| |
| /* |
| rbd.c -- Export ceph rados objects as a Linux block device |
| |
| |
| based on drivers/block/osdblk.c: |
| |
| Copyright 2009 Red Hat, Inc. |
| |
| This program is free software; you can redistribute it and/or modify |
| it under the terms of the GNU General Public License as published by |
| the Free Software Foundation. |
| |
| This program is distributed in the hope that it will be useful, |
| but WITHOUT ANY WARRANTY; without even the implied warranty of |
| MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| GNU General Public License for more details. |
| |
| You should have received a copy of the GNU General Public License |
| along with this program; see the file COPYING. If not, write to |
| the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA. |
| |
| |
| |
| For usage instructions, please refer to: |
| |
| Documentation/ABI/testing/sysfs-bus-rbd |
| |
| */ |
| |
| #include <linux/ceph/libceph.h> |
| #include <linux/ceph/osd_client.h> |
| #include <linux/ceph/mon_client.h> |
| #include <linux/ceph/cls_lock_client.h> |
| #include <linux/ceph/striper.h> |
| #include <linux/ceph/decode.h> |
| #include <linux/fs_parser.h> |
| #include <linux/bsearch.h> |
| |
| #include <linux/kernel.h> |
| #include <linux/device.h> |
| #include <linux/module.h> |
| #include <linux/blk-mq.h> |
| #include <linux/fs.h> |
| #include <linux/blkdev.h> |
| #include <linux/slab.h> |
| #include <linux/idr.h> |
| #include <linux/workqueue.h> |
| |
| #include "rbd_types.h" |
| |
| #define RBD_DEBUG /* Activate rbd_assert() calls */ |
| |
| /* |
| * Increment the given counter and return its updated value. |
| * If the counter is already 0 it will not be incremented. |
| * If the counter is already at its maximum value returns |
| * -EINVAL without updating it. |
| */ |
| static int atomic_inc_return_safe(atomic_t *v) |
| { |
| unsigned int counter; |
| |
| counter = (unsigned int)atomic_fetch_add_unless(v, 1, 0); |
| if (counter <= (unsigned int)INT_MAX) |
| return (int)counter; |
| |
| atomic_dec(v); |
| |
| return -EINVAL; |
| } |
| |
| /* Decrement the counter. Return the resulting value, or -EINVAL */ |
| static int atomic_dec_return_safe(atomic_t *v) |
| { |
| int counter; |
| |
| counter = atomic_dec_return(v); |
| if (counter >= 0) |
| return counter; |
| |
| atomic_inc(v); |
| |
| return -EINVAL; |
| } |
| |
| #define RBD_DRV_NAME "rbd" |
| |
| #define RBD_MINORS_PER_MAJOR 256 |
| #define RBD_SINGLE_MAJOR_PART_SHIFT 4 |
| |
| #define RBD_MAX_PARENT_CHAIN_LEN 16 |
| |
| #define RBD_SNAP_DEV_NAME_PREFIX "snap_" |
| #define RBD_MAX_SNAP_NAME_LEN \ |
| (NAME_MAX - (sizeof (RBD_SNAP_DEV_NAME_PREFIX) - 1)) |
| |
| #define RBD_MAX_SNAP_COUNT 510 /* allows max snapc to fit in 4KB */ |
| |
| #define RBD_SNAP_HEAD_NAME "-" |
| |
| #define BAD_SNAP_INDEX U32_MAX /* invalid index into snap array */ |
| |
| /* This allows a single page to hold an image name sent by OSD */ |
| #define RBD_IMAGE_NAME_LEN_MAX (PAGE_SIZE - sizeof (__le32) - 1) |
| #define RBD_IMAGE_ID_LEN_MAX 64 |
| |
| #define RBD_OBJ_PREFIX_LEN_MAX 64 |
| |
| #define RBD_NOTIFY_TIMEOUT 5 /* seconds */ |
| #define RBD_RETRY_DELAY msecs_to_jiffies(1000) |
| |
| /* Feature bits */ |
| |
| #define RBD_FEATURE_LAYERING (1ULL<<0) |
| #define RBD_FEATURE_STRIPINGV2 (1ULL<<1) |
| #define RBD_FEATURE_EXCLUSIVE_LOCK (1ULL<<2) |
| #define RBD_FEATURE_OBJECT_MAP (1ULL<<3) |
| #define RBD_FEATURE_FAST_DIFF (1ULL<<4) |
| #define RBD_FEATURE_DEEP_FLATTEN (1ULL<<5) |
| #define RBD_FEATURE_DATA_POOL (1ULL<<7) |
| #define RBD_FEATURE_OPERATIONS (1ULL<<8) |
| |
| #define RBD_FEATURES_ALL (RBD_FEATURE_LAYERING | \ |
| RBD_FEATURE_STRIPINGV2 | \ |
| RBD_FEATURE_EXCLUSIVE_LOCK | \ |
| RBD_FEATURE_OBJECT_MAP | \ |
| RBD_FEATURE_FAST_DIFF | \ |
| RBD_FEATURE_DEEP_FLATTEN | \ |
| RBD_FEATURE_DATA_POOL | \ |
| RBD_FEATURE_OPERATIONS) |
| |
| /* Features supported by this (client software) implementation. */ |
| |
| #define RBD_FEATURES_SUPPORTED (RBD_FEATURES_ALL) |
| |
| /* |
| * An RBD device name will be "rbd#", where the "rbd" comes from |
| * RBD_DRV_NAME above, and # is a unique integer identifier. |
| */ |
| #define DEV_NAME_LEN 32 |
| |
| /* |
| * block device image metadata (in-memory version) |
| */ |
| struct rbd_image_header { |
| /* These six fields never change for a given rbd image */ |
| char *object_prefix; |
| __u8 obj_order; |
| u64 stripe_unit; |
| u64 stripe_count; |
| s64 data_pool_id; |
| u64 features; /* Might be changeable someday? */ |
| |
| /* The remaining fields need to be updated occasionally */ |
| u64 image_size; |
| struct ceph_snap_context *snapc; |
| char *snap_names; /* format 1 only */ |
| u64 *snap_sizes; /* format 1 only */ |
| }; |
| |
| /* |
| * An rbd image specification. |
| * |
| * The tuple (pool_id, image_id, snap_id) is sufficient to uniquely |
| * identify an image. Each rbd_dev structure includes a pointer to |
| * an rbd_spec structure that encapsulates this identity. |
| * |
| * Each of the id's in an rbd_spec has an associated name. For a |
| * user-mapped image, the names are supplied and the id's associated |
| * with them are looked up. For a layered image, a parent image is |
| * defined by the tuple, and the names are looked up. |
| * |
| * An rbd_dev structure contains a parent_spec pointer which is |
| * non-null if the image it represents is a child in a layered |
| * image. This pointer will refer to the rbd_spec structure used |
| * by the parent rbd_dev for its own identity (i.e., the structure |
| * is shared between the parent and child). |
| * |
| * Since these structures are populated once, during the discovery |
| * phase of image construction, they are effectively immutable so |
| * we make no effort to synchronize access to them. |
| * |
| * Note that code herein does not assume the image name is known (it |
| * could be a null pointer). |
| */ |
| struct rbd_spec { |
| u64 pool_id; |
| const char *pool_name; |
| const char *pool_ns; /* NULL if default, never "" */ |
| |
| const char *image_id; |
| const char *image_name; |
| |
| u64 snap_id; |
| const char *snap_name; |
| |
| struct kref kref; |
| }; |
| |
| /* |
| * an instance of the client. multiple devices may share an rbd client. |
| */ |
| struct rbd_client { |
| struct ceph_client *client; |
| struct kref kref; |
| struct list_head node; |
| }; |
| |
| struct pending_result { |
| int result; /* first nonzero result */ |
| int num_pending; |
| }; |
| |
| struct rbd_img_request; |
| |
| enum obj_request_type { |
| OBJ_REQUEST_NODATA = 1, |
| OBJ_REQUEST_BIO, /* pointer into provided bio (list) */ |
| OBJ_REQUEST_BVECS, /* pointer into provided bio_vec array */ |
| OBJ_REQUEST_OWN_BVECS, /* private bio_vec array, doesn't own pages */ |
| }; |
| |
| enum obj_operation_type { |
| OBJ_OP_READ = 1, |
| OBJ_OP_WRITE, |
| OBJ_OP_DISCARD, |
| OBJ_OP_ZEROOUT, |
| }; |
| |
| #define RBD_OBJ_FLAG_DELETION (1U << 0) |
| #define RBD_OBJ_FLAG_COPYUP_ENABLED (1U << 1) |
| #define RBD_OBJ_FLAG_COPYUP_ZEROS (1U << 2) |
| #define RBD_OBJ_FLAG_MAY_EXIST (1U << 3) |
| #define RBD_OBJ_FLAG_NOOP_FOR_NONEXISTENT (1U << 4) |
| |
| enum rbd_obj_read_state { |
| RBD_OBJ_READ_START = 1, |
| RBD_OBJ_READ_OBJECT, |
| RBD_OBJ_READ_PARENT, |
| }; |
| |
| /* |
| * Writes go through the following state machine to deal with |
| * layering: |
| * |
| * . . . . . RBD_OBJ_WRITE_GUARD. . . . . . . . . . . . . . |
| * . | . |
| * . v . |
| * . RBD_OBJ_WRITE_READ_FROM_PARENT. . . . |
| * . | . . |
| * . v v (deep-copyup . |
| * (image . RBD_OBJ_WRITE_COPYUP_EMPTY_SNAPC . not needed) . |
| * flattened) v | . . |
| * . v . . |
| * . . . .RBD_OBJ_WRITE_COPYUP_OPS. . . . . (copyup . |
| * | not needed) v |
| * v . |
| * done . . . . . . . . . . . . . . . . . . |
| * ^ |
| * | |
| * RBD_OBJ_WRITE_FLAT |
| * |
| * Writes start in RBD_OBJ_WRITE_GUARD or _FLAT, depending on whether |
| * assert_exists guard is needed or not (in some cases it's not needed |
| * even if there is a parent). |
| */ |
| enum rbd_obj_write_state { |
| RBD_OBJ_WRITE_START = 1, |
| RBD_OBJ_WRITE_PRE_OBJECT_MAP, |
| RBD_OBJ_WRITE_OBJECT, |
| __RBD_OBJ_WRITE_COPYUP, |
| RBD_OBJ_WRITE_COPYUP, |
| RBD_OBJ_WRITE_POST_OBJECT_MAP, |
| }; |
| |
| enum rbd_obj_copyup_state { |
| RBD_OBJ_COPYUP_START = 1, |
| RBD_OBJ_COPYUP_READ_PARENT, |
| __RBD_OBJ_COPYUP_OBJECT_MAPS, |
| RBD_OBJ_COPYUP_OBJECT_MAPS, |
| __RBD_OBJ_COPYUP_WRITE_OBJECT, |
| RBD_OBJ_COPYUP_WRITE_OBJECT, |
| }; |
| |
| struct rbd_obj_request { |
| struct ceph_object_extent ex; |
| unsigned int flags; /* RBD_OBJ_FLAG_* */ |
| union { |
| enum rbd_obj_read_state read_state; /* for reads */ |
| enum rbd_obj_write_state write_state; /* for writes */ |
| }; |
| |
| struct rbd_img_request *img_request; |
| struct ceph_file_extent *img_extents; |
| u32 num_img_extents; |
| |
| union { |
| struct ceph_bio_iter bio_pos; |
| struct { |
| struct ceph_bvec_iter bvec_pos; |
| u32 bvec_count; |
| u32 bvec_idx; |
| }; |
| }; |
| |
| enum rbd_obj_copyup_state copyup_state; |
| struct bio_vec *copyup_bvecs; |
| u32 copyup_bvec_count; |
| |
| struct list_head osd_reqs; /* w/ r_private_item */ |
| |
| struct mutex state_mutex; |
| struct pending_result pending; |
| struct kref kref; |
| }; |
| |
| enum img_req_flags { |
| IMG_REQ_CHILD, /* initiator: block = 0, child image = 1 */ |
| IMG_REQ_LAYERED, /* ENOENT handling: normal = 0, layered = 1 */ |
| }; |
| |
| enum rbd_img_state { |
| RBD_IMG_START = 1, |
| RBD_IMG_EXCLUSIVE_LOCK, |
| __RBD_IMG_OBJECT_REQUESTS, |
| RBD_IMG_OBJECT_REQUESTS, |
| }; |
| |
| struct rbd_img_request { |
| struct rbd_device *rbd_dev; |
| enum obj_operation_type op_type; |
| enum obj_request_type data_type; |
| unsigned long flags; |
| enum rbd_img_state state; |
| union { |
| u64 snap_id; /* for reads */ |
| struct ceph_snap_context *snapc; /* for writes */ |
| }; |
| struct rbd_obj_request *obj_request; /* obj req initiator */ |
| |
| struct list_head lock_item; |
| struct list_head object_extents; /* obj_req.ex structs */ |
| |
| struct mutex state_mutex; |
| struct pending_result pending; |
| struct work_struct work; |
| int work_result; |
| }; |
| |
| #define for_each_obj_request(ireq, oreq) \ |
| list_for_each_entry(oreq, &(ireq)->object_extents, ex.oe_item) |
| #define for_each_obj_request_safe(ireq, oreq, n) \ |
| list_for_each_entry_safe(oreq, n, &(ireq)->object_extents, ex.oe_item) |
| |
| enum rbd_watch_state { |
| RBD_WATCH_STATE_UNREGISTERED, |
| RBD_WATCH_STATE_REGISTERED, |
| RBD_WATCH_STATE_ERROR, |
| }; |
| |
| enum rbd_lock_state { |
| RBD_LOCK_STATE_UNLOCKED, |
| RBD_LOCK_STATE_LOCKED, |
| RBD_LOCK_STATE_RELEASING, |
| }; |
| |
| /* WatchNotify::ClientId */ |
| struct rbd_client_id { |
| u64 gid; |
| u64 handle; |
| }; |
| |
| struct rbd_mapping { |
| u64 size; |
| }; |
| |
| /* |
| * a single device |
| */ |
| struct rbd_device { |
| int dev_id; /* blkdev unique id */ |
| |
| int major; /* blkdev assigned major */ |
| int minor; |
| struct gendisk *disk; /* blkdev's gendisk and rq */ |
| |
| u32 image_format; /* Either 1 or 2 */ |
| struct rbd_client *rbd_client; |
| |
| char name[DEV_NAME_LEN]; /* blkdev name, e.g. rbd3 */ |
| |
| spinlock_t lock; /* queue, flags, open_count */ |
| |
| struct rbd_image_header header; |
| unsigned long flags; /* possibly lock protected */ |
| struct rbd_spec *spec; |
| struct rbd_options *opts; |
| char *config_info; /* add{,_single_major} string */ |
| |
| struct ceph_object_id header_oid; |
| struct ceph_object_locator header_oloc; |
| |
| struct ceph_file_layout layout; /* used for all rbd requests */ |
| |
| struct mutex watch_mutex; |
| enum rbd_watch_state watch_state; |
| struct ceph_osd_linger_request *watch_handle; |
| u64 watch_cookie; |
| struct delayed_work watch_dwork; |
| |
| struct rw_semaphore lock_rwsem; |
| enum rbd_lock_state lock_state; |
| char lock_cookie[32]; |
| struct rbd_client_id owner_cid; |
| struct work_struct acquired_lock_work; |
| struct work_struct released_lock_work; |
| struct delayed_work lock_dwork; |
| struct work_struct unlock_work; |
| spinlock_t lock_lists_lock; |
| struct list_head acquiring_list; |
| struct list_head running_list; |
| struct completion acquire_wait; |
| int acquire_err; |
| struct completion releasing_wait; |
| |
| spinlock_t object_map_lock; |
| u8 *object_map; |
| u64 object_map_size; /* in objects */ |
| u64 object_map_flags; |
| |
| struct workqueue_struct *task_wq; |
| |
| struct rbd_spec *parent_spec; |
| u64 parent_overlap; |
| atomic_t parent_ref; |
| struct rbd_device *parent; |
| |
| /* Block layer tags. */ |
| struct blk_mq_tag_set tag_set; |
| |
| /* protects updating the header */ |
| struct rw_semaphore header_rwsem; |
| |
| struct rbd_mapping mapping; |
| |
| struct list_head node; |
| |
| /* sysfs related */ |
| struct device dev; |
| unsigned long open_count; /* protected by lock */ |
| }; |
| |
| /* |
| * Flag bits for rbd_dev->flags: |
| * - REMOVING (which is coupled with rbd_dev->open_count) is protected |
| * by rbd_dev->lock |
| */ |
| enum rbd_dev_flags { |
| RBD_DEV_FLAG_EXISTS, /* rbd_dev_device_setup() ran */ |
| RBD_DEV_FLAG_REMOVING, /* this mapping is being removed */ |
| RBD_DEV_FLAG_READONLY, /* -o ro or snapshot */ |
| }; |
| |
| static DEFINE_MUTEX(client_mutex); /* Serialize client creation */ |
| |
| static LIST_HEAD(rbd_dev_list); /* devices */ |
| static DEFINE_SPINLOCK(rbd_dev_list_lock); |
| |
| static LIST_HEAD(rbd_client_list); /* clients */ |
| static DEFINE_SPINLOCK(rbd_client_list_lock); |
| |
| /* Slab caches for frequently-allocated structures */ |
| |
| static struct kmem_cache *rbd_img_request_cache; |
| static struct kmem_cache *rbd_obj_request_cache; |
| |
| static int rbd_major; |
| static DEFINE_IDA(rbd_dev_id_ida); |
| |
| static struct workqueue_struct *rbd_wq; |
| |
| static struct ceph_snap_context rbd_empty_snapc = { |
| .nref = REFCOUNT_INIT(1), |
| }; |
| |
| /* |
| * single-major requires >= 0.75 version of userspace rbd utility. |
| */ |
| static bool single_major = true; |
| module_param(single_major, bool, 0444); |
| MODULE_PARM_DESC(single_major, "Use a single major number for all rbd devices (default: true)"); |
| |
| static ssize_t add_store(struct bus_type *bus, const char *buf, size_t count); |
| static ssize_t remove_store(struct bus_type *bus, const char *buf, |
| size_t count); |
| static ssize_t add_single_major_store(struct bus_type *bus, const char *buf, |
| size_t count); |
| static ssize_t remove_single_major_store(struct bus_type *bus, const char *buf, |
| size_t count); |
| static int rbd_dev_image_probe(struct rbd_device *rbd_dev, int depth); |
| |
| static int rbd_dev_id_to_minor(int dev_id) |
| { |
| return dev_id << RBD_SINGLE_MAJOR_PART_SHIFT; |
| } |
| |
| static int minor_to_rbd_dev_id(int minor) |
| { |
| return minor >> RBD_SINGLE_MAJOR_PART_SHIFT; |
| } |
| |
| static bool rbd_is_ro(struct rbd_device *rbd_dev) |
| { |
| return test_bit(RBD_DEV_FLAG_READONLY, &rbd_dev->flags); |
| } |
| |
| static bool rbd_is_snap(struct rbd_device *rbd_dev) |
| { |
| return rbd_dev->spec->snap_id != CEPH_NOSNAP; |
| } |
| |
| static bool __rbd_is_lock_owner(struct rbd_device *rbd_dev) |
| { |
| lockdep_assert_held(&rbd_dev->lock_rwsem); |
| |
| return rbd_dev->lock_state == RBD_LOCK_STATE_LOCKED || |
| rbd_dev->lock_state == RBD_LOCK_STATE_RELEASING; |
| } |
| |
| static bool rbd_is_lock_owner(struct rbd_device *rbd_dev) |
| { |
| bool is_lock_owner; |
| |
| down_read(&rbd_dev->lock_rwsem); |
| is_lock_owner = __rbd_is_lock_owner(rbd_dev); |
| up_read(&rbd_dev->lock_rwsem); |
| return is_lock_owner; |
| } |
| |
| static ssize_t supported_features_show(struct bus_type *bus, char *buf) |
| { |
| return sprintf(buf, "0x%llx\n", RBD_FEATURES_SUPPORTED); |
| } |
| |
| static BUS_ATTR_WO(add); |
| static BUS_ATTR_WO(remove); |
| static BUS_ATTR_WO(add_single_major); |
| static BUS_ATTR_WO(remove_single_major); |
| static BUS_ATTR_RO(supported_features); |
| |
| static struct attribute *rbd_bus_attrs[] = { |
| &bus_attr_add.attr, |
| &bus_attr_remove.attr, |
| &bus_attr_add_single_major.attr, |
| &bus_attr_remove_single_major.attr, |
| &bus_attr_supported_features.attr, |
| NULL, |
| }; |
| |
| static umode_t rbd_bus_is_visible(struct kobject *kobj, |
| struct attribute *attr, int index) |
| { |
| if (!single_major && |
| (attr == &bus_attr_add_single_major.attr || |
| attr == &bus_attr_remove_single_major.attr)) |
| return 0; |
| |
| return attr->mode; |
| } |
| |
| static const struct attribute_group rbd_bus_group = { |
| .attrs = rbd_bus_attrs, |
| .is_visible = rbd_bus_is_visible, |
| }; |
| __ATTRIBUTE_GROUPS(rbd_bus); |
| |
| static struct bus_type rbd_bus_type = { |
| .name = "rbd", |
| .bus_groups = rbd_bus_groups, |
| }; |
| |
| static void rbd_root_dev_release(struct device *dev) |
| { |
| } |
| |
| static struct device rbd_root_dev = { |
| .init_name = "rbd", |
| .release = rbd_root_dev_release, |
| }; |
| |
| static __printf(2, 3) |
| void rbd_warn(struct rbd_device *rbd_dev, const char *fmt, ...) |
| { |
| struct va_format vaf; |
| va_list args; |
| |
| va_start(args, fmt); |
| vaf.fmt = fmt; |
| vaf.va = &args; |
| |
| if (!rbd_dev) |
| printk(KERN_WARNING "%s: %pV\n", RBD_DRV_NAME, &vaf); |
| else if (rbd_dev->disk) |
| printk(KERN_WARNING "%s: %s: %pV\n", |
| RBD_DRV_NAME, rbd_dev->disk->disk_name, &vaf); |
| else if (rbd_dev->spec && rbd_dev->spec->image_name) |
| printk(KERN_WARNING "%s: image %s: %pV\n", |
| RBD_DRV_NAME, rbd_dev->spec->image_name, &vaf); |
| else if (rbd_dev->spec && rbd_dev->spec->image_id) |
| printk(KERN_WARNING "%s: id %s: %pV\n", |
| RBD_DRV_NAME, rbd_dev->spec->image_id, &vaf); |
| else /* punt */ |
| printk(KERN_WARNING "%s: rbd_dev %p: %pV\n", |
| RBD_DRV_NAME, rbd_dev, &vaf); |
| va_end(args); |
| } |
| |
| #ifdef RBD_DEBUG |
| #define rbd_assert(expr) \ |
| if (unlikely(!(expr))) { \ |
| printk(KERN_ERR "\nAssertion failure in %s() " \ |
| "at line %d:\n\n" \ |
| "\trbd_assert(%s);\n\n", \ |
| __func__, __LINE__, #expr); \ |
| BUG(); \ |
| } |
| #else /* !RBD_DEBUG */ |
| # define rbd_assert(expr) ((void) 0) |
| #endif /* !RBD_DEBUG */ |
| |
| static void rbd_dev_remove_parent(struct rbd_device *rbd_dev); |
| |
| static int rbd_dev_refresh(struct rbd_device *rbd_dev); |
| static int rbd_dev_v2_header_onetime(struct rbd_device *rbd_dev); |
| static int rbd_dev_header_info(struct rbd_device *rbd_dev); |
| static int rbd_dev_v2_parent_info(struct rbd_device *rbd_dev); |
| static const char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev, |
| u64 snap_id); |
| static int _rbd_dev_v2_snap_size(struct rbd_device *rbd_dev, u64 snap_id, |
| u8 *order, u64 *snap_size); |
| static int rbd_dev_v2_get_flags(struct rbd_device *rbd_dev); |
| |
| static void rbd_obj_handle_request(struct rbd_obj_request *obj_req, int result); |
| static void rbd_img_handle_request(struct rbd_img_request *img_req, int result); |
| |
| /* |
| * Return true if nothing else is pending. |
| */ |
| static bool pending_result_dec(struct pending_result *pending, int *result) |
| { |
| rbd_assert(pending->num_pending > 0); |
| |
| if (*result && !pending->result) |
| pending->result = *result; |
| if (--pending->num_pending) |
| return false; |
| |
| *result = pending->result; |
| return true; |
| } |
| |
| static int rbd_open(struct block_device *bdev, fmode_t mode) |
| { |
| struct rbd_device *rbd_dev = bdev->bd_disk->private_data; |
| bool removing = false; |
| |
| spin_lock_irq(&rbd_dev->lock); |
| if (test_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags)) |
| removing = true; |
| else |
| rbd_dev->open_count++; |
| spin_unlock_irq(&rbd_dev->lock); |
| if (removing) |
| return -ENOENT; |
| |
| (void) get_device(&rbd_dev->dev); |
| |
| return 0; |
| } |
| |
| static void rbd_release(struct gendisk *disk, fmode_t mode) |
| { |
| struct rbd_device *rbd_dev = disk->private_data; |
| unsigned long open_count_before; |
| |
| spin_lock_irq(&rbd_dev->lock); |
| open_count_before = rbd_dev->open_count--; |
| spin_unlock_irq(&rbd_dev->lock); |
| rbd_assert(open_count_before > 0); |
| |
| put_device(&rbd_dev->dev); |
| } |
| |
| static int rbd_ioctl_set_ro(struct rbd_device *rbd_dev, unsigned long arg) |
| { |
| int ro; |
| |
| if (get_user(ro, (int __user *)arg)) |
| return -EFAULT; |
| |
| /* |
| * Both images mapped read-only and snapshots can't be marked |
| * read-write. |
| */ |
| if (!ro) { |
| if (rbd_is_ro(rbd_dev)) |
| return -EROFS; |
| |
| rbd_assert(!rbd_is_snap(rbd_dev)); |
| } |
| |
| /* Let blkdev_roset() handle it */ |
| return -ENOTTY; |
| } |
| |
| static int rbd_ioctl(struct block_device *bdev, fmode_t mode, |
| unsigned int cmd, unsigned long arg) |
| { |
| struct rbd_device *rbd_dev = bdev->bd_disk->private_data; |
| int ret; |
| |
| switch (cmd) { |
| case BLKROSET: |
| ret = rbd_ioctl_set_ro(rbd_dev, arg); |
| break; |
| default: |
| ret = -ENOTTY; |
| } |
| |
| return ret; |
| } |
| |
| #ifdef CONFIG_COMPAT |
| static int rbd_compat_ioctl(struct block_device *bdev, fmode_t mode, |
| unsigned int cmd, unsigned long arg) |
| { |
| return rbd_ioctl(bdev, mode, cmd, arg); |
| } |
| #endif /* CONFIG_COMPAT */ |
| |
| static const struct block_device_operations rbd_bd_ops = { |
| .owner = THIS_MODULE, |
| .open = rbd_open, |
| .release = rbd_release, |
| .ioctl = rbd_ioctl, |
| #ifdef CONFIG_COMPAT |
| .compat_ioctl = rbd_compat_ioctl, |
| #endif |
| }; |
| |
| /* |
| * Initialize an rbd client instance. Success or not, this function |
| * consumes ceph_opts. Caller holds client_mutex. |
| */ |
| static struct rbd_client *rbd_client_create(struct ceph_options *ceph_opts) |
| { |
| struct rbd_client *rbdc; |
| int ret = -ENOMEM; |
| |
| dout("%s:\n", __func__); |
| rbdc = kmalloc(sizeof(struct rbd_client), GFP_KERNEL); |
| if (!rbdc) |
| goto out_opt; |
| |
| kref_init(&rbdc->kref); |
| INIT_LIST_HEAD(&rbdc->node); |
| |
| rbdc->client = ceph_create_client(ceph_opts, rbdc); |
| if (IS_ERR(rbdc->client)) |
| goto out_rbdc; |
| ceph_opts = NULL; /* Now rbdc->client is responsible for ceph_opts */ |
| |
| ret = ceph_open_session(rbdc->client); |
| if (ret < 0) |
| goto out_client; |
| |
| spin_lock(&rbd_client_list_lock); |
| list_add_tail(&rbdc->node, &rbd_client_list); |
| spin_unlock(&rbd_client_list_lock); |
| |
| dout("%s: rbdc %p\n", __func__, rbdc); |
| |
| return rbdc; |
| out_client: |
| ceph_destroy_client(rbdc->client); |
| out_rbdc: |
| kfree(rbdc); |
| out_opt: |
| if (ceph_opts) |
| ceph_destroy_options(ceph_opts); |
| dout("%s: error %d\n", __func__, ret); |
| |
| return ERR_PTR(ret); |
| } |
| |
| static struct rbd_client *__rbd_get_client(struct rbd_client *rbdc) |
| { |
| kref_get(&rbdc->kref); |
| |
| return rbdc; |
| } |
| |
| /* |
| * Find a ceph client with specific addr and configuration. If |
| * found, bump its reference count. |
| */ |
| static struct rbd_client *rbd_client_find(struct ceph_options *ceph_opts) |
| { |
| struct rbd_client *client_node; |
| bool found = false; |
| |
| if (ceph_opts->flags & CEPH_OPT_NOSHARE) |
| return NULL; |
| |
| spin_lock(&rbd_client_list_lock); |
| list_for_each_entry(client_node, &rbd_client_list, node) { |
| if (!ceph_compare_options(ceph_opts, client_node->client)) { |
| __rbd_get_client(client_node); |
| |
| found = true; |
| break; |
| } |
| } |
| spin_unlock(&rbd_client_list_lock); |
| |
| return found ? client_node : NULL; |
| } |
| |
| /* |
| * (Per device) rbd map options |
| */ |
| enum { |
| Opt_queue_depth, |
| Opt_alloc_size, |
| Opt_lock_timeout, |
| /* int args above */ |
| Opt_pool_ns, |
| Opt_compression_hint, |
| /* string args above */ |
| Opt_read_only, |
| Opt_read_write, |
| Opt_lock_on_read, |
| Opt_exclusive, |
| Opt_notrim, |
| }; |
| |
| enum { |
| Opt_compression_hint_none, |
| Opt_compression_hint_compressible, |
| Opt_compression_hint_incompressible, |
| }; |
| |
| static const struct constant_table rbd_param_compression_hint[] = { |
| {"none", Opt_compression_hint_none}, |
| {"compressible", Opt_compression_hint_compressible}, |
| {"incompressible", Opt_compression_hint_incompressible}, |
| {} |
| }; |
| |
| static const struct fs_parameter_spec rbd_parameters[] = { |
| fsparam_u32 ("alloc_size", Opt_alloc_size), |
| fsparam_enum ("compression_hint", Opt_compression_hint, |
| rbd_param_compression_hint), |
| fsparam_flag ("exclusive", Opt_exclusive), |
| fsparam_flag ("lock_on_read", Opt_lock_on_read), |
| fsparam_u32 ("lock_timeout", Opt_lock_timeout), |
| fsparam_flag ("notrim", Opt_notrim), |
| fsparam_string ("_pool_ns", Opt_pool_ns), |
| fsparam_u32 ("queue_depth", Opt_queue_depth), |
| fsparam_flag ("read_only", Opt_read_only), |
| fsparam_flag ("read_write", Opt_read_write), |
| fsparam_flag ("ro", Opt_read_only), |
| fsparam_flag ("rw", Opt_read_write), |
| {} |
| }; |
| |
| struct rbd_options { |
| int queue_depth; |
| int alloc_size; |
| unsigned long lock_timeout; |
| bool read_only; |
| bool lock_on_read; |
| bool exclusive; |
| bool trim; |
| |
| u32 alloc_hint_flags; /* CEPH_OSD_OP_ALLOC_HINT_FLAG_* */ |
| }; |
| |
| #define RBD_QUEUE_DEPTH_DEFAULT BLKDEV_MAX_RQ |
| #define RBD_ALLOC_SIZE_DEFAULT (64 * 1024) |
| #define RBD_LOCK_TIMEOUT_DEFAULT 0 /* no timeout */ |
| #define RBD_READ_ONLY_DEFAULT false |
| #define RBD_LOCK_ON_READ_DEFAULT false |
| #define RBD_EXCLUSIVE_DEFAULT false |
| #define RBD_TRIM_DEFAULT true |
| |
| struct rbd_parse_opts_ctx { |
| struct rbd_spec *spec; |
| struct ceph_options *copts; |
| struct rbd_options *opts; |
| }; |
| |
| static char* obj_op_name(enum obj_operation_type op_type) |
| { |
| switch (op_type) { |
| case OBJ_OP_READ: |
| return "read"; |
| case OBJ_OP_WRITE: |
| return "write"; |
| case OBJ_OP_DISCARD: |
| return "discard"; |
| case OBJ_OP_ZEROOUT: |
| return "zeroout"; |
| default: |
| return "???"; |
| } |
| } |
| |
| /* |
| * Destroy ceph client |
| * |
| * Caller must hold rbd_client_list_lock. |
| */ |
| static void rbd_client_release(struct kref *kref) |
| { |
| struct rbd_client *rbdc = container_of(kref, struct rbd_client, kref); |
| |
| dout("%s: rbdc %p\n", __func__, rbdc); |
| spin_lock(&rbd_client_list_lock); |
| list_del(&rbdc->node); |
| spin_unlock(&rbd_client_list_lock); |
| |
| ceph_destroy_client(rbdc->client); |
| kfree(rbdc); |
| } |
| |
| /* |
| * Drop reference to ceph client node. If it's not referenced anymore, release |
| * it. |
| */ |
| static void rbd_put_client(struct rbd_client *rbdc) |
| { |
| if (rbdc) |
| kref_put(&rbdc->kref, rbd_client_release); |
| } |
| |
| /* |
| * Get a ceph client with specific addr and configuration, if one does |
| * not exist create it. Either way, ceph_opts is consumed by this |
| * function. |
| */ |
| static struct rbd_client *rbd_get_client(struct ceph_options *ceph_opts) |
| { |
| struct rbd_client *rbdc; |
| int ret; |
| |
| mutex_lock(&client_mutex); |
| rbdc = rbd_client_find(ceph_opts); |
| if (rbdc) { |
| ceph_destroy_options(ceph_opts); |
| |
| /* |
| * Using an existing client. Make sure ->pg_pools is up to |
| * date before we look up the pool id in do_rbd_add(). |
| */ |
| ret = ceph_wait_for_latest_osdmap(rbdc->client, |
| rbdc->client->options->mount_timeout); |
| if (ret) { |
| rbd_warn(NULL, "failed to get latest osdmap: %d", ret); |
| rbd_put_client(rbdc); |
| rbdc = ERR_PTR(ret); |
| } |
| } else { |
| rbdc = rbd_client_create(ceph_opts); |
| } |
| mutex_unlock(&client_mutex); |
| |
| return rbdc; |
| } |
| |
| static bool rbd_image_format_valid(u32 image_format) |
| { |
| return image_format == 1 || image_format == 2; |
| } |
| |
| static bool rbd_dev_ondisk_valid(struct rbd_image_header_ondisk *ondisk) |
| { |
| size_t size; |
| u32 snap_count; |
| |
| /* The header has to start with the magic rbd header text */ |
| if (memcmp(&ondisk->text, RBD_HEADER_TEXT, sizeof (RBD_HEADER_TEXT))) |
| return false; |
| |
| /* The bio layer requires at least sector-sized I/O */ |
| |
| if (ondisk->options.order < SECTOR_SHIFT) |
| return false; |
| |
| /* If we use u64 in a few spots we may be able to loosen this */ |
| |
| if (ondisk->options.order > 8 * sizeof (int) - 1) |
| return false; |
| |
| /* |
| * The size of a snapshot header has to fit in a size_t, and |
| * that limits the number of snapshots. |
| */ |
| snap_count = le32_to_cpu(ondisk->snap_count); |
| size = SIZE_MAX - sizeof (struct ceph_snap_context); |
| if (snap_count > size / sizeof (__le64)) |
| return false; |
| |
| /* |
| * Not only that, but the size of the entire the snapshot |
| * header must also be representable in a size_t. |
| */ |
| size -= snap_count * sizeof (__le64); |
| if ((u64) size < le64_to_cpu(ondisk->snap_names_len)) |
| return false; |
| |
| return true; |
| } |
| |
| /* |
| * returns the size of an object in the image |
| */ |
| static u32 rbd_obj_bytes(struct rbd_image_header *header) |
| { |
| return 1U << header->obj_order; |
| } |
| |
| static void rbd_init_layout(struct rbd_device *rbd_dev) |
| { |
| if (rbd_dev->header.stripe_unit == 0 || |
| rbd_dev->header.stripe_count == 0) { |
| rbd_dev->header.stripe_unit = rbd_obj_bytes(&rbd_dev->header); |
| rbd_dev->header.stripe_count = 1; |
| } |
| |
| rbd_dev->layout.stripe_unit = rbd_dev->header.stripe_unit; |
| rbd_dev->layout.stripe_count = rbd_dev->header.stripe_count; |
| rbd_dev->layout.object_size = rbd_obj_bytes(&rbd_dev->header); |
| rbd_dev->layout.pool_id = rbd_dev->header.data_pool_id == CEPH_NOPOOL ? |
| rbd_dev->spec->pool_id : rbd_dev->header.data_pool_id; |
| RCU_INIT_POINTER(rbd_dev->layout.pool_ns, NULL); |
| } |
| |
| /* |
| * Fill an rbd image header with information from the given format 1 |
| * on-disk header. |
| */ |
| static int rbd_header_from_disk(struct rbd_device *rbd_dev, |
| struct rbd_image_header_ondisk *ondisk) |
| { |
| struct rbd_image_header *header = &rbd_dev->header; |
| bool first_time = header->object_prefix == NULL; |
| struct ceph_snap_context *snapc; |
| char *object_prefix = NULL; |
| char *snap_names = NULL; |
| u64 *snap_sizes = NULL; |
| u32 snap_count; |
| int ret = -ENOMEM; |
| u32 i; |
| |
| /* Allocate this now to avoid having to handle failure below */ |
| |
| if (first_time) { |
| object_prefix = kstrndup(ondisk->object_prefix, |
| sizeof(ondisk->object_prefix), |
| GFP_KERNEL); |
| if (!object_prefix) |
| return -ENOMEM; |
| } |
| |
| /* Allocate the snapshot context and fill it in */ |
| |
| snap_count = le32_to_cpu(ondisk->snap_count); |
| snapc = ceph_create_snap_context(snap_count, GFP_KERNEL); |
| if (!snapc) |
| goto out_err; |
| snapc->seq = le64_to_cpu(ondisk->snap_seq); |
| if (snap_count) { |
| struct rbd_image_snap_ondisk *snaps; |
| u64 snap_names_len = le64_to_cpu(ondisk->snap_names_len); |
| |
| /* We'll keep a copy of the snapshot names... */ |
| |
| if (snap_names_len > (u64)SIZE_MAX) |
| goto out_2big; |
| snap_names = kmalloc(snap_names_len, GFP_KERNEL); |
| if (!snap_names) |
| goto out_err; |
| |
| /* ...as well as the array of their sizes. */ |
| snap_sizes = kmalloc_array(snap_count, |
| sizeof(*header->snap_sizes), |
| GFP_KERNEL); |
| if (!snap_sizes) |
| goto out_err; |
| |
| /* |
| * Copy the names, and fill in each snapshot's id |
| * and size. |
| * |
| * Note that rbd_dev_v1_header_info() guarantees the |
| * ondisk buffer we're working with has |
| * snap_names_len bytes beyond the end of the |
| * snapshot id array, this memcpy() is safe. |
| */ |
| memcpy(snap_names, &ondisk->snaps[snap_count], snap_names_len); |
| snaps = ondisk->snaps; |
| for (i = 0; i < snap_count; i++) { |
| snapc->snaps[i] = le64_to_cpu(snaps[i].id); |
| snap_sizes[i] = le64_to_cpu(snaps[i].image_size); |
| } |
| } |
| |
| /* We won't fail any more, fill in the header */ |
| |
| if (first_time) { |
| header->object_prefix = object_prefix; |
| header->obj_order = ondisk->options.order; |
| rbd_init_layout(rbd_dev); |
| } else { |
| ceph_put_snap_context(header->snapc); |
| kfree(header->snap_names); |
| kfree(header->snap_sizes); |
| } |
| |
| /* The remaining fields always get updated (when we refresh) */ |
| |
| header->image_size = le64_to_cpu(ondisk->image_size); |
| header->snapc = snapc; |
| header->snap_names = snap_names; |
| header->snap_sizes = snap_sizes; |
| |
| return 0; |
| out_2big: |
| ret = -EIO; |
| out_err: |
| kfree(snap_sizes); |
| kfree(snap_names); |
| ceph_put_snap_context(snapc); |
| kfree(object_prefix); |
| |
| return ret; |
| } |
| |
| static const char *_rbd_dev_v1_snap_name(struct rbd_device *rbd_dev, u32 which) |
| { |
| const char *snap_name; |
| |
| rbd_assert(which < rbd_dev->header.snapc->num_snaps); |
| |
| /* Skip over names until we find the one we are looking for */ |
| |
| snap_name = rbd_dev->header.snap_names; |
| while (which--) |
| snap_name += strlen(snap_name) + 1; |
| |
| return kstrdup(snap_name, GFP_KERNEL); |
| } |
| |
| /* |
| * Snapshot id comparison function for use with qsort()/bsearch(). |
| * Note that result is for snapshots in *descending* order. |
| */ |
| static int snapid_compare_reverse(const void *s1, const void *s2) |
| { |
| u64 snap_id1 = *(u64 *)s1; |
| u64 snap_id2 = *(u64 *)s2; |
| |
| if (snap_id1 < snap_id2) |
| return 1; |
| return snap_id1 == snap_id2 ? 0 : -1; |
| } |
| |
| /* |
| * Search a snapshot context to see if the given snapshot id is |
| * present. |
| * |
| * Returns the position of the snapshot id in the array if it's found, |
| * or BAD_SNAP_INDEX otherwise. |
| * |
| * Note: The snapshot array is in kept sorted (by the osd) in |
| * reverse order, highest snapshot id first. |
| */ |
| static u32 rbd_dev_snap_index(struct rbd_device *rbd_dev, u64 snap_id) |
| { |
| struct ceph_snap_context *snapc = rbd_dev->header.snapc; |
| u64 *found; |
| |
| found = bsearch(&snap_id, &snapc->snaps, snapc->num_snaps, |
| sizeof (snap_id), snapid_compare_reverse); |
| |
| return found ? (u32)(found - &snapc->snaps[0]) : BAD_SNAP_INDEX; |
| } |
| |
| static const char *rbd_dev_v1_snap_name(struct rbd_device *rbd_dev, |
| u64 snap_id) |
| { |
| u32 which; |
| const char *snap_name; |
| |
| which = rbd_dev_snap_index(rbd_dev, snap_id); |
| if (which == BAD_SNAP_INDEX) |
| return ERR_PTR(-ENOENT); |
| |
| snap_name = _rbd_dev_v1_snap_name(rbd_dev, which); |
| return snap_name ? snap_name : ERR_PTR(-ENOMEM); |
| } |
| |
| static const char *rbd_snap_name(struct rbd_device *rbd_dev, u64 snap_id) |
| { |
| if (snap_id == CEPH_NOSNAP) |
| return RBD_SNAP_HEAD_NAME; |
| |
| rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); |
| if (rbd_dev->image_format == 1) |
| return rbd_dev_v1_snap_name(rbd_dev, snap_id); |
| |
| return rbd_dev_v2_snap_name(rbd_dev, snap_id); |
| } |
| |
| static int rbd_snap_size(struct rbd_device *rbd_dev, u64 snap_id, |
| u64 *snap_size) |
| { |
| rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); |
| if (snap_id == CEPH_NOSNAP) { |
| *snap_size = rbd_dev->header.image_size; |
| } else if (rbd_dev->image_format == 1) { |
| u32 which; |
| |
| which = rbd_dev_snap_index(rbd_dev, snap_id); |
| if (which == BAD_SNAP_INDEX) |
| return -ENOENT; |
| |
| *snap_size = rbd_dev->header.snap_sizes[which]; |
| } else { |
| u64 size = 0; |
| int ret; |
| |
| ret = _rbd_dev_v2_snap_size(rbd_dev, snap_id, NULL, &size); |
| if (ret) |
| return ret; |
| |
| *snap_size = size; |
| } |
| return 0; |
| } |
| |
| static int rbd_dev_mapping_set(struct rbd_device *rbd_dev) |
| { |
| u64 snap_id = rbd_dev->spec->snap_id; |
| u64 size = 0; |
| int ret; |
| |
| ret = rbd_snap_size(rbd_dev, snap_id, &size); |
| if (ret) |
| return ret; |
| |
| rbd_dev->mapping.size = size; |
| return 0; |
| } |
| |
| static void rbd_dev_mapping_clear(struct rbd_device *rbd_dev) |
| { |
| rbd_dev->mapping.size = 0; |
| } |
| |
| static void zero_bvec(struct bio_vec *bv) |
| { |
| void *buf; |
| unsigned long flags; |
| |
| buf = bvec_kmap_irq(bv, &flags); |
| memset(buf, 0, bv->bv_len); |
| flush_dcache_page(bv->bv_page); |
| bvec_kunmap_irq(buf, &flags); |
| } |
| |
| static void zero_bios(struct ceph_bio_iter *bio_pos, u32 off, u32 bytes) |
| { |
| struct ceph_bio_iter it = *bio_pos; |
| |
| ceph_bio_iter_advance(&it, off); |
| ceph_bio_iter_advance_step(&it, bytes, ({ |
| zero_bvec(&bv); |
| })); |
| } |
| |
| static void zero_bvecs(struct ceph_bvec_iter *bvec_pos, u32 off, u32 bytes) |
| { |
| struct ceph_bvec_iter it = *bvec_pos; |
| |
| ceph_bvec_iter_advance(&it, off); |
| ceph_bvec_iter_advance_step(&it, bytes, ({ |
| zero_bvec(&bv); |
| })); |
| } |
| |
| /* |
| * Zero a range in @obj_req data buffer defined by a bio (list) or |
| * (private) bio_vec array. |
| * |
| * @off is relative to the start of the data buffer. |
| */ |
| static void rbd_obj_zero_range(struct rbd_obj_request *obj_req, u32 off, |
| u32 bytes) |
| { |
| dout("%s %p data buf %u~%u\n", __func__, obj_req, off, bytes); |
| |
| switch (obj_req->img_request->data_type) { |
| case OBJ_REQUEST_BIO: |
| zero_bios(&obj_req->bio_pos, off, bytes); |
| break; |
| case OBJ_REQUEST_BVECS: |
| case OBJ_REQUEST_OWN_BVECS: |
| zero_bvecs(&obj_req->bvec_pos, off, bytes); |
| break; |
| default: |
| BUG(); |
| } |
| } |
| |
| static void rbd_obj_request_destroy(struct kref *kref); |
| static void rbd_obj_request_put(struct rbd_obj_request *obj_request) |
| { |
| rbd_assert(obj_request != NULL); |
| dout("%s: obj %p (was %d)\n", __func__, obj_request, |
| kref_read(&obj_request->kref)); |
| kref_put(&obj_request->kref, rbd_obj_request_destroy); |
| } |
| |
| static inline void rbd_img_obj_request_add(struct rbd_img_request *img_request, |
| struct rbd_obj_request *obj_request) |
| { |
| rbd_assert(obj_request->img_request == NULL); |
| |
| /* Image request now owns object's original reference */ |
| obj_request->img_request = img_request; |
| dout("%s: img %p obj %p\n", __func__, img_request, obj_request); |
| } |
| |
| static inline void rbd_img_obj_request_del(struct rbd_img_request *img_request, |
| struct rbd_obj_request *obj_request) |
| { |
| dout("%s: img %p obj %p\n", __func__, img_request, obj_request); |
| list_del(&obj_request->ex.oe_item); |
| rbd_assert(obj_request->img_request == img_request); |
| rbd_obj_request_put(obj_request); |
| } |
| |
| static void rbd_osd_submit(struct ceph_osd_request *osd_req) |
| { |
| struct rbd_obj_request *obj_req = osd_req->r_priv; |
| |
| dout("%s osd_req %p for obj_req %p objno %llu %llu~%llu\n", |
| __func__, osd_req, obj_req, obj_req->ex.oe_objno, |
| obj_req->ex.oe_off, obj_req->ex.oe_len); |
| ceph_osdc_start_request(osd_req->r_osdc, osd_req, false); |
| } |
| |
| /* |
| * The default/initial value for all image request flags is 0. Each |
| * is conditionally set to 1 at image request initialization time |
| * and currently never change thereafter. |
| */ |
| static void img_request_layered_set(struct rbd_img_request *img_request) |
| { |
| set_bit(IMG_REQ_LAYERED, &img_request->flags); |
| } |
| |
| static bool img_request_layered_test(struct rbd_img_request *img_request) |
| { |
| return test_bit(IMG_REQ_LAYERED, &img_request->flags) != 0; |
| } |
| |
| static bool rbd_obj_is_entire(struct rbd_obj_request *obj_req) |
| { |
| struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev; |
| |
| return !obj_req->ex.oe_off && |
| obj_req->ex.oe_len == rbd_dev->layout.object_size; |
| } |
| |
| static bool rbd_obj_is_tail(struct rbd_obj_request *obj_req) |
| { |
| struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev; |
| |
| return obj_req->ex.oe_off + obj_req->ex.oe_len == |
| rbd_dev->layout.object_size; |
| } |
| |
| /* |
| * Must be called after rbd_obj_calc_img_extents(). |
| */ |
| static bool rbd_obj_copyup_enabled(struct rbd_obj_request *obj_req) |
| { |
| if (!obj_req->num_img_extents || |
| (rbd_obj_is_entire(obj_req) && |
| !obj_req->img_request->snapc->num_snaps)) |
| return false; |
| |
| return true; |
| } |
| |
| static u64 rbd_obj_img_extents_bytes(struct rbd_obj_request *obj_req) |
| { |
| return ceph_file_extents_bytes(obj_req->img_extents, |
| obj_req->num_img_extents); |
| } |
| |
| static bool rbd_img_is_write(struct rbd_img_request *img_req) |
| { |
| switch (img_req->op_type) { |
| case OBJ_OP_READ: |
| return false; |
| case OBJ_OP_WRITE: |
| case OBJ_OP_DISCARD: |
| case OBJ_OP_ZEROOUT: |
| return true; |
| default: |
| BUG(); |
| } |
| } |
| |
| static void rbd_osd_req_callback(struct ceph_osd_request *osd_req) |
| { |
| struct rbd_obj_request *obj_req = osd_req->r_priv; |
| int result; |
| |
| dout("%s osd_req %p result %d for obj_req %p\n", __func__, osd_req, |
| osd_req->r_result, obj_req); |
| |
| /* |
| * Writes aren't allowed to return a data payload. In some |
| * guarded write cases (e.g. stat + zero on an empty object) |
| * a stat response makes it through, but we don't care. |
| */ |
| if (osd_req->r_result > 0 && rbd_img_is_write(obj_req->img_request)) |
| result = 0; |
| else |
| result = osd_req->r_result; |
| |
| rbd_obj_handle_request(obj_req, result); |
| } |
| |
| static void rbd_osd_format_read(struct ceph_osd_request *osd_req) |
| { |
| struct rbd_obj_request *obj_request = osd_req->r_priv; |
| |
| osd_req->r_flags = CEPH_OSD_FLAG_READ; |
| osd_req->r_snapid = obj_request->img_request->snap_id; |
| } |
| |
| static void rbd_osd_format_write(struct ceph_osd_request *osd_req) |
| { |
| struct rbd_obj_request *obj_request = osd_req->r_priv; |
| |
| osd_req->r_flags = CEPH_OSD_FLAG_WRITE; |
| ktime_get_real_ts64(&osd_req->r_mtime); |
| osd_req->r_data_offset = obj_request->ex.oe_off; |
| } |
| |
| static struct ceph_osd_request * |
| __rbd_obj_add_osd_request(struct rbd_obj_request *obj_req, |
| struct ceph_snap_context *snapc, int num_ops) |
| { |
| struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev; |
| struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
| struct ceph_osd_request *req; |
| const char *name_format = rbd_dev->image_format == 1 ? |
| RBD_V1_DATA_FORMAT : RBD_V2_DATA_FORMAT; |
| int ret; |
| |
| req = ceph_osdc_alloc_request(osdc, snapc, num_ops, false, GFP_NOIO); |
| if (!req) |
| return ERR_PTR(-ENOMEM); |
| |
| list_add_tail(&req->r_private_item, &obj_req->osd_reqs); |
| req->r_callback = rbd_osd_req_callback; |
| req->r_priv = obj_req; |
| |
| /* |
| * Data objects may be stored in a separate pool, but always in |
| * the same namespace in that pool as the header in its pool. |
| */ |
| ceph_oloc_copy(&req->r_base_oloc, &rbd_dev->header_oloc); |
| req->r_base_oloc.pool = rbd_dev->layout.pool_id; |
| |
| ret = ceph_oid_aprintf(&req->r_base_oid, GFP_NOIO, name_format, |
| rbd_dev->header.object_prefix, |
| obj_req->ex.oe_objno); |
| if (ret) |
| return ERR_PTR(ret); |
| |
| return req; |
| } |
| |
| static struct ceph_osd_request * |
| rbd_obj_add_osd_request(struct rbd_obj_request *obj_req, int num_ops) |
| { |
| return __rbd_obj_add_osd_request(obj_req, obj_req->img_request->snapc, |
| num_ops); |
| } |
| |
| static struct rbd_obj_request *rbd_obj_request_create(void) |
| { |
| struct rbd_obj_request *obj_request; |
| |
| obj_request = kmem_cache_zalloc(rbd_obj_request_cache, GFP_NOIO); |
| if (!obj_request) |
| return NULL; |
| |
| ceph_object_extent_init(&obj_request->ex); |
| INIT_LIST_HEAD(&obj_request->osd_reqs); |
| mutex_init(&obj_request->state_mutex); |
| kref_init(&obj_request->kref); |
| |
| dout("%s %p\n", __func__, obj_request); |
| return obj_request; |
| } |
| |
| static void rbd_obj_request_destroy(struct kref *kref) |
| { |
| struct rbd_obj_request *obj_request; |
| struct ceph_osd_request *osd_req; |
| u32 i; |
| |
| obj_request = container_of(kref, struct rbd_obj_request, kref); |
| |
| dout("%s: obj %p\n", __func__, obj_request); |
| |
| while (!list_empty(&obj_request->osd_reqs)) { |
| osd_req = list_first_entry(&obj_request->osd_reqs, |
| struct ceph_osd_request, r_private_item); |
| list_del_init(&osd_req->r_private_item); |
| ceph_osdc_put_request(osd_req); |
| } |
| |
| switch (obj_request->img_request->data_type) { |
| case OBJ_REQUEST_NODATA: |
| case OBJ_REQUEST_BIO: |
| case OBJ_REQUEST_BVECS: |
| break; /* Nothing to do */ |
| case OBJ_REQUEST_OWN_BVECS: |
| kfree(obj_request->bvec_pos.bvecs); |
| break; |
| default: |
| BUG(); |
| } |
| |
| kfree(obj_request->img_extents); |
| if (obj_request->copyup_bvecs) { |
| for (i = 0; i < obj_request->copyup_bvec_count; i++) { |
| if (obj_request->copyup_bvecs[i].bv_page) |
| __free_page(obj_request->copyup_bvecs[i].bv_page); |
| } |
| kfree(obj_request->copyup_bvecs); |
| } |
| |
| kmem_cache_free(rbd_obj_request_cache, obj_request); |
| } |
| |
| /* It's OK to call this for a device with no parent */ |
| |
| static void rbd_spec_put(struct rbd_spec *spec); |
| static void rbd_dev_unparent(struct rbd_device *rbd_dev) |
| { |
| rbd_dev_remove_parent(rbd_dev); |
| rbd_spec_put(rbd_dev->parent_spec); |
| rbd_dev->parent_spec = NULL; |
| rbd_dev->parent_overlap = 0; |
| } |
| |
| /* |
| * Parent image reference counting is used to determine when an |
| * image's parent fields can be safely torn down--after there are no |
| * more in-flight requests to the parent image. When the last |
| * reference is dropped, cleaning them up is safe. |
| */ |
| static void rbd_dev_parent_put(struct rbd_device *rbd_dev) |
| { |
| int counter; |
| |
| if (!rbd_dev->parent_spec) |
| return; |
| |
| counter = atomic_dec_return_safe(&rbd_dev->parent_ref); |
| if (counter > 0) |
| return; |
| |
| /* Last reference; clean up parent data structures */ |
| |
| if (!counter) |
| rbd_dev_unparent(rbd_dev); |
| else |
| rbd_warn(rbd_dev, "parent reference underflow"); |
| } |
| |
| /* |
| * If an image has a non-zero parent overlap, get a reference to its |
| * parent. |
| * |
| * Returns true if the rbd device has a parent with a non-zero |
| * overlap and a reference for it was successfully taken, or |
| * false otherwise. |
| */ |
| static bool rbd_dev_parent_get(struct rbd_device *rbd_dev) |
| { |
| int counter = 0; |
| |
| if (!rbd_dev->parent_spec) |
| return false; |
| |
| if (rbd_dev->parent_overlap) |
| counter = atomic_inc_return_safe(&rbd_dev->parent_ref); |
| |
| if (counter < 0) |
| rbd_warn(rbd_dev, "parent reference overflow"); |
| |
| return counter > 0; |
| } |
| |
| static void rbd_img_request_init(struct rbd_img_request *img_request, |
| struct rbd_device *rbd_dev, |
| enum obj_operation_type op_type) |
| { |
| memset(img_request, 0, sizeof(*img_request)); |
| |
| img_request->rbd_dev = rbd_dev; |
| img_request->op_type = op_type; |
| |
| INIT_LIST_HEAD(&img_request->lock_item); |
| INIT_LIST_HEAD(&img_request->object_extents); |
| mutex_init(&img_request->state_mutex); |
| } |
| |
| static void rbd_img_capture_header(struct rbd_img_request *img_req) |
| { |
| struct rbd_device *rbd_dev = img_req->rbd_dev; |
| |
| lockdep_assert_held(&rbd_dev->header_rwsem); |
| |
| if (rbd_img_is_write(img_req)) |
| img_req->snapc = ceph_get_snap_context(rbd_dev->header.snapc); |
| else |
| img_req->snap_id = rbd_dev->spec->snap_id; |
| |
| if (rbd_dev_parent_get(rbd_dev)) |
| img_request_layered_set(img_req); |
| } |
| |
| static void rbd_img_request_destroy(struct rbd_img_request *img_request) |
| { |
| struct rbd_obj_request *obj_request; |
| struct rbd_obj_request *next_obj_request; |
| |
| dout("%s: img %p\n", __func__, img_request); |
| |
| WARN_ON(!list_empty(&img_request->lock_item)); |
| for_each_obj_request_safe(img_request, obj_request, next_obj_request) |
| rbd_img_obj_request_del(img_request, obj_request); |
| |
| if (img_request_layered_test(img_request)) |
| rbd_dev_parent_put(img_request->rbd_dev); |
| |
| if (rbd_img_is_write(img_request)) |
| ceph_put_snap_context(img_request->snapc); |
| |
| if (test_bit(IMG_REQ_CHILD, &img_request->flags)) |
| kmem_cache_free(rbd_img_request_cache, img_request); |
| } |
| |
| #define BITS_PER_OBJ 2 |
| #define OBJS_PER_BYTE (BITS_PER_BYTE / BITS_PER_OBJ) |
| #define OBJ_MASK ((1 << BITS_PER_OBJ) - 1) |
| |
| static void __rbd_object_map_index(struct rbd_device *rbd_dev, u64 objno, |
| u64 *index, u8 *shift) |
| { |
| u32 off; |
| |
| rbd_assert(objno < rbd_dev->object_map_size); |
| *index = div_u64_rem(objno, OBJS_PER_BYTE, &off); |
| *shift = (OBJS_PER_BYTE - off - 1) * BITS_PER_OBJ; |
| } |
| |
| static u8 __rbd_object_map_get(struct rbd_device *rbd_dev, u64 objno) |
| { |
| u64 index; |
| u8 shift; |
| |
| lockdep_assert_held(&rbd_dev->object_map_lock); |
| __rbd_object_map_index(rbd_dev, objno, &index, &shift); |
| return (rbd_dev->object_map[index] >> shift) & OBJ_MASK; |
| } |
| |
| static void __rbd_object_map_set(struct rbd_device *rbd_dev, u64 objno, u8 val) |
| { |
| u64 index; |
| u8 shift; |
| u8 *p; |
| |
| lockdep_assert_held(&rbd_dev->object_map_lock); |
| rbd_assert(!(val & ~OBJ_MASK)); |
| |
| __rbd_object_map_index(rbd_dev, objno, &index, &shift); |
| p = &rbd_dev->object_map[index]; |
| *p = (*p & ~(OBJ_MASK << shift)) | (val << shift); |
| } |
| |
| static u8 rbd_object_map_get(struct rbd_device *rbd_dev, u64 objno) |
| { |
| u8 state; |
| |
| spin_lock(&rbd_dev->object_map_lock); |
| state = __rbd_object_map_get(rbd_dev, objno); |
| spin_unlock(&rbd_dev->object_map_lock); |
| return state; |
| } |
| |
| static bool use_object_map(struct rbd_device *rbd_dev) |
| { |
| /* |
| * An image mapped read-only can't use the object map -- it isn't |
| * loaded because the header lock isn't acquired. Someone else can |
| * write to the image and update the object map behind our back. |
| * |
| * A snapshot can't be written to, so using the object map is always |
| * safe. |
| */ |
| if (!rbd_is_snap(rbd_dev) && rbd_is_ro(rbd_dev)) |
| return false; |
| |
| return ((rbd_dev->header.features & RBD_FEATURE_OBJECT_MAP) && |
| !(rbd_dev->object_map_flags & RBD_FLAG_OBJECT_MAP_INVALID)); |
| } |
| |
| static bool rbd_object_map_may_exist(struct rbd_device *rbd_dev, u64 objno) |
| { |
| u8 state; |
| |
| /* fall back to default logic if object map is disabled or invalid */ |
| if (!use_object_map(rbd_dev)) |
| return true; |
| |
| state = rbd_object_map_get(rbd_dev, objno); |
| return state != OBJECT_NONEXISTENT; |
| } |
| |
| static void rbd_object_map_name(struct rbd_device *rbd_dev, u64 snap_id, |
| struct ceph_object_id *oid) |
| { |
| if (snap_id == CEPH_NOSNAP) |
| ceph_oid_printf(oid, "%s%s", RBD_OBJECT_MAP_PREFIX, |
| rbd_dev->spec->image_id); |
| else |
| ceph_oid_printf(oid, "%s%s.%016llx", RBD_OBJECT_MAP_PREFIX, |
| rbd_dev->spec->image_id, snap_id); |
| } |
| |
| static int rbd_object_map_lock(struct rbd_device *rbd_dev) |
| { |
| struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
| CEPH_DEFINE_OID_ONSTACK(oid); |
| u8 lock_type; |
| char *lock_tag; |
| struct ceph_locker *lockers; |
| u32 num_lockers; |
| bool broke_lock = false; |
| int ret; |
| |
| rbd_object_map_name(rbd_dev, CEPH_NOSNAP, &oid); |
| |
| again: |
| ret = ceph_cls_lock(osdc, &oid, &rbd_dev->header_oloc, RBD_LOCK_NAME, |
| CEPH_CLS_LOCK_EXCLUSIVE, "", "", "", 0); |
| if (ret != -EBUSY || broke_lock) { |
| if (ret == -EEXIST) |
| ret = 0; /* already locked by myself */ |
| if (ret) |
| rbd_warn(rbd_dev, "failed to lock object map: %d", ret); |
| return ret; |
| } |
| |
| ret = ceph_cls_lock_info(osdc, &oid, &rbd_dev->header_oloc, |
| RBD_LOCK_NAME, &lock_type, &lock_tag, |
| &lockers, &num_lockers); |
| if (ret) { |
| if (ret == -ENOENT) |
| goto again; |
| |
| rbd_warn(rbd_dev, "failed to get object map lockers: %d", ret); |
| return ret; |
| } |
| |
| kfree(lock_tag); |
| if (num_lockers == 0) |
| goto again; |
| |
| rbd_warn(rbd_dev, "breaking object map lock owned by %s%llu", |
| ENTITY_NAME(lockers[0].id.name)); |
| |
| ret = ceph_cls_break_lock(osdc, &oid, &rbd_dev->header_oloc, |
| RBD_LOCK_NAME, lockers[0].id.cookie, |
| &lockers[0].id.name); |
| ceph_free_lockers(lockers, num_lockers); |
| if (ret) { |
| if (ret == -ENOENT) |
| goto again; |
| |
| rbd_warn(rbd_dev, "failed to break object map lock: %d", ret); |
| return ret; |
| } |
| |
| broke_lock = true; |
| goto again; |
| } |
| |
| static void rbd_object_map_unlock(struct rbd_device *rbd_dev) |
| { |
| struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
| CEPH_DEFINE_OID_ONSTACK(oid); |
| int ret; |
| |
| rbd_object_map_name(rbd_dev, CEPH_NOSNAP, &oid); |
| |
| ret = ceph_cls_unlock(osdc, &oid, &rbd_dev->header_oloc, RBD_LOCK_NAME, |
| ""); |
| if (ret && ret != -ENOENT) |
| rbd_warn(rbd_dev, "failed to unlock object map: %d", ret); |
| } |
| |
| static int decode_object_map_header(void **p, void *end, u64 *object_map_size) |
| { |
| u8 struct_v; |
| u32 struct_len; |
| u32 header_len; |
| void *header_end; |
| int ret; |
| |
| ceph_decode_32_safe(p, end, header_len, e_inval); |
| header_end = *p + header_len; |
| |
| ret = ceph_start_decoding(p, end, 1, "BitVector header", &struct_v, |
| &struct_len); |
| if (ret) |
| return ret; |
| |
| ceph_decode_64_safe(p, end, *object_map_size, e_inval); |
| |
| *p = header_end; |
| return 0; |
| |
| e_inval: |
| return -EINVAL; |
| } |
| |
| static int __rbd_object_map_load(struct rbd_device *rbd_dev) |
| { |
| struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
| CEPH_DEFINE_OID_ONSTACK(oid); |
| struct page **pages; |
| void *p, *end; |
| size_t reply_len; |
| u64 num_objects; |
| u64 object_map_bytes; |
| u64 object_map_size; |
| int num_pages; |
| int ret; |
| |
| rbd_assert(!rbd_dev->object_map && !rbd_dev->object_map_size); |
| |
| num_objects = ceph_get_num_objects(&rbd_dev->layout, |
| rbd_dev->mapping.size); |
| object_map_bytes = DIV_ROUND_UP_ULL(num_objects * BITS_PER_OBJ, |
| BITS_PER_BYTE); |
| num_pages = calc_pages_for(0, object_map_bytes) + 1; |
| pages = ceph_alloc_page_vector(num_pages, GFP_KERNEL); |
| if (IS_ERR(pages)) |
| return PTR_ERR(pages); |
| |
| reply_len = num_pages * PAGE_SIZE; |
| rbd_object_map_name(rbd_dev, rbd_dev->spec->snap_id, &oid); |
| ret = ceph_osdc_call(osdc, &oid, &rbd_dev->header_oloc, |
| "rbd", "object_map_load", CEPH_OSD_FLAG_READ, |
| NULL, 0, pages, &reply_len); |
| if (ret) |
| goto out; |
| |
| p = page_address(pages[0]); |
| end = p + min(reply_len, (size_t)PAGE_SIZE); |
| ret = decode_object_map_header(&p, end, &object_map_size); |
| if (ret) |
| goto out; |
| |
| if (object_map_size != num_objects) { |
| rbd_warn(rbd_dev, "object map size mismatch: %llu vs %llu", |
| object_map_size, num_objects); |
| ret = -EINVAL; |
| goto out; |
| } |
| |
| if (offset_in_page(p) + object_map_bytes > reply_len) { |
| ret = -EINVAL; |
| goto out; |
| } |
| |
| rbd_dev->object_map = kvmalloc(object_map_bytes, GFP_KERNEL); |
| if (!rbd_dev->object_map) { |
| ret = -ENOMEM; |
| goto out; |
| } |
| |
| rbd_dev->object_map_size = object_map_size; |
| ceph_copy_from_page_vector(pages, rbd_dev->object_map, |
| offset_in_page(p), object_map_bytes); |
| |
| out: |
| ceph_release_page_vector(pages, num_pages); |
| return ret; |
| } |
| |
| static void rbd_object_map_free(struct rbd_device *rbd_dev) |
| { |
| kvfree(rbd_dev->object_map); |
| rbd_dev->object_map = NULL; |
| rbd_dev->object_map_size = 0; |
| } |
| |
| static int rbd_object_map_load(struct rbd_device *rbd_dev) |
| { |
| int ret; |
| |
| ret = __rbd_object_map_load(rbd_dev); |
| if (ret) |
| return ret; |
| |
| ret = rbd_dev_v2_get_flags(rbd_dev); |
| if (ret) { |
| rbd_object_map_free(rbd_dev); |
| return ret; |
| } |
| |
| if (rbd_dev->object_map_flags & RBD_FLAG_OBJECT_MAP_INVALID) |
| rbd_warn(rbd_dev, "object map is invalid"); |
| |
| return 0; |
| } |
| |
| static int rbd_object_map_open(struct rbd_device *rbd_dev) |
| { |
| int ret; |
| |
| ret = rbd_object_map_lock(rbd_dev); |
| if (ret) |
| return ret; |
| |
| ret = rbd_object_map_load(rbd_dev); |
| if (ret) { |
| rbd_object_map_unlock(rbd_dev); |
| return ret; |
| } |
| |
| return 0; |
| } |
| |
| static void rbd_object_map_close(struct rbd_device *rbd_dev) |
| { |
| rbd_object_map_free(rbd_dev); |
| rbd_object_map_unlock(rbd_dev); |
| } |
| |
| /* |
| * This function needs snap_id (or more precisely just something to |
| * distinguish between HEAD and snapshot object maps), new_state and |
| * current_state that were passed to rbd_object_map_update(). |
| * |
| * To avoid allocating and stashing a context we piggyback on the OSD |
| * request. A HEAD update has two ops (assert_locked). For new_state |
| * and current_state we decode our own object_map_update op, encoded in |
| * rbd_cls_object_map_update(). |
| */ |
| static int rbd_object_map_update_finish(struct rbd_obj_request *obj_req, |
| struct ceph_osd_request *osd_req) |
| { |
| struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev; |
| struct ceph_osd_data *osd_data; |
| u64 objno; |
| u8 state, new_state, uninitialized_var(current_state); |
| bool has_current_state; |
| void *p; |
| |
| if (osd_req->r_result) |
| return osd_req->r_result; |
| |
| /* |
| * Nothing to do for a snapshot object map. |
| */ |
| if (osd_req->r_num_ops == 1) |
| return 0; |
| |
| /* |
| * Update in-memory HEAD object map. |
| */ |
| rbd_assert(osd_req->r_num_ops == 2); |
| osd_data = osd_req_op_data(osd_req, 1, cls, request_data); |
| rbd_assert(osd_data->type == CEPH_OSD_DATA_TYPE_PAGES); |
| |
| p = page_address(osd_data->pages[0]); |
| objno = ceph_decode_64(&p); |
| rbd_assert(objno == obj_req->ex.oe_objno); |
| rbd_assert(ceph_decode_64(&p) == objno + 1); |
| new_state = ceph_decode_8(&p); |
| has_current_state = ceph_decode_8(&p); |
| if (has_current_state) |
| current_state = ceph_decode_8(&p); |
| |
| spin_lock(&rbd_dev->object_map_lock); |
| state = __rbd_object_map_get(rbd_dev, objno); |
| if (!has_current_state || current_state == state || |
| (current_state == OBJECT_EXISTS && state == OBJECT_EXISTS_CLEAN)) |
| __rbd_object_map_set(rbd_dev, objno, new_state); |
| spin_unlock(&rbd_dev->object_map_lock); |
| |
| return 0; |
| } |
| |
| static void rbd_object_map_callback(struct ceph_osd_request *osd_req) |
| { |
| struct rbd_obj_request *obj_req = osd_req->r_priv; |
| int result; |
| |
| dout("%s osd_req %p result %d for obj_req %p\n", __func__, osd_req, |
| osd_req->r_result, obj_req); |
| |
| result = rbd_object_map_update_finish(obj_req, osd_req); |
| rbd_obj_handle_request(obj_req, result); |
| } |
| |
| static bool update_needed(struct rbd_device *rbd_dev, u64 objno, u8 new_state) |
| { |
| u8 state = rbd_object_map_get(rbd_dev, objno); |
| |
| if (state == new_state || |
| (new_state == OBJECT_PENDING && state == OBJECT_NONEXISTENT) || |
| (new_state == OBJECT_NONEXISTENT && state != OBJECT_PENDING)) |
| return false; |
| |
| return true; |
| } |
| |
| static int rbd_cls_object_map_update(struct ceph_osd_request *req, |
| int which, u64 objno, u8 new_state, |
| const u8 *current_state) |
| { |
| struct page **pages; |
| void *p, *start; |
| int ret; |
| |
| ret = osd_req_op_cls_init(req, which, "rbd", "object_map_update"); |
| if (ret) |
| return ret; |
| |
| pages = ceph_alloc_page_vector(1, GFP_NOIO); |
| if (IS_ERR(pages)) |
| return PTR_ERR(pages); |
| |
| p = start = page_address(pages[0]); |
| ceph_encode_64(&p, objno); |
| ceph_encode_64(&p, objno + 1); |
| ceph_encode_8(&p, new_state); |
| if (current_state) { |
| ceph_encode_8(&p, 1); |
| ceph_encode_8(&p, *current_state); |
| } else { |
| ceph_encode_8(&p, 0); |
| } |
| |
| osd_req_op_cls_request_data_pages(req, which, pages, p - start, 0, |
| false, true); |
| return 0; |
| } |
| |
| /* |
| * Return: |
| * 0 - object map update sent |
| * 1 - object map update isn't needed |
| * <0 - error |
| */ |
| static int rbd_object_map_update(struct rbd_obj_request *obj_req, u64 snap_id, |
| u8 new_state, const u8 *current_state) |
| { |
| struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev; |
| struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
| struct ceph_osd_request *req; |
| int num_ops = 1; |
| int which = 0; |
| int ret; |
| |
| if (snap_id == CEPH_NOSNAP) { |
| if (!update_needed(rbd_dev, obj_req->ex.oe_objno, new_state)) |
| return 1; |
| |
| num_ops++; /* assert_locked */ |
| } |
| |
| req = ceph_osdc_alloc_request(osdc, NULL, num_ops, false, GFP_NOIO); |
| if (!req) |
| return -ENOMEM; |
| |
| list_add_tail(&req->r_private_item, &obj_req->osd_reqs); |
| req->r_callback = rbd_object_map_callback; |
| req->r_priv = obj_req; |
| |
| rbd_object_map_name(rbd_dev, snap_id, &req->r_base_oid); |
| ceph_oloc_copy(&req->r_base_oloc, &rbd_dev->header_oloc); |
| req->r_flags = CEPH_OSD_FLAG_WRITE; |
| ktime_get_real_ts64(&req->r_mtime); |
| |
| if (snap_id == CEPH_NOSNAP) { |
| /* |
| * Protect against possible race conditions during lock |
| * ownership transitions. |
| */ |
| ret = ceph_cls_assert_locked(req, which++, RBD_LOCK_NAME, |
| CEPH_CLS_LOCK_EXCLUSIVE, "", ""); |
| if (ret) |
| return ret; |
| } |
| |
| ret = rbd_cls_object_map_update(req, which, obj_req->ex.oe_objno, |
| new_state, current_state); |
| if (ret) |
| return ret; |
| |
| ret = ceph_osdc_alloc_messages(req, GFP_NOIO); |
| if (ret) |
| return ret; |
| |
| ceph_osdc_start_request(osdc, req, false); |
| return 0; |
| } |
| |
| static void prune_extents(struct ceph_file_extent *img_extents, |
| u32 *num_img_extents, u64 overlap) |
| { |
| u32 cnt = *num_img_extents; |
| |
| /* drop extents completely beyond the overlap */ |
| while (cnt && img_extents[cnt - 1].fe_off >= overlap) |
| cnt--; |
| |
| if (cnt) { |
| struct ceph_file_extent *ex = &img_extents[cnt - 1]; |
| |
| /* trim final overlapping extent */ |
| if (ex->fe_off + ex->fe_len > overlap) |
| ex->fe_len = overlap - ex->fe_off; |
| } |
| |
| *num_img_extents = cnt; |
| } |
| |
| /* |
| * Determine the byte range(s) covered by either just the object extent |
| * or the entire object in the parent image. |
| */ |
| static int rbd_obj_calc_img_extents(struct rbd_obj_request *obj_req, |
| bool entire) |
| { |
| struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev; |
| int ret; |
| |
| if (!rbd_dev->parent_overlap) |
| return 0; |
| |
| ret = ceph_extent_to_file(&rbd_dev->layout, obj_req->ex.oe_objno, |
| entire ? 0 : obj_req->ex.oe_off, |
| entire ? rbd_dev->layout.object_size : |
| obj_req->ex.oe_len, |
| &obj_req->img_extents, |
| &obj_req->num_img_extents); |
| if (ret) |
| return ret; |
| |
| prune_extents(obj_req->img_extents, &obj_req->num_img_extents, |
| rbd_dev->parent_overlap); |
| return 0; |
| } |
| |
| static void rbd_osd_setup_data(struct ceph_osd_request *osd_req, int which) |
| { |
| struct rbd_obj_request *obj_req = osd_req->r_priv; |
| |
| switch (obj_req->img_request->data_type) { |
| case OBJ_REQUEST_BIO: |
| osd_req_op_extent_osd_data_bio(osd_req, which, |
| &obj_req->bio_pos, |
| obj_req->ex.oe_len); |
| break; |
| case OBJ_REQUEST_BVECS: |
| case OBJ_REQUEST_OWN_BVECS: |
| rbd_assert(obj_req->bvec_pos.iter.bi_size == |
| obj_req->ex.oe_len); |
| rbd_assert(obj_req->bvec_idx == obj_req->bvec_count); |
| osd_req_op_extent_osd_data_bvec_pos(osd_req, which, |
| &obj_req->bvec_pos); |
| break; |
| default: |
| BUG(); |
| } |
| } |
| |
| static int rbd_osd_setup_stat(struct ceph_osd_request *osd_req, int which) |
| { |
| struct page **pages; |
| |
| /* |
| * The response data for a STAT call consists of: |
| * le64 length; |
| * struct { |
| * le32 tv_sec; |
| * le32 tv_nsec; |
| * } mtime; |
| */ |
| pages = ceph_alloc_page_vector(1, GFP_NOIO); |
| if (IS_ERR(pages)) |
| return PTR_ERR(pages); |
| |
| osd_req_op_init(osd_req, which, CEPH_OSD_OP_STAT, 0); |
| osd_req_op_raw_data_in_pages(osd_req, which, pages, |
| 8 + sizeof(struct ceph_timespec), |
| 0, false, true); |
| return 0; |
| } |
| |
| static int rbd_osd_setup_copyup(struct ceph_osd_request *osd_req, int which, |
| u32 bytes) |
| { |
| struct rbd_obj_request *obj_req = osd_req->r_priv; |
| int ret; |
| |
| ret = osd_req_op_cls_init(osd_req, which, "rbd", "copyup"); |
| if (ret) |
| return ret; |
| |
| osd_req_op_cls_request_data_bvecs(osd_req, which, obj_req->copyup_bvecs, |
| obj_req->copyup_bvec_count, bytes); |
| return 0; |
| } |
| |
| static int rbd_obj_init_read(struct rbd_obj_request *obj_req) |
| { |
| obj_req->read_state = RBD_OBJ_READ_START; |
| return 0; |
| } |
| |
| static void __rbd_osd_setup_write_ops(struct ceph_osd_request *osd_req, |
| int which) |
| { |
| struct rbd_obj_request *obj_req = osd_req->r_priv; |
| struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev; |
| u16 opcode; |
| |
| if (!use_object_map(rbd_dev) || |
| !(obj_req->flags & RBD_OBJ_FLAG_MAY_EXIST)) { |
| osd_req_op_alloc_hint_init(osd_req, which++, |
| rbd_dev->layout.object_size, |
| rbd_dev->layout.object_size, |
| rbd_dev->opts->alloc_hint_flags); |
| } |
| |
| if (rbd_obj_is_entire(obj_req)) |
| opcode = CEPH_OSD_OP_WRITEFULL; |
| else |
| opcode = CEPH_OSD_OP_WRITE; |
| |
| osd_req_op_extent_init(osd_req, which, opcode, |
| obj_req->ex.oe_off, obj_req->ex.oe_len, 0, 0); |
| rbd_osd_setup_data(osd_req, which); |
| } |
| |
| static int rbd_obj_init_write(struct rbd_obj_request *obj_req) |
| { |
| int ret; |
| |
| /* reverse map the entire object onto the parent */ |
| ret = rbd_obj_calc_img_extents(obj_req, true); |
| if (ret) |
| return ret; |
| |
| if (rbd_obj_copyup_enabled(obj_req)) |
| obj_req->flags |= RBD_OBJ_FLAG_COPYUP_ENABLED; |
| |
| obj_req->write_state = RBD_OBJ_WRITE_START; |
| return 0; |
| } |
| |
| static u16 truncate_or_zero_opcode(struct rbd_obj_request *obj_req) |
| { |
| return rbd_obj_is_tail(obj_req) ? CEPH_OSD_OP_TRUNCATE : |
| CEPH_OSD_OP_ZERO; |
| } |
| |
| static void __rbd_osd_setup_discard_ops(struct ceph_osd_request *osd_req, |
| int which) |
| { |
| struct rbd_obj_request *obj_req = osd_req->r_priv; |
| |
| if (rbd_obj_is_entire(obj_req) && !obj_req->num_img_extents) { |
| rbd_assert(obj_req->flags & RBD_OBJ_FLAG_DELETION); |
| osd_req_op_init(osd_req, which, CEPH_OSD_OP_DELETE, 0); |
| } else { |
| osd_req_op_extent_init(osd_req, which, |
| truncate_or_zero_opcode(obj_req), |
| obj_req->ex.oe_off, obj_req->ex.oe_len, |
| 0, 0); |
| } |
| } |
| |
| static int rbd_obj_init_discard(struct rbd_obj_request *obj_req) |
| { |
| struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev; |
| u64 off, next_off; |
| int ret; |
| |
| /* |
| * Align the range to alloc_size boundary and punt on discards |
| * that are too small to free up any space. |
| * |
| * alloc_size == object_size && is_tail() is a special case for |
| * filestore with filestore_punch_hole = false, needed to allow |
| * truncate (in addition to delete). |
| */ |
| if (rbd_dev->opts->alloc_size != rbd_dev->layout.object_size || |
| !rbd_obj_is_tail(obj_req)) { |
| off = round_up(obj_req->ex.oe_off, rbd_dev->opts->alloc_size); |
| next_off = round_down(obj_req->ex.oe_off + obj_req->ex.oe_len, |
| rbd_dev->opts->alloc_size); |
| if (off >= next_off) |
| return 1; |
| |
| dout("%s %p %llu~%llu -> %llu~%llu\n", __func__, |
| obj_req, obj_req->ex.oe_off, obj_req->ex.oe_len, |
| off, next_off - off); |
| obj_req->ex.oe_off = off; |
| obj_req->ex.oe_len = next_off - off; |
| } |
| |
| /* reverse map the entire object onto the parent */ |
| ret = rbd_obj_calc_img_extents(obj_req, true); |
| if (ret) |
| return ret; |
| |
| obj_req->flags |= RBD_OBJ_FLAG_NOOP_FOR_NONEXISTENT; |
| if (rbd_obj_is_entire(obj_req) && !obj_req->num_img_extents) |
| obj_req->flags |= RBD_OBJ_FLAG_DELETION; |
| |
| obj_req->write_state = RBD_OBJ_WRITE_START; |
| return 0; |
| } |
| |
| static void __rbd_osd_setup_zeroout_ops(struct ceph_osd_request *osd_req, |
| int which) |
| { |
| struct rbd_obj_request *obj_req = osd_req->r_priv; |
| u16 opcode; |
| |
| if (rbd_obj_is_entire(obj_req)) { |
| if (obj_req->num_img_extents) { |
| if (!(obj_req->flags & RBD_OBJ_FLAG_COPYUP_ENABLED)) |
| osd_req_op_init(osd_req, which++, |
| CEPH_OSD_OP_CREATE, 0); |
| opcode = CEPH_OSD_OP_TRUNCATE; |
| } else { |
| rbd_assert(obj_req->flags & RBD_OBJ_FLAG_DELETION); |
| osd_req_op_init(osd_req, which++, |
| CEPH_OSD_OP_DELETE, 0); |
| opcode = 0; |
| } |
| } else { |
| opcode = truncate_or_zero_opcode(obj_req); |
| } |
| |
| if (opcode) |
| osd_req_op_extent_init(osd_req, which, opcode, |
| obj_req->ex.oe_off, obj_req->ex.oe_len, |
| 0, 0); |
| } |
| |
| static int rbd_obj_init_zeroout(struct rbd_obj_request *obj_req) |
| { |
| int ret; |
| |
| /* reverse map the entire object onto the parent */ |
| ret = rbd_obj_calc_img_extents(obj_req, true); |
| if (ret) |
| return ret; |
| |
| if (rbd_obj_copyup_enabled(obj_req)) |
| obj_req->flags |= RBD_OBJ_FLAG_COPYUP_ENABLED; |
| if (!obj_req->num_img_extents) { |
| obj_req->flags |= RBD_OBJ_FLAG_NOOP_FOR_NONEXISTENT; |
| if (rbd_obj_is_entire(obj_req)) |
| obj_req->flags |= RBD_OBJ_FLAG_DELETION; |
| } |
| |
| obj_req->write_state = RBD_OBJ_WRITE_START; |
| return 0; |
| } |
| |
| static int count_write_ops(struct rbd_obj_request *obj_req) |
| { |
| struct rbd_img_request *img_req = obj_req->img_request; |
| |
| switch (img_req->op_type) { |
| case OBJ_OP_WRITE: |
| if (!use_object_map(img_req->rbd_dev) || |
| !(obj_req->flags & RBD_OBJ_FLAG_MAY_EXIST)) |
| return 2; /* setallochint + write/writefull */ |
| |
| return 1; /* write/writefull */ |
| case OBJ_OP_DISCARD: |
| return 1; /* delete/truncate/zero */ |
| case OBJ_OP_ZEROOUT: |
| if (rbd_obj_is_entire(obj_req) && obj_req->num_img_extents && |
| !(obj_req->flags & RBD_OBJ_FLAG_COPYUP_ENABLED)) |
| return 2; /* create + truncate */ |
| |
| return 1; /* delete/truncate/zero */ |
| default: |
| BUG(); |
| } |
| } |
| |
| static void rbd_osd_setup_write_ops(struct ceph_osd_request *osd_req, |
| int which) |
| { |
| struct rbd_obj_request *obj_req = osd_req->r_priv; |
| |
| switch (obj_req->img_request->op_type) { |
| case OBJ_OP_WRITE: |
| __rbd_osd_setup_write_ops(osd_req, which); |
| break; |
| case OBJ_OP_DISCARD: |
| __rbd_osd_setup_discard_ops(osd_req, which); |
| break; |
| case OBJ_OP_ZEROOUT: |
| __rbd_osd_setup_zeroout_ops(osd_req, which); |
| break; |
| default: |
| BUG(); |
| } |
| } |
| |
| /* |
| * Prune the list of object requests (adjust offset and/or length, drop |
| * redundant requests). Prepare object request state machines and image |
| * request state machine for execution. |
| */ |
| static int __rbd_img_fill_request(struct rbd_img_request *img_req) |
| { |
| struct rbd_obj_request *obj_req, *next_obj_req; |
| int ret; |
| |
| for_each_obj_request_safe(img_req, obj_req, next_obj_req) { |
| switch (img_req->op_type) { |
| case OBJ_OP_READ: |
| ret = rbd_obj_init_read(obj_req); |
| break; |
| case OBJ_OP_WRITE: |
| ret = rbd_obj_init_write(obj_req); |
| break; |
| case OBJ_OP_DISCARD: |
| ret = rbd_obj_init_discard(obj_req); |
| break; |
| case OBJ_OP_ZEROOUT: |
| ret = rbd_obj_init_zeroout(obj_req); |
| break; |
| default: |
| BUG(); |
| } |
| if (ret < 0) |
| return ret; |
| if (ret > 0) { |
| rbd_img_obj_request_del(img_req, obj_req); |
| continue; |
| } |
| } |
| |
| img_req->state = RBD_IMG_START; |
| return 0; |
| } |
| |
| union rbd_img_fill_iter { |
| struct ceph_bio_iter bio_iter; |
| struct ceph_bvec_iter bvec_iter; |
| }; |
| |
| struct rbd_img_fill_ctx { |
| enum obj_request_type pos_type; |
| union rbd_img_fill_iter *pos; |
| union rbd_img_fill_iter iter; |
| ceph_object_extent_fn_t set_pos_fn; |
| ceph_object_extent_fn_t count_fn; |
| ceph_object_extent_fn_t copy_fn; |
| }; |
| |
| static struct ceph_object_extent *alloc_object_extent(void *arg) |
| { |
| struct rbd_img_request *img_req = arg; |
| struct rbd_obj_request *obj_req; |
| |
| obj_req = rbd_obj_request_create(); |
| if (!obj_req) |
| return NULL; |
| |
| rbd_img_obj_request_add(img_req, obj_req); |
| return &obj_req->ex; |
| } |
| |
| /* |
| * While su != os && sc == 1 is technically not fancy (it's the same |
| * layout as su == os && sc == 1), we can't use the nocopy path for it |
| * because ->set_pos_fn() should be called only once per object. |
| * ceph_file_to_extents() invokes action_fn once per stripe unit, so |
| * treat su != os && sc == 1 as fancy. |
| */ |
| static bool rbd_layout_is_fancy(struct ceph_file_layout *l) |
| { |
| return l->stripe_unit != l->object_size; |
| } |
| |
| static int rbd_img_fill_request_nocopy(struct rbd_img_request *img_req, |
| struct ceph_file_extent *img_extents, |
| u32 num_img_extents, |
| struct rbd_img_fill_ctx *fctx) |
| { |
| u32 i; |
| int ret; |
| |
| img_req->data_type = fctx->pos_type; |
| |
| /* |
| * Create object requests and set each object request's starting |
| * position in the provided bio (list) or bio_vec array. |
| */ |
| fctx->iter = *fctx->pos; |
| for (i = 0; i < num_img_extents; i++) { |
| ret = ceph_file_to_extents(&img_req->rbd_dev->layout, |
| img_extents[i].fe_off, |
| img_extents[i].fe_len, |
| &img_req->object_extents, |
| alloc_object_extent, img_req, |
| fctx->set_pos_fn, &fctx->iter); |
| if (ret) |
| return ret; |
| } |
| |
| return __rbd_img_fill_request(img_req); |
| } |
| |
| /* |
| * Map a list of image extents to a list of object extents, create the |
| * corresponding object requests (normally each to a different object, |
| * but not always) and add them to @img_req. For each object request, |
| * set up its data descriptor to point to the corresponding chunk(s) of |
| * @fctx->pos data buffer. |
| * |
| * Because ceph_file_to_extents() will merge adjacent object extents |
| * together, each object request's data descriptor may point to multiple |
| * different chunks of @fctx->pos data buffer. |
| * |
| * @fctx->pos data buffer is assumed to be large enough. |
| */ |
| static int rbd_img_fill_request(struct rbd_img_request *img_req, |
| struct ceph_file_extent *img_extents, |
| u32 num_img_extents, |
| struct rbd_img_fill_ctx *fctx) |
| { |
| struct rbd_device *rbd_dev = img_req->rbd_dev; |
| struct rbd_obj_request *obj_req; |
| u32 i; |
| int ret; |
| |
| if (fctx->pos_type == OBJ_REQUEST_NODATA || |
| !rbd_layout_is_fancy(&rbd_dev->layout)) |
| return rbd_img_fill_request_nocopy(img_req, img_extents, |
| num_img_extents, fctx); |
| |
| img_req->data_type = OBJ_REQUEST_OWN_BVECS; |
| |
| /* |
| * Create object requests and determine ->bvec_count for each object |
| * request. Note that ->bvec_count sum over all object requests may |
| * be greater than the number of bio_vecs in the provided bio (list) |
| * or bio_vec array because when mapped, those bio_vecs can straddle |
| * stripe unit boundaries. |
| */ |
| fctx->iter = *fctx->pos; |
| for (i = 0; i < num_img_extents; i++) { |
| ret = ceph_file_to_extents(&rbd_dev->layout, |
| img_extents[i].fe_off, |
| img_extents[i].fe_len, |
| &img_req->object_extents, |
| alloc_object_extent, img_req, |
| fctx->count_fn, &fctx->iter); |
| if (ret) |
| return ret; |
| } |
| |
| for_each_obj_request(img_req, obj_req) { |
| obj_req->bvec_pos.bvecs = kmalloc_array(obj_req->bvec_count, |
| sizeof(*obj_req->bvec_pos.bvecs), |
| GFP_NOIO); |
| if (!obj_req->bvec_pos.bvecs) |
| return -ENOMEM; |
| } |
| |
| /* |
| * Fill in each object request's private bio_vec array, splitting and |
| * rearranging the provided bio_vecs in stripe unit chunks as needed. |
| */ |
| fctx->iter = *fctx->pos; |
| for (i = 0; i < num_img_extents; i++) { |
| ret = ceph_iterate_extents(&rbd_dev->layout, |
| img_extents[i].fe_off, |
| img_extents[i].fe_len, |
| &img_req->object_extents, |
| fctx->copy_fn, &fctx->iter); |
| if (ret) |
| return ret; |
| } |
| |
| return __rbd_img_fill_request(img_req); |
| } |
| |
| static int rbd_img_fill_nodata(struct rbd_img_request *img_req, |
| u64 off, u64 len) |
| { |
| struct ceph_file_extent ex = { off, len }; |
| union rbd_img_fill_iter dummy = {}; |
| struct rbd_img_fill_ctx fctx = { |
| .pos_type = OBJ_REQUEST_NODATA, |
| .pos = &dummy, |
| }; |
| |
| return rbd_img_fill_request(img_req, &ex, 1, &fctx); |
| } |
| |
| static void set_bio_pos(struct ceph_object_extent *ex, u32 bytes, void *arg) |
| { |
| struct rbd_obj_request *obj_req = |
| container_of(ex, struct rbd_obj_request, ex); |
| struct ceph_bio_iter *it = arg; |
| |
| dout("%s objno %llu bytes %u\n", __func__, ex->oe_objno, bytes); |
| obj_req->bio_pos = *it; |
| ceph_bio_iter_advance(it, bytes); |
| } |
| |
| static void count_bio_bvecs(struct ceph_object_extent *ex, u32 bytes, void *arg) |
| { |
| struct rbd_obj_request *obj_req = |
| container_of(ex, struct rbd_obj_request, ex); |
| struct ceph_bio_iter *it = arg; |
| |
| dout("%s objno %llu bytes %u\n", __func__, ex->oe_objno, bytes); |
| ceph_bio_iter_advance_step(it, bytes, ({ |
| obj_req->bvec_count++; |
| })); |
| |
| } |
| |
| static void copy_bio_bvecs(struct ceph_object_extent *ex, u32 bytes, void *arg) |
| { |
| struct rbd_obj_request *obj_req = |
| container_of(ex, struct rbd_obj_request, ex); |
| struct ceph_bio_iter *it = arg; |
| |
| dout("%s objno %llu bytes %u\n", __func__, ex->oe_objno, bytes); |
| ceph_bio_iter_advance_step(it, bytes, ({ |
| obj_req->bvec_pos.bvecs[obj_req->bvec_idx++] = bv; |
| obj_req->bvec_pos.iter.bi_size += bv.bv_len; |
| })); |
| } |
| |
| static int __rbd_img_fill_from_bio(struct rbd_img_request *img_req, |
| struct ceph_file_extent *img_extents, |
| u32 num_img_extents, |
| struct ceph_bio_iter *bio_pos) |
| { |
| struct rbd_img_fill_ctx fctx = { |
| .pos_type = OBJ_REQUEST_BIO, |
| .pos = (union rbd_img_fill_iter *)bio_pos, |
| .set_pos_fn = set_bio_pos, |
| .count_fn = count_bio_bvecs, |
| .copy_fn = copy_bio_bvecs, |
| }; |
| |
| return rbd_img_fill_request(img_req, img_extents, num_img_extents, |
| &fctx); |
| } |
| |
| static int rbd_img_fill_from_bio(struct rbd_img_request *img_req, |
| u64 off, u64 len, struct bio *bio) |
| { |
| struct ceph_file_extent ex = { off, len }; |
| struct ceph_bio_iter it = { .bio = bio, .iter = bio->bi_iter }; |
| |
| return __rbd_img_fill_from_bio(img_req, &ex, 1, &it); |
| } |
| |
| static void set_bvec_pos(struct ceph_object_extent *ex, u32 bytes, void *arg) |
| { |
| struct rbd_obj_request *obj_req = |
| container_of(ex, struct rbd_obj_request, ex); |
| struct ceph_bvec_iter *it = arg; |
| |
| obj_req->bvec_pos = *it; |
| ceph_bvec_iter_shorten(&obj_req->bvec_pos, bytes); |
| ceph_bvec_iter_advance(it, bytes); |
| } |
| |
| static void count_bvecs(struct ceph_object_extent *ex, u32 bytes, void *arg) |
| { |
| struct rbd_obj_request *obj_req = |
| container_of(ex, struct rbd_obj_request, ex); |
| struct ceph_bvec_iter *it = arg; |
| |
| ceph_bvec_iter_advance_step(it, bytes, ({ |
| obj_req->bvec_count++; |
| })); |
| } |
| |
| static void copy_bvecs(struct ceph_object_extent *ex, u32 bytes, void *arg) |
| { |
| struct rbd_obj_request *obj_req = |
| container_of(ex, struct rbd_obj_request, ex); |
| struct ceph_bvec_iter *it = arg; |
| |
| ceph_bvec_iter_advance_step(it, bytes, ({ |
| obj_req->bvec_pos.bvecs[obj_req->bvec_idx++] = bv; |
| obj_req->bvec_pos.iter.bi_size += bv.bv_len; |
| })); |
| } |
| |
| static int __rbd_img_fill_from_bvecs(struct rbd_img_request *img_req, |
| struct ceph_file_extent *img_extents, |
| u32 num_img_extents, |
| struct ceph_bvec_iter *bvec_pos) |
| { |
| struct rbd_img_fill_ctx fctx = { |
| .pos_type = OBJ_REQUEST_BVECS, |
| .pos = (union rbd_img_fill_iter *)bvec_pos, |
| .set_pos_fn = set_bvec_pos, |
| .count_fn = count_bvecs, |
| .copy_fn = copy_bvecs, |
| }; |
| |
| return rbd_img_fill_request(img_req, img_extents, num_img_extents, |
| &fctx); |
| } |
| |
| static int rbd_img_fill_from_bvecs(struct rbd_img_request *img_req, |
| struct ceph_file_extent *img_extents, |
| u32 num_img_extents, |
| struct bio_vec *bvecs) |
| { |
| struct ceph_bvec_iter it = { |
| .bvecs = bvecs, |
| .iter = { .bi_size = ceph_file_extents_bytes(img_extents, |
| num_img_extents) }, |
| }; |
| |
| return __rbd_img_fill_from_bvecs(img_req, img_extents, num_img_extents, |
| &it); |
| } |
| |
| static void rbd_img_handle_request_work(struct work_struct *work) |
| { |
| struct rbd_img_request *img_req = |
| container_of(work, struct rbd_img_request, work); |
| |
| rbd_img_handle_request(img_req, img_req->work_result); |
| } |
| |
| static void rbd_img_schedule(struct rbd_img_request *img_req, int result) |
| { |
| INIT_WORK(&img_req->work, rbd_img_handle_request_work); |
| img_req->work_result = result; |
| queue_work(rbd_wq, &img_req->work); |
| } |
| |
| static bool rbd_obj_may_exist(struct rbd_obj_request *obj_req) |
| { |
| struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev; |
| |
| if (rbd_object_map_may_exist(rbd_dev, obj_req->ex.oe_objno)) { |
| obj_req->flags |= RBD_OBJ_FLAG_MAY_EXIST; |
| return true; |
| } |
| |
| dout("%s %p objno %llu assuming dne\n", __func__, obj_req, |
| obj_req->ex.oe_objno); |
| return false; |
| } |
| |
| static int rbd_obj_read_object(struct rbd_obj_request *obj_req) |
| { |
| struct ceph_osd_request *osd_req; |
| int ret; |
| |
| osd_req = __rbd_obj_add_osd_request(obj_req, NULL, 1); |
| if (IS_ERR(osd_req)) |
| return PTR_ERR(osd_req); |
| |
| osd_req_op_extent_init(osd_req, 0, CEPH_OSD_OP_READ, |
| obj_req->ex.oe_off, obj_req->ex.oe_len, 0, 0); |
| rbd_osd_setup_data(osd_req, 0); |
| rbd_osd_format_read(osd_req); |
| |
| ret = ceph_osdc_alloc_messages(osd_req, GFP_NOIO); |
| if (ret) |
| return ret; |
| |
| rbd_osd_submit(osd_req); |
| return 0; |
| } |
| |
| static int rbd_obj_read_from_parent(struct rbd_obj_request *obj_req) |
| { |
| struct rbd_img_request *img_req = obj_req->img_request; |
| struct rbd_device *parent = img_req->rbd_dev->parent; |
| struct rbd_img_request *child_img_req; |
| int ret; |
| |
| child_img_req = kmem_cache_alloc(rbd_img_request_cache, GFP_NOIO); |
| if (!child_img_req) |
| return -ENOMEM; |
| |
| rbd_img_request_init(child_img_req, parent, OBJ_OP_READ); |
| __set_bit(IMG_REQ_CHILD, &child_img_req->flags); |
| child_img_req->obj_request = obj_req; |
| |
| down_read(&parent->header_rwsem); |
| rbd_img_capture_header(child_img_req); |
| up_read(&parent->header_rwsem); |
| |
| dout("%s child_img_req %p for obj_req %p\n", __func__, child_img_req, |
| obj_req); |
| |
| if (!rbd_img_is_write(img_req)) { |
| switch (img_req->data_type) { |
| case OBJ_REQUEST_BIO: |
| ret = __rbd_img_fill_from_bio(child_img_req, |
| obj_req->img_extents, |
| obj_req->num_img_extents, |
| &obj_req->bio_pos); |
| break; |
| case OBJ_REQUEST_BVECS: |
| case OBJ_REQUEST_OWN_BVECS: |
| ret = __rbd_img_fill_from_bvecs(child_img_req, |
| obj_req->img_extents, |
| obj_req->num_img_extents, |
| &obj_req->bvec_pos); |
| break; |
| default: |
| BUG(); |
| } |
| } else { |
| ret = rbd_img_fill_from_bvecs(child_img_req, |
| obj_req->img_extents, |
| obj_req->num_img_extents, |
| obj_req->copyup_bvecs); |
| } |
| if (ret) { |
| rbd_img_request_destroy(child_img_req); |
| return ret; |
| } |
| |
| /* avoid parent chain recursion */ |
| rbd_img_schedule(child_img_req, 0); |
| return 0; |
| } |
| |
| static bool rbd_obj_advance_read(struct rbd_obj_request *obj_req, int *result) |
| { |
| struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev; |
| int ret; |
| |
| again: |
| switch (obj_req->read_state) { |
| case RBD_OBJ_READ_START: |
| rbd_assert(!*result); |
| |
| if (!rbd_obj_may_exist(obj_req)) { |
| *result = -ENOENT; |
| obj_req->read_state = RBD_OBJ_READ_OBJECT; |
| goto again; |
| } |
| |
| ret = rbd_obj_read_object(obj_req); |
| if (ret) { |
| *result = ret; |
| return true; |
| } |
| obj_req->read_state = RBD_OBJ_READ_OBJECT; |
| return false; |
| case RBD_OBJ_READ_OBJECT: |
| if (*result == -ENOENT && rbd_dev->parent_overlap) { |
| /* reverse map this object extent onto the parent */ |
| ret = rbd_obj_calc_img_extents(obj_req, false); |
| if (ret) { |
| *result = ret; |
| return true; |
| } |
| if (obj_req->num_img_extents) { |
| ret = rbd_obj_read_from_parent(obj_req); |
| if (ret) { |
| *result = ret; |
| return true; |
| } |
| obj_req->read_state = RBD_OBJ_READ_PARENT; |
| return false; |
| } |
| } |
| |
| /* |
| * -ENOENT means a hole in the image -- zero-fill the entire |
| * length of the request. A short read also implies zero-fill |
| * to the end of the request. |
| */ |
| if (*result == -ENOENT) { |
| rbd_obj_zero_range(obj_req, 0, obj_req->ex.oe_len); |
| *result = 0; |
| } else if (*result >= 0) { |
| if (*result < obj_req->ex.oe_len) |
| rbd_obj_zero_range(obj_req, *result, |
| obj_req->ex.oe_len - *result); |
| else |
| rbd_assert(*result == obj_req->ex.oe_len); |
| *result = 0; |
| } |
| return true; |
| case RBD_OBJ_READ_PARENT: |
| /* |
| * The parent image is read only up to the overlap -- zero-fill |
| * from the overlap to the end of the request. |
| */ |
| if (!*result) { |
| u32 obj_overlap = rbd_obj_img_extents_bytes(obj_req); |
| |
| if (obj_overlap < obj_req->ex.oe_len) |
| rbd_obj_zero_range(obj_req, obj_overlap, |
| obj_req->ex.oe_len - obj_overlap); |
| } |
| return true; |
| default: |
| BUG(); |
| } |
| } |
| |
| static bool rbd_obj_write_is_noop(struct rbd_obj_request *obj_req) |
| { |
| struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev; |
| |
| if (rbd_object_map_may_exist(rbd_dev, obj_req->ex.oe_objno)) |
| obj_req->flags |= RBD_OBJ_FLAG_MAY_EXIST; |
| |
| if (!(obj_req->flags & RBD_OBJ_FLAG_MAY_EXIST) && |
| (obj_req->flags & RBD_OBJ_FLAG_NOOP_FOR_NONEXISTENT)) { |
| dout("%s %p noop for nonexistent\n", __func__, obj_req); |
| return true; |
| } |
| |
| return false; |
| } |
| |
| /* |
| * Return: |
| * 0 - object map update sent |
| * 1 - object map update isn't needed |
| * <0 - error |
| */ |
| static int rbd_obj_write_pre_object_map(struct rbd_obj_request *obj_req) |
| { |
| struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev; |
| u8 new_state; |
| |
| if (!(rbd_dev->header.features & RBD_FEATURE_OBJECT_MAP)) |
| return 1; |
| |
| if (obj_req->flags & RBD_OBJ_FLAG_DELETION) |
| new_state = OBJECT_PENDING; |
| else |
| new_state = OBJECT_EXISTS; |
| |
| return rbd_object_map_update(obj_req, CEPH_NOSNAP, new_state, NULL); |
| } |
| |
| static int rbd_obj_write_object(struct rbd_obj_request *obj_req) |
| { |
| struct ceph_osd_request *osd_req; |
| int num_ops = count_write_ops(obj_req); |
| int which = 0; |
| int ret; |
| |
| if (obj_req->flags & RBD_OBJ_FLAG_COPYUP_ENABLED) |
| num_ops++; /* stat */ |
| |
| osd_req = rbd_obj_add_osd_request(obj_req, num_ops); |
| if (IS_ERR(osd_req)) |
| return PTR_ERR(osd_req); |
| |
| if (obj_req->flags & RBD_OBJ_FLAG_COPYUP_ENABLED) { |
| ret = rbd_osd_setup_stat(osd_req, which++); |
| if (ret) |
| return ret; |
| } |
| |
| rbd_osd_setup_write_ops(osd_req, which); |
| rbd_osd_format_write(osd_req); |
| |
| ret = ceph_osdc_alloc_messages(osd_req, GFP_NOIO); |
| if (ret) |
| return ret; |
| |
| rbd_osd_submit(osd_req); |
| return 0; |
| } |
| |
| /* |
| * copyup_bvecs pages are never highmem pages |
| */ |
| static bool is_zero_bvecs(struct bio_vec *bvecs, u32 bytes) |
| { |
| struct ceph_bvec_iter it = { |
| .bvecs = bvecs, |
| .iter = { .bi_size = bytes }, |
| }; |
| |
| ceph_bvec_iter_advance_step(&it, bytes, ({ |
| if (memchr_inv(page_address(bv.bv_page) + bv.bv_offset, 0, |
| bv.bv_len)) |
| return false; |
| })); |
| return true; |
| } |
| |
| #define MODS_ONLY U32_MAX |
| |
| static int rbd_obj_copyup_empty_snapc(struct rbd_obj_request *obj_req, |
| u32 bytes) |
| { |
| struct ceph_osd_request *osd_req; |
| int ret; |
| |
| dout("%s obj_req %p bytes %u\n", __func__, obj_req, bytes); |
| rbd_assert(bytes > 0 && bytes != MODS_ONLY); |
| |
| osd_req = __rbd_obj_add_osd_request(obj_req, &rbd_empty_snapc, 1); |
| if (IS_ERR(osd_req)) |
| return PTR_ERR(osd_req); |
| |
| ret = rbd_osd_setup_copyup(osd_req, 0, bytes); |
| if (ret) |
| return ret; |
| |
| rbd_osd_format_write(osd_req); |
| |
| ret = ceph_osdc_alloc_messages(osd_req, GFP_NOIO); |
| if (ret) |
| return ret; |
| |
| rbd_osd_submit(osd_req); |
| return 0; |
| } |
| |
| static int rbd_obj_copyup_current_snapc(struct rbd_obj_request *obj_req, |
| u32 bytes) |
| { |
| struct ceph_osd_request *osd_req; |
| int num_ops = count_write_ops(obj_req); |
| int which = 0; |
| int ret; |
| |
| dout("%s obj_req %p bytes %u\n", __func__, obj_req, bytes); |
| |
| if (bytes != MODS_ONLY) |
| num_ops++; /* copyup */ |
| |
| osd_req = rbd_obj_add_osd_request(obj_req, num_ops); |
| if (IS_ERR(osd_req)) |
| return PTR_ERR(osd_req); |
| |
| if (bytes != MODS_ONLY) { |
| ret = rbd_osd_setup_copyup(osd_req, which++, bytes); |
| if (ret) |
| return ret; |
| } |
| |
| rbd_osd_setup_write_ops(osd_req, which); |
| rbd_osd_format_write(osd_req); |
| |
| ret = ceph_osdc_alloc_messages(osd_req, GFP_NOIO); |
| if (ret) |
| return ret; |
| |
| rbd_osd_submit(osd_req); |
| return 0; |
| } |
| |
| static int setup_copyup_bvecs(struct rbd_obj_request *obj_req, u64 obj_overlap) |
| { |
| u32 i; |
| |
| rbd_assert(!obj_req->copyup_bvecs); |
| obj_req->copyup_bvec_count = calc_pages_for(0, obj_overlap); |
| obj_req->copyup_bvecs = kcalloc(obj_req->copyup_bvec_count, |
| sizeof(*obj_req->copyup_bvecs), |
| GFP_NOIO); |
| if (!obj_req->copyup_bvecs) |
| return -ENOMEM; |
| |
| for (i = 0; i < obj_req->copyup_bvec_count; i++) { |
| unsigned int len = min(obj_overlap, (u64)PAGE_SIZE); |
| |
| obj_req->copyup_bvecs[i].bv_page = alloc_page(GFP_NOIO); |
| if (!obj_req->copyup_bvecs[i].bv_page) |
| return -ENOMEM; |
| |
| obj_req->copyup_bvecs[i].bv_offset = 0; |
| obj_req->copyup_bvecs[i].bv_len = len; |
| obj_overlap -= len; |
| } |
| |
| rbd_assert(!obj_overlap); |
| return 0; |
| } |
| |
| /* |
| * The target object doesn't exist. Read the data for the entire |
| * target object up to the overlap point (if any) from the parent, |
| * so we can use it for a copyup. |
| */ |
| static int rbd_obj_copyup_read_parent(struct rbd_obj_request *obj_req) |
| { |
| struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev; |
| int ret; |
| |
| rbd_assert(obj_req->num_img_extents); |
| prune_extents(obj_req->img_extents, &obj_req->num_img_extents, |
| rbd_dev->parent_overlap); |
| if (!obj_req->num_img_extents) { |
| /* |
| * The overlap has become 0 (most likely because the |
| * image has been flattened). Re-submit the original write |
| * request -- pass MODS_ONLY since the copyup isn't needed |
| * anymore. |
| */ |
| return rbd_obj_copyup_current_snapc(obj_req, MODS_ONLY); |
| } |
| |
| ret = setup_copyup_bvecs(obj_req, rbd_obj_img_extents_bytes(obj_req)); |
| if (ret) |
| return ret; |
| |
| return rbd_obj_read_from_parent(obj_req); |
| } |
| |
| static void rbd_obj_copyup_object_maps(struct rbd_obj_request *obj_req) |
| { |
| struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev; |
| struct ceph_snap_context *snapc = obj_req->img_request->snapc; |
| u8 new_state; |
| u32 i; |
| int ret; |
| |
| rbd_assert(!obj_req->pending.result && !obj_req->pending.num_pending); |
| |
| if (!(rbd_dev->header.features & RBD_FEATURE_OBJECT_MAP)) |
| return; |
| |
| if (obj_req->flags & RBD_OBJ_FLAG_COPYUP_ZEROS) |
| return; |
| |
| for (i = 0; i < snapc->num_snaps; i++) { |
| if ((rbd_dev->header.features & RBD_FEATURE_FAST_DIFF) && |
| i + 1 < snapc->num_snaps) |
| new_state = OBJECT_EXISTS_CLEAN; |
| else |
| new_state = OBJECT_EXISTS; |
| |
| ret = rbd_object_map_update(obj_req, snapc->snaps[i], |
| new_state, NULL); |
| if (ret < 0) { |
| obj_req->pending.result = ret; |
|