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>
45 #include <linux/workqueue.h>
47 #include "rbd_types.h"
49 #define RBD_DEBUG /* Activate rbd_assert() calls */
52 * The basic unit of block I/O is a sector. It is interpreted in a
53 * number of contexts in Linux (blk, bio, genhd), but the default is
54 * universally 512 bytes. These symbols are just slightly more
55 * meaningful than the bare numbers they represent.
57 #define SECTOR_SHIFT 9
58 #define SECTOR_SIZE (1ULL << SECTOR_SHIFT)
61 * Increment the given counter and return its updated value.
62 * If the counter is already 0 it will not be incremented.
63 * If the counter is already at its maximum value returns
64 * -EINVAL without updating it.
66 static int atomic_inc_return_safe(atomic_t *v)
70 counter = (unsigned int)__atomic_add_unless(v, 1, 0);
71 if (counter <= (unsigned int)INT_MAX)
79 /* Decrement the counter. Return the resulting value, or -EINVAL */
80 static int atomic_dec_return_safe(atomic_t *v)
84 counter = atomic_dec_return(v);
93 #define RBD_DRV_NAME "rbd"
95 #define RBD_MINORS_PER_MAJOR 256
96 #define RBD_SINGLE_MAJOR_PART_SHIFT 4
98 #define RBD_SNAP_DEV_NAME_PREFIX "snap_"
99 #define RBD_MAX_SNAP_NAME_LEN \
100 (NAME_MAX - (sizeof (RBD_SNAP_DEV_NAME_PREFIX) - 1))
102 #define RBD_MAX_SNAP_COUNT 510 /* allows max snapc to fit in 4KB */
104 #define RBD_SNAP_HEAD_NAME "-"
106 #define BAD_SNAP_INDEX U32_MAX /* invalid index into snap array */
108 /* This allows a single page to hold an image name sent by OSD */
109 #define RBD_IMAGE_NAME_LEN_MAX (PAGE_SIZE - sizeof (__le32) - 1)
110 #define RBD_IMAGE_ID_LEN_MAX 64
112 #define RBD_OBJ_PREFIX_LEN_MAX 64
116 #define RBD_FEATURE_LAYERING (1<<0)
117 #define RBD_FEATURE_STRIPINGV2 (1<<1)
118 #define RBD_FEATURES_ALL \
119 (RBD_FEATURE_LAYERING | RBD_FEATURE_STRIPINGV2)
121 /* Features supported by this (client software) implementation. */
123 #define RBD_FEATURES_SUPPORTED (RBD_FEATURES_ALL)
126 * An RBD device name will be "rbd#", where the "rbd" comes from
127 * RBD_DRV_NAME above, and # is a unique integer identifier.
128 * MAX_INT_FORMAT_WIDTH is used in ensuring DEV_NAME_LEN is big
129 * enough to hold all possible device names.
131 #define DEV_NAME_LEN 32
132 #define MAX_INT_FORMAT_WIDTH ((5 * sizeof (int)) / 2 + 1)
135 * block device image metadata (in-memory version)
137 struct rbd_image_header {
138 /* These six fields never change for a given rbd image */
145 u64 features; /* Might be changeable someday? */
147 /* The remaining fields need to be updated occasionally */
149 struct ceph_snap_context *snapc;
150 char *snap_names; /* format 1 only */
151 u64 *snap_sizes; /* format 1 only */
155 * An rbd image specification.
157 * The tuple (pool_id, image_id, snap_id) is sufficient to uniquely
158 * identify an image. Each rbd_dev structure includes a pointer to
159 * an rbd_spec structure that encapsulates this identity.
161 * Each of the id's in an rbd_spec has an associated name. For a
162 * user-mapped image, the names are supplied and the id's associated
163 * with them are looked up. For a layered image, a parent image is
164 * defined by the tuple, and the names are looked up.
166 * An rbd_dev structure contains a parent_spec pointer which is
167 * non-null if the image it represents is a child in a layered
168 * image. This pointer will refer to the rbd_spec structure used
169 * by the parent rbd_dev for its own identity (i.e., the structure
170 * is shared between the parent and child).
172 * Since these structures are populated once, during the discovery
173 * phase of image construction, they are effectively immutable so
174 * we make no effort to synchronize access to them.
176 * Note that code herein does not assume the image name is known (it
177 * could be a null pointer).
181 const char *pool_name;
183 const char *image_id;
184 const char *image_name;
187 const char *snap_name;
193 * an instance of the client. multiple devices may share an rbd client.
196 struct ceph_client *client;
198 struct list_head node;
201 struct rbd_img_request;
202 typedef void (*rbd_img_callback_t)(struct rbd_img_request *);
204 #define BAD_WHICH U32_MAX /* Good which or bad which, which? */
206 struct rbd_obj_request;
207 typedef void (*rbd_obj_callback_t)(struct rbd_obj_request *);
209 enum obj_request_type {
210 OBJ_REQUEST_NODATA, OBJ_REQUEST_BIO, OBJ_REQUEST_PAGES
213 enum obj_operation_type {
220 OBJ_REQ_DONE, /* completion flag: not done = 0, done = 1 */
221 OBJ_REQ_IMG_DATA, /* object usage: standalone = 0, image = 1 */
222 OBJ_REQ_KNOWN, /* EXISTS flag valid: no = 0, yes = 1 */
223 OBJ_REQ_EXISTS, /* target exists: no = 0, yes = 1 */
226 struct rbd_obj_request {
227 const char *object_name;
228 u64 offset; /* object start byte */
229 u64 length; /* bytes from offset */
233 * An object request associated with an image will have its
234 * img_data flag set; a standalone object request will not.
236 * A standalone object request will have which == BAD_WHICH
237 * and a null obj_request pointer.
239 * An object request initiated in support of a layered image
240 * object (to check for its existence before a write) will
241 * have which == BAD_WHICH and a non-null obj_request pointer.
243 * Finally, an object request for rbd image data will have
244 * which != BAD_WHICH, and will have a non-null img_request
245 * pointer. The value of which will be in the range
246 * 0..(img_request->obj_request_count-1).
249 struct rbd_obj_request *obj_request; /* STAT op */
251 struct rbd_img_request *img_request;
253 /* links for img_request->obj_requests list */
254 struct list_head links;
257 u32 which; /* posn image request list */
259 enum obj_request_type type;
261 struct bio *bio_list;
267 struct page **copyup_pages;
268 u32 copyup_page_count;
270 struct ceph_osd_request *osd_req;
272 u64 xferred; /* bytes transferred */
275 rbd_obj_callback_t callback;
276 struct completion completion;
282 IMG_REQ_WRITE, /* I/O direction: read = 0, write = 1 */
283 IMG_REQ_CHILD, /* initiator: block = 0, child image = 1 */
284 IMG_REQ_LAYERED, /* ENOENT handling: normal = 0, layered = 1 */
285 IMG_REQ_DISCARD, /* discard: normal = 0, discard request = 1 */
288 struct rbd_img_request {
289 struct rbd_device *rbd_dev;
290 u64 offset; /* starting image byte offset */
291 u64 length; /* byte count from offset */
294 u64 snap_id; /* for reads */
295 struct ceph_snap_context *snapc; /* for writes */
298 struct request *rq; /* block request */
299 struct rbd_obj_request *obj_request; /* obj req initiator */
301 struct page **copyup_pages;
302 u32 copyup_page_count;
303 spinlock_t completion_lock;/* protects next_completion */
305 rbd_img_callback_t callback;
306 u64 xferred;/* aggregate bytes transferred */
307 int result; /* first nonzero obj_request result */
309 u32 obj_request_count;
310 struct list_head obj_requests; /* rbd_obj_request structs */
315 #define for_each_obj_request(ireq, oreq) \
316 list_for_each_entry(oreq, &(ireq)->obj_requests, links)
317 #define for_each_obj_request_from(ireq, oreq) \
318 list_for_each_entry_from(oreq, &(ireq)->obj_requests, links)
319 #define for_each_obj_request_safe(ireq, oreq, n) \
320 list_for_each_entry_safe_reverse(oreq, n, &(ireq)->obj_requests, links)
332 int dev_id; /* blkdev unique id */
334 int major; /* blkdev assigned major */
336 struct gendisk *disk; /* blkdev's gendisk and rq */
338 u32 image_format; /* Either 1 or 2 */
339 struct rbd_client *rbd_client;
341 char name[DEV_NAME_LEN]; /* blkdev name, e.g. rbd3 */
343 struct list_head rq_queue; /* incoming rq queue */
344 spinlock_t lock; /* queue, flags, open_count */
345 struct work_struct rq_work;
347 struct rbd_image_header header;
348 unsigned long flags; /* possibly lock protected */
349 struct rbd_spec *spec;
353 struct ceph_file_layout layout;
355 struct ceph_osd_event *watch_event;
356 struct rbd_obj_request *watch_request;
358 struct rbd_spec *parent_spec;
361 struct rbd_device *parent;
363 /* protects updating the header */
364 struct rw_semaphore header_rwsem;
366 struct rbd_mapping mapping;
368 struct list_head node;
372 unsigned long open_count; /* protected by lock */
376 * Flag bits for rbd_dev->flags. If atomicity is required,
377 * rbd_dev->lock is used to protect access.
379 * Currently, only the "removing" flag (which is coupled with the
380 * "open_count" field) requires atomic access.
383 RBD_DEV_FLAG_EXISTS, /* mapped snapshot has not been deleted */
384 RBD_DEV_FLAG_REMOVING, /* this mapping is being removed */
387 static DEFINE_MUTEX(client_mutex); /* Serialize client creation */
389 static LIST_HEAD(rbd_dev_list); /* devices */
390 static DEFINE_SPINLOCK(rbd_dev_list_lock);
392 static LIST_HEAD(rbd_client_list); /* clients */
393 static DEFINE_SPINLOCK(rbd_client_list_lock);
395 /* Slab caches for frequently-allocated structures */
397 static struct kmem_cache *rbd_img_request_cache;
398 static struct kmem_cache *rbd_obj_request_cache;
399 static struct kmem_cache *rbd_segment_name_cache;
401 static int rbd_major;
402 static DEFINE_IDA(rbd_dev_id_ida);
404 static struct workqueue_struct *rbd_wq;
407 * Default to false for now, as single-major requires >= 0.75 version of
408 * userspace rbd utility.
410 static bool single_major = false;
411 module_param(single_major, bool, S_IRUGO);
412 MODULE_PARM_DESC(single_major, "Use a single major number for all rbd devices (default: false)");
414 static int rbd_img_request_submit(struct rbd_img_request *img_request);
416 static void rbd_dev_device_release(struct device *dev);
418 static ssize_t rbd_add(struct bus_type *bus, const char *buf,
420 static ssize_t rbd_remove(struct bus_type *bus, const char *buf,
422 static ssize_t rbd_add_single_major(struct bus_type *bus, const char *buf,
424 static ssize_t rbd_remove_single_major(struct bus_type *bus, const char *buf,
426 static int rbd_dev_image_probe(struct rbd_device *rbd_dev, bool mapping);
427 static void rbd_spec_put(struct rbd_spec *spec);
429 static int rbd_dev_id_to_minor(int dev_id)
431 return dev_id << RBD_SINGLE_MAJOR_PART_SHIFT;
434 static int minor_to_rbd_dev_id(int minor)
436 return minor >> RBD_SINGLE_MAJOR_PART_SHIFT;
439 static BUS_ATTR(add, S_IWUSR, NULL, rbd_add);
440 static BUS_ATTR(remove, S_IWUSR, NULL, rbd_remove);
441 static BUS_ATTR(add_single_major, S_IWUSR, NULL, rbd_add_single_major);
442 static BUS_ATTR(remove_single_major, S_IWUSR, NULL, rbd_remove_single_major);
444 static struct attribute *rbd_bus_attrs[] = {
446 &bus_attr_remove.attr,
447 &bus_attr_add_single_major.attr,
448 &bus_attr_remove_single_major.attr,
452 static umode_t rbd_bus_is_visible(struct kobject *kobj,
453 struct attribute *attr, int index)
456 (attr == &bus_attr_add_single_major.attr ||
457 attr == &bus_attr_remove_single_major.attr))
463 static const struct attribute_group rbd_bus_group = {
464 .attrs = rbd_bus_attrs,
465 .is_visible = rbd_bus_is_visible,
467 __ATTRIBUTE_GROUPS(rbd_bus);
469 static struct bus_type rbd_bus_type = {
471 .bus_groups = rbd_bus_groups,
474 static void rbd_root_dev_release(struct device *dev)
478 static struct device rbd_root_dev = {
480 .release = rbd_root_dev_release,
483 static __printf(2, 3)
484 void rbd_warn(struct rbd_device *rbd_dev, const char *fmt, ...)
486 struct va_format vaf;
494 printk(KERN_WARNING "%s: %pV\n", RBD_DRV_NAME, &vaf);
495 else if (rbd_dev->disk)
496 printk(KERN_WARNING "%s: %s: %pV\n",
497 RBD_DRV_NAME, rbd_dev->disk->disk_name, &vaf);
498 else if (rbd_dev->spec && rbd_dev->spec->image_name)
499 printk(KERN_WARNING "%s: image %s: %pV\n",
500 RBD_DRV_NAME, rbd_dev->spec->image_name, &vaf);
501 else if (rbd_dev->spec && rbd_dev->spec->image_id)
502 printk(KERN_WARNING "%s: id %s: %pV\n",
503 RBD_DRV_NAME, rbd_dev->spec->image_id, &vaf);
505 printk(KERN_WARNING "%s: rbd_dev %p: %pV\n",
506 RBD_DRV_NAME, rbd_dev, &vaf);
511 #define rbd_assert(expr) \
512 if (unlikely(!(expr))) { \
513 printk(KERN_ERR "\nAssertion failure in %s() " \
515 "\trbd_assert(%s);\n\n", \
516 __func__, __LINE__, #expr); \
519 #else /* !RBD_DEBUG */
520 # define rbd_assert(expr) ((void) 0)
521 #endif /* !RBD_DEBUG */
523 static int rbd_img_obj_request_submit(struct rbd_obj_request *obj_request);
524 static void rbd_img_parent_read(struct rbd_obj_request *obj_request);
525 static void rbd_dev_remove_parent(struct rbd_device *rbd_dev);
527 static int rbd_dev_refresh(struct rbd_device *rbd_dev);
528 static int rbd_dev_v2_header_onetime(struct rbd_device *rbd_dev);
529 static int rbd_dev_header_info(struct rbd_device *rbd_dev);
530 static int rbd_dev_v2_parent_info(struct rbd_device *rbd_dev);
531 static const char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev,
533 static int _rbd_dev_v2_snap_size(struct rbd_device *rbd_dev, u64 snap_id,
534 u8 *order, u64 *snap_size);
535 static int _rbd_dev_v2_snap_features(struct rbd_device *rbd_dev, u64 snap_id,
537 static u64 rbd_snap_id_by_name(struct rbd_device *rbd_dev, const char *name);
539 static int rbd_open(struct block_device *bdev, fmode_t mode)
541 struct rbd_device *rbd_dev = bdev->bd_disk->private_data;
542 bool removing = false;
544 if ((mode & FMODE_WRITE) && rbd_dev->mapping.read_only)
547 spin_lock_irq(&rbd_dev->lock);
548 if (test_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags))
551 rbd_dev->open_count++;
552 spin_unlock_irq(&rbd_dev->lock);
556 (void) get_device(&rbd_dev->dev);
561 static void rbd_release(struct gendisk *disk, fmode_t mode)
563 struct rbd_device *rbd_dev = disk->private_data;
564 unsigned long open_count_before;
566 spin_lock_irq(&rbd_dev->lock);
567 open_count_before = rbd_dev->open_count--;
568 spin_unlock_irq(&rbd_dev->lock);
569 rbd_assert(open_count_before > 0);
571 put_device(&rbd_dev->dev);
574 static int rbd_ioctl_set_ro(struct rbd_device *rbd_dev, unsigned long arg)
579 bool ro_changed = false;
581 /* get_user() may sleep, so call it before taking rbd_dev->lock */
582 if (get_user(val, (int __user *)(arg)))
585 ro = val ? true : false;
586 /* Snapshot doesn't allow to write*/
587 if (rbd_dev->spec->snap_id != CEPH_NOSNAP && !ro)
590 spin_lock_irq(&rbd_dev->lock);
591 /* prevent others open this device */
592 if (rbd_dev->open_count > 1) {
597 if (rbd_dev->mapping.read_only != ro) {
598 rbd_dev->mapping.read_only = ro;
603 spin_unlock_irq(&rbd_dev->lock);
604 /* set_disk_ro() may sleep, so call it after releasing rbd_dev->lock */
605 if (ret == 0 && ro_changed)
606 set_disk_ro(rbd_dev->disk, ro ? 1 : 0);
611 static int rbd_ioctl(struct block_device *bdev, fmode_t mode,
612 unsigned int cmd, unsigned long arg)
614 struct rbd_device *rbd_dev = bdev->bd_disk->private_data;
619 ret = rbd_ioctl_set_ro(rbd_dev, arg);
629 static int rbd_compat_ioctl(struct block_device *bdev, fmode_t mode,
630 unsigned int cmd, unsigned long arg)
632 return rbd_ioctl(bdev, mode, cmd, arg);
634 #endif /* CONFIG_COMPAT */
636 static const struct block_device_operations rbd_bd_ops = {
637 .owner = THIS_MODULE,
639 .release = rbd_release,
642 .compat_ioctl = rbd_compat_ioctl,
647 * Initialize an rbd client instance. Success or not, this function
648 * consumes ceph_opts. Caller holds client_mutex.
650 static struct rbd_client *rbd_client_create(struct ceph_options *ceph_opts)
652 struct rbd_client *rbdc;
655 dout("%s:\n", __func__);
656 rbdc = kmalloc(sizeof(struct rbd_client), GFP_KERNEL);
660 kref_init(&rbdc->kref);
661 INIT_LIST_HEAD(&rbdc->node);
663 rbdc->client = ceph_create_client(ceph_opts, rbdc, 0, 0);
664 if (IS_ERR(rbdc->client))
666 ceph_opts = NULL; /* Now rbdc->client is responsible for ceph_opts */
668 ret = ceph_open_session(rbdc->client);
672 spin_lock(&rbd_client_list_lock);
673 list_add_tail(&rbdc->node, &rbd_client_list);
674 spin_unlock(&rbd_client_list_lock);
676 dout("%s: rbdc %p\n", __func__, rbdc);
680 ceph_destroy_client(rbdc->client);
685 ceph_destroy_options(ceph_opts);
686 dout("%s: error %d\n", __func__, ret);
691 static struct rbd_client *__rbd_get_client(struct rbd_client *rbdc)
693 kref_get(&rbdc->kref);
699 * Find a ceph client with specific addr and configuration. If
700 * found, bump its reference count.
702 static struct rbd_client *rbd_client_find(struct ceph_options *ceph_opts)
704 struct rbd_client *client_node;
707 if (ceph_opts->flags & CEPH_OPT_NOSHARE)
710 spin_lock(&rbd_client_list_lock);
711 list_for_each_entry(client_node, &rbd_client_list, node) {
712 if (!ceph_compare_options(ceph_opts, client_node->client)) {
713 __rbd_get_client(client_node);
719 spin_unlock(&rbd_client_list_lock);
721 return found ? client_node : NULL;
731 /* string args above */
734 /* Boolean args above */
738 static match_table_t rbd_opts_tokens = {
740 /* string args above */
741 {Opt_read_only, "read_only"},
742 {Opt_read_only, "ro"}, /* Alternate spelling */
743 {Opt_read_write, "read_write"},
744 {Opt_read_write, "rw"}, /* Alternate spelling */
745 /* Boolean args above */
753 #define RBD_READ_ONLY_DEFAULT false
755 static int parse_rbd_opts_token(char *c, void *private)
757 struct rbd_options *rbd_opts = private;
758 substring_t argstr[MAX_OPT_ARGS];
759 int token, intval, ret;
761 token = match_token(c, rbd_opts_tokens, argstr);
765 if (token < Opt_last_int) {
766 ret = match_int(&argstr[0], &intval);
768 pr_err("bad mount option arg (not int) "
772 dout("got int token %d val %d\n", token, intval);
773 } else if (token > Opt_last_int && token < Opt_last_string) {
774 dout("got string token %d val %s\n", token,
776 } else if (token > Opt_last_string && token < Opt_last_bool) {
777 dout("got Boolean token %d\n", token);
779 dout("got token %d\n", token);
784 rbd_opts->read_only = true;
787 rbd_opts->read_only = false;
796 static char* obj_op_name(enum obj_operation_type op_type)
811 * Get a ceph client with specific addr and configuration, if one does
812 * not exist create it. Either way, ceph_opts is consumed by this
815 static struct rbd_client *rbd_get_client(struct ceph_options *ceph_opts)
817 struct rbd_client *rbdc;
819 mutex_lock_nested(&client_mutex, SINGLE_DEPTH_NESTING);
820 rbdc = rbd_client_find(ceph_opts);
821 if (rbdc) /* using an existing client */
822 ceph_destroy_options(ceph_opts);
824 rbdc = rbd_client_create(ceph_opts);
825 mutex_unlock(&client_mutex);
831 * Destroy ceph client
833 * Caller must hold rbd_client_list_lock.
835 static void rbd_client_release(struct kref *kref)
837 struct rbd_client *rbdc = container_of(kref, struct rbd_client, kref);
839 dout("%s: rbdc %p\n", __func__, rbdc);
840 spin_lock(&rbd_client_list_lock);
841 list_del(&rbdc->node);
842 spin_unlock(&rbd_client_list_lock);
844 ceph_destroy_client(rbdc->client);
849 * Drop reference to ceph client node. If it's not referenced anymore, release
852 static void rbd_put_client(struct rbd_client *rbdc)
855 kref_put(&rbdc->kref, rbd_client_release);
858 static bool rbd_image_format_valid(u32 image_format)
860 return image_format == 1 || image_format == 2;
863 static bool rbd_dev_ondisk_valid(struct rbd_image_header_ondisk *ondisk)
868 /* The header has to start with the magic rbd header text */
869 if (memcmp(&ondisk->text, RBD_HEADER_TEXT, sizeof (RBD_HEADER_TEXT)))
872 /* The bio layer requires at least sector-sized I/O */
874 if (ondisk->options.order < SECTOR_SHIFT)
877 /* If we use u64 in a few spots we may be able to loosen this */
879 if (ondisk->options.order > 8 * sizeof (int) - 1)
883 * The size of a snapshot header has to fit in a size_t, and
884 * that limits the number of snapshots.
886 snap_count = le32_to_cpu(ondisk->snap_count);
887 size = SIZE_MAX - sizeof (struct ceph_snap_context);
888 if (snap_count > size / sizeof (__le64))
892 * Not only that, but the size of the entire the snapshot
893 * header must also be representable in a size_t.
895 size -= snap_count * sizeof (__le64);
896 if ((u64) size < le64_to_cpu(ondisk->snap_names_len))
903 * Fill an rbd image header with information from the given format 1
906 static int rbd_header_from_disk(struct rbd_device *rbd_dev,
907 struct rbd_image_header_ondisk *ondisk)
909 struct rbd_image_header *header = &rbd_dev->header;
910 bool first_time = header->object_prefix == NULL;
911 struct ceph_snap_context *snapc;
912 char *object_prefix = NULL;
913 char *snap_names = NULL;
914 u64 *snap_sizes = NULL;
920 /* Allocate this now to avoid having to handle failure below */
925 len = strnlen(ondisk->object_prefix,
926 sizeof (ondisk->object_prefix));
927 object_prefix = kmalloc(len + 1, GFP_KERNEL);
930 memcpy(object_prefix, ondisk->object_prefix, len);
931 object_prefix[len] = '\0';
934 /* Allocate the snapshot context and fill it in */
936 snap_count = le32_to_cpu(ondisk->snap_count);
937 snapc = ceph_create_snap_context(snap_count, GFP_KERNEL);
940 snapc->seq = le64_to_cpu(ondisk->snap_seq);
942 struct rbd_image_snap_ondisk *snaps;
943 u64 snap_names_len = le64_to_cpu(ondisk->snap_names_len);
945 /* We'll keep a copy of the snapshot names... */
947 if (snap_names_len > (u64)SIZE_MAX)
949 snap_names = kmalloc(snap_names_len, GFP_KERNEL);
953 /* ...as well as the array of their sizes. */
955 size = snap_count * sizeof (*header->snap_sizes);
956 snap_sizes = kmalloc(size, GFP_KERNEL);
961 * Copy the names, and fill in each snapshot's id
964 * Note that rbd_dev_v1_header_info() guarantees the
965 * ondisk buffer we're working with has
966 * snap_names_len bytes beyond the end of the
967 * snapshot id array, this memcpy() is safe.
969 memcpy(snap_names, &ondisk->snaps[snap_count], snap_names_len);
970 snaps = ondisk->snaps;
971 for (i = 0; i < snap_count; i++) {
972 snapc->snaps[i] = le64_to_cpu(snaps[i].id);
973 snap_sizes[i] = le64_to_cpu(snaps[i].image_size);
977 /* We won't fail any more, fill in the header */
980 header->object_prefix = object_prefix;
981 header->obj_order = ondisk->options.order;
982 header->crypt_type = ondisk->options.crypt_type;
983 header->comp_type = ondisk->options.comp_type;
984 /* The rest aren't used for format 1 images */
985 header->stripe_unit = 0;
986 header->stripe_count = 0;
987 header->features = 0;
989 ceph_put_snap_context(header->snapc);
990 kfree(header->snap_names);
991 kfree(header->snap_sizes);
994 /* The remaining fields always get updated (when we refresh) */
996 header->image_size = le64_to_cpu(ondisk->image_size);
997 header->snapc = snapc;
998 header->snap_names = snap_names;
999 header->snap_sizes = snap_sizes;
1007 ceph_put_snap_context(snapc);
1008 kfree(object_prefix);
1013 static const char *_rbd_dev_v1_snap_name(struct rbd_device *rbd_dev, u32 which)
1015 const char *snap_name;
1017 rbd_assert(which < rbd_dev->header.snapc->num_snaps);
1019 /* Skip over names until we find the one we are looking for */
1021 snap_name = rbd_dev->header.snap_names;
1023 snap_name += strlen(snap_name) + 1;
1025 return kstrdup(snap_name, GFP_KERNEL);
1029 * Snapshot id comparison function for use with qsort()/bsearch().
1030 * Note that result is for snapshots in *descending* order.
1032 static int snapid_compare_reverse(const void *s1, const void *s2)
1034 u64 snap_id1 = *(u64 *)s1;
1035 u64 snap_id2 = *(u64 *)s2;
1037 if (snap_id1 < snap_id2)
1039 return snap_id1 == snap_id2 ? 0 : -1;
1043 * Search a snapshot context to see if the given snapshot id is
1046 * Returns the position of the snapshot id in the array if it's found,
1047 * or BAD_SNAP_INDEX otherwise.
1049 * Note: The snapshot array is in kept sorted (by the osd) in
1050 * reverse order, highest snapshot id first.
1052 static u32 rbd_dev_snap_index(struct rbd_device *rbd_dev, u64 snap_id)
1054 struct ceph_snap_context *snapc = rbd_dev->header.snapc;
1057 found = bsearch(&snap_id, &snapc->snaps, snapc->num_snaps,
1058 sizeof (snap_id), snapid_compare_reverse);
1060 return found ? (u32)(found - &snapc->snaps[0]) : BAD_SNAP_INDEX;
1063 static const char *rbd_dev_v1_snap_name(struct rbd_device *rbd_dev,
1067 const char *snap_name;
1069 which = rbd_dev_snap_index(rbd_dev, snap_id);
1070 if (which == BAD_SNAP_INDEX)
1071 return ERR_PTR(-ENOENT);
1073 snap_name = _rbd_dev_v1_snap_name(rbd_dev, which);
1074 return snap_name ? snap_name : ERR_PTR(-ENOMEM);
1077 static const char *rbd_snap_name(struct rbd_device *rbd_dev, u64 snap_id)
1079 if (snap_id == CEPH_NOSNAP)
1080 return RBD_SNAP_HEAD_NAME;
1082 rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
1083 if (rbd_dev->image_format == 1)
1084 return rbd_dev_v1_snap_name(rbd_dev, snap_id);
1086 return rbd_dev_v2_snap_name(rbd_dev, snap_id);
1089 static int rbd_snap_size(struct rbd_device *rbd_dev, u64 snap_id,
1092 rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
1093 if (snap_id == CEPH_NOSNAP) {
1094 *snap_size = rbd_dev->header.image_size;
1095 } else if (rbd_dev->image_format == 1) {
1098 which = rbd_dev_snap_index(rbd_dev, snap_id);
1099 if (which == BAD_SNAP_INDEX)
1102 *snap_size = rbd_dev->header.snap_sizes[which];
1107 ret = _rbd_dev_v2_snap_size(rbd_dev, snap_id, NULL, &size);
1116 static int rbd_snap_features(struct rbd_device *rbd_dev, u64 snap_id,
1119 rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
1120 if (snap_id == CEPH_NOSNAP) {
1121 *snap_features = rbd_dev->header.features;
1122 } else if (rbd_dev->image_format == 1) {
1123 *snap_features = 0; /* No features for format 1 */
1128 ret = _rbd_dev_v2_snap_features(rbd_dev, snap_id, &features);
1132 *snap_features = features;
1137 static int rbd_dev_mapping_set(struct rbd_device *rbd_dev)
1139 u64 snap_id = rbd_dev->spec->snap_id;
1144 ret = rbd_snap_size(rbd_dev, snap_id, &size);
1147 ret = rbd_snap_features(rbd_dev, snap_id, &features);
1151 rbd_dev->mapping.size = size;
1152 rbd_dev->mapping.features = features;
1157 static void rbd_dev_mapping_clear(struct rbd_device *rbd_dev)
1159 rbd_dev->mapping.size = 0;
1160 rbd_dev->mapping.features = 0;
1163 static void rbd_segment_name_free(const char *name)
1165 /* The explicit cast here is needed to drop the const qualifier */
1167 kmem_cache_free(rbd_segment_name_cache, (void *)name);
1170 static const char *rbd_segment_name(struct rbd_device *rbd_dev, u64 offset)
1177 name = kmem_cache_alloc(rbd_segment_name_cache, GFP_NOIO);
1180 segment = offset >> rbd_dev->header.obj_order;
1181 name_format = "%s.%012llx";
1182 if (rbd_dev->image_format == 2)
1183 name_format = "%s.%016llx";
1184 ret = snprintf(name, CEPH_MAX_OID_NAME_LEN + 1, name_format,
1185 rbd_dev->header.object_prefix, segment);
1186 if (ret < 0 || ret > CEPH_MAX_OID_NAME_LEN) {
1187 pr_err("error formatting segment name for #%llu (%d)\n",
1189 rbd_segment_name_free(name);
1196 static u64 rbd_segment_offset(struct rbd_device *rbd_dev, u64 offset)
1198 u64 segment_size = (u64) 1 << rbd_dev->header.obj_order;
1200 return offset & (segment_size - 1);
1203 static u64 rbd_segment_length(struct rbd_device *rbd_dev,
1204 u64 offset, u64 length)
1206 u64 segment_size = (u64) 1 << rbd_dev->header.obj_order;
1208 offset &= segment_size - 1;
1210 rbd_assert(length <= U64_MAX - offset);
1211 if (offset + length > segment_size)
1212 length = segment_size - offset;
1218 * returns the size of an object in the image
1220 static u64 rbd_obj_bytes(struct rbd_image_header *header)
1222 return 1 << header->obj_order;
1229 static void bio_chain_put(struct bio *chain)
1235 chain = chain->bi_next;
1241 * zeros a bio chain, starting at specific offset
1243 static void zero_bio_chain(struct bio *chain, int start_ofs)
1246 struct bvec_iter iter;
1247 unsigned long flags;
1252 bio_for_each_segment(bv, chain, iter) {
1253 if (pos + bv.bv_len > start_ofs) {
1254 int remainder = max(start_ofs - pos, 0);
1255 buf = bvec_kmap_irq(&bv, &flags);
1256 memset(buf + remainder, 0,
1257 bv.bv_len - remainder);
1258 flush_dcache_page(bv.bv_page);
1259 bvec_kunmap_irq(buf, &flags);
1264 chain = chain->bi_next;
1269 * similar to zero_bio_chain(), zeros data defined by a page array,
1270 * starting at the given byte offset from the start of the array and
1271 * continuing up to the given end offset. The pages array is
1272 * assumed to be big enough to hold all bytes up to the end.
1274 static void zero_pages(struct page **pages, u64 offset, u64 end)
1276 struct page **page = &pages[offset >> PAGE_SHIFT];
1278 rbd_assert(end > offset);
1279 rbd_assert(end - offset <= (u64)SIZE_MAX);
1280 while (offset < end) {
1283 unsigned long flags;
1286 page_offset = offset & ~PAGE_MASK;
1287 length = min_t(size_t, PAGE_SIZE - page_offset, end - offset);
1288 local_irq_save(flags);
1289 kaddr = kmap_atomic(*page);
1290 memset(kaddr + page_offset, 0, length);
1291 flush_dcache_page(*page);
1292 kunmap_atomic(kaddr);
1293 local_irq_restore(flags);
1301 * Clone a portion of a bio, starting at the given byte offset
1302 * and continuing for the number of bytes indicated.
1304 static struct bio *bio_clone_range(struct bio *bio_src,
1305 unsigned int offset,
1311 bio = bio_clone(bio_src, gfpmask);
1313 return NULL; /* ENOMEM */
1315 bio_advance(bio, offset);
1316 bio->bi_iter.bi_size = len;
1322 * Clone a portion of a bio chain, starting at the given byte offset
1323 * into the first bio in the source chain and continuing for the
1324 * number of bytes indicated. The result is another bio chain of
1325 * exactly the given length, or a null pointer on error.
1327 * The bio_src and offset parameters are both in-out. On entry they
1328 * refer to the first source bio and the offset into that bio where
1329 * the start of data to be cloned is located.
1331 * On return, bio_src is updated to refer to the bio in the source
1332 * chain that contains first un-cloned byte, and *offset will
1333 * contain the offset of that byte within that bio.
1335 static struct bio *bio_chain_clone_range(struct bio **bio_src,
1336 unsigned int *offset,
1340 struct bio *bi = *bio_src;
1341 unsigned int off = *offset;
1342 struct bio *chain = NULL;
1345 /* Build up a chain of clone bios up to the limit */
1347 if (!bi || off >= bi->bi_iter.bi_size || !len)
1348 return NULL; /* Nothing to clone */
1352 unsigned int bi_size;
1356 rbd_warn(NULL, "bio_chain exhausted with %u left", len);
1357 goto out_err; /* EINVAL; ran out of bio's */
1359 bi_size = min_t(unsigned int, bi->bi_iter.bi_size - off, len);
1360 bio = bio_clone_range(bi, off, bi_size, gfpmask);
1362 goto out_err; /* ENOMEM */
1365 end = &bio->bi_next;
1368 if (off == bi->bi_iter.bi_size) {
1379 bio_chain_put(chain);
1385 * The default/initial value for all object request flags is 0. For
1386 * each flag, once its value is set to 1 it is never reset to 0
1389 static void obj_request_img_data_set(struct rbd_obj_request *obj_request)
1391 if (test_and_set_bit(OBJ_REQ_IMG_DATA, &obj_request->flags)) {
1392 struct rbd_device *rbd_dev;
1394 rbd_dev = obj_request->img_request->rbd_dev;
1395 rbd_warn(rbd_dev, "obj_request %p already marked img_data",
1400 static bool obj_request_img_data_test(struct rbd_obj_request *obj_request)
1403 return test_bit(OBJ_REQ_IMG_DATA, &obj_request->flags) != 0;
1406 static void obj_request_done_set(struct rbd_obj_request *obj_request)
1408 if (test_and_set_bit(OBJ_REQ_DONE, &obj_request->flags)) {
1409 struct rbd_device *rbd_dev = NULL;
1411 if (obj_request_img_data_test(obj_request))
1412 rbd_dev = obj_request->img_request->rbd_dev;
1413 rbd_warn(rbd_dev, "obj_request %p already marked done",
1418 static bool obj_request_done_test(struct rbd_obj_request *obj_request)
1421 return test_bit(OBJ_REQ_DONE, &obj_request->flags) != 0;
1425 * This sets the KNOWN flag after (possibly) setting the EXISTS
1426 * flag. The latter is set based on the "exists" value provided.
1428 * Note that for our purposes once an object exists it never goes
1429 * away again. It's possible that the response from two existence
1430 * checks are separated by the creation of the target object, and
1431 * the first ("doesn't exist") response arrives *after* the second
1432 * ("does exist"). In that case we ignore the second one.
1434 static void obj_request_existence_set(struct rbd_obj_request *obj_request,
1438 set_bit(OBJ_REQ_EXISTS, &obj_request->flags);
1439 set_bit(OBJ_REQ_KNOWN, &obj_request->flags);
1443 static bool obj_request_known_test(struct rbd_obj_request *obj_request)
1446 return test_bit(OBJ_REQ_KNOWN, &obj_request->flags) != 0;
1449 static bool obj_request_exists_test(struct rbd_obj_request *obj_request)
1452 return test_bit(OBJ_REQ_EXISTS, &obj_request->flags) != 0;
1455 static bool obj_request_overlaps_parent(struct rbd_obj_request *obj_request)
1457 struct rbd_device *rbd_dev = obj_request->img_request->rbd_dev;
1459 return obj_request->img_offset <
1460 round_up(rbd_dev->parent_overlap, rbd_obj_bytes(&rbd_dev->header));
1463 static void rbd_obj_request_get(struct rbd_obj_request *obj_request)
1465 dout("%s: obj %p (was %d)\n", __func__, obj_request,
1466 atomic_read(&obj_request->kref.refcount));
1467 kref_get(&obj_request->kref);
1470 static void rbd_obj_request_destroy(struct kref *kref);
1471 static void rbd_obj_request_put(struct rbd_obj_request *obj_request)
1473 rbd_assert(obj_request != NULL);
1474 dout("%s: obj %p (was %d)\n", __func__, obj_request,
1475 atomic_read(&obj_request->kref.refcount));
1476 kref_put(&obj_request->kref, rbd_obj_request_destroy);
1479 static void rbd_img_request_get(struct rbd_img_request *img_request)
1481 dout("%s: img %p (was %d)\n", __func__, img_request,
1482 atomic_read(&img_request->kref.refcount));
1483 kref_get(&img_request->kref);
1486 static bool img_request_child_test(struct rbd_img_request *img_request);
1487 static void rbd_parent_request_destroy(struct kref *kref);
1488 static void rbd_img_request_destroy(struct kref *kref);
1489 static void rbd_img_request_put(struct rbd_img_request *img_request)
1491 rbd_assert(img_request != NULL);
1492 dout("%s: img %p (was %d)\n", __func__, img_request,
1493 atomic_read(&img_request->kref.refcount));
1494 if (img_request_child_test(img_request))
1495 kref_put(&img_request->kref, rbd_parent_request_destroy);
1497 kref_put(&img_request->kref, rbd_img_request_destroy);
1500 static inline void rbd_img_obj_request_add(struct rbd_img_request *img_request,
1501 struct rbd_obj_request *obj_request)
1503 rbd_assert(obj_request->img_request == NULL);
1505 /* Image request now owns object's original reference */
1506 obj_request->img_request = img_request;
1507 obj_request->which = img_request->obj_request_count;
1508 rbd_assert(!obj_request_img_data_test(obj_request));
1509 obj_request_img_data_set(obj_request);
1510 rbd_assert(obj_request->which != BAD_WHICH);
1511 img_request->obj_request_count++;
1512 list_add_tail(&obj_request->links, &img_request->obj_requests);
1513 dout("%s: img %p obj %p w=%u\n", __func__, img_request, obj_request,
1514 obj_request->which);
1517 static inline void rbd_img_obj_request_del(struct rbd_img_request *img_request,
1518 struct rbd_obj_request *obj_request)
1520 rbd_assert(obj_request->which != BAD_WHICH);
1522 dout("%s: img %p obj %p w=%u\n", __func__, img_request, obj_request,
1523 obj_request->which);
1524 list_del(&obj_request->links);
1525 rbd_assert(img_request->obj_request_count > 0);
1526 img_request->obj_request_count--;
1527 rbd_assert(obj_request->which == img_request->obj_request_count);
1528 obj_request->which = BAD_WHICH;
1529 rbd_assert(obj_request_img_data_test(obj_request));
1530 rbd_assert(obj_request->img_request == img_request);
1531 obj_request->img_request = NULL;
1532 obj_request->callback = NULL;
1533 rbd_obj_request_put(obj_request);
1536 static bool obj_request_type_valid(enum obj_request_type type)
1539 case OBJ_REQUEST_NODATA:
1540 case OBJ_REQUEST_BIO:
1541 case OBJ_REQUEST_PAGES:
1548 static int rbd_obj_request_submit(struct ceph_osd_client *osdc,
1549 struct rbd_obj_request *obj_request)
1551 dout("%s %p\n", __func__, obj_request);
1552 return ceph_osdc_start_request(osdc, obj_request->osd_req, false);
1555 static void rbd_obj_request_end(struct rbd_obj_request *obj_request)
1557 dout("%s %p\n", __func__, obj_request);
1558 ceph_osdc_cancel_request(obj_request->osd_req);
1562 * Wait for an object request to complete. If interrupted, cancel the
1563 * underlying osd request.
1565 static int rbd_obj_request_wait(struct rbd_obj_request *obj_request)
1569 dout("%s %p\n", __func__, obj_request);
1571 ret = wait_for_completion_interruptible(&obj_request->completion);
1573 dout("%s %p interrupted\n", __func__, obj_request);
1574 rbd_obj_request_end(obj_request);
1578 dout("%s %p done\n", __func__, obj_request);
1582 static void rbd_img_request_complete(struct rbd_img_request *img_request)
1585 dout("%s: img %p\n", __func__, img_request);
1588 * If no error occurred, compute the aggregate transfer
1589 * count for the image request. We could instead use
1590 * atomic64_cmpxchg() to update it as each object request
1591 * completes; not clear which way is better off hand.
1593 if (!img_request->result) {
1594 struct rbd_obj_request *obj_request;
1597 for_each_obj_request(img_request, obj_request)
1598 xferred += obj_request->xferred;
1599 img_request->xferred = xferred;
1602 if (img_request->callback)
1603 img_request->callback(img_request);
1605 rbd_img_request_put(img_request);
1609 * The default/initial value for all image request flags is 0. Each
1610 * is conditionally set to 1 at image request initialization time
1611 * and currently never change thereafter.
1613 static void img_request_write_set(struct rbd_img_request *img_request)
1615 set_bit(IMG_REQ_WRITE, &img_request->flags);
1619 static bool img_request_write_test(struct rbd_img_request *img_request)
1622 return test_bit(IMG_REQ_WRITE, &img_request->flags) != 0;
1626 * Set the discard flag when the img_request is an discard request
1628 static void img_request_discard_set(struct rbd_img_request *img_request)
1630 set_bit(IMG_REQ_DISCARD, &img_request->flags);
1634 static bool img_request_discard_test(struct rbd_img_request *img_request)
1637 return test_bit(IMG_REQ_DISCARD, &img_request->flags) != 0;
1640 static void img_request_child_set(struct rbd_img_request *img_request)
1642 set_bit(IMG_REQ_CHILD, &img_request->flags);
1646 static void img_request_child_clear(struct rbd_img_request *img_request)
1648 clear_bit(IMG_REQ_CHILD, &img_request->flags);
1652 static bool img_request_child_test(struct rbd_img_request *img_request)
1655 return test_bit(IMG_REQ_CHILD, &img_request->flags) != 0;
1658 static void img_request_layered_set(struct rbd_img_request *img_request)
1660 set_bit(IMG_REQ_LAYERED, &img_request->flags);
1664 static void img_request_layered_clear(struct rbd_img_request *img_request)
1666 clear_bit(IMG_REQ_LAYERED, &img_request->flags);
1670 static bool img_request_layered_test(struct rbd_img_request *img_request)
1673 return test_bit(IMG_REQ_LAYERED, &img_request->flags) != 0;
1676 static enum obj_operation_type
1677 rbd_img_request_op_type(struct rbd_img_request *img_request)
1679 if (img_request_write_test(img_request))
1680 return OBJ_OP_WRITE;
1681 else if (img_request_discard_test(img_request))
1682 return OBJ_OP_DISCARD;
1688 rbd_img_obj_request_read_callback(struct rbd_obj_request *obj_request)
1690 u64 xferred = obj_request->xferred;
1691 u64 length = obj_request->length;
1693 dout("%s: obj %p img %p result %d %llu/%llu\n", __func__,
1694 obj_request, obj_request->img_request, obj_request->result,
1697 * ENOENT means a hole in the image. We zero-fill the entire
1698 * length of the request. A short read also implies zero-fill
1699 * to the end of the request. An error requires the whole
1700 * length of the request to be reported finished with an error
1701 * to the block layer. In each case we update the xferred
1702 * count to indicate the whole request was satisfied.
1704 rbd_assert(obj_request->type != OBJ_REQUEST_NODATA);
1705 if (obj_request->result == -ENOENT) {
1706 if (obj_request->type == OBJ_REQUEST_BIO)
1707 zero_bio_chain(obj_request->bio_list, 0);
1709 zero_pages(obj_request->pages, 0, length);
1710 obj_request->result = 0;
1711 } else if (xferred < length && !obj_request->result) {
1712 if (obj_request->type == OBJ_REQUEST_BIO)
1713 zero_bio_chain(obj_request->bio_list, xferred);
1715 zero_pages(obj_request->pages, xferred, length);
1717 obj_request->xferred = length;
1718 obj_request_done_set(obj_request);
1721 static void rbd_obj_request_complete(struct rbd_obj_request *obj_request)
1723 dout("%s: obj %p cb %p\n", __func__, obj_request,
1724 obj_request->callback);
1725 if (obj_request->callback)
1726 obj_request->callback(obj_request);
1728 complete_all(&obj_request->completion);
1731 static void rbd_osd_trivial_callback(struct rbd_obj_request *obj_request)
1733 dout("%s: obj %p\n", __func__, obj_request);
1734 obj_request_done_set(obj_request);
1737 static void rbd_osd_read_callback(struct rbd_obj_request *obj_request)
1739 struct rbd_img_request *img_request = NULL;
1740 struct rbd_device *rbd_dev = NULL;
1741 bool layered = false;
1743 if (obj_request_img_data_test(obj_request)) {
1744 img_request = obj_request->img_request;
1745 layered = img_request && img_request_layered_test(img_request);
1746 rbd_dev = img_request->rbd_dev;
1749 dout("%s: obj %p img %p result %d %llu/%llu\n", __func__,
1750 obj_request, img_request, obj_request->result,
1751 obj_request->xferred, obj_request->length);
1752 if (layered && obj_request->result == -ENOENT &&
1753 obj_request->img_offset < rbd_dev->parent_overlap)
1754 rbd_img_parent_read(obj_request);
1755 else if (img_request)
1756 rbd_img_obj_request_read_callback(obj_request);
1758 obj_request_done_set(obj_request);
1761 static void rbd_osd_write_callback(struct rbd_obj_request *obj_request)
1763 dout("%s: obj %p result %d %llu\n", __func__, obj_request,
1764 obj_request->result, obj_request->length);
1766 * There is no such thing as a successful short write. Set
1767 * it to our originally-requested length.
1769 obj_request->xferred = obj_request->length;
1770 obj_request_done_set(obj_request);
1773 static void rbd_osd_discard_callback(struct rbd_obj_request *obj_request)
1775 dout("%s: obj %p result %d %llu\n", __func__, obj_request,
1776 obj_request->result, obj_request->length);
1778 * There is no such thing as a successful short discard. Set
1779 * it to our originally-requested length.
1781 obj_request->xferred = obj_request->length;
1782 /* discarding a non-existent object is not a problem */
1783 if (obj_request->result == -ENOENT)
1784 obj_request->result = 0;
1785 obj_request_done_set(obj_request);
1789 * For a simple stat call there's nothing to do. We'll do more if
1790 * this is part of a write sequence for a layered image.
1792 static void rbd_osd_stat_callback(struct rbd_obj_request *obj_request)
1794 dout("%s: obj %p\n", __func__, obj_request);
1795 obj_request_done_set(obj_request);
1798 static void rbd_osd_req_callback(struct ceph_osd_request *osd_req,
1799 struct ceph_msg *msg)
1801 struct rbd_obj_request *obj_request = osd_req->r_priv;
1804 dout("%s: osd_req %p msg %p\n", __func__, osd_req, msg);
1805 rbd_assert(osd_req == obj_request->osd_req);
1806 if (obj_request_img_data_test(obj_request)) {
1807 rbd_assert(obj_request->img_request);
1808 rbd_assert(obj_request->which != BAD_WHICH);
1810 rbd_assert(obj_request->which == BAD_WHICH);
1813 if (osd_req->r_result < 0)
1814 obj_request->result = osd_req->r_result;
1816 rbd_assert(osd_req->r_num_ops <= CEPH_OSD_MAX_OP);
1819 * We support a 64-bit length, but ultimately it has to be
1820 * passed to blk_end_request(), which takes an unsigned int.
1822 obj_request->xferred = osd_req->r_reply_op_len[0];
1823 rbd_assert(obj_request->xferred < (u64)UINT_MAX);
1825 opcode = osd_req->r_ops[0].op;
1827 case CEPH_OSD_OP_READ:
1828 rbd_osd_read_callback(obj_request);
1830 case CEPH_OSD_OP_SETALLOCHINT:
1831 rbd_assert(osd_req->r_ops[1].op == CEPH_OSD_OP_WRITE);
1833 case CEPH_OSD_OP_WRITE:
1834 rbd_osd_write_callback(obj_request);
1836 case CEPH_OSD_OP_STAT:
1837 rbd_osd_stat_callback(obj_request);
1839 case CEPH_OSD_OP_DELETE:
1840 case CEPH_OSD_OP_TRUNCATE:
1841 case CEPH_OSD_OP_ZERO:
1842 rbd_osd_discard_callback(obj_request);
1844 case CEPH_OSD_OP_CALL:
1845 case CEPH_OSD_OP_NOTIFY_ACK:
1846 case CEPH_OSD_OP_WATCH:
1847 rbd_osd_trivial_callback(obj_request);
1850 rbd_warn(NULL, "%s: unsupported op %hu",
1851 obj_request->object_name, (unsigned short) opcode);
1855 if (obj_request_done_test(obj_request))
1856 rbd_obj_request_complete(obj_request);
1859 static void rbd_osd_req_format_read(struct rbd_obj_request *obj_request)
1861 struct rbd_img_request *img_request = obj_request->img_request;
1862 struct ceph_osd_request *osd_req = obj_request->osd_req;
1865 rbd_assert(osd_req != NULL);
1867 snap_id = img_request ? img_request->snap_id : CEPH_NOSNAP;
1868 ceph_osdc_build_request(osd_req, obj_request->offset,
1869 NULL, snap_id, NULL);
1872 static void rbd_osd_req_format_write(struct rbd_obj_request *obj_request)
1874 struct rbd_img_request *img_request = obj_request->img_request;
1875 struct ceph_osd_request *osd_req = obj_request->osd_req;
1876 struct ceph_snap_context *snapc;
1877 struct timespec mtime = CURRENT_TIME;
1879 rbd_assert(osd_req != NULL);
1881 snapc = img_request ? img_request->snapc : NULL;
1882 ceph_osdc_build_request(osd_req, obj_request->offset,
1883 snapc, CEPH_NOSNAP, &mtime);
1887 * Create an osd request. A read request has one osd op (read).
1888 * A write request has either one (watch) or two (hint+write) osd ops.
1889 * (All rbd data writes are prefixed with an allocation hint op, but
1890 * technically osd watch is a write request, hence this distinction.)
1892 static struct ceph_osd_request *rbd_osd_req_create(
1893 struct rbd_device *rbd_dev,
1894 enum obj_operation_type op_type,
1895 unsigned int num_ops,
1896 struct rbd_obj_request *obj_request)
1898 struct ceph_snap_context *snapc = NULL;
1899 struct ceph_osd_client *osdc;
1900 struct ceph_osd_request *osd_req;
1902 if (obj_request_img_data_test(obj_request) &&
1903 (op_type == OBJ_OP_DISCARD || op_type == OBJ_OP_WRITE)) {
1904 struct rbd_img_request *img_request = obj_request->img_request;
1905 if (op_type == OBJ_OP_WRITE) {
1906 rbd_assert(img_request_write_test(img_request));
1908 rbd_assert(img_request_discard_test(img_request));
1910 snapc = img_request->snapc;
1913 rbd_assert(num_ops == 1 || ((op_type == OBJ_OP_WRITE) && num_ops == 2));
1915 /* Allocate and initialize the request, for the num_ops ops */
1917 osdc = &rbd_dev->rbd_client->client->osdc;
1918 osd_req = ceph_osdc_alloc_request(osdc, snapc, num_ops, false,
1921 return NULL; /* ENOMEM */
1923 if (op_type == OBJ_OP_WRITE || op_type == OBJ_OP_DISCARD)
1924 osd_req->r_flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK;
1926 osd_req->r_flags = CEPH_OSD_FLAG_READ;
1928 osd_req->r_callback = rbd_osd_req_callback;
1929 osd_req->r_priv = obj_request;
1931 osd_req->r_base_oloc.pool = ceph_file_layout_pg_pool(rbd_dev->layout);
1932 ceph_oid_set_name(&osd_req->r_base_oid, obj_request->object_name);
1938 * Create a copyup osd request based on the information in the object
1939 * request supplied. A copyup request has two or three osd ops, a
1940 * copyup method call, potentially a hint op, and a write or truncate
1943 static struct ceph_osd_request *
1944 rbd_osd_req_create_copyup(struct rbd_obj_request *obj_request)
1946 struct rbd_img_request *img_request;
1947 struct ceph_snap_context *snapc;
1948 struct rbd_device *rbd_dev;
1949 struct ceph_osd_client *osdc;
1950 struct ceph_osd_request *osd_req;
1951 int num_osd_ops = 3;
1953 rbd_assert(obj_request_img_data_test(obj_request));
1954 img_request = obj_request->img_request;
1955 rbd_assert(img_request);
1956 rbd_assert(img_request_write_test(img_request) ||
1957 img_request_discard_test(img_request));
1959 if (img_request_discard_test(img_request))
1962 /* Allocate and initialize the request, for all the ops */
1964 snapc = img_request->snapc;
1965 rbd_dev = img_request->rbd_dev;
1966 osdc = &rbd_dev->rbd_client->client->osdc;
1967 osd_req = ceph_osdc_alloc_request(osdc, snapc, num_osd_ops,
1970 return NULL; /* ENOMEM */
1972 osd_req->r_flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK;
1973 osd_req->r_callback = rbd_osd_req_callback;
1974 osd_req->r_priv = obj_request;
1976 osd_req->r_base_oloc.pool = ceph_file_layout_pg_pool(rbd_dev->layout);
1977 ceph_oid_set_name(&osd_req->r_base_oid, obj_request->object_name);
1983 static void rbd_osd_req_destroy(struct ceph_osd_request *osd_req)
1985 ceph_osdc_put_request(osd_req);
1988 /* object_name is assumed to be a non-null pointer and NUL-terminated */
1990 static struct rbd_obj_request *rbd_obj_request_create(const char *object_name,
1991 u64 offset, u64 length,
1992 enum obj_request_type type)
1994 struct rbd_obj_request *obj_request;
1998 rbd_assert(obj_request_type_valid(type));
2000 size = strlen(object_name) + 1;
2001 name = kmalloc(size, GFP_KERNEL);
2005 obj_request = kmem_cache_zalloc(rbd_obj_request_cache, GFP_KERNEL);
2011 obj_request->object_name = memcpy(name, object_name, size);
2012 obj_request->offset = offset;
2013 obj_request->length = length;
2014 obj_request->flags = 0;
2015 obj_request->which = BAD_WHICH;
2016 obj_request->type = type;
2017 INIT_LIST_HEAD(&obj_request->links);
2018 init_completion(&obj_request->completion);
2019 kref_init(&obj_request->kref);
2021 dout("%s: \"%s\" %llu/%llu %d -> obj %p\n", __func__, object_name,
2022 offset, length, (int)type, obj_request);
2027 static void rbd_obj_request_destroy(struct kref *kref)
2029 struct rbd_obj_request *obj_request;
2031 obj_request = container_of(kref, struct rbd_obj_request, kref);
2033 dout("%s: obj %p\n", __func__, obj_request);
2035 rbd_assert(obj_request->img_request == NULL);
2036 rbd_assert(obj_request->which == BAD_WHICH);
2038 if (obj_request->osd_req)
2039 rbd_osd_req_destroy(obj_request->osd_req);
2041 rbd_assert(obj_request_type_valid(obj_request->type));
2042 switch (obj_request->type) {
2043 case OBJ_REQUEST_NODATA:
2044 break; /* Nothing to do */
2045 case OBJ_REQUEST_BIO:
2046 if (obj_request->bio_list)
2047 bio_chain_put(obj_request->bio_list);
2049 case OBJ_REQUEST_PAGES:
2050 if (obj_request->pages)
2051 ceph_release_page_vector(obj_request->pages,
2052 obj_request->page_count);
2056 kfree(obj_request->object_name);
2057 obj_request->object_name = NULL;
2058 kmem_cache_free(rbd_obj_request_cache, obj_request);
2061 /* It's OK to call this for a device with no parent */
2063 static void rbd_spec_put(struct rbd_spec *spec);
2064 static void rbd_dev_unparent(struct rbd_device *rbd_dev)
2066 rbd_dev_remove_parent(rbd_dev);
2067 rbd_spec_put(rbd_dev->parent_spec);
2068 rbd_dev->parent_spec = NULL;
2069 rbd_dev->parent_overlap = 0;
2073 * Parent image reference counting is used to determine when an
2074 * image's parent fields can be safely torn down--after there are no
2075 * more in-flight requests to the parent image. When the last
2076 * reference is dropped, cleaning them up is safe.
2078 static void rbd_dev_parent_put(struct rbd_device *rbd_dev)
2082 if (!rbd_dev->parent_spec)
2085 counter = atomic_dec_return_safe(&rbd_dev->parent_ref);
2089 /* Last reference; clean up parent data structures */
2092 rbd_dev_unparent(rbd_dev);
2094 rbd_warn(rbd_dev, "parent reference underflow");
2098 * If an image has a non-zero parent overlap, get a reference to its
2101 * We must get the reference before checking for the overlap to
2102 * coordinate properly with zeroing the parent overlap in
2103 * rbd_dev_v2_parent_info() when an image gets flattened. We
2104 * drop it again if there is no overlap.
2106 * Returns true if the rbd device has a parent with a non-zero
2107 * overlap and a reference for it was successfully taken, or
2110 static bool rbd_dev_parent_get(struct rbd_device *rbd_dev)
2114 if (!rbd_dev->parent_spec)
2117 counter = atomic_inc_return_safe(&rbd_dev->parent_ref);
2118 if (counter > 0 && rbd_dev->parent_overlap)
2121 /* Image was flattened, but parent is not yet torn down */
2124 rbd_warn(rbd_dev, "parent reference overflow");
2130 * Caller is responsible for filling in the list of object requests
2131 * that comprises the image request, and the Linux request pointer
2132 * (if there is one).
2134 static struct rbd_img_request *rbd_img_request_create(
2135 struct rbd_device *rbd_dev,
2136 u64 offset, u64 length,
2137 enum obj_operation_type op_type,
2138 struct ceph_snap_context *snapc)
2140 struct rbd_img_request *img_request;
2142 img_request = kmem_cache_alloc(rbd_img_request_cache, GFP_NOIO);
2146 img_request->rq = NULL;
2147 img_request->rbd_dev = rbd_dev;
2148 img_request->offset = offset;
2149 img_request->length = length;
2150 img_request->flags = 0;
2151 if (op_type == OBJ_OP_DISCARD) {
2152 img_request_discard_set(img_request);
2153 img_request->snapc = snapc;
2154 } else if (op_type == OBJ_OP_WRITE) {
2155 img_request_write_set(img_request);
2156 img_request->snapc = snapc;
2158 img_request->snap_id = rbd_dev->spec->snap_id;
2160 if (rbd_dev_parent_get(rbd_dev))
2161 img_request_layered_set(img_request);
2162 spin_lock_init(&img_request->completion_lock);
2163 img_request->next_completion = 0;
2164 img_request->callback = NULL;
2165 img_request->result = 0;
2166 img_request->obj_request_count = 0;
2167 INIT_LIST_HEAD(&img_request->obj_requests);
2168 kref_init(&img_request->kref);
2170 dout("%s: rbd_dev %p %s %llu/%llu -> img %p\n", __func__, rbd_dev,
2171 obj_op_name(op_type), offset, length, img_request);
2176 static void rbd_img_request_destroy(struct kref *kref)
2178 struct rbd_img_request *img_request;
2179 struct rbd_obj_request *obj_request;
2180 struct rbd_obj_request *next_obj_request;
2182 img_request = container_of(kref, struct rbd_img_request, kref);
2184 dout("%s: img %p\n", __func__, img_request);
2186 for_each_obj_request_safe(img_request, obj_request, next_obj_request)
2187 rbd_img_obj_request_del(img_request, obj_request);
2188 rbd_assert(img_request->obj_request_count == 0);
2190 if (img_request_layered_test(img_request)) {
2191 img_request_layered_clear(img_request);
2192 rbd_dev_parent_put(img_request->rbd_dev);
2195 if (img_request_write_test(img_request) ||
2196 img_request_discard_test(img_request))
2197 ceph_put_snap_context(img_request->snapc);
2199 kmem_cache_free(rbd_img_request_cache, img_request);
2202 static struct rbd_img_request *rbd_parent_request_create(
2203 struct rbd_obj_request *obj_request,
2204 u64 img_offset, u64 length)
2206 struct rbd_img_request *parent_request;
2207 struct rbd_device *rbd_dev;
2209 rbd_assert(obj_request->img_request);
2210 rbd_dev = obj_request->img_request->rbd_dev;
2212 parent_request = rbd_img_request_create(rbd_dev->parent, img_offset,
2213 length, OBJ_OP_READ, NULL);
2214 if (!parent_request)
2217 img_request_child_set(parent_request);
2218 rbd_obj_request_get(obj_request);
2219 parent_request->obj_request = obj_request;
2221 return parent_request;
2224 static void rbd_parent_request_destroy(struct kref *kref)
2226 struct rbd_img_request *parent_request;
2227 struct rbd_obj_request *orig_request;
2229 parent_request = container_of(kref, struct rbd_img_request, kref);
2230 orig_request = parent_request->obj_request;
2232 parent_request->obj_request = NULL;
2233 rbd_obj_request_put(orig_request);
2234 img_request_child_clear(parent_request);
2236 rbd_img_request_destroy(kref);
2239 static bool rbd_img_obj_end_request(struct rbd_obj_request *obj_request)
2241 struct rbd_img_request *img_request;
2242 unsigned int xferred;
2246 rbd_assert(obj_request_img_data_test(obj_request));
2247 img_request = obj_request->img_request;
2249 rbd_assert(obj_request->xferred <= (u64)UINT_MAX);
2250 xferred = (unsigned int)obj_request->xferred;
2251 result = obj_request->result;
2253 struct rbd_device *rbd_dev = img_request->rbd_dev;
2254 enum obj_operation_type op_type;
2256 if (img_request_discard_test(img_request))
2257 op_type = OBJ_OP_DISCARD;
2258 else if (img_request_write_test(img_request))
2259 op_type = OBJ_OP_WRITE;
2261 op_type = OBJ_OP_READ;
2263 rbd_warn(rbd_dev, "%s %llx at %llx (%llx)",
2264 obj_op_name(op_type), obj_request->length,
2265 obj_request->img_offset, obj_request->offset);
2266 rbd_warn(rbd_dev, " result %d xferred %x",
2268 if (!img_request->result)
2269 img_request->result = result;
2272 /* Image object requests don't own their page array */
2274 if (obj_request->type == OBJ_REQUEST_PAGES) {
2275 obj_request->pages = NULL;
2276 obj_request->page_count = 0;
2279 if (img_request_child_test(img_request)) {
2280 rbd_assert(img_request->obj_request != NULL);
2281 more = obj_request->which < img_request->obj_request_count - 1;
2283 rbd_assert(img_request->rq != NULL);
2284 more = blk_end_request(img_request->rq, result, xferred);
2290 static void rbd_img_obj_callback(struct rbd_obj_request *obj_request)
2292 struct rbd_img_request *img_request;
2293 u32 which = obj_request->which;
2296 rbd_assert(obj_request_img_data_test(obj_request));
2297 img_request = obj_request->img_request;
2299 dout("%s: img %p obj %p\n", __func__, img_request, obj_request);
2300 rbd_assert(img_request != NULL);
2301 rbd_assert(img_request->obj_request_count > 0);
2302 rbd_assert(which != BAD_WHICH);
2303 rbd_assert(which < img_request->obj_request_count);
2305 spin_lock_irq(&img_request->completion_lock);
2306 if (which != img_request->next_completion)
2309 for_each_obj_request_from(img_request, obj_request) {
2311 rbd_assert(which < img_request->obj_request_count);
2313 if (!obj_request_done_test(obj_request))
2315 more = rbd_img_obj_end_request(obj_request);
2319 rbd_assert(more ^ (which == img_request->obj_request_count));
2320 img_request->next_completion = which;
2322 spin_unlock_irq(&img_request->completion_lock);
2323 rbd_img_request_put(img_request);
2326 rbd_img_request_complete(img_request);
2330 * Add individual osd ops to the given ceph_osd_request and prepare
2331 * them for submission. num_ops is the current number of
2332 * osd operations already to the object request.
2334 static void rbd_img_obj_request_fill(struct rbd_obj_request *obj_request,
2335 struct ceph_osd_request *osd_request,
2336 enum obj_operation_type op_type,
2337 unsigned int num_ops)
2339 struct rbd_img_request *img_request = obj_request->img_request;
2340 struct rbd_device *rbd_dev = img_request->rbd_dev;
2341 u64 object_size = rbd_obj_bytes(&rbd_dev->header);
2342 u64 offset = obj_request->offset;
2343 u64 length = obj_request->length;
2347 if (op_type == OBJ_OP_DISCARD) {
2348 if (!offset && length == object_size &&
2349 (!img_request_layered_test(img_request) ||
2350 !obj_request_overlaps_parent(obj_request))) {
2351 opcode = CEPH_OSD_OP_DELETE;
2352 } else if ((offset + length == object_size)) {
2353 opcode = CEPH_OSD_OP_TRUNCATE;
2355 down_read(&rbd_dev->header_rwsem);
2356 img_end = rbd_dev->header.image_size;
2357 up_read(&rbd_dev->header_rwsem);
2359 if (obj_request->img_offset + length == img_end)
2360 opcode = CEPH_OSD_OP_TRUNCATE;
2362 opcode = CEPH_OSD_OP_ZERO;
2364 } else if (op_type == OBJ_OP_WRITE) {
2365 opcode = CEPH_OSD_OP_WRITE;
2366 osd_req_op_alloc_hint_init(osd_request, num_ops,
2367 object_size, object_size);
2370 opcode = CEPH_OSD_OP_READ;
2373 osd_req_op_extent_init(osd_request, num_ops, opcode, offset, length,
2375 if (obj_request->type == OBJ_REQUEST_BIO)
2376 osd_req_op_extent_osd_data_bio(osd_request, num_ops,
2377 obj_request->bio_list, length);
2378 else if (obj_request->type == OBJ_REQUEST_PAGES)
2379 osd_req_op_extent_osd_data_pages(osd_request, num_ops,
2380 obj_request->pages, length,
2381 offset & ~PAGE_MASK, false, false);
2383 /* Discards are also writes */
2384 if (op_type == OBJ_OP_WRITE || op_type == OBJ_OP_DISCARD)
2385 rbd_osd_req_format_write(obj_request);
2387 rbd_osd_req_format_read(obj_request);
2391 * Split up an image request into one or more object requests, each
2392 * to a different object. The "type" parameter indicates whether
2393 * "data_desc" is the pointer to the head of a list of bio
2394 * structures, or the base of a page array. In either case this
2395 * function assumes data_desc describes memory sufficient to hold
2396 * all data described by the image request.
2398 static int rbd_img_request_fill(struct rbd_img_request *img_request,
2399 enum obj_request_type type,
2402 struct rbd_device *rbd_dev = img_request->rbd_dev;
2403 struct rbd_obj_request *obj_request = NULL;
2404 struct rbd_obj_request *next_obj_request;
2405 struct bio *bio_list = NULL;
2406 unsigned int bio_offset = 0;
2407 struct page **pages = NULL;
2408 enum obj_operation_type op_type;
2412 dout("%s: img %p type %d data_desc %p\n", __func__, img_request,
2413 (int)type, data_desc);
2415 img_offset = img_request->offset;
2416 resid = img_request->length;
2417 rbd_assert(resid > 0);
2418 op_type = rbd_img_request_op_type(img_request);
2420 if (type == OBJ_REQUEST_BIO) {
2421 bio_list = data_desc;
2422 rbd_assert(img_offset ==
2423 bio_list->bi_iter.bi_sector << SECTOR_SHIFT);
2424 } else if (type == OBJ_REQUEST_PAGES) {
2429 struct ceph_osd_request *osd_req;
2430 const char *object_name;
2434 object_name = rbd_segment_name(rbd_dev, img_offset);
2437 offset = rbd_segment_offset(rbd_dev, img_offset);
2438 length = rbd_segment_length(rbd_dev, img_offset, resid);
2439 obj_request = rbd_obj_request_create(object_name,
2440 offset, length, type);
2441 /* object request has its own copy of the object name */
2442 rbd_segment_name_free(object_name);
2447 * set obj_request->img_request before creating the
2448 * osd_request so that it gets the right snapc
2450 rbd_img_obj_request_add(img_request, obj_request);
2452 if (type == OBJ_REQUEST_BIO) {
2453 unsigned int clone_size;
2455 rbd_assert(length <= (u64)UINT_MAX);
2456 clone_size = (unsigned int)length;
2457 obj_request->bio_list =
2458 bio_chain_clone_range(&bio_list,
2462 if (!obj_request->bio_list)
2464 } else if (type == OBJ_REQUEST_PAGES) {
2465 unsigned int page_count;
2467 obj_request->pages = pages;
2468 page_count = (u32)calc_pages_for(offset, length);
2469 obj_request->page_count = page_count;
2470 if ((offset + length) & ~PAGE_MASK)
2471 page_count--; /* more on last page */
2472 pages += page_count;
2475 osd_req = rbd_osd_req_create(rbd_dev, op_type,
2476 (op_type == OBJ_OP_WRITE) ? 2 : 1,
2481 obj_request->osd_req = osd_req;
2482 obj_request->callback = rbd_img_obj_callback;
2483 obj_request->img_offset = img_offset;
2485 rbd_img_obj_request_fill(obj_request, osd_req, op_type, 0);
2487 rbd_img_request_get(img_request);
2489 img_offset += length;
2496 for_each_obj_request_safe(img_request, obj_request, next_obj_request)
2497 rbd_img_obj_request_del(img_request, obj_request);
2503 rbd_img_obj_copyup_callback(struct rbd_obj_request *obj_request)
2505 struct rbd_img_request *img_request;
2506 struct rbd_device *rbd_dev;
2507 struct page **pages;
2510 rbd_assert(obj_request->type == OBJ_REQUEST_BIO ||
2511 obj_request->type == OBJ_REQUEST_NODATA);
2512 rbd_assert(obj_request_img_data_test(obj_request));
2513 img_request = obj_request->img_request;
2514 rbd_assert(img_request);
2516 rbd_dev = img_request->rbd_dev;
2517 rbd_assert(rbd_dev);
2519 pages = obj_request->copyup_pages;
2520 rbd_assert(pages != NULL);
2521 obj_request->copyup_pages = NULL;
2522 page_count = obj_request->copyup_page_count;
2523 rbd_assert(page_count);
2524 obj_request->copyup_page_count = 0;
2525 ceph_release_page_vector(pages, page_count);
2528 * We want the transfer count to reflect the size of the
2529 * original write request. There is no such thing as a
2530 * successful short write, so if the request was successful
2531 * we can just set it to the originally-requested length.
2533 if (!obj_request->result)
2534 obj_request->xferred = obj_request->length;
2536 /* Finish up with the normal image object callback */
2538 rbd_img_obj_callback(obj_request);
2542 rbd_img_obj_parent_read_full_callback(struct rbd_img_request *img_request)
2544 struct rbd_obj_request *orig_request;
2545 struct ceph_osd_request *osd_req;
2546 struct ceph_osd_client *osdc;
2547 struct rbd_device *rbd_dev;
2548 struct page **pages;
2549 enum obj_operation_type op_type;
2554 rbd_assert(img_request_child_test(img_request));
2556 /* First get what we need from the image request */
2558 pages = img_request->copyup_pages;
2559 rbd_assert(pages != NULL);
2560 img_request->copyup_pages = NULL;
2561 page_count = img_request->copyup_page_count;
2562 rbd_assert(page_count);
2563 img_request->copyup_page_count = 0;
2565 orig_request = img_request->obj_request;
2566 rbd_assert(orig_request != NULL);
2567 rbd_assert(obj_request_type_valid(orig_request->type));
2568 img_result = img_request->result;
2569 parent_length = img_request->length;
2570 rbd_assert(parent_length == img_request->xferred);
2571 rbd_img_request_put(img_request);
2573 rbd_assert(orig_request->img_request);
2574 rbd_dev = orig_request->img_request->rbd_dev;
2575 rbd_assert(rbd_dev);
2578 * If the overlap has become 0 (most likely because the
2579 * image has been flattened) we need to free the pages
2580 * and re-submit the original write request.
2582 if (!rbd_dev->parent_overlap) {
2583 struct ceph_osd_client *osdc;
2585 ceph_release_page_vector(pages, page_count);
2586 osdc = &rbd_dev->rbd_client->client->osdc;
2587 img_result = rbd_obj_request_submit(osdc, orig_request);
2596 * The original osd request is of no use to use any more.
2597 * We need a new one that can hold the three ops in a copyup
2598 * request. Allocate the new copyup osd request for the
2599 * original request, and release the old one.
2601 img_result = -ENOMEM;
2602 osd_req = rbd_osd_req_create_copyup(orig_request);
2605 rbd_osd_req_destroy(orig_request->osd_req);
2606 orig_request->osd_req = osd_req;
2607 orig_request->copyup_pages = pages;
2608 orig_request->copyup_page_count = page_count;
2610 /* Initialize the copyup op */
2612 osd_req_op_cls_init(osd_req, 0, CEPH_OSD_OP_CALL, "rbd", "copyup");
2613 osd_req_op_cls_request_data_pages(osd_req, 0, pages, parent_length, 0,
2616 /* Add the other op(s) */
2618 op_type = rbd_img_request_op_type(orig_request->img_request);
2619 rbd_img_obj_request_fill(orig_request, osd_req, op_type, 1);
2621 /* All set, send it off. */
2623 orig_request->callback = rbd_img_obj_copyup_callback;
2624 osdc = &rbd_dev->rbd_client->client->osdc;
2625 img_result = rbd_obj_request_submit(osdc, orig_request);
2629 /* Record the error code and complete the request */
2631 orig_request->result = img_result;
2632 orig_request->xferred = 0;
2633 obj_request_done_set(orig_request);
2634 rbd_obj_request_complete(orig_request);
2638 * Read from the parent image the range of data that covers the
2639 * entire target of the given object request. This is used for
2640 * satisfying a layered image write request when the target of an
2641 * object request from the image request does not exist.
2643 * A page array big enough to hold the returned data is allocated
2644 * and supplied to rbd_img_request_fill() as the "data descriptor."
2645 * When the read completes, this page array will be transferred to
2646 * the original object request for the copyup operation.
2648 * If an error occurs, record it as the result of the original
2649 * object request and mark it done so it gets completed.
2651 static int rbd_img_obj_parent_read_full(struct rbd_obj_request *obj_request)
2653 struct rbd_img_request *img_request = NULL;
2654 struct rbd_img_request *parent_request = NULL;
2655 struct rbd_device *rbd_dev;
2658 struct page **pages = NULL;
2662 rbd_assert(obj_request_img_data_test(obj_request));
2663 rbd_assert(obj_request_type_valid(obj_request->type));
2665 img_request = obj_request->img_request;
2666 rbd_assert(img_request != NULL);
2667 rbd_dev = img_request->rbd_dev;
2668 rbd_assert(rbd_dev->parent != NULL);
2671 * Determine the byte range covered by the object in the
2672 * child image to which the original request was to be sent.
2674 img_offset = obj_request->img_offset - obj_request->offset;
2675 length = (u64)1 << rbd_dev->header.obj_order;
2678 * There is no defined parent data beyond the parent
2679 * overlap, so limit what we read at that boundary if
2682 if (img_offset + length > rbd_dev->parent_overlap) {
2683 rbd_assert(img_offset < rbd_dev->parent_overlap);
2684 length = rbd_dev->parent_overlap - img_offset;
2688 * Allocate a page array big enough to receive the data read
2691 page_count = (u32)calc_pages_for(0, length);
2692 pages = ceph_alloc_page_vector(page_count, GFP_KERNEL);
2693 if (IS_ERR(pages)) {
2694 result = PTR_ERR(pages);
2700 parent_request = rbd_parent_request_create(obj_request,
2701 img_offset, length);
2702 if (!parent_request)
2705 result = rbd_img_request_fill(parent_request, OBJ_REQUEST_PAGES, pages);
2708 parent_request->copyup_pages = pages;
2709 parent_request->copyup_page_count = page_count;
2711 parent_request->callback = rbd_img_obj_parent_read_full_callback;
2712 result = rbd_img_request_submit(parent_request);
2716 parent_request->copyup_pages = NULL;
2717 parent_request->copyup_page_count = 0;
2718 parent_request->obj_request = NULL;
2719 rbd_obj_request_put(obj_request);
2722 ceph_release_page_vector(pages, page_count);
2724 rbd_img_request_put(parent_request);
2725 obj_request->result = result;
2726 obj_request->xferred = 0;
2727 obj_request_done_set(obj_request);
2732 static void rbd_img_obj_exists_callback(struct rbd_obj_request *obj_request)
2734 struct rbd_obj_request *orig_request;
2735 struct rbd_device *rbd_dev;
2738 rbd_assert(!obj_request_img_data_test(obj_request));
2741 * All we need from the object request is the original
2742 * request and the result of the STAT op. Grab those, then
2743 * we're done with the request.
2745 orig_request = obj_request->obj_request;
2746 obj_request->obj_request = NULL;
2747 rbd_obj_request_put(orig_request);
2748 rbd_assert(orig_request);
2749 rbd_assert(orig_request->img_request);
2751 result = obj_request->result;
2752 obj_request->result = 0;
2754 dout("%s: obj %p for obj %p result %d %llu/%llu\n", __func__,
2755 obj_request, orig_request, result,
2756 obj_request->xferred, obj_request->length);
2757 rbd_obj_request_put(obj_request);
2760 * If the overlap has become 0 (most likely because the
2761 * image has been flattened) we need to free the pages
2762 * and re-submit the original write request.
2764 rbd_dev = orig_request->img_request->rbd_dev;
2765 if (!rbd_dev->parent_overlap) {
2766 struct ceph_osd_client *osdc;
2768 osdc = &rbd_dev->rbd_client->client->osdc;
2769 result = rbd_obj_request_submit(osdc, orig_request);
2775 * Our only purpose here is to determine whether the object
2776 * exists, and we don't want to treat the non-existence as
2777 * an error. If something else comes back, transfer the
2778 * error to the original request and complete it now.
2781 obj_request_existence_set(orig_request, true);
2782 } else if (result == -ENOENT) {
2783 obj_request_existence_set(orig_request, false);
2784 } else if (result) {
2785 orig_request->result = result;
2790 * Resubmit the original request now that we have recorded
2791 * whether the target object exists.
2793 orig_request->result = rbd_img_obj_request_submit(orig_request);
2795 if (orig_request->result)
2796 rbd_obj_request_complete(orig_request);
2799 static int rbd_img_obj_exists_submit(struct rbd_obj_request *obj_request)
2801 struct rbd_obj_request *stat_request;
2802 struct rbd_device *rbd_dev;
2803 struct ceph_osd_client *osdc;
2804 struct page **pages = NULL;
2810 * The response data for a STAT call consists of:
2817 size = sizeof (__le64) + sizeof (__le32) + sizeof (__le32);
2818 page_count = (u32)calc_pages_for(0, size);
2819 pages = ceph_alloc_page_vector(page_count, GFP_KERNEL);
2821 return PTR_ERR(pages);
2824 stat_request = rbd_obj_request_create(obj_request->object_name, 0, 0,
2829 rbd_obj_request_get(obj_request);
2830 stat_request->obj_request = obj_request;
2831 stat_request->pages = pages;
2832 stat_request->page_count = page_count;
2834 rbd_assert(obj_request->img_request);
2835 rbd_dev = obj_request->img_request->rbd_dev;
2836 stat_request->osd_req = rbd_osd_req_create(rbd_dev, OBJ_OP_READ, 1,
2838 if (!stat_request->osd_req)
2840 stat_request->callback = rbd_img_obj_exists_callback;
2842 osd_req_op_init(stat_request->osd_req, 0, CEPH_OSD_OP_STAT);
2843 osd_req_op_raw_data_in_pages(stat_request->osd_req, 0, pages, size, 0,
2845 rbd_osd_req_format_read(stat_request);
2847 osdc = &rbd_dev->rbd_client->client->osdc;
2848 ret = rbd_obj_request_submit(osdc, stat_request);
2851 rbd_obj_request_put(obj_request);
2856 static bool img_obj_request_simple(struct rbd_obj_request *obj_request)
2858 struct rbd_img_request *img_request;
2859 struct rbd_device *rbd_dev;
2861 rbd_assert(obj_request_img_data_test(obj_request));
2863 img_request = obj_request->img_request;
2864 rbd_assert(img_request);
2865 rbd_dev = img_request->rbd_dev;
2868 if (!img_request_write_test(img_request) &&
2869 !img_request_discard_test(img_request))
2872 /* Non-layered writes */
2873 if (!img_request_layered_test(img_request))
2877 * Layered writes outside of the parent overlap range don't
2878 * share any data with the parent.
2880 if (!obj_request_overlaps_parent(obj_request))
2884 * Entire-object layered writes - we will overwrite whatever
2885 * parent data there is anyway.
2887 if (!obj_request->offset &&
2888 obj_request->length == rbd_obj_bytes(&rbd_dev->header))
2892 * If the object is known to already exist, its parent data has
2893 * already been copied.
2895 if (obj_request_known_test(obj_request) &&
2896 obj_request_exists_test(obj_request))
2902 static int rbd_img_obj_request_submit(struct rbd_obj_request *obj_request)
2904 if (img_obj_request_simple(obj_request)) {
2905 struct rbd_device *rbd_dev;
2906 struct ceph_osd_client *osdc;
2908 rbd_dev = obj_request->img_request->rbd_dev;
2909 osdc = &rbd_dev->rbd_client->client->osdc;
2911 return rbd_obj_request_submit(osdc, obj_request);
2915 * It's a layered write. The target object might exist but
2916 * we may not know that yet. If we know it doesn't exist,
2917 * start by reading the data for the full target object from
2918 * the parent so we can use it for a copyup to the target.
2920 if (obj_request_known_test(obj_request))
2921 return rbd_img_obj_parent_read_full(obj_request);
2923 /* We don't know whether the target exists. Go find out. */
2925 return rbd_img_obj_exists_submit(obj_request);
2928 static int rbd_img_request_submit(struct rbd_img_request *img_request)
2930 struct rbd_obj_request *obj_request;
2931 struct rbd_obj_request *next_obj_request;
2933 dout("%s: img %p\n", __func__, img_request);
2934 for_each_obj_request_safe(img_request, obj_request, next_obj_request) {
2937 ret = rbd_img_obj_request_submit(obj_request);
2945 static void rbd_img_parent_read_callback(struct rbd_img_request *img_request)
2947 struct rbd_obj_request *obj_request;
2948 struct rbd_device *rbd_dev;
2953 rbd_assert(img_request_child_test(img_request));
2955 /* First get what we need from the image request and release it */
2957 obj_request = img_request->obj_request;
2958 img_xferred = img_request->xferred;
2959 img_result = img_request->result;
2960 rbd_img_request_put(img_request);
2963 * If the overlap has become 0 (most likely because the
2964 * image has been flattened) we need to re-submit the
2967 rbd_assert(obj_request);
2968 rbd_assert(obj_request->img_request);
2969 rbd_dev = obj_request->img_request->rbd_dev;
2970 if (!rbd_dev->parent_overlap) {
2971 struct ceph_osd_client *osdc;
2973 osdc = &rbd_dev->rbd_client->client->osdc;
2974 img_result = rbd_obj_request_submit(osdc, obj_request);
2979 obj_request->result = img_result;
2980 if (obj_request->result)
2984 * We need to zero anything beyond the parent overlap
2985 * boundary. Since rbd_img_obj_request_read_callback()
2986 * will zero anything beyond the end of a short read, an
2987 * easy way to do this is to pretend the data from the
2988 * parent came up short--ending at the overlap boundary.
2990 rbd_assert(obj_request->img_offset < U64_MAX - obj_request->length);
2991 obj_end = obj_request->img_offset + obj_request->length;
2992 if (obj_end > rbd_dev->parent_overlap) {
2995 if (obj_request->img_offset < rbd_dev->parent_overlap)
2996 xferred = rbd_dev->parent_overlap -
2997 obj_request->img_offset;
2999 obj_request->xferred = min(img_xferred, xferred);
3001 obj_request->xferred = img_xferred;
3004 rbd_img_obj_request_read_callback(obj_request);
3005 rbd_obj_request_complete(obj_request);
3008 static void rbd_img_parent_read(struct rbd_obj_request *obj_request)
3010 struct rbd_img_request *img_request;
3013 rbd_assert(obj_request_img_data_test(obj_request));
3014 rbd_assert(obj_request->img_request != NULL);
3015 rbd_assert(obj_request->result == (s32) -ENOENT);
3016 rbd_assert(obj_request_type_valid(obj_request->type));
3018 /* rbd_read_finish(obj_request, obj_request->length); */
3019 img_request = rbd_parent_request_create(obj_request,
3020 obj_request->img_offset,
3021 obj_request->length);
3026 if (obj_request->type == OBJ_REQUEST_BIO)
3027 result = rbd_img_request_fill(img_request, OBJ_REQUEST_BIO,
3028 obj_request->bio_list);
3030 result = rbd_img_request_fill(img_request, OBJ_REQUEST_PAGES,
3031 obj_request->pages);
3035 img_request->callback = rbd_img_parent_read_callback;
3036 result = rbd_img_request_submit(img_request);
3043 rbd_img_request_put(img_request);
3044 obj_request->result = result;
3045 obj_request->xferred = 0;
3046 obj_request_done_set(obj_request);
3049 static int rbd_obj_notify_ack_sync(struct rbd_device *rbd_dev, u64 notify_id)
3051 struct rbd_obj_request *obj_request;
3052 struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
3055 obj_request = rbd_obj_request_create(rbd_dev->header_name, 0, 0,
3056 OBJ_REQUEST_NODATA);
3061 obj_request->osd_req = rbd_osd_req_create(rbd_dev, OBJ_OP_READ, 1,
3063 if (!obj_request->osd_req)
3066 osd_req_op_watch_init(obj_request->osd_req, 0, CEPH_OSD_OP_NOTIFY_ACK,
3068 rbd_osd_req_format_read(obj_request);
3070 ret = rbd_obj_request_submit(osdc, obj_request);
3073 ret = rbd_obj_request_wait(obj_request);
3075 rbd_obj_request_put(obj_request);
3080 static void rbd_watch_cb(u64 ver, u64 notify_id, u8 opcode, void *data)
3082 struct rbd_device *rbd_dev = (struct rbd_device *)data;
3088 dout("%s: \"%s\" notify_id %llu opcode %u\n", __func__,
3089 rbd_dev->header_name, (unsigned long long)notify_id,
3090 (unsigned int)opcode);
3093 * Until adequate refresh error handling is in place, there is
3094 * not much we can do here, except warn.
3096 * See http://tracker.ceph.com/issues/5040
3098 ret = rbd_dev_refresh(rbd_dev);
3100 rbd_warn(rbd_dev, "refresh failed: %d", ret);
3102 ret = rbd_obj_notify_ack_sync(rbd_dev, notify_id);
3104 rbd_warn(rbd_dev, "notify_ack ret %d", ret);
3108 * Send a (un)watch request and wait for the ack. Return a request
3109 * with a ref held on success or error.
3111 static struct rbd_obj_request *rbd_obj_watch_request_helper(
3112 struct rbd_device *rbd_dev,
3115 struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
3116 struct rbd_obj_request *obj_request;
3119 obj_request = rbd_obj_request_create(rbd_dev->header_name, 0, 0,
3120 OBJ_REQUEST_NODATA);
3122 return ERR_PTR(-ENOMEM);
3124 obj_request->osd_req = rbd_osd_req_create(rbd_dev, OBJ_OP_WRITE, 1,
3126 if (!obj_request->osd_req) {
3131 osd_req_op_watch_init(obj_request->osd_req, 0, CEPH_OSD_OP_WATCH,
3132 rbd_dev->watch_event->cookie, 0, watch);
3133 rbd_osd_req_format_write(obj_request);
3136 ceph_osdc_set_request_linger(osdc, obj_request->osd_req);
3138 ret = rbd_obj_request_submit(osdc, obj_request);
3142 ret = rbd_obj_request_wait(obj_request);
3146 ret = obj_request->result;
3149 rbd_obj_request_end(obj_request);
3156 rbd_obj_request_put(obj_request);
3157 return ERR_PTR(ret);
3161 * Initiate a watch request, synchronously.
3163 static int rbd_dev_header_watch_sync(struct rbd_device *rbd_dev)
3165 struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
3166 struct rbd_obj_request *obj_request;
3169 rbd_assert(!rbd_dev->watch_event);
3170 rbd_assert(!rbd_dev->watch_request);
3172 ret = ceph_osdc_create_event(osdc, rbd_watch_cb, rbd_dev,
3173 &rbd_dev->watch_event);
3177 obj_request = rbd_obj_watch_request_helper(rbd_dev, true);
3178 if (IS_ERR(obj_request)) {
3179 ceph_osdc_cancel_event(rbd_dev->watch_event);
3180 rbd_dev->watch_event = NULL;
3181 return PTR_ERR(obj_request);
3185 * A watch request is set to linger, so the underlying osd
3186 * request won't go away until we unregister it. We retain
3187 * a pointer to the object request during that time (in
3188 * rbd_dev->watch_request), so we'll keep a reference to it.
3189 * We'll drop that reference after we've unregistered it in
3190 * rbd_dev_header_unwatch_sync().
3192 rbd_dev->watch_request = obj_request;
3198 * Tear down a watch request, synchronously.
3200 static void rbd_dev_header_unwatch_sync(struct rbd_device *rbd_dev)
3202 struct rbd_obj_request *obj_request;
3204 rbd_assert(rbd_dev->watch_event);
3205 rbd_assert(rbd_dev->watch_request);
3207 rbd_obj_request_end(rbd_dev->watch_request);
3208 rbd_obj_request_put(rbd_dev->watch_request);
3209 rbd_dev->watch_request = NULL;
3211 obj_request = rbd_obj_watch_request_helper(rbd_dev, false);
3212 if (!IS_ERR(obj_request))
3213 rbd_obj_request_put(obj_request);
3215 rbd_warn(rbd_dev, "unable to tear down watch request (%ld)",
3216 PTR_ERR(obj_request));
3218 ceph_osdc_cancel_event(rbd_dev->watch_event);
3219 rbd_dev->watch_event = NULL;
3223 * Synchronous osd object method call. Returns the number of bytes
3224 * returned in the outbound buffer, or a negative error code.
3226 static int rbd_obj_method_sync(struct rbd_device *rbd_dev,
3227 const char *object_name,
3228 const char *class_name,
3229 const char *method_name,
3230 const void *outbound,
3231 size_t outbound_size,
3233 size_t inbound_size)
3235 struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
3236 struct rbd_obj_request *obj_request;
3237 struct page **pages;
3242 * Method calls are ultimately read operations. The result
3243 * should placed into the inbound buffer provided. They
3244 * also supply outbound data--parameters for the object
3245 * method. Currently if this is present it will be a
3248 page_count = (u32)calc_pages_for(0, inbound_size);
3249 pages = ceph_alloc_page_vector(page_count, GFP_KERNEL);
3251 return PTR_ERR(pages);
3254 obj_request = rbd_obj_request_create(object_name, 0, inbound_size,
3259 obj_request->pages = pages;
3260 obj_request->page_count = page_count;
3262 obj_request->osd_req = rbd_osd_req_create(rbd_dev, OBJ_OP_READ, 1,
3264 if (!obj_request->osd_req)
3267 osd_req_op_cls_init(obj_request->osd_req, 0, CEPH_OSD_OP_CALL,
3268 class_name, method_name);
3269 if (outbound_size) {
3270 struct ceph_pagelist *pagelist;
3272 pagelist = kmalloc(sizeof (*pagelist), GFP_NOFS);
3276 ceph_pagelist_init(pagelist);
3277 ceph_pagelist_append(pagelist, outbound, outbound_size);
3278 osd_req_op_cls_request_data_pagelist(obj_request->osd_req, 0,
3281 osd_req_op_cls_response_data_pages(obj_request->osd_req, 0,
3282 obj_request->pages, inbound_size,
3284 rbd_osd_req_format_read(obj_request);
3286 ret = rbd_obj_request_submit(osdc, obj_request);
3289 ret = rbd_obj_request_wait(obj_request);
3293 ret = obj_request->result;
3297 rbd_assert(obj_request->xferred < (u64)INT_MAX);
3298 ret = (int)obj_request->xferred;
3299 ceph_copy_from_page_vector(pages, inbound, 0, obj_request->xferred);
3302 rbd_obj_request_put(obj_request);
3304 ceph_release_page_vector(pages, page_count);
3309 static void rbd_handle_request(struct rbd_device *rbd_dev, struct request *rq)
3311 struct rbd_img_request *img_request;
3312 struct ceph_snap_context *snapc = NULL;
3313 u64 offset = (u64)blk_rq_pos(rq) << SECTOR_SHIFT;
3314 u64 length = blk_rq_bytes(rq);
3315 enum obj_operation_type op_type;
3319 if (rq->cmd_flags & REQ_DISCARD)
3320 op_type = OBJ_OP_DISCARD;
3321 else if (rq->cmd_flags & REQ_WRITE)
3322 op_type = OBJ_OP_WRITE;
3324 op_type = OBJ_OP_READ;
3326 /* Ignore/skip any zero-length requests */
3329 dout("%s: zero-length request\n", __func__);
3334 /* Only reads are allowed to a read-only device */
3336 if (op_type != OBJ_OP_READ) {
3337 if (rbd_dev->mapping.read_only) {
3341 rbd_assert(rbd_dev->spec->snap_id == CEPH_NOSNAP);
3345 * Quit early if the mapped snapshot no longer exists. It's
3346 * still possible the snapshot will have disappeared by the
3347 * time our request arrives at the osd, but there's no sense in
3348 * sending it if we already know.
3350 if (!test_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags)) {
3351 dout("request for non-existent snapshot");
3352 rbd_assert(rbd_dev->spec->snap_id != CEPH_NOSNAP);
3357 if (offset && length > U64_MAX - offset + 1) {
3358 rbd_warn(rbd_dev, "bad request range (%llu~%llu)", offset,
3361 goto err_rq; /* Shouldn't happen */
3364 down_read(&rbd_dev->header_rwsem);
3365 mapping_size = rbd_dev->mapping.size;
3366 if (op_type != OBJ_OP_READ) {
3367 snapc = rbd_dev->header.snapc;
3368 ceph_get_snap_context(snapc);
3370 up_read(&rbd_dev->header_rwsem);
3372 if (offset + length > mapping_size) {
3373 rbd_warn(rbd_dev, "beyond EOD (%llu~%llu > %llu)", offset,
3374 length, mapping_size);
3379 img_request = rbd_img_request_create(rbd_dev, offset, length, op_type,
3385 img_request->rq = rq;
3387 if (op_type == OBJ_OP_DISCARD)
3388 result = rbd_img_request_fill(img_request, OBJ_REQUEST_NODATA,
3391 result = rbd_img_request_fill(img_request, OBJ_REQUEST_BIO,
3394 goto err_img_request;
3396 result = rbd_img_request_submit(img_request);
3398 goto err_img_request;
3403 rbd_img_request_put(img_request);
3406 rbd_warn(rbd_dev, "%s %llx at %llx result %d",
3407 obj_op_name(op_type), length, offset, result);
3409 ceph_put_snap_context(snapc);
3410 blk_end_request_all(rq, result);
3413 static void rbd_request_workfn(struct work_struct *work)
3415 struct rbd_device *rbd_dev =
3416 container_of(work, struct rbd_device, rq_work);
3417 struct request *rq, *next;
3418 LIST_HEAD(requests);
3420 spin_lock_irq(&rbd_dev->lock); /* rq->q->queue_lock */
3421 list_splice_init(&rbd_dev->rq_queue, &requests);
3422 spin_unlock_irq(&rbd_dev->lock);
3424 list_for_each_entry_safe(rq, next, &requests, queuelist) {
3425 list_del_init(&rq->queuelist);
3426 rbd_handle_request(rbd_dev, rq);
3431 * Called with q->queue_lock held and interrupts disabled, possibly on
3432 * the way to schedule(). Do not sleep here!
3434 static void rbd_request_fn(struct request_queue *q)
3436 struct rbd_device *rbd_dev = q->queuedata;
3440 rbd_assert(rbd_dev);
3442 while ((rq = blk_fetch_request(q))) {
3443 /* Ignore any non-FS requests that filter through. */
3444 if (rq->cmd_type != REQ_TYPE_FS) {
3445 dout("%s: non-fs request type %d\n", __func__,
3446 (int) rq->cmd_type);
3447 __blk_end_request_all(rq, 0);
3451 list_add_tail(&rq->queuelist, &rbd_dev->rq_queue);
3456 queue_work(rbd_wq, &rbd_dev->rq_work);
3460 * a queue callback. Makes sure that we don't create a bio that spans across
3461 * multiple osd objects. One exception would be with a single page bios,
3462 * which we handle later at bio_chain_clone_range()
3464 static int rbd_merge_bvec(struct request_queue *q, struct bvec_merge_data *bmd,
3465 struct bio_vec *bvec)
3467 struct rbd_device *rbd_dev = q->queuedata;
3468 sector_t sector_offset;
3469 sector_t sectors_per_obj;
3470 sector_t obj_sector_offset;
3474 * Find how far into its rbd object the partition-relative
3475 * bio start sector is to offset relative to the enclosing
3478 sector_offset = get_start_sect(bmd->bi_bdev) + bmd->bi_sector;
3479 sectors_per_obj = 1 << (rbd_dev->header.obj_order - SECTOR_SHIFT);
3480 obj_sector_offset = sector_offset & (sectors_per_obj - 1);
3483 * Compute the number of bytes from that offset to the end
3484 * of the object. Account for what's already used by the bio.
3486 ret = (int) (sectors_per_obj - obj_sector_offset) << SECTOR_SHIFT;
3487 if (ret > bmd->bi_size)
3488 ret -= bmd->bi_size;
3493 * Don't send back more than was asked for. And if the bio
3494 * was empty, let the whole thing through because: "Note
3495 * that a block device *must* allow a single page to be
3496 * added to an empty bio."
3498 rbd_assert(bvec->bv_len <= PAGE_SIZE);
3499 if (ret > (int) bvec->bv_len || !bmd->bi_size)
3500 ret = (int) bvec->bv_len;
3505 static void rbd_free_disk(struct rbd_device *rbd_dev)
3507 struct gendisk *disk = rbd_dev->disk;
3512 rbd_dev->disk = NULL;
3513 if (disk->flags & GENHD_FL_UP) {
3516 blk_cleanup_queue(disk->queue);
3521 static int rbd_obj_read_sync(struct rbd_device *rbd_dev,
3522 const char *object_name,
3523 u64 offset, u64 length, void *buf)
3526 struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
3527 struct rbd_obj_request *obj_request;
3528 struct page **pages = NULL;
3533 page_count = (u32) calc_pages_for(offset, length);
3534 pages = ceph_alloc_page_vector(page_count, GFP_KERNEL);
3536 return PTR_ERR(pages);
3539 obj_request = rbd_obj_request_create(object_name, offset, length,
3544 obj_request->pages = pages;
3545 obj_request->page_count = page_count;
3547 obj_request->osd_req = rbd_osd_req_create(rbd_dev, OBJ_OP_READ, 1,
3549 if (!obj_request->osd_req)
3552 osd_req_op_extent_init(obj_request->osd_req, 0, CEPH_OSD_OP_READ,
3553 offset, length, 0, 0);
3554 osd_req_op_extent_osd_data_pages(obj_request->osd_req, 0,
3556 obj_request->length,
3557 obj_request->offset & ~PAGE_MASK,
3559 rbd_osd_req_format_read(obj_request);
3561 ret = rbd_obj_request_submit(osdc, obj_request);
3564 ret = rbd_obj_request_wait(obj_request);
3568 ret = obj_request->result;
3572 rbd_assert(obj_request->xferred <= (u64) SIZE_MAX);
3573 size = (size_t) obj_request->xferred;
3574 ceph_copy_from_page_vector(pages, buf, 0, size);
3575 rbd_assert(size <= (size_t)INT_MAX);
3579 rbd_obj_request_put(obj_request);
3581 ceph_release_page_vector(pages, page_count);
3587 * Read the complete header for the given rbd device. On successful
3588 * return, the rbd_dev->header field will contain up-to-date
3589 * information about the image.
3591 static int rbd_dev_v1_header_info(struct rbd_device *rbd_dev)
3593 struct rbd_image_header_ondisk *ondisk = NULL;
3600 * The complete header will include an array of its 64-bit
3601 * snapshot ids, followed by the names of those snapshots as
3602 * a contiguous block of NUL-terminated strings. Note that
3603 * the number of snapshots could change by the time we read
3604 * it in, in which case we re-read it.
3611 size = sizeof (*ondisk);
3612 size += snap_count * sizeof (struct rbd_image_snap_ondisk);
3614 ondisk = kmalloc(size, GFP_KERNEL);
3618 ret = rbd_obj_read_sync(rbd_dev, rbd_dev->header_name,
3622 if ((size_t)ret < size) {
3624 rbd_warn(rbd_dev, "short header read (want %zd got %d)",
3628 if (!rbd_dev_ondisk_valid(ondisk)) {
3630 rbd_warn(rbd_dev, "invalid header");
3634 names_size = le64_to_cpu(ondisk->snap_names_len);
3635 want_count = snap_count;
3636 snap_count = le32_to_cpu(ondisk->snap_count);
3637 } while (snap_count != want_count);
3639 ret = rbd_header_from_disk(rbd_dev, ondisk);
3647 * Clear the rbd device's EXISTS flag if the snapshot it's mapped to
3648 * has disappeared from the (just updated) snapshot context.
3650 static void rbd_exists_validate(struct rbd_device *rbd_dev)
3654 if (!test_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags))
3657 snap_id = rbd_dev->spec->snap_id;
3658 if (snap_id == CEPH_NOSNAP)
3661 if (rbd_dev_snap_index(rbd_dev, snap_id) == BAD_SNAP_INDEX)
3662 clear_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags);
3665 static void rbd_dev_update_size(struct rbd_device *rbd_dev)
3671 * Don't hold the lock while doing disk operations,
3672 * or lock ordering will conflict with the bdev mutex via:
3673 * rbd_add() -> blkdev_get() -> rbd_open()
3675 spin_lock_irq(&rbd_dev->lock);
3676 removing = test_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags);
3677 spin_unlock_irq(&rbd_dev->lock);
3679 * If the device is being removed, rbd_dev->disk has
3680 * been destroyed, so don't try to update its size
3683 size = (sector_t)rbd_dev->mapping.size / SECTOR_SIZE;
3684 dout("setting size to %llu sectors", (unsigned long long)size);
3685 set_capacity(rbd_dev->disk, size);
3686 revalidate_disk(rbd_dev->disk);
3690 static int rbd_dev_refresh(struct rbd_device *rbd_dev)
3695 down_write(&rbd_dev->header_rwsem);
3696 mapping_size = rbd_dev->mapping.size;
3698 ret = rbd_dev_header_info(rbd_dev);
3703 * If there is a parent, see if it has disappeared due to the
3704 * mapped image getting flattened.
3706 if (rbd_dev->parent) {
3707 ret = rbd_dev_v2_parent_info(rbd_dev);
3712 if (rbd_dev->spec->snap_id == CEPH_NOSNAP) {
3713 if (rbd_dev->mapping.size != rbd_dev->header.image_size)
3714 rbd_dev->mapping.size = rbd_dev->header.image_size;
3716 /* validate mapped snapshot's EXISTS flag */
3717 rbd_exists_validate(rbd_dev);
3720 up_write(&rbd_dev->header_rwsem);
3722 if (mapping_size != rbd_dev->mapping.size)
3723 rbd_dev_update_size(rbd_dev);
3728 static int rbd_init_disk(struct rbd_device *rbd_dev)
3730 struct gendisk *disk;
3731 struct request_queue *q;
3734 /* create gendisk info */
3735 disk = alloc_disk(single_major ?
3736 (1 << RBD_SINGLE_MAJOR_PART_SHIFT) :
3737 RBD_MINORS_PER_MAJOR);
3741 snprintf(disk->disk_name, sizeof(disk->disk_name), RBD_DRV_NAME "%d",
3743 disk->major = rbd_dev->major;
3744 disk->first_minor = rbd_dev->minor;
3746 disk->flags |= GENHD_FL_EXT_DEVT;
3747 disk->fops = &rbd_bd_ops;
3748 disk->private_data = rbd_dev;
3750 q = blk_init_queue(rbd_request_fn, &rbd_dev->lock);
3754 /* We use the default size, but let's be explicit about it. */
3755 blk_queue_physical_block_size(q, SECTOR_SIZE);
3757 /* set io sizes to object size */
3758 segment_size = rbd_obj_bytes(&rbd_dev->header);
3759 blk_queue_max_hw_sectors(q, segment_size / SECTOR_SIZE);
3760 blk_queue_max_segment_size(q, segment_size);
3761 blk_queue_io_min(q, segment_size);
3762 blk_queue_io_opt(q, segment_size);
3764 /* enable the discard support */
3765 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
3766 q->limits.discard_granularity = segment_size;
3767 q->limits.discard_alignment = segment_size;
3768 q->limits.max_discard_sectors = segment_size / SECTOR_SIZE;
3769 q->limits.discard_zeroes_data = 1;
3771 blk_queue_merge_bvec(q, rbd_merge_bvec);
3774 q->queuedata = rbd_dev;
3776 rbd_dev->disk = disk;
3789 static struct rbd_device *dev_to_rbd_dev(struct device *dev)
3791 return container_of(dev, struct rbd_device, dev);
3794 static ssize_t rbd_size_show(struct device *dev,
3795 struct device_attribute *attr, char *buf)
3797 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3799 return sprintf(buf, "%llu\n",
3800 (unsigned long long)rbd_dev->mapping.size);
3804 * Note this shows the features for whatever's mapped, which is not
3805 * necessarily the base image.
3807 static ssize_t rbd_features_show(struct device *dev,
3808 struct device_attribute *attr, char *buf)
3810 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3812 return sprintf(buf, "0x%016llx\n",
3813 (unsigned long long)rbd_dev->mapping.features);
3816 static ssize_t rbd_major_show(struct device *dev,
3817 struct device_attribute *attr, char *buf)
3819 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3822 return sprintf(buf, "%d\n", rbd_dev->major);
3824 return sprintf(buf, "(none)\n");
3827 static ssize_t rbd_minor_show(struct device *dev,
3828 struct device_attribute *attr, char *buf)
3830 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3832 return sprintf(buf, "%d\n", rbd_dev->minor);
3835 static ssize_t rbd_client_id_show(struct device *dev,
3836 struct device_attribute *attr, char *buf)
3838 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3840 return sprintf(buf, "client%lld\n",
3841 ceph_client_id(rbd_dev->rbd_client->client));
3844 static ssize_t rbd_pool_show(struct device *dev,
3845 struct device_attribute *attr, char *buf)
3847 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3849 return sprintf(buf, "%s\n", rbd_dev->spec->pool_name);
3852 static ssize_t rbd_pool_id_show(struct device *dev,
3853 struct device_attribute *attr, char *buf)
3855 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3857 return sprintf(buf, "%llu\n",
3858 (unsigned long long) rbd_dev->spec->pool_id);
3861 static ssize_t rbd_name_show(struct device *dev,
3862 struct device_attribute *attr, char *buf)
3864 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3866 if (rbd_dev->spec->image_name)
3867 return sprintf(buf, "%s\n", rbd_dev->spec->image_name);
3869 return sprintf(buf, "(unknown)\n");
3872 static ssize_t rbd_image_id_show(struct device *dev,
3873 struct device_attribute *attr, char *buf)
3875 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3877 return sprintf(buf, "%s\n", rbd_dev->spec->image_id);
3881 * Shows the name of the currently-mapped snapshot (or
3882 * RBD_SNAP_HEAD_NAME for the base image).
3884 static ssize_t rbd_snap_show(struct device *dev,
3885 struct device_attribute *attr,
3888 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3890 return sprintf(buf, "%s\n", rbd_dev->spec->snap_name);
3894 * For a v2 image, shows the chain of parent images, separated by empty
3895 * lines. For v1 images or if there is no parent, shows "(no parent
3898 static ssize_t rbd_parent_show(struct device *dev,
3899 struct device_attribute *attr,
3902 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3905 if (!rbd_dev->parent)
3906 return sprintf(buf, "(no parent image)\n");
3908 for ( ; rbd_dev->parent; rbd_dev = rbd_dev->parent) {
3909 struct rbd_spec *spec = rbd_dev->parent_spec;
3911 count += sprintf(&buf[count], "%s"
3912 "pool_id %llu\npool_name %s\n"
3913 "image_id %s\nimage_name %s\n"
3914 "snap_id %llu\nsnap_name %s\n"
3916 !count ? "" : "\n", /* first? */
3917 spec->pool_id, spec->pool_name,
3918 spec->image_id, spec->image_name ?: "(unknown)",
3919 spec->snap_id, spec->snap_name,
3920 rbd_dev->parent_overlap);
3926 static ssize_t rbd_image_refresh(struct device *dev,
3927 struct device_attribute *attr,
3931 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3934 ret = rbd_dev_refresh(rbd_dev);
3941 static DEVICE_ATTR(size, S_IRUGO, rbd_size_show, NULL);
3942 static DEVICE_ATTR(features, S_IRUGO, rbd_features_show, NULL);
3943 static DEVICE_ATTR(major, S_IRUGO, rbd_major_show, NULL);
3944 static DEVICE_ATTR(minor, S_IRUGO, rbd_minor_show, NULL);
3945 static DEVICE_ATTR(client_id, S_IRUGO, rbd_client_id_show, NULL);
3946 static DEVICE_ATTR(pool, S_IRUGO, rbd_pool_show, NULL);
3947 static DEVICE_ATTR(pool_id, S_IRUGO, rbd_pool_id_show, NULL);
3948 static DEVICE_ATTR(name, S_IRUGO, rbd_name_show, NULL);
3949 static DEVICE_ATTR(image_id, S_IRUGO, rbd_image_id_show, NULL);
3950 static DEVICE_ATTR(refresh, S_IWUSR, NULL, rbd_image_refresh);
3951 static DEVICE_ATTR(current_snap, S_IRUGO, rbd_snap_show, NULL);
3952 static DEVICE_ATTR(parent, S_IRUGO, rbd_parent_show, NULL);
3954 static struct attribute *rbd_attrs[] = {
3955 &dev_attr_size.attr,
3956 &dev_attr_features.attr,
3957 &dev_attr_major.attr,
3958 &dev_attr_minor.attr,
3959 &dev_attr_client_id.attr,
3960 &dev_attr_pool.attr,
3961 &dev_attr_pool_id.attr,
3962 &dev_attr_name.attr,
3963 &dev_attr_image_id.attr,
3964 &dev_attr_current_snap.attr,
3965 &dev_attr_parent.attr,
3966 &dev_attr_refresh.attr,
3970 static struct attribute_group rbd_attr_group = {
3974 static const struct attribute_group *rbd_attr_groups[] = {
3979 static void rbd_sysfs_dev_release(struct device *dev)
3983 static struct device_type rbd_device_type = {
3985 .groups = rbd_attr_groups,
3986 .release = rbd_sysfs_dev_release,
3989 static struct rbd_spec *rbd_spec_get(struct rbd_spec *spec)
3991 kref_get(&spec->kref);
3996 static void rbd_spec_free(struct kref *kref);
3997 static void rbd_spec_put(struct rbd_spec *spec)
4000 kref_put(&spec->kref, rbd_spec_free);
4003 static struct rbd_spec *rbd_spec_alloc(void)
4005 struct rbd_spec *spec;
4007 spec = kzalloc(sizeof (*spec), GFP_KERNEL);
4011 spec->pool_id = CEPH_NOPOOL;
4012 spec->snap_id = CEPH_NOSNAP;
4013 kref_init(&spec->kref);
4018 static void rbd_spec_free(struct kref *kref)
4020 struct rbd_spec *spec = container_of(kref, struct rbd_spec, kref);
4022 kfree(spec->pool_name);
4023 kfree(spec->image_id);
4024 kfree(spec->image_name);
4025 kfree(spec->snap_name);
4029 static struct rbd_device *rbd_dev_create(struct rbd_client *rbdc,
4030 struct rbd_spec *spec)
4032 struct rbd_device *rbd_dev;
4034 rbd_dev = kzalloc(sizeof (*rbd_dev), GFP_KERNEL);
4038 spin_lock_init(&rbd_dev->lock);
4039 INIT_LIST_HEAD(&rbd_dev->rq_queue);
4040 INIT_WORK(&rbd_dev->rq_work, rbd_request_workfn);
4042 atomic_set(&rbd_dev->parent_ref, 0);
4043 INIT_LIST_HEAD(&rbd_dev->node);
4044 init_rwsem(&rbd_dev->header_rwsem);
4046 rbd_dev->spec = spec;
4047 rbd_dev->rbd_client = rbdc;
4049 /* Initialize the layout used for all rbd requests */
4051 rbd_dev->layout.fl_stripe_unit = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER);
4052 rbd_dev->layout.fl_stripe_count = cpu_to_le32(1);
4053 rbd_dev->layout.fl_object_size = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER);
4054 rbd_dev->layout.fl_pg_pool = cpu_to_le32((u32) spec->pool_id);
4059 static void rbd_dev_destroy(struct rbd_device *rbd_dev)
4061 rbd_put_client(rbd_dev->rbd_client);
4062 rbd_spec_put(rbd_dev->spec);
4067 * Get the size and object order for an image snapshot, or if
4068 * snap_id is CEPH_NOSNAP, gets this information for the base
4071 static int _rbd_dev_v2_snap_size(struct rbd_device *rbd_dev, u64 snap_id,
4072 u8 *order, u64 *snap_size)
4074 __le64 snapid = cpu_to_le64(snap_id);
4079 } __attribute__ ((packed)) size_buf = { 0 };
4081 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
4083 &snapid, sizeof (snapid),
4084 &size_buf, sizeof (size_buf));
4085 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
4088 if (ret < sizeof (size_buf))
4092 *order = size_buf.order;
4093 dout(" order %u", (unsigned int)*order);
4095 *snap_size = le64_to_cpu(size_buf.size);
4097 dout(" snap_id 0x%016llx snap_size = %llu\n",
4098 (unsigned long long)snap_id,
4099 (unsigned long long)*snap_size);
4104 static int rbd_dev_v2_image_size(struct rbd_device *rbd_dev)
4106 return _rbd_dev_v2_snap_size(rbd_dev, CEPH_NOSNAP,
4107 &rbd_dev->header.obj_order,
4108 &rbd_dev->header.image_size);
4111 static int rbd_dev_v2_object_prefix(struct rbd_device *rbd_dev)
4117 reply_buf = kzalloc(RBD_OBJ_PREFIX_LEN_MAX, GFP_KERNEL);
4121 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
4122 "rbd", "get_object_prefix", NULL, 0,
4123 reply_buf, RBD_OBJ_PREFIX_LEN_MAX);
4124 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
4129 rbd_dev->header.object_prefix = ceph_extract_encoded_string(&p,
4130 p + ret, NULL, GFP_NOIO);
4133 if (IS_ERR(rbd_dev->header.object_prefix)) {
4134 ret = PTR_ERR(rbd_dev->header.object_prefix);
4135 rbd_dev->header.object_prefix = NULL;
4137 dout(" object_prefix = %s\n", rbd_dev->header.object_prefix);
4145 static int _rbd_dev_v2_snap_features(struct rbd_device *rbd_dev, u64 snap_id,
4148 __le64 snapid = cpu_to_le64(snap_id);
4152 } __attribute__ ((packed)) features_buf = { 0 };
4156 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
4157 "rbd", "get_features",
4158 &snapid, sizeof (snapid),
4159 &features_buf, sizeof (features_buf));
4160 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
4163 if (ret < sizeof (features_buf))
4166 incompat = le64_to_cpu(features_buf.incompat);
4167 if (incompat & ~RBD_FEATURES_SUPPORTED)
4170 *snap_features = le64_to_cpu(features_buf.features);
4172 dout(" snap_id 0x%016llx features = 0x%016llx incompat = 0x%016llx\n",
4173 (unsigned long long)snap_id,
4174 (unsigned long long)*snap_features,
4175 (unsigned long long)le64_to_cpu(features_buf.incompat));
4180 static int rbd_dev_v2_features(struct rbd_device *rbd_dev)
4182 return _rbd_dev_v2_snap_features(rbd_dev, CEPH_NOSNAP,
4183 &rbd_dev->header.features);
4186 static int rbd_dev_v2_parent_info(struct rbd_device *rbd_dev)
4188 struct rbd_spec *parent_spec;
4190 void *reply_buf = NULL;
4200 parent_spec = rbd_spec_alloc();
4204 size = sizeof (__le64) + /* pool_id */
4205 sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX + /* image_id */
4206 sizeof (__le64) + /* snap_id */
4207 sizeof (__le64); /* overlap */
4208 reply_buf = kmalloc(size, GFP_KERNEL);
4214 snapid = cpu_to_le64(rbd_dev->spec->snap_id);
4215 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
4216 "rbd", "get_parent",
4217 &snapid, sizeof (snapid),
4219 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
4224 end = reply_buf + ret;
4226 ceph_decode_64_safe(&p, end, pool_id, out_err);
4227 if (pool_id == CEPH_NOPOOL) {
4229 * Either the parent never existed, or we have
4230 * record of it but the image got flattened so it no
4231 * longer has a parent. When the parent of a
4232 * layered image disappears we immediately set the
4233 * overlap to 0. The effect of this is that all new
4234 * requests will be treated as if the image had no
4237 if (rbd_dev->parent_overlap) {
4238 rbd_dev->parent_overlap = 0;
4240 rbd_dev_parent_put(rbd_dev);
4241 pr_info("%s: clone image has been flattened\n",
4242 rbd_dev->disk->disk_name);
4245 goto out; /* No parent? No problem. */
4248 /* The ceph file layout needs to fit pool id in 32 bits */
4251 if (pool_id > (u64)U32_MAX) {
4252 rbd_warn(NULL, "parent pool id too large (%llu > %u)",
4253 (unsigned long long)pool_id, U32_MAX);
4257 image_id = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL);
4258 if (IS_ERR(image_id)) {
4259 ret = PTR_ERR(image_id);
4262 ceph_decode_64_safe(&p, end, snap_id, out_err);
4263 ceph_decode_64_safe(&p, end, overlap, out_err);
4266 * The parent won't change (except when the clone is
4267 * flattened, already handled that). So we only need to
4268 * record the parent spec we have not already done so.
4270 if (!rbd_dev->parent_spec) {
4271 parent_spec->pool_id = pool_id;
4272 parent_spec->image_id = image_id;
4273 parent_spec->snap_id = snap_id;
4274 rbd_dev->parent_spec = parent_spec;
4275 parent_spec = NULL; /* rbd_dev now owns this */
4281 * We always update the parent overlap. If it's zero we
4282 * treat it specially.
4284 rbd_dev->parent_overlap = overlap;
4288 /* A null parent_spec indicates it's the initial probe */
4292 * The overlap has become zero, so the clone
4293 * must have been resized down to 0 at some
4294 * point. Treat this the same as a flatten.
4296 rbd_dev_parent_put(rbd_dev);
4297 pr_info("%s: clone image now standalone\n",
4298 rbd_dev->disk->disk_name);
4301 * For the initial probe, if we find the
4302 * overlap is zero we just pretend there was
4305 rbd_warn(rbd_dev, "ignoring parent with overlap 0");
4312 rbd_spec_put(parent_spec);
4317 static int rbd_dev_v2_striping_info(struct rbd_device *rbd_dev)
4321 __le64 stripe_count;
4322 } __attribute__ ((packed)) striping_info_buf = { 0 };
4323 size_t size = sizeof (striping_info_buf);
4330 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
4331 "rbd", "get_stripe_unit_count", NULL, 0,
4332 (char *)&striping_info_buf, size);
4333 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
4340 * We don't actually support the "fancy striping" feature
4341 * (STRIPINGV2) yet, but if the striping sizes are the
4342 * defaults the behavior is the same as before. So find
4343 * out, and only fail if the image has non-default values.
4346 obj_size = (u64)1 << rbd_dev->header.obj_order;
4347 p = &striping_info_buf;
4348 stripe_unit = ceph_decode_64(&p);
4349 if (stripe_unit != obj_size) {
4350 rbd_warn(rbd_dev, "unsupported stripe unit "
4351 "(got %llu want %llu)",
4352 stripe_unit, obj_size);
4355 stripe_count = ceph_decode_64(&p);
4356 if (stripe_count != 1) {
4357 rbd_warn(rbd_dev, "unsupported stripe count "
4358 "(got %llu want 1)", stripe_count);
4361 rbd_dev->header.stripe_unit = stripe_unit;
4362 rbd_dev->header.stripe_count = stripe_count;
4367 static char *rbd_dev_image_name(struct rbd_device *rbd_dev)
4369 size_t image_id_size;
4374 void *reply_buf = NULL;
4376 char *image_name = NULL;
4379 rbd_assert(!rbd_dev->spec->image_name);
4381 len = strlen(rbd_dev->spec->image_id);
4382 image_id_size = sizeof (__le32) + len;
4383 image_id = kmalloc(image_id_size, GFP_KERNEL);
4388 end = image_id + image_id_size;
4389 ceph_encode_string(&p, end, rbd_dev->spec->image_id, (u32)len);
4391 size = sizeof (__le32) + RBD_IMAGE_NAME_LEN_MAX;
4392 reply_buf = kmalloc(size, GFP_KERNEL);
4396 ret = rbd_obj_method_sync(rbd_dev, RBD_DIRECTORY,
4397 "rbd", "dir_get_name",
4398 image_id, image_id_size,
4403 end = reply_buf + ret;
4405 image_name = ceph_extract_encoded_string(&p, end, &len, GFP_KERNEL);
4406 if (IS_ERR(image_name))
4409 dout("%s: name is %s len is %zd\n", __func__, image_name, len);
4417 static u64 rbd_v1_snap_id_by_name(struct rbd_device *rbd_dev, const char *name)
4419 struct ceph_snap_context *snapc = rbd_dev->header.snapc;
4420 const char *snap_name;
4423 /* Skip over names until we find the one we are looking for */
4425 snap_name = rbd_dev->header.snap_names;
4426 while (which < snapc->num_snaps) {
4427 if (!strcmp(name, snap_name))
4428 return snapc->snaps[which];
4429 snap_name += strlen(snap_name) + 1;
4435 static u64 rbd_v2_snap_id_by_name(struct rbd_device *rbd_dev, const char *name)
4437 struct ceph_snap_context *snapc = rbd_dev->header.snapc;
4442 for (which = 0; !found && which < snapc->num_snaps; which++) {
4443 const char *snap_name;
4445 snap_id = snapc->snaps[which];
4446 snap_name = rbd_dev_v2_snap_name(rbd_dev, snap_id);
4447 if (IS_ERR(snap_name)) {
4448 /* ignore no-longer existing snapshots */
4449 if (PTR_ERR(snap_name) == -ENOENT)
4454 found = !strcmp(name, snap_name);
4457 return found ? snap_id : CEPH_NOSNAP;
4461 * Assumes name is never RBD_SNAP_HEAD_NAME; returns CEPH_NOSNAP if
4462 * no snapshot by that name is found, or if an error occurs.
4464 static u64 rbd_snap_id_by_name(struct rbd_device *rbd_dev, const char *name)
4466 if (rbd_dev->image_format == 1)
4467 return rbd_v1_snap_id_by_name(rbd_dev, name);
4469 return rbd_v2_snap_id_by_name(rbd_dev, name);
4473 * An image being mapped will have everything but the snap id.
4475 static int rbd_spec_fill_snap_id(struct rbd_device *rbd_dev)
4477 struct rbd_spec *spec = rbd_dev->spec;
4479 rbd_assert(spec->pool_id != CEPH_NOPOOL && spec->pool_name);
4480 rbd_assert(spec->image_id && spec->image_name);
4481 rbd_assert(spec->snap_name);
4483 if (strcmp(spec->snap_name, RBD_SNAP_HEAD_NAME)) {
4486 snap_id = rbd_snap_id_by_name(rbd_dev, spec->snap_name);
4487 if (snap_id == CEPH_NOSNAP)
4490 spec->snap_id = snap_id;
4492 spec->snap_id = CEPH_NOSNAP;
4499 * A parent image will have all ids but none of the names.
4501 * All names in an rbd spec are dynamically allocated. It's OK if we
4502 * can't figure out the name for an image id.
4504 static int rbd_spec_fill_names(struct rbd_device *rbd_dev)
4506 struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
4507 struct rbd_spec *spec = rbd_dev->spec;
4508 const char *pool_name;
4509 const char *image_name;
4510 const char *snap_name;
4513 rbd_assert(spec->pool_id != CEPH_NOPOOL);
4514 rbd_assert(spec->image_id);
4515 rbd_assert(spec->snap_id != CEPH_NOSNAP);
4517 /* Get the pool name; we have to make our own copy of this */
4519 pool_name = ceph_pg_pool_name_by_id(osdc->osdmap, spec->pool_id);
4521 rbd_warn(rbd_dev, "no pool with id %llu", spec->pool_id);
4524 pool_name = kstrdup(pool_name, GFP_KERNEL);
4528 /* Fetch the image name; tolerate failure here */
4530 image_name = rbd_dev_image_name(rbd_dev);
4532 rbd_warn(rbd_dev, "unable to get image name");
4534 /* Fetch the snapshot name */
4536 snap_name = rbd_snap_name(rbd_dev, spec->snap_id);
4537 if (IS_ERR(snap_name)) {
4538 ret = PTR_ERR(snap_name);
4542 spec->pool_name = pool_name;
4543 spec->image_name = image_name;
4544 spec->snap_name = snap_name;
4554 static int rbd_dev_v2_snap_context(struct rbd_device *rbd_dev)
4563 struct ceph_snap_context *snapc;
4567 * We'll need room for the seq value (maximum snapshot id),
4568 * snapshot count, and array of that many snapshot ids.
4569 * For now we have a fixed upper limit on the number we're
4570 * prepared to receive.
4572 size = sizeof (__le64) + sizeof (__le32) +
4573 RBD_MAX_SNAP_COUNT * sizeof (__le64);
4574 reply_buf = kzalloc(size, GFP_KERNEL);
4578 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
4579 "rbd", "get_snapcontext", NULL, 0,
4581 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
4586 end = reply_buf + ret;
4588 ceph_decode_64_safe(&p, end, seq, out);
4589 ceph_decode_32_safe(&p, end, snap_count, out);
4592 * Make sure the reported number of snapshot ids wouldn't go
4593 * beyond the end of our buffer. But before checking that,
4594 * make sure the computed size of the snapshot context we
4595 * allocate is representable in a size_t.
4597 if (snap_count > (SIZE_MAX - sizeof (struct ceph_snap_context))
4602 if (!ceph_has_room(&p, end, snap_count * sizeof (__le64)))
4606 snapc = ceph_create_snap_context(snap_count, GFP_KERNEL);
4612 for (i = 0; i < snap_count; i++)
4613 snapc->snaps[i] = ceph_decode_64(&p);
4615 ceph_put_snap_context(rbd_dev->header.snapc);
4616 rbd_dev->header.snapc = snapc;
4618 dout(" snap context seq = %llu, snap_count = %u\n",
4619 (unsigned long long)seq, (unsigned int)snap_count);
4626 static const char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev,
4637 size = sizeof (__le32) + RBD_MAX_SNAP_NAME_LEN;
4638 reply_buf = kmalloc(size, GFP_KERNEL);
4640 return ERR_PTR(-ENOMEM);
4642 snapid = cpu_to_le64(snap_id);
4643 ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
4644 "rbd", "get_snapshot_name",
4645 &snapid, sizeof (snapid),
4647 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
4649 snap_name = ERR_PTR(ret);
4654 end = reply_buf + ret;
4655 snap_name = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL);
4656 if (IS_ERR(snap_name))
4659 dout(" snap_id 0x%016llx snap_name = %s\n",
4660 (unsigned long long)snap_id, snap_name);
4667 static int rbd_dev_v2_header_info(struct rbd_device *rbd_dev)
4669 bool first_time = rbd_dev->header.object_prefix == NULL;
4672 ret = rbd_dev_v2_image_size(rbd_dev);
4677 ret = rbd_dev_v2_header_onetime(rbd_dev);
4682 ret = rbd_dev_v2_snap_context(rbd_dev);
4683 dout("rbd_dev_v2_snap_context returned %d\n", ret);
4688 static int rbd_dev_header_info(struct rbd_device *rbd_dev)
4690 rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
4692 if (rbd_dev->image_format == 1)
4693 return rbd_dev_v1_header_info(rbd_dev);
4695 return rbd_dev_v2_header_info(rbd_dev);
4698 static int rbd_bus_add_dev(struct rbd_device *rbd_dev)
4703 dev = &rbd_dev->dev;
4704 dev->bus = &rbd_bus_type;
4705 dev->type = &rbd_device_type;
4706 dev->parent = &rbd_root_dev;
4707 dev->release = rbd_dev_device_release;
4708 dev_set_name(dev, "%d", rbd_dev->dev_id);
4709 ret = device_register(dev);
4714 static void rbd_bus_del_dev(struct rbd_device *rbd_dev)
4716 device_unregister(&rbd_dev->dev);
4720 * Get a unique rbd identifier for the given new rbd_dev, and add
4721 * the rbd_dev to the global list.
4723 static int rbd_dev_id_get(struct rbd_device *rbd_dev)
4727 new_dev_id = ida_simple_get(&rbd_dev_id_ida,
4728 0, minor_to_rbd_dev_id(1 << MINORBITS),
4733 rbd_dev->dev_id = new_dev_id;
4735 spin_lock(&rbd_dev_list_lock);
4736 list_add_tail(&rbd_dev->node, &rbd_dev_list);
4737 spin_unlock(&rbd_dev_list_lock);
4739 dout("rbd_dev %p given dev id %d\n", rbd_dev, rbd_dev->dev_id);
4745 * Remove an rbd_dev from the global list, and record that its
4746 * identifier is no longer in use.
4748 static void rbd_dev_id_put(struct rbd_device *rbd_dev)
4750 spin_lock(&rbd_dev_list_lock);
4751 list_del_init(&rbd_dev->node);
4752 spin_unlock(&rbd_dev_list_lock);
4754 ida_simple_remove(&rbd_dev_id_ida, rbd_dev->dev_id);
4756 dout("rbd_dev %p released dev id %d\n", rbd_dev, rbd_dev->dev_id);
4760 * Skips over white space at *buf, and updates *buf to point to the
4761 * first found non-space character (if any). Returns the length of
4762 * the token (string of non-white space characters) found. Note
4763 * that *buf must be terminated with '\0'.
4765 static inline size_t next_token(const char **buf)
4768 * These are the characters that produce nonzero for
4769 * isspace() in the "C" and "POSIX" locales.
4771 const char *spaces = " \f\n\r\t\v";
4773 *buf += strspn(*buf, spaces); /* Find start of token */
4775 return strcspn(*buf, spaces); /* Return token length */
4779 * Finds the next token in *buf, and if the provided token buffer is
4780 * big enough, copies the found token into it. The result, if
4781 * copied, is guaranteed to be terminated with '\0'. Note that *buf
4782 * must be terminated with '\0' on entry.
4784 * Returns the length of the token found (not including the '\0').
4785 * Return value will be 0 if no token is found, and it will be >=
4786 * token_size if the token would not fit.
4788 * The *buf pointer will be updated to point beyond the end of the
4789 * found token. Note that this occurs even if the token buffer is
4790 * too small to hold it.
4792 static inline size_t copy_token(const char **buf,
4798 len = next_token(buf);
4799 if (len < token_size) {
4800 memcpy(token, *buf, len);
4801 *(token + len) = '\0';
4809 * Finds the next token in *buf, dynamically allocates a buffer big
4810 * enough to hold a copy of it, and copies the token into the new
4811 * buffer. The copy is guaranteed to be terminated with '\0'. Note
4812 * that a duplicate buffer is created even for a zero-length token.
4814 * Returns a pointer to the newly-allocated duplicate, or a null
4815 * pointer if memory for the duplicate was not available. If
4816 * the lenp argument is a non-null pointer, the length of the token
4817 * (not including the '\0') is returned in *lenp.
4819 * If successful, the *buf pointer will be updated to point beyond
4820 * the end of the found token.
4822 * Note: uses GFP_KERNEL for allocation.
4824 static inline char *dup_token(const char **buf, size_t *lenp)
4829 len = next_token(buf);
4830 dup = kmemdup(*buf, len + 1, GFP_KERNEL);
4833 *(dup + len) = '\0';
4843 * Parse the options provided for an "rbd add" (i.e., rbd image
4844 * mapping) request. These arrive via a write to /sys/bus/rbd/add,
4845 * and the data written is passed here via a NUL-terminated buffer.
4846 * Returns 0 if successful or an error code otherwise.
4848 * The information extracted from these options is recorded in
4849 * the other parameters which return dynamically-allocated
4852 * The address of a pointer that will refer to a ceph options
4853 * structure. Caller must release the returned pointer using
4854 * ceph_destroy_options() when it is no longer needed.
4856 * Address of an rbd options pointer. Fully initialized by
4857 * this function; caller must release with kfree().
4859 * Address of an rbd image specification pointer. Fully
4860 * initialized by this function based on parsed options.
4861 * Caller must release with rbd_spec_put().
4863 * The options passed take this form:
4864 * <mon_addrs> <options> <pool_name> <image_name> [<snap_id>]
4867 * A comma-separated list of one or more monitor addresses.
4868 * A monitor address is an ip address, optionally followed
4869 * by a port number (separated by a colon).
4870 * I.e.: ip1[:port1][,ip2[:port2]...]
4872 * A comma-separated list of ceph and/or rbd options.
4874 * The name of the rados pool containing the rbd image.
4876 * The name of the image in that pool to map.
4878 * An optional snapshot id. If provided, the mapping will
4879 * present data from the image at the time that snapshot was
4880 * created. The image head is used if no snapshot id is
4881 * provided. Snapshot mappings are always read-only.
4883 static int rbd_add_parse_args(const char *buf,
4884 struct ceph_options **ceph_opts,
4885 struct rbd_options **opts,
4886 struct rbd_spec **rbd_spec)
4890 const char *mon_addrs;
4892 size_t mon_addrs_size;
4893 struct rbd_spec *spec = NULL;
4894 struct rbd_options *rbd_opts = NULL;
4895 struct ceph_options *copts;
4898 /* The first four tokens are required */
4900 len = next_token(&buf);
4902 rbd_warn(NULL, "no monitor address(es) provided");
4906 mon_addrs_size = len + 1;
4910 options = dup_token(&buf, NULL);
4914 rbd_warn(NULL, "no options provided");
4918 spec = rbd_spec_alloc();
4922 spec->pool_name = dup_token(&buf, NULL);
4923 if (!spec->pool_name)
4925 if (!*spec->pool_name) {
4926 rbd_warn(NULL, "no pool name provided");
4930 spec->image_name = dup_token(&buf, NULL);
4931 if (!spec->image_name)
4933 if (!*spec->image_name) {
4934 rbd_warn(NULL, "no image name provided");
4939 * Snapshot name is optional; default is to use "-"
4940 * (indicating the head/no snapshot).
4942 len = next_token(&buf);
4944 buf = RBD_SNAP_HEAD_NAME; /* No snapshot supplied */
4945 len = sizeof (RBD_SNAP_HEAD_NAME) - 1;
4946 } else if (len > RBD_MAX_SNAP_NAME_LEN) {
4947 ret = -ENAMETOOLONG;
4950 snap_name = kmemdup(buf, len + 1, GFP_KERNEL);
4953 *(snap_name + len) = '\0';
4954 spec->snap_name = snap_name;
4956 /* Initialize all rbd options to the defaults */
4958 rbd_opts = kzalloc(sizeof (*rbd_opts), GFP_KERNEL);
4962 rbd_opts->read_only = RBD_READ_ONLY_DEFAULT;
4964 copts = ceph_parse_options(options, mon_addrs,
4965 mon_addrs + mon_addrs_size - 1,
4966 parse_rbd_opts_token, rbd_opts);
4967 if (IS_ERR(copts)) {
4968 ret = PTR_ERR(copts);
4989 * Return pool id (>= 0) or a negative error code.
4991 static int rbd_add_get_pool_id(struct rbd_client *rbdc, const char *pool_name)
4994 unsigned long timeout = rbdc->client->options->mount_timeout * HZ;
4999 ret = ceph_pg_poolid_by_name(rbdc->client->osdc.osdmap, pool_name);
5000 if (ret == -ENOENT && tries++ < 1) {
5001 ret = ceph_monc_do_get_version(&rbdc->client->monc, "osdmap",
5006 if (rbdc->client->osdc.osdmap->epoch < newest_epoch) {
5007 ceph_monc_request_next_osdmap(&rbdc->client->monc);
5008 (void) ceph_monc_wait_osdmap(&rbdc->client->monc,
5009 newest_epoch, timeout);
5012 /* the osdmap we have is new enough */
5021 * An rbd format 2 image has a unique identifier, distinct from the
5022 * name given to it by the user. Internally, that identifier is
5023 * what's used to specify the names of objects related to the image.
5025 * A special "rbd id" object is used to map an rbd image name to its
5026 * id. If that object doesn't exist, then there is no v2 rbd image
5027 * with the supplied name.
5029 * This function will record the given rbd_dev's image_id field if
5030 * it can be determined, and in that case will return 0. If any
5031 * errors occur a negative errno will be returned and the rbd_dev's
5032 * image_id field will be unchanged (and should be NULL).
5034 static int rbd_dev_image_id(struct rbd_device *rbd_dev)
5043 * When probing a parent image, the image id is already
5044 * known (and the image name likely is not). There's no
5045 * need to fetch the image id again in this case. We
5046 * do still need to set the image format though.
5048 if (rbd_dev->spec->image_id) {
5049 rbd_dev->image_format = *rbd_dev->spec->image_id ? 2 : 1;
5055 * First, see if the format 2 image id file exists, and if
5056 * so, get the image's persistent id from it.
5058 size = sizeof (RBD_ID_PREFIX) + strlen(rbd_dev->spec->image_name);
5059 object_name = kmalloc(size, GFP_NOIO);
5062 sprintf(object_name, "%s%s", RBD_ID_PREFIX, rbd_dev->spec->image_name);
5063 dout("rbd id object name is %s\n", object_name);
5065 /* Response will be an encoded string, which includes a length */
5067 size = sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX;
5068 response = kzalloc(size, GFP_NOIO);
5074 /* If it doesn't exist we'll assume it's a format 1 image */
5076 ret = rbd_obj_method_sync(rbd_dev, object_name,
5077 "rbd", "get_id", NULL, 0,
5078 response, RBD_IMAGE_ID_LEN_MAX);
5079 dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
5080 if (ret == -ENOENT) {
5081 image_id = kstrdup("", GFP_KERNEL);
5082 ret = image_id ? 0 : -ENOMEM;
5084 rbd_dev->image_format = 1;
5085 } else if (ret >= 0) {
5088 image_id = ceph_extract_encoded_string(&p, p + ret,
5090 ret = PTR_ERR_OR_ZERO(image_id);
5092 rbd_dev->image_format = 2;
5096 rbd_dev->spec->image_id = image_id;
5097 dout("image_id is %s\n", image_id);
5107 * Undo whatever state changes are made by v1 or v2 header info
5110 static void rbd_dev_unprobe(struct rbd_device *rbd_dev)
5112 struct rbd_image_header *header;
5114 /* Drop parent reference unless it's already been done (or none) */
5116 if (rbd_dev->parent_overlap)
5117 rbd_dev_parent_put(rbd_dev);
5119 /* Free dynamic fields from the header, then zero it out */
5121 header = &rbd_dev->header;
5122 ceph_put_snap_context(header->snapc);
5123 kfree(header->snap_sizes);
5124 kfree(header->snap_names);
5125 kfree(header->object_prefix);
5126 memset(header, 0, sizeof (*header));
5129 static int rbd_dev_v2_header_onetime(struct rbd_device *rbd_dev)
5133 ret = rbd_dev_v2_object_prefix(rbd_dev);
5138 * Get the and check features for the image. Currently the
5139 * features are assumed to never change.
5141 ret = rbd_dev_v2_features(rbd_dev);
5145 /* If the image supports fancy striping, get its parameters */
5147 if (rbd_dev->header.features & RBD_FEATURE_STRIPINGV2) {
5148 ret = rbd_dev_v2_striping_info(rbd_dev);
5152 /* No support for crypto and compression type format 2 images */
5156 rbd_dev->header.features = 0;
5157 kfree(rbd_dev->header.object_prefix);
5158 rbd_dev->header.object_prefix = NULL;
5163 static int rbd_dev_probe_parent(struct rbd_device *rbd_dev)
5165 struct rbd_device *parent = NULL;
5166 struct rbd_spec *parent_spec;
5167 struct rbd_client *rbdc;
5170 if (!rbd_dev->parent_spec)
5173 * We need to pass a reference to the client and the parent
5174 * spec when creating the parent rbd_dev. Images related by
5175 * parent/child relationships always share both.
5177 parent_spec = rbd_spec_get(rbd_dev->parent_spec);
5178 rbdc = __rbd_get_client(rbd_dev->rbd_client);
5181 parent = rbd_dev_create(rbdc, parent_spec);
5185 ret = rbd_dev_image_probe(parent, false);
5188 rbd_dev->parent = parent;
5189 atomic_set(&rbd_dev->parent_ref, 1);
5194 rbd_dev_unparent(rbd_dev);
5195 kfree(rbd_dev->header_name);
5196 rbd_dev_destroy(parent);
5198 rbd_put_client(rbdc);
5199 rbd_spec_put(parent_spec);
5205 static int rbd_dev_device_setup(struct rbd_device *rbd_dev)
5209 /* Get an id and fill in device name. */
5211 ret = rbd_dev_id_get(rbd_dev);
5215 BUILD_BUG_ON(DEV_NAME_LEN
5216 < sizeof (RBD_DRV_NAME) + MAX_INT_FORMAT_WIDTH);
5217 sprintf(rbd_dev->name, "%s%d", RBD_DRV_NAME, rbd_dev->dev_id);
5219 /* Record our major and minor device numbers. */
5221 if (!single_major) {
5222 ret = register_blkdev(0, rbd_dev->name);
5226 rbd_dev->major = ret;
5229 rbd_dev->major = rbd_major;
5230 rbd_dev->minor = rbd_dev_id_to_minor(rbd_dev->dev_id);
5233 /* Set up the blkdev mapping. */
5235 ret = rbd_init_disk(rbd_dev);
5237 goto err_out_blkdev;
5239 ret = rbd_dev_mapping_set(rbd_dev);
5243 set_capacity(rbd_dev->disk, rbd_dev->mapping.size / SECTOR_SIZE);
5244 set_disk_ro(rbd_dev->disk, rbd_dev->mapping.read_only);
5246 ret = rbd_bus_add_dev(rbd_dev);
5248 goto err_out_mapping;
5250 /* Everything's ready. Announce the disk to the world. */
5252 set_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags);
5253 add_disk(rbd_dev->disk);
5255 pr_info("%s: added with size 0x%llx\n", rbd_dev->disk->disk_name,
5256 (unsigned long long) rbd_dev->mapping.size);
5261 rbd_dev_mapping_clear(rbd_dev);
5263 rbd_free_disk(rbd_dev);
5266 unregister_blkdev(rbd_dev->major, rbd_dev->name);
5268 rbd_dev_id_put(rbd_dev);
5269 rbd_dev_mapping_clear(rbd_dev);
5274 static int rbd_dev_header_name(struct rbd_device *rbd_dev)
5276 struct rbd_spec *spec = rbd_dev->spec;
5279 /* Record the header object name for this rbd image. */
5281 rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
5283 if (rbd_dev->image_format == 1)
5284 size = strlen(spec->image_name) + sizeof (RBD_SUFFIX);
5286 size = sizeof (RBD_HEADER_PREFIX) + strlen(spec->image_id);
5288 rbd_dev->header_name = kmalloc(size, GFP_KERNEL);
5289 if (!rbd_dev->header_name)
5292 if (rbd_dev->image_format == 1)
5293 sprintf(rbd_dev->header_name, "%s%s",
5294 spec->image_name, RBD_SUFFIX);
5296 sprintf(rbd_dev->header_name, "%s%s",
5297 RBD_HEADER_PREFIX, spec->image_id);
5301 static void rbd_dev_image_release(struct rbd_device *rbd_dev)
5303 rbd_dev_unprobe(rbd_dev);
5304 kfree(rbd_dev->header_name);
5305 rbd_dev->header_name = NULL;
5306 rbd_dev->image_format = 0;
5307 kfree(rbd_dev->spec->image_id);
5308 rbd_dev->spec->image_id = NULL;
5310 rbd_dev_destroy(rbd_dev);
5314 * Probe for the existence of the header object for the given rbd
5315 * device. If this image is the one being mapped (i.e., not a
5316 * parent), initiate a watch on its header object before using that
5317 * object to get detailed information about the rbd image.
5319 static int rbd_dev_image_probe(struct rbd_device *rbd_dev, bool mapping)
5324 * Get the id from the image id object. Unless there's an
5325 * error, rbd_dev->spec->image_id will be filled in with
5326 * a dynamically-allocated string, and rbd_dev->image_format
5327 * will be set to either 1 or 2.
5329 ret = rbd_dev_image_id(rbd_dev);
5333 ret = rbd_dev_header_name(rbd_dev);
5335 goto err_out_format;
5338 ret = rbd_dev_header_watch_sync(rbd_dev);
5340 goto out_header_name;
5343 ret = rbd_dev_header_info(rbd_dev);
5348 * If this image is the one being mapped, we have pool name and
5349 * id, image name and id, and snap name - need to fill snap id.
5350 * Otherwise this is a parent image, identified by pool, image
5351 * and snap ids - need to fill in names for those ids.
5354 ret = rbd_spec_fill_snap_id(rbd_dev);
5356 ret = rbd_spec_fill_names(rbd_dev);
5360 if (rbd_dev->header.features & RBD_FEATURE_LAYERING) {
5361 ret = rbd_dev_v2_parent_info(rbd_dev);
5366 * Need to warn users if this image is the one being
5367 * mapped and has a parent.
5369 if (mapping && rbd_dev->parent_spec)
5371 "WARNING: kernel layering is EXPERIMENTAL!");
5374 ret = rbd_dev_probe_parent(rbd_dev);
5378 dout("discovered format %u image, header name is %s\n",
5379 rbd_dev->image_format, rbd_dev->header_name);
5383 rbd_dev_unprobe(rbd_dev);
5386 rbd_dev_header_unwatch_sync(rbd_dev);
5388 kfree(rbd_dev->header_name);
5389 rbd_dev->header_name = NULL;
5391 rbd_dev->image_format = 0;
5392 kfree(rbd_dev->spec->image_id);
5393 rbd_dev->spec->image_id = NULL;
5397 static ssize_t do_rbd_add(struct bus_type *bus,
5401 struct rbd_device *rbd_dev = NULL;
5402 struct ceph_options *ceph_opts = NULL;
5403 struct rbd_options *rbd_opts = NULL;
5404 struct rbd_spec *spec = NULL;
5405 struct rbd_client *rbdc;
5409 if (!try_module_get(THIS_MODULE))
5412 /* parse add command */
5413 rc = rbd_add_parse_args(buf, &ceph_opts, &rbd_opts, &spec);
5415 goto err_out_module;
5416 read_only = rbd_opts->read_only;
5418 rbd_opts = NULL; /* done with this */
5420 rbdc = rbd_get_client(ceph_opts);
5427 rc = rbd_add_get_pool_id(rbdc, spec->pool_name);
5429 goto err_out_client;
5430 spec->pool_id = (u64)rc;
5432 /* The ceph file layout needs to fit pool id in 32 bits */
5434 if (spec->pool_id > (u64)U32_MAX) {
5435 rbd_warn(NULL, "pool id too large (%llu > %u)",
5436 (unsigned long long)spec->pool_id, U32_MAX);
5438 goto err_out_client;
5441 rbd_dev = rbd_dev_create(rbdc, spec);
5443 goto err_out_client;
5444 rbdc = NULL; /* rbd_dev now owns this */
5445 spec = NULL; /* rbd_dev now owns this */
5447 rc = rbd_dev_image_probe(rbd_dev, true);
5449 goto err_out_rbd_dev;
5451 /* If we are mapping a snapshot it must be marked read-only */
5453 if (rbd_dev->spec->snap_id != CEPH_NOSNAP)
5455 rbd_dev->mapping.read_only = read_only;
5457 rc = rbd_dev_device_setup(rbd_dev);
5460 * rbd_dev_header_unwatch_sync() can't be moved into
5461 * rbd_dev_image_release() without refactoring, see
5462 * commit 1f3ef78861ac.
5464 rbd_dev_header_unwatch_sync(rbd_dev);
5465 rbd_dev_image_release(rbd_dev);
5466 goto err_out_module;
5472 rbd_dev_destroy(rbd_dev);
5474 rbd_put_client(rbdc);
5478 module_put(THIS_MODULE);
5480 dout("Error adding device %s\n", buf);
5485 static ssize_t rbd_add(struct bus_type *bus,
5492 return do_rbd_add(bus, buf, count);
5495 static ssize_t rbd_add_single_major(struct bus_type *bus,
5499 return do_rbd_add(bus, buf, count);
5502 static void rbd_dev_device_release(struct device *dev)
5504 struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
5506 rbd_free_disk(rbd_dev);
5507 clear_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags);
5508 rbd_dev_mapping_clear(rbd_dev);
5510 unregister_blkdev(rbd_dev->major, rbd_dev->name);
5511 rbd_dev_id_put(rbd_dev);
5512 rbd_dev_mapping_clear(rbd_dev);
5515 static void rbd_dev_remove_parent(struct rbd_device *rbd_dev)
5517 while (rbd_dev->parent) {
5518 struct rbd_device *first = rbd_dev;
5519 struct rbd_device *second = first->parent;
5520 struct rbd_device *third;
5523 * Follow to the parent with no grandparent and
5526 while (second && (third = second->parent)) {
5531 rbd_dev_image_release(second);
5532 first->parent = NULL;
5533 first->parent_overlap = 0;
5535 rbd_assert(first->parent_spec);
5536 rbd_spec_put(first->parent_spec);
5537 first->parent_spec = NULL;
5541 static ssize_t do_rbd_remove(struct bus_type *bus,
5545 struct rbd_device *rbd_dev = NULL;
5546 struct list_head *tmp;
5549 bool already = false;
5552 ret = kstrtoul(buf, 10, &ul);
5556 /* convert to int; abort if we lost anything in the conversion */
5562 spin_lock(&rbd_dev_list_lock);
5563 list_for_each(tmp, &rbd_dev_list) {
5564 rbd_dev = list_entry(tmp, struct rbd_device, node);
5565 if (rbd_dev->dev_id == dev_id) {
5571 spin_lock_irq(&rbd_dev->lock);
5572 if (rbd_dev->open_count)
5575 already = test_and_set_bit(RBD_DEV_FLAG_REMOVING,
5577 spin_unlock_irq(&rbd_dev->lock);
5579 spin_unlock(&rbd_dev_list_lock);
5580 if (ret < 0 || already)
5583 rbd_dev_header_unwatch_sync(rbd_dev);
5585 * flush remaining watch callbacks - these must be complete
5586 * before the osd_client is shutdown
5588 dout("%s: flushing notifies", __func__);
5589 ceph_osdc_flush_notifies(&rbd_dev->rbd_client->client->osdc);
5592 * Don't free anything from rbd_dev->disk until after all
5593 * notifies are completely processed. Otherwise
5594 * rbd_bus_del_dev() will race with rbd_watch_cb(), resulting
5595 * in a potential use after free of rbd_dev->disk or rbd_dev.
5597 rbd_bus_del_dev(rbd_dev);
5598 rbd_dev_image_release(rbd_dev);
5599 module_put(THIS_MODULE);
5604 static ssize_t rbd_remove(struct bus_type *bus,
5611 return do_rbd_remove(bus, buf, count);
5614 static ssize_t rbd_remove_single_major(struct bus_type *bus,
5618 return do_rbd_remove(bus, buf, count);
5622 * create control files in sysfs
5625 static int rbd_sysfs_init(void)
5629 ret = device_register(&rbd_root_dev);
5633 ret = bus_register(&rbd_bus_type);
5635 device_unregister(&rbd_root_dev);
5640 static void rbd_sysfs_cleanup(void)
5642 bus_unregister(&rbd_bus_type);
5643 device_unregister(&rbd_root_dev);
5646 static int rbd_slab_init(void)
5648 rbd_assert(!rbd_img_request_cache);
5649 rbd_img_request_cache = kmem_cache_create("rbd_img_request",
5650 sizeof (struct rbd_img_request),
5651 __alignof__(struct rbd_img_request),
5653 if (!rbd_img_request_cache)
5656 rbd_assert(!rbd_obj_request_cache);
5657 rbd_obj_request_cache = kmem_cache_create("rbd_obj_request",
5658 sizeof (struct rbd_obj_request),
5659 __alignof__(struct rbd_obj_request),
5661 if (!rbd_obj_request_cache)
5664 rbd_assert(!rbd_segment_name_cache);
5665 rbd_segment_name_cache = kmem_cache_create("rbd_segment_name",
5666 CEPH_MAX_OID_NAME_LEN + 1, 1, 0, NULL);
5667 if (rbd_segment_name_cache)
5670 if (rbd_obj_request_cache) {
5671 kmem_cache_destroy(rbd_obj_request_cache);
5672 rbd_obj_request_cache = NULL;
5675 kmem_cache_destroy(rbd_img_request_cache);
5676 rbd_img_request_cache = NULL;
5681 static void rbd_slab_exit(void)
5683 rbd_assert(rbd_segment_name_cache);
5684 kmem_cache_destroy(rbd_segment_name_cache);
5685 rbd_segment_name_cache = NULL;
5687 rbd_assert(rbd_obj_request_cache);
5688 kmem_cache_destroy(rbd_obj_request_cache);
5689 rbd_obj_request_cache = NULL;
5691 rbd_assert(rbd_img_request_cache);
5692 kmem_cache_destroy(rbd_img_request_cache);
5693 rbd_img_request_cache = NULL;
5696 static int __init rbd_init(void)
5700 if (!libceph_compatible(NULL)) {
5701 rbd_warn(NULL, "libceph incompatibility (quitting)");
5705 rc = rbd_slab_init();
5710 * The number of active work items is limited by the number of
5711 * rbd devices, so leave @max_active at default.
5713 rbd_wq = alloc_workqueue(RBD_DRV_NAME, WQ_MEM_RECLAIM, 0);
5720 rbd_major = register_blkdev(0, RBD_DRV_NAME);
5721 if (rbd_major < 0) {
5727 rc = rbd_sysfs_init();
5729 goto err_out_blkdev;
5732 pr_info("loaded (major %d)\n", rbd_major);
5734 pr_info("loaded\n");
5740 unregister_blkdev(rbd_major, RBD_DRV_NAME);
5742 destroy_workqueue(rbd_wq);
5748 static void __exit rbd_exit(void)
5750 ida_destroy(&rbd_dev_id_ida);
5751 rbd_sysfs_cleanup();
5753 unregister_blkdev(rbd_major, RBD_DRV_NAME);
5754 destroy_workqueue(rbd_wq);
5758 module_init(rbd_init);
5759 module_exit(rbd_exit);
5761 MODULE_AUTHOR("Alex Elder <elder@inktank.com>");
5762 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
5763 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
5764 /* following authorship retained from original osdblk.c */
5765 MODULE_AUTHOR("Jeff Garzik <jeff@garzik.org>");
5767 MODULE_DESCRIPTION("RADOS Block Device (RBD) driver");
5768 MODULE_LICENSE("GPL");