\n");
3219 static ssize_t rbd_image_id_show(struct device *dev,
3220 struct device_attribute *attr, char *buf)
3222 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3224 return sprintf(buf, "%s\n", rbd_dev->spec->image_id);
3228 * Shows the name of the currently-mapped snapshot (or
3229 * RBD_SNAP_HEAD_NAME for the base image).
3231 static ssize_t rbd_snap_show(struct device *dev,
3232 struct device_attribute *attr,
3235 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3237 return sprintf(buf, "%s\n", rbd_dev->spec->snap_name);
3241 * For an rbd v2 image, shows the pool id, image id, and snapshot id
3242 * for the parent image. If there is no parent, simply shows
3243 * "(no parent image)".
3245 static ssize_t rbd_parent_show(struct device *dev,
3246 struct device_attribute *attr,
3249 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3250 struct rbd_spec *spec = rbd_dev->parent_spec;
3255 return sprintf(buf, "(no parent image)\n");
3257 count = sprintf(bufp, "pool_id %llu\npool_name %s\n",
3258 (unsigned long long) spec->pool_id, spec->pool_name);
3263 count = sprintf(bufp, "image_id %s\nimage_name %s\n", spec->image_id,
3264 spec->image_name ? spec->image_name : "(unknown)");
3269 count = sprintf(bufp, "snap_id %llu\nsnap_name %s\n",
3270 (unsigned long long) spec->snap_id, spec->snap_name);
3275 count = sprintf(bufp, "overlap %llu\n", rbd_dev->parent_overlap);
3280 return (ssize_t) (bufp - buf);
3283 static ssize_t rbd_image_refresh(struct device *dev,
3284 struct device_attribute *attr,
3288 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3291 ret = rbd_dev_refresh(rbd_dev, NULL);
3293 return ret < 0 ? ret : size;
3296 static DEVICE_ATTR(size, S_IRUGO, rbd_size_show, NULL);
3297 static DEVICE_ATTR(features, S_IRUGO, rbd_features_show, NULL);
3298 static DEVICE_ATTR(major, S_IRUGO, rbd_major_show, NULL);
3299 static DEVICE_ATTR(client_id, S_IRUGO, rbd_client_id_show, NULL);
3300 static DEVICE_ATTR(pool, S_IRUGO, rbd_pool_show, NULL);
3301 static DEVICE_ATTR(pool_id, S_IRUGO, rbd_pool_id_show, NULL);
3302 static DEVICE_ATTR(name, S_IRUGO, rbd_name_show, NULL);
3303 static DEVICE_ATTR(image_id, S_IRUGO, rbd_image_id_show, NULL);
3304 static DEVICE_ATTR(refresh, S_IWUSR, NULL, rbd_image_refresh);
3305 static DEVICE_ATTR(current_snap, S_IRUGO, rbd_snap_show, NULL);
3306 static DEVICE_ATTR(parent, S_IRUGO, rbd_parent_show, NULL);
3308 static struct attribute *rbd_attrs[] = {
3309 &dev_attr_size.attr,
3310 &dev_attr_features.attr,
3311 &dev_attr_major.attr,
3312 &dev_attr_client_id.attr,
3313 &dev_attr_pool.attr,
3314 &dev_attr_pool_id.attr,
3315 &dev_attr_name.attr,
3316 &dev_attr_image_id.attr,
3317 &dev_attr_current_snap.attr,
3318 &dev_attr_parent.attr,
3319 &dev_attr_refresh.attr,
3323 static struct attribute_group rbd_attr_group = {
3327 static const struct attribute_group *rbd_attr_groups[] = {
3332 static void rbd_sysfs_dev_release(struct device *dev)
3336 static struct device_type rbd_device_type = {
3338 .groups = rbd_attr_groups,
3339 .release = rbd_sysfs_dev_release,
3347 static ssize_t rbd_snap_size_show(struct device *dev,
3348 struct device_attribute *attr,
3351 struct rbd_snap *snap = container_of(dev, struct rbd_snap, dev);
3353 return sprintf(buf, "%llu\n", (unsigned long long)snap->size);
3356 static ssize_t rbd_snap_id_show(struct device *dev,
3357 struct device_attribute *attr,
3360 struct rbd_snap *snap = container_of(dev, struct rbd_snap, dev);
3362 return sprintf(buf, "%llu\n", (unsigned long long)snap->id);
3365 static ssize_t rbd_snap_features_show(struct device *dev,
3366 struct device_attribute *attr,
3369 struct rbd_snap *snap = container_of(dev, struct rbd_snap, dev);
3371 return sprintf(buf, "0x%016llx\n",
3372 (unsigned long long) snap->features);
3375 static DEVICE_ATTR(snap_size, S_IRUGO, rbd_snap_size_show, NULL);
3376 static DEVICE_ATTR(snap_id, S_IRUGO, rbd_snap_id_show, NULL);
3377 static DEVICE_ATTR(snap_features, S_IRUGO, rbd_snap_features_show, NULL);
3379 static struct attribute *rbd_snap_attrs[] = {
3380 &dev_attr_snap_size.attr,
3381 &dev_attr_snap_id.attr,
3382 &dev_attr_snap_features.attr,
3386 static struct attribute_group rbd_snap_attr_group = {
3387 .attrs = rbd_snap_attrs,
3390 static void rbd_snap_dev_release(struct device *dev)
3392 struct rbd_snap *snap = container_of(dev, struct rbd_snap, dev);
3397 static const struct attribute_group *rbd_snap_attr_groups[] = {
3398 &rbd_snap_attr_group,
3402 static struct device_type rbd_snap_device_type = {
3403 .groups = rbd_snap_attr_groups,
3404 .release = rbd_snap_dev_release,
3407 static struct rbd_spec *rbd_spec_get(struct rbd_spec *spec)
3409 kref_get(&spec->kref);
3414 static void rbd_spec_free(struct kref *kref);
3415 static void rbd_spec_put(struct rbd_spec *spec)
3418 kref_put(&spec->kref, rbd_spec_free);
3421 static struct rbd_spec *rbd_spec_alloc(void)
3423 struct rbd_spec *spec;
3425 spec = kzalloc(sizeof (*spec), GFP_KERNEL);
3428 kref_init(&spec->kref);
3433 static void rbd_spec_free(struct kref *kref)
3435 struct rbd_spec *spec = container_of(kref, struct rbd_spec, kref);
3437 kfree(spec->pool_name);
3438 kfree(spec->image_id);
3439 kfree(spec->image_name);
3440 kfree(spec->snap_name);
3444 static struct rbd_device *rbd_dev_create(struct rbd_client *rbdc,
3445 struct rbd_spec *spec)
3447 struct rbd_device *rbd_dev;
3449 rbd_dev = kzalloc(sizeof (*rbd_dev), GFP_KERNEL);
3453 spin_lock_init(&rbd_dev->lock);
3455 INIT_LIST_HEAD(&rbd_dev->node);
3456 INIT_LIST_HEAD(&rbd_dev->snaps);
3457 init_rwsem(&rbd_dev->header_rwsem);
3459 rbd_dev->spec = spec;
3460 rbd_dev->rbd_client = rbdc;
3462 /* Initialize the layout used for all rbd requests */
3464 rbd_dev->layout.fl_stripe_unit = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER);
3465 rbd_dev->layout.fl_stripe_count = cpu_to_le32(1);
3466 rbd_dev->layout.fl_object_size = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER);
3467 rbd_dev->layout.fl_pg_pool = cpu_to_le32((u32) spec->pool_id);
3472 static void rbd_dev_destroy(struct rbd_device *rbd_dev)
3474 rbd_spec_put(rbd_dev->parent_spec);
3475 kfree(rbd_dev->header_name);
3476 rbd_put_client(rbd_dev->rbd_client);
3477 rbd_spec_put(rbd_dev->spec);
3481 static bool rbd_snap_registered(struct rbd_snap *snap)
3483 bool ret = snap->dev.type == &rbd_snap_device_type;
3484 bool reg = device_is_registered(&snap->dev);
3486 rbd_assert(!ret ^ reg);
3491 static void rbd_remove_snap_dev(struct rbd_snap *snap)
3493 list_del(&snap->node);
3494 if (device_is_registered(&snap->dev))
3495 device_unregister(&snap->dev);
3498 static int rbd_register_snap_dev(struct rbd_snap *snap,
3499 struct device *parent)
3501 struct device *dev = &snap->dev;
3504 dev->type = &rbd_snap_device_type;
3505 dev->parent = parent;
3506 dev->release = rbd_snap_dev_release;
3507 dev_set_name(dev, "%s%s", RBD_SNAP_DEV_NAME_PREFIX, snap->name);
3508 dout("%s: registering device for snapshot %s\n", __func__, snap->name);
3510 ret = device_register(dev);
3515 static struct rbd_snap *__rbd_add_snap_dev(struct rbd_device *rbd_dev,
3516 const char *snap_name,
3517 u64 snap_id, u64 snap_size,
3520 struct rbd_snap *snap;
3523 snap = kzalloc(sizeof (*snap), GFP_KERNEL);
3525 return ERR_PTR(-ENOMEM);
3528 snap->name = kstrdup(snap_name, GFP_KERNEL);
3533 snap->size = snap_size;
3534 snap->features = snap_features;
3542 return ERR_PTR(ret);
3545 static char *rbd_dev_v1_snap_info(struct rbd_device *rbd_dev, u32 which,
3546 u64 *snap_size, u64 *snap_features)
3550 rbd_assert(which < rbd_dev->header.snapc->num_snaps);
3552 *snap_size = rbd_dev->header.snap_sizes[which];
3553 *snap_features = 0; /* No features for v1 */
3555 /* Skip over names until we find the one we are looking for */
3557 snap_name = rbd_dev->header.snap_names;
3559 snap_name += strlen(snap_name) + 1;
3565 * Get the size and object order for an image snapshot, or if
3566 * snap_id is CEPH_NOSNAP, gets this information for the base
3569 static int _rbd_dev_v2_snap_size(struct rbd_device *rbd_dev, u64 snap_id,
3570 u8 *order, u64 *snap_size)
3572 __le64 snapid = cpu_to_le64(snap_id);
3577 } __attribute__ ((packed)) size_buf = { 0 };
3579 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
3581 &snapid, sizeof (snapid),
3582 &size_buf, sizeof (size_buf), NULL);
3583 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
3587 *order = size_buf.order;
3588 *snap_size = le64_to_cpu(size_buf.size);
3590 dout(" snap_id 0x%016llx order = %u, snap_size = %llu\n",
3591 (unsigned long long) snap_id, (unsigned int) *order,
3592 (unsigned long long) *snap_size);
3597 static int rbd_dev_v2_image_size(struct rbd_device *rbd_dev)
3599 return _rbd_dev_v2_snap_size(rbd_dev, CEPH_NOSNAP,
3600 &rbd_dev->header.obj_order,
3601 &rbd_dev->header.image_size);
3604 static int rbd_dev_v2_object_prefix(struct rbd_device *rbd_dev)
3610 reply_buf = kzalloc(RBD_OBJ_PREFIX_LEN_MAX, GFP_KERNEL);
3614 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
3615 "rbd", "get_object_prefix", NULL, 0,
3616 reply_buf, RBD_OBJ_PREFIX_LEN_MAX, NULL);
3617 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
3622 rbd_dev->header.object_prefix = ceph_extract_encoded_string(&p,
3623 p + RBD_OBJ_PREFIX_LEN_MAX,
3626 if (IS_ERR(rbd_dev->header.object_prefix)) {
3627 ret = PTR_ERR(rbd_dev->header.object_prefix);
3628 rbd_dev->header.object_prefix = NULL;
3630 dout(" object_prefix = %s\n", rbd_dev->header.object_prefix);
3639 static int _rbd_dev_v2_snap_features(struct rbd_device *rbd_dev, u64 snap_id,
3642 __le64 snapid = cpu_to_le64(snap_id);
3646 } __attribute__ ((packed)) features_buf = { 0 };
3650 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
3651 "rbd", "get_features",
3652 &snapid, sizeof (snapid),
3653 &features_buf, sizeof (features_buf), NULL);
3654 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
3658 incompat = le64_to_cpu(features_buf.incompat);
3659 if (incompat & ~RBD_FEATURES_SUPPORTED)
3662 *snap_features = le64_to_cpu(features_buf.features);
3664 dout(" snap_id 0x%016llx features = 0x%016llx incompat = 0x%016llx\n",
3665 (unsigned long long) snap_id,
3666 (unsigned long long) *snap_features,
3667 (unsigned long long) le64_to_cpu(features_buf.incompat));
3672 static int rbd_dev_v2_features(struct rbd_device *rbd_dev)
3674 return _rbd_dev_v2_snap_features(rbd_dev, CEPH_NOSNAP,
3675 &rbd_dev->header.features);
3678 static int rbd_dev_v2_parent_info(struct rbd_device *rbd_dev)
3680 struct rbd_spec *parent_spec;
3682 void *reply_buf = NULL;
3690 parent_spec = rbd_spec_alloc();
3694 size = sizeof (__le64) + /* pool_id */
3695 sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX + /* image_id */
3696 sizeof (__le64) + /* snap_id */
3697 sizeof (__le64); /* overlap */
3698 reply_buf = kmalloc(size, GFP_KERNEL);
3704 snapid = cpu_to_le64(CEPH_NOSNAP);
3705 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
3706 "rbd", "get_parent",
3707 &snapid, sizeof (snapid),
3708 reply_buf, size, NULL);
3709 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
3715 end = reply_buf + size;
3716 ceph_decode_64_safe(&p, end, parent_spec->pool_id, out_err);
3717 if (parent_spec->pool_id == CEPH_NOPOOL)
3718 goto out; /* No parent? No problem. */
3720 /* The ceph file layout needs to fit pool id in 32 bits */
3723 if (WARN_ON(parent_spec->pool_id > (u64) U32_MAX))
3726 image_id = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL);
3727 if (IS_ERR(image_id)) {
3728 ret = PTR_ERR(image_id);
3731 parent_spec->image_id = image_id;
3732 ceph_decode_64_safe(&p, end, parent_spec->snap_id, out_err);
3733 ceph_decode_64_safe(&p, end, overlap, out_err);
3735 rbd_dev->parent_overlap = overlap;
3736 rbd_dev->parent_spec = parent_spec;
3737 parent_spec = NULL; /* rbd_dev now owns this */
3742 rbd_spec_put(parent_spec);
3747 static char *rbd_dev_image_name(struct rbd_device *rbd_dev)
3749 size_t image_id_size;
3754 void *reply_buf = NULL;
3756 char *image_name = NULL;
3759 rbd_assert(!rbd_dev->spec->image_name);
3761 len = strlen(rbd_dev->spec->image_id);
3762 image_id_size = sizeof (__le32) + len;
3763 image_id = kmalloc(image_id_size, GFP_KERNEL);
3768 end = image_id + image_id_size;
3769 ceph_encode_string(&p, end, rbd_dev->spec->image_id, (u32) len);
3771 size = sizeof (__le32) + RBD_IMAGE_NAME_LEN_MAX;
3772 reply_buf = kmalloc(size, GFP_KERNEL);
3776 ret = rbd_obj_method_sync(rbd_dev, RBD_DIRECTORY,
3777 "rbd", "dir_get_name",
3778 image_id, image_id_size,
3779 reply_buf, size, NULL);
3783 end = reply_buf + size;
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);
3797 * When a parent image gets probed, we only have the pool, image,
3798 * and snapshot ids but not the names of any of them. This call
3799 * is made later to fill in those names. It has to be done after
3800 * rbd_dev_snaps_update() has completed because some of the
3801 * information (in particular, snapshot name) is not available
3804 static int rbd_dev_probe_update_spec(struct rbd_device *rbd_dev)
3806 struct ceph_osd_client *osdc;
3808 void *reply_buf = NULL;
3811 if (rbd_dev->spec->pool_name)
3812 return 0; /* Already have the names */
3814 /* Look up the pool name */
3816 osdc = &rbd_dev->rbd_client->client->osdc;
3817 name = ceph_pg_pool_name_by_id(osdc->osdmap, rbd_dev->spec->pool_id);
3819 rbd_warn(rbd_dev, "there is no pool with id %llu",
3820 rbd_dev->spec->pool_id); /* Really a BUG() */
3824 rbd_dev->spec->pool_name = kstrdup(name, GFP_KERNEL);
3825 if (!rbd_dev->spec->pool_name)
3828 /* Fetch the image name; tolerate failure here */
3830 name = rbd_dev_image_name(rbd_dev);
3832 rbd_dev->spec->image_name = (char *)name;
3834 rbd_warn(rbd_dev, "unable to get image name");
3836 /* Look up the snapshot name. */
3838 name = rbd_snap_name(rbd_dev, rbd_dev->spec->snap_id);
3840 rbd_warn(rbd_dev, "no snapshot with id %llu",
3841 rbd_dev->spec->snap_id); /* Really a BUG() */
3845 rbd_dev->spec->snap_name = kstrdup(name, GFP_KERNEL);
3846 if(!rbd_dev->spec->snap_name)
3852 kfree(rbd_dev->spec->pool_name);
3853 rbd_dev->spec->pool_name = NULL;
3858 static int rbd_dev_v2_snap_context(struct rbd_device *rbd_dev, u64 *ver)
3867 struct ceph_snap_context *snapc;
3871 * We'll need room for the seq value (maximum snapshot id),
3872 * snapshot count, and array of that many snapshot ids.
3873 * For now we have a fixed upper limit on the number we're
3874 * prepared to receive.
3876 size = sizeof (__le64) + sizeof (__le32) +
3877 RBD_MAX_SNAP_COUNT * sizeof (__le64);
3878 reply_buf = kzalloc(size, GFP_KERNEL);
3882 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
3883 "rbd", "get_snapcontext", NULL, 0,
3884 reply_buf, size, ver);
3885 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
3891 end = reply_buf + size;
3892 ceph_decode_64_safe(&p, end, seq, out);
3893 ceph_decode_32_safe(&p, end, snap_count, out);
3896 * Make sure the reported number of snapshot ids wouldn't go
3897 * beyond the end of our buffer. But before checking that,
3898 * make sure the computed size of the snapshot context we
3899 * allocate is representable in a size_t.
3901 if (snap_count > (SIZE_MAX - sizeof (struct ceph_snap_context))
3906 if (!ceph_has_room(&p, end, snap_count * sizeof (__le64)))
3909 size = sizeof (struct ceph_snap_context) +
3910 snap_count * sizeof (snapc->snaps[0]);
3911 snapc = kmalloc(size, GFP_KERNEL);
3917 atomic_set(&snapc->nref, 1);
3919 snapc->num_snaps = snap_count;
3920 for (i = 0; i < snap_count; i++)
3921 snapc->snaps[i] = ceph_decode_64(&p);
3923 rbd_dev->header.snapc = snapc;
3925 dout(" snap context seq = %llu, snap_count = %u\n",
3926 (unsigned long long) seq, (unsigned int) snap_count);
3934 static char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev, u32 which)
3944 size = sizeof (__le32) + RBD_MAX_SNAP_NAME_LEN;
3945 reply_buf = kmalloc(size, GFP_KERNEL);
3947 return ERR_PTR(-ENOMEM);
3949 snap_id = cpu_to_le64(rbd_dev->header.snapc->snaps[which]);
3950 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
3951 "rbd", "get_snapshot_name",
3952 &snap_id, sizeof (snap_id),
3953 reply_buf, size, NULL);
3954 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
3959 end = reply_buf + size;
3960 snap_name = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL);
3961 if (IS_ERR(snap_name)) {
3962 ret = PTR_ERR(snap_name);
3965 dout(" snap_id 0x%016llx snap_name = %s\n",
3966 (unsigned long long) le64_to_cpu(snap_id), snap_name);
3974 return ERR_PTR(ret);
3977 static char *rbd_dev_v2_snap_info(struct rbd_device *rbd_dev, u32 which,
3978 u64 *snap_size, u64 *snap_features)
3984 snap_id = rbd_dev->header.snapc->snaps[which];
3985 ret = _rbd_dev_v2_snap_size(rbd_dev, snap_id, &order, snap_size);
3987 return ERR_PTR(ret);
3988 ret = _rbd_dev_v2_snap_features(rbd_dev, snap_id, snap_features);
3990 return ERR_PTR(ret);
3992 return rbd_dev_v2_snap_name(rbd_dev, which);
3995 static char *rbd_dev_snap_info(struct rbd_device *rbd_dev, u32 which,
3996 u64 *snap_size, u64 *snap_features)
3998 if (rbd_dev->image_format == 1)
3999 return rbd_dev_v1_snap_info(rbd_dev, which,
4000 snap_size, snap_features);
4001 if (rbd_dev->image_format == 2)
4002 return rbd_dev_v2_snap_info(rbd_dev, which,
4003 snap_size, snap_features);
4004 return ERR_PTR(-EINVAL);
4007 static int rbd_dev_v2_refresh(struct rbd_device *rbd_dev, u64 *hver)
4012 down_write(&rbd_dev->header_rwsem);
4014 /* Grab old order first, to see if it changes */
4016 obj_order = rbd_dev->header.obj_order,
4017 ret = rbd_dev_v2_image_size(rbd_dev);
4020 if (rbd_dev->header.obj_order != obj_order) {
4024 rbd_update_mapping_size(rbd_dev);
4026 ret = rbd_dev_v2_snap_context(rbd_dev, hver);
4027 dout("rbd_dev_v2_snap_context returned %d\n", ret);
4030 ret = rbd_dev_snaps_update(rbd_dev);
4031 dout("rbd_dev_snaps_update returned %d\n", ret);
4034 ret = rbd_dev_snaps_register(rbd_dev);
4035 dout("rbd_dev_snaps_register returned %d\n", ret);
4037 up_write(&rbd_dev->header_rwsem);
4043 * Scan the rbd device's current snapshot list and compare it to the
4044 * newly-received snapshot context. Remove any existing snapshots
4045 * not present in the new snapshot context. Add a new snapshot for
4046 * any snaphots in the snapshot context not in the current list.
4047 * And verify there are no changes to snapshots we already know
4050 * Assumes the snapshots in the snapshot context are sorted by
4051 * snapshot id, highest id first. (Snapshots in the rbd_dev's list
4052 * are also maintained in that order.)
4054 static int rbd_dev_snaps_update(struct rbd_device *rbd_dev)
4056 struct ceph_snap_context *snapc = rbd_dev->header.snapc;
4057 const u32 snap_count = snapc->num_snaps;
4058 struct list_head *head = &rbd_dev->snaps;
4059 struct list_head *links = head->next;
4062 dout("%s: snap count is %u\n", __func__, (unsigned int) snap_count);
4063 while (index < snap_count || links != head) {
4065 struct rbd_snap *snap;
4068 u64 snap_features = 0;
4070 snap_id = index < snap_count ? snapc->snaps[index]
4072 snap = links != head ? list_entry(links, struct rbd_snap, node)
4074 rbd_assert(!snap || snap->id != CEPH_NOSNAP);
4076 if (snap_id == CEPH_NOSNAP || (snap && snap->id > snap_id)) {
4077 struct list_head *next = links->next;
4080 * A previously-existing snapshot is not in
4081 * the new snap context.
4083 * If the now missing snapshot is the one the
4084 * image is mapped to, clear its exists flag
4085 * so we can avoid sending any more requests
4088 if (rbd_dev->spec->snap_id == snap->id)
4089 clear_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags);
4090 rbd_remove_snap_dev(snap);
4091 dout("%ssnap id %llu has been removed\n",
4092 rbd_dev->spec->snap_id == snap->id ?
4094 (unsigned long long) snap->id);
4096 /* Done with this list entry; advance */
4102 snap_name = rbd_dev_snap_info(rbd_dev, index,
4103 &snap_size, &snap_features);
4104 if (IS_ERR(snap_name))
4105 return PTR_ERR(snap_name);
4107 dout("entry %u: snap_id = %llu\n", (unsigned int) snap_count,
4108 (unsigned long long) snap_id);
4109 if (!snap || (snap_id != CEPH_NOSNAP && snap->id < snap_id)) {
4110 struct rbd_snap *new_snap;
4112 /* We haven't seen this snapshot before */
4114 new_snap = __rbd_add_snap_dev(rbd_dev, snap_name,
4115 snap_id, snap_size, snap_features);
4116 if (IS_ERR(new_snap)) {
4117 int err = PTR_ERR(new_snap);
4119 dout(" failed to add dev, error %d\n", err);
4124 /* New goes before existing, or at end of list */
4126 dout(" added dev%s\n", snap ? "" : " at end\n");
4128 list_add_tail(&new_snap->node, &snap->node);
4130 list_add_tail(&new_snap->node, head);
4132 /* Already have this one */
4134 dout(" already present\n");
4136 rbd_assert(snap->size == snap_size);
4137 rbd_assert(!strcmp(snap->name, snap_name));
4138 rbd_assert(snap->features == snap_features);
4140 /* Done with this list entry; advance */
4142 links = links->next;
4145 /* Advance to the next entry in the snapshot context */
4149 dout("%s: done\n", __func__);
4155 * Scan the list of snapshots and register the devices for any that
4156 * have not already been registered.
4158 static int rbd_dev_snaps_register(struct rbd_device *rbd_dev)
4160 struct rbd_snap *snap;
4163 dout("%s:\n", __func__);
4164 if (WARN_ON(!device_is_registered(&rbd_dev->dev)))
4167 list_for_each_entry(snap, &rbd_dev->snaps, node) {
4168 if (!rbd_snap_registered(snap)) {
4169 ret = rbd_register_snap_dev(snap, &rbd_dev->dev);
4174 dout("%s: returning %d\n", __func__, ret);
4179 static int rbd_bus_add_dev(struct rbd_device *rbd_dev)
4184 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
4186 dev = &rbd_dev->dev;
4187 dev->bus = &rbd_bus_type;
4188 dev->type = &rbd_device_type;
4189 dev->parent = &rbd_root_dev;
4190 dev->release = rbd_dev_release;
4191 dev_set_name(dev, "%d", rbd_dev->dev_id);
4192 ret = device_register(dev);
4194 mutex_unlock(&ctl_mutex);
4199 static void rbd_bus_del_dev(struct rbd_device *rbd_dev)
4201 device_unregister(&rbd_dev->dev);
4204 static atomic64_t rbd_dev_id_max = ATOMIC64_INIT(0);
4207 * Get a unique rbd identifier for the given new rbd_dev, and add
4208 * the rbd_dev to the global list. The minimum rbd id is 1.
4210 static void rbd_dev_id_get(struct rbd_device *rbd_dev)
4212 rbd_dev->dev_id = atomic64_inc_return(&rbd_dev_id_max);
4214 spin_lock(&rbd_dev_list_lock);
4215 list_add_tail(&rbd_dev->node, &rbd_dev_list);
4216 spin_unlock(&rbd_dev_list_lock);
4217 dout("rbd_dev %p given dev id %llu\n", rbd_dev,
4218 (unsigned long long) rbd_dev->dev_id);
4222 * Remove an rbd_dev from the global list, and record that its
4223 * identifier is no longer in use.
4225 static void rbd_dev_id_put(struct rbd_device *rbd_dev)
4227 struct list_head *tmp;
4228 int rbd_id = rbd_dev->dev_id;
4231 rbd_assert(rbd_id > 0);
4233 dout("rbd_dev %p released dev id %llu\n", rbd_dev,
4234 (unsigned long long) rbd_dev->dev_id);
4235 spin_lock(&rbd_dev_list_lock);
4236 list_del_init(&rbd_dev->node);
4239 * If the id being "put" is not the current maximum, there
4240 * is nothing special we need to do.
4242 if (rbd_id != atomic64_read(&rbd_dev_id_max)) {
4243 spin_unlock(&rbd_dev_list_lock);
4248 * We need to update the current maximum id. Search the
4249 * list to find out what it is. We're more likely to find
4250 * the maximum at the end, so search the list backward.
4253 list_for_each_prev(tmp, &rbd_dev_list) {
4254 struct rbd_device *rbd_dev;
4256 rbd_dev = list_entry(tmp, struct rbd_device, node);
4257 if (rbd_dev->dev_id > max_id)
4258 max_id = rbd_dev->dev_id;
4260 spin_unlock(&rbd_dev_list_lock);
4263 * The max id could have been updated by rbd_dev_id_get(), in
4264 * which case it now accurately reflects the new maximum.
4265 * Be careful not to overwrite the maximum value in that
4268 atomic64_cmpxchg(&rbd_dev_id_max, rbd_id, max_id);
4269 dout(" max dev id has been reset\n");
4273 * Skips over white space at *buf, and updates *buf to point to the
4274 * first found non-space character (if any). Returns the length of
4275 * the token (string of non-white space characters) found. Note
4276 * that *buf must be terminated with '\0'.
4278 static inline size_t next_token(const char **buf)
4281 * These are the characters that produce nonzero for
4282 * isspace() in the "C" and "POSIX" locales.
4284 const char *spaces = " \f\n\r\t\v";
4286 *buf += strspn(*buf, spaces); /* Find start of token */
4288 return strcspn(*buf, spaces); /* Return token length */
4292 * Finds the next token in *buf, and if the provided token buffer is
4293 * big enough, copies the found token into it. The result, if
4294 * copied, is guaranteed to be terminated with '\0'. Note that *buf
4295 * must be terminated with '\0' on entry.
4297 * Returns the length of the token found (not including the '\0').
4298 * Return value will be 0 if no token is found, and it will be >=
4299 * token_size if the token would not fit.
4301 * The *buf pointer will be updated to point beyond the end of the
4302 * found token. Note that this occurs even if the token buffer is
4303 * too small to hold it.
4305 static inline size_t copy_token(const char **buf,
4311 len = next_token(buf);
4312 if (len < token_size) {
4313 memcpy(token, *buf, len);
4314 *(token + len) = '\0';
4322 * Finds the next token in *buf, dynamically allocates a buffer big
4323 * enough to hold a copy of it, and copies the token into the new
4324 * buffer. The copy is guaranteed to be terminated with '\0'. Note
4325 * that a duplicate buffer is created even for a zero-length token.
4327 * Returns a pointer to the newly-allocated duplicate, or a null
4328 * pointer if memory for the duplicate was not available. If
4329 * the lenp argument is a non-null pointer, the length of the token
4330 * (not including the '\0') is returned in *lenp.
4332 * If successful, the *buf pointer will be updated to point beyond
4333 * the end of the found token.
4335 * Note: uses GFP_KERNEL for allocation.
4337 static inline char *dup_token(const char **buf, size_t *lenp)
4342 len = next_token(buf);
4343 dup = kmemdup(*buf, len + 1, GFP_KERNEL);
4346 *(dup + len) = '\0';
4356 * Parse the options provided for an "rbd add" (i.e., rbd image
4357 * mapping) request. These arrive via a write to /sys/bus/rbd/add,
4358 * and the data written is passed here via a NUL-terminated buffer.
4359 * Returns 0 if successful or an error code otherwise.
4361 * The information extracted from these options is recorded in
4362 * the other parameters which return dynamically-allocated
4365 * The address of a pointer that will refer to a ceph options
4366 * structure. Caller must release the returned pointer using
4367 * ceph_destroy_options() when it is no longer needed.
4369 * Address of an rbd options pointer. Fully initialized by
4370 * this function; caller must release with kfree().
4372 * Address of an rbd image specification pointer. Fully
4373 * initialized by this function based on parsed options.
4374 * Caller must release with rbd_spec_put().
4376 * The options passed take this form:
4377 * <mon_addrs> <options> <pool_name> <image_name> [<snap_id>]
4380 * A comma-separated list of one or more monitor addresses.
4381 * A monitor address is an ip address, optionally followed
4382 * by a port number (separated by a colon).
4383 * I.e.: ip1[:port1][,ip2[:port2]...]
4385 * A comma-separated list of ceph and/or rbd options.
4387 * The name of the rados pool containing the rbd image.
4389 * The name of the image in that pool to map.
4391 * An optional snapshot id. If provided, the mapping will
4392 * present data from the image at the time that snapshot was
4393 * created. The image head is used if no snapshot id is
4394 * provided. Snapshot mappings are always read-only.
4396 static int rbd_add_parse_args(const char *buf,
4397 struct ceph_options **ceph_opts,
4398 struct rbd_options **opts,
4399 struct rbd_spec **rbd_spec)
4403 const char *mon_addrs;
4404 size_t mon_addrs_size;
4405 struct rbd_spec *spec = NULL;
4406 struct rbd_options *rbd_opts = NULL;
4407 struct ceph_options *copts;
4410 /* The first four tokens are required */
4412 len = next_token(&buf);
4414 rbd_warn(NULL, "no monitor address(es) provided");
4418 mon_addrs_size = len + 1;
4422 options = dup_token(&buf, NULL);
4426 rbd_warn(NULL, "no options provided");
4430 spec = rbd_spec_alloc();
4434 spec->pool_name = dup_token(&buf, NULL);
4435 if (!spec->pool_name)
4437 if (!*spec->pool_name) {
4438 rbd_warn(NULL, "no pool name provided");
4442 spec->image_name = dup_token(&buf, NULL);
4443 if (!spec->image_name)
4445 if (!*spec->image_name) {
4446 rbd_warn(NULL, "no image name provided");
4451 * Snapshot name is optional; default is to use "-"
4452 * (indicating the head/no snapshot).
4454 len = next_token(&buf);
4456 buf = RBD_SNAP_HEAD_NAME; /* No snapshot supplied */
4457 len = sizeof (RBD_SNAP_HEAD_NAME) - 1;
4458 } else if (len > RBD_MAX_SNAP_NAME_LEN) {
4459 ret = -ENAMETOOLONG;
4462 spec->snap_name = kmemdup(buf, len + 1, GFP_KERNEL);
4463 if (!spec->snap_name)
4465 *(spec->snap_name + len) = '\0';
4467 /* Initialize all rbd options to the defaults */
4469 rbd_opts = kzalloc(sizeof (*rbd_opts), GFP_KERNEL);
4473 rbd_opts->read_only = RBD_READ_ONLY_DEFAULT;
4475 copts = ceph_parse_options(options, mon_addrs,
4476 mon_addrs + mon_addrs_size - 1,
4477 parse_rbd_opts_token, rbd_opts);
4478 if (IS_ERR(copts)) {
4479 ret = PTR_ERR(copts);
4500 * An rbd format 2 image has a unique identifier, distinct from the
4501 * name given to it by the user. Internally, that identifier is
4502 * what's used to specify the names of objects related to the image.
4504 * A special "rbd id" object is used to map an rbd image name to its
4505 * id. If that object doesn't exist, then there is no v2 rbd image
4506 * with the supplied name.
4508 * This function will record the given rbd_dev's image_id field if
4509 * it can be determined, and in that case will return 0. If any
4510 * errors occur a negative errno will be returned and the rbd_dev's
4511 * image_id field will be unchanged (and should be NULL).
4513 static int rbd_dev_image_id(struct rbd_device *rbd_dev)
4521 /* If we already have it we don't need to look it up */
4523 if (rbd_dev->spec->image_id)
4527 * When probing a parent image, the image id is already
4528 * known (and the image name likely is not). There's no
4529 * need to fetch the image id again in this case.
4531 if (rbd_dev->spec->image_id)
4535 * First, see if the format 2 image id file exists, and if
4536 * so, get the image's persistent id from it.
4538 size = sizeof (RBD_ID_PREFIX) + strlen(rbd_dev->spec->image_name);
4539 object_name = kmalloc(size, GFP_NOIO);
4542 sprintf(object_name, "%s%s", RBD_ID_PREFIX, rbd_dev->spec->image_name);
4543 dout("rbd id object name is %s\n", object_name);
4545 /* Response will be an encoded string, which includes a length */
4547 size = sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX;
4548 response = kzalloc(size, GFP_NOIO);
4554 ret = rbd_obj_method_sync(rbd_dev, object_name,
4555 "rbd", "get_id", NULL, 0,
4556 response, RBD_IMAGE_ID_LEN_MAX, NULL);
4557 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
4562 rbd_dev->spec->image_id = ceph_extract_encoded_string(&p,
4563 p + RBD_IMAGE_ID_LEN_MAX,
4565 if (IS_ERR(rbd_dev->spec->image_id)) {
4566 ret = PTR_ERR(rbd_dev->spec->image_id);
4567 rbd_dev->spec->image_id = NULL;
4569 dout("image_id is %s\n", rbd_dev->spec->image_id);
4578 static int rbd_dev_v1_probe(struct rbd_device *rbd_dev)
4583 /* Version 1 images have no id; empty string is used */
4585 rbd_dev->spec->image_id = kstrdup("", GFP_KERNEL);
4586 if (!rbd_dev->spec->image_id)
4589 /* Record the header object name for this rbd image. */
4591 size = strlen(rbd_dev->spec->image_name) + sizeof (RBD_SUFFIX);
4592 rbd_dev->header_name = kmalloc(size, GFP_KERNEL);
4593 if (!rbd_dev->header_name) {
4597 sprintf(rbd_dev->header_name, "%s%s",
4598 rbd_dev->spec->image_name, RBD_SUFFIX);
4600 /* Populate rbd image metadata */
4602 ret = rbd_read_header(rbd_dev, &rbd_dev->header);
4606 /* Version 1 images have no parent (no layering) */
4608 rbd_dev->parent_spec = NULL;
4609 rbd_dev->parent_overlap = 0;
4611 rbd_dev->image_format = 1;
4613 dout("discovered version 1 image, header name is %s\n",
4614 rbd_dev->header_name);
4619 kfree(rbd_dev->header_name);
4620 rbd_dev->header_name = NULL;
4621 kfree(rbd_dev->spec->image_id);
4622 rbd_dev->spec->image_id = NULL;
4627 static int rbd_dev_v2_probe(struct rbd_device *rbd_dev)
4634 * Image id was filled in by the caller. Record the header
4635 * object name for this rbd image.
4637 size = sizeof (RBD_HEADER_PREFIX) + strlen(rbd_dev->spec->image_id);
4638 rbd_dev->header_name = kmalloc(size, GFP_KERNEL);
4639 if (!rbd_dev->header_name)
4641 sprintf(rbd_dev->header_name, "%s%s",
4642 RBD_HEADER_PREFIX, rbd_dev->spec->image_id);
4644 /* Get the size and object order for the image */
4646 ret = rbd_dev_v2_image_size(rbd_dev);
4650 /* Get the object prefix (a.k.a. block_name) for the image */
4652 ret = rbd_dev_v2_object_prefix(rbd_dev);
4656 /* Get the and check features for the image */
4658 ret = rbd_dev_v2_features(rbd_dev);
4662 /* If the image supports layering, get the parent info */
4664 if (rbd_dev->header.features & RBD_FEATURE_LAYERING) {
4665 ret = rbd_dev_v2_parent_info(rbd_dev);
4670 /* crypto and compression type aren't (yet) supported for v2 images */
4672 rbd_dev->header.crypt_type = 0;
4673 rbd_dev->header.comp_type = 0;
4675 /* Get the snapshot context, plus the header version */
4677 ret = rbd_dev_v2_snap_context(rbd_dev, &ver);
4680 rbd_dev->header.obj_version = ver;
4682 rbd_dev->image_format = 2;
4684 dout("discovered version 2 image, header name is %s\n",
4685 rbd_dev->header_name);
4689 rbd_dev->parent_overlap = 0;
4690 rbd_spec_put(rbd_dev->parent_spec);
4691 rbd_dev->parent_spec = NULL;
4692 kfree(rbd_dev->header_name);
4693 rbd_dev->header_name = NULL;
4694 kfree(rbd_dev->header.object_prefix);
4695 rbd_dev->header.object_prefix = NULL;
4700 static int rbd_dev_probe_finish(struct rbd_device *rbd_dev)
4702 struct rbd_device *parent = NULL;
4703 struct rbd_spec *parent_spec = NULL;
4704 struct rbd_client *rbdc = NULL;
4707 /* no need to lock here, as rbd_dev is not registered yet */
4708 ret = rbd_dev_snaps_update(rbd_dev);
4712 ret = rbd_dev_probe_update_spec(rbd_dev);
4716 ret = rbd_dev_set_mapping(rbd_dev);
4720 /* generate unique id: find highest unique id, add one */
4721 rbd_dev_id_get(rbd_dev);
4723 /* Fill in the device name, now that we have its id. */
4724 BUILD_BUG_ON(DEV_NAME_LEN
4725 < sizeof (RBD_DRV_NAME) + MAX_INT_FORMAT_WIDTH);
4726 sprintf(rbd_dev->name, "%s%d", RBD_DRV_NAME, rbd_dev->dev_id);
4728 /* Get our block major device number. */
4730 ret = register_blkdev(0, rbd_dev->name);
4733 rbd_dev->major = ret;
4735 /* Set up the blkdev mapping. */
4737 ret = rbd_init_disk(rbd_dev);
4739 goto err_out_blkdev;
4741 ret = rbd_bus_add_dev(rbd_dev);
4746 * At this point cleanup in the event of an error is the job
4747 * of the sysfs code (initiated by rbd_bus_del_dev()).
4749 /* Probe the parent if there is one */
4751 if (rbd_dev->parent_spec) {
4753 * We need to pass a reference to the client and the
4754 * parent spec when creating the parent rbd_dev.
4755 * Images related by parent/child relationships
4756 * always share both.
4758 parent_spec = rbd_spec_get(rbd_dev->parent_spec);
4759 rbdc = __rbd_get_client(rbd_dev->rbd_client);
4761 parent = rbd_dev_create(rbdc, parent_spec);
4766 rbdc = NULL; /* parent now owns reference */
4767 parent_spec = NULL; /* parent now owns reference */
4768 ret = rbd_dev_probe(parent);
4770 goto err_out_parent;
4771 rbd_dev->parent = parent;
4774 down_write(&rbd_dev->header_rwsem);
4775 ret = rbd_dev_snaps_register(rbd_dev);
4776 up_write(&rbd_dev->header_rwsem);
4780 ret = rbd_dev_header_watch_sync(rbd_dev, 1);
4784 /* Everything's ready. Announce the disk to the world. */
4786 add_disk(rbd_dev->disk);
4788 pr_info("%s: added with size 0x%llx\n", rbd_dev->disk->disk_name,
4789 (unsigned long long) rbd_dev->mapping.size);
4794 rbd_dev_destroy(parent);
4796 rbd_spec_put(parent_spec);
4797 rbd_put_client(rbdc);
4799 /* this will also clean up rest of rbd_dev stuff */
4801 rbd_bus_del_dev(rbd_dev);
4805 rbd_free_disk(rbd_dev);
4807 unregister_blkdev(rbd_dev->major, rbd_dev->name);
4809 rbd_dev_id_put(rbd_dev);
4811 rbd_remove_all_snaps(rbd_dev);
4817 * Probe for the existence of the header object for the given rbd
4818 * device. For format 2 images this includes determining the image
4821 static int rbd_dev_probe(struct rbd_device *rbd_dev)
4826 * Get the id from the image id object. If it's not a
4827 * format 2 image, we'll get ENOENT back, and we'll assume
4828 * it's a format 1 image.
4830 ret = rbd_dev_image_id(rbd_dev);
4832 ret = rbd_dev_v1_probe(rbd_dev);
4834 ret = rbd_dev_v2_probe(rbd_dev);
4836 dout("probe failed, returning %d\n", ret);
4841 ret = rbd_dev_probe_finish(rbd_dev);
4843 rbd_header_free(&rbd_dev->header);
4848 static ssize_t rbd_add(struct bus_type *bus,
4852 struct rbd_device *rbd_dev = NULL;
4853 struct ceph_options *ceph_opts = NULL;
4854 struct rbd_options *rbd_opts = NULL;
4855 struct rbd_spec *spec = NULL;
4856 struct rbd_client *rbdc;
4857 struct ceph_osd_client *osdc;
4860 if (!try_module_get(THIS_MODULE))
4863 /* parse add command */
4864 rc = rbd_add_parse_args(buf, &ceph_opts, &rbd_opts, &spec);
4866 goto err_out_module;
4868 rbdc = rbd_get_client(ceph_opts);
4873 ceph_opts = NULL; /* rbd_dev client now owns this */
4876 osdc = &rbdc->client->osdc;
4877 rc = ceph_pg_poolid_by_name(osdc->osdmap, spec->pool_name);
4879 goto err_out_client;
4880 spec->pool_id = (u64) rc;
4882 /* The ceph file layout needs to fit pool id in 32 bits */
4884 if (WARN_ON(spec->pool_id > (u64) U32_MAX)) {
4886 goto err_out_client;
4889 rbd_dev = rbd_dev_create(rbdc, spec);
4891 goto err_out_client;
4892 rbdc = NULL; /* rbd_dev now owns this */
4893 spec = NULL; /* rbd_dev now owns this */
4895 rbd_dev->mapping.read_only = rbd_opts->read_only;
4897 rbd_opts = NULL; /* done with this */
4899 rc = rbd_dev_probe(rbd_dev);
4901 goto err_out_rbd_dev;
4905 rbd_dev_destroy(rbd_dev);
4907 rbd_put_client(rbdc);
4910 ceph_destroy_options(ceph_opts);
4914 module_put(THIS_MODULE);
4916 dout("Error adding device %s\n", buf);
4918 return (ssize_t) rc;
4921 static struct rbd_device *__rbd_get_dev(unsigned long dev_id)
4923 struct list_head *tmp;
4924 struct rbd_device *rbd_dev;
4926 spin_lock(&rbd_dev_list_lock);
4927 list_for_each(tmp, &rbd_dev_list) {
4928 rbd_dev = list_entry(tmp, struct rbd_device, node);
4929 if (rbd_dev->dev_id == dev_id) {
4930 spin_unlock(&rbd_dev_list_lock);
4934 spin_unlock(&rbd_dev_list_lock);
4938 static void rbd_dev_release(struct device *dev)
4940 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
4942 if (rbd_dev->watch_event)
4943 rbd_dev_header_watch_sync(rbd_dev, 0);
4945 /* clean up and free blkdev */
4946 rbd_free_disk(rbd_dev);
4947 unregister_blkdev(rbd_dev->major, rbd_dev->name);
4949 /* release allocated disk header fields */
4950 rbd_header_free(&rbd_dev->header);
4952 /* done with the id, and with the rbd_dev */
4953 rbd_dev_id_put(rbd_dev);
4954 rbd_assert(rbd_dev->rbd_client != NULL);
4955 rbd_dev_destroy(rbd_dev);
4957 /* release module ref */
4958 module_put(THIS_MODULE);
4961 static void __rbd_remove(struct rbd_device *rbd_dev)
4963 rbd_remove_all_snaps(rbd_dev);
4964 rbd_bus_del_dev(rbd_dev);
4967 static ssize_t rbd_remove(struct bus_type *bus,
4971 struct rbd_device *rbd_dev = NULL;
4976 rc = strict_strtoul(buf, 10, &ul);
4980 /* convert to int; abort if we lost anything in the conversion */
4981 target_id = (int) ul;
4982 if (target_id != ul)
4985 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
4987 rbd_dev = __rbd_get_dev(target_id);
4993 spin_lock_irq(&rbd_dev->lock);
4994 if (rbd_dev->open_count)
4997 set_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags);
4998 spin_unlock_irq(&rbd_dev->lock);
5002 while (rbd_dev->parent_spec) {
5003 struct rbd_device *first = rbd_dev;
5004 struct rbd_device *second = first->parent;
5005 struct rbd_device *third;
5008 * Follow to the parent with no grandparent and
5011 while (second && (third = second->parent)) {
5015 __rbd_remove(second);
5016 rbd_spec_put(first->parent_spec);
5017 first->parent_spec = NULL;
5018 first->parent_overlap = 0;
5019 first->parent = NULL;
5021 __rbd_remove(rbd_dev);
5024 mutex_unlock(&ctl_mutex);
5030 * create control files in sysfs
5033 static int rbd_sysfs_init(void)
5037 ret = device_register(&rbd_root_dev);
5041 ret = bus_register(&rbd_bus_type);
5043 device_unregister(&rbd_root_dev);
5048 static void rbd_sysfs_cleanup(void)
5050 bus_unregister(&rbd_bus_type);
5051 device_unregister(&rbd_root_dev);
5054 static int __init rbd_init(void)
5058 if (!libceph_compatible(NULL)) {
5059 rbd_warn(NULL, "libceph incompatibility (quitting)");
5063 rc = rbd_sysfs_init();
5066 pr_info("loaded " RBD_DRV_NAME_LONG "\n");
5070 static void __exit rbd_exit(void)
5072 rbd_sysfs_cleanup();
5075 module_init(rbd_init);
5076 module_exit(rbd_exit);
5078 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
5079 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
5080 MODULE_DESCRIPTION("rados block device");
5082 /* following authorship retained from original osdblk.c */
5083 MODULE_AUTHOR("Jeff Garzik <jeff@garzik.org>");
5085 MODULE_LICENSE("GPL");