3 rbd.c -- Export ceph rados objects as a Linux block device
6 based on drivers/block/osdblk.c:
8 Copyright 2009 Red Hat, Inc.
10 This program is free software; you can redistribute it and/or modify
11 it under the terms of the GNU General Public License as published by
12 the Free Software Foundation.
14 This program is distributed in the hope that it will be useful,
15 but WITHOUT ANY WARRANTY; without even the implied warranty of
16 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
17 GNU General Public License for more details.
19 You should have received a copy of the GNU General Public License
20 along with this program; see the file COPYING. If not, write to
21 the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA.
25 For usage instructions, please refer to:
27 Documentation/ABI/testing/sysfs-bus-rbd
31 #include <linux/ceph/libceph.h>
32 #include <linux/ceph/osd_client.h>
33 #include <linux/ceph/mon_client.h>
34 #include <linux/ceph/decode.h>
35 #include <linux/parser.h>
36 #include <linux/bsearch.h>
38 #include <linux/kernel.h>
39 #include <linux/device.h>
40 #include <linux/module.h>
42 #include <linux/blkdev.h>
43 #include <linux/slab.h>
44 #include <linux/idr.h>
46 #include "rbd_types.h"
48 #define RBD_DEBUG /* Activate rbd_assert() calls */
51 * The basic unit of block I/O is a sector. It is interpreted in a
52 * number of contexts in Linux (blk, bio, genhd), but the default is
53 * universally 512 bytes. These symbols are just slightly more
54 * meaningful than the bare numbers they represent.
56 #define SECTOR_SHIFT 9
57 #define SECTOR_SIZE (1ULL << SECTOR_SHIFT)
60 * Increment the given counter and return its updated value.
61 * If the counter is already 0 it will not be incremented.
62 * If the counter is already at its maximum value returns
63 * -EINVAL without updating it.
65 static int atomic_inc_return_safe(atomic_t *v)
69 counter = (unsigned int)__atomic_add_unless(v, 1, 0);
70 if (counter <= (unsigned int)INT_MAX)
78 /* Decrement the counter. Return the resulting value, or -EINVAL */
79 static int atomic_dec_return_safe(atomic_t *v)
83 counter = atomic_dec_return(v);
92 #define RBD_DRV_NAME "rbd"
94 #define RBD_MINORS_PER_MAJOR 256
95 #define RBD_SINGLE_MAJOR_PART_SHIFT 4
97 #define RBD_SNAP_DEV_NAME_PREFIX "snap_"
98 #define RBD_MAX_SNAP_NAME_LEN \
99 (NAME_MAX - (sizeof (RBD_SNAP_DEV_NAME_PREFIX) - 1))
101 #define RBD_MAX_SNAP_COUNT 510 /* allows max snapc to fit in 4KB */
103 #define RBD_SNAP_HEAD_NAME "-"
105 #define BAD_SNAP_INDEX U32_MAX /* invalid index into snap array */
107 /* This allows a single page to hold an image name sent by OSD */
108 #define RBD_IMAGE_NAME_LEN_MAX (PAGE_SIZE - sizeof (__le32) - 1)
109 #define RBD_IMAGE_ID_LEN_MAX 64
111 #define RBD_OBJ_PREFIX_LEN_MAX 64
115 #define RBD_FEATURE_LAYERING (1<<0)
116 #define RBD_FEATURE_STRIPINGV2 (1<<1)
117 #define RBD_FEATURES_ALL \
118 (RBD_FEATURE_LAYERING | RBD_FEATURE_STRIPINGV2)
120 /* Features supported by this (client software) implementation. */
122 #define RBD_FEATURES_SUPPORTED (RBD_FEATURES_ALL)
125 * An RBD device name will be "rbd#", where the "rbd" comes from
126 * RBD_DRV_NAME above, and # is a unique integer identifier.
127 * MAX_INT_FORMAT_WIDTH is used in ensuring DEV_NAME_LEN is big
128 * enough to hold all possible device names.
130 #define DEV_NAME_LEN 32
131 #define MAX_INT_FORMAT_WIDTH ((5 * sizeof (int)) / 2 + 1)
134 * block device image metadata (in-memory version)
136 struct rbd_image_header {
137 /* These six fields never change for a given rbd image */
144 u64 features; /* Might be changeable someday? */
146 /* The remaining fields need to be updated occasionally */
148 struct ceph_snap_context *snapc;
149 char *snap_names; /* format 1 only */
150 u64 *snap_sizes; /* format 1 only */
154 * An rbd image specification.
156 * The tuple (pool_id, image_id, snap_id) is sufficient to uniquely
157 * identify an image. Each rbd_dev structure includes a pointer to
158 * an rbd_spec structure that encapsulates this identity.
160 * Each of the id's in an rbd_spec has an associated name. For a
161 * user-mapped image, the names are supplied and the id's associated
162 * with them are looked up. For a layered image, a parent image is
163 * defined by the tuple, and the names are looked up.
165 * An rbd_dev structure contains a parent_spec pointer which is
166 * non-null if the image it represents is a child in a layered
167 * image. This pointer will refer to the rbd_spec structure used
168 * by the parent rbd_dev for its own identity (i.e., the structure
169 * is shared between the parent and child).
171 * Since these structures are populated once, during the discovery
172 * phase of image construction, they are effectively immutable so
173 * we make no effort to synchronize access to them.
175 * Note that code herein does not assume the image name is known (it
176 * could be a null pointer).
180 const char *pool_name;
182 const char *image_id;
183 const char *image_name;
186 const char *snap_name;
192 * an instance of the client. multiple devices may share an rbd client.
195 struct ceph_client *client;
197 struct list_head node;
200 struct rbd_img_request;
201 typedef void (*rbd_img_callback_t)(struct rbd_img_request *);
203 #define BAD_WHICH U32_MAX /* Good which or bad which, which? */
205 struct rbd_obj_request;
206 typedef void (*rbd_obj_callback_t)(struct rbd_obj_request *);
208 enum obj_request_type {
209 OBJ_REQUEST_NODATA, OBJ_REQUEST_BIO, OBJ_REQUEST_PAGES
213 OBJ_REQ_DONE, /* completion flag: not done = 0, done = 1 */
214 OBJ_REQ_IMG_DATA, /* object usage: standalone = 0, image = 1 */
215 OBJ_REQ_KNOWN, /* EXISTS flag valid: no = 0, yes = 1 */
216 OBJ_REQ_EXISTS, /* target exists: no = 0, yes = 1 */
219 struct rbd_obj_request {
220 const char *object_name;
221 u64 offset; /* object start byte */
222 u64 length; /* bytes from offset */
226 * An object request associated with an image will have its
227 * img_data flag set; a standalone object request will not.
229 * A standalone object request will have which == BAD_WHICH
230 * and a null obj_request pointer.
232 * An object request initiated in support of a layered image
233 * object (to check for its existence before a write) will
234 * have which == BAD_WHICH and a non-null obj_request pointer.
236 * Finally, an object request for rbd image data will have
237 * which != BAD_WHICH, and will have a non-null img_request
238 * pointer. The value of which will be in the range
239 * 0..(img_request->obj_request_count-1).
242 struct rbd_obj_request *obj_request; /* STAT op */
244 struct rbd_img_request *img_request;
246 /* links for img_request->obj_requests list */
247 struct list_head links;
250 u32 which; /* posn image request list */
252 enum obj_request_type type;
254 struct bio *bio_list;
260 struct page **copyup_pages;
261 u32 copyup_page_count;
263 struct ceph_osd_request *osd_req;
265 u64 xferred; /* bytes transferred */
268 rbd_obj_callback_t callback;
269 struct completion completion;
275 IMG_REQ_WRITE, /* I/O direction: read = 0, write = 1 */
276 IMG_REQ_CHILD, /* initiator: block = 0, child image = 1 */
277 IMG_REQ_LAYERED, /* ENOENT handling: normal = 0, layered = 1 */
280 struct rbd_img_request {
281 struct rbd_device *rbd_dev;
282 u64 offset; /* starting image byte offset */
283 u64 length; /* byte count from offset */
286 u64 snap_id; /* for reads */
287 struct ceph_snap_context *snapc; /* for writes */
290 struct request *rq; /* block request */
291 struct rbd_obj_request *obj_request; /* obj req initiator */
293 struct page **copyup_pages;
294 u32 copyup_page_count;
295 spinlock_t completion_lock;/* protects next_completion */
297 rbd_img_callback_t callback;
298 u64 xferred;/* aggregate bytes transferred */
299 int result; /* first nonzero obj_request result */
301 u32 obj_request_count;
302 struct list_head obj_requests; /* rbd_obj_request structs */
307 #define for_each_obj_request(ireq, oreq) \
308 list_for_each_entry(oreq, &(ireq)->obj_requests, links)
309 #define for_each_obj_request_from(ireq, oreq) \
310 list_for_each_entry_from(oreq, &(ireq)->obj_requests, links)
311 #define for_each_obj_request_safe(ireq, oreq, n) \
312 list_for_each_entry_safe_reverse(oreq, n, &(ireq)->obj_requests, links)
324 int dev_id; /* blkdev unique id */
326 int major; /* blkdev assigned major */
328 struct gendisk *disk; /* blkdev's gendisk and rq */
330 u32 image_format; /* Either 1 or 2 */
331 struct rbd_client *rbd_client;
333 char name[DEV_NAME_LEN]; /* blkdev name, e.g. rbd3 */
335 spinlock_t lock; /* queue, flags, open_count */
337 struct rbd_image_header header;
338 unsigned long flags; /* possibly lock protected */
339 struct rbd_spec *spec;
343 struct ceph_file_layout layout;
345 struct ceph_osd_event *watch_event;
346 struct rbd_obj_request *watch_request;
348 struct rbd_spec *parent_spec;
351 struct rbd_device *parent;
353 /* protects updating the header */
354 struct rw_semaphore header_rwsem;
356 struct rbd_mapping mapping;
358 struct list_head node;
362 unsigned long open_count; /* protected by lock */
366 * Flag bits for rbd_dev->flags. If atomicity is required,
367 * rbd_dev->lock is used to protect access.
369 * Currently, only the "removing" flag (which is coupled with the
370 * "open_count" field) requires atomic access.
373 RBD_DEV_FLAG_EXISTS, /* mapped snapshot has not been deleted */
374 RBD_DEV_FLAG_REMOVING, /* this mapping is being removed */
377 static DEFINE_MUTEX(client_mutex); /* Serialize client creation */
379 static LIST_HEAD(rbd_dev_list); /* devices */
380 static DEFINE_SPINLOCK(rbd_dev_list_lock);
382 static LIST_HEAD(rbd_client_list); /* clients */
383 static DEFINE_SPINLOCK(rbd_client_list_lock);
385 /* Slab caches for frequently-allocated structures */
387 static struct kmem_cache *rbd_img_request_cache;
388 static struct kmem_cache *rbd_obj_request_cache;
389 static struct kmem_cache *rbd_segment_name_cache;
391 static int rbd_major;
392 static DEFINE_IDA(rbd_dev_id_ida);
395 * Default to false for now, as single-major requires >= 0.75 version of
396 * userspace rbd utility.
398 static bool single_major = false;
399 module_param(single_major, bool, S_IRUGO);
400 MODULE_PARM_DESC(single_major, "Use a single major number for all rbd devices (default: false)");
402 static int rbd_img_request_submit(struct rbd_img_request *img_request);
404 static void rbd_dev_device_release(struct device *dev);
406 static ssize_t rbd_add(struct bus_type *bus, const char *buf,
408 static ssize_t rbd_remove(struct bus_type *bus, const char *buf,
410 static ssize_t rbd_add_single_major(struct bus_type *bus, const char *buf,
412 static ssize_t rbd_remove_single_major(struct bus_type *bus, const char *buf,
414 static int rbd_dev_image_probe(struct rbd_device *rbd_dev, bool mapping);
415 static void rbd_spec_put(struct rbd_spec *spec);
417 static int rbd_dev_id_to_minor(int dev_id)
419 return dev_id << RBD_SINGLE_MAJOR_PART_SHIFT;
422 static int minor_to_rbd_dev_id(int minor)
424 return minor >> RBD_SINGLE_MAJOR_PART_SHIFT;
427 static BUS_ATTR(add, S_IWUSR, NULL, rbd_add);
428 static BUS_ATTR(remove, S_IWUSR, NULL, rbd_remove);
429 static BUS_ATTR(add_single_major, S_IWUSR, NULL, rbd_add_single_major);
430 static BUS_ATTR(remove_single_major, S_IWUSR, NULL, rbd_remove_single_major);
432 static struct attribute *rbd_bus_attrs[] = {
434 &bus_attr_remove.attr,
435 &bus_attr_add_single_major.attr,
436 &bus_attr_remove_single_major.attr,
440 static umode_t rbd_bus_is_visible(struct kobject *kobj,
441 struct attribute *attr, int index)
444 (attr == &bus_attr_add_single_major.attr ||
445 attr == &bus_attr_remove_single_major.attr))
451 static const struct attribute_group rbd_bus_group = {
452 .attrs = rbd_bus_attrs,
453 .is_visible = rbd_bus_is_visible,
455 __ATTRIBUTE_GROUPS(rbd_bus);
457 static struct bus_type rbd_bus_type = {
459 .bus_groups = rbd_bus_groups,
462 static void rbd_root_dev_release(struct device *dev)
466 static struct device rbd_root_dev = {
468 .release = rbd_root_dev_release,
471 static __printf(2, 3)
472 void rbd_warn(struct rbd_device *rbd_dev, const char *fmt, ...)
474 struct va_format vaf;
482 printk(KERN_WARNING "%s: %pV\n", RBD_DRV_NAME, &vaf);
483 else if (rbd_dev->disk)
484 printk(KERN_WARNING "%s: %s: %pV\n",
485 RBD_DRV_NAME, rbd_dev->disk->disk_name, &vaf);
486 else if (rbd_dev->spec && rbd_dev->spec->image_name)
487 printk(KERN_WARNING "%s: image %s: %pV\n",
488 RBD_DRV_NAME, rbd_dev->spec->image_name, &vaf);
489 else if (rbd_dev->spec && rbd_dev->spec->image_id)
490 printk(KERN_WARNING "%s: id %s: %pV\n",
491 RBD_DRV_NAME, rbd_dev->spec->image_id, &vaf);
493 printk(KERN_WARNING "%s: rbd_dev %p: %pV\n",
494 RBD_DRV_NAME, rbd_dev, &vaf);
499 #define rbd_assert(expr) \
500 if (unlikely(!(expr))) { \
501 printk(KERN_ERR "\nAssertion failure in %s() " \
503 "\trbd_assert(%s);\n\n", \
504 __func__, __LINE__, #expr); \
507 #else /* !RBD_DEBUG */
508 # define rbd_assert(expr) ((void) 0)
509 #endif /* !RBD_DEBUG */
511 static int rbd_img_obj_request_submit(struct rbd_obj_request *obj_request);
512 static void rbd_img_parent_read(struct rbd_obj_request *obj_request);
513 static void rbd_dev_remove_parent(struct rbd_device *rbd_dev);
515 static int rbd_dev_refresh(struct rbd_device *rbd_dev);
516 static int rbd_dev_v2_header_onetime(struct rbd_device *rbd_dev);
517 static int rbd_dev_v2_header_info(struct rbd_device *rbd_dev);
518 static const char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev,
520 static int _rbd_dev_v2_snap_size(struct rbd_device *rbd_dev, u64 snap_id,
521 u8 *order, u64 *snap_size);
522 static int _rbd_dev_v2_snap_features(struct rbd_device *rbd_dev, u64 snap_id,
524 static u64 rbd_snap_id_by_name(struct rbd_device *rbd_dev, const char *name);
526 static int rbd_open(struct block_device *bdev, fmode_t mode)
528 struct rbd_device *rbd_dev = bdev->bd_disk->private_data;
529 bool removing = false;
531 if ((mode & FMODE_WRITE) && rbd_dev->mapping.read_only)
534 spin_lock_irq(&rbd_dev->lock);
535 if (test_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags))
538 rbd_dev->open_count++;
539 spin_unlock_irq(&rbd_dev->lock);
543 (void) get_device(&rbd_dev->dev);
544 set_device_ro(bdev, rbd_dev->mapping.read_only);
549 static void rbd_release(struct gendisk *disk, fmode_t mode)
551 struct rbd_device *rbd_dev = disk->private_data;
552 unsigned long open_count_before;
554 spin_lock_irq(&rbd_dev->lock);
555 open_count_before = rbd_dev->open_count--;
556 spin_unlock_irq(&rbd_dev->lock);
557 rbd_assert(open_count_before > 0);
559 put_device(&rbd_dev->dev);
562 static const struct block_device_operations rbd_bd_ops = {
563 .owner = THIS_MODULE,
565 .release = rbd_release,
569 * Initialize an rbd client instance. Success or not, this function
570 * consumes ceph_opts. Caller holds client_mutex.
572 static struct rbd_client *rbd_client_create(struct ceph_options *ceph_opts)
574 struct rbd_client *rbdc;
577 dout("%s:\n", __func__);
578 rbdc = kmalloc(sizeof(struct rbd_client), GFP_KERNEL);
582 kref_init(&rbdc->kref);
583 INIT_LIST_HEAD(&rbdc->node);
585 rbdc->client = ceph_create_client(ceph_opts, rbdc, 0, 0);
586 if (IS_ERR(rbdc->client))
588 ceph_opts = NULL; /* Now rbdc->client is responsible for ceph_opts */
590 ret = ceph_open_session(rbdc->client);
594 spin_lock(&rbd_client_list_lock);
595 list_add_tail(&rbdc->node, &rbd_client_list);
596 spin_unlock(&rbd_client_list_lock);
598 dout("%s: rbdc %p\n", __func__, rbdc);
602 ceph_destroy_client(rbdc->client);
607 ceph_destroy_options(ceph_opts);
608 dout("%s: error %d\n", __func__, ret);
613 static struct rbd_client *__rbd_get_client(struct rbd_client *rbdc)
615 kref_get(&rbdc->kref);
621 * Find a ceph client with specific addr and configuration. If
622 * found, bump its reference count.
624 static struct rbd_client *rbd_client_find(struct ceph_options *ceph_opts)
626 struct rbd_client *client_node;
629 if (ceph_opts->flags & CEPH_OPT_NOSHARE)
632 spin_lock(&rbd_client_list_lock);
633 list_for_each_entry(client_node, &rbd_client_list, node) {
634 if (!ceph_compare_options(ceph_opts, client_node->client)) {
635 __rbd_get_client(client_node);
641 spin_unlock(&rbd_client_list_lock);
643 return found ? client_node : NULL;
653 /* string args above */
656 /* Boolean args above */
660 static match_table_t rbd_opts_tokens = {
662 /* string args above */
663 {Opt_read_only, "read_only"},
664 {Opt_read_only, "ro"}, /* Alternate spelling */
665 {Opt_read_write, "read_write"},
666 {Opt_read_write, "rw"}, /* Alternate spelling */
667 /* Boolean args above */
675 #define RBD_READ_ONLY_DEFAULT false
677 static int parse_rbd_opts_token(char *c, void *private)
679 struct rbd_options *rbd_opts = private;
680 substring_t argstr[MAX_OPT_ARGS];
681 int token, intval, ret;
683 token = match_token(c, rbd_opts_tokens, argstr);
687 if (token < Opt_last_int) {
688 ret = match_int(&argstr[0], &intval);
690 pr_err("bad mount option arg (not int) "
694 dout("got int token %d val %d\n", token, intval);
695 } else if (token > Opt_last_int && token < Opt_last_string) {
696 dout("got string token %d val %s\n", token,
698 } else if (token > Opt_last_string && token < Opt_last_bool) {
699 dout("got Boolean token %d\n", token);
701 dout("got token %d\n", token);
706 rbd_opts->read_only = true;
709 rbd_opts->read_only = false;
719 * Get a ceph client with specific addr and configuration, if one does
720 * not exist create it. Either way, ceph_opts is consumed by this
723 static struct rbd_client *rbd_get_client(struct ceph_options *ceph_opts)
725 struct rbd_client *rbdc;
727 mutex_lock_nested(&client_mutex, SINGLE_DEPTH_NESTING);
728 rbdc = rbd_client_find(ceph_opts);
729 if (rbdc) /* using an existing client */
730 ceph_destroy_options(ceph_opts);
732 rbdc = rbd_client_create(ceph_opts);
733 mutex_unlock(&client_mutex);
739 * Destroy ceph client
741 * Caller must hold rbd_client_list_lock.
743 static void rbd_client_release(struct kref *kref)
745 struct rbd_client *rbdc = container_of(kref, struct rbd_client, kref);
747 dout("%s: rbdc %p\n", __func__, rbdc);
748 spin_lock(&rbd_client_list_lock);
749 list_del(&rbdc->node);
750 spin_unlock(&rbd_client_list_lock);
752 ceph_destroy_client(rbdc->client);
757 * Drop reference to ceph client node. If it's not referenced anymore, release
760 static void rbd_put_client(struct rbd_client *rbdc)
763 kref_put(&rbdc->kref, rbd_client_release);
766 static bool rbd_image_format_valid(u32 image_format)
768 return image_format == 1 || image_format == 2;
771 static bool rbd_dev_ondisk_valid(struct rbd_image_header_ondisk *ondisk)
776 /* The header has to start with the magic rbd header text */
777 if (memcmp(&ondisk->text, RBD_HEADER_TEXT, sizeof (RBD_HEADER_TEXT)))
780 /* The bio layer requires at least sector-sized I/O */
782 if (ondisk->options.order < SECTOR_SHIFT)
785 /* If we use u64 in a few spots we may be able to loosen this */
787 if (ondisk->options.order > 8 * sizeof (int) - 1)
791 * The size of a snapshot header has to fit in a size_t, and
792 * that limits the number of snapshots.
794 snap_count = le32_to_cpu(ondisk->snap_count);
795 size = SIZE_MAX - sizeof (struct ceph_snap_context);
796 if (snap_count > size / sizeof (__le64))
800 * Not only that, but the size of the entire the snapshot
801 * header must also be representable in a size_t.
803 size -= snap_count * sizeof (__le64);
804 if ((u64) size < le64_to_cpu(ondisk->snap_names_len))
811 * Fill an rbd image header with information from the given format 1
814 static int rbd_header_from_disk(struct rbd_device *rbd_dev,
815 struct rbd_image_header_ondisk *ondisk)
817 struct rbd_image_header *header = &rbd_dev->header;
818 bool first_time = header->object_prefix == NULL;
819 struct ceph_snap_context *snapc;
820 char *object_prefix = NULL;
821 char *snap_names = NULL;
822 u64 *snap_sizes = NULL;
828 /* Allocate this now to avoid having to handle failure below */
833 len = strnlen(ondisk->object_prefix,
834 sizeof (ondisk->object_prefix));
835 object_prefix = kmalloc(len + 1, GFP_KERNEL);
838 memcpy(object_prefix, ondisk->object_prefix, len);
839 object_prefix[len] = '\0';
842 /* Allocate the snapshot context and fill it in */
844 snap_count = le32_to_cpu(ondisk->snap_count);
845 snapc = ceph_create_snap_context(snap_count, GFP_KERNEL);
848 snapc->seq = le64_to_cpu(ondisk->snap_seq);
850 struct rbd_image_snap_ondisk *snaps;
851 u64 snap_names_len = le64_to_cpu(ondisk->snap_names_len);
853 /* We'll keep a copy of the snapshot names... */
855 if (snap_names_len > (u64)SIZE_MAX)
857 snap_names = kmalloc(snap_names_len, GFP_KERNEL);
861 /* ...as well as the array of their sizes. */
863 size = snap_count * sizeof (*header->snap_sizes);
864 snap_sizes = kmalloc(size, GFP_KERNEL);
869 * Copy the names, and fill in each snapshot's id
872 * Note that rbd_dev_v1_header_info() guarantees the
873 * ondisk buffer we're working with has
874 * snap_names_len bytes beyond the end of the
875 * snapshot id array, this memcpy() is safe.
877 memcpy(snap_names, &ondisk->snaps[snap_count], snap_names_len);
878 snaps = ondisk->snaps;
879 for (i = 0; i < snap_count; i++) {
880 snapc->snaps[i] = le64_to_cpu(snaps[i].id);
881 snap_sizes[i] = le64_to_cpu(snaps[i].image_size);
885 /* We won't fail any more, fill in the header */
888 header->object_prefix = object_prefix;
889 header->obj_order = ondisk->options.order;
890 header->crypt_type = ondisk->options.crypt_type;
891 header->comp_type = ondisk->options.comp_type;
892 /* The rest aren't used for format 1 images */
893 header->stripe_unit = 0;
894 header->stripe_count = 0;
895 header->features = 0;
897 ceph_put_snap_context(header->snapc);
898 kfree(header->snap_names);
899 kfree(header->snap_sizes);
902 /* The remaining fields always get updated (when we refresh) */
904 header->image_size = le64_to_cpu(ondisk->image_size);
905 header->snapc = snapc;
906 header->snap_names = snap_names;
907 header->snap_sizes = snap_sizes;
909 /* Make sure mapping size is consistent with header info */
911 if (rbd_dev->spec->snap_id == CEPH_NOSNAP || first_time)
912 if (rbd_dev->mapping.size != header->image_size)
913 rbd_dev->mapping.size = header->image_size;
921 ceph_put_snap_context(snapc);
922 kfree(object_prefix);
927 static const char *_rbd_dev_v1_snap_name(struct rbd_device *rbd_dev, u32 which)
929 const char *snap_name;
931 rbd_assert(which < rbd_dev->header.snapc->num_snaps);
933 /* Skip over names until we find the one we are looking for */
935 snap_name = rbd_dev->header.snap_names;
937 snap_name += strlen(snap_name) + 1;
939 return kstrdup(snap_name, GFP_KERNEL);
943 * Snapshot id comparison function for use with qsort()/bsearch().
944 * Note that result is for snapshots in *descending* order.
946 static int snapid_compare_reverse(const void *s1, const void *s2)
948 u64 snap_id1 = *(u64 *)s1;
949 u64 snap_id2 = *(u64 *)s2;
951 if (snap_id1 < snap_id2)
953 return snap_id1 == snap_id2 ? 0 : -1;
957 * Search a snapshot context to see if the given snapshot id is
960 * Returns the position of the snapshot id in the array if it's found,
961 * or BAD_SNAP_INDEX otherwise.
963 * Note: The snapshot array is in kept sorted (by the osd) in
964 * reverse order, highest snapshot id first.
966 static u32 rbd_dev_snap_index(struct rbd_device *rbd_dev, u64 snap_id)
968 struct ceph_snap_context *snapc = rbd_dev->header.snapc;
971 found = bsearch(&snap_id, &snapc->snaps, snapc->num_snaps,
972 sizeof (snap_id), snapid_compare_reverse);
974 return found ? (u32)(found - &snapc->snaps[0]) : BAD_SNAP_INDEX;
977 static const char *rbd_dev_v1_snap_name(struct rbd_device *rbd_dev,
981 const char *snap_name;
983 which = rbd_dev_snap_index(rbd_dev, snap_id);
984 if (which == BAD_SNAP_INDEX)
985 return ERR_PTR(-ENOENT);
987 snap_name = _rbd_dev_v1_snap_name(rbd_dev, which);
988 return snap_name ? snap_name : ERR_PTR(-ENOMEM);
991 static const char *rbd_snap_name(struct rbd_device *rbd_dev, u64 snap_id)
993 if (snap_id == CEPH_NOSNAP)
994 return RBD_SNAP_HEAD_NAME;
996 rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
997 if (rbd_dev->image_format == 1)
998 return rbd_dev_v1_snap_name(rbd_dev, snap_id);
1000 return rbd_dev_v2_snap_name(rbd_dev, snap_id);
1003 static int rbd_snap_size(struct rbd_device *rbd_dev, u64 snap_id,
1006 rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
1007 if (snap_id == CEPH_NOSNAP) {
1008 *snap_size = rbd_dev->header.image_size;
1009 } else if (rbd_dev->image_format == 1) {
1012 which = rbd_dev_snap_index(rbd_dev, snap_id);
1013 if (which == BAD_SNAP_INDEX)
1016 *snap_size = rbd_dev->header.snap_sizes[which];
1021 ret = _rbd_dev_v2_snap_size(rbd_dev, snap_id, NULL, &size);
1030 static int rbd_snap_features(struct rbd_device *rbd_dev, u64 snap_id,
1033 rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
1034 if (snap_id == CEPH_NOSNAP) {
1035 *snap_features = rbd_dev->header.features;
1036 } else if (rbd_dev->image_format == 1) {
1037 *snap_features = 0; /* No features for format 1 */
1042 ret = _rbd_dev_v2_snap_features(rbd_dev, snap_id, &features);
1046 *snap_features = features;
1051 static int rbd_dev_mapping_set(struct rbd_device *rbd_dev)
1053 u64 snap_id = rbd_dev->spec->snap_id;
1058 ret = rbd_snap_size(rbd_dev, snap_id, &size);
1061 ret = rbd_snap_features(rbd_dev, snap_id, &features);
1065 rbd_dev->mapping.size = size;
1066 rbd_dev->mapping.features = features;
1071 static void rbd_dev_mapping_clear(struct rbd_device *rbd_dev)
1073 rbd_dev->mapping.size = 0;
1074 rbd_dev->mapping.features = 0;
1077 static const char *rbd_segment_name(struct rbd_device *rbd_dev, u64 offset)
1084 name = kmem_cache_alloc(rbd_segment_name_cache, GFP_NOIO);
1087 segment = offset >> rbd_dev->header.obj_order;
1088 name_format = "%s.%012llx";
1089 if (rbd_dev->image_format == 2)
1090 name_format = "%s.%016llx";
1091 ret = snprintf(name, CEPH_MAX_OID_NAME_LEN + 1, name_format,
1092 rbd_dev->header.object_prefix, segment);
1093 if (ret < 0 || ret > CEPH_MAX_OID_NAME_LEN) {
1094 pr_err("error formatting segment name for #%llu (%d)\n",
1103 static void rbd_segment_name_free(const char *name)
1105 /* The explicit cast here is needed to drop the const qualifier */
1107 kmem_cache_free(rbd_segment_name_cache, (void *)name);
1110 static u64 rbd_segment_offset(struct rbd_device *rbd_dev, u64 offset)
1112 u64 segment_size = (u64) 1 << rbd_dev->header.obj_order;
1114 return offset & (segment_size - 1);
1117 static u64 rbd_segment_length(struct rbd_device *rbd_dev,
1118 u64 offset, u64 length)
1120 u64 segment_size = (u64) 1 << rbd_dev->header.obj_order;
1122 offset &= segment_size - 1;
1124 rbd_assert(length <= U64_MAX - offset);
1125 if (offset + length > segment_size)
1126 length = segment_size - offset;
1132 * returns the size of an object in the image
1134 static u64 rbd_obj_bytes(struct rbd_image_header *header)
1136 return 1 << header->obj_order;
1143 static void bio_chain_put(struct bio *chain)
1149 chain = chain->bi_next;
1155 * zeros a bio chain, starting at specific offset
1157 static void zero_bio_chain(struct bio *chain, int start_ofs)
1160 unsigned long flags;
1166 bio_for_each_segment(bv, chain, i) {
1167 if (pos + bv->bv_len > start_ofs) {
1168 int remainder = max(start_ofs - pos, 0);
1169 buf = bvec_kmap_irq(bv, &flags);
1170 memset(buf + remainder, 0,
1171 bv->bv_len - remainder);
1172 flush_dcache_page(bv->bv_page);
1173 bvec_kunmap_irq(buf, &flags);
1178 chain = chain->bi_next;
1183 * similar to zero_bio_chain(), zeros data defined by a page array,
1184 * starting at the given byte offset from the start of the array and
1185 * continuing up to the given end offset. The pages array is
1186 * assumed to be big enough to hold all bytes up to the end.
1188 static void zero_pages(struct page **pages, u64 offset, u64 end)
1190 struct page **page = &pages[offset >> PAGE_SHIFT];
1192 rbd_assert(end > offset);
1193 rbd_assert(end - offset <= (u64)SIZE_MAX);
1194 while (offset < end) {
1197 unsigned long flags;
1200 page_offset = offset & ~PAGE_MASK;
1201 length = min_t(size_t, PAGE_SIZE - page_offset, end - offset);
1202 local_irq_save(flags);
1203 kaddr = kmap_atomic(*page);
1204 memset(kaddr + page_offset, 0, length);
1205 flush_dcache_page(*page);
1206 kunmap_atomic(kaddr);
1207 local_irq_restore(flags);
1215 * Clone a portion of a bio, starting at the given byte offset
1216 * and continuing for the number of bytes indicated.
1218 static struct bio *bio_clone_range(struct bio *bio_src,
1219 unsigned int offset,
1227 unsigned short end_idx;
1228 unsigned short vcnt;
1231 /* Handle the easy case for the caller */
1233 if (!offset && len == bio_src->bi_size)
1234 return bio_clone(bio_src, gfpmask);
1236 if (WARN_ON_ONCE(!len))
1238 if (WARN_ON_ONCE(len > bio_src->bi_size))
1240 if (WARN_ON_ONCE(offset > bio_src->bi_size - len))
1243 /* Find first affected segment... */
1246 bio_for_each_segment(bv, bio_src, idx) {
1247 if (resid < bv->bv_len)
1249 resid -= bv->bv_len;
1253 /* ...and the last affected segment */
1256 __bio_for_each_segment(bv, bio_src, end_idx, idx) {
1257 if (resid <= bv->bv_len)
1259 resid -= bv->bv_len;
1261 vcnt = end_idx - idx + 1;
1263 /* Build the clone */
1265 bio = bio_alloc(gfpmask, (unsigned int) vcnt);
1267 return NULL; /* ENOMEM */
1269 bio->bi_bdev = bio_src->bi_bdev;
1270 bio->bi_sector = bio_src->bi_sector + (offset >> SECTOR_SHIFT);
1271 bio->bi_rw = bio_src->bi_rw;
1272 bio->bi_flags |= 1 << BIO_CLONED;
1275 * Copy over our part of the bio_vec, then update the first
1276 * and last (or only) entries.
1278 memcpy(&bio->bi_io_vec[0], &bio_src->bi_io_vec[idx],
1279 vcnt * sizeof (struct bio_vec));
1280 bio->bi_io_vec[0].bv_offset += voff;
1282 bio->bi_io_vec[0].bv_len -= voff;
1283 bio->bi_io_vec[vcnt - 1].bv_len = resid;
1285 bio->bi_io_vec[0].bv_len = len;
1288 bio->bi_vcnt = vcnt;
1296 * Clone a portion of a bio chain, starting at the given byte offset
1297 * into the first bio in the source chain and continuing for the
1298 * number of bytes indicated. The result is another bio chain of
1299 * exactly the given length, or a null pointer on error.
1301 * The bio_src and offset parameters are both in-out. On entry they
1302 * refer to the first source bio and the offset into that bio where
1303 * the start of data to be cloned is located.
1305 * On return, bio_src is updated to refer to the bio in the source
1306 * chain that contains first un-cloned byte, and *offset will
1307 * contain the offset of that byte within that bio.
1309 static struct bio *bio_chain_clone_range(struct bio **bio_src,
1310 unsigned int *offset,
1314 struct bio *bi = *bio_src;
1315 unsigned int off = *offset;
1316 struct bio *chain = NULL;
1319 /* Build up a chain of clone bios up to the limit */
1321 if (!bi || off >= bi->bi_size || !len)
1322 return NULL; /* Nothing to clone */
1326 unsigned int bi_size;
1330 rbd_warn(NULL, "bio_chain exhausted with %u left", len);
1331 goto out_err; /* EINVAL; ran out of bio's */
1333 bi_size = min_t(unsigned int, bi->bi_size - off, len);
1334 bio = bio_clone_range(bi, off, bi_size, gfpmask);
1336 goto out_err; /* ENOMEM */
1339 end = &bio->bi_next;
1342 if (off == bi->bi_size) {
1353 bio_chain_put(chain);
1359 * The default/initial value for all object request flags is 0. For
1360 * each flag, once its value is set to 1 it is never reset to 0
1363 static void obj_request_img_data_set(struct rbd_obj_request *obj_request)
1365 if (test_and_set_bit(OBJ_REQ_IMG_DATA, &obj_request->flags)) {
1366 struct rbd_device *rbd_dev;
1368 rbd_dev = obj_request->img_request->rbd_dev;
1369 rbd_warn(rbd_dev, "obj_request %p already marked img_data\n",
1374 static bool obj_request_img_data_test(struct rbd_obj_request *obj_request)
1377 return test_bit(OBJ_REQ_IMG_DATA, &obj_request->flags) != 0;
1380 static void obj_request_done_set(struct rbd_obj_request *obj_request)
1382 if (test_and_set_bit(OBJ_REQ_DONE, &obj_request->flags)) {
1383 struct rbd_device *rbd_dev = NULL;
1385 if (obj_request_img_data_test(obj_request))
1386 rbd_dev = obj_request->img_request->rbd_dev;
1387 rbd_warn(rbd_dev, "obj_request %p already marked done\n",
1392 static bool obj_request_done_test(struct rbd_obj_request *obj_request)
1395 return test_bit(OBJ_REQ_DONE, &obj_request->flags) != 0;
1399 * This sets the KNOWN flag after (possibly) setting the EXISTS
1400 * flag. The latter is set based on the "exists" value provided.
1402 * Note that for our purposes once an object exists it never goes
1403 * away again. It's possible that the response from two existence
1404 * checks are separated by the creation of the target object, and
1405 * the first ("doesn't exist") response arrives *after* the second
1406 * ("does exist"). In that case we ignore the second one.
1408 static void obj_request_existence_set(struct rbd_obj_request *obj_request,
1412 set_bit(OBJ_REQ_EXISTS, &obj_request->flags);
1413 set_bit(OBJ_REQ_KNOWN, &obj_request->flags);
1417 static bool obj_request_known_test(struct rbd_obj_request *obj_request)
1420 return test_bit(OBJ_REQ_KNOWN, &obj_request->flags) != 0;
1423 static bool obj_request_exists_test(struct rbd_obj_request *obj_request)
1426 return test_bit(OBJ_REQ_EXISTS, &obj_request->flags) != 0;
1429 static void rbd_obj_request_get(struct rbd_obj_request *obj_request)
1431 dout("%s: obj %p (was %d)\n", __func__, obj_request,
1432 atomic_read(&obj_request->kref.refcount));
1433 kref_get(&obj_request->kref);
1436 static void rbd_obj_request_destroy(struct kref *kref);
1437 static void rbd_obj_request_put(struct rbd_obj_request *obj_request)
1439 rbd_assert(obj_request != NULL);
1440 dout("%s: obj %p (was %d)\n", __func__, obj_request,
1441 atomic_read(&obj_request->kref.refcount));
1442 kref_put(&obj_request->kref, rbd_obj_request_destroy);
1445 static bool img_request_child_test(struct rbd_img_request *img_request);
1446 static void rbd_parent_request_destroy(struct kref *kref);
1447 static void rbd_img_request_destroy(struct kref *kref);
1448 static void rbd_img_request_put(struct rbd_img_request *img_request)
1450 rbd_assert(img_request != NULL);
1451 dout("%s: img %p (was %d)\n", __func__, img_request,
1452 atomic_read(&img_request->kref.refcount));
1453 if (img_request_child_test(img_request))
1454 kref_put(&img_request->kref, rbd_parent_request_destroy);
1456 kref_put(&img_request->kref, rbd_img_request_destroy);
1459 static inline void rbd_img_obj_request_add(struct rbd_img_request *img_request,
1460 struct rbd_obj_request *obj_request)
1462 rbd_assert(obj_request->img_request == NULL);
1464 /* Image request now owns object's original reference */
1465 obj_request->img_request = img_request;
1466 obj_request->which = img_request->obj_request_count;
1467 rbd_assert(!obj_request_img_data_test(obj_request));
1468 obj_request_img_data_set(obj_request);
1469 rbd_assert(obj_request->which != BAD_WHICH);
1470 img_request->obj_request_count++;
1471 list_add_tail(&obj_request->links, &img_request->obj_requests);
1472 dout("%s: img %p obj %p w=%u\n", __func__, img_request, obj_request,
1473 obj_request->which);
1476 static inline void rbd_img_obj_request_del(struct rbd_img_request *img_request,
1477 struct rbd_obj_request *obj_request)
1479 rbd_assert(obj_request->which != BAD_WHICH);
1481 dout("%s: img %p obj %p w=%u\n", __func__, img_request, obj_request,
1482 obj_request->which);
1483 list_del(&obj_request->links);
1484 rbd_assert(img_request->obj_request_count > 0);
1485 img_request->obj_request_count--;
1486 rbd_assert(obj_request->which == img_request->obj_request_count);
1487 obj_request->which = BAD_WHICH;
1488 rbd_assert(obj_request_img_data_test(obj_request));
1489 rbd_assert(obj_request->img_request == img_request);
1490 obj_request->img_request = NULL;
1491 obj_request->callback = NULL;
1492 rbd_obj_request_put(obj_request);
1495 static bool obj_request_type_valid(enum obj_request_type type)
1498 case OBJ_REQUEST_NODATA:
1499 case OBJ_REQUEST_BIO:
1500 case OBJ_REQUEST_PAGES:
1507 static int rbd_obj_request_submit(struct ceph_osd_client *osdc,
1508 struct rbd_obj_request *obj_request)
1510 dout("%s: osdc %p obj %p\n", __func__, osdc, obj_request);
1512 return ceph_osdc_start_request(osdc, obj_request->osd_req, false);
1515 static void rbd_img_request_complete(struct rbd_img_request *img_request)
1518 dout("%s: img %p\n", __func__, img_request);
1521 * If no error occurred, compute the aggregate transfer
1522 * count for the image request. We could instead use
1523 * atomic64_cmpxchg() to update it as each object request
1524 * completes; not clear which way is better off hand.
1526 if (!img_request->result) {
1527 struct rbd_obj_request *obj_request;
1530 for_each_obj_request(img_request, obj_request)
1531 xferred += obj_request->xferred;
1532 img_request->xferred = xferred;
1535 if (img_request->callback)
1536 img_request->callback(img_request);
1538 rbd_img_request_put(img_request);
1541 /* Caller is responsible for rbd_obj_request_destroy(obj_request) */
1543 static int rbd_obj_request_wait(struct rbd_obj_request *obj_request)
1545 dout("%s: obj %p\n", __func__, obj_request);
1547 return wait_for_completion_interruptible(&obj_request->completion);
1551 * The default/initial value for all image request flags is 0. Each
1552 * is conditionally set to 1 at image request initialization time
1553 * and currently never change thereafter.
1555 static void img_request_write_set(struct rbd_img_request *img_request)
1557 set_bit(IMG_REQ_WRITE, &img_request->flags);
1561 static bool img_request_write_test(struct rbd_img_request *img_request)
1564 return test_bit(IMG_REQ_WRITE, &img_request->flags) != 0;
1567 static void img_request_child_set(struct rbd_img_request *img_request)
1569 set_bit(IMG_REQ_CHILD, &img_request->flags);
1573 static void img_request_child_clear(struct rbd_img_request *img_request)
1575 clear_bit(IMG_REQ_CHILD, &img_request->flags);
1579 static bool img_request_child_test(struct rbd_img_request *img_request)
1582 return test_bit(IMG_REQ_CHILD, &img_request->flags) != 0;
1585 static void img_request_layered_set(struct rbd_img_request *img_request)
1587 set_bit(IMG_REQ_LAYERED, &img_request->flags);
1591 static void img_request_layered_clear(struct rbd_img_request *img_request)
1593 clear_bit(IMG_REQ_LAYERED, &img_request->flags);
1597 static bool img_request_layered_test(struct rbd_img_request *img_request)
1600 return test_bit(IMG_REQ_LAYERED, &img_request->flags) != 0;
1604 rbd_img_obj_request_read_callback(struct rbd_obj_request *obj_request)
1606 u64 xferred = obj_request->xferred;
1607 u64 length = obj_request->length;
1609 dout("%s: obj %p img %p result %d %llu/%llu\n", __func__,
1610 obj_request, obj_request->img_request, obj_request->result,
1613 * ENOENT means a hole in the image. We zero-fill the entire
1614 * length of the request. A short read also implies zero-fill
1615 * to the end of the request. An error requires the whole
1616 * length of the request to be reported finished with an error
1617 * to the block layer. In each case we update the xferred
1618 * count to indicate the whole request was satisfied.
1620 rbd_assert(obj_request->type != OBJ_REQUEST_NODATA);
1621 if (obj_request->result == -ENOENT) {
1622 if (obj_request->type == OBJ_REQUEST_BIO)
1623 zero_bio_chain(obj_request->bio_list, 0);
1625 zero_pages(obj_request->pages, 0, length);
1626 obj_request->result = 0;
1627 } else if (xferred < length && !obj_request->result) {
1628 if (obj_request->type == OBJ_REQUEST_BIO)
1629 zero_bio_chain(obj_request->bio_list, xferred);
1631 zero_pages(obj_request->pages, xferred, length);
1633 obj_request->xferred = length;
1634 obj_request_done_set(obj_request);
1637 static void rbd_obj_request_complete(struct rbd_obj_request *obj_request)
1639 dout("%s: obj %p cb %p\n", __func__, obj_request,
1640 obj_request->callback);
1641 if (obj_request->callback)
1642 obj_request->callback(obj_request);
1644 complete_all(&obj_request->completion);
1647 static void rbd_osd_trivial_callback(struct rbd_obj_request *obj_request)
1649 dout("%s: obj %p\n", __func__, obj_request);
1650 obj_request_done_set(obj_request);
1653 static void rbd_osd_read_callback(struct rbd_obj_request *obj_request)
1655 struct rbd_img_request *img_request = NULL;
1656 struct rbd_device *rbd_dev = NULL;
1657 bool layered = false;
1659 if (obj_request_img_data_test(obj_request)) {
1660 img_request = obj_request->img_request;
1661 layered = img_request && img_request_layered_test(img_request);
1662 rbd_dev = img_request->rbd_dev;
1665 dout("%s: obj %p img %p result %d %llu/%llu\n", __func__,
1666 obj_request, img_request, obj_request->result,
1667 obj_request->xferred, obj_request->length);
1668 if (layered && obj_request->result == -ENOENT &&
1669 obj_request->img_offset < rbd_dev->parent_overlap)
1670 rbd_img_parent_read(obj_request);
1671 else if (img_request)
1672 rbd_img_obj_request_read_callback(obj_request);
1674 obj_request_done_set(obj_request);
1677 static void rbd_osd_write_callback(struct rbd_obj_request *obj_request)
1679 dout("%s: obj %p result %d %llu\n", __func__, obj_request,
1680 obj_request->result, obj_request->length);
1682 * There is no such thing as a successful short write. Set
1683 * it to our originally-requested length.
1685 obj_request->xferred = obj_request->length;
1686 obj_request_done_set(obj_request);
1690 * For a simple stat call there's nothing to do. We'll do more if
1691 * this is part of a write sequence for a layered image.
1693 static void rbd_osd_stat_callback(struct rbd_obj_request *obj_request)
1695 dout("%s: obj %p\n", __func__, obj_request);
1696 obj_request_done_set(obj_request);
1699 static void rbd_osd_req_callback(struct ceph_osd_request *osd_req,
1700 struct ceph_msg *msg)
1702 struct rbd_obj_request *obj_request = osd_req->r_priv;
1705 dout("%s: osd_req %p msg %p\n", __func__, osd_req, msg);
1706 rbd_assert(osd_req == obj_request->osd_req);
1707 if (obj_request_img_data_test(obj_request)) {
1708 rbd_assert(obj_request->img_request);
1709 rbd_assert(obj_request->which != BAD_WHICH);
1711 rbd_assert(obj_request->which == BAD_WHICH);
1714 if (osd_req->r_result < 0)
1715 obj_request->result = osd_req->r_result;
1717 BUG_ON(osd_req->r_num_ops > 2);
1720 * We support a 64-bit length, but ultimately it has to be
1721 * passed to blk_end_request(), which takes an unsigned int.
1723 obj_request->xferred = osd_req->r_reply_op_len[0];
1724 rbd_assert(obj_request->xferred < (u64)UINT_MAX);
1725 opcode = osd_req->r_ops[0].op;
1727 case CEPH_OSD_OP_READ:
1728 rbd_osd_read_callback(obj_request);
1730 case CEPH_OSD_OP_WRITE:
1731 rbd_osd_write_callback(obj_request);
1733 case CEPH_OSD_OP_STAT:
1734 rbd_osd_stat_callback(obj_request);
1736 case CEPH_OSD_OP_CALL:
1737 case CEPH_OSD_OP_NOTIFY_ACK:
1738 case CEPH_OSD_OP_WATCH:
1739 rbd_osd_trivial_callback(obj_request);
1742 rbd_warn(NULL, "%s: unsupported op %hu\n",
1743 obj_request->object_name, (unsigned short) opcode);
1747 if (obj_request_done_test(obj_request))
1748 rbd_obj_request_complete(obj_request);
1751 static void rbd_osd_req_format_read(struct rbd_obj_request *obj_request)
1753 struct rbd_img_request *img_request = obj_request->img_request;
1754 struct ceph_osd_request *osd_req = obj_request->osd_req;
1757 rbd_assert(osd_req != NULL);
1759 snap_id = img_request ? img_request->snap_id : CEPH_NOSNAP;
1760 ceph_osdc_build_request(osd_req, obj_request->offset,
1761 NULL, snap_id, NULL);
1764 static void rbd_osd_req_format_write(struct rbd_obj_request *obj_request)
1766 struct rbd_img_request *img_request = obj_request->img_request;
1767 struct ceph_osd_request *osd_req = obj_request->osd_req;
1768 struct ceph_snap_context *snapc;
1769 struct timespec mtime = CURRENT_TIME;
1771 rbd_assert(osd_req != NULL);
1773 snapc = img_request ? img_request->snapc : NULL;
1774 ceph_osdc_build_request(osd_req, obj_request->offset,
1775 snapc, CEPH_NOSNAP, &mtime);
1778 static struct ceph_osd_request *rbd_osd_req_create(
1779 struct rbd_device *rbd_dev,
1781 struct rbd_obj_request *obj_request)
1783 struct ceph_snap_context *snapc = NULL;
1784 struct ceph_osd_client *osdc;
1785 struct ceph_osd_request *osd_req;
1787 if (obj_request_img_data_test(obj_request)) {
1788 struct rbd_img_request *img_request = obj_request->img_request;
1790 rbd_assert(write_request ==
1791 img_request_write_test(img_request));
1793 snapc = img_request->snapc;
1796 /* Allocate and initialize the request, for the single op */
1798 osdc = &rbd_dev->rbd_client->client->osdc;
1799 osd_req = ceph_osdc_alloc_request(osdc, snapc, 1, false, GFP_ATOMIC);
1801 return NULL; /* ENOMEM */
1804 osd_req->r_flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK;
1806 osd_req->r_flags = CEPH_OSD_FLAG_READ;
1808 osd_req->r_callback = rbd_osd_req_callback;
1809 osd_req->r_priv = obj_request;
1811 osd_req->r_base_oloc.pool = ceph_file_layout_pg_pool(rbd_dev->layout);
1812 ceph_oid_set_name(&osd_req->r_base_oid, obj_request->object_name);
1818 * Create a copyup osd request based on the information in the
1819 * object request supplied. A copyup request has two osd ops,
1820 * a copyup method call, and a "normal" write request.
1822 static struct ceph_osd_request *
1823 rbd_osd_req_create_copyup(struct rbd_obj_request *obj_request)
1825 struct rbd_img_request *img_request;
1826 struct ceph_snap_context *snapc;
1827 struct rbd_device *rbd_dev;
1828 struct ceph_osd_client *osdc;
1829 struct ceph_osd_request *osd_req;
1831 rbd_assert(obj_request_img_data_test(obj_request));
1832 img_request = obj_request->img_request;
1833 rbd_assert(img_request);
1834 rbd_assert(img_request_write_test(img_request));
1836 /* Allocate and initialize the request, for the two ops */
1838 snapc = img_request->snapc;
1839 rbd_dev = img_request->rbd_dev;
1840 osdc = &rbd_dev->rbd_client->client->osdc;
1841 osd_req = ceph_osdc_alloc_request(osdc, snapc, 2, false, GFP_ATOMIC);
1843 return NULL; /* ENOMEM */
1845 osd_req->r_flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK;
1846 osd_req->r_callback = rbd_osd_req_callback;
1847 osd_req->r_priv = obj_request;
1849 osd_req->r_base_oloc.pool = ceph_file_layout_pg_pool(rbd_dev->layout);
1850 ceph_oid_set_name(&osd_req->r_base_oid, obj_request->object_name);
1856 static void rbd_osd_req_destroy(struct ceph_osd_request *osd_req)
1858 ceph_osdc_put_request(osd_req);
1861 /* object_name is assumed to be a non-null pointer and NUL-terminated */
1863 static struct rbd_obj_request *rbd_obj_request_create(const char *object_name,
1864 u64 offset, u64 length,
1865 enum obj_request_type type)
1867 struct rbd_obj_request *obj_request;
1871 rbd_assert(obj_request_type_valid(type));
1873 size = strlen(object_name) + 1;
1874 name = kmalloc(size, GFP_KERNEL);
1878 obj_request = kmem_cache_zalloc(rbd_obj_request_cache, GFP_KERNEL);
1884 obj_request->object_name = memcpy(name, object_name, size);
1885 obj_request->offset = offset;
1886 obj_request->length = length;
1887 obj_request->flags = 0;
1888 obj_request->which = BAD_WHICH;
1889 obj_request->type = type;
1890 INIT_LIST_HEAD(&obj_request->links);
1891 init_completion(&obj_request->completion);
1892 kref_init(&obj_request->kref);
1894 dout("%s: \"%s\" %llu/%llu %d -> obj %p\n", __func__, object_name,
1895 offset, length, (int)type, obj_request);
1900 static void rbd_obj_request_destroy(struct kref *kref)
1902 struct rbd_obj_request *obj_request;
1904 obj_request = container_of(kref, struct rbd_obj_request, kref);
1906 dout("%s: obj %p\n", __func__, obj_request);
1908 rbd_assert(obj_request->img_request == NULL);
1909 rbd_assert(obj_request->which == BAD_WHICH);
1911 if (obj_request->osd_req)
1912 rbd_osd_req_destroy(obj_request->osd_req);
1914 rbd_assert(obj_request_type_valid(obj_request->type));
1915 switch (obj_request->type) {
1916 case OBJ_REQUEST_NODATA:
1917 break; /* Nothing to do */
1918 case OBJ_REQUEST_BIO:
1919 if (obj_request->bio_list)
1920 bio_chain_put(obj_request->bio_list);
1922 case OBJ_REQUEST_PAGES:
1923 if (obj_request->pages)
1924 ceph_release_page_vector(obj_request->pages,
1925 obj_request->page_count);
1929 kfree(obj_request->object_name);
1930 obj_request->object_name = NULL;
1931 kmem_cache_free(rbd_obj_request_cache, obj_request);
1934 /* It's OK to call this for a device with no parent */
1936 static void rbd_spec_put(struct rbd_spec *spec);
1937 static void rbd_dev_unparent(struct rbd_device *rbd_dev)
1939 rbd_dev_remove_parent(rbd_dev);
1940 rbd_spec_put(rbd_dev->parent_spec);
1941 rbd_dev->parent_spec = NULL;
1942 rbd_dev->parent_overlap = 0;
1946 * Parent image reference counting is used to determine when an
1947 * image's parent fields can be safely torn down--after there are no
1948 * more in-flight requests to the parent image. When the last
1949 * reference is dropped, cleaning them up is safe.
1951 static void rbd_dev_parent_put(struct rbd_device *rbd_dev)
1955 if (!rbd_dev->parent_spec)
1958 counter = atomic_dec_return_safe(&rbd_dev->parent_ref);
1962 /* Last reference; clean up parent data structures */
1965 rbd_dev_unparent(rbd_dev);
1967 rbd_warn(rbd_dev, "parent reference underflow\n");
1971 * If an image has a non-zero parent overlap, get a reference to its
1974 * We must get the reference before checking for the overlap to
1975 * coordinate properly with zeroing the parent overlap in
1976 * rbd_dev_v2_parent_info() when an image gets flattened. We
1977 * drop it again if there is no overlap.
1979 * Returns true if the rbd device has a parent with a non-zero
1980 * overlap and a reference for it was successfully taken, or
1983 static bool rbd_dev_parent_get(struct rbd_device *rbd_dev)
1987 if (!rbd_dev->parent_spec)
1990 counter = atomic_inc_return_safe(&rbd_dev->parent_ref);
1991 if (counter > 0 && rbd_dev->parent_overlap)
1994 /* Image was flattened, but parent is not yet torn down */
1997 rbd_warn(rbd_dev, "parent reference overflow\n");
2003 * Caller is responsible for filling in the list of object requests
2004 * that comprises the image request, and the Linux request pointer
2005 * (if there is one).
2007 static struct rbd_img_request *rbd_img_request_create(
2008 struct rbd_device *rbd_dev,
2009 u64 offset, u64 length,
2012 struct rbd_img_request *img_request;
2014 img_request = kmem_cache_alloc(rbd_img_request_cache, GFP_ATOMIC);
2018 if (write_request) {
2019 down_read(&rbd_dev->header_rwsem);
2020 ceph_get_snap_context(rbd_dev->header.snapc);
2021 up_read(&rbd_dev->header_rwsem);
2024 img_request->rq = NULL;
2025 img_request->rbd_dev = rbd_dev;
2026 img_request->offset = offset;
2027 img_request->length = length;
2028 img_request->flags = 0;
2029 if (write_request) {
2030 img_request_write_set(img_request);
2031 img_request->snapc = rbd_dev->header.snapc;
2033 img_request->snap_id = rbd_dev->spec->snap_id;
2035 if (rbd_dev_parent_get(rbd_dev))
2036 img_request_layered_set(img_request);
2037 spin_lock_init(&img_request->completion_lock);
2038 img_request->next_completion = 0;
2039 img_request->callback = NULL;
2040 img_request->result = 0;
2041 img_request->obj_request_count = 0;
2042 INIT_LIST_HEAD(&img_request->obj_requests);
2043 kref_init(&img_request->kref);
2045 dout("%s: rbd_dev %p %s %llu/%llu -> img %p\n", __func__, rbd_dev,
2046 write_request ? "write" : "read", offset, length,
2052 static void rbd_img_request_destroy(struct kref *kref)
2054 struct rbd_img_request *img_request;
2055 struct rbd_obj_request *obj_request;
2056 struct rbd_obj_request *next_obj_request;
2058 img_request = container_of(kref, struct rbd_img_request, kref);
2060 dout("%s: img %p\n", __func__, img_request);
2062 for_each_obj_request_safe(img_request, obj_request, next_obj_request)
2063 rbd_img_obj_request_del(img_request, obj_request);
2064 rbd_assert(img_request->obj_request_count == 0);
2066 if (img_request_layered_test(img_request)) {
2067 img_request_layered_clear(img_request);
2068 rbd_dev_parent_put(img_request->rbd_dev);
2071 if (img_request_write_test(img_request))
2072 ceph_put_snap_context(img_request->snapc);
2074 kmem_cache_free(rbd_img_request_cache, img_request);
2077 static struct rbd_img_request *rbd_parent_request_create(
2078 struct rbd_obj_request *obj_request,
2079 u64 img_offset, u64 length)
2081 struct rbd_img_request *parent_request;
2082 struct rbd_device *rbd_dev;
2084 rbd_assert(obj_request->img_request);
2085 rbd_dev = obj_request->img_request->rbd_dev;
2087 parent_request = rbd_img_request_create(rbd_dev->parent,
2088 img_offset, length, false);
2089 if (!parent_request)
2092 img_request_child_set(parent_request);
2093 rbd_obj_request_get(obj_request);
2094 parent_request->obj_request = obj_request;
2096 return parent_request;
2099 static void rbd_parent_request_destroy(struct kref *kref)
2101 struct rbd_img_request *parent_request;
2102 struct rbd_obj_request *orig_request;
2104 parent_request = container_of(kref, struct rbd_img_request, kref);
2105 orig_request = parent_request->obj_request;
2107 parent_request->obj_request = NULL;
2108 rbd_obj_request_put(orig_request);
2109 img_request_child_clear(parent_request);
2111 rbd_img_request_destroy(kref);
2114 static bool rbd_img_obj_end_request(struct rbd_obj_request *obj_request)
2116 struct rbd_img_request *img_request;
2117 unsigned int xferred;
2121 rbd_assert(obj_request_img_data_test(obj_request));
2122 img_request = obj_request->img_request;
2124 rbd_assert(obj_request->xferred <= (u64)UINT_MAX);
2125 xferred = (unsigned int)obj_request->xferred;
2126 result = obj_request->result;
2128 struct rbd_device *rbd_dev = img_request->rbd_dev;
2130 rbd_warn(rbd_dev, "%s %llx at %llx (%llx)\n",
2131 img_request_write_test(img_request) ? "write" : "read",
2132 obj_request->length, obj_request->img_offset,
2133 obj_request->offset);
2134 rbd_warn(rbd_dev, " result %d xferred %x\n",
2136 if (!img_request->result)
2137 img_request->result = result;
2140 /* Image object requests don't own their page array */
2142 if (obj_request->type == OBJ_REQUEST_PAGES) {
2143 obj_request->pages = NULL;
2144 obj_request->page_count = 0;
2147 if (img_request_child_test(img_request)) {
2148 rbd_assert(img_request->obj_request != NULL);
2149 more = obj_request->which < img_request->obj_request_count - 1;
2151 rbd_assert(img_request->rq != NULL);
2152 more = blk_end_request(img_request->rq, result, xferred);
2158 static void rbd_img_obj_callback(struct rbd_obj_request *obj_request)
2160 struct rbd_img_request *img_request;
2161 u32 which = obj_request->which;
2164 rbd_assert(obj_request_img_data_test(obj_request));
2165 img_request = obj_request->img_request;
2167 dout("%s: img %p obj %p\n", __func__, img_request, obj_request);
2168 rbd_assert(img_request != NULL);
2169 rbd_assert(img_request->obj_request_count > 0);
2170 rbd_assert(which != BAD_WHICH);
2171 rbd_assert(which < img_request->obj_request_count);
2172 rbd_assert(which >= img_request->next_completion);
2174 spin_lock_irq(&img_request->completion_lock);
2175 if (which != img_request->next_completion)
2178 for_each_obj_request_from(img_request, obj_request) {
2180 rbd_assert(which < img_request->obj_request_count);
2182 if (!obj_request_done_test(obj_request))
2184 more = rbd_img_obj_end_request(obj_request);
2188 rbd_assert(more ^ (which == img_request->obj_request_count));
2189 img_request->next_completion = which;
2191 spin_unlock_irq(&img_request->completion_lock);
2194 rbd_img_request_complete(img_request);
2198 * Split up an image request into one or more object requests, each
2199 * to a different object. The "type" parameter indicates whether
2200 * "data_desc" is the pointer to the head of a list of bio
2201 * structures, or the base of a page array. In either case this
2202 * function assumes data_desc describes memory sufficient to hold
2203 * all data described by the image request.
2205 static int rbd_img_request_fill(struct rbd_img_request *img_request,
2206 enum obj_request_type type,
2209 struct rbd_device *rbd_dev = img_request->rbd_dev;
2210 struct rbd_obj_request *obj_request = NULL;
2211 struct rbd_obj_request *next_obj_request;
2212 bool write_request = img_request_write_test(img_request);
2213 struct bio *bio_list = NULL;
2214 unsigned int bio_offset = 0;
2215 struct page **pages = NULL;
2220 dout("%s: img %p type %d data_desc %p\n", __func__, img_request,
2221 (int)type, data_desc);
2223 opcode = write_request ? CEPH_OSD_OP_WRITE : CEPH_OSD_OP_READ;
2224 img_offset = img_request->offset;
2225 resid = img_request->length;
2226 rbd_assert(resid > 0);
2228 if (type == OBJ_REQUEST_BIO) {
2229 bio_list = data_desc;
2230 rbd_assert(img_offset == bio_list->bi_sector << SECTOR_SHIFT);
2232 rbd_assert(type == OBJ_REQUEST_PAGES);
2237 struct ceph_osd_request *osd_req;
2238 const char *object_name;
2242 object_name = rbd_segment_name(rbd_dev, img_offset);
2245 offset = rbd_segment_offset(rbd_dev, img_offset);
2246 length = rbd_segment_length(rbd_dev, img_offset, resid);
2247 obj_request = rbd_obj_request_create(object_name,
2248 offset, length, type);
2249 /* object request has its own copy of the object name */
2250 rbd_segment_name_free(object_name);
2254 * set obj_request->img_request before creating the
2255 * osd_request so that it gets the right snapc
2257 rbd_img_obj_request_add(img_request, obj_request);
2259 if (type == OBJ_REQUEST_BIO) {
2260 unsigned int clone_size;
2262 rbd_assert(length <= (u64)UINT_MAX);
2263 clone_size = (unsigned int)length;
2264 obj_request->bio_list =
2265 bio_chain_clone_range(&bio_list,
2269 if (!obj_request->bio_list)
2272 unsigned int page_count;
2274 obj_request->pages = pages;
2275 page_count = (u32)calc_pages_for(offset, length);
2276 obj_request->page_count = page_count;
2277 if ((offset + length) & ~PAGE_MASK)
2278 page_count--; /* more on last page */
2279 pages += page_count;
2282 osd_req = rbd_osd_req_create(rbd_dev, write_request,
2286 obj_request->osd_req = osd_req;
2287 obj_request->callback = rbd_img_obj_callback;
2289 osd_req_op_extent_init(osd_req, 0, opcode, offset, length,
2291 if (type == OBJ_REQUEST_BIO)
2292 osd_req_op_extent_osd_data_bio(osd_req, 0,
2293 obj_request->bio_list, length);
2295 osd_req_op_extent_osd_data_pages(osd_req, 0,
2296 obj_request->pages, length,
2297 offset & ~PAGE_MASK, false, false);
2300 rbd_osd_req_format_write(obj_request);
2302 rbd_osd_req_format_read(obj_request);
2304 obj_request->img_offset = img_offset;
2306 img_offset += length;
2313 rbd_obj_request_put(obj_request);
2315 for_each_obj_request_safe(img_request, obj_request, next_obj_request)
2316 rbd_obj_request_put(obj_request);
2322 rbd_img_obj_copyup_callback(struct rbd_obj_request *obj_request)
2324 struct rbd_img_request *img_request;
2325 struct rbd_device *rbd_dev;
2326 struct page **pages;
2329 rbd_assert(obj_request->type == OBJ_REQUEST_BIO);
2330 rbd_assert(obj_request_img_data_test(obj_request));
2331 img_request = obj_request->img_request;
2332 rbd_assert(img_request);
2334 rbd_dev = img_request->rbd_dev;
2335 rbd_assert(rbd_dev);
2337 pages = obj_request->copyup_pages;
2338 rbd_assert(pages != NULL);
2339 obj_request->copyup_pages = NULL;
2340 page_count = obj_request->copyup_page_count;
2341 rbd_assert(page_count);
2342 obj_request->copyup_page_count = 0;
2343 ceph_release_page_vector(pages, page_count);
2346 * We want the transfer count to reflect the size of the
2347 * original write request. There is no such thing as a
2348 * successful short write, so if the request was successful
2349 * we can just set it to the originally-requested length.
2351 if (!obj_request->result)
2352 obj_request->xferred = obj_request->length;
2354 /* Finish up with the normal image object callback */
2356 rbd_img_obj_callback(obj_request);
2360 rbd_img_obj_parent_read_full_callback(struct rbd_img_request *img_request)
2362 struct rbd_obj_request *orig_request;
2363 struct ceph_osd_request *osd_req;
2364 struct ceph_osd_client *osdc;
2365 struct rbd_device *rbd_dev;
2366 struct page **pages;
2373 rbd_assert(img_request_child_test(img_request));
2375 /* First get what we need from the image request */
2377 pages = img_request->copyup_pages;
2378 rbd_assert(pages != NULL);
2379 img_request->copyup_pages = NULL;
2380 page_count = img_request->copyup_page_count;
2381 rbd_assert(page_count);
2382 img_request->copyup_page_count = 0;
2384 orig_request = img_request->obj_request;
2385 rbd_assert(orig_request != NULL);
2386 rbd_assert(obj_request_type_valid(orig_request->type));
2387 img_result = img_request->result;
2388 parent_length = img_request->length;
2389 rbd_assert(parent_length == img_request->xferred);
2390 rbd_img_request_put(img_request);
2392 rbd_assert(orig_request->img_request);
2393 rbd_dev = orig_request->img_request->rbd_dev;
2394 rbd_assert(rbd_dev);
2397 * If the overlap has become 0 (most likely because the
2398 * image has been flattened) we need to free the pages
2399 * and re-submit the original write request.
2401 if (!rbd_dev->parent_overlap) {
2402 struct ceph_osd_client *osdc;
2404 ceph_release_page_vector(pages, page_count);
2405 osdc = &rbd_dev->rbd_client->client->osdc;
2406 img_result = rbd_obj_request_submit(osdc, orig_request);
2415 * The original osd request is of no use to use any more.
2416 * We need a new one that can hold the two ops in a copyup
2417 * request. Allocate the new copyup osd request for the
2418 * original request, and release the old one.
2420 img_result = -ENOMEM;
2421 osd_req = rbd_osd_req_create_copyup(orig_request);
2424 rbd_osd_req_destroy(orig_request->osd_req);
2425 orig_request->osd_req = osd_req;
2426 orig_request->copyup_pages = pages;
2427 orig_request->copyup_page_count = page_count;
2429 /* Initialize the copyup op */
2431 osd_req_op_cls_init(osd_req, 0, CEPH_OSD_OP_CALL, "rbd", "copyup");
2432 osd_req_op_cls_request_data_pages(osd_req, 0, pages, parent_length, 0,
2435 /* Then the original write request op */
2437 offset = orig_request->offset;
2438 length = orig_request->length;
2439 osd_req_op_extent_init(osd_req, 1, CEPH_OSD_OP_WRITE,
2440 offset, length, 0, 0);
2441 if (orig_request->type == OBJ_REQUEST_BIO)
2442 osd_req_op_extent_osd_data_bio(osd_req, 1,
2443 orig_request->bio_list, length);
2445 osd_req_op_extent_osd_data_pages(osd_req, 1,
2446 orig_request->pages, length,
2447 offset & ~PAGE_MASK, false, false);
2449 rbd_osd_req_format_write(orig_request);
2451 /* All set, send it off. */
2453 orig_request->callback = rbd_img_obj_copyup_callback;
2454 osdc = &rbd_dev->rbd_client->client->osdc;
2455 img_result = rbd_obj_request_submit(osdc, orig_request);
2459 /* Record the error code and complete the request */
2461 orig_request->result = img_result;
2462 orig_request->xferred = 0;
2463 obj_request_done_set(orig_request);
2464 rbd_obj_request_complete(orig_request);
2468 * Read from the parent image the range of data that covers the
2469 * entire target of the given object request. This is used for
2470 * satisfying a layered image write request when the target of an
2471 * object request from the image request does not exist.
2473 * A page array big enough to hold the returned data is allocated
2474 * and supplied to rbd_img_request_fill() as the "data descriptor."
2475 * When the read completes, this page array will be transferred to
2476 * the original object request for the copyup operation.
2478 * If an error occurs, record it as the result of the original
2479 * object request and mark it done so it gets completed.
2481 static int rbd_img_obj_parent_read_full(struct rbd_obj_request *obj_request)
2483 struct rbd_img_request *img_request = NULL;
2484 struct rbd_img_request *parent_request = NULL;
2485 struct rbd_device *rbd_dev;
2488 struct page **pages = NULL;
2492 rbd_assert(obj_request_img_data_test(obj_request));
2493 rbd_assert(obj_request_type_valid(obj_request->type));
2495 img_request = obj_request->img_request;
2496 rbd_assert(img_request != NULL);
2497 rbd_dev = img_request->rbd_dev;
2498 rbd_assert(rbd_dev->parent != NULL);
2501 * Determine the byte range covered by the object in the
2502 * child image to which the original request was to be sent.
2504 img_offset = obj_request->img_offset - obj_request->offset;
2505 length = (u64)1 << rbd_dev->header.obj_order;
2508 * There is no defined parent data beyond the parent
2509 * overlap, so limit what we read at that boundary if
2512 if (img_offset + length > rbd_dev->parent_overlap) {
2513 rbd_assert(img_offset < rbd_dev->parent_overlap);
2514 length = rbd_dev->parent_overlap - img_offset;
2518 * Allocate a page array big enough to receive the data read
2521 page_count = (u32)calc_pages_for(0, length);
2522 pages = ceph_alloc_page_vector(page_count, GFP_KERNEL);
2523 if (IS_ERR(pages)) {
2524 result = PTR_ERR(pages);
2530 parent_request = rbd_parent_request_create(obj_request,
2531 img_offset, length);
2532 if (!parent_request)
2535 result = rbd_img_request_fill(parent_request, OBJ_REQUEST_PAGES, pages);
2538 parent_request->copyup_pages = pages;
2539 parent_request->copyup_page_count = page_count;
2541 parent_request->callback = rbd_img_obj_parent_read_full_callback;
2542 result = rbd_img_request_submit(parent_request);
2546 parent_request->copyup_pages = NULL;
2547 parent_request->copyup_page_count = 0;
2548 parent_request->obj_request = NULL;
2549 rbd_obj_request_put(obj_request);
2552 ceph_release_page_vector(pages, page_count);
2554 rbd_img_request_put(parent_request);
2555 obj_request->result = result;
2556 obj_request->xferred = 0;
2557 obj_request_done_set(obj_request);
2562 static void rbd_img_obj_exists_callback(struct rbd_obj_request *obj_request)
2564 struct rbd_obj_request *orig_request;
2565 struct rbd_device *rbd_dev;
2568 rbd_assert(!obj_request_img_data_test(obj_request));
2571 * All we need from the object request is the original
2572 * request and the result of the STAT op. Grab those, then
2573 * we're done with the request.
2575 orig_request = obj_request->obj_request;
2576 obj_request->obj_request = NULL;
2577 rbd_obj_request_put(orig_request);
2578 rbd_assert(orig_request);
2579 rbd_assert(orig_request->img_request);
2581 result = obj_request->result;
2582 obj_request->result = 0;
2584 dout("%s: obj %p for obj %p result %d %llu/%llu\n", __func__,
2585 obj_request, orig_request, result,
2586 obj_request->xferred, obj_request->length);
2587 rbd_obj_request_put(obj_request);
2590 * If the overlap has become 0 (most likely because the
2591 * image has been flattened) we need to free the pages
2592 * and re-submit the original write request.
2594 rbd_dev = orig_request->img_request->rbd_dev;
2595 if (!rbd_dev->parent_overlap) {
2596 struct ceph_osd_client *osdc;
2598 osdc = &rbd_dev->rbd_client->client->osdc;
2599 result = rbd_obj_request_submit(osdc, orig_request);
2605 * Our only purpose here is to determine whether the object
2606 * exists, and we don't want to treat the non-existence as
2607 * an error. If something else comes back, transfer the
2608 * error to the original request and complete it now.
2611 obj_request_existence_set(orig_request, true);
2612 } else if (result == -ENOENT) {
2613 obj_request_existence_set(orig_request, false);
2614 } else if (result) {
2615 orig_request->result = result;
2620 * Resubmit the original request now that we have recorded
2621 * whether the target object exists.
2623 orig_request->result = rbd_img_obj_request_submit(orig_request);
2625 if (orig_request->result)
2626 rbd_obj_request_complete(orig_request);
2629 static int rbd_img_obj_exists_submit(struct rbd_obj_request *obj_request)
2631 struct rbd_obj_request *stat_request;
2632 struct rbd_device *rbd_dev;
2633 struct ceph_osd_client *osdc;
2634 struct page **pages = NULL;
2640 * The response data for a STAT call consists of:
2647 size = sizeof (__le64) + sizeof (__le32) + sizeof (__le32);
2648 page_count = (u32)calc_pages_for(0, size);
2649 pages = ceph_alloc_page_vector(page_count, GFP_KERNEL);
2651 return PTR_ERR(pages);
2654 stat_request = rbd_obj_request_create(obj_request->object_name, 0, 0,
2659 rbd_obj_request_get(obj_request);
2660 stat_request->obj_request = obj_request;
2661 stat_request->pages = pages;
2662 stat_request->page_count = page_count;
2664 rbd_assert(obj_request->img_request);
2665 rbd_dev = obj_request->img_request->rbd_dev;
2666 stat_request->osd_req = rbd_osd_req_create(rbd_dev, false,
2668 if (!stat_request->osd_req)
2670 stat_request->callback = rbd_img_obj_exists_callback;
2672 osd_req_op_init(stat_request->osd_req, 0, CEPH_OSD_OP_STAT);
2673 osd_req_op_raw_data_in_pages(stat_request->osd_req, 0, pages, size, 0,
2675 rbd_osd_req_format_read(stat_request);
2677 osdc = &rbd_dev->rbd_client->client->osdc;
2678 ret = rbd_obj_request_submit(osdc, stat_request);
2681 rbd_obj_request_put(obj_request);
2686 static int rbd_img_obj_request_submit(struct rbd_obj_request *obj_request)
2688 struct rbd_img_request *img_request;
2689 struct rbd_device *rbd_dev;
2692 rbd_assert(obj_request_img_data_test(obj_request));
2694 img_request = obj_request->img_request;
2695 rbd_assert(img_request);
2696 rbd_dev = img_request->rbd_dev;
2699 * Only writes to layered images need special handling.
2700 * Reads and non-layered writes are simple object requests.
2701 * Layered writes that start beyond the end of the overlap
2702 * with the parent have no parent data, so they too are
2703 * simple object requests. Finally, if the target object is
2704 * known to already exist, its parent data has already been
2705 * copied, so a write to the object can also be handled as a
2706 * simple object request.
2708 if (!img_request_write_test(img_request) ||
2709 !img_request_layered_test(img_request) ||
2710 rbd_dev->parent_overlap <= obj_request->img_offset ||
2711 ((known = obj_request_known_test(obj_request)) &&
2712 obj_request_exists_test(obj_request))) {
2714 struct rbd_device *rbd_dev;
2715 struct ceph_osd_client *osdc;
2717 rbd_dev = obj_request->img_request->rbd_dev;
2718 osdc = &rbd_dev->rbd_client->client->osdc;
2720 return rbd_obj_request_submit(osdc, obj_request);
2724 * It's a layered write. The target object might exist but
2725 * we may not know that yet. If we know it doesn't exist,
2726 * start by reading the data for the full target object from
2727 * the parent so we can use it for a copyup to the target.
2730 return rbd_img_obj_parent_read_full(obj_request);
2732 /* We don't know whether the target exists. Go find out. */
2734 return rbd_img_obj_exists_submit(obj_request);
2737 static int rbd_img_request_submit(struct rbd_img_request *img_request)
2739 struct rbd_obj_request *obj_request;
2740 struct rbd_obj_request *next_obj_request;
2742 dout("%s: img %p\n", __func__, img_request);
2743 for_each_obj_request_safe(img_request, obj_request, next_obj_request) {
2746 ret = rbd_img_obj_request_submit(obj_request);
2754 static void rbd_img_parent_read_callback(struct rbd_img_request *img_request)
2756 struct rbd_obj_request *obj_request;
2757 struct rbd_device *rbd_dev;
2762 rbd_assert(img_request_child_test(img_request));
2764 /* First get what we need from the image request and release it */
2766 obj_request = img_request->obj_request;
2767 img_xferred = img_request->xferred;
2768 img_result = img_request->result;
2769 rbd_img_request_put(img_request);
2772 * If the overlap has become 0 (most likely because the
2773 * image has been flattened) we need to re-submit the
2776 rbd_assert(obj_request);
2777 rbd_assert(obj_request->img_request);
2778 rbd_dev = obj_request->img_request->rbd_dev;
2779 if (!rbd_dev->parent_overlap) {
2780 struct ceph_osd_client *osdc;
2782 osdc = &rbd_dev->rbd_client->client->osdc;
2783 img_result = rbd_obj_request_submit(osdc, obj_request);
2788 obj_request->result = img_result;
2789 if (obj_request->result)
2793 * We need to zero anything beyond the parent overlap
2794 * boundary. Since rbd_img_obj_request_read_callback()
2795 * will zero anything beyond the end of a short read, an
2796 * easy way to do this is to pretend the data from the
2797 * parent came up short--ending at the overlap boundary.
2799 rbd_assert(obj_request->img_offset < U64_MAX - obj_request->length);
2800 obj_end = obj_request->img_offset + obj_request->length;
2801 if (obj_end > rbd_dev->parent_overlap) {
2804 if (obj_request->img_offset < rbd_dev->parent_overlap)
2805 xferred = rbd_dev->parent_overlap -
2806 obj_request->img_offset;
2808 obj_request->xferred = min(img_xferred, xferred);
2810 obj_request->xferred = img_xferred;
2813 rbd_img_obj_request_read_callback(obj_request);
2814 rbd_obj_request_complete(obj_request);
2817 static void rbd_img_parent_read(struct rbd_obj_request *obj_request)
2819 struct rbd_img_request *img_request;
2822 rbd_assert(obj_request_img_data_test(obj_request));
2823 rbd_assert(obj_request->img_request != NULL);
2824 rbd_assert(obj_request->result == (s32) -ENOENT);
2825 rbd_assert(obj_request_type_valid(obj_request->type));
2827 /* rbd_read_finish(obj_request, obj_request->length); */
2828 img_request = rbd_parent_request_create(obj_request,
2829 obj_request->img_offset,
2830 obj_request->length);
2835 if (obj_request->type == OBJ_REQUEST_BIO)
2836 result = rbd_img_request_fill(img_request, OBJ_REQUEST_BIO,
2837 obj_request->bio_list);
2839 result = rbd_img_request_fill(img_request, OBJ_REQUEST_PAGES,
2840 obj_request->pages);
2844 img_request->callback = rbd_img_parent_read_callback;
2845 result = rbd_img_request_submit(img_request);
2852 rbd_img_request_put(img_request);
2853 obj_request->result = result;
2854 obj_request->xferred = 0;
2855 obj_request_done_set(obj_request);
2858 static int rbd_obj_notify_ack_sync(struct rbd_device *rbd_dev, u64 notify_id)
2860 struct rbd_obj_request *obj_request;
2861 struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
2864 obj_request = rbd_obj_request_create(rbd_dev->header_name, 0, 0,
2865 OBJ_REQUEST_NODATA);
2870 obj_request->osd_req = rbd_osd_req_create(rbd_dev, false, obj_request);
2871 if (!obj_request->osd_req)
2874 osd_req_op_watch_init(obj_request->osd_req, 0, CEPH_OSD_OP_NOTIFY_ACK,
2876 rbd_osd_req_format_read(obj_request);
2878 ret = rbd_obj_request_submit(osdc, obj_request);
2881 ret = rbd_obj_request_wait(obj_request);
2883 rbd_obj_request_put(obj_request);
2888 static void rbd_watch_cb(u64 ver, u64 notify_id, u8 opcode, void *data)
2890 struct rbd_device *rbd_dev = (struct rbd_device *)data;
2896 dout("%s: \"%s\" notify_id %llu opcode %u\n", __func__,
2897 rbd_dev->header_name, (unsigned long long)notify_id,
2898 (unsigned int)opcode);
2899 ret = rbd_dev_refresh(rbd_dev);
2901 rbd_warn(rbd_dev, "header refresh error (%d)\n", ret);
2903 rbd_obj_notify_ack_sync(rbd_dev, notify_id);
2907 * Request sync osd watch/unwatch. The value of "start" determines
2908 * whether a watch request is being initiated or torn down.
2910 static int __rbd_dev_header_watch_sync(struct rbd_device *rbd_dev, bool start)
2912 struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
2913 struct rbd_obj_request *obj_request;
2916 rbd_assert(start ^ !!rbd_dev->watch_event);
2917 rbd_assert(start ^ !!rbd_dev->watch_request);
2920 ret = ceph_osdc_create_event(osdc, rbd_watch_cb, rbd_dev,
2921 &rbd_dev->watch_event);
2924 rbd_assert(rbd_dev->watch_event != NULL);
2928 obj_request = rbd_obj_request_create(rbd_dev->header_name, 0, 0,
2929 OBJ_REQUEST_NODATA);
2933 obj_request->osd_req = rbd_osd_req_create(rbd_dev, true, obj_request);
2934 if (!obj_request->osd_req)
2938 ceph_osdc_set_request_linger(osdc, obj_request->osd_req);
2940 ceph_osdc_unregister_linger_request(osdc,
2941 rbd_dev->watch_request->osd_req);
2943 osd_req_op_watch_init(obj_request->osd_req, 0, CEPH_OSD_OP_WATCH,
2944 rbd_dev->watch_event->cookie, 0, start ? 1 : 0);
2945 rbd_osd_req_format_write(obj_request);
2947 ret = rbd_obj_request_submit(osdc, obj_request);
2950 ret = rbd_obj_request_wait(obj_request);
2953 ret = obj_request->result;
2958 * A watch request is set to linger, so the underlying osd
2959 * request won't go away until we unregister it. We retain
2960 * a pointer to the object request during that time (in
2961 * rbd_dev->watch_request), so we'll keep a reference to
2962 * it. We'll drop that reference (below) after we've
2966 rbd_dev->watch_request = obj_request;
2971 /* We have successfully torn down the watch request */
2973 rbd_obj_request_put(rbd_dev->watch_request);
2974 rbd_dev->watch_request = NULL;
2976 /* Cancel the event if we're tearing down, or on error */
2977 ceph_osdc_cancel_event(rbd_dev->watch_event);
2978 rbd_dev->watch_event = NULL;
2980 rbd_obj_request_put(obj_request);
2985 static int rbd_dev_header_watch_sync(struct rbd_device *rbd_dev)
2987 return __rbd_dev_header_watch_sync(rbd_dev, true);
2990 static void rbd_dev_header_unwatch_sync(struct rbd_device *rbd_dev)
2994 ret = __rbd_dev_header_watch_sync(rbd_dev, false);
2996 rbd_warn(rbd_dev, "unable to tear down watch request: %d\n",
3002 * Synchronous osd object method call. Returns the number of bytes
3003 * returned in the outbound buffer, or a negative error code.
3005 static int rbd_obj_method_sync(struct rbd_device *rbd_dev,
3006 const char *object_name,
3007 const char *class_name,
3008 const char *method_name,
3009 const void *outbound,
3010 size_t outbound_size,
3012 size_t inbound_size)
3014 struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
3015 struct rbd_obj_request *obj_request;
3016 struct page **pages;
3021 * Method calls are ultimately read operations. The result
3022 * should placed into the inbound buffer provided. They
3023 * also supply outbound data--parameters for the object
3024 * method. Currently if this is present it will be a
3027 page_count = (u32)calc_pages_for(0, inbound_size);
3028 pages = ceph_alloc_page_vector(page_count, GFP_KERNEL);
3030 return PTR_ERR(pages);
3033 obj_request = rbd_obj_request_create(object_name, 0, inbound_size,
3038 obj_request->pages = pages;
3039 obj_request->page_count = page_count;
3041 obj_request->osd_req = rbd_osd_req_create(rbd_dev, false, obj_request);
3042 if (!obj_request->osd_req)
3045 osd_req_op_cls_init(obj_request->osd_req, 0, CEPH_OSD_OP_CALL,
3046 class_name, method_name);
3047 if (outbound_size) {
3048 struct ceph_pagelist *pagelist;
3050 pagelist = kmalloc(sizeof (*pagelist), GFP_NOFS);
3054 ceph_pagelist_init(pagelist);
3055 ceph_pagelist_append(pagelist, outbound, outbound_size);
3056 osd_req_op_cls_request_data_pagelist(obj_request->osd_req, 0,
3059 osd_req_op_cls_response_data_pages(obj_request->osd_req, 0,
3060 obj_request->pages, inbound_size,
3062 rbd_osd_req_format_read(obj_request);
3064 ret = rbd_obj_request_submit(osdc, obj_request);
3067 ret = rbd_obj_request_wait(obj_request);
3071 ret = obj_request->result;
3075 rbd_assert(obj_request->xferred < (u64)INT_MAX);
3076 ret = (int)obj_request->xferred;
3077 ceph_copy_from_page_vector(pages, inbound, 0, obj_request->xferred);
3080 rbd_obj_request_put(obj_request);
3082 ceph_release_page_vector(pages, page_count);
3087 static void rbd_request_fn(struct request_queue *q)
3088 __releases(q->queue_lock) __acquires(q->queue_lock)
3090 struct rbd_device *rbd_dev = q->queuedata;
3091 bool read_only = rbd_dev->mapping.read_only;
3095 while ((rq = blk_fetch_request(q))) {
3096 bool write_request = rq_data_dir(rq) == WRITE;
3097 struct rbd_img_request *img_request;
3101 /* Ignore any non-FS requests that filter through. */
3103 if (rq->cmd_type != REQ_TYPE_FS) {
3104 dout("%s: non-fs request type %d\n", __func__,
3105 (int) rq->cmd_type);
3106 __blk_end_request_all(rq, 0);
3110 /* Ignore/skip any zero-length requests */
3112 offset = (u64) blk_rq_pos(rq) << SECTOR_SHIFT;
3113 length = (u64) blk_rq_bytes(rq);
3116 dout("%s: zero-length request\n", __func__);
3117 __blk_end_request_all(rq, 0);
3121 spin_unlock_irq(q->queue_lock);
3123 /* Disallow writes to a read-only device */
3125 if (write_request) {
3129 rbd_assert(rbd_dev->spec->snap_id == CEPH_NOSNAP);
3133 * Quit early if the mapped snapshot no longer
3134 * exists. It's still possible the snapshot will
3135 * have disappeared by the time our request arrives
3136 * at the osd, but there's no sense in sending it if
3139 if (!test_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags)) {
3140 dout("request for non-existent snapshot");
3141 rbd_assert(rbd_dev->spec->snap_id != CEPH_NOSNAP);
3147 if (offset && length > U64_MAX - offset + 1) {
3148 rbd_warn(rbd_dev, "bad request range (%llu~%llu)\n",
3150 goto end_request; /* Shouldn't happen */
3154 if (offset + length > rbd_dev->mapping.size) {
3155 rbd_warn(rbd_dev, "beyond EOD (%llu~%llu > %llu)\n",
3156 offset, length, rbd_dev->mapping.size);
3161 img_request = rbd_img_request_create(rbd_dev, offset, length,
3166 img_request->rq = rq;
3168 result = rbd_img_request_fill(img_request, OBJ_REQUEST_BIO,
3171 result = rbd_img_request_submit(img_request);
3173 rbd_img_request_put(img_request);
3175 spin_lock_irq(q->queue_lock);
3177 rbd_warn(rbd_dev, "%s %llx at %llx result %d\n",
3178 write_request ? "write" : "read",
3179 length, offset, result);
3181 __blk_end_request_all(rq, result);
3187 * a queue callback. Makes sure that we don't create a bio that spans across
3188 * multiple osd objects. One exception would be with a single page bios,
3189 * which we handle later at bio_chain_clone_range()
3191 static int rbd_merge_bvec(struct request_queue *q, struct bvec_merge_data *bmd,
3192 struct bio_vec *bvec)
3194 struct rbd_device *rbd_dev = q->queuedata;
3195 sector_t sector_offset;
3196 sector_t sectors_per_obj;
3197 sector_t obj_sector_offset;
3201 * Find how far into its rbd object the partition-relative
3202 * bio start sector is to offset relative to the enclosing
3205 sector_offset = get_start_sect(bmd->bi_bdev) + bmd->bi_sector;
3206 sectors_per_obj = 1 << (rbd_dev->header.obj_order - SECTOR_SHIFT);
3207 obj_sector_offset = sector_offset & (sectors_per_obj - 1);
3210 * Compute the number of bytes from that offset to the end
3211 * of the object. Account for what's already used by the bio.
3213 ret = (int) (sectors_per_obj - obj_sector_offset) << SECTOR_SHIFT;
3214 if (ret > bmd->bi_size)
3215 ret -= bmd->bi_size;
3220 * Don't send back more than was asked for. And if the bio
3221 * was empty, let the whole thing through because: "Note
3222 * that a block device *must* allow a single page to be
3223 * added to an empty bio."
3225 rbd_assert(bvec->bv_len <= PAGE_SIZE);
3226 if (ret > (int) bvec->bv_len || !bmd->bi_size)
3227 ret = (int) bvec->bv_len;
3232 static void rbd_free_disk(struct rbd_device *rbd_dev)
3234 struct gendisk *disk = rbd_dev->disk;
3239 rbd_dev->disk = NULL;
3240 if (disk->flags & GENHD_FL_UP) {
3243 blk_cleanup_queue(disk->queue);
3248 static int rbd_obj_read_sync(struct rbd_device *rbd_dev,
3249 const char *object_name,
3250 u64 offset, u64 length, void *buf)
3253 struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
3254 struct rbd_obj_request *obj_request;
3255 struct page **pages = NULL;
3260 page_count = (u32) calc_pages_for(offset, length);
3261 pages = ceph_alloc_page_vector(page_count, GFP_KERNEL);
3263 ret = PTR_ERR(pages);
3266 obj_request = rbd_obj_request_create(object_name, offset, length,
3271 obj_request->pages = pages;
3272 obj_request->page_count = page_count;
3274 obj_request->osd_req = rbd_osd_req_create(rbd_dev, false, obj_request);
3275 if (!obj_request->osd_req)
3278 osd_req_op_extent_init(obj_request->osd_req, 0, CEPH_OSD_OP_READ,
3279 offset, length, 0, 0);
3280 osd_req_op_extent_osd_data_pages(obj_request->osd_req, 0,
3282 obj_request->length,
3283 obj_request->offset & ~PAGE_MASK,
3285 rbd_osd_req_format_read(obj_request);
3287 ret = rbd_obj_request_submit(osdc, obj_request);
3290 ret = rbd_obj_request_wait(obj_request);
3294 ret = obj_request->result;
3298 rbd_assert(obj_request->xferred <= (u64) SIZE_MAX);
3299 size = (size_t) obj_request->xferred;
3300 ceph_copy_from_page_vector(pages, buf, 0, size);
3301 rbd_assert(size <= (size_t)INT_MAX);
3305 rbd_obj_request_put(obj_request);
3307 ceph_release_page_vector(pages, page_count);
3313 * Read the complete header for the given rbd device. On successful
3314 * return, the rbd_dev->header field will contain up-to-date
3315 * information about the image.
3317 static int rbd_dev_v1_header_info(struct rbd_device *rbd_dev)
3319 struct rbd_image_header_ondisk *ondisk = NULL;
3326 * The complete header will include an array of its 64-bit
3327 * snapshot ids, followed by the names of those snapshots as
3328 * a contiguous block of NUL-terminated strings. Note that
3329 * the number of snapshots could change by the time we read
3330 * it in, in which case we re-read it.
3337 size = sizeof (*ondisk);
3338 size += snap_count * sizeof (struct rbd_image_snap_ondisk);
3340 ondisk = kmalloc(size, GFP_KERNEL);
3344 ret = rbd_obj_read_sync(rbd_dev, rbd_dev->header_name,
3348 if ((size_t)ret < size) {
3350 rbd_warn(rbd_dev, "short header read (want %zd got %d)",
3354 if (!rbd_dev_ondisk_valid(ondisk)) {
3356 rbd_warn(rbd_dev, "invalid header");
3360 names_size = le64_to_cpu(ondisk->snap_names_len);
3361 want_count = snap_count;
3362 snap_count = le32_to_cpu(ondisk->snap_count);
3363 } while (snap_count != want_count);
3365 ret = rbd_header_from_disk(rbd_dev, ondisk);
3373 * Clear the rbd device's EXISTS flag if the snapshot it's mapped to
3374 * has disappeared from the (just updated) snapshot context.
3376 static void rbd_exists_validate(struct rbd_device *rbd_dev)
3380 if (!test_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags))
3383 snap_id = rbd_dev->spec->snap_id;
3384 if (snap_id == CEPH_NOSNAP)
3387 if (rbd_dev_snap_index(rbd_dev, snap_id) == BAD_SNAP_INDEX)
3388 clear_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags);
3391 static void rbd_dev_update_size(struct rbd_device *rbd_dev)
3397 * Don't hold the lock while doing disk operations,
3398 * or lock ordering will conflict with the bdev mutex via:
3399 * rbd_add() -> blkdev_get() -> rbd_open()
3401 spin_lock_irq(&rbd_dev->lock);
3402 removing = test_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags);
3403 spin_unlock_irq(&rbd_dev->lock);
3405 * If the device is being removed, rbd_dev->disk has
3406 * been destroyed, so don't try to update its size
3409 size = (sector_t)rbd_dev->mapping.size / SECTOR_SIZE;
3410 dout("setting size to %llu sectors", (unsigned long long)size);
3411 set_capacity(rbd_dev->disk, size);
3412 revalidate_disk(rbd_dev->disk);
3416 static int rbd_dev_refresh(struct rbd_device *rbd_dev)
3421 rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
3422 down_write(&rbd_dev->header_rwsem);
3423 mapping_size = rbd_dev->mapping.size;
3424 if (rbd_dev->image_format == 1)
3425 ret = rbd_dev_v1_header_info(rbd_dev);
3427 ret = rbd_dev_v2_header_info(rbd_dev);
3429 /* If it's a mapped snapshot, validate its EXISTS flag */
3431 rbd_exists_validate(rbd_dev);
3432 up_write(&rbd_dev->header_rwsem);
3434 if (mapping_size != rbd_dev->mapping.size) {
3435 rbd_dev_update_size(rbd_dev);
3441 static int rbd_init_disk(struct rbd_device *rbd_dev)
3443 struct gendisk *disk;
3444 struct request_queue *q;
3447 /* create gendisk info */
3448 disk = alloc_disk(single_major ?
3449 (1 << RBD_SINGLE_MAJOR_PART_SHIFT) :
3450 RBD_MINORS_PER_MAJOR);
3454 snprintf(disk->disk_name, sizeof(disk->disk_name), RBD_DRV_NAME "%d",
3456 disk->major = rbd_dev->major;
3457 disk->first_minor = rbd_dev->minor;
3459 disk->flags |= GENHD_FL_EXT_DEVT;
3460 disk->fops = &rbd_bd_ops;
3461 disk->private_data = rbd_dev;
3463 q = blk_init_queue(rbd_request_fn, &rbd_dev->lock);
3467 /* We use the default size, but let's be explicit about it. */
3468 blk_queue_physical_block_size(q, SECTOR_SIZE);
3470 /* set io sizes to object size */
3471 segment_size = rbd_obj_bytes(&rbd_dev->header);
3472 blk_queue_max_hw_sectors(q, segment_size / SECTOR_SIZE);
3473 blk_queue_max_segment_size(q, segment_size);
3474 blk_queue_io_min(q, segment_size);
3475 blk_queue_io_opt(q, segment_size);
3477 blk_queue_merge_bvec(q, rbd_merge_bvec);
3480 q->queuedata = rbd_dev;
3482 rbd_dev->disk = disk;
3495 static struct rbd_device *dev_to_rbd_dev(struct device *dev)
3497 return container_of(dev, struct rbd_device, dev);
3500 static ssize_t rbd_size_show(struct device *dev,
3501 struct device_attribute *attr, char *buf)
3503 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3505 return sprintf(buf, "%llu\n",
3506 (unsigned long long)rbd_dev->mapping.size);
3510 * Note this shows the features for whatever's mapped, which is not
3511 * necessarily the base image.
3513 static ssize_t rbd_features_show(struct device *dev,
3514 struct device_attribute *attr, char *buf)
3516 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3518 return sprintf(buf, "0x%016llx\n",
3519 (unsigned long long)rbd_dev->mapping.features);
3522 static ssize_t rbd_major_show(struct device *dev,
3523 struct device_attribute *attr, char *buf)
3525 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3528 return sprintf(buf, "%d\n", rbd_dev->major);
3530 return sprintf(buf, "(none)\n");
3533 static ssize_t rbd_minor_show(struct device *dev,
3534 struct device_attribute *attr, char *buf)
3536 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3538 return sprintf(buf, "%d\n", rbd_dev->minor);
3541 static ssize_t rbd_client_id_show(struct device *dev,
3542 struct device_attribute *attr, char *buf)
3544 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3546 return sprintf(buf, "client%lld\n",
3547 ceph_client_id(rbd_dev->rbd_client->client));
3550 static ssize_t rbd_pool_show(struct device *dev,
3551 struct device_attribute *attr, char *buf)
3553 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3555 return sprintf(buf, "%s\n", rbd_dev->spec->pool_name);
3558 static ssize_t rbd_pool_id_show(struct device *dev,
3559 struct device_attribute *attr, char *buf)
3561 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3563 return sprintf(buf, "%llu\n",
3564 (unsigned long long) rbd_dev->spec->pool_id);
3567 static ssize_t rbd_name_show(struct device *dev,
3568 struct device_attribute *attr, char *buf)
3570 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3572 if (rbd_dev->spec->image_name)
3573 return sprintf(buf, "%s\n", rbd_dev->spec->image_name);
3575 return sprintf(buf, "(unknown)\n");
3578 static ssize_t rbd_image_id_show(struct device *dev,
3579 struct device_attribute *attr, char *buf)
3581 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3583 return sprintf(buf, "%s\n", rbd_dev->spec->image_id);
3587 * Shows the name of the currently-mapped snapshot (or
3588 * RBD_SNAP_HEAD_NAME for the base image).
3590 static ssize_t rbd_snap_show(struct device *dev,
3591 struct device_attribute *attr,
3594 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3596 return sprintf(buf, "%s\n", rbd_dev->spec->snap_name);
3600 * For an rbd v2 image, shows the pool id, image id, and snapshot id
3601 * for the parent image. If there is no parent, simply shows
3602 * "(no parent image)".
3604 static ssize_t rbd_parent_show(struct device *dev,
3605 struct device_attribute *attr,
3608 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3609 struct rbd_spec *spec = rbd_dev->parent_spec;
3614 return sprintf(buf, "(no parent image)\n");
3616 count = sprintf(bufp, "pool_id %llu\npool_name %s\n",
3617 (unsigned long long) spec->pool_id, spec->pool_name);
3622 count = sprintf(bufp, "image_id %s\nimage_name %s\n", spec->image_id,
3623 spec->image_name ? spec->image_name : "(unknown)");
3628 count = sprintf(bufp, "snap_id %llu\nsnap_name %s\n",
3629 (unsigned long long) spec->snap_id, spec->snap_name);
3634 count = sprintf(bufp, "overlap %llu\n", rbd_dev->parent_overlap);
3639 return (ssize_t) (bufp - buf);
3642 static ssize_t rbd_image_refresh(struct device *dev,
3643 struct device_attribute *attr,
3647 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3650 ret = rbd_dev_refresh(rbd_dev);
3652 rbd_warn(rbd_dev, ": manual header refresh error (%d)\n", ret);
3654 return ret < 0 ? ret : size;
3657 static DEVICE_ATTR(size, S_IRUGO, rbd_size_show, NULL);
3658 static DEVICE_ATTR(features, S_IRUGO, rbd_features_show, NULL);
3659 static DEVICE_ATTR(major, S_IRUGO, rbd_major_show, NULL);
3660 static DEVICE_ATTR(minor, S_IRUGO, rbd_minor_show, NULL);
3661 static DEVICE_ATTR(client_id, S_IRUGO, rbd_client_id_show, NULL);
3662 static DEVICE_ATTR(pool, S_IRUGO, rbd_pool_show, NULL);
3663 static DEVICE_ATTR(pool_id, S_IRUGO, rbd_pool_id_show, NULL);
3664 static DEVICE_ATTR(name, S_IRUGO, rbd_name_show, NULL);
3665 static DEVICE_ATTR(image_id, S_IRUGO, rbd_image_id_show, NULL);
3666 static DEVICE_ATTR(refresh, S_IWUSR, NULL, rbd_image_refresh);
3667 static DEVICE_ATTR(current_snap, S_IRUGO, rbd_snap_show, NULL);
3668 static DEVICE_ATTR(parent, S_IRUGO, rbd_parent_show, NULL);
3670 static struct attribute *rbd_attrs[] = {
3671 &dev_attr_size.attr,
3672 &dev_attr_features.attr,
3673 &dev_attr_major.attr,
3674 &dev_attr_minor.attr,
3675 &dev_attr_client_id.attr,
3676 &dev_attr_pool.attr,
3677 &dev_attr_pool_id.attr,
3678 &dev_attr_name.attr,
3679 &dev_attr_image_id.attr,
3680 &dev_attr_current_snap.attr,
3681 &dev_attr_parent.attr,
3682 &dev_attr_refresh.attr,
3686 static struct attribute_group rbd_attr_group = {
3690 static const struct attribute_group *rbd_attr_groups[] = {
3695 static void rbd_sysfs_dev_release(struct device *dev)
3699 static struct device_type rbd_device_type = {
3701 .groups = rbd_attr_groups,
3702 .release = rbd_sysfs_dev_release,
3705 static struct rbd_spec *rbd_spec_get(struct rbd_spec *spec)
3707 kref_get(&spec->kref);
3712 static void rbd_spec_free(struct kref *kref);
3713 static void rbd_spec_put(struct rbd_spec *spec)
3716 kref_put(&spec->kref, rbd_spec_free);
3719 static struct rbd_spec *rbd_spec_alloc(void)
3721 struct rbd_spec *spec;
3723 spec = kzalloc(sizeof (*spec), GFP_KERNEL);
3726 kref_init(&spec->kref);
3731 static void rbd_spec_free(struct kref *kref)
3733 struct rbd_spec *spec = container_of(kref, struct rbd_spec, kref);
3735 kfree(spec->pool_name);
3736 kfree(spec->image_id);
3737 kfree(spec->image_name);
3738 kfree(spec->snap_name);
3742 static struct rbd_device *rbd_dev_create(struct rbd_client *rbdc,
3743 struct rbd_spec *spec)
3745 struct rbd_device *rbd_dev;
3747 rbd_dev = kzalloc(sizeof (*rbd_dev), GFP_KERNEL);
3751 spin_lock_init(&rbd_dev->lock);
3753 atomic_set(&rbd_dev->parent_ref, 0);
3754 INIT_LIST_HEAD(&rbd_dev->node);
3755 init_rwsem(&rbd_dev->header_rwsem);
3757 rbd_dev->spec = spec;
3758 rbd_dev->rbd_client = rbdc;
3760 /* Initialize the layout used for all rbd requests */
3762 rbd_dev->layout.fl_stripe_unit = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER);
3763 rbd_dev->layout.fl_stripe_count = cpu_to_le32(1);
3764 rbd_dev->layout.fl_object_size = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER);
3765 rbd_dev->layout.fl_pg_pool = cpu_to_le32((u32) spec->pool_id);
3770 static void rbd_dev_destroy(struct rbd_device *rbd_dev)
3772 rbd_put_client(rbd_dev->rbd_client);
3773 rbd_spec_put(rbd_dev->spec);
3778 * Get the size and object order for an image snapshot, or if
3779 * snap_id is CEPH_NOSNAP, gets this information for the base
3782 static int _rbd_dev_v2_snap_size(struct rbd_device *rbd_dev, u64 snap_id,
3783 u8 *order, u64 *snap_size)
3785 __le64 snapid = cpu_to_le64(snap_id);
3790 } __attribute__ ((packed)) size_buf = { 0 };
3792 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
3794 &snapid, sizeof (snapid),
3795 &size_buf, sizeof (size_buf));
3796 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
3799 if (ret < sizeof (size_buf))
3803 *order = size_buf.order;
3804 dout(" order %u", (unsigned int)*order);
3806 *snap_size = le64_to_cpu(size_buf.size);
3808 dout(" snap_id 0x%016llx snap_size = %llu\n",
3809 (unsigned long long)snap_id,
3810 (unsigned long long)*snap_size);
3815 static int rbd_dev_v2_image_size(struct rbd_device *rbd_dev)
3817 return _rbd_dev_v2_snap_size(rbd_dev, CEPH_NOSNAP,
3818 &rbd_dev->header.obj_order,
3819 &rbd_dev->header.image_size);
3822 static int rbd_dev_v2_object_prefix(struct rbd_device *rbd_dev)
3828 reply_buf = kzalloc(RBD_OBJ_PREFIX_LEN_MAX, GFP_KERNEL);
3832 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
3833 "rbd", "get_object_prefix", NULL, 0,
3834 reply_buf, RBD_OBJ_PREFIX_LEN_MAX);
3835 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
3840 rbd_dev->header.object_prefix = ceph_extract_encoded_string(&p,
3841 p + ret, NULL, GFP_NOIO);
3844 if (IS_ERR(rbd_dev->header.object_prefix)) {
3845 ret = PTR_ERR(rbd_dev->header.object_prefix);
3846 rbd_dev->header.object_prefix = NULL;
3848 dout(" object_prefix = %s\n", rbd_dev->header.object_prefix);
3856 static int _rbd_dev_v2_snap_features(struct rbd_device *rbd_dev, u64 snap_id,
3859 __le64 snapid = cpu_to_le64(snap_id);
3863 } __attribute__ ((packed)) features_buf = { 0 };
3867 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
3868 "rbd", "get_features",
3869 &snapid, sizeof (snapid),
3870 &features_buf, sizeof (features_buf));
3871 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
3874 if (ret < sizeof (features_buf))
3877 incompat = le64_to_cpu(features_buf.incompat);
3878 if (incompat & ~RBD_FEATURES_SUPPORTED)
3881 *snap_features = le64_to_cpu(features_buf.features);
3883 dout(" snap_id 0x%016llx features = 0x%016llx incompat = 0x%016llx\n",
3884 (unsigned long long)snap_id,
3885 (unsigned long long)*snap_features,
3886 (unsigned long long)le64_to_cpu(features_buf.incompat));
3891 static int rbd_dev_v2_features(struct rbd_device *rbd_dev)
3893 return _rbd_dev_v2_snap_features(rbd_dev, CEPH_NOSNAP,
3894 &rbd_dev->header.features);
3897 static int rbd_dev_v2_parent_info(struct rbd_device *rbd_dev)
3899 struct rbd_spec *parent_spec;
3901 void *reply_buf = NULL;
3911 parent_spec = rbd_spec_alloc();
3915 size = sizeof (__le64) + /* pool_id */
3916 sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX + /* image_id */
3917 sizeof (__le64) + /* snap_id */
3918 sizeof (__le64); /* overlap */
3919 reply_buf = kmalloc(size, GFP_KERNEL);
3925 snapid = cpu_to_le64(CEPH_NOSNAP);
3926 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
3927 "rbd", "get_parent",
3928 &snapid, sizeof (snapid),
3930 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
3935 end = reply_buf + ret;
3937 ceph_decode_64_safe(&p, end, pool_id, out_err);
3938 if (pool_id == CEPH_NOPOOL) {
3940 * Either the parent never existed, or we have
3941 * record of it but the image got flattened so it no
3942 * longer has a parent. When the parent of a
3943 * layered image disappears we immediately set the
3944 * overlap to 0. The effect of this is that all new
3945 * requests will be treated as if the image had no
3948 if (rbd_dev->parent_overlap) {
3949 rbd_dev->parent_overlap = 0;
3951 rbd_dev_parent_put(rbd_dev);
3952 pr_info("%s: clone image has been flattened\n",
3953 rbd_dev->disk->disk_name);
3956 goto out; /* No parent? No problem. */
3959 /* The ceph file layout needs to fit pool id in 32 bits */
3962 if (pool_id > (u64)U32_MAX) {
3963 rbd_warn(NULL, "parent pool id too large (%llu > %u)\n",
3964 (unsigned long long)pool_id, U32_MAX);
3968 image_id = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL);
3969 if (IS_ERR(image_id)) {
3970 ret = PTR_ERR(image_id);
3973 ceph_decode_64_safe(&p, end, snap_id, out_err);
3974 ceph_decode_64_safe(&p, end, overlap, out_err);
3977 * The parent won't change (except when the clone is
3978 * flattened, already handled that). So we only need to
3979 * record the parent spec we have not already done so.
3981 if (!rbd_dev->parent_spec) {
3982 parent_spec->pool_id = pool_id;
3983 parent_spec->image_id = image_id;
3984 parent_spec->snap_id = snap_id;
3985 rbd_dev->parent_spec = parent_spec;
3986 parent_spec = NULL; /* rbd_dev now owns this */
3990 * We always update the parent overlap. If it's zero we
3991 * treat it specially.
3993 rbd_dev->parent_overlap = overlap;
3997 /* A null parent_spec indicates it's the initial probe */
4001 * The overlap has become zero, so the clone
4002 * must have been resized down to 0 at some
4003 * point. Treat this the same as a flatten.
4005 rbd_dev_parent_put(rbd_dev);
4006 pr_info("%s: clone image now standalone\n",
4007 rbd_dev->disk->disk_name);
4010 * For the initial probe, if we find the
4011 * overlap is zero we just pretend there was
4014 rbd_warn(rbd_dev, "ignoring parent of "
4015 "clone with overlap 0\n");
4022 rbd_spec_put(parent_spec);
4027 static int rbd_dev_v2_striping_info(struct rbd_device *rbd_dev)
4031 __le64 stripe_count;
4032 } __attribute__ ((packed)) striping_info_buf = { 0 };
4033 size_t size = sizeof (striping_info_buf);
4040 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
4041 "rbd", "get_stripe_unit_count", NULL, 0,
4042 (char *)&striping_info_buf, size);
4043 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
4050 * We don't actually support the "fancy striping" feature
4051 * (STRIPINGV2) yet, but if the striping sizes are the
4052 * defaults the behavior is the same as before. So find
4053 * out, and only fail if the image has non-default values.
4056 obj_size = (u64)1 << rbd_dev->header.obj_order;
4057 p = &striping_info_buf;
4058 stripe_unit = ceph_decode_64(&p);
4059 if (stripe_unit != obj_size) {
4060 rbd_warn(rbd_dev, "unsupported stripe unit "
4061 "(got %llu want %llu)",
4062 stripe_unit, obj_size);
4065 stripe_count = ceph_decode_64(&p);
4066 if (stripe_count != 1) {
4067 rbd_warn(rbd_dev, "unsupported stripe count "
4068 "(got %llu want 1)", stripe_count);
4071 rbd_dev->header.stripe_unit = stripe_unit;
4072 rbd_dev->header.stripe_count = stripe_count;
4077 static char *rbd_dev_image_name(struct rbd_device *rbd_dev)
4079 size_t image_id_size;
4084 void *reply_buf = NULL;
4086 char *image_name = NULL;
4089 rbd_assert(!rbd_dev->spec->image_name);
4091 len = strlen(rbd_dev->spec->image_id);
4092 image_id_size = sizeof (__le32) + len;
4093 image_id = kmalloc(image_id_size, GFP_KERNEL);
4098 end = image_id + image_id_size;
4099 ceph_encode_string(&p, end, rbd_dev->spec->image_id, (u32)len);
4101 size = sizeof (__le32) + RBD_IMAGE_NAME_LEN_MAX;
4102 reply_buf = kmalloc(size, GFP_KERNEL);
4106 ret = rbd_obj_method_sync(rbd_dev, RBD_DIRECTORY,
4107 "rbd", "dir_get_name",
4108 image_id, image_id_size,
4113 end = reply_buf + ret;
4115 image_name = ceph_extract_encoded_string(&p, end, &len, GFP_KERNEL);
4116 if (IS_ERR(image_name))
4119 dout("%s: name is %s len is %zd\n", __func__, image_name, len);
4127 static u64 rbd_v1_snap_id_by_name(struct rbd_device *rbd_dev, const char *name)
4129 struct ceph_snap_context *snapc = rbd_dev->header.snapc;
4130 const char *snap_name;
4133 /* Skip over names until we find the one we are looking for */
4135 snap_name = rbd_dev->header.snap_names;
4136 while (which < snapc->num_snaps) {
4137 if (!strcmp(name, snap_name))
4138 return snapc->snaps[which];
4139 snap_name += strlen(snap_name) + 1;
4145 static u64 rbd_v2_snap_id_by_name(struct rbd_device *rbd_dev, const char *name)
4147 struct ceph_snap_context *snapc = rbd_dev->header.snapc;
4152 for (which = 0; !found && which < snapc->num_snaps; which++) {
4153 const char *snap_name;
4155 snap_id = snapc->snaps[which];
4156 snap_name = rbd_dev_v2_snap_name(rbd_dev, snap_id);
4157 if (IS_ERR(snap_name)) {
4158 /* ignore no-longer existing snapshots */
4159 if (PTR_ERR(snap_name) == -ENOENT)
4164 found = !strcmp(name, snap_name);
4167 return found ? snap_id : CEPH_NOSNAP;
4171 * Assumes name is never RBD_SNAP_HEAD_NAME; returns CEPH_NOSNAP if
4172 * no snapshot by that name is found, or if an error occurs.
4174 static u64 rbd_snap_id_by_name(struct rbd_device *rbd_dev, const char *name)
4176 if (rbd_dev->image_format == 1)
4177 return rbd_v1_snap_id_by_name(rbd_dev, name);
4179 return rbd_v2_snap_id_by_name(rbd_dev, name);
4183 * When an rbd image has a parent image, it is identified by the
4184 * pool, image, and snapshot ids (not names). This function fills
4185 * in the names for those ids. (It's OK if we can't figure out the
4186 * name for an image id, but the pool and snapshot ids should always
4187 * exist and have names.) All names in an rbd spec are dynamically
4190 * When an image being mapped (not a parent) is probed, we have the
4191 * pool name and pool id, image name and image id, and the snapshot
4192 * name. The only thing we're missing is the snapshot id.
4194 static int rbd_dev_spec_update(struct rbd_device *rbd_dev)
4196 struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
4197 struct rbd_spec *spec = rbd_dev->spec;
4198 const char *pool_name;
4199 const char *image_name;
4200 const char *snap_name;
4204 * An image being mapped will have the pool name (etc.), but
4205 * we need to look up the snapshot id.
4207 if (spec->pool_name) {
4208 if (strcmp(spec->snap_name, RBD_SNAP_HEAD_NAME)) {
4211 snap_id = rbd_snap_id_by_name(rbd_dev, spec->snap_name);
4212 if (snap_id == CEPH_NOSNAP)
4214 spec->snap_id = snap_id;
4216 spec->snap_id = CEPH_NOSNAP;
4222 /* Get the pool name; we have to make our own copy of this */
4224 pool_name = ceph_pg_pool_name_by_id(osdc->osdmap, spec->pool_id);
4226 rbd_warn(rbd_dev, "no pool with id %llu", spec->pool_id);
4229 pool_name = kstrdup(pool_name, GFP_KERNEL);
4233 /* Fetch the image name; tolerate failure here */
4235 image_name = rbd_dev_image_name(rbd_dev);
4237 rbd_warn(rbd_dev, "unable to get image name");
4239 /* Look up the snapshot name, and make a copy */
4241 snap_name = rbd_snap_name(rbd_dev, spec->snap_id);
4242 if (IS_ERR(snap_name)) {
4243 ret = PTR_ERR(snap_name);
4247 spec->pool_name = pool_name;
4248 spec->image_name = image_name;
4249 spec->snap_name = snap_name;
4259 static int rbd_dev_v2_snap_context(struct rbd_device *rbd_dev)
4268 struct ceph_snap_context *snapc;
4272 * We'll need room for the seq value (maximum snapshot id),
4273 * snapshot count, and array of that many snapshot ids.
4274 * For now we have a fixed upper limit on the number we're
4275 * prepared to receive.
4277 size = sizeof (__le64) + sizeof (__le32) +
4278 RBD_MAX_SNAP_COUNT * sizeof (__le64);
4279 reply_buf = kzalloc(size, GFP_KERNEL);
4283 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
4284 "rbd", "get_snapcontext", NULL, 0,
4286 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
4291 end = reply_buf + ret;
4293 ceph_decode_64_safe(&p, end, seq, out);
4294 ceph_decode_32_safe(&p, end, snap_count, out);
4297 * Make sure the reported number of snapshot ids wouldn't go
4298 * beyond the end of our buffer. But before checking that,
4299 * make sure the computed size of the snapshot context we
4300 * allocate is representable in a size_t.
4302 if (snap_count > (SIZE_MAX - sizeof (struct ceph_snap_context))
4307 if (!ceph_has_room(&p, end, snap_count * sizeof (__le64)))
4311 snapc = ceph_create_snap_context(snap_count, GFP_KERNEL);
4317 for (i = 0; i < snap_count; i++)
4318 snapc->snaps[i] = ceph_decode_64(&p);
4320 ceph_put_snap_context(rbd_dev->header.snapc);
4321 rbd_dev->header.snapc = snapc;
4323 dout(" snap context seq = %llu, snap_count = %u\n",
4324 (unsigned long long)seq, (unsigned int)snap_count);
4331 static const char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev,
4342 size = sizeof (__le32) + RBD_MAX_SNAP_NAME_LEN;
4343 reply_buf = kmalloc(size, GFP_KERNEL);
4345 return ERR_PTR(-ENOMEM);
4347 snapid = cpu_to_le64(snap_id);
4348 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
4349 "rbd", "get_snapshot_name",
4350 &snapid, sizeof (snapid),
4352 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
4354 snap_name = ERR_PTR(ret);
4359 end = reply_buf + ret;
4360 snap_name = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL);
4361 if (IS_ERR(snap_name))
4364 dout(" snap_id 0x%016llx snap_name = %s\n",
4365 (unsigned long long)snap_id, snap_name);
4372 static int rbd_dev_v2_header_info(struct rbd_device *rbd_dev)
4374 bool first_time = rbd_dev->header.object_prefix == NULL;
4377 ret = rbd_dev_v2_image_size(rbd_dev);
4382 ret = rbd_dev_v2_header_onetime(rbd_dev);
4388 * If the image supports layering, get the parent info. We
4389 * need to probe the first time regardless. Thereafter we
4390 * only need to if there's a parent, to see if it has
4391 * disappeared due to the mapped image getting flattened.
4393 if (rbd_dev->header.features & RBD_FEATURE_LAYERING &&
4394 (first_time || rbd_dev->parent_spec)) {
4397 ret = rbd_dev_v2_parent_info(rbd_dev);
4402 * Print a warning if this is the initial probe and
4403 * the image has a parent. Don't print it if the
4404 * image now being probed is itself a parent. We
4405 * can tell at this point because we won't know its
4406 * pool name yet (just its pool id).
4408 warn = rbd_dev->parent_spec && rbd_dev->spec->pool_name;
4409 if (first_time && warn)
4410 rbd_warn(rbd_dev, "WARNING: kernel layering "
4411 "is EXPERIMENTAL!");
4414 if (rbd_dev->spec->snap_id == CEPH_NOSNAP)
4415 if (rbd_dev->mapping.size != rbd_dev->header.image_size)
4416 rbd_dev->mapping.size = rbd_dev->header.image_size;
4418 ret = rbd_dev_v2_snap_context(rbd_dev);
4419 dout("rbd_dev_v2_snap_context returned %d\n", ret);
4424 static int rbd_bus_add_dev(struct rbd_device *rbd_dev)
4429 dev = &rbd_dev->dev;
4430 dev->bus = &rbd_bus_type;
4431 dev->type = &rbd_device_type;
4432 dev->parent = &rbd_root_dev;
4433 dev->release = rbd_dev_device_release;
4434 dev_set_name(dev, "%d", rbd_dev->dev_id);
4435 ret = device_register(dev);
4440 static void rbd_bus_del_dev(struct rbd_device *rbd_dev)
4442 device_unregister(&rbd_dev->dev);
4446 * Get a unique rbd identifier for the given new rbd_dev, and add
4447 * the rbd_dev to the global list.
4449 static int rbd_dev_id_get(struct rbd_device *rbd_dev)
4453 new_dev_id = ida_simple_get(&rbd_dev_id_ida,
4454 0, minor_to_rbd_dev_id(1 << MINORBITS),
4459 rbd_dev->dev_id = new_dev_id;
4461 spin_lock(&rbd_dev_list_lock);
4462 list_add_tail(&rbd_dev->node, &rbd_dev_list);
4463 spin_unlock(&rbd_dev_list_lock);
4465 dout("rbd_dev %p given dev id %d\n", rbd_dev, rbd_dev->dev_id);
4471 * Remove an rbd_dev from the global list, and record that its
4472 * identifier is no longer in use.
4474 static void rbd_dev_id_put(struct rbd_device *rbd_dev)
4476 spin_lock(&rbd_dev_list_lock);
4477 list_del_init(&rbd_dev->node);
4478 spin_unlock(&rbd_dev_list_lock);
4480 ida_simple_remove(&rbd_dev_id_ida, rbd_dev->dev_id);
4482 dout("rbd_dev %p released dev id %d\n", rbd_dev, rbd_dev->dev_id);
4486 * Skips over white space at *buf, and updates *buf to point to the
4487 * first found non-space character (if any). Returns the length of
4488 * the token (string of non-white space characters) found. Note
4489 * that *buf must be terminated with '\0'.
4491 static inline size_t next_token(const char **buf)
4494 * These are the characters that produce nonzero for
4495 * isspace() in the "C" and "POSIX" locales.
4497 const char *spaces = " \f\n\r\t\v";
4499 *buf += strspn(*buf, spaces); /* Find start of token */
4501 return strcspn(*buf, spaces); /* Return token length */
4505 * Finds the next token in *buf, and if the provided token buffer is
4506 * big enough, copies the found token into it. The result, if
4507 * copied, is guaranteed to be terminated with '\0'. Note that *buf
4508 * must be terminated with '\0' on entry.
4510 * Returns the length of the token found (not including the '\0').
4511 * Return value will be 0 if no token is found, and it will be >=
4512 * token_size if the token would not fit.
4514 * The *buf pointer will be updated to point beyond the end of the
4515 * found token. Note that this occurs even if the token buffer is
4516 * too small to hold it.
4518 static inline size_t copy_token(const char **buf,
4524 len = next_token(buf);
4525 if (len < token_size) {
4526 memcpy(token, *buf, len);
4527 *(token + len) = '\0';
4535 * Finds the next token in *buf, dynamically allocates a buffer big
4536 * enough to hold a copy of it, and copies the token into the new
4537 * buffer. The copy is guaranteed to be terminated with '\0'. Note
4538 * that a duplicate buffer is created even for a zero-length token.
4540 * Returns a pointer to the newly-allocated duplicate, or a null
4541 * pointer if memory for the duplicate was not available. If
4542 * the lenp argument is a non-null pointer, the length of the token
4543 * (not including the '\0') is returned in *lenp.
4545 * If successful, the *buf pointer will be updated to point beyond
4546 * the end of the found token.
4548 * Note: uses GFP_KERNEL for allocation.
4550 static inline char *dup_token(const char **buf, size_t *lenp)
4555 len = next_token(buf);
4556 dup = kmemdup(*buf, len + 1, GFP_KERNEL);
4559 *(dup + len) = '\0';
4569 * Parse the options provided for an "rbd add" (i.e., rbd image
4570 * mapping) request. These arrive via a write to /sys/bus/rbd/add,
4571 * and the data written is passed here via a NUL-terminated buffer.
4572 * Returns 0 if successful or an error code otherwise.
4574 * The information extracted from these options is recorded in
4575 * the other parameters which return dynamically-allocated
4578 * The address of a pointer that will refer to a ceph options
4579 * structure. Caller must release the returned pointer using
4580 * ceph_destroy_options() when it is no longer needed.
4582 * Address of an rbd options pointer. Fully initialized by
4583 * this function; caller must release with kfree().
4585 * Address of an rbd image specification pointer. Fully
4586 * initialized by this function based on parsed options.
4587 * Caller must release with rbd_spec_put().
4589 * The options passed take this form:
4590 * <mon_addrs> <options> <pool_name> <image_name> [<snap_id>]
4593 * A comma-separated list of one or more monitor addresses.
4594 * A monitor address is an ip address, optionally followed
4595 * by a port number (separated by a colon).
4596 * I.e.: ip1[:port1][,ip2[:port2]...]
4598 * A comma-separated list of ceph and/or rbd options.
4600 * The name of the rados pool containing the rbd image.
4602 * The name of the image in that pool to map.
4604 * An optional snapshot id. If provided, the mapping will
4605 * present data from the image at the time that snapshot was
4606 * created. The image head is used if no snapshot id is
4607 * provided. Snapshot mappings are always read-only.
4609 static int rbd_add_parse_args(const char *buf,
4610 struct ceph_options **ceph_opts,
4611 struct rbd_options **opts,
4612 struct rbd_spec **rbd_spec)
4616 const char *mon_addrs;
4618 size_t mon_addrs_size;
4619 struct rbd_spec *spec = NULL;
4620 struct rbd_options *rbd_opts = NULL;
4621 struct ceph_options *copts;
4624 /* The first four tokens are required */
4626 len = next_token(&buf);
4628 rbd_warn(NULL, "no monitor address(es) provided");
4632 mon_addrs_size = len + 1;
4636 options = dup_token(&buf, NULL);
4640 rbd_warn(NULL, "no options provided");
4644 spec = rbd_spec_alloc();
4648 spec->pool_name = dup_token(&buf, NULL);
4649 if (!spec->pool_name)
4651 if (!*spec->pool_name) {
4652 rbd_warn(NULL, "no pool name provided");
4656 spec->image_name = dup_token(&buf, NULL);
4657 if (!spec->image_name)
4659 if (!*spec->image_name) {
4660 rbd_warn(NULL, "no image name provided");
4665 * Snapshot name is optional; default is to use "-"
4666 * (indicating the head/no snapshot).
4668 len = next_token(&buf);
4670 buf = RBD_SNAP_HEAD_NAME; /* No snapshot supplied */
4671 len = sizeof (RBD_SNAP_HEAD_NAME) - 1;
4672 } else if (len > RBD_MAX_SNAP_NAME_LEN) {
4673 ret = -ENAMETOOLONG;
4676 snap_name = kmemdup(buf, len + 1, GFP_KERNEL);
4679 *(snap_name + len) = '\0';
4680 spec->snap_name = snap_name;
4682 /* Initialize all rbd options to the defaults */
4684 rbd_opts = kzalloc(sizeof (*rbd_opts), GFP_KERNEL);
4688 rbd_opts->read_only = RBD_READ_ONLY_DEFAULT;
4690 copts = ceph_parse_options(options, mon_addrs,
4691 mon_addrs + mon_addrs_size - 1,
4692 parse_rbd_opts_token, rbd_opts);
4693 if (IS_ERR(copts)) {
4694 ret = PTR_ERR(copts);
4715 * An rbd format 2 image has a unique identifier, distinct from the
4716 * name given to it by the user. Internally, that identifier is
4717 * what's used to specify the names of objects related to the image.
4719 * A special "rbd id" object is used to map an rbd image name to its
4720 * id. If that object doesn't exist, then there is no v2 rbd image
4721 * with the supplied name.
4723 * This function will record the given rbd_dev's image_id field if
4724 * it can be determined, and in that case will return 0. If any
4725 * errors occur a negative errno will be returned and the rbd_dev's
4726 * image_id field will be unchanged (and should be NULL).
4728 static int rbd_dev_image_id(struct rbd_device *rbd_dev)
4737 * When probing a parent image, the image id is already
4738 * known (and the image name likely is not). There's no
4739 * need to fetch the image id again in this case. We
4740 * do still need to set the image format though.
4742 if (rbd_dev->spec->image_id) {
4743 rbd_dev->image_format = *rbd_dev->spec->image_id ? 2 : 1;
4749 * First, see if the format 2 image id file exists, and if
4750 * so, get the image's persistent id from it.
4752 size = sizeof (RBD_ID_PREFIX) + strlen(rbd_dev->spec->image_name);
4753 object_name = kmalloc(size, GFP_NOIO);
4756 sprintf(object_name, "%s%s", RBD_ID_PREFIX, rbd_dev->spec->image_name);
4757 dout("rbd id object name is %s\n", object_name);
4759 /* Response will be an encoded string, which includes a length */
4761 size = sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX;
4762 response = kzalloc(size, GFP_NOIO);
4768 /* If it doesn't exist we'll assume it's a format 1 image */
4770 ret = rbd_obj_method_sync(rbd_dev, object_name,
4771 "rbd", "get_id", NULL, 0,
4772 response, RBD_IMAGE_ID_LEN_MAX);
4773 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
4774 if (ret == -ENOENT) {
4775 image_id = kstrdup("", GFP_KERNEL);
4776 ret = image_id ? 0 : -ENOMEM;
4778 rbd_dev->image_format = 1;
4779 } else if (ret > sizeof (__le32)) {
4782 image_id = ceph_extract_encoded_string(&p, p + ret,
4784 ret = IS_ERR(image_id) ? PTR_ERR(image_id) : 0;
4786 rbd_dev->image_format = 2;
4792 rbd_dev->spec->image_id = image_id;
4793 dout("image_id is %s\n", image_id);
4803 * Undo whatever state changes are made by v1 or v2 header info
4806 static void rbd_dev_unprobe(struct rbd_device *rbd_dev)
4808 struct rbd_image_header *header;
4810 /* Drop parent reference unless it's already been done (or none) */
4812 if (rbd_dev->parent_overlap)
4813 rbd_dev_parent_put(rbd_dev);
4815 /* Free dynamic fields from the header, then zero it out */
4817 header = &rbd_dev->header;
4818 ceph_put_snap_context(header->snapc);
4819 kfree(header->snap_sizes);
4820 kfree(header->snap_names);
4821 kfree(header->object_prefix);
4822 memset(header, 0, sizeof (*header));
4825 static int rbd_dev_v2_header_onetime(struct rbd_device *rbd_dev)
4829 ret = rbd_dev_v2_object_prefix(rbd_dev);
4834 * Get the and check features for the image. Currently the
4835 * features are assumed to never change.
4837 ret = rbd_dev_v2_features(rbd_dev);
4841 /* If the image supports fancy striping, get its parameters */
4843 if (rbd_dev->header.features & RBD_FEATURE_STRIPINGV2) {
4844 ret = rbd_dev_v2_striping_info(rbd_dev);
4848 /* No support for crypto and compression type format 2 images */
4852 rbd_dev->header.features = 0;
4853 kfree(rbd_dev->header.object_prefix);
4854 rbd_dev->header.object_prefix = NULL;
4859 static int rbd_dev_probe_parent(struct rbd_device *rbd_dev)
4861 struct rbd_device *parent = NULL;
4862 struct rbd_spec *parent_spec;
4863 struct rbd_client *rbdc;
4866 if (!rbd_dev->parent_spec)
4869 * We need to pass a reference to the client and the parent
4870 * spec when creating the parent rbd_dev. Images related by
4871 * parent/child relationships always share both.
4873 parent_spec = rbd_spec_get(rbd_dev->parent_spec);
4874 rbdc = __rbd_get_client(rbd_dev->rbd_client);
4877 parent = rbd_dev_create(rbdc, parent_spec);
4881 ret = rbd_dev_image_probe(parent, false);
4884 rbd_dev->parent = parent;
4885 atomic_set(&rbd_dev->parent_ref, 1);
4890 rbd_dev_unparent(rbd_dev);
4891 kfree(rbd_dev->header_name);
4892 rbd_dev_destroy(parent);
4894 rbd_put_client(rbdc);
4895 rbd_spec_put(parent_spec);
4901 static int rbd_dev_device_setup(struct rbd_device *rbd_dev)
4905 /* Get an id and fill in device name. */
4907 ret = rbd_dev_id_get(rbd_dev);
4911 BUILD_BUG_ON(DEV_NAME_LEN
4912 < sizeof (RBD_DRV_NAME) + MAX_INT_FORMAT_WIDTH);
4913 sprintf(rbd_dev->name, "%s%d", RBD_DRV_NAME, rbd_dev->dev_id);
4915 /* Record our major and minor device numbers. */
4917 if (!single_major) {
4918 ret = register_blkdev(0, rbd_dev->name);
4922 rbd_dev->major = ret;
4925 rbd_dev->major = rbd_major;
4926 rbd_dev->minor = rbd_dev_id_to_minor(rbd_dev->dev_id);
4929 /* Set up the blkdev mapping. */
4931 ret = rbd_init_disk(rbd_dev);
4933 goto err_out_blkdev;
4935 ret = rbd_dev_mapping_set(rbd_dev);
4938 set_capacity(rbd_dev->disk, rbd_dev->mapping.size / SECTOR_SIZE);
4940 ret = rbd_bus_add_dev(rbd_dev);
4942 goto err_out_mapping;
4944 /* Everything's ready. Announce the disk to the world. */
4946 set_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags);
4947 add_disk(rbd_dev->disk);
4949 pr_info("%s: added with size 0x%llx\n", rbd_dev->disk->disk_name,
4950 (unsigned long long) rbd_dev->mapping.size);
4955 rbd_dev_mapping_clear(rbd_dev);
4957 rbd_free_disk(rbd_dev);
4960 unregister_blkdev(rbd_dev->major, rbd_dev->name);
4962 rbd_dev_id_put(rbd_dev);
4963 rbd_dev_mapping_clear(rbd_dev);
4968 static int rbd_dev_header_name(struct rbd_device *rbd_dev)
4970 struct rbd_spec *spec = rbd_dev->spec;
4973 /* Record the header object name for this rbd image. */
4975 rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
4977 if (rbd_dev->image_format == 1)
4978 size = strlen(spec->image_name) + sizeof (RBD_SUFFIX);
4980 size = sizeof (RBD_HEADER_PREFIX) + strlen(spec->image_id);
4982 rbd_dev->header_name = kmalloc(size, GFP_KERNEL);
4983 if (!rbd_dev->header_name)
4986 if (rbd_dev->image_format == 1)
4987 sprintf(rbd_dev->header_name, "%s%s",
4988 spec->image_name, RBD_SUFFIX);
4990 sprintf(rbd_dev->header_name, "%s%s",
4991 RBD_HEADER_PREFIX, spec->image_id);
4995 static void rbd_dev_image_release(struct rbd_device *rbd_dev)
4997 rbd_dev_unprobe(rbd_dev);
4998 kfree(rbd_dev->header_name);
4999 rbd_dev->header_name = NULL;
5000 rbd_dev->image_format = 0;
5001 kfree(rbd_dev->spec->image_id);
5002 rbd_dev->spec->image_id = NULL;
5004 rbd_dev_destroy(rbd_dev);
5008 * Probe for the existence of the header object for the given rbd
5009 * device. If this image is the one being mapped (i.e., not a
5010 * parent), initiate a watch on its header object before using that
5011 * object to get detailed information about the rbd image.
5013 static int rbd_dev_image_probe(struct rbd_device *rbd_dev, bool mapping)
5018 * Get the id from the image id object. Unless there's an
5019 * error, rbd_dev->spec->image_id will be filled in with
5020 * a dynamically-allocated string, and rbd_dev->image_format
5021 * will be set to either 1 or 2.
5023 ret = rbd_dev_image_id(rbd_dev);
5026 rbd_assert(rbd_dev->spec->image_id);
5027 rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
5029 ret = rbd_dev_header_name(rbd_dev);
5031 goto err_out_format;
5034 ret = rbd_dev_header_watch_sync(rbd_dev);
5036 goto out_header_name;
5039 if (rbd_dev->image_format == 1)
5040 ret = rbd_dev_v1_header_info(rbd_dev);
5042 ret = rbd_dev_v2_header_info(rbd_dev);
5046 ret = rbd_dev_spec_update(rbd_dev);
5050 ret = rbd_dev_probe_parent(rbd_dev);
5054 dout("discovered format %u image, header name is %s\n",
5055 rbd_dev->image_format, rbd_dev->header_name);
5059 rbd_dev_unprobe(rbd_dev);
5062 rbd_dev_header_unwatch_sync(rbd_dev);
5064 kfree(rbd_dev->header_name);
5065 rbd_dev->header_name = NULL;
5067 rbd_dev->image_format = 0;
5068 kfree(rbd_dev->spec->image_id);
5069 rbd_dev->spec->image_id = NULL;
5071 dout("probe failed, returning %d\n", ret);
5076 static ssize_t do_rbd_add(struct bus_type *bus,
5080 struct rbd_device *rbd_dev = NULL;
5081 struct ceph_options *ceph_opts = NULL;
5082 struct rbd_options *rbd_opts = NULL;
5083 struct rbd_spec *spec = NULL;
5084 struct rbd_client *rbdc;
5085 struct ceph_osd_client *osdc;
5089 if (!try_module_get(THIS_MODULE))
5092 /* parse add command */
5093 rc = rbd_add_parse_args(buf, &ceph_opts, &rbd_opts, &spec);
5095 goto err_out_module;
5096 read_only = rbd_opts->read_only;
5098 rbd_opts = NULL; /* done with this */
5100 rbdc = rbd_get_client(ceph_opts);
5107 osdc = &rbdc->client->osdc;
5108 rc = ceph_pg_poolid_by_name(osdc->osdmap, spec->pool_name);
5110 goto err_out_client;
5111 spec->pool_id = (u64)rc;
5113 /* The ceph file layout needs to fit pool id in 32 bits */
5115 if (spec->pool_id > (u64)U32_MAX) {
5116 rbd_warn(NULL, "pool id too large (%llu > %u)\n",
5117 (unsigned long long)spec->pool_id, U32_MAX);
5119 goto err_out_client;
5122 rbd_dev = rbd_dev_create(rbdc, spec);
5124 goto err_out_client;
5125 rbdc = NULL; /* rbd_dev now owns this */
5126 spec = NULL; /* rbd_dev now owns this */
5128 rc = rbd_dev_image_probe(rbd_dev, true);
5130 goto err_out_rbd_dev;
5132 /* If we are mapping a snapshot it must be marked read-only */
5134 if (rbd_dev->spec->snap_id != CEPH_NOSNAP)
5136 rbd_dev->mapping.read_only = read_only;
5138 rc = rbd_dev_device_setup(rbd_dev);
5141 * rbd_dev_header_unwatch_sync() can't be moved into
5142 * rbd_dev_image_release() without refactoring, see
5143 * commit 1f3ef78861ac.
5145 rbd_dev_header_unwatch_sync(rbd_dev);
5146 rbd_dev_image_release(rbd_dev);
5147 goto err_out_module;
5153 rbd_dev_destroy(rbd_dev);
5155 rbd_put_client(rbdc);
5159 module_put(THIS_MODULE);
5161 dout("Error adding device %s\n", buf);
5166 static ssize_t rbd_add(struct bus_type *bus,
5173 return do_rbd_add(bus, buf, count);
5176 static ssize_t rbd_add_single_major(struct bus_type *bus,
5180 return do_rbd_add(bus, buf, count);
5183 static void rbd_dev_device_release(struct device *dev)
5185 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
5187 rbd_free_disk(rbd_dev);
5188 clear_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags);
5189 rbd_dev_mapping_clear(rbd_dev);
5191 unregister_blkdev(rbd_dev->major, rbd_dev->name);
5192 rbd_dev_id_put(rbd_dev);
5193 rbd_dev_mapping_clear(rbd_dev);
5196 static void rbd_dev_remove_parent(struct rbd_device *rbd_dev)
5198 while (rbd_dev->parent) {
5199 struct rbd_device *first = rbd_dev;
5200 struct rbd_device *second = first->parent;
5201 struct rbd_device *third;
5204 * Follow to the parent with no grandparent and
5207 while (second && (third = second->parent)) {
5212 rbd_dev_image_release(second);
5213 first->parent = NULL;
5214 first->parent_overlap = 0;
5216 rbd_assert(first->parent_spec);
5217 rbd_spec_put(first->parent_spec);
5218 first->parent_spec = NULL;
5222 static ssize_t do_rbd_remove(struct bus_type *bus,
5226 struct rbd_device *rbd_dev = NULL;
5227 struct list_head *tmp;
5230 bool already = false;
5233 ret = kstrtoul(buf, 10, &ul);
5237 /* convert to int; abort if we lost anything in the conversion */
5243 spin_lock(&rbd_dev_list_lock);
5244 list_for_each(tmp, &rbd_dev_list) {
5245 rbd_dev = list_entry(tmp, struct rbd_device, node);
5246 if (rbd_dev->dev_id == dev_id) {
5252 spin_lock_irq(&rbd_dev->lock);
5253 if (rbd_dev->open_count)
5256 already = test_and_set_bit(RBD_DEV_FLAG_REMOVING,
5258 spin_unlock_irq(&rbd_dev->lock);
5260 spin_unlock(&rbd_dev_list_lock);
5261 if (ret < 0 || already)
5264 rbd_dev_header_unwatch_sync(rbd_dev);
5266 * flush remaining watch callbacks - these must be complete
5267 * before the osd_client is shutdown
5269 dout("%s: flushing notifies", __func__);
5270 ceph_osdc_flush_notifies(&rbd_dev->rbd_client->client->osdc);
5273 * Don't free anything from rbd_dev->disk until after all
5274 * notifies are completely processed. Otherwise
5275 * rbd_bus_del_dev() will race with rbd_watch_cb(), resulting
5276 * in a potential use after free of rbd_dev->disk or rbd_dev.
5278 rbd_bus_del_dev(rbd_dev);
5279 rbd_dev_image_release(rbd_dev);
5280 module_put(THIS_MODULE);
5285 static ssize_t rbd_remove(struct bus_type *bus,
5292 return do_rbd_remove(bus, buf, count);
5295 static ssize_t rbd_remove_single_major(struct bus_type *bus,
5299 return do_rbd_remove(bus, buf, count);
5303 * create control files in sysfs
5306 static int rbd_sysfs_init(void)
5310 ret = device_register(&rbd_root_dev);
5314 ret = bus_register(&rbd_bus_type);
5316 device_unregister(&rbd_root_dev);
5321 static void rbd_sysfs_cleanup(void)
5323 bus_unregister(&rbd_bus_type);
5324 device_unregister(&rbd_root_dev);
5327 static int rbd_slab_init(void)
5329 rbd_assert(!rbd_img_request_cache);
5330 rbd_img_request_cache = kmem_cache_create("rbd_img_request",
5331 sizeof (struct rbd_img_request),
5332 __alignof__(struct rbd_img_request),
5334 if (!rbd_img_request_cache)
5337 rbd_assert(!rbd_obj_request_cache);
5338 rbd_obj_request_cache = kmem_cache_create("rbd_obj_request",
5339 sizeof (struct rbd_obj_request),
5340 __alignof__(struct rbd_obj_request),
5342 if (!rbd_obj_request_cache)
5345 rbd_assert(!rbd_segment_name_cache);
5346 rbd_segment_name_cache = kmem_cache_create("rbd_segment_name",
5347 CEPH_MAX_OID_NAME_LEN + 1, 1, 0, NULL);
5348 if (rbd_segment_name_cache)
5351 if (rbd_obj_request_cache) {
5352 kmem_cache_destroy(rbd_obj_request_cache);
5353 rbd_obj_request_cache = NULL;
5356 kmem_cache_destroy(rbd_img_request_cache);
5357 rbd_img_request_cache = NULL;
5362 static void rbd_slab_exit(void)
5364 rbd_assert(rbd_segment_name_cache);
5365 kmem_cache_destroy(rbd_segment_name_cache);
5366 rbd_segment_name_cache = NULL;
5368 rbd_assert(rbd_obj_request_cache);
5369 kmem_cache_destroy(rbd_obj_request_cache);
5370 rbd_obj_request_cache = NULL;
5372 rbd_assert(rbd_img_request_cache);
5373 kmem_cache_destroy(rbd_img_request_cache);
5374 rbd_img_request_cache = NULL;
5377 static int __init rbd_init(void)
5381 if (!libceph_compatible(NULL)) {
5382 rbd_warn(NULL, "libceph incompatibility (quitting)");
5386 rc = rbd_slab_init();
5391 rbd_major = register_blkdev(0, RBD_DRV_NAME);
5392 if (rbd_major < 0) {
5398 rc = rbd_sysfs_init();
5400 goto err_out_blkdev;
5403 pr_info("loaded (major %d)\n", rbd_major);
5405 pr_info("loaded\n");
5411 unregister_blkdev(rbd_major, RBD_DRV_NAME);
5417 static void __exit rbd_exit(void)
5419 rbd_sysfs_cleanup();
5421 unregister_blkdev(rbd_major, RBD_DRV_NAME);
5425 module_init(rbd_init);
5426 module_exit(rbd_exit);
5428 MODULE_AUTHOR("Alex Elder <elder@inktank.com>");
5429 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
5430 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
5431 /* following authorship retained from original osdblk.c */
5432 MODULE_AUTHOR("Jeff Garzik <jeff@garzik.org>");
5434 MODULE_DESCRIPTION("RADOS Block Device (RBD) driver");
5435 MODULE_LICENSE("GPL");