/*
   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/decode.h>
#include <linux/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 */

/*
 * The basic unit of block I/O is a sector.  It is interpreted in a
 * number of contexts in Linux (blk, bio, genhd), but the default is
 * universally 512 bytes.  These symbols are just slightly more
 * meaningful than the bare numbers they represent.
 */
#define SECTOR_SHIFT    9
#define SECTOR_SIZE     (1ULL << SECTOR_SHIFT)

/*
 * 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_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    (1<<0)
#define RBD_FEATURE_STRIPINGV2  (1<<1)
#define RBD_FEATURE_EXCLUSIVE_LOCK (1<<2)
#define RBD_FEATURES_ALL        (RBD_FEATURE_LAYERING |         \
                                 RBD_FEATURE_STRIPINGV2 |       \
                                 RBD_FEATURE_EXCLUSIVE_LOCK)

/* 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;
        __u8 crypt_type;
        __u8 comp_type;
        u64 stripe_unit;
        u64 stripe_count;
        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      *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 rbd_img_request;
typedef void (*rbd_img_callback_t)(struct rbd_img_request *);

#define BAD_WHICH       U32_MAX         /* Good which or bad which, which? */

struct rbd_obj_request;
typedef void (*rbd_obj_callback_t)(struct rbd_obj_request *);

enum obj_request_type {
        OBJ_REQUEST_NODATA, OBJ_REQUEST_BIO, OBJ_REQUEST_PAGES
};

enum obj_operation_type {
        OBJ_OP_WRITE,
        OBJ_OP_READ,
        OBJ_OP_DISCARD,
};

enum obj_req_flags {
        OBJ_REQ_DONE,           /* completion flag: not done = 0, done = 1 */
        OBJ_REQ_IMG_DATA,       /* object usage: standalone = 0, image = 1 */
        OBJ_REQ_KNOWN,          /* EXISTS flag valid: no = 0, yes = 1 */
        OBJ_REQ_EXISTS,         /* target exists: no = 0, yes = 1 */
};

struct rbd_obj_request {
        const char              *object_name;
        u64                     offset;         /* object start byte */
        u64                     length;         /* bytes from offset */
        unsigned long           flags;

        /*
         * An object request associated with an image will have its
         * img_data flag set; a standalone object request will not.
         *
         * A standalone object request will have which == BAD_WHICH
         * and a null obj_request pointer.
         *
         * An object request initiated in support of a layered image
         * object (to check for its existence before a write) will
         * have which == BAD_WHICH and a non-null obj_request pointer.
         *
         * Finally, an object request for rbd image data will have
         * which != BAD_WHICH, and will have a non-null img_request
         * pointer.  The value of which will be in the range
         * 0..(img_request->obj_request_count-1).
         */
        union {
                struct rbd_obj_request  *obj_request;   /* STAT op */
                struct {
                        struct rbd_img_request  *img_request;
                        u64                     img_offset;
                        /* links for img_request->obj_requests list */
                        struct list_head        links;
                };
        };
        u32                     which;          /* posn image request list */

        enum obj_request_type   type;
        union {
                struct bio      *bio_list;
                struct {
                        struct page     **pages;
                        u32             page_count;
                };
        };
        struct page             **copyup_pages;
        u32                     copyup_page_count;

        struct ceph_osd_request *osd_req;

        u64                     xferred;        /* bytes transferred */
        int                     result;

        rbd_obj_callback_t      callback;
        struct completion       completion;

        struct kref             kref;
};

enum img_req_flags {
        IMG_REQ_WRITE,          /* I/O direction: read = 0, write = 1 */
        IMG_REQ_CHILD,          /* initiator: block = 0, child image = 1 */
        IMG_REQ_LAYERED,        /* ENOENT handling: normal = 0, layered = 1 */
        IMG_REQ_DISCARD,        /* discard: normal = 0, discard request = 1 */
};

struct rbd_img_request {
        struct rbd_device       *rbd_dev;
        u64                     offset; /* starting image byte offset */
        u64                     length; /* byte count from offset */
        unsigned long           flags;
        union {
                u64                     snap_id;        /* for reads */
                struct ceph_snap_context *snapc;        /* for writes */
        };
        union {
                struct request          *rq;            /* block request */
                struct rbd_obj_request  *obj_request;   /* obj req initiator */
        };
        struct page             **copyup_pages;
        u32                     copyup_page_count;
        spinlock_t              completion_lock;/* protects next_completion */
        u32                     next_completion;
        rbd_img_callback_t      callback;
        u64                     xferred;/* aggregate bytes transferred */
        int                     result; /* first nonzero obj_request result */

        u32                     obj_request_count;
        struct list_head        obj_requests;   /* rbd_obj_request structs */

        struct kref             kref;
};

#define for_each_obj_request(ireq, oreq) \
        list_for_each_entry(oreq, &(ireq)->obj_requests, links)
#define for_each_obj_request_from(ireq, oreq) \
        list_for_each_entry_from(oreq, &(ireq)->obj_requests, links)
#define for_each_obj_request_safe(ireq, oreq, n) \
        list_for_each_entry_safe_reverse(oreq, n, &(ireq)->obj_requests, links)

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;
        u64                     features;
        bool                    read_only;
};

/*
 * 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;
        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;
        wait_queue_head_t       lock_waitq;

        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
 * - BLACKLISTED is protected by rbd_dev->lock_rwsem
 */
enum rbd_dev_flags {
        RBD_DEV_FLAG_EXISTS,    /* mapped snapshot has not been deleted */
        RBD_DEV_FLAG_REMOVING,  /* this mapping is being removed */
        RBD_DEV_FLAG_BLACKLISTED, /* our ceph_client is blacklisted */
};

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 struct kmem_cache        *rbd_segment_name_cache;

static int rbd_major;
static DEFINE_IDA(rbd_dev_id_ida);

static struct workqueue_struct *rbd_wq;

/*
 * Default to false for now, as single-major requires >= 0.75 version of
 * userspace rbd utility.
 */
static bool single_major = false;
module_param(single_major, bool, S_IRUGO);
MODULE_PARM_DESC(single_major, "Use a single major number for all rbd devices (default: false)");

static int rbd_img_request_submit(struct rbd_img_request *img_request);

static ssize_t rbd_add(struct bus_type *bus, const char *buf,
                       size_t count);
static ssize_t rbd_remove(struct bus_type *bus, const char *buf,
                          size_t count);
static ssize_t rbd_add_single_major(struct bus_type *bus, const char *buf,
                                    size_t count);
static ssize_t rbd_remove_single_major(struct bus_type *bus, const char *buf,
                                       size_t count);
static int rbd_dev_image_probe(struct rbd_device *rbd_dev, int depth);
static void rbd_spec_put(struct rbd_spec *spec);

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_lock_supported(struct rbd_device *rbd_dev)
{
        return (rbd_dev->header.features & RBD_FEATURE_EXCLUSIVE_LOCK) &&
               rbd_dev->spec->snap_id == CEPH_NOSNAP &&
               !rbd_dev->mapping.read_only;
}

static bool __rbd_is_lock_owner(struct rbd_device *rbd_dev)
{
        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 BUS_ATTR(add, S_IWUSR, NULL, rbd_add);
static BUS_ATTR(remove, S_IWUSR, NULL, rbd_remove);
static BUS_ATTR(add_single_major, S_IWUSR, NULL, rbd_add_single_major);
static BUS_ATTR(remove_single_major, S_IWUSR, NULL, rbd_remove_single_major);

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,
        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_osd_copyup_callback(struct rbd_obj_request *obj_request);
static int rbd_img_obj_request_submit(struct rbd_obj_request *obj_request);
static void rbd_img_parent_read(struct rbd_obj_request *obj_request);
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_snap_features(struct rbd_device *rbd_dev, u64 snap_id,
                u64 *snap_features);

static int rbd_open(struct block_device *bdev, fmode_t mode)
{
        struct rbd_device *rbd_dev = bdev->bd_disk->private_data;
        bool removing = false;

        if ((mode & FMODE_WRITE) && rbd_dev->mapping.read_only)
                return -EROFS;

        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 ret = 0;
        int val;
        bool ro;
        bool ro_changed = false;

        /* get_user() may sleep, so call it before taking rbd_dev->lock */
        if (get_user(val, (int __user *)(arg)))
                return -EFAULT;

        ro = val ? true : false;
        /* Snapshot doesn't allow to write*/
        if (rbd_dev->spec->snap_id != CEPH_NOSNAP && !ro)
                return -EROFS;

        spin_lock_irq(&rbd_dev->lock);
        /* prevent others open this device */
        if (rbd_dev->open_count > 1) {
                ret = -EBUSY;
                goto out;
        }

        if (rbd_dev->mapping.read_only != ro) {
                rbd_dev->mapping.read_only = ro;
                ro_changed = true;
        }

out:
        spin_unlock_irq(&rbd_dev->lock);
        /* set_disk_ro() may sleep, so call it after releasing rbd_dev->lock */
        if (ret == 0 && ro_changed)
                set_disk_ro(rbd_dev->disk, ro ? 1 : 0);

        return ret;
}

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 = 0;

        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, 0, 0);
        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_last_int,
        /* int args above */
        Opt_last_string,
        /* string args above */
        Opt_read_only,
        Opt_read_write,
        Opt_lock_on_read,
        Opt_err
};

static match_table_t rbd_opts_tokens = {
        {Opt_queue_depth, "queue_depth=%d"},
        /* int args above */
        /* string args above */
        {Opt_read_only, "read_only"},
        {Opt_read_only, "ro"},          /* Alternate spelling */
        {Opt_read_write, "read_write"},
        {Opt_read_write, "rw"},         /* Alternate spelling */
        {Opt_lock_on_read, "lock_on_read"},
        {Opt_err, NULL}
};

struct rbd_options {
        int     queue_depth;
        bool    read_only;
        bool    lock_on_read;
};

#define RBD_QUEUE_DEPTH_DEFAULT BLKDEV_MAX_RQ
#define RBD_READ_ONLY_DEFAULT   false
#define RBD_LOCK_ON_READ_DEFAULT false

static int parse_rbd_opts_token(char *c, void *private)
{
        struct rbd_options *rbd_opts = private;
        substring_t argstr[MAX_OPT_ARGS];
        int token, intval, ret;

        token = match_token(c, rbd_opts_tokens, argstr);
        if (token < Opt_last_int) {
                ret = match_int(&argstr[0], &intval);
                if (ret < 0) {
                        pr_err("bad mount option arg (not int) at '%s'\n", c);
                        return ret;
                }
                dout("got int token %d val %d\n", token, intval);
        } else if (token > Opt_last_int && token < Opt_last_string) {
                dout("got string token %d val %s\n", token, argstr[0].from);
        } else {
                dout("got token %d\n", token);
        }

        switch (token) {
        case Opt_queue_depth:
                if (intval < 1) {
                        pr_err("queue_depth out of range\n");
                        return -EINVAL;
                }
                rbd_opts->queue_depth = intval;
                break;
        case Opt_read_only:
                rbd_opts->read_only = true;
                break;
        case Opt_read_write:
                rbd_opts->read_only = false;
                break;
        case Opt_lock_on_read:
                rbd_opts->lock_on_read = true;
                break;
        default:
                /* libceph prints "bad option" msg */
                return -EINVAL;
        }

        return 0;
}

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";
        default:
                return "???";
        }
}

/*
 * 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;

        mutex_lock_nested(&client_mutex, SINGLE_DEPTH_NESTING);
        rbdc = rbd_client_find(ceph_opts);
        if (rbdc)       /* using an existing client */
                ceph_destroy_options(ceph_opts);
        else
                rbdc = rbd_client_create(ceph_opts);
        mutex_unlock(&client_mutex);

        return rbdc;
}

/*
 * 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);
}

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;
}

/*
 * 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) {
                size_t len;

                len = strnlen(ondisk->object_prefix,
                                sizeof (ondisk->object_prefix));
                object_prefix = kmalloc(len + 1, GFP_KERNEL);
                if (!object_prefix)

                        return -ENOMEM;
                memcpy(object_prefix, ondisk->object_prefix, len);
                object_prefix[len] = '\0';
        }

        /* 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;
                header->crypt_type = ondisk->options.crypt_type;
                header->comp_type = ondisk->options.comp_type;
                /* The rest aren't used for format 1 images */
                header->stripe_unit = 0;
                header->stripe_count = 0;
                header->features = 0;
        } 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_snap_features(struct rbd_device *rbd_dev, u64 snap_id,
                        u64 *snap_features)
{
        rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
        if (snap_id == CEPH_NOSNAP) {
                *snap_features = rbd_dev->header.features;
        } else if (rbd_dev->image_format == 1) {
                *snap_features = 0;     /* No features for format 1 */
        } else {
                u64 features = 0;
                int ret;

                ret = _rbd_dev_v2_snap_features(rbd_dev, snap_id, &features);
                if (ret)
                        return ret;

                *snap_features = features;
        }
        return 0;
}

static int rbd_dev_mapping_set(struct rbd_device *rbd_dev)
{
        u64 snap_id = rbd_dev->spec->snap_id;
        u64 size = 0;
        u64 features = 0;
        int ret;

        ret = rbd_snap_size(rbd_dev, snap_id, &size);
        if (ret)
                return ret;
        ret = rbd_snap_features(rbd_dev, snap_id, &features);
        if (ret)
                return ret;

        rbd_dev->mapping.size = size;
        rbd_dev->mapping.features = features;

        return 0;
}

static void rbd_dev_mapping_clear(struct rbd_device *rbd_dev)
{
        rbd_dev->mapping.size = 0;
        rbd_dev->mapping.features = 0;
}

static void rbd_segment_name_free(const char *name)
{
        /* The explicit cast here is needed to drop the const qualifier */

        kmem_cache_free(rbd_segment_name_cache, (void *)name);
}

static const char *rbd_segment_name(struct rbd_device *rbd_dev, u64 offset)
{
        char *name;
        u64 segment;
        int ret;
        char *name_format;

        name = kmem_cache_alloc(rbd_segment_name_cache, GFP_NOIO);
        if (!name)
                return NULL;
        segment = offset >> rbd_dev->header.obj_order;
        name_format = "%s.%012llx";
        if (rbd_dev->image_format == 2)
                name_format = "%s.%016llx";
        ret = snprintf(name, CEPH_MAX_OID_NAME_LEN + 1, name_format,
                        rbd_dev->header.object_prefix, segment);
        if (ret < 0 || ret > CEPH_MAX_OID_NAME_LEN) {
                pr_err("error formatting segment name for #%llu (%d)\n",
                        segment, ret);
                rbd_segment_name_free(name);
                name = NULL;
        }

        return name;
}

static u64 rbd_segment_offset(struct rbd_device *rbd_dev, u64 offset)
{
        u64 segment_size = (u64) 1 << rbd_dev->header.obj_order;

        return offset & (segment_size - 1);
}

static u64 rbd_segment_length(struct rbd_device *rbd_dev,
                                u64 offset, u64 length)
{
        u64 segment_size = (u64) 1 << rbd_dev->header.obj_order;

        offset &= segment_size - 1;

        rbd_assert(length <= U64_MAX - offset);
        if (offset + length > segment_size)
                length = segment_size - offset;

        return length;
}

/*
 * returns the size of an object in the image
 */
static u64 rbd_obj_bytes(struct rbd_image_header *header)
{
        return 1 << header->obj_order;
}

/*
 * bio helpers
 */

static void bio_chain_put(struct bio *chain)
{
        struct bio *tmp;

        while (chain) {
                tmp = chain;
                chain = chain->bi_next;
                bio_put(tmp);
        }
}

/*
 * zeros a bio chain, starting at specific offset
 */
static void zero_bio_chain(struct bio *chain, int start_ofs)
{
        struct bio_vec bv;
        struct bvec_iter iter;
        unsigned long flags;
        void *buf;
        int pos = 0;

        while (chain) {
                bio_for_each_segment(bv, chain, iter) {
                        if (pos + bv.bv_len > start_ofs) {
                                int remainder = max(start_ofs - pos, 0);
                                buf = bvec_kmap_irq(&bv, &flags);
                                memset(buf + remainder, 0,
                                       bv.bv_len - remainder);
                                flush_dcache_page(bv.bv_page);
                                bvec_kunmap_irq(buf, &flags);
                        }
                        pos += bv.bv_len;
                }

                chain = chain->bi_next;
        }
}

/*
 * similar to zero_bio_chain(), zeros data defined by a page array,
 * starting at the given byte offset from the start of the array and
 * continuing up to the given end offset.  The pages array is
 * assumed to be big enough to hold all bytes up to the end.
 */
static void zero_pages(struct page **pages, u64 offset, u64 end)
{
        struct page **page = &pages[offset >> PAGE_SHIFT];

        rbd_assert(end > offset);
        rbd_assert(end - offset <= (u64)SIZE_MAX);
        while (offset < end) {
                size_t page_offset;
                size_t length;
                unsigned long flags;
                void *kaddr;

                page_offset = offset & ~PAGE_MASK;
                length = min_t(size_t, PAGE_SIZE - page_offset, end - offset);
                local_irq_save(flags);
                kaddr = kmap_atomic(*page);
                memset(kaddr + page_offset, 0, length);
                flush_dcache_page(*page);
                kunmap_atomic(kaddr);
                local_irq_restore(flags);

                offset += length;
                page++;
        }
}

/*
 * Clone a portion of a bio, starting at the given byte offset
 * and continuing for the number of bytes indicated.
 */
static struct bio *bio_clone_range(struct bio *bio_src,
                                        unsigned int offset,
                                        unsigned int len,
                                        gfp_t gfpmask)
{
        struct bio *bio;

        bio = bio_clone(bio_src, gfpmask);
        if (!bio)
                return NULL;    /* ENOMEM */

        bio_advance(bio, offset);
        bio->bi_iter.bi_size = len;

        return bio;
}

/*
 * Clone a portion of a bio chain, starting at the given byte offset
 * into the first bio in the source chain and continuing for the
 * number of bytes indicated.  The result is another bio chain of
 * exactly the given length, or a null pointer on error.
 *
 * The bio_src and offset parameters are both in-out.  On entry they
 * refer to the first source bio and the offset into that bio where
 * the start of data to be cloned is located.
 *
 * On return, bio_src is updated to refer to the bio in the source
 * chain that contains first un-cloned byte, and *offset will
 * contain the offset of that byte within that bio.
 */
static struct bio *bio_chain_clone_range(struct bio **bio_src,
                                        unsigned int *offset,
                                        unsigned int len,
                                        gfp_t gfpmask)
{
        struct bio *bi = *bio_src;
        unsigned int off = *offset;
        struct bio *chain = NULL;
        struct bio **end;

        /* Build up a chain of clone bios up to the limit */

        if (!bi || off >= bi->bi_iter.bi_size || !len)
                return NULL;            /* Nothing to clone */

        end = &chain;
        while (len) {
                unsigned int bi_size;
                struct bio *bio;

                if (!bi) {
                        rbd_warn(NULL, "bio_chain exhausted with %u left", len);
                        goto out_err;   /* EINVAL; ran out of bio's */
                }
                bi_size = min_t(unsigned int, bi->bi_iter.bi_size - off, len);
                bio = bio_clone_range(bi, off, bi_size, gfpmask);
                if (!bio)
                        goto out_err;   /* ENOMEM */

                *end = bio;
                end = &bio->bi_next;

                off += bi_size;
                if (off == bi->bi_iter.bi_size) {
                        bi = bi->bi_next;
                        off = 0;
                }
                len -= bi_size;
        }
        *bio_src = bi;
        *offset = off;

        return chain;
out_err:
        bio_chain_put(chain);

        return NULL;
}

/*
 * The default/initial value for all object request flags is 0.  For
 * each flag, once its value is set to 1 it is never reset to 0
 * again.
 */
static void obj_request_img_data_set(struct rbd_obj_request *obj_request)
{
        if (test_and_set_bit(OBJ_REQ_IMG_DATA, &obj_request->flags)) {
                struct rbd_device *rbd_dev;

                rbd_dev = obj_request->img_request->rbd_dev;
                rbd_warn(rbd_dev, "obj_request %p already marked img_data",
                        obj_request);
        }
}

static bool obj_request_img_data_test(struct rbd_obj_request *obj_request)
{
        smp_mb();
        return test_bit(OBJ_REQ_IMG_DATA, &obj_request->flags) != 0;
}

static void obj_request_done_set(struct rbd_obj_request *obj_request)
{
        if (test_and_set_bit(OBJ_REQ_DONE, &obj_request->flags)) {
                struct rbd_device *rbd_dev = NULL;

                if (obj_request_img_data_test(obj_request))
                        rbd_dev = obj_request->img_request->rbd_dev;
                rbd_warn(rbd_dev, "obj_request %p already marked done",
                        obj_request);
        }
}

static bool obj_request_done_test(struct rbd_obj_request *obj_request)
{
        smp_mb();
        return test_bit(OBJ_REQ_DONE, &obj_request->flags) != 0;
}

/*
 * This sets the KNOWN flag after (possibly) setting the EXISTS
 * flag.  The latter is set based on the "exists" value provided.
 *
 * Note that for our purposes once an object exists it never goes
 * away again.  It's possible that the response from two existence
 * checks are separated by the creation of the target object, and
 * the first ("doesn't exist") response arrives *after* the second
 * ("does exist").  In that case we ignore the second one.
 */
static void obj_request_existence_set(struct rbd_obj_request *obj_request,
                                bool exists)
{
        if (exists)
                set_bit(OBJ_REQ_EXISTS, &obj_request->flags);
        set_bit(OBJ_REQ_KNOWN, &obj_request->flags);
        smp_mb();
}

static bool obj_request_known_test(struct rbd_obj_request *obj_request)
{
        smp_mb();
        return test_bit(OBJ_REQ_KNOWN, &obj_request->flags) != 0;
}

static bool obj_request_exists_test(struct rbd_obj_request *obj_request)
{
        smp_mb();
        return test_bit(OBJ_REQ_EXISTS, &obj_request->flags) != 0;
}

static bool obj_request_overlaps_parent(struct rbd_obj_request *obj_request)
{
        struct rbd_device *rbd_dev = obj_request->img_request->rbd_dev;

        return obj_request->img_offset <
            round_up(rbd_dev->parent_overlap, rbd_obj_bytes(&rbd_dev->header));
}

static void rbd_obj_request_get(struct rbd_obj_request *obj_request)
{
        dout("%s: obj %p (was %d)\n", __func__, obj_request,
                atomic_read(&obj_request->kref.refcount));
        kref_get(&obj_request->kref);
}

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,
                atomic_read(&obj_request->kref.refcount));
        kref_put(&obj_request->kref, rbd_obj_request_destroy);
}

static void rbd_img_request_get(struct rbd_img_request *img_request)
{
        dout("%s: img %p (was %d)\n", __func__, img_request,
             atomic_read(&img_request->kref.refcount));
        kref_get(&img_request->kref);
}

static bool img_request_child_test(struct rbd_img_request *img_request);
static void rbd_parent_request_destroy(struct kref *kref);
static void rbd_img_request_destroy(struct kref *kref);
static void rbd_img_request_put(struct rbd_img_request *img_request)
{
        rbd_assert(img_request != NULL);
        dout("%s: img %p (was %d)\n", __func__, img_request,
                atomic_read(&img_request->kref.refcount));
        if (img_request_child_test(img_request))
                kref_put(&img_request->kref, rbd_parent_request_destroy);
        else
                kref_put(&img_request->kref, rbd_img_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;
        obj_request->which = img_request->obj_request_count;
        rbd_assert(!obj_request_img_data_test(obj_request));
        obj_request_img_data_set(obj_request);
        rbd_assert(obj_request->which != BAD_WHICH);
        img_request->obj_request_count++;
        list_add_tail(&obj_request->links, &img_request->obj_requests);
        dout("%s: img %p obj %p w=%u\n", __func__, img_request, obj_request,
                obj_request->which);
}

static inline void rbd_img_obj_request_del(struct rbd_img_request *img_request,
                                        struct rbd_obj_request *obj_request)
{
        rbd_assert(obj_request->which != BAD_WHICH);

        dout("%s: img %p obj %p w=%u\n", __func__, img_request, obj_request,
                obj_request->which);
        list_del(&obj_request->links);
        rbd_assert(img_request->obj_request_count > 0);
        img_request->obj_request_count--;
        rbd_assert(obj_request->which == img_request->obj_request_count);
        obj_request->which = BAD_WHICH;
        rbd_assert(obj_request_img_data_test(obj_request));
        rbd_assert(obj_request->img_request == img_request);
        obj_request->img_request = NULL;
        obj_request->callback = NULL;
        rbd_obj_request_put(obj_request);
}

static bool obj_request_type_valid(enum obj_request_type type)
{
        switch (type) {
        case OBJ_REQUEST_NODATA:
        case OBJ_REQUEST_BIO:
        case OBJ_REQUEST_PAGES:
                return true;
        default:
                return false;
        }
}

static void rbd_img_obj_callback(struct rbd_obj_request *obj_request);

static void rbd_obj_request_submit(struct rbd_obj_request *obj_request)
{
        struct ceph_osd_request *osd_req = obj_request->osd_req;

        dout("%s %p osd_req %p\n", __func__, obj_request, osd_req);
        if (obj_request_img_data_test(obj_request)) {
                WARN_ON(obj_request->callback != rbd_img_obj_callback);
                rbd_img_request_get(obj_request->img_request);
        }
        ceph_osdc_start_request(osd_req->r_osdc, osd_req, false);
}

static void rbd_obj_request_end(struct rbd_obj_request *obj_request)
{
        dout("%s %p\n", __func__, obj_request);
        ceph_osdc_cancel_request(obj_request->osd_req);
}

/*
 * Wait for an object request to complete.  If interrupted, cancel the
 * underlying osd request.
 *
 * @timeout: in jiffies, 0 means "wait forever"
 */
static int __rbd_obj_request_wait(struct rbd_obj_request *obj_request,
                                  unsigned long timeout)
{
        long ret;

        dout("%s %p\n", __func__, obj_request);
        ret = wait_for_completion_interruptible_timeout(
                                        &obj_request->completion,
                                        ceph_timeout_jiffies(timeout));
        if (ret <= 0) {
                if (ret == 0)
                        ret = -ETIMEDOUT;
                rbd_obj_request_end(obj_request);
        } else {
                ret = 0;
        }

        dout("%s %p ret %d\n", __func__, obj_request, (int)ret);
        return ret;
}

static int rbd_obj_request_wait(struct rbd_obj_request *obj_request)
{
        return __rbd_obj_request_wait(obj_request, 0);
}

static void rbd_img_request_complete(struct rbd_img_request *img_request)
{

        dout("%s: img %p\n", __func__, img_request);

        /*
         * If no error occurred, compute the aggregate transfer
         * count for the image request.  We could instead use
         * atomic64_cmpxchg() to update it as each object request
         * completes; not clear which way is better off hand.
         */
        if (!img_request->result) {
                struct rbd_obj_request *obj_request;
                u64 xferred = 0;

                for_each_obj_request(img_request, obj_request)
                        xferred += obj_request->xferred;
                img_request->xferred = xferred;
        }

        if (img_request->callback)
                img_request->callback(img_request);
        else
                rbd_img_request_put(img_request);
}

/*
 * 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_write_set(struct rbd_img_request *img_request)
{
        set_bit(IMG_REQ_WRITE, &img_request->flags);
        smp_mb();
}

static bool img_request_write_test(struct rbd_img_request *img_request)
{
        smp_mb();
        return test_bit(IMG_REQ_WRITE, &img_request->flags) != 0;
}

/*
 * Set the discard flag when the img_request is an discard request
 */
static void img_request_discard_set(struct rbd_img_request *img_request)
{
        set_bit(IMG_REQ_DISCARD, &img_request->flags);
        smp_mb();
}

static bool img_request_discard_test(struct rbd_img_request *img_request)
{
        smp_mb();
        return test_bit(IMG_REQ_DISCARD, &img_request->flags) != 0;
}

static void img_request_child_set(struct rbd_img_request *img_request)
{
        set_bit(IMG_REQ_CHILD, &img_request->flags);
        smp_mb();
}

static void img_request_child_clear(struct rbd_img_request *img_request)
{
        clear_bit(IMG_REQ_CHILD, &img_request->flags);
        smp_mb();
}

static bool img_request_child_test(struct rbd_img_request *img_request)
{
        smp_mb();
        return test_bit(IMG_REQ_CHILD, &img_request->flags) != 0;
}

static void img_request_layered_set(struct rbd_img_request *img_request)
{
        set_bit(IMG_REQ_LAYERED, &img_request->flags);
        smp_mb();
}

static void img_request_layered_clear(struct rbd_img_request *img_request)
{
        clear_bit(IMG_REQ_LAYERED, &img_request->flags);
        smp_mb();
}

static bool img_request_layered_test(struct rbd_img_request *img_request)
{
        smp_mb();
        return test_bit(IMG_REQ_LAYERED, &img_request->flags) != 0;
}

static enum obj_operation_type
rbd_img_request_op_type(struct rbd_img_request *img_request)
{
        if (img_request_write_test(img_request))
                return OBJ_OP_WRITE;
        else if (img_request_discard_test(img_request))
                return OBJ_OP_DISCARD;
        else
                return OBJ_OP_READ;
}

static void
rbd_img_obj_request_read_callback(struct rbd_obj_request *obj_request)
{
        u64 xferred = obj_request->xferred;
        u64 length = obj_request->length;

        dout("%s: obj %p img %p result %d %llu/%llu\n", __func__,
                obj_request, obj_request->img_request, obj_request->result,
                xferred, length);
        /*
         * ENOENT means a hole in the image.  We zero-fill the entire
         * length of the request.  A short read also implies zero-fill
         * to the end of the request.  An error requires the whole
         * length of the request to be reported finished with an error
         * to the block layer.  In each case we update the xferred
         * count to indicate the whole request was satisfied.
         */
        rbd_assert(obj_request->type != OBJ_REQUEST_NODATA);
        if (obj_request->result == -ENOENT) {
                if (obj_request->type == OBJ_REQUEST_BIO)
                        zero_bio_chain(obj_request->bio_list, 0);
                else
                        zero_pages(obj_request->pages, 0, length);
                obj_request->result = 0;
        } else if (xferred < length && !obj_request->result) {
                if (obj_request->type == OBJ_REQUEST_BIO)
                        zero_bio_chain(obj_request->bio_list, xferred);
                else
                        zero_pages(obj_request->pages, xferred, length);
        }
        obj_request->xferred = length;
        obj_request_done_set(obj_request);
}

static void rbd_obj_request_complete(struct rbd_obj_request *obj_request)
{
        dout("%s: obj %p cb %p\n", __func__, obj_request,
                obj_request->callback);
        if (obj_request->callback)
                obj_request->callback(obj_request);
        else
                complete_all(&obj_request->completion);
}

static void rbd_obj_request_error(struct rbd_obj_request *obj_request, int err)
{
        obj_request->result = err;
        obj_request->xferred = 0;
        /*
         * kludge - mirror rbd_obj_request_submit() to match a put in
         * rbd_img_obj_callback()
         */
        if (obj_request_img_data_test(obj_request)) {
                WARN_ON(obj_request->callback != rbd_img_obj_callback);
                rbd_img_request_get(obj_request->img_request);
        }
        obj_request_done_set(obj_request);
        rbd_obj_request_complete(obj_request);
}

static void rbd_osd_read_callback(struct rbd_obj_request *obj_request)
{
        struct rbd_img_request *img_request = NULL;
        struct rbd_device *rbd_dev = NULL;
        bool layered = false;

        if (obj_request_img_data_test(obj_request)) {
                img_request = obj_request->img_request;
                layered = img_request && img_request_layered_test(img_request);
                rbd_dev = img_request->rbd_dev;
        }

        dout("%s: obj %p img %p result %d %llu/%llu\n", __func__,
                obj_request, img_request, obj_request->result,
                obj_request->xferred, obj_request->length);
        if (layered && obj_request->result == -ENOENT &&
                        obj_request->img_offset < rbd_dev->parent_overlap)
                rbd_img_parent_read(obj_request);
        else if (img_request)
                rbd_img_obj_request_read_callback(obj_request);
        else
                obj_request_done_set(obj_request);
}

static void rbd_osd_write_callback(struct rbd_obj_request *obj_request)
{
        dout("%s: obj %p result %d %llu\n", __func__, obj_request,
                obj_request->result, obj_request->length);
        /*
         * There is no such thing as a successful short write.  Set
         * it to our originally-requested length.
         */
        obj_request->xferred = obj_request->length;
        obj_request_done_set(obj_request);
}

static void rbd_osd_discard_callback(struct rbd_obj_request *obj_request)
{
        dout("%s: obj %p result %d %llu\n", __func__, obj_request,
                obj_request->result, obj_request->length);
        /*
         * There is no such thing as a successful short discard.  Set
         * it to our originally-requested length.
         */
        obj_request->xferred = obj_request->length;
        /* discarding a non-existent object is not a problem */
        if (obj_request->result == -ENOENT)
                obj_request->result = 0;
        obj_request_done_set(obj_request);
}

/*
 * For a simple stat call there's nothing to do.  We'll do more if
 * this is part of a write sequence for a layered image.
 */
static void rbd_osd_stat_callback(struct rbd_obj_request *obj_request)
{
        dout("%s: obj %p\n", __func__, obj_request);
        obj_request_done_set(obj_request);
}

static void rbd_osd_call_callback(struct rbd_obj_request *obj_request)
{
        dout("%s: obj %p\n", __func__, obj_request);

        if (obj_request_img_data_test(obj_request))
                rbd_osd_copyup_callback(obj_request);
        else
                obj_request_done_set(obj_request);
}

static void rbd_osd_req_callback(struct ceph_osd_request *osd_req)
{
        struct rbd_obj_request *obj_request = osd_req->r_priv;
        u16 opcode;

        dout("%s: osd_req %p\n", __func__, osd_req);
        rbd_assert(osd_req == obj_request->osd_req);
        if (obj_request_img_data_test(obj_request)) {
                rbd_assert(obj_request->img_request);
                rbd_assert(obj_request->which != BAD_WHICH);
        } else {
                rbd_assert(obj_request->which == BAD_WHICH);
        }

        if (osd_req->r_result < 0)
                obj_request->result = osd_req->r_result;

        /*
         * We support a 64-bit length, but ultimately it has to be
         * passed to the block layer, which just supports a 32-bit
         * length field.
         */
        obj_request->xferred = osd_req->r_ops[0].outdata_len;
        rbd_assert(obj_request->xferred < (u64)UINT_MAX);

        opcode = osd_req->r_ops[0].op;
        switch (opcode) {
        case CEPH_OSD_OP_READ:
                rbd_osd_read_callback(obj_request);
                break;
        case CEPH_OSD_OP_SETALLOCHINT:
                rbd_assert(osd_req->r_ops[1].op == CEPH_OSD_OP_WRITE ||
                           osd_req->r_ops[1].op == CEPH_OSD_OP_WRITEFULL);
                /* fall through */
        case CEPH_OSD_OP_WRITE:
        case CEPH_OSD_OP_WRITEFULL:
                rbd_osd_write_callback(obj_request);
                break;
        case CEPH_OSD_OP_STAT:
                rbd_osd_stat_callback(obj_request);
                break;
        case CEPH_OSD_OP_DELETE:
        case CEPH_OSD_OP_TRUNCATE:
        case CEPH_OSD_OP_ZERO:
                rbd_osd_discard_callback(obj_request);
                break;
        case CEPH_OSD_OP_CALL:
                rbd_osd_call_callback(obj_request);
                break;
        default:
                rbd_warn(NULL, "%s: unsupported op %hu",
                        obj_request->object_name, (unsigned short) opcode);
                break;
        }

        if (obj_request_done_test(obj_request))
                rbd_obj_request_complete(obj_request);
}

static void rbd_osd_req_format_read(struct rbd_obj_request *obj_request)
{
        struct ceph_osd_request *osd_req = obj_request->osd_req;

        rbd_assert(obj_request_img_data_test(obj_request));
        osd_req->r_snapid = obj_request->img_request->snap_id;
}

static void rbd_osd_req_format_write(struct rbd_obj_request *obj_request)
{
        struct ceph_osd_request *osd_req = obj_request->osd_req;

        osd_req->r_mtime = CURRENT_TIME;
        osd_req->r_data_offset = obj_request->offset;
}

/*
 * Create an osd request.  A read request has one osd op (read).
 * A write request has either one (watch) or two (hint+write) osd ops.
 * (All rbd data writes are prefixed with an allocation hint op, but
 * technically osd watch is a write request, hence this distinction.)
 */
static struct ceph_osd_request *rbd_osd_req_create(
                                        struct rbd_device *rbd_dev,
                                        enum obj_operation_type op_type,
                                        unsigned int num_ops,
                                        struct rbd_obj_request *obj_request)
{
        struct ceph_snap_context *snapc = NULL;
        struct ceph_osd_client *osdc;
        struct ceph_osd_request *osd_req;

        if (obj_request_img_data_test(obj_request) &&
                (op_type == OBJ_OP_DISCARD || op_type == OBJ_OP_WRITE)) {

                struct rbd_img_request *img_request = obj_request->img_request;
                if (op_type == OBJ_OP_WRITE) {
                        rbd_assert(img_request_write_test(img_request));
                } else {
                        rbd_assert(img_request_discard_test(img_request));
                }
                snapc = img_request->snapc;
        }

        rbd_assert(num_ops == 1 || ((op_type == OBJ_OP_WRITE) && num_ops == 2));

        /* Allocate and initialize the request, for the num_ops ops */

        osdc = &rbd_dev->rbd_client->client->osdc;
        osd_req = ceph_osdc_alloc_request(osdc, snapc, num_ops, false,
                                          GFP_NOIO);
        if (!osd_req)
                goto fail;

        if (op_type == OBJ_OP_WRITE || op_type == OBJ_OP_DISCARD)
                osd_req->r_flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK;
        else
                osd_req->r_flags = CEPH_OSD_FLAG_READ;

        osd_req->r_callback = rbd_osd_req_callback;
        osd_req->r_priv = obj_request;

        osd_req->r_base_oloc.pool = rbd_dev->layout.pool_id;
        if (ceph_oid_aprintf(&osd_req->r_base_oid, GFP_NOIO, "%s",
                             obj_request->object_name))
                goto fail;

        if (ceph_osdc_alloc_messages(osd_req, GFP_NOIO))
                goto fail;

        return osd_req;

fail:
        ceph_osdc_put_request(osd_req);
        return NULL;
}

/*
 * Create a copyup osd request based on the information in the object
 * request supplied.  A copyup request has two or three osd ops, a
 * copyup method call, potentially a hint op, and a write or truncate
 * or zero op.
 */
static struct ceph_osd_request *
rbd_osd_req_create_copyup(struct rbd_obj_request *obj_request)
{
        struct rbd_img_request *img_request;
        struct ceph_snap_context *snapc;
        struct rbd_device *rbd_dev;
        struct ceph_osd_client *osdc;
        struct ceph_osd_request *osd_req;
        int num_osd_ops = 3;

        rbd_assert(obj_request_img_data_test(obj_request));
        img_request = obj_request->img_request;
        rbd_assert(img_request);
        rbd_assert(img_request_write_test(img_request) ||
                        img_request_discard_test(img_request));

        if (img_request_discard_test(img_request))
                num_osd_ops = 2;

        /* Allocate and initialize the request, for all the ops */

        snapc = img_request->snapc;
        rbd_dev = img_request->rbd_dev;
        osdc = &rbd_dev->rbd_client->client->osdc;
        osd_req = ceph_osdc_alloc_request(osdc, snapc, num_osd_ops,
                                                false, GFP_NOIO);
        if (!osd_req)
                goto fail;

        osd_req->r_flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK;
        osd_req->r_callback = rbd_osd_req_callback;
        osd_req->r_priv = obj_request;

        osd_req->r_base_oloc.pool = rbd_dev->layout.pool_id;
        if (ceph_oid_aprintf(&osd_req->r_base_oid, GFP_NOIO, "%s",
                             obj_request->object_name))
                goto fail;

        if (ceph_osdc_alloc_messages(osd_req, GFP_NOIO))
                goto fail;

        return osd_req;

fail:
        ceph_osdc_put_request(osd_req);
        return NULL;
}


static void rbd_osd_req_destroy(struct ceph_osd_request *osd_req)
{
        ceph_osdc_put_request(osd_req);
}

/* object_name is assumed to be a non-null pointer and NUL-terminated */

static struct rbd_obj_request *rbd_obj_request_create(const char *object_name,
                                                u64 offset, u64 length,
                                                enum obj_request_type type)
{
        struct rbd_obj_request *obj_request;
        size_t size;
        char *name;

        rbd_assert(obj_request_type_valid(type));

        size = strlen(object_name) + 1;
        name = kmalloc(size, GFP_NOIO);
        if (!name)
                return NULL;

        obj_request = kmem_cache_zalloc(rbd_obj_request_cache, GFP_NOIO);
        if (!obj_request) {
                kfree(name);
                return NULL;
        }

        obj_request->object_name = memcpy(name, object_name, size);
        obj_request->offset = offset;
        obj_request->length = length;
        obj_request->flags = 0;
        obj_request->which = BAD_WHICH;
        obj_request->type = type;
        INIT_LIST_HEAD(&obj_request->links);
        init_completion(&obj_request->completion);
        kref_init(&obj_request->kref);

        dout("%s: \"%s\" %llu/%llu %d -> obj %p\n", __func__, object_name,
                offset, length, (int)type, obj_request);

        return obj_request;
}

static void rbd_obj_request_destroy(struct kref *kref)
{
        struct rbd_obj_request *obj_request;

        obj_request = container_of(kref, struct rbd_obj_request, kref);

        dout("%s: obj %p\n", __func__, obj_request);

        rbd_assert(obj_request->img_request == NULL);
        rbd_assert(obj_request->which == BAD_WHICH);

        if (obj_request->osd_req)
                rbd_osd_req_destroy(obj_request->osd_req);

        rbd_assert(obj_request_type_valid(obj_request->type));
        switch (obj_request->type) {
        case OBJ_REQUEST_NODATA:
                break;          /* Nothing to do */
        case OBJ_REQUEST_BIO:
                if (obj_request->bio_list)
                        bio_chain_put(obj_request->bio_list);
                break;
        case OBJ_REQUEST_PAGES:
                /* img_data requests don't own their page array */
                if (obj_request->pages &&
                    !obj_request_img_data_test(obj_request))
                        ceph_release_page_vector(obj_request->pages,
                                                obj_request->page_count);
                break;
        }

        kfree(obj_request->object_name);
        obj_request->object_name = NULL;
        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;

        down_read(&rbd_dev->header_rwsem);
        if (rbd_dev->parent_overlap)
                counter = atomic_inc_return_safe(&rbd_dev->parent_ref);
        up_read(&rbd_dev->header_rwsem);

        if (counter < 0)
                rbd_warn(rbd_dev, "parent reference overflow");

        return counter > 0;
}

/*
 * Caller is responsible for filling in the list of object requests
 * that comprises the image request, and the Linux request pointer
 * (if there is one).
 */
static struct rbd_img_request *rbd_img_request_create(
                                        struct rbd_device *rbd_dev,
                                        u64 offset, u64 length,
                                        enum obj_operation_type op_type,
                                        struct ceph_snap_context *snapc)
{
        struct rbd_img_request *img_request;

        img_request = kmem_cache_alloc(rbd_img_request_cache, GFP_NOIO);
        if (!img_request)
                return NULL;

        img_request->rq = NULL;
        img_request->rbd_dev = rbd_dev;
        img_request->offset = offset;
        img_request->length = length;
        img_request->flags = 0;
        if (op_type == OBJ_OP_DISCARD) {
                img_request_discard_set(img_request);
                img_request->snapc = snapc;
        } else if (op_type == OBJ_OP_WRITE) {
                img_request_write_set(img_request);
                img_request->snapc = snapc;
        } else {
                img_request->snap_id = rbd_dev->spec->snap_id;
        }
        if (rbd_dev_parent_get(rbd_dev))
                img_request_layered_set(img_request);
        spin_lock_init(&img_request->completion_lock);
        img_request->next_completion = 0;
        img_request->callback = NULL;
        img_request->result = 0;
        img_request->obj_request_count = 0;
        INIT_LIST_HEAD(&img_request->obj_requests);
        kref_init(&img_request->kref);

        dout("%s: rbd_dev %p %s %llu/%llu -> img %p\n", __func__, rbd_dev,
                obj_op_name(op_type), offset, length, img_request);

        return img_request;
}

static void rbd_img_request_destroy(struct kref *kref)
{
        struct rbd_img_request *img_request;
        struct rbd_obj_request *obj_request;
        struct rbd_obj_request *next_obj_request;

        img_request = container_of(kref, struct rbd_img_request, kref);

        dout("%s: img %p\n", __func__, img_request);

        for_each_obj_request_safe(img_request, obj_request, next_obj_request)
                rbd_img_obj_request_del(img_request, obj_request);
        rbd_assert(img_request->obj_request_count == 0);

        if (img_request_layered_test(img_request)) {
                img_request_layered_clear(img_request);
                rbd_dev_parent_put(img_request->rbd_dev);
        }

        if (img_request_write_test(img_request) ||
                img_request_discard_test(img_request))
                ceph_put_snap_context(img_request->snapc);

        kmem_cache_free(rbd_img_request_cache, img_request);
}

static struct rbd_img_request *rbd_parent_request_create(
                                        struct rbd_obj_request *obj_request,
                                        u64 img_offset, u64 length)
{
        struct rbd_img_request *parent_request;
        struct rbd_device *rbd_dev;

        rbd_assert(obj_request->img_request);
        rbd_dev = obj_request->img_request->rbd_dev;

        parent_request = rbd_img_request_create(rbd_dev->parent, img_offset,
                                                length, OBJ_OP_READ, NULL);
        if (!parent_request)
                return NULL;

        img_request_child_set(parent_request);
        rbd_obj_request_get(obj_request);
        parent_request->obj_request = obj_request;

        return parent_request;
}

static void rbd_parent_request_destroy(struct kref *kref)
{
        struct rbd_img_request *parent_request;
        struct rbd_obj_request *orig_request;

        parent_request = container_of(kref, struct rbd_img_request, kref);
        orig_request = parent_request->obj_request;

        parent_request->obj_request = NULL;
        rbd_obj_request_put(orig_request);
        img_request_child_clear(parent_request);

        rbd_img_request_destroy(kref);
}

static bool rbd_img_obj_end_request(struct rbd_obj_request *obj_request)
{
        struct rbd_img_request *img_request;
        unsigned int xferred;
        int result;
        bool more;

        rbd_assert(obj_request_img_data_test(obj_request));
        img_request = obj_request->img_request;

        rbd_assert(obj_request->xferred <= (u64)UINT_MAX);
        xferred = (unsigned int)obj_request->xferred;
        result = obj_request->result;
        if (result) {
                struct rbd_device *rbd_dev = img_request->rbd_dev;
                enum obj_operation_type op_type;

                if (img_request_discard_test(img_request))
                        op_type = OBJ_OP_DISCARD;
                else if (img_request_write_test(img_request))
                        op_type = OBJ_OP_WRITE;
                else
                        op_type = OBJ_OP_READ;

                rbd_warn(rbd_dev, "%s %llx at %llx (%llx)",
                        obj_op_name(op_type), obj_request->length,
                        obj_request->img_offset, obj_request->offset);
                rbd_warn(rbd_dev, "  result %d xferred %x",
                        result, xferred);
                if (!img_request->result)
                        img_request->result = result;
                /*
                 * Need to end I/O on the entire obj_request worth of
                 * bytes in case of error.
                 */
                xferred = obj_request->length;
        }

        if (img_request_child_test(img_request)) {
                rbd_assert(img_request->obj_request != NULL);
                more = obj_request->which < img_request->obj_request_count - 1;
        } else {
                rbd_assert(img_request->rq != NULL);

                more = blk_update_request(img_request->rq, result, xferred);
                if (!more)
                        __blk_mq_end_request(img_request->rq, result);
        }

        return more;
}

static void rbd_img_obj_callback(struct rbd_obj_request *obj_request)
{
        struct rbd_img_request *img_request;
        u32 which = obj_request->which;
        bool more = true;

        rbd_assert(obj_request_img_data_test(obj_request));
        img_request = obj_request->img_request;

        dout("%s: img %p obj %p\n", __func__, img_request, obj_request);
        rbd_assert(img_request != NULL);
        rbd_assert(img_request->obj_request_count > 0);
        rbd_assert(which != BAD_WHICH);
        rbd_assert(which < img_request->obj_request_count);

        spin_lock_irq(&img_request->completion_lock);
        if (which != img_request->next_completion)
                goto out;

        for_each_obj_request_from(img_request, obj_request) {
                rbd_assert(more);
                rbd_assert(which < img_request->obj_request_count);

                if (!obj_request_done_test(obj_request))
                        break;
                more = rbd_img_obj_end_request(obj_request);
                which++;
        }

        rbd_assert(more ^ (which == img_request->obj_request_count));
        img_request->next_completion = which;
out:
        spin_unlock_irq(&img_request->completion_lock);
        rbd_img_request_put(img_request);

        if (!more)
                rbd_img_request_complete(img_request);
}

/*
 * Add individual osd ops to the given ceph_osd_request and prepare
 * them for submission. num_ops is the current number of
 * osd operations already to the object request.
 */
static void rbd_img_obj_request_fill(struct rbd_obj_request *obj_request,
                                struct ceph_osd_request *osd_request,
                                enum obj_operation_type op_type,
                                unsigned int num_ops)
{
        struct rbd_img_request *img_request = obj_request->img_request;
        struct rbd_device *rbd_dev = img_request->rbd_dev;
        u64 object_size = rbd_obj_bytes(&rbd_dev->header);
        u64 offset = obj_request->offset;
        u64 length = obj_request->length;
        u64 img_end;
        u16 opcode;

        if (op_type == OBJ_OP_DISCARD) {
                if (!offset && length == object_size &&
                    (!img_request_layered_test(img_request) ||
                     !obj_request_overlaps_parent(obj_request))) {
                        opcode = CEPH_OSD_OP_DELETE;
                } else if ((offset + length == object_size)) {
                        opcode = CEPH_OSD_OP_TRUNCATE;
                } else {
                        down_read(&rbd_dev->header_rwsem);
                        img_end = rbd_dev->header.image_size;
                        up_read(&rbd_dev->header_rwsem);

                        if (obj_request->img_offset + length == img_end)
                                opcode = CEPH_OSD_OP_TRUNCATE;
                        else
                                opcode = CEPH_OSD_OP_ZERO;
                }
        } else if (op_type == OBJ_OP_WRITE) {
                if (!offset && length == object_size)
                        opcode = CEPH_OSD_OP_WRITEFULL;
                else
                        opcode = CEPH_OSD_OP_WRITE;
                osd_req_op_alloc_hint_init(osd_request, num_ops,
                                        object_size, object_size);
                num_ops++;
        } else {
                opcode = CEPH_OSD_OP_READ;
        }

        if (opcode == CEPH_OSD_OP_DELETE)
                osd_req_op_init(osd_request, num_ops, opcode, 0);
        else
                osd_req_op_extent_init(osd_request, num_ops, opcode,
                                       offset, length, 0, 0);

        if (obj_request->type == OBJ_REQUEST_BIO)
                osd_req_op_extent_osd_data_bio(osd_request, num_ops,
                                        obj_request->bio_list, length);
        else if (obj_request->type == OBJ_REQUEST_PAGES)
                osd_req_op_extent_osd_data_pages(osd_request, num_ops,
                                        obj_request->pages, length,
                                        offset & ~PAGE_MASK, false, false);

        /* Discards are also writes */
        if (op_type == OBJ_OP_WRITE || op_type == OBJ_OP_DISCARD)
                rbd_osd_req_format_write(obj_request);
        else
                rbd_osd_req_format_read(obj_request);
}

/*
 * Split up an image request into one or more object requests, each
 * to a different object.  The "type" parameter indicates whether
 * "data_desc" is the pointer to the head of a list of bio
 * structures, or the base of a page array.  In either case this
 * function assumes data_desc describes memory sufficient to hold
 * all data described by the image request.
 */
static int rbd_img_request_fill(struct rbd_img_request *img_request,
                                        enum obj_request_type type,
                                        void *data_desc)
{
        struct rbd_device *rbd_dev = img_request->rbd_dev;
        struct rbd_obj_request *obj_request = NULL;
        struct rbd_obj_request *next_obj_request;
        struct bio *bio_list = NULL;
        unsigned int bio_offset = 0;
        struct page **pages = NULL;
        enum obj_operation_type op_type;
        u64 img_offset;
        u64 resid;

        dout("%s: img %p type %d data_desc %p\n", __func__, img_request,
                (int)type, data_desc);

        img_offset = img_request->offset;
        resid = img_request->length;
        rbd_assert(resid > 0);
        op_type = rbd_img_request_op_type(img_request);

        if (type == OBJ_REQUEST_BIO) {
                bio_list = data_desc;
                rbd_assert(img_offset ==
                           bio_list->bi_iter.bi_sector << SECTOR_SHIFT);
        } else if (type == OBJ_REQUEST_PAGES) {
                pages = data_desc;
        }

        while (resid) {
                struct ceph_osd_request *osd_req;
                const char *object_name;
                u64 offset;
                u64 length;

                object_name = rbd_segment_name(rbd_dev, img_offset);
                if (!object_name)
                        goto out_unwind;
                offset = rbd_segment_offset(rbd_dev, img_offset);
                length = rbd_segment_length(rbd_dev, img_offset, resid);
                obj_request = rbd_obj_request_create(object_name,
                                                offset, length, type);
                /* object request has its own copy of the object name */
                rbd_segment_name_free(object_name);
                if (!obj_request)
                        goto out_unwind;

                /*
                 * set obj_request->img_request before creating the
                 * osd_request so that it gets the right snapc
                 */
                rbd_img_obj_request_add(img_request, obj_request);

                if (type == OBJ_REQUEST_BIO) {
                        unsigned int clone_size;

                        rbd_assert(length <= (u64)UINT_MAX);
                        clone_size = (unsigned int)length;
                        obj_request->bio_list =
                                        bio_chain_clone_range(&bio_list,
                                                                &bio_offset,
                                                                clone_size,
                                                                GFP_NOIO);
                        if (!obj_request->bio_list)
                                goto out_unwind;
                } else if (type == OBJ_REQUEST_PAGES) {
                        unsigned int page_count;

                        obj_request->pages = pages;
                        page_count = (u32)calc_pages_for(offset, length);
                        obj_request->page_count = page_count;
                        if ((offset + length) & ~PAGE_MASK)
                                page_count--;   /* more on last page */
                        pages += page_count;
                }

                osd_req = rbd_osd_req_create(rbd_dev, op_type,
                                        (op_type == OBJ_OP_WRITE) ? 2 : 1,
                                        obj_request);
                if (!osd_req)
                        goto out_unwind;

                obj_request->osd_req = osd_req;
                obj_request->callback = rbd_img_obj_callback;
                obj_request->img_offset = img_offset;

                rbd_img_obj_request_fill(obj_request, osd_req, op_type, 0);

                img_offset += length;
                resid -= length;
        }

        return 0;

out_unwind:
        for_each_obj_request_safe(img_request, obj_request, next_obj_request)
                rbd_img_obj_request_del(img_request, obj_request);

        return -ENOMEM;
}

static void
rbd_osd_copyup_callback(struct rbd_obj_request *obj_request)
{
        struct rbd_img_request *img_request;
        struct rbd_device *rbd_dev;
        struct page **pages;
        u32 page_count;

        dout("%s: obj %p\n", __func__, obj_request);

        rbd_assert(obj_request->type == OBJ_REQUEST_BIO ||
                obj_request->type == OBJ_REQUEST_NODATA);
        rbd_assert(obj_request_img_data_test(obj_request));
        img_request = obj_request->img_request;
        rbd_assert(img_request);

        rbd_dev = img_request->rbd_dev;
        rbd_assert(rbd_dev);

        pages = obj_request->copyup_pages;
        rbd_assert(pages != NULL);
        obj_request->copyup_pages = NULL;
        page_count = obj_request->copyup_page_count;
        rbd_assert(page_count);
        obj_request->copyup_page_count = 0;
        ceph_release_page_vector(pages, page_count);

        /*
         * We want the transfer count to reflect the size of the
         * original write request.  There is no such thing as a
         * successful short write, so if the request was successful
         * we can just set it to the originally-requested length.
         */
        if (!obj_request->result)
                obj_request->xferred = obj_request->length;

        obj_request_done_set(obj_request);
}

static void
rbd_img_obj_parent_read_full_callback(struct rbd_img_request *img_request)
{
        struct rbd_obj_request *orig_request;
        struct ceph_osd_request *osd_req;
        struct rbd_device *rbd_dev;
        struct page **pages;
        enum obj_operation_type op_type;
        u32 page_count;
        int img_result;
        u64 parent_length;

        rbd_assert(img_request_child_test(img_request));

        /* First get what we need from the image request */

        pages = img_request->copyup_pages;
        rbd_assert(pages != NULL);
        img_request->copyup_pages = NULL;
        page_count = img_request->copyup_page_count;
        rbd_assert(page_count);
        img_request->copyup_page_count = 0;

        orig_request = img_request->obj_request;
        rbd_assert(orig_request != NULL);
        rbd_assert(obj_request_type_valid(orig_request->type));
        img_result = img_request->result;
        parent_length = img_request->length;
        rbd_assert(img_result || parent_length == img_request->xferred);
        rbd_img_request_put(img_request);

        rbd_assert(orig_request->img_request);
        rbd_dev = orig_request->img_request->rbd_dev;
        rbd_assert(rbd_dev);

        /*
         * If the overlap has become 0 (most likely because the
         * image has been flattened) we need to free the pages
         * and re-submit the original write request.
         */
        if (!rbd_dev->parent_overlap) {
                ceph_release_page_vector(pages, page_count);
                rbd_obj_request_submit(orig_request);
                return;
        }

        if (img_result)
                goto out_err;

        /*
         * The original osd request is of no use to use any more.
         * We need a new one that can hold the three ops in a copyup
         * request.  Allocate the new copyup osd request for the
         * original request, and release the old one.
         */
        img_result = -ENOMEM;
        osd_req = rbd_osd_req_create_copyup(orig_request);
        if (!osd_req)
                goto out_err;
        rbd_osd_req_destroy(orig_request->osd_req);
        orig_request->osd_req = osd_req;
        orig_request->copyup_pages = pages;
        orig_request->copyup_page_count = page_count;

        /* Initialize the copyup op */

        osd_req_op_cls_init(osd_req, 0, CEPH_OSD_OP_CALL, "rbd", "copyup");
        osd_req_op_cls_request_data_pages(osd_req, 0, pages, parent_length, 0,
                                                false, false);

        /* Add the other op(s) */

        op_type = rbd_img_request_op_type(orig_request->img_request);
        rbd_img_obj_request_fill(orig_request, osd_req, op_type, 1);

        /* All set, send it off. */

        rbd_obj_request_submit(orig_request);
        return;

out_err:
        ceph_release_page_vector(pages, page_count);
        rbd_obj_request_error(orig_request, img_result);
}

/*
 * Read from the parent image the range of data that covers the
 * entire target of the given object request.  This is used for
 * satisfying a layered image write request when the target of an
 * object request from the image request does not exist.
 *
 * A page array big enough to hold the returned data is allocated
 * and supplied to rbd_img_request_fill() as the "data descriptor."
 * When the read completes, this page array will be transferred to
 * the original object request for the copyup operation.
 *
 * If an error occurs, it is recorded as the result of the original
 * object request in rbd_img_obj_exists_callback().
 */
static int rbd_img_obj_parent_read_full(struct rbd_obj_request *obj_request)
{
        struct rbd_device *rbd_dev = obj_request->img_request->rbd_dev;
        struct rbd_img_request *parent_request = NULL;
        u64 img_offset;
        u64 length;
        struct page **pages = NULL;
        u32 page_count;
        int result;

        rbd_assert(rbd_dev->parent != NULL);

        /*
         * Determine the byte range covered by the object in the
         * child image to which the original request was to be sent.
         */
        img_offset = obj_request->img_offset - obj_request->offset;
        length = (u64)1 << rbd_dev->header.obj_order;

        /*
         * There is no defined parent data beyond the parent
         * overlap, so limit what we read at that boundary if
         * necessary.
         */
        if (img_offset + length > rbd_dev->parent_overlap) {
                rbd_assert(img_offset < rbd_dev->parent_overlap);
                length = rbd_dev->parent_overlap - img_offset;
        }

        /*
         * Allocate a page array big enough to receive the data read
         * from the parent.
         */
        page_count = (u32)calc_pages_for(0, length);
        pages = ceph_alloc_page_vector(page_count, GFP_KERNEL);
        if (IS_ERR(pages)) {
                result = PTR_ERR(pages);
                pages = NULL;
                goto out_err;
        }

        result = -ENOMEM;
        parent_request = rbd_parent_request_create(obj_request,
                                                img_offset, length);
        if (!parent_request)
                goto out_err;

        result = rbd_img_request_fill(parent_request, OBJ_REQUEST_PAGES, pages);
        if (result)
                goto out_err;

        parent_request->copyup_pages = pages;
        parent_request->copyup_page_count = page_count;
        parent_request->callback = rbd_img_obj_parent_read_full_callback;

        result = rbd_img_request_submit(parent_request);
        if (!result)
                return 0;

        parent_request->copyup_pages = NULL;
        parent_request->copyup_page_count = 0;
        parent_request->obj_request = NULL;
        rbd_obj_request_put(obj_request);
out_err:
        if (pages)
                ceph_release_page_vector(pages, page_count);
        if (parent_request)
                rbd_img_request_put(parent_request);
        return result;
}

static void rbd_img_obj_exists_callback(struct rbd_obj_request *obj_request)
{
        struct rbd_obj_request *orig_request;
        struct rbd_device *rbd_dev;
        int result;

        rbd_assert(!obj_request_img_data_test(obj_request));

        /*
         * All we need from the object request is the original
         * request and the result of the STAT op.  Grab those, then
         * we're done with the request.
         */
        orig_request = obj_request->obj_request;
        obj_request->obj_request = NULL;
        rbd_obj_request_put(orig_request);
        rbd_assert(orig_request);
        rbd_assert(orig_request->img_request);

        result = obj_request->result;
        obj_request->result = 0;

        dout("%s: obj %p for obj %p result %d %llu/%llu\n", __func__,
                obj_request, orig_request, result,
                obj_request->xferred, obj_request->length);
        rbd_obj_request_put(obj_request);

        /*
         * If the overlap has become 0 (most likely because the
         * image has been flattened) we need to re-submit the
         * original request.
         */
        rbd_dev = orig_request->img_request->rbd_dev;
        if (!rbd_dev->parent_overlap) {
                rbd_obj_request_submit(orig_request);
                return;
        }

        /*
         * Our only purpose here is to determine whether the object
         * exists, and we don't want to treat the non-existence as
         * an error.  If something else comes back, transfer the
         * error to the original request and complete it now.
         */
        if (!result) {
                obj_request_existence_set(orig_request, true);
        } else if (result == -ENOENT) {
                obj_request_existence_set(orig_request, false);
        } else {
                goto fail_orig_request;
        }

        /*
         * Resubmit the original request now that we have recorded
         * whether the target object exists.
         */
        result = rbd_img_obj_request_submit(orig_request);
        if (result)
                goto fail_orig_request;

        return;

fail_orig_request:
        rbd_obj_request_error(orig_request, result);
}

static int rbd_img_obj_exists_submit(struct rbd_obj_request *obj_request)
{
        struct rbd_device *rbd_dev = obj_request->img_request->rbd_dev;
        struct rbd_obj_request *stat_request;
        struct page **pages;
        u32 page_count;
        size_t size;
        int ret;

        stat_request = rbd_obj_request_create(obj_request->object_name, 0, 0,
                                              OBJ_REQUEST_PAGES);
        if (!stat_request)
                return -ENOMEM;

        stat_request->osd_req = rbd_osd_req_create(rbd_dev, OBJ_OP_READ, 1,
                                                   stat_request);
        if (!stat_request->osd_req) {
                ret = -ENOMEM;
                goto fail_stat_request;
        }

        /*
         * The response data for a STAT call consists of:
         *     le64 length;
         *     struct {
         *         le32 tv_sec;
         *         le32 tv_nsec;
         *     } mtime;
         */
        size = sizeof (__le64) + sizeof (__le32) + sizeof (__le32);
        page_count = (u32)calc_pages_for(0, size);
        pages = ceph_alloc_page_vector(page_count, GFP_KERNEL);
        if (IS_ERR(pages)) {
                ret = PTR_ERR(pages);
                goto fail_stat_request;
        }

        osd_req_op_init(stat_request->osd_req, 0, CEPH_OSD_OP_STAT, 0);
        osd_req_op_raw_data_in_pages(stat_request->osd_req, 0, pages, size, 0,
                                     false, false);

        rbd_obj_request_get(obj_request);
        stat_request->obj_request = obj_request;
        stat_request->pages = pages;
        stat_request->page_count = page_count;
        stat_request->callback = rbd_img_obj_exists_callback;

        rbd_obj_request_submit(stat_request);
        return 0;

fail_stat_request:
        rbd_obj_request_put(stat_request);
        return ret;
}

static bool img_obj_request_simple(struct rbd_obj_request *obj_request)
{
        struct rbd_img_request *img_request = obj_request->img_request;
        struct rbd_device *rbd_dev = img_request->rbd_dev;

        /* Reads */
        if (!img_request_write_test(img_request) &&
            !img_request_discard_test(img_request))
                return true;

        /* Non-layered writes */
        if (!img_request_layered_test(img_request))
                return true;

        /*
         * Layered writes outside of the parent overlap range don't
         * share any data with the parent.
         */
        if (!obj_request_overlaps_parent(obj_request))
                return true;

        /*
         * Entire-object layered writes - we will overwrite whatever
         * parent data there is anyway.
         */
        if (!obj_request->offset &&
            obj_request->length == rbd_obj_bytes(&rbd_dev->header))
                return true;

        /*
         * If the object is known to already exist, its parent data has
         * already been copied.
         */
        if (obj_request_known_test(obj_request) &&
            obj_request_exists_test(obj_request))
                return true;

        return false;
}

static int rbd_img_obj_request_submit(struct rbd_obj_request *obj_request)
{
        rbd_assert(obj_request_img_data_test(obj_request));
        rbd_assert(obj_request_type_valid(obj_request->type));
        rbd_assert(obj_request->img_request);

        if (img_obj_request_simple(obj_request)) {
                rbd_obj_request_submit(obj_request);
                return 0;
        }

        /*

         * It's a layered write.  The target object might exist but
         * we may not know that yet.  If we know it doesn't exist,
         * start by reading the data for the full target object from
         * the parent so we can use it for a copyup to the target.
         */
        if (obj_request_known_test(obj_request))
                return rbd_img_obj_parent_read_full(obj_request);

        /* We don't know whether the target exists.  Go find out. */

        return rbd_img_obj_exists_submit(obj_request);
}

static int rbd_img_request_submit(struct rbd_img_request *img_request)
{
        struct rbd_obj_request *obj_request;
        struct rbd_obj_request *next_obj_request;
        int ret = 0;

        dout("%s: img %p\n", __func__, img_request);

        rbd_img_request_get(img_request);
        for_each_obj_request_safe(img_request, obj_request, next_obj_request) {
                ret = rbd_img_obj_request_submit(obj_request);
                if (ret)
                        goto out_put_ireq;
        }

out_put_ireq:
        rbd_img_request_put(img_request);
        return ret;
}

static void rbd_img_parent_read_callback(struct rbd_img_request *img_request)
{
        struct rbd_obj_request *obj_request;
        struct rbd_device *rbd_dev;
        u64 obj_end;
        u64 img_xferred;
        int img_result;

        rbd_assert(img_request_child_test(img_request));

        /* First get what we need from the image request and release it */

        obj_request = img_request->obj_request;
        img_xferred = img_request->xferred;
        img_result = img_request->result;
        rbd_img_request_put(img_request);

        /*
         * If the overlap has become 0 (most likely because the
         * image has been flattened) we need to re-submit the
         * original request.
         */
        rbd_assert(obj_request);
        rbd_assert(obj_request->img_request);
        rbd_dev = obj_request->img_request->rbd_dev;
        if (!rbd_dev->parent_overlap) {
                rbd_obj_request_submit(obj_request);
                return;
        }

        obj_request->result = img_result;
        if (obj_request->result)
                goto out;

        /*
         * We need to zero anything beyond the parent overlap
         * boundary.  Since rbd_img_obj_request_read_callback()
         * will zero anything beyond the end of a short read, an
         * easy way to do this is to pretend the data from the
         * parent came up short--ending at the overlap boundary.
         */
        rbd_assert(obj_request->img_offset < U64_MAX - obj_request->length);
        obj_end = obj_request->img_offset + obj_request->length;
        if (obj_end > rbd_dev->parent_overlap) {
                u64 xferred = 0;

                if (obj_request->img_offset < rbd_dev->parent_overlap)
                        xferred = rbd_dev->parent_overlap -
                                        obj_request->img_offset;

                obj_request->xferred = min(img_xferred, xferred);
        } else {
                obj_request->xferred = img_xferred;
        }
out:
        rbd_img_obj_request_read_callback(obj_request);
        rbd_obj_request_complete(obj_request);
}

static void rbd_img_parent_read(struct rbd_obj_request *obj_request)
{
        struct rbd_img_request *img_request;
        int result;

        rbd_assert(obj_request_img_data_test(obj_request));
        rbd_assert(obj_request->img_request != NULL);
        rbd_assert(obj_request->result == (s32) -ENOENT);
        rbd_assert(obj_request_type_valid(obj_request->type));

        /* rbd_read_finish(obj_request, obj_request->length); */
        img_request = rbd_parent_request_create(obj_request,
                                                obj_request->img_offset,
                                                obj_request->length);
        result = -ENOMEM;
        if (!img_request)
                goto out_err;

        if (obj_request->type == OBJ_REQUEST_BIO)
                result = rbd_img_request_fill(img_request, OBJ_REQUEST_BIO,
                                                obj_request->bio_list);
        else
                result = rbd_img_request_fill(img_request, OBJ_REQUEST_PAGES,
                                                obj_request->pages);
        if (result)
                goto out_err;

        img_request->callback = rbd_img_parent_read_callback;
        result = rbd_img_request_submit(img_request);
        if (result)
                goto out_err;

        return;
out_err:
        if (img_request)
                rbd_img_request_put(img_request);
        obj_request->result = result;
        obj_request->xferred = 0;
        obj_request_done_set(obj_request);
}

static const struct rbd_client_id rbd_empty_cid;

static bool rbd_cid_equal(const struct rbd_client_id *lhs,
                          const struct rbd_client_id *rhs)
{
        return lhs->gid == rhs->gid && lhs->handle == rhs->handle;
}

static struct rbd_client_id rbd_get_cid(struct rbd_device *rbd_dev)
{
        struct rbd_client_id cid;

        mutex_lock(&rbd_dev->watch_mutex);
        cid.gid = ceph_client_gid(rbd_dev->rbd_client->client);
        cid.handle = rbd_dev->watch_cookie;
        mutex_unlock(&rbd_dev->watch_mutex);
        return cid;
}

/*
 * lock_rwsem must be held for write
 */
static void rbd_set_owner_cid(struct rbd_device *rbd_dev,
                              const struct rbd_client_id *cid)
{
        dout("%s rbd_dev %p %llu-%llu -> %llu-%llu\n", __func__, rbd_dev,
             rbd_dev->owner_cid.gid, rbd_dev->owner_cid.handle,
             cid->gid, cid->handle);
        rbd_dev->owner_cid = *cid; /* struct */
}

static void format_lock_cookie(struct rbd_device *rbd_dev, char *buf)
{
        mutex_lock(&rbd_dev->watch_mutex);
        sprintf(buf, "%s %llu", RBD_LOCK_COOKIE_PREFIX, rbd_dev->watch_cookie);
        mutex_unlock(&rbd_dev->watch_mutex);
}

/*
 * lock_rwsem must be held for write
 */
static int rbd_lock(struct rbd_device *rbd_dev)
{
        struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
        struct rbd_client_id cid = rbd_get_cid(rbd_dev);
        char cookie[32];
        int ret;

        WARN_ON(__rbd_is_lock_owner(rbd_dev));

        format_lock_cookie(rbd_dev, cookie);
        ret = ceph_cls_lock(osdc, &rbd_dev->header_oid, &rbd_dev->header_oloc,
                            RBD_LOCK_NAME, CEPH_CLS_LOCK_EXCLUSIVE, cookie,
                            RBD_LOCK_TAG, "", 0);
        if (ret)
                return ret;

        rbd_dev->lock_state = RBD_LOCK_STATE_LOCKED;
        rbd_set_owner_cid(rbd_dev, &cid);
        queue_work(rbd_dev->task_wq, &rbd_dev->acquired_lock_work);
        return 0;
}

/*
 * lock_rwsem must be held for write
 */
static int rbd_unlock(struct rbd_device *rbd_dev)
{
        struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
        char cookie[32];
        int ret;

        WARN_ON(!__rbd_is_lock_owner(rbd_dev));

        rbd_dev->lock_state = RBD_LOCK_STATE_UNLOCKED;

        format_lock_cookie(rbd_dev, cookie);
        ret = ceph_cls_unlock(osdc, &rbd_dev->header_oid, &rbd_dev->header_oloc,
                              RBD_LOCK_NAME, cookie);
        if (ret && ret != -ENOENT) {
                rbd_warn(rbd_dev, "cls_unlock failed: %d", ret);
                return ret;
        }

        rbd_set_owner_cid(rbd_dev, &rbd_empty_cid);
        queue_work(rbd_dev->task_wq, &rbd_dev->released_lock_work);
        return 0;
}

static int __rbd_notify_op_lock(struct rbd_device *rbd_dev,
                                enum rbd_notify_op notify_op,
                                struct page ***preply_pages,
                                size_t *preply_len)
{
        struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
        struct rbd_client_id cid = rbd_get_cid(rbd_dev);
        int buf_size = 4 + 8 + 8 + CEPH_ENCODING_START_BLK_LEN;
        char buf[buf_size];
        void *p = buf;

        dout("%s rbd_dev %p notify_op %d\n", __func__, rbd_dev, notify_op);

        /* encode *LockPayload NotifyMessage (op + ClientId) */
        ceph_start_encoding(&p, 2, 1, buf_size - CEPH_ENCODING_START_BLK_LEN);
        ceph_encode_32(&p, notify_op);
        ceph_encode_64(&p, cid.gid);
        ceph_encode_64(&p, cid.handle);

        return ceph_osdc_notify(osdc, &rbd_dev->header_oid,
                                &rbd_dev->header_oloc, buf, buf_size,
                                RBD_NOTIFY_TIMEOUT, preply_pages, preply_len);
}

static void rbd_notify_op_lock(struct rbd_device *rbd_dev,
                               enum rbd_notify_op notify_op)
{
        struct page **reply_pages;
        size_t reply_len;

        __rbd_notify_op_lock(rbd_dev, notify_op, &reply_pages, &reply_len);
        ceph_release_page_vector(reply_pages, calc_pages_for(0, reply_len));
}

static void rbd_notify_acquired_lock(struct work_struct *work)
{
        struct rbd_device *rbd_dev = container_of(work, struct rbd_device,
                                                  acquired_lock_work);

        rbd_notify_op_lock(rbd_dev, RBD_NOTIFY_OP_ACQUIRED_LOCK);
}

static void rbd_notify_released_lock(struct work_struct *work)
{
        struct rbd_device *rbd_dev = container_of(work, struct rbd_device,
                                                  released_lock_work);

        rbd_notify_op_lock(rbd_dev, RBD_NOTIFY_OP_RELEASED_LOCK);
}

static int rbd_request_lock(struct rbd_device *rbd_dev)
{
        struct page **reply_pages;
        size_t reply_len;
        bool lock_owner_responded = false;
        int ret;

        dout("%s rbd_dev %p\n", __func__, rbd_dev);

        ret = __rbd_notify_op_lock(rbd_dev, RBD_NOTIFY_OP_REQUEST_LOCK,
                                   &reply_pages, &reply_len);
        if (ret && ret != -ETIMEDOUT) {
                rbd_warn(rbd_dev, "failed to request lock: %d", ret);
                goto out;
        }

        if (reply_len > 0 && reply_len <= PAGE_SIZE) {
                void *p = page_address(reply_pages[0]);
                void *const end = p + reply_len;
                u32 n;

                ceph_decode_32_safe(&p, end, n, e_inval); /* num_acks */
                while (n--) {
                        u8 struct_v;
                        u32 len;

                        ceph_decode_need(&p, end, 8 + 8, e_inval);
                        p += 8 + 8; /* skip gid and cookie */

                        ceph_decode_32_safe(&p, end, len, e_inval);
                        if (!len)
                                continue;

                        if (lock_owner_responded) {
                                rbd_warn(rbd_dev,
                                         "duplicate lock owners detected");
                                ret = -EIO;
                                goto out;
                        }

                        lock_owner_responded = true;
                        ret = ceph_start_decoding(&p, end, 1, "ResponseMessage",
                                                  &struct_v, &len);
                        if (ret) {
                                rbd_warn(rbd_dev,
                                         "failed to decode ResponseMessage: %d",
                                         ret);
                                goto e_inval;
                        }

                        ret = ceph_decode_32(&p);
                }
        }

        if (!lock_owner_responded) {
                rbd_warn(rbd_dev, "no lock owners detected");
                ret = -ETIMEDOUT;
        }

out:
        ceph_release_page_vector(reply_pages, calc_pages_for(0, reply_len));
        return ret;

e_inval:
        ret = -EINVAL;
        goto out;
}

static void wake_requests(struct rbd_device *rbd_dev, bool wake_all)
{
        dout("%s rbd_dev %p wake_all %d\n", __func__, rbd_dev, wake_all);

        cancel_delayed_work(&rbd_dev->lock_dwork);
        if (wake_all)
                wake_up_all(&rbd_dev->lock_waitq);
        else
                wake_up(&rbd_dev->lock_waitq);
}

static int get_lock_owner_info(struct rbd_device *rbd_dev,
                               struct ceph_locker **lockers, u32 *num_lockers)
{
        struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
        u8 lock_type;
        char *lock_tag;
        int ret;

        dout("%s rbd_dev %p\n", __func__, rbd_dev);

        ret = ceph_cls_lock_info(osdc, &rbd_dev->header_oid,
                                 &rbd_dev->header_oloc, RBD_LOCK_NAME,
                                 &lock_type, &lock_tag, lockers, num_lockers);
        if (ret)
                return ret;

        if (*num_lockers == 0) {
                dout("%s rbd_dev %p no lockers detected\n", __func__, rbd_dev);
                goto out;
        }

        if (strcmp(lock_tag, RBD_LOCK_TAG)) {
                rbd_warn(rbd_dev, "locked by external mechanism, tag %s",
                         lock_tag);
                ret = -EBUSY;
                goto out;
        }

        if (lock_type == CEPH_CLS_LOCK_SHARED) {
                rbd_warn(rbd_dev, "shared lock type detected");
                ret = -EBUSY;
                goto out;
        }

        if (strncmp((*lockers)[0].id.cookie, RBD_LOCK_COOKIE_PREFIX,
                    strlen(RBD_LOCK_COOKIE_PREFIX))) {
                rbd_warn(rbd_dev, "locked by external mechanism, cookie %s",
                         (*lockers)[0].id.cookie);
                ret = -EBUSY;
                goto out;
        }

out:
        kfree(lock_tag);
        return ret;
}

static int find_watcher(struct rbd_device *rbd_dev,
                        const struct ceph_locker *locker)
{
        struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
        struct ceph_watch_item *watchers;
        u32 num_watchers;
        u64 cookie;
        int i;
        int ret;

        ret = ceph_osdc_list_watchers(osdc, &rbd_dev->header_oid,
                                      &rbd_dev->header_oloc, &watchers,
                                      &num_watchers);
        if (ret)
                return ret;

        sscanf(locker->id.cookie, RBD_LOCK_COOKIE_PREFIX " %llu", &cookie);
        for (i = 0; i < num_watchers; i++) {
                if (!memcmp(&watchers[i].addr, &locker->info.addr,
                            sizeof(locker->info.addr)) &&
                    watchers[i].cookie == cookie) {
                        struct rbd_client_id cid = {
                                .gid = le64_to_cpu(watchers[i].name.num),
                                .handle = cookie,
                        };

                        dout("%s rbd_dev %p found cid %llu-%llu\n", __func__,
                             rbd_dev, cid.gid, cid.handle);
                        rbd_set_owner_cid(rbd_dev, &cid);
                        ret = 1;
                        goto out;
                }
        }

        dout("%s rbd_dev %p no watchers\n", __func__, rbd_dev);
        ret = 0;
out:
        kfree(watchers);
        return ret;
}

/*
 * lock_rwsem must be held for write
 */
static int rbd_try_lock(struct rbd_device *rbd_dev)
{
        struct ceph_client *client = rbd_dev->rbd_client->client;
        struct ceph_locker *lockers;
        u32 num_lockers;
        int ret;

        for (;;) {
                ret = rbd_lock(rbd_dev);
                if (ret != -EBUSY)
                        return ret;

                /* determine if the current lock holder is still alive */
                ret = get_lock_owner_info(rbd_dev, &lockers, &num_lockers);
                if (ret)
                        return ret;

                if (num_lockers == 0)
                        goto again;

                ret = find_watcher(rbd_dev, lockers);
                if (ret) {
                        if (ret > 0)
                                ret = 0; /* have to request lock */
                        goto out;
                }

                rbd_warn(rbd_dev, "%s%llu seems dead, breaking lock",
                         ENTITY_NAME(lockers[0].id.name));

                ret = ceph_monc_blacklist_add(&client->monc,
                                              &lockers[0].info.addr);
                if (ret) {
                        rbd_warn(rbd_dev, "blacklist of %s%llu failed: %d",
                                 ENTITY_NAME(lockers[0].id.name), ret);
                        goto out;
                }

                ret = ceph_cls_break_lock(&client->osdc, &rbd_dev->header_oid,
                                          &rbd_dev->header_oloc, RBD_LOCK_NAME,
                                          lockers[0].id.cookie,
                                          &lockers[0].id.name);
                if (ret && ret != -ENOENT)
                        goto out;

again:
                ceph_free_lockers(lockers, num_lockers);
        }

out:
        ceph_free_lockers(lockers, num_lockers);
        return ret;
}

/*
 * ret is set only if lock_state is RBD_LOCK_STATE_UNLOCKED
 */
static enum rbd_lock_state rbd_try_acquire_lock(struct rbd_device *rbd_dev,
                                                int *pret)
{
        enum rbd_lock_state lock_state;

        down_read(&rbd_dev->lock_rwsem);
        dout("%s rbd_dev %p read lock_state %d\n", __func__, rbd_dev,
             rbd_dev->lock_state);
        if (__rbd_is_lock_owner(rbd_dev)) {
                lock_state = rbd_dev->lock_state;
                up_read(&rbd_dev->lock_rwsem);
                return lock_state;
        }

        up_read(&rbd_dev->lock_rwsem);
        down_write(&rbd_dev->lock_rwsem);
        dout("%s rbd_dev %p write lock_state %d\n", __func__, rbd_dev,
             rbd_dev->lock_state);
        if (!__rbd_is_lock_owner(rbd_dev)) {
                *pret = rbd_try_lock(rbd_dev);
                if (*pret)
                        rbd_warn(rbd_dev, "failed to acquire lock: %d", *pret);
        }

        lock_state = rbd_dev->lock_state;
        up_write(&rbd_dev->lock_rwsem);
        return lock_state;
}

static void rbd_acquire_lock(struct work_struct *work)
{
        struct rbd_device *rbd_dev = container_of(to_delayed_work(work),
                                            struct rbd_device, lock_dwork);
        enum rbd_lock_state lock_state;
        int ret;

        dout("%s rbd_dev %p\n", __func__, rbd_dev);
again:
        lock_state = rbd_try_acquire_lock(rbd_dev, &ret);
        if (lock_state != RBD_LOCK_STATE_UNLOCKED || ret == -EBLACKLISTED) {
                if (lock_state == RBD_LOCK_STATE_LOCKED)
                        wake_requests(rbd_dev, true);
                dout("%s rbd_dev %p lock_state %d ret %d - done\n", __func__,
                     rbd_dev, lock_state, ret);
                return;
        }

        ret = rbd_request_lock(rbd_dev);
        if (ret == -ETIMEDOUT) {
                goto again; /* treat this as a dead client */
        } else if (ret < 0) {
                rbd_warn(rbd_dev, "error requesting lock: %d", ret);
                mod_delayed_work(rbd_dev->task_wq, &rbd_dev->lock_dwork,
                                 RBD_RETRY_DELAY);
        } else {
                /*
                 * lock owner acked, but resend if we don't see them
                 * release the lock
                 */
                dout("%s rbd_dev %p requeueing lock_dwork\n", __func__,
                     rbd_dev);
                mod_delayed_work(rbd_dev->task_wq, &rbd_dev->lock_dwork,
                    msecs_to_jiffies(2 * RBD_NOTIFY_TIMEOUT * MSEC_PER_SEC));
        }
}

/*
 * lock_rwsem must be held for write
 */
static bool rbd_release_lock(struct rbd_device *rbd_dev)
{
        dout("%s rbd_dev %p read lock_state %d\n", __func__, rbd_dev,
             rbd_dev->lock_state);
        if (rbd_dev->lock_state != RBD_LOCK_STATE_LOCKED)
                return false;

        rbd_dev->lock_state = RBD_LOCK_STATE_RELEASING;
        downgrade_write(&rbd_dev->lock_rwsem);
        /*
         * Ensure that all in-flight IO is flushed.
         *
         * FIXME: ceph_osdc_sync() flushes the entire OSD client, which
         * may be shared with other devices.
         */
        ceph_osdc_sync(&rbd_dev->rbd_client->client->osdc);
        up_read(&rbd_dev->lock_rwsem);

        down_write(&rbd_dev->lock_rwsem);
        dout("%s rbd_dev %p write lock_state %d\n", __func__, rbd_dev,
             rbd_dev->lock_state);
        if (rbd_dev->lock_state != RBD_LOCK_STATE_RELEASING)
                return false;

        if (!rbd_unlock(rbd_dev))
                /*
                 * Give others a chance to grab the lock - we would re-acquire
                 * almost immediately if we got new IO during ceph_osdc_sync()
                 * otherwise.  We need to ack our own notifications, so this
                 * lock_dwork will be requeued from rbd_wait_state_locked()
                 * after wake_requests() in rbd_handle_released_lock().
                 */
                cancel_delayed_work(&rbd_dev->lock_dwork);

        return true;
}

static void rbd_release_lock_work(struct work_struct *work)
{
        struct rbd_device *rbd_dev = container_of(work, struct rbd_device,
                                                  unlock_work);

        down_write(&rbd_dev->lock_rwsem);
        rbd_release_lock(rbd_dev);
        up_write(&rbd_dev->lock_rwsem);
}

static void rbd_handle_acquired_lock(struct rbd_device *rbd_dev, u8 struct_v,
                                     void **p)
{
        struct rbd_client_id cid = { 0 };

        if (struct_v >= 2) {
                cid.gid = ceph_decode_64(p);
                cid.handle = ceph_decode_64(p);
        }

        dout("%s rbd_dev %p cid %llu-%llu\n", __func__, rbd_dev, cid.gid,
             cid.handle);
        if (!rbd_cid_equal(&cid, &rbd_empty_cid)) {
                down_write(&rbd_dev->lock_rwsem);
                if (rbd_cid_equal(&cid, &rbd_dev->owner_cid)) {
                        /*
                         * we already know that the remote client is
                         * the owner
                         */
                        up_write(&rbd_dev->lock_rwsem);
                        return;
                }

                rbd_set_owner_cid(rbd_dev, &cid);
                downgrade_write(&rbd_dev->lock_rwsem);
        } else {
                down_read(&rbd_dev->lock_rwsem);
        }

        if (!__rbd_is_lock_owner(rbd_dev))
                wake_requests(rbd_dev, false);
        up_read(&rbd_dev->lock_rwsem);
}

static void rbd_handle_released_lock(struct rbd_device *rbd_dev, u8 struct_v,
                                     void **p)
{
        struct rbd_client_id cid = { 0 };

        if (struct_v >= 2) {
                cid.gid = ceph_decode_64(p);
                cid.handle = ceph_decode_64(p);
        }

        dout("%s rbd_dev %p cid %llu-%llu\n", __func__, rbd_dev, cid.gid,
             cid.handle);
        if (!rbd_cid_equal(&cid, &rbd_empty_cid)) {
                down_write(&rbd_dev->lock_rwsem);
                if (!rbd_cid_equal(&cid, &rbd_dev->owner_cid)) {
                        dout("%s rbd_dev %p unexpected owner, cid %llu-%llu != owner_cid %llu-%llu\n",
                             __func__, rbd_dev, cid.gid, cid.handle,
                             rbd_dev->owner_cid.gid, rbd_dev->owner_cid.handle);
                        up_write(&rbd_dev->lock_rwsem);
                        return;
                }

                rbd_set_owner_cid(rbd_dev, &rbd_empty_cid);
                downgrade_write(&rbd_dev->lock_rwsem);
        } else {
                down_read(&rbd_dev->lock_rwsem);
        }

        if (!__rbd_is_lock_owner(rbd_dev))
                wake_requests(rbd_dev, false);
        up_read(&rbd_dev->lock_rwsem);
}

static bool rbd_handle_request_lock(struct rbd_device *rbd_dev, u8 struct_v,
                                    void **p)
{
        struct rbd_client_id my_cid = rbd_get_cid(rbd_dev);
        struct rbd_client_id cid = { 0 };
        bool need_to_send;

        if (struct_v >= 2) {
                cid.gid = ceph_decode_64(p);
                cid.handle = ceph_decode_64(p);
        }

        dout("%s rbd_dev %p cid %llu-%llu\n", __func__, rbd_dev, cid.gid,
             cid.handle);
        if (rbd_cid_equal(&cid, &my_cid))
                return false;

        down_read(&rbd_dev->lock_rwsem);
        need_to_send = __rbd_is_lock_owner(rbd_dev);
        if (rbd_dev->lock_state == RBD_LOCK_STATE_LOCKED) {
                if (!rbd_cid_equal(&rbd_dev->owner_cid, &rbd_empty_cid)) {
                        dout("%s rbd_dev %p queueing unlock_work\n", __func__,
                             rbd_dev);
                        queue_work(rbd_dev->task_wq, &rbd_dev->unlock_work);
                }
        }
        up_read(&rbd_dev->lock_rwsem);
        return need_to_send;
}

static void __rbd_acknowledge_notify(struct rbd_device *rbd_dev,
                                     u64 notify_id, u64 cookie, s32 *result)
{
        struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
        int buf_size = 4 + CEPH_ENCODING_START_BLK_LEN;
        char buf[buf_size];
        int ret;

        if (result) {
                void *p = buf;

                /* encode ResponseMessage */
                ceph_start_encoding(&p, 1, 1,
                                    buf_size - CEPH_ENCODING_START_BLK_LEN);
                ceph_encode_32(&p, *result);
        } else {
                buf_size = 0;
        }

        ret = ceph_osdc_notify_ack(osdc, &rbd_dev->header_oid,
                                   &rbd_dev->header_oloc, notify_id, cookie,
                                   buf, buf_size);
        if (ret)
                rbd_warn(rbd_dev, "acknowledge_notify failed: %d", ret);
}

static void rbd_acknowledge_notify(struct rbd_device *rbd_dev, u64 notify_id,
                                   u64 cookie)
{
        dout("%s rbd_dev %p\n", __func__, rbd_dev);
        __rbd_acknowledge_notify(rbd_dev, notify_id, cookie, NULL);
}

static void rbd_acknowledge_notify_result(struct rbd_device *rbd_dev,
                                          u64 notify_id, u64 cookie, s32 result)
{
        dout("%s rbd_dev %p result %d\n", __func__, rbd_dev, result);
        __rbd_acknowledge_notify(rbd_dev, notify_id, cookie, &result);
}

static void rbd_watch_cb(void *arg, u64 notify_id, u64 cookie,
                         u64 notifier_id, void *data, size_t data_len)
{
        struct rbd_device *rbd_dev = arg;
        void *p = data;
        void *const end = p + data_len;
        u8 struct_v = 0;
        u32 len;
        u32 notify_op;
        int ret;

        dout("%s rbd_dev %p cookie %llu notify_id %llu data_len %zu\n",
             __func__, rbd_dev, cookie, notify_id, data_len);
        if (data_len) {
                ret = ceph_start_decoding(&p, end, 1, "NotifyMessage",
                                          &struct_v, &len);
                if (ret) {
                        rbd_warn(rbd_dev, "failed to decode NotifyMessage: %d",
                                 ret);
                        return;
                }

                notify_op = ceph_decode_32(&p);
        } else {
                /* legacy notification for header updates */
                notify_op = RBD_NOTIFY_OP_HEADER_UPDATE;
                len = 0;
        }

        dout("%s rbd_dev %p notify_op %u\n", __func__, rbd_dev, notify_op);
        switch (notify_op) {
        case RBD_NOTIFY_OP_ACQUIRED_LOCK:
                rbd_handle_acquired_lock(rbd_dev, struct_v, &p);
                rbd_acknowledge_notify(rbd_dev, notify_id, cookie);
                break;
        case RBD_NOTIFY_OP_RELEASED_LOCK:
                rbd_handle_released_lock(rbd_dev, struct_v, &p);
                rbd_acknowledge_notify(rbd_dev, notify_id, cookie);
                break;
        case RBD_NOTIFY_OP_REQUEST_LOCK:
                if (rbd_handle_request_lock(rbd_dev, struct_v, &p))
                        /*
                         * send ResponseMessage(0) back so the client
                         * can detect a missing owner
                         */
                        rbd_acknowledge_notify_result(rbd_dev, notify_id,
                                                      cookie, 0);
                else
                        rbd_acknowledge_notify(rbd_dev, notify_id, cookie);
                break;
        case RBD_NOTIFY_OP_HEADER_UPDATE:
                ret = rbd_dev_refresh(rbd_dev);
                if (ret)
                        rbd_warn(rbd_dev, "refresh failed: %d", ret);

                rbd_acknowledge_notify(rbd_dev, notify_id, cookie);
                break;
        default:
                if (rbd_is_lock_owner(rbd_dev))
                        rbd_acknowledge_notify_result(rbd_dev, notify_id,
                                                      cookie, -EOPNOTSUPP);
                else
                        rbd_acknowledge_notify(rbd_dev, notify_id, cookie);
                break;
        }
}

static void __rbd_unregister_watch(struct rbd_device *rbd_dev);

static void rbd_watch_errcb(void *arg, u64 cookie, int err)
{
        struct rbd_device *rbd_dev = arg;

        rbd_warn(rbd_dev, "encountered watch error: %d", err);

        down_write(&rbd_dev->lock_rwsem);
        rbd_set_owner_cid(rbd_dev, &rbd_empty_cid);
        up_write(&rbd_dev->lock_rwsem);

        mutex_lock(&rbd_dev->watch_mutex);
        if (rbd_dev->watch_state == RBD_WATCH_STATE_REGISTERED) {
                __rbd_unregister_watch(rbd_dev);
                rbd_dev->watch_state = RBD_WATCH_STATE_ERROR;

                queue_delayed_work(rbd_dev->task_wq, &rbd_dev->watch_dwork, 0);
        }
        mutex_unlock(&rbd_dev->watch_mutex);
}

/*
 * watch_mutex must be locked
 */
static int __rbd_register_watch(struct rbd_device *rbd_dev)
{
        struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
        struct ceph_osd_linger_request *handle;

        rbd_assert(!rbd_dev->watch_handle);
        dout("%s rbd_dev %p\n", __func__, rbd_dev);

        handle = ceph_osdc_watch(osdc, &rbd_dev->header_oid,
                                 &rbd_dev->header_oloc, rbd_watch_cb,
                                 rbd_watch_errcb, rbd_dev);
        if (IS_ERR(handle))
                return PTR_ERR(handle);

        rbd_dev->watch_handle = handle;
        return 0;
}

/*
 * watch_mutex must be locked
 */
static void __rbd_unregister_watch(struct rbd_device *rbd_dev)
{
        struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
        int ret;

        rbd_assert(rbd_dev->watch_handle);
        dout("%s rbd_dev %p\n", __func__, rbd_dev);

        ret = ceph_osdc_unwatch(osdc, rbd_dev->watch_handle);
        if (ret)
                rbd_warn(rbd_dev, "failed to unwatch: %d", ret);

        rbd_dev->watch_handle = NULL;
}

static int rbd_register_watch(struct rbd_device *rbd_dev)
{
        int ret;

        mutex_lock(&rbd_dev->watch_mutex);
        rbd_assert(rbd_dev->watch_state == RBD_WATCH_STATE_UNREGISTERED);
        ret = __rbd_register_watch(rbd_dev);
        if (ret)
                goto out;

        rbd_dev->watch_state = RBD_WATCH_STATE_REGISTERED;
        rbd_dev->watch_cookie = rbd_dev->watch_handle->linger_id;

out:
        mutex_unlock(&rbd_dev->watch_mutex);
        return ret;
}

static void cancel_tasks_sync(struct rbd_device *rbd_dev)
{
        dout("%s rbd_dev %p\n", __func__, rbd_dev);

        cancel_delayed_work_sync(&rbd_dev->watch_dwork);
        cancel_work_sync(&rbd_dev->acquired_lock_work);
        cancel_work_sync(&rbd_dev->released_lock_work);
        cancel_delayed_work_sync(&rbd_dev->lock_dwork);
        cancel_work_sync(&rbd_dev->unlock_work);
}

static void rbd_unregister_watch(struct rbd_device *rbd_dev)
{
        WARN_ON(waitqueue_active(&rbd_dev->lock_waitq));
        cancel_tasks_sync(rbd_dev);

        mutex_lock(&rbd_dev->watch_mutex);
        if (rbd_dev->watch_state == RBD_WATCH_STATE_REGISTERED)
                __rbd_unregister_watch(rbd_dev);
        rbd_dev->watch_state = RBD_WATCH_STATE_UNREGISTERED;
        mutex_unlock(&rbd_dev->watch_mutex);

        ceph_osdc_flush_notifies(&rbd_dev->rbd_client->client->osdc);
}

static void rbd_reregister_watch(struct work_struct *work)
{
        struct rbd_device *rbd_dev = container_of(to_delayed_work(work),
                                            struct rbd_device, watch_dwork);
        bool was_lock_owner = false;
        bool need_to_wake = false;
        int ret;

        dout("%s rbd_dev %p\n", __func__, rbd_dev);

        down_write(&rbd_dev->lock_rwsem);
        if (rbd_dev->lock_state == RBD_LOCK_STATE_LOCKED)
                was_lock_owner = rbd_release_lock(rbd_dev);

        mutex_lock(&rbd_dev->watch_mutex);
        if (rbd_dev->watch_state != RBD_WATCH_STATE_ERROR) {
                mutex_unlock(&rbd_dev->watch_mutex);
                goto out;
        }

        ret = __rbd_register_watch(rbd_dev);
        if (ret) {
                rbd_warn(rbd_dev, "failed to reregister watch: %d", ret);
                if (ret == -EBLACKLISTED || ret == -ENOENT) {
                        set_bit(RBD_DEV_FLAG_BLACKLISTED, &rbd_dev->flags);
                        need_to_wake = true;
                } else {
                        queue_delayed_work(rbd_dev->task_wq,
                                           &rbd_dev->watch_dwork,
                                           RBD_RETRY_DELAY);
                }
                mutex_unlock(&rbd_dev->watch_mutex);
                goto out;
        }

        need_to_wake = true;
        rbd_dev->watch_state = RBD_WATCH_STATE_REGISTERED;
        rbd_dev->watch_cookie = rbd_dev->watch_handle->linger_id;
        mutex_unlock(&rbd_dev->watch_mutex);

        ret = rbd_dev_refresh(rbd_dev);
        if (ret)
                rbd_warn(rbd_dev, "reregisteration refresh failed: %d", ret);

        if (was_lock_owner) {
                ret = rbd_try_lock(rbd_dev);
                if (ret)
                        rbd_warn(rbd_dev, "reregisteration lock failed: %d",
                                 ret);
        }

out:
        up_write(&rbd_dev->lock_rwsem);
        if (need_to_wake)
                wake_requests(rbd_dev, true);
}

/*
 * Synchronous osd object method call.  Returns the number of bytes
 * returned in the outbound buffer, or a negative error code.
 */
static int rbd_obj_method_sync(struct rbd_device *rbd_dev,
                             const char *object_name,
                             const char *class_name,
                             const char *method_name,
                             const void *outbound,
                             size_t outbound_size,
                             void *inbound,
                             size_t inbound_size)
{
        struct rbd_obj_request *obj_request;
        struct page **pages;
        u32 page_count;
        int ret;

        /*
         * Method calls are ultimately read operations.  The result