LU-3259 clio: Revise read ahead implementation

[fs/lustre-release.git] / lustre / include / cl_object.h

/*
 * GPL HEADER START
 *
 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
 *
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License version 2 only,
 * 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 version 2 for more details (a copy is included
 * in the LICENSE file that accompanied this code).
 *
 * You should have received a copy of the GNU General Public License
 * version 2 along with this program; If not, see
 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
 *
 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
 * CA 95054 USA or visit www.sun.com if you need additional information or
 * have any questions.
 *
 * GPL HEADER END
 */
/*
 * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
 * Use is subject to license terms.
 *
 * Copyright (c) 2011, 2014, Intel Corporation.
 */
/*
 * This file is part of Lustre, http://www.lustre.org/
 * Lustre is a trademark of Sun Microsystems, Inc.
 */
#ifndef _LUSTRE_CL_OBJECT_H
#define _LUSTRE_CL_OBJECT_H

/** \defgroup clio clio
 *
 * Client objects implement io operations and cache pages.
 *
 * Examples: lov and osc are implementations of cl interface.
 *
 * Big Theory Statement.
 *
 * Layered objects.
 *
 * Client implementation is based on the following data-types:
 *
 *   - cl_object
 *
 *   - cl_page
 *
 *   - cl_lock     represents an extent lock on an object.
 *
 *   - cl_io       represents high-level i/o activity such as whole read/write
 *                 system call, or write-out of pages from under the lock being
 *                 canceled. cl_io has sub-ios that can be stopped and resumed
 *                 independently, thus achieving high degree of transfer
 *                 parallelism. Single cl_io can be advanced forward by
 *                 the multiple threads (although in the most usual case of
 *                 read/write system call it is associated with the single user
 *                 thread, that issued the system call).
 *
 *   - cl_req      represents a collection of pages for a transfer. cl_req is
 *                 constructed by req-forming engine that tries to saturate
 *                 transport with large and continuous transfers.
 *
 * Terminology
 *
 *     - to avoid confusion high-level I/O operation like read or write system
 *     call is referred to as "an io", whereas low-level I/O operation, like
 *     RPC, is referred to as "a transfer"
 *
 *     - "generic code" means generic (not file system specific) code in the
 *     hosting environment. "cl-code" means code (mostly in cl_*.c files) that
 *     is not layer specific.
 *
 * Locking.
 *
 *  - i_mutex
 *      - PG_locked
 *          - cl_object_header::coh_page_guard
 *          - lu_site::ls_guard
 *
 * See the top comment in cl_object.c for the description of overall locking and
 * reference-counting design.
 *
 * See comments below for the description of i/o, page, and dlm-locking
 * design.
 *
 * @{
 */

/*
 * super-class definitions.
 */
#include <libcfs/libcfs.h>
#include <lu_object.h>
#include <linux/atomic.h>
#include <linux/mutex.h>
#include <linux/radix-tree.h>
#include <linux/spinlock.h>
#include <linux/wait.h>
#include <lustre_dlm.h>

struct inode;

struct cl_device;
struct cl_device_operations;

struct cl_object;
struct cl_object_page_operations;
struct cl_object_lock_operations;

struct cl_page;
struct cl_page_slice;
struct cl_lock;
struct cl_lock_slice;

struct cl_lock_operations;
struct cl_page_operations;

struct cl_io;
struct cl_io_slice;

struct cl_req;
struct cl_req_slice;

/**
 * Operations for each data device in the client stack.
 *
 * \see vvp_cl_ops, lov_cl_ops, lovsub_cl_ops, osc_cl_ops
 */
struct cl_device_operations {
        /**
         * Initialize cl_req. This method is called top-to-bottom on all
         * devices in the stack to get them a chance to allocate layer-private
         * data, and to attach them to the cl_req by calling
         * cl_req_slice_add().
         *
         * \see osc_req_init(), lov_req_init(), lovsub_req_init()
         * \see vvp_req_init()
         */
        int (*cdo_req_init)(const struct lu_env *env, struct cl_device *dev,
                            struct cl_req *req);
};

/**
 * Device in the client stack.
 *
 * \see vvp_device, lov_device, lovsub_device, osc_device
 */
struct cl_device {
        /** Super-class. */
        struct lu_device                   cd_lu_dev;
        /** Per-layer operation vector. */
        const struct cl_device_operations *cd_ops;
};

/** \addtogroup cl_object cl_object
 * @{ */
/**
 * "Data attributes" of cl_object. Data attributes can be updated
 * independently for a sub-object, and top-object's attributes are calculated
 * from sub-objects' ones.
 */
struct cl_attr {
        /** Object size, in bytes */
        loff_t cat_size;
        /**
         * Known minimal size, in bytes.
         *
         * This is only valid when at least one DLM lock is held.
         */
        loff_t cat_kms;
        /** Modification time. Measured in seconds since epoch. */
        time_t cat_mtime;
        /** Access time. Measured in seconds since epoch. */
        time_t cat_atime;
        /** Change time. Measured in seconds since epoch. */
        time_t cat_ctime;
        /**
         * Blocks allocated to this cl_object on the server file system.
         *
         * \todo XXX An interface for block size is needed.
         */
        __u64  cat_blocks;
        /**
         * User identifier for quota purposes.
         */
        uid_t  cat_uid;
        /**
         * Group identifier for quota purposes.
         */
        gid_t  cat_gid;

        /* nlink of the directory */
        __u64  cat_nlink;
};

/**
 * Fields in cl_attr that are being set.
 */
enum cl_attr_valid {
        CAT_SIZE   = 1 << 0,
        CAT_KMS    = 1 << 1,
        CAT_MTIME  = 1 << 3,
        CAT_ATIME  = 1 << 4,
        CAT_CTIME  = 1 << 5,
        CAT_BLOCKS = 1 << 6,
        CAT_UID    = 1 << 7,
        CAT_GID    = 1 << 8
};

/**
 * Sub-class of lu_object with methods common for objects on the client
 * stacks.
 *
 * cl_object: represents a regular file system object, both a file and a
 *    stripe. cl_object is based on lu_object: it is identified by a fid,
 *    layered, cached, hashed, and lrued. Important distinction with the server
 *    side, where md_object and dt_object are used, is that cl_object "fans out"
 *    at the lov/sns level: depending on the file layout, single file is
 *    represented as a set of "sub-objects" (stripes). At the implementation
 *    level, struct lov_object contains an array of cl_objects. Each sub-object
 *    is a full-fledged cl_object, having its fid, living in the lru and hash
 *    table.
 *
 *    This leads to the next important difference with the server side: on the
 *    client, it's quite usual to have objects with the different sequence of
 *    layers. For example, typical top-object is composed of the following
 *    layers:
 *
 *        - vvp
 *        - lov
 *
 *    whereas its sub-objects are composed of
 *
 *        - lovsub
 *        - osc
 *
 *    layers. Here "lovsub" is a mostly dummy layer, whose purpose is to keep
 *    track of the object-subobject relationship.
 *
 *    Sub-objects are not cached independently: when top-object is about to
 *    be discarded from the memory, all its sub-objects are torn-down and
 *    destroyed too.
 *
 * \see vvp_object, lov_object, lovsub_object, osc_object
 */
struct cl_object {
        /** super class */
        struct lu_object                   co_lu;
        /** per-object-layer operations */
        const struct cl_object_operations *co_ops;
        /** offset of page slice in cl_page buffer */
        int                                co_slice_off;
};

/**
 * Description of the client object configuration. This is used for the
 * creation of a new client object that is identified by a more state than
 * fid.
 */
struct cl_object_conf {
        /** Super-class. */
        struct lu_object_conf     coc_lu;
        union {
                /**
                 * Object layout. This is consumed by lov.
                 */
                struct lustre_md *coc_md;
                /**
                 * Description of particular stripe location in the
                 * cluster. This is consumed by osc.
                 */
                struct lov_oinfo *coc_oinfo;
        } u;
        /**
         * VFS inode. This is consumed by vvp.
         */
        struct inode             *coc_inode;
        /**
         * Layout lock handle.
         */
        struct ldlm_lock         *coc_lock;
        /**
         * Operation to handle layout, OBJECT_CONF_XYZ.
         */
        int                       coc_opc;
};

enum {
        /** configure layout, set up a new stripe, must be called while
         * holding layout lock. */
        OBJECT_CONF_SET = 0,
        /** invalidate the current stripe configuration due to losing
         * layout lock. */
        OBJECT_CONF_INVALIDATE = 1,
        /** wait for old layout to go away so that new layout can be
         * set up. */
        OBJECT_CONF_WAIT = 2
};

/**
 * Operations implemented for each cl object layer.
 *
 * \see vvp_ops, lov_ops, lovsub_ops, osc_ops
 */
struct cl_object_operations {
        /**
         * Initialize page slice for this layer. Called top-to-bottom through
         * every object layer when a new cl_page is instantiated. Layer
         * keeping private per-page data, or requiring its own page operations
         * vector should allocate these data here, and attach then to the page
         * by calling cl_page_slice_add(). \a vmpage is locked (in the VM
         * sense). Optional.
         *
         * \retval NULL success.
         *
         * \retval ERR_PTR(errno) failure code.
         *
         * \retval valid-pointer pointer to already existing referenced page
         *         to be used instead of newly created.
         */
        int  (*coo_page_init)(const struct lu_env *env, struct cl_object *obj,
                                struct cl_page *page, pgoff_t index);
        /**
         * Initialize lock slice for this layer. Called top-to-bottom through
         * every object layer when a new cl_lock is instantiated. Layer
         * keeping private per-lock data, or requiring its own lock operations
         * vector should allocate these data here, and attach then to the lock
         * by calling cl_lock_slice_add(). Mandatory.
         */
        int  (*coo_lock_init)(const struct lu_env *env,
                              struct cl_object *obj, struct cl_lock *lock,
                              const struct cl_io *io);
        /**
         * Initialize io state for a given layer.
         *
         * called top-to-bottom once per io existence to initialize io
         * state. If layer wants to keep some state for this type of io, it
         * has to embed struct cl_io_slice in lu_env::le_ses, and register
         * slice with cl_io_slice_add(). It is guaranteed that all threads
         * participating in this io share the same session.
         */
        int  (*coo_io_init)(const struct lu_env *env,
                            struct cl_object *obj, struct cl_io *io);
        /**
         * Fill portion of \a attr that this layer controls. This method is
         * called top-to-bottom through all object layers.
         *
         * \pre cl_object_header::coh_attr_guard of the top-object is locked.
         *
         * \return   0: to continue
         * \return +ve: to stop iterating through layers (but 0 is returned
         * from enclosing cl_object_attr_get())
         * \return -ve: to signal error
         */
        int (*coo_attr_get)(const struct lu_env *env, struct cl_object *obj,
                            struct cl_attr *attr);
        /**
         * Update attributes.
         *
         * \a valid is a bitmask composed from enum #cl_attr_valid, and
         * indicating what attributes are to be set.
         *
         * \pre cl_object_header::coh_attr_guard of the top-object is locked.
         *
         * \return the same convention as for
         * cl_object_operations::coo_attr_get() is used.
         */
        int (*coo_attr_update)(const struct lu_env *env, struct cl_object *obj,
                               const struct cl_attr *attr, unsigned valid);
        /**
         * Update object configuration. Called top-to-bottom to modify object
         * configuration.
         *
         * XXX error conditions and handling.
         */
        int (*coo_conf_set)(const struct lu_env *env, struct cl_object *obj,
                            const struct cl_object_conf *conf);
        /**
         * Glimpse ast. Executed when glimpse ast arrives for a lock on this
         * object. Layers are supposed to fill parts of \a lvb that will be
         * shipped to the glimpse originator as a glimpse result.
         *
         * \see vvp_object_glimpse(), lovsub_object_glimpse(),
         * \see osc_object_glimpse()
         */
        int (*coo_glimpse)(const struct lu_env *env,
                           const struct cl_object *obj, struct ost_lvb *lvb);
        /**
         * Object prune method. Called when the layout is going to change on
         * this object, therefore each layer has to clean up their cache,
         * mainly pages and locks.
         */
        int (*coo_prune)(const struct lu_env *env, struct cl_object *obj);
        /**
         * Object getstripe method.
         */
        int (*coo_getstripe)(const struct lu_env *env, struct cl_object *obj,
                             struct lov_user_md __user *lum);
        /**
         * Find whether there is any callback data (ldlm lock) attached upon
         * the object.
         */
        int (*coo_find_cbdata)(const struct lu_env *env, struct cl_object *obj,
                               ldlm_iterator_t iter, void *data);
};

/**
 * Extended header for client object.
 */
struct cl_object_header {
        /** Standard lu_object_header. cl_object::co_lu::lo_header points
         * here. */
        struct lu_object_header coh_lu;

        /**
         * Parent object. It is assumed that an object has a well-defined
         * parent, but not a well-defined child (there may be multiple
         * sub-objects, for the same top-object). cl_object_header::coh_parent
         * field allows certain code to be written generically, without
         * limiting possible cl_object layouts unduly.
         */
        struct cl_object_header *coh_parent;
        /**
         * Protects consistency between cl_attr of parent object and
         * attributes of sub-objects, that the former is calculated ("merged")
         * from.
         *
         * \todo XXX this can be read/write lock if needed.
         */
        spinlock_t               coh_attr_guard;
        /**
         * Size of cl_page + page slices
         */
        unsigned short           coh_page_bufsize;
        /**
         * Number of objects above this one: 0 for a top-object, 1 for its
         * sub-object, etc.
         */
        unsigned char            coh_nesting;
};

/**
 * Helper macro: iterate over all layers of the object \a obj, assigning every
 * layer top-to-bottom to \a slice.
 */
#define cl_object_for_each(slice, obj)                          \
        list_for_each_entry((slice),                            \
                            &(obj)->co_lu.lo_header->loh_layers,\
                            co_lu.lo_linkage)

/**
 * Helper macro: iterate over all layers of the object \a obj, assigning every
 * layer bottom-to-top to \a slice.
 */
#define cl_object_for_each_reverse(slice, obj)                          \
        list_for_each_entry_reverse((slice),                            \
                                    &(obj)->co_lu.lo_header->loh_layers,\
                                    co_lu.lo_linkage)

/** @} cl_object */

#define CL_PAGE_EOF ((pgoff_t)~0ull)

/** \addtogroup cl_page cl_page
 * @{ */

/** \struct cl_page
 * Layered client page.
 *
 * cl_page: represents a portion of a file, cached in the memory. All pages
 *    of the given file are of the same size, and are kept in the radix tree
 *    hanging off the cl_object. cl_page doesn't fan out, but as sub-objects
 *    of the top-level file object are first class cl_objects, they have their
 *    own radix trees of pages and hence page is implemented as a sequence of
 *    struct cl_pages's, linked into double-linked list through
 *    cl_page::cp_parent and cl_page::cp_child pointers, each residing in the
 *    corresponding radix tree at the corresponding logical offset.
 *
 * cl_page is associated with VM page of the hosting environment (struct
 *    page in Linux kernel, for example), struct page. It is assumed, that this
 *    association is implemented by one of cl_page layers (top layer in the
 *    current design) that
 *
 *        - intercepts per-VM-page call-backs made by the environment (e.g.,
 *          memory pressure),
 *
 *        - translates state (page flag bits) and locking between lustre and
 *          environment.
 *
 *    The association between cl_page and struct page is immutable and
 *    established when cl_page is created.
 *
 * cl_page can be "owned" by a particular cl_io (see below), guaranteeing
 *    this io an exclusive access to this page w.r.t. other io attempts and
 *    various events changing page state (such as transfer completion, or
 *    eviction of the page from the memory). Note, that in general cl_io
 *    cannot be identified with a particular thread, and page ownership is not
 *    exactly equal to the current thread holding a lock on the page. Layer
 *    implementing association between cl_page and struct page has to implement
 *    ownership on top of available synchronization mechanisms.
 *
 *    While lustre client maintains the notion of an page ownership by io,
 *    hosting MM/VM usually has its own page concurrency control
 *    mechanisms. For example, in Linux, page access is synchronized by the
 *    per-page PG_locked bit-lock, and generic kernel code (generic_file_*())
 *    takes care to acquire and release such locks as necessary around the
 *    calls to the file system methods (->readpage(), ->prepare_write(),
 *    ->commit_write(), etc.). This leads to the situation when there are two
 *    different ways to own a page in the client:
 *
 *        - client code explicitly and voluntary owns the page (cl_page_own());
 *
 *        - VM locks a page and then calls the client, that has "to assume"
 *          the ownership from the VM (cl_page_assume()).
 *
 *    Dual methods to release ownership are cl_page_disown() and
 *    cl_page_unassume().
 *
 * cl_page is reference counted (cl_page::cp_ref). When reference counter
 *    drops to 0, the page is returned to the cache, unless it is in
 *    cl_page_state::CPS_FREEING state, in which case it is immediately
 *    destroyed.
 *
 *    The general logic guaranteeing the absence of "existential races" for
 *    pages is the following:
 *
 *        - there are fixed known ways for a thread to obtain a new reference
 *          to a page:
 *
 *            - by doing a lookup in the cl_object radix tree, protected by the
 *              spin-lock;
 *
 *            - by starting from VM-locked struct page and following some
 *              hosting environment method (e.g., following ->private pointer in
 *              the case of Linux kernel), see cl_vmpage_page();
 *
 *        - when the page enters cl_page_state::CPS_FREEING state, all these
 *          ways are severed with the proper synchronization
 *          (cl_page_delete());
 *
 *        - entry into cl_page_state::CPS_FREEING is serialized by the VM page
 *          lock;
 *
 *        - no new references to the page in cl_page_state::CPS_FREEING state
 *          are allowed (checked in cl_page_get()).
 *
 *    Together this guarantees that when last reference to a
 *    cl_page_state::CPS_FREEING page is released, it is safe to destroy the
 *    page, as neither references to it can be acquired at that point, nor
 *    ones exist.
 *
 * cl_page is a state machine. States are enumerated in enum
 *    cl_page_state. Possible state transitions are enumerated in
 *    cl_page_state_set(). State transition process (i.e., actual changing of
 *    cl_page::cp_state field) is protected by the lock on the underlying VM
 *    page.
 *
 * Linux Kernel implementation.
 *
 *    Binding between cl_page and struct page (which is a typedef for
 *    struct page) is implemented in the vvp layer. cl_page is attached to the
 *    ->private pointer of the struct page, together with the setting of
 *    PG_private bit in page->flags, and acquiring additional reference on the
 *    struct page (much like struct buffer_head, or any similar file system
 *    private data structures).
 *
 *    PG_locked lock is used to implement both ownership and transfer
 *    synchronization, that is, page is VM-locked in CPS_{OWNED,PAGE{IN,OUT}}
 *    states. No additional references are acquired for the duration of the
 *    transfer.
 *
 * \warning *THIS IS NOT* the behavior expected by the Linux kernel, where
 *          write-out is "protected" by the special PG_writeback bit.
 */

/**
 * States of cl_page. cl_page.c assumes particular order here.
 *
 * The page state machine is rather crude, as it doesn't recognize finer page
 * states like "dirty" or "up to date". This is because such states are not
 * always well defined for the whole stack (see, for example, the
 * implementation of the read-ahead, that hides page up-to-dateness to track
 * cache hits accurately). Such sub-states are maintained by the layers that
 * are interested in them.
 */
enum cl_page_state {
        /**
         * Page is in the cache, un-owned. Page leaves cached state in the
         * following cases:
         *
         *     - [cl_page_state::CPS_OWNED] io comes across the page and
         *     owns it;
         *
         *     - [cl_page_state::CPS_PAGEOUT] page is dirty, the
         *     req-formation engine decides that it wants to include this page
         *     into an cl_req being constructed, and yanks it from the cache;
         *
         *     - [cl_page_state::CPS_FREEING] VM callback is executed to
         *     evict the page form the memory;
         *
         * \invariant cl_page::cp_owner == NULL && cl_page::cp_req == NULL
         */
        CPS_CACHED,
        /**
         * Page is exclusively owned by some cl_io. Page may end up in this
         * state as a result of
         *
         *     - io creating new page and immediately owning it;
         *
         *     - [cl_page_state::CPS_CACHED] io finding existing cached page
         *     and owning it;
         *
         *     - [cl_page_state::CPS_OWNED] io finding existing owned page
         *     and waiting for owner to release the page;
         *
         * Page leaves owned state in the following cases:
         *
         *     - [cl_page_state::CPS_CACHED] io decides to leave the page in
         *     the cache, doing nothing;
         *
         *     - [cl_page_state::CPS_PAGEIN] io starts read transfer for
         *     this page;
         *
         *     - [cl_page_state::CPS_PAGEOUT] io starts immediate write
         *     transfer for this page;
         *
         *     - [cl_page_state::CPS_FREEING] io decides to destroy this
         *     page (e.g., as part of truncate or extent lock cancellation).
         *
         * \invariant cl_page::cp_owner != NULL && cl_page::cp_req == NULL
         */
        CPS_OWNED,
        /**
         * Page is being written out, as a part of a transfer. This state is
         * entered when req-formation logic decided that it wants this page to
         * be sent through the wire _now_. Specifically, it means that once
         * this state is achieved, transfer completion handler (with either
         * success or failure indication) is guaranteed to be executed against
         * this page independently of any locks and any scheduling decisions
         * made by the hosting environment (that effectively means that the
         * page is never put into cl_page_state::CPS_PAGEOUT state "in
         * advance". This property is mentioned, because it is important when
         * reasoning about possible dead-locks in the system). The page can
         * enter this state as a result of
         *
         *     - [cl_page_state::CPS_OWNED] an io requesting an immediate
         *     write-out of this page, or
         *
         *     - [cl_page_state::CPS_CACHED] req-forming engine deciding
         *     that it has enough dirty pages cached to issue a "good"
         *     transfer.
         *
         * The page leaves cl_page_state::CPS_PAGEOUT state when the transfer
         * is completed---it is moved into cl_page_state::CPS_CACHED state.
         *
         * Underlying VM page is locked for the duration of transfer.
         *
         * \invariant: cl_page::cp_owner == NULL && cl_page::cp_req != NULL
         */
        CPS_PAGEOUT,
        /**
         * Page is being read in, as a part of a transfer. This is quite
         * similar to the cl_page_state::CPS_PAGEOUT state, except that
         * read-in is always "immediate"---there is no such thing a sudden
         * construction of read cl_req from cached, presumably not up to date,
         * pages.
         *
         * Underlying VM page is locked for the duration of transfer.
         *
         * \invariant: cl_page::cp_owner == NULL && cl_page::cp_req != NULL
         */
        CPS_PAGEIN,
        /**
         * Page is being destroyed. This state is entered when client decides
         * that page has to be deleted from its host object, as, e.g., a part
         * of truncate.
         *
         * Once this state is reached, there is no way to escape it.
         *
         * \invariant: cl_page::cp_owner == NULL && cl_page::cp_req == NULL
         */
        CPS_FREEING,
        CPS_NR
};

enum cl_page_type {
        /** Host page, the page is from the host inode which the cl_page
         * belongs to. */
        CPT_CACHEABLE = 1,

        /** Transient page, the transient cl_page is used to bind a cl_page
         *  to vmpage which is not belonging to the same object of cl_page.
         *  it is used in DirectIO, lockless IO and liblustre. */
        CPT_TRANSIENT,
};

/**
 * Fields are protected by the lock on struct page, except for atomics and
 * immutables.
 *
 * \invariant Data type invariants are in cl_page_invariant(). Basically:
 * cl_page::cp_parent and cl_page::cp_child are a well-formed double-linked
 * list, consistent with the parent/child pointers in the cl_page::cp_obj and
 * cl_page::cp_owner (when set).
 */
struct cl_page {
        /** Reference counter. */
        atomic_t                 cp_ref;
        /** Transfer error. */
        int                      cp_error;
        /** An object this page is a part of. Immutable after creation. */
        struct cl_object        *cp_obj;
        /** vmpage */
        struct page             *cp_vmpage;
        /** Linkage of pages within group. Pages must be owned */
        struct list_head         cp_batch;
        /** List of slices. Immutable after creation. */
        struct list_head         cp_layers;
        /** Linkage of pages within cl_req. */
        struct list_head         cp_flight;
        /**
         * Page state. This field is const to avoid accidental update, it is
         * modified only internally within cl_page.c. Protected by a VM lock.
         */
        const enum cl_page_state cp_state;
        /**
         * Page type. Only CPT_TRANSIENT is used so far. Immutable after
         * creation.
         */
        enum cl_page_type        cp_type;

        /**
         * Owning IO in cl_page_state::CPS_OWNED state. Sub-page can be owned
         * by sub-io. Protected by a VM lock.
         */
        struct cl_io            *cp_owner;
        /**
         * Owning IO request in cl_page_state::CPS_PAGEOUT and
         * cl_page_state::CPS_PAGEIN states. This field is maintained only in
         * the top-level pages. Protected by a VM lock.
         */
        struct cl_req           *cp_req;
        /** List of references to this page, for debugging. */
        struct lu_ref            cp_reference;
        /** Link to an object, for debugging. */
        struct lu_ref_link       cp_obj_ref;
        /** Link to a queue, for debugging. */
        struct lu_ref_link       cp_queue_ref;
        /** Assigned if doing a sync_io */
        struct cl_sync_io       *cp_sync_io;
};

/**
 * Per-layer part of cl_page.
 *
 * \see vvp_page, lov_page, osc_page
 */
struct cl_page_slice {
        struct cl_page                  *cpl_page;
        pgoff_t                          cpl_index;
        /**
         * Object slice corresponding to this page slice. Immutable after
         * creation.
         */
        struct cl_object                *cpl_obj;
        const struct cl_page_operations *cpl_ops;
        /** Linkage into cl_page::cp_layers. Immutable after creation. */
        struct list_head                 cpl_linkage;
};

/**
 * Lock mode. For the client extent locks.
 *
 * \ingroup cl_lock
 */
enum cl_lock_mode {
        CLM_READ,
        CLM_WRITE,
        CLM_GROUP,
        CLM_MAX,
};

/**
 * Requested transfer type.
 * \ingroup cl_req
 */
enum cl_req_type {
        CRT_READ,
        CRT_WRITE,
        CRT_NR
};

/**
 * Per-layer page operations.
 *
 * Methods taking an \a io argument are for the activity happening in the
 * context of given \a io. Page is assumed to be owned by that io, except for
 * the obvious cases (like cl_page_operations::cpo_own()).
 *
 * \see vvp_page_ops, lov_page_ops, osc_page_ops
 */
struct cl_page_operations {
        /**
         * cl_page<->struct page methods. Only one layer in the stack has to
         * implement these. Current code assumes that this functionality is
         * provided by the topmost layer, see cl_page_disown0() as an example.
         */

        /**
         * Called when \a io acquires this page into the exclusive
         * ownership. When this method returns, it is guaranteed that the is
         * not owned by other io, and no transfer is going on against
         * it. Optional.
         *
         * \see cl_page_own()
         * \see vvp_page_own(), lov_page_own()
         */
        int  (*cpo_own)(const struct lu_env *env,
                        const struct cl_page_slice *slice,
                        struct cl_io *io, int nonblock);
        /** Called when ownership it yielded. Optional.
         *
         * \see cl_page_disown()
         * \see vvp_page_disown()
         */
        void (*cpo_disown)(const struct lu_env *env,
                           const struct cl_page_slice *slice, struct cl_io *io);
        /**
         * Called for a page that is already "owned" by \a io from VM point of
         * view. Optional.
         *
         * \see cl_page_assume()
         * \see vvp_page_assume(), lov_page_assume()
         */
        void (*cpo_assume)(const struct lu_env *env,
                           const struct cl_page_slice *slice, struct cl_io *io);
        /** Dual to cl_page_operations::cpo_assume(). Optional. Called
         * bottom-to-top when IO releases a page without actually unlocking
         * it.
         *
         * \see cl_page_unassume()
         * \see vvp_page_unassume()
         */
        void (*cpo_unassume)(const struct lu_env *env,
                             const struct cl_page_slice *slice,
                             struct cl_io *io);
        /**
         * Announces whether the page contains valid data or not by \a uptodate.
         *
         * \see cl_page_export()
         * \see vvp_page_export()
         */
        void  (*cpo_export)(const struct lu_env *env,
                            const struct cl_page_slice *slice, int uptodate);
        /**
         * Checks whether underlying VM page is locked (in the suitable
         * sense). Used for assertions.
         *
         * \retval    -EBUSY: page is protected by a lock of a given mode;
         * \retval  -ENODATA: page is not protected by a lock;
         * \retval         0: this layer cannot decide. (Should never happen.)
         */
        int (*cpo_is_vmlocked)(const struct lu_env *env,
                               const struct cl_page_slice *slice);
        /**
         * Page destruction.
         */

        /**
         * Called when page is truncated from the object. Optional.
         *
         * \see cl_page_discard()
         * \see vvp_page_discard(), osc_page_discard()
         */
        void (*cpo_discard)(const struct lu_env *env,
                            const struct cl_page_slice *slice,
                            struct cl_io *io);
        /**
         * Called when page is removed from the cache, and is about to being
         * destroyed. Optional.
         *
         * \see cl_page_delete()
         * \see vvp_page_delete(), osc_page_delete()
         */
        void (*cpo_delete)(const struct lu_env *env,
                           const struct cl_page_slice *slice);
        /** Destructor. Frees resources and slice itself. */
        void (*cpo_fini)(const struct lu_env *env,
                         struct cl_page_slice *slice);
        /**
         * Optional debugging helper. Prints given page slice.
         *
         * \see cl_page_print()
         */
        int (*cpo_print)(const struct lu_env *env,
                         const struct cl_page_slice *slice,
                         void *cookie, lu_printer_t p);
        /**
         * \name transfer
         *
         * Transfer methods. See comment on cl_req for a description of
         * transfer formation and life-cycle.
         *
         * @{
         */
        /**
         * Request type dependent vector of operations.
         *
         * Transfer operations depend on transfer mode (cl_req_type). To avoid
         * passing transfer mode to each and every of these methods, and to
         * avoid branching on request type inside of the methods, separate
         * methods for cl_req_type:CRT_READ and cl_req_type:CRT_WRITE are
         * provided. That is, method invocation usually looks like
         *
         *         slice->cp_ops.io[req->crq_type].cpo_method(env, slice, ...);
         */
        struct {
                /**
                 * Called when a page is submitted for a transfer as a part of
                 * cl_page_list.
                 *
                 * \return    0         : page is eligible for submission;
                 * \return    -EALREADY : skip this page;
                 * \return    -ve       : error.
                 *
                 * \see cl_page_prep()
                 */
                int  (*cpo_prep)(const struct lu_env *env,
                                 const struct cl_page_slice *slice,
                                 struct cl_io *io);
                /**
                 * Completion handler. This is guaranteed to be eventually
                 * fired after cl_page_operations::cpo_prep() or
                 * cl_page_operations::cpo_make_ready() call.
                 *
                 * This method can be called in a non-blocking context. It is
                 * guaranteed however, that the page involved and its object
                 * are pinned in memory (and, hence, calling cl_page_put() is
                 * safe).
                 *
                 * \see cl_page_completion()
                 */
                void (*cpo_completion)(const struct lu_env *env,
                                       const struct cl_page_slice *slice,
                                       int ioret);
                /**
                 * Called when cached page is about to be added to the
                 * cl_req as a part of req formation.
                 *
                 * \return    0       : proceed with this page;
                 * \return    -EAGAIN : skip this page;
                 * \return    -ve     : error.
                 *
                 * \see cl_page_make_ready()
                 */
                int  (*cpo_make_ready)(const struct lu_env *env,
                                       const struct cl_page_slice *slice);
        } io[CRT_NR];
        /**
         * Tell transfer engine that only [to, from] part of a page should be
         * transmitted.
         *
         * This is used for immediate transfers.
         *
         * \todo XXX this is not very good interface. It would be much better
         * if all transfer parameters were supplied as arguments to
         * cl_io_operations::cio_submit() call, but it is not clear how to do
         * this for page queues.
         *
         * \see cl_page_clip()
         */
        void (*cpo_clip)(const struct lu_env *env,
                         const struct cl_page_slice *slice,
                         int from, int to);
        /**
         * \pre  the page was queued for transferring.
         * \post page is removed from client's pending list, or -EBUSY
         *       is returned if it has already been in transferring.
         *
         * This is one of seldom page operation which is:
         * 0. called from top level;
         * 1. don't have vmpage locked;
         * 2. every layer should synchronize execution of its ->cpo_cancel()
         *    with completion handlers. Osc uses client obd lock for this
         *    purpose. Based on there is no vvp_page_cancel and
         *    lov_page_cancel(), cpo_cancel is defacto protected by client lock.
         *
         * \see osc_page_cancel().
         */
        int (*cpo_cancel)(const struct lu_env *env,
                          const struct cl_page_slice *slice);
        /**
         * Write out a page by kernel. This is only called by ll_writepage
         * right now.
         *
         * \see cl_page_flush()
         */
        int (*cpo_flush)(const struct lu_env *env,
                         const struct cl_page_slice *slice,
                         struct cl_io *io);
        /** @} transfer */
};

/**
 * Helper macro, dumping detailed information about \a page into a log.
 */
#define CL_PAGE_DEBUG(mask, env, page, format, ...)                     \
do {                                                                    \
        if (cfs_cdebug_show(mask, DEBUG_SUBSYSTEM)) {                   \
                LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, mask, NULL);        \
                cl_page_print(env, &msgdata, lu_cdebug_printer, page);  \
                CDEBUG(mask, format , ## __VA_ARGS__);                  \
        }                                                               \
} while (0)

/**
 * Helper macro, dumping shorter information about \a page into a log.
 */
#define CL_PAGE_HEADER(mask, env, page, format, ...)                          \
do {                                                                          \
        if (cfs_cdebug_show(mask, DEBUG_SUBSYSTEM)) {                         \
                LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, mask, NULL);              \
                cl_page_header_print(env, &msgdata, lu_cdebug_printer, page); \
                CDEBUG(mask, format , ## __VA_ARGS__);                        \
        }                                                                     \
} while (0)

static inline struct page *cl_page_vmpage(const struct cl_page *page)
{
        LASSERT(page->cp_vmpage != NULL);
        return page->cp_vmpage;
}

/**
 * Check if a cl_page is in use.
 *
 * Client cache holds a refcount, this refcount will be dropped when
 * the page is taken out of cache, see vvp_page_delete().
 */
static inline bool __page_in_use(const struct cl_page *page, int refc)
{
        return (atomic_read(&page->cp_ref) > refc + 1);
}

/**
 * Caller itself holds a refcount of cl_page.
 */
#define cl_page_in_use(pg)       __page_in_use(pg, 1)
/**
 * Caller doesn't hold a refcount.
 */
#define cl_page_in_use_noref(pg) __page_in_use(pg, 0)

/** @} cl_page */

/** \addtogroup cl_lock cl_lock
 * @{ */
/** \struct cl_lock
 *
 * Extent locking on the client.
 *
 * LAYERING
 *
 * The locking model of the new client code is built around
 *
 *        struct cl_lock
 *
 * data-type representing an extent lock on a regular file. cl_lock is a
 * layered object (much like cl_object and cl_page), it consists of a header
 * (struct cl_lock) and a list of layers (struct cl_lock_slice), linked to
 * cl_lock::cll_layers list through cl_lock_slice::cls_linkage.
 *
 * Typical cl_lock consists of the two layers:
 *
 *     - vvp_lock (vvp specific data), and
 *     - lov_lock (lov specific data).
 *
 * lov_lock contains an array of sub-locks. Each of these sub-locks is a
 * normal cl_lock: it has a header (struct cl_lock) and a list of layers:
 *
 *     - lovsub_lock, and
 *     - osc_lock
 *
 * Each sub-lock is associated with a cl_object (representing stripe
 * sub-object or the file to which top-level cl_lock is associated to), and is
 * linked into that cl_object::coh_locks. In this respect cl_lock is similar to
 * cl_object (that at lov layer also fans out into multiple sub-objects), and
 * is different from cl_page, that doesn't fan out (there is usually exactly
 * one osc_page for every vvp_page). We shall call vvp-lov portion of the lock
 * a "top-lock" and its lovsub-osc portion a "sub-lock".
 *
 * LIFE CYCLE
 *
 * cl_lock is a cacheless data container for the requirements of locks to
 * complete the IO. cl_lock is created before I/O starts and destroyed when the
 * I/O is complete.
 *
 * cl_lock depends on LDLM lock to fulfill lock semantics. LDLM lock is attached
 * to cl_lock at OSC layer. LDLM lock is still cacheable.
 *
 * INTERFACE AND USAGE
 *
 * Two major methods are supported for cl_lock: clo_enqueue and clo_cancel.  A
 * cl_lock is enqueued by cl_lock_request(), which will call clo_enqueue()
 * methods for each layer to enqueue the lock. At the LOV layer, if a cl_lock
 * consists of multiple sub cl_locks, each sub locks will be enqueued
 * correspondingly. At OSC layer, the lock enqueue request will tend to reuse
 * cached LDLM lock; otherwise a new LDLM lock will have to be requested from
 * OST side.
 *
 * cl_lock_cancel() must be called to release a cl_lock after use. clo_cancel()
 * method will be called for each layer to release the resource held by this
 * lock. At OSC layer, the reference count of LDLM lock, which is held at
 * clo_enqueue time, is released.
 *
 * LDLM lock can only be canceled if there is no cl_lock using it.
 *
 * Overall process of the locking during IO operation is as following:
 *
 *     - once parameters for IO are setup in cl_io, cl_io_operations::cio_lock()
 *       is called on each layer. Responsibility of this method is to add locks,
 *       needed by a given layer into cl_io.ci_lockset.
 *
 *     - once locks for all layers were collected, they are sorted to avoid
 *       dead-locks (cl_io_locks_sort()), and enqueued.
 *
 *     - when all locks are acquired, IO is performed;
 *
 *     - locks are released after IO is complete.
 *
 * Striping introduces major additional complexity into locking. The
 * fundamental problem is that it is generally unsafe to actively use (hold)
 * two locks on the different OST servers at the same time, as this introduces
 * inter-server dependency and can lead to cascading evictions.
 *
 * Basic solution is to sub-divide large read/write IOs into smaller pieces so
 * that no multi-stripe locks are taken (note that this design abandons POSIX
 * read/write semantics). Such pieces ideally can be executed concurrently. At
 * the same time, certain types of IO cannot be sub-divived, without
 * sacrificing correctness. This includes:
 *
 *  - O_APPEND write, where [0, EOF] lock has to be taken, to guarantee
 *  atomicity;
 *
 *  - ftruncate(fd, offset), where [offset, EOF] lock has to be taken.
 *
 * Also, in the case of read(fd, buf, count) or write(fd, buf, count), where
 * buf is a part of memory mapped Lustre file, a lock or locks protecting buf
 * has to be held together with the usual lock on [offset, offset + count].
 *
 * Interaction with DLM
 *
 * In the expected setup, cl_lock is ultimately backed up by a collection of
 * DLM locks (struct ldlm_lock). Association between cl_lock and DLM lock is
 * implemented in osc layer, that also matches DLM events (ASTs, cancellation,
 * etc.) into cl_lock_operation calls. See struct osc_lock for a more detailed
 * description of interaction with DLM.
 */

/**
 * Lock description.
 */
struct cl_lock_descr {
        /** Object this lock is granted for. */
        struct cl_object *cld_obj;
        /** Index of the first page protected by this lock. */
        pgoff_t           cld_start;
        /** Index of the last page (inclusive) protected by this lock. */
        pgoff_t           cld_end;
        /** Group ID, for group lock */
        __u64             cld_gid;
        /** Lock mode. */
        enum cl_lock_mode cld_mode;
        /**
         * flags to enqueue lock. A combination of bit-flags from
         * enum cl_enq_flags.
         */
        __u32             cld_enq_flags;
};

#define DDESCR "%s(%d):[%lu, %lu]:%x"
#define PDESCR(descr)                                                   \
        cl_lock_mode_name((descr)->cld_mode), (descr)->cld_mode,        \
        (descr)->cld_start, (descr)->cld_end, (descr)->cld_enq_flags

const char *cl_lock_mode_name(const enum cl_lock_mode mode);

/**
 * Layered client lock.
 */
struct cl_lock {
        /** List of slices. Immutable after creation. */
        struct list_head      cll_layers;
        /** lock attribute, extent, cl_object, etc. */
        struct cl_lock_descr  cll_descr;
};

/**
 * Per-layer part of cl_lock
 *
 * \see vvp_lock, lov_lock, lovsub_lock, osc_lock
 */
struct cl_lock_slice {
        struct cl_lock                  *cls_lock;
        /** Object slice corresponding to this lock slice. Immutable after
         * creation. */
        struct cl_object                *cls_obj;
        const struct cl_lock_operations *cls_ops;
        /** Linkage into cl_lock::cll_layers. Immutable after creation. */
        struct list_head                 cls_linkage;
};

/**
 *
 * \see vvp_lock_ops, lov_lock_ops, lovsub_lock_ops, osc_lock_ops
 */
struct cl_lock_operations {
        /** @{ */
        /**
         * Attempts to enqueue the lock. Called top-to-bottom.
         *
         * \retval 0    this layer has enqueued the lock successfully
         * \retval >0   this layer has enqueued the lock, but need to wait on
         *              @anchor for resources
         * \retval -ve  failure
         *
         * \see vvp_lock_enqueue(), lov_lock_enqueue(), lovsub_lock_enqueue(),
         * \see osc_lock_enqueue()
         */
        int  (*clo_enqueue)(const struct lu_env *env,
                            const struct cl_lock_slice *slice,
                            struct cl_io *io, struct cl_sync_io *anchor);
        /**
         * Cancel a lock, release its DLM lock ref, while does not cancel the
         * DLM lock
         */
        void (*clo_cancel)(const struct lu_env *env,
                           const struct cl_lock_slice *slice);
        /** @} */
        /**
         * Destructor. Frees resources and the slice.
         *
         * \see vvp_lock_fini(), lov_lock_fini(), lovsub_lock_fini(),
         * \see osc_lock_fini()
         */
        void (*clo_fini)(const struct lu_env *env, struct cl_lock_slice *slice);
        /**
         * Optional debugging helper. Prints given lock slice.
         */
        int (*clo_print)(const struct lu_env *env,
                         void *cookie, lu_printer_t p,
                         const struct cl_lock_slice *slice);
};

#define CL_LOCK_DEBUG(mask, env, lock, format, ...)                     \
do {                                                                    \
        if (cfs_cdebug_show(mask, DEBUG_SUBSYSTEM)) {                   \
                LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, mask, NULL);        \
                cl_lock_print(env, &msgdata, lu_cdebug_printer, lock);  \
                CDEBUG(mask, format , ## __VA_ARGS__);                  \
        }                                                               \
} while (0)

#define CL_LOCK_ASSERT(expr, env, lock) do {                            \
        if (likely(expr))                                               \
                break;                                                  \
                                                                        \
        CL_LOCK_DEBUG(D_ERROR, env, lock, "failed at %s.\n", #expr);    \
        LBUG();                                                         \
} while (0)

/** @} cl_lock */

/** \addtogroup cl_page_list cl_page_list
 * Page list used to perform collective operations on a group of pages.
 *
 * Pages are added to the list one by one. cl_page_list acquires a reference
 * for every page in it. Page list is used to perform collective operations on
 * pages:
 *
 *     - submit pages for an immediate transfer,
 *
 *     - own pages on behalf of certain io (waiting for each page in turn),
 *
 *     - discard pages.
 *
 * When list is finalized, it releases references on all pages it still has.
 *
 * \todo XXX concurrency control.
 *
 * @{
 */
struct cl_page_list {
        unsigned                 pl_nr;
        struct list_head         pl_pages;
        struct task_struct      *pl_owner;
};

/** 
 * A 2-queue of pages. A convenience data-type for common use case, 2-queue
 * contains an incoming page list and an outgoing page list.
 */
struct cl_2queue {
        struct cl_page_list c2_qin;
        struct cl_page_list c2_qout;
};

/** @} cl_page_list */

/** \addtogroup cl_io cl_io
 * @{ */
/** \struct cl_io
 * I/O
 *
 * cl_io represents a high level I/O activity like
 * read(2)/write(2)/truncate(2) system call, or cancellation of an extent
 * lock.
 *
 * cl_io is a layered object, much like cl_{object,page,lock} but with one
 * important distinction. We want to minimize number of calls to the allocator
 * in the fast path, e.g., in the case of read(2) when everything is cached:
 * client already owns the lock over region being read, and data are cached
 * due to read-ahead. To avoid allocation of cl_io layers in such situations,
 * per-layer io state is stored in the session, associated with the io, see
 * struct {vvp,lov,osc}_io for example. Sessions allocation is amortized
 * by using free-lists, see cl_env_get().
 *
 * There is a small predefined number of possible io types, enumerated in enum
 * cl_io_type.
 *
 * cl_io is a state machine, that can be advanced concurrently by the multiple
 * threads. It is up to these threads to control the concurrency and,
 * specifically, to detect when io is done, and its state can be safely
 * released.
 *
 * For read/write io overall execution plan is as following:
 *
 *     (0) initialize io state through all layers;
 *
 *     (1) loop: prepare chunk of work to do
 *
 *     (2) call all layers to collect locks they need to process current chunk
 *
 *     (3) sort all locks to avoid dead-locks, and acquire them
 *
 *     (4) process the chunk: call per-page methods
 *         cl_io_operations::cio_prepare_write(),
 *         cl_io_operations::cio_commit_write() for write)
 *
 *     (5) release locks
 *
 *     (6) repeat loop.
 *
 * To implement the "parallel IO mode", lov layer creates sub-io's (lazily to
 * address allocation efficiency issues mentioned above), and returns with the
 * special error condition from per-page method when current sub-io has to
 * block. This causes io loop to be repeated, and lov switches to the next
 * sub-io in its cl_io_operations::cio_iter_init() implementation.
 */

/** IO types */
enum cl_io_type {
        /** read system call */
        CIT_READ,
        /** write system call */
        CIT_WRITE,
        /** truncate, utime system calls */
        CIT_SETATTR,
        /**
         * page fault handling
         */
        CIT_FAULT,
        /**
         * fsync system call handling
         * To write out a range of file
         */
        CIT_FSYNC,
        /**
         * Miscellaneous io. This is used for occasional io activity that
         * doesn't fit into other types. Currently this is used for:
         *
         *     - cancellation of an extent lock. This io exists as a context
         *     to write dirty pages from under the lock being canceled back
         *     to the server;
         *
         *     - VM induced page write-out. An io context for writing page out
         *     for memory cleansing;
         *
         *     - glimpse. An io context to acquire glimpse lock.
         *
         *     - grouplock. An io context to acquire group lock.
         *
         * CIT_MISC io is used simply as a context in which locks and pages
         * are manipulated. Such io has no internal "process", that is,
         * cl_io_loop() is never called for it.
         */
        CIT_MISC,
        CIT_OP_NR
};

/**
 * States of cl_io state machine
 */
enum cl_io_state {
        /** Not initialized. */
        CIS_ZERO,
        /** Initialized. */
        CIS_INIT,
        /** IO iteration started. */
        CIS_IT_STARTED,
        /** Locks taken. */
        CIS_LOCKED,
        /** Actual IO is in progress. */
        CIS_IO_GOING,
        /** IO for the current iteration finished. */
        CIS_IO_FINISHED,
        /** Locks released. */
        CIS_UNLOCKED,
        /** Iteration completed. */
        CIS_IT_ENDED,
        /** cl_io finalized. */
        CIS_FINI
};

/**
 * IO state private for a layer.
 *
 * This is usually embedded into layer session data, rather than allocated
 * dynamically.
 *
 * \see vvp_io, lov_io, osc_io
 */
struct cl_io_slice {
        struct cl_io                    *cis_io;
        /** corresponding object slice. Immutable after creation. */
        struct cl_object                *cis_obj;
        /** io operations. Immutable after creation. */
        const struct cl_io_operations   *cis_iop;
        /**
         * linkage into a list of all slices for a given cl_io, hanging off
         * cl_io::ci_layers. Immutable after creation.
         */
        struct list_head                cis_linkage;
};

typedef void (*cl_commit_cbt)(const struct lu_env *, struct cl_io *,
                              struct cl_page *);

struct cl_read_ahead {
        /* Maximum page index the readahead window will end.
         * This is determined DLM lock coverage, RPC and stripe boundary.
         * cra_end is included. */
        pgoff_t cra_end;
        /* Release routine. If readahead holds resources underneath, this
         * function should be called to release it. */
        void    (*cra_release)(const struct lu_env *env, void *cbdata);
        /* Callback data for cra_release routine */
        void    *cra_cbdata;
};

static inline void cl_read_ahead_release(const struct lu_env *env,
                                         struct cl_read_ahead *ra)
{
        if (ra->cra_release != NULL)
                ra->cra_release(env, ra->cra_cbdata);
        memset(ra, 0, sizeof(*ra));
}


/**
 * Per-layer io operations.
 * \see vvp_io_ops, lov_io_ops, lovsub_io_ops, osc_io_ops
 */
struct cl_io_operations {
        /**
         * Vector of io state transition methods for every io type.
         *
         * \see cl_page_operations::io
         */
        struct {
                /**
                 * Prepare io iteration at a given layer.
                 *
                 * Called top-to-bottom at the beginning of each iteration of
                 * "io loop" (if it makes sense for this type of io). Here
                 * layer selects what work it will do during this iteration.
                 *
                 * \see cl_io_operations::cio_iter_fini()
                 */
                int (*cio_iter_init) (const struct lu_env *env,
                                      const struct cl_io_slice *slice);
                /**
                 * Finalize io iteration.
                 *
                 * Called bottom-to-top at the end of each iteration of "io
                 * loop". Here layers can decide whether IO has to be
                 * continued.
                 *
                 * \see cl_io_operations::cio_iter_init()
                 */
                void (*cio_iter_fini) (const struct lu_env *env,
                                       const struct cl_io_slice *slice);
                /**
                 * Collect locks for the current iteration of io.
                 *
                 * Called top-to-bottom to collect all locks necessary for
                 * this iteration. This methods shouldn't actually enqueue
                 * anything, instead it should post a lock through
                 * cl_io_lock_add(). Once all locks are collected, they are
                 * sorted and enqueued in the proper order.
                 */
                int  (*cio_lock) (const struct lu_env *env,
                                  const struct cl_io_slice *slice);
                /**
                 * Finalize unlocking.
                 *
                 * Called bottom-to-top to finish layer specific unlocking
                 * functionality, after generic code released all locks
                 * acquired by cl_io_operations::cio_lock().
                 */
                void  (*cio_unlock)(const struct lu_env *env,
                                    const struct cl_io_slice *slice);
                /**
                 * Start io iteration.
                 *
                 * Once all locks are acquired, called top-to-bottom to
                 * commence actual IO. In the current implementation,
                 * top-level vvp_io_{read,write}_start() does all the work
                 * synchronously by calling generic_file_*(), so other layers
                 * are called when everything is done.
                 */
                int  (*cio_start)(const struct lu_env *env,
                                  const struct cl_io_slice *slice);
                /**
                 * Called top-to-bottom at the end of io loop. Here layer
                 * might wait for an unfinished asynchronous io.
                 */
                void (*cio_end)  (const struct lu_env *env,
                                  const struct cl_io_slice *slice);
                /**
                 * Called bottom-to-top to notify layers that read/write IO
                 * iteration finished, with \a nob bytes transferred.
                 */
                void (*cio_advance)(const struct lu_env *env,
                                    const struct cl_io_slice *slice,
                                    size_t nob);
                /**
                 * Called once per io, bottom-to-top to release io resources.
                 */
                void (*cio_fini) (const struct lu_env *env,
                                  const struct cl_io_slice *slice);
        } op[CIT_OP_NR];

        /**
         * Submit pages from \a queue->c2_qin for IO, and move
         * successfully submitted pages into \a queue->c2_qout. Return
         * non-zero if failed to submit even the single page. If
         * submission failed after some pages were moved into \a
         * queue->c2_qout, completion callback with non-zero ioret is
         * executed on them.
         */
        int  (*cio_submit)(const struct lu_env *env,
                        const struct cl_io_slice *slice,
                        enum cl_req_type crt,
                        struct cl_2queue *queue);
        /**
         * Queue async page for write.
         * The difference between cio_submit and cio_queue is that
         * cio_submit is for urgent request.
         */
        int  (*cio_commit_async)(const struct lu_env *env,
                        const struct cl_io_slice *slice,
                        struct cl_page_list *queue, int from, int to,
                        cl_commit_cbt cb);
        /**
         * Decide maximum read ahead extent
         *
         * \pre io->ci_type == CIT_READ
         */
        int (*cio_read_ahead)(const struct lu_env *env,
                              const struct cl_io_slice *slice,
                              pgoff_t start, struct cl_read_ahead *ra);
        /**
         * Optional debugging helper. Print given io slice.
         */
        int (*cio_print)(const struct lu_env *env, void *cookie,
                         lu_printer_t p, const struct cl_io_slice *slice);
};

/**
 * Flags to lock enqueue procedure.
 * \ingroup cl_lock
 */
enum cl_enq_flags {
        /**
         * instruct server to not block, if conflicting lock is found. Instead
         * -EWOULDBLOCK is returned immediately.
         */
        CEF_NONBLOCK     = 0x00000001,
        /**
         * take lock asynchronously (out of order), as it cannot
         * deadlock. This is for LDLM_FL_HAS_INTENT locks used for glimpsing.
         */
        CEF_ASYNC        = 0x00000002,
        /**
         * tell the server to instruct (though a flag in the blocking ast) an
         * owner of the conflicting lock, that it can drop dirty pages
         * protected by this lock, without sending them to the server.
         */
        CEF_DISCARD_DATA = 0x00000004,
        /**
         * tell the sub layers that it must be a `real' lock. This is used for
         * mmapped-buffer locks and glimpse locks that must be never converted
         * into lockless mode.
         *
         * \see vvp_mmap_locks(), cl_glimpse_lock().
         */
        CEF_MUST         = 0x00000008,
        /**
         * tell the sub layers that never request a `real' lock. This flag is
         * not used currently.
         *
         * cl_io::ci_lockreq and CEF_{MUST,NEVER} flags specify lockless
         * conversion policy: ci_lockreq describes generic information of lock
         * requirement for this IO, especially for locks which belong to the
         * object doing IO; however, lock itself may have precise requirements
         * that are described by the enqueue flags.
         */
        CEF_NEVER        = 0x00000010,
        /**
         * for async glimpse lock.
         */
        CEF_AGL          = 0x00000020,
        /**
         * enqueue a lock to test DLM lock existence.
         */
        CEF_PEEK        = 0x00000040,
        /**
         * mask of enq_flags.
         */
        CEF_MASK         = 0x0000007f,
};

/**
 * Link between lock and io. Intermediate structure is needed, because the
 * same lock can be part of multiple io's simultaneously.
 */
struct cl_io_lock_link {
        /** linkage into one of cl_lockset lists. */
        struct list_head        cill_linkage;
        struct cl_lock          cill_lock;
        /** optional destructor */
        void                    (*cill_fini)(const struct lu_env *env,
                                             struct cl_io_lock_link *link);
};
#define cill_descr      cill_lock.cll_descr

/**
 * Lock-set represents a collection of locks, that io needs at a
 * time. Generally speaking, client tries to avoid holding multiple locks when
 * possible, because
 *
 *      - holding extent locks over multiple ost's introduces the danger of
 *        "cascading timeouts";
 *
 *      - holding multiple locks over the same ost is still dead-lock prone,
 *        see comment in osc_lock_enqueue(),
 *
 * but there are certain situations where this is unavoidable:
 *
 *      - O_APPEND writes have to take [0, EOF] lock for correctness;
 *
 *      - truncate has to take [new-size, EOF] lock for correctness;
 *
 *      - SNS has to take locks across full stripe for correctness;
 *
 *      - in the case when user level buffer, supplied to {read,write}(file0),
 *        is a part of a memory mapped lustre file, client has to take a dlm
 *        locks on file0, and all files that back up the buffer (or a part of
 *        the buffer, that is being processed in the current chunk, in any
 *        case, there are situations where at least 2 locks are necessary).
 *
 * In such cases we at least try to take locks in the same consistent
 * order. To this end, all locks are first collected, then sorted, and then
 * enqueued.
 */
struct cl_lockset {
        /** locks to be acquired. */
        struct list_head  cls_todo;
        /** locks acquired. */
        struct list_head  cls_done;
};

/**
 * Lock requirements(demand) for IO. It should be cl_io_lock_req,
 * but 'req' is always to be thought as 'request' :-)
 */
enum cl_io_lock_dmd {
        /** Always lock data (e.g., O_APPEND). */
        CILR_MANDATORY = 0,
        /** Layers are free to decide between local and global locking. */
        CILR_MAYBE,
        /** Never lock: there is no cache (e.g., liblustre). */
        CILR_NEVER
};

enum cl_fsync_mode {
        /** start writeback, do not wait for them to finish */
        CL_FSYNC_NONE  = 0,
        /** start writeback and wait for them to finish */
        CL_FSYNC_LOCAL = 1,
        /** discard all of dirty pages in a specific file range */
        CL_FSYNC_DISCARD = 2,
        /** start writeback and make sure they have reached storage before
         * return. OST_SYNC RPC must be issued and finished */
        CL_FSYNC_ALL   = 3
};

struct cl_io_rw_common {
        loff_t      crw_pos;
        size_t      crw_count;
        int         crw_nonblock;
};


/**
 * State for io.
 *
 * cl_io is shared by all threads participating in this IO (in current
 * implementation only one thread advances IO, but parallel IO design and
 * concurrent copy_*_user() require multiple threads acting on the same IO. It
 * is up to these threads to serialize their activities, including updates to
 * mutable cl_io fields.
 */
struct cl_io {
        /** type of this IO. Immutable after creation. */
        enum cl_io_type                ci_type;
        /** current state of cl_io state machine. */
        enum cl_io_state               ci_state;
        /** main object this io is against. Immutable after creation. */
        struct cl_object              *ci_obj;
        /**
         * Upper layer io, of which this io is a part of. Immutable after
         * creation.
         */
        struct cl_io                  *ci_parent;
        /** List of slices. Immutable after creation. */
        struct list_head                ci_layers;
        /** list of locks (to be) acquired by this io. */
        struct cl_lockset              ci_lockset;
        /** lock requirements, this is just a help info for sublayers. */
        enum cl_io_lock_dmd            ci_lockreq;
        union {
                struct cl_rd_io {
                        struct cl_io_rw_common rd;
                } ci_rd;
                struct cl_wr_io {
                        struct cl_io_rw_common wr;
                        int                    wr_append;
                        int                    wr_sync;
                } ci_wr;
                struct cl_io_rw_common ci_rw;
                struct cl_setattr_io {
                        struct ost_lvb   sa_attr;
                        unsigned int     sa_valid;
                        int              sa_stripe_index;
                        struct lu_fid    *sa_parent_fid;
                        struct obd_capa  *sa_capa;
                } ci_setattr;
                struct cl_fault_io {
                        /** page index within file. */
                        pgoff_t         ft_index;
                        /** bytes valid byte on a faulted page. */
                        size_t          ft_nob;
                        /** writable page? for nopage() only */
                        int             ft_writable;
                        /** page of an executable? */
                        int             ft_executable;
                        /** page_mkwrite() */
                        int             ft_mkwrite;
                        /** resulting page */
                        struct cl_page *ft_page;
                } ci_fault;
                struct cl_fsync_io {
                        loff_t             fi_start;
                        loff_t             fi_end;
                        struct obd_capa   *fi_capa;
                        /** file system level fid */
                        struct lu_fid     *fi_fid;
                        enum cl_fsync_mode fi_mode;
                        /* how many pages were written/discarded */
                        unsigned int       fi_nr_written;
                } ci_fsync;
        } u;
        struct cl_2queue     ci_queue;
        size_t               ci_nob;
        int                  ci_result;
        unsigned int         ci_continue:1,
        /**
         * This io has held grouplock, to inform sublayers that
         * don't do lockless i/o.
         */
                             ci_no_srvlock:1,
        /**
         * The whole IO need to be restarted because layout has been changed
         */
                             ci_need_restart:1,
        /**
         * to not refresh layout - the IO issuer knows that the layout won't
         * change(page operations, layout change causes all page to be
         * discarded), or it doesn't matter if it changes(sync).
         */
                             ci_ignore_layout:1,
        /**
         * Check if layout changed after the IO finishes. Mainly for HSM
         * requirement. If IO occurs to openning files, it doesn't need to
         * verify layout because HSM won't release openning files.
         * Right now, only two opertaions need to verify layout: glimpse
         * and setattr.
         */
                             ci_verify_layout:1,
        /**
         * file is released, restore has to to be triggered by vvp layer
         */
                             ci_restore_needed:1,
        /**
         * O_NOATIME
         */
                             ci_noatime:1;
        /**
         * Number of pages owned by this IO. For invariant checking.
         */
        unsigned             ci_owned_nr;
};

/** @} cl_io */

/** \addtogroup cl_req cl_req
 * @{ */
/** \struct cl_req
 * Transfer.
 *
 * There are two possible modes of transfer initiation on the client:
 *
 *     - immediate transfer: this is started when a high level io wants a page
 *       or a collection of pages to be transferred right away. Examples:
 *       read-ahead, synchronous read in the case of non-page aligned write,
 *       page write-out as a part of extent lock cancellation, page write-out
 *       as a part of memory cleansing. Immediate transfer can be both
 *       cl_req_type::CRT_READ and cl_req_type::CRT_WRITE;
 *
 *     - opportunistic transfer (cl_req_type::CRT_WRITE only), that happens
 *       when io wants to transfer a page to the server some time later, when
 *       it can be done efficiently. Example: pages dirtied by the write(2)
 *       path.
 *
 * In any case, transfer takes place in the form of a cl_req, which is a
 * representation for a network RPC.
 *
 * Pages queued for an opportunistic transfer are cached until it is decided
 * that efficient RPC can be composed of them. This decision is made by "a
 * req-formation engine", currently implemented as a part of osc
 * layer. Req-formation depends on many factors: the size of the resulting
 * RPC, whether or not multi-object RPCs are supported by the server,
 * max-rpc-in-flight limitations, size of the dirty cache, etc.
 *
 * For the immediate transfer io submits a cl_page_list, that req-formation
 * engine slices into cl_req's, possibly adding cached pages to some of
 * the resulting req's.
 *
 * Whenever a page from cl_page_list is added to a newly constructed req, its
 * cl_page_operations::cpo_prep() layer methods are called. At that moment,
 * page state is atomically changed from cl_page_state::CPS_OWNED to
 * cl_page_state::CPS_PAGEOUT or cl_page_state::CPS_PAGEIN, cl_page::cp_owner
 * is zeroed, and cl_page::cp_req is set to the
 * req. cl_page_operations::cpo_prep() method at the particular layer might
 * return -EALREADY to indicate that it does not need to submit this page
 * at all. This is possible, for example, if page, submitted for read,
 * became up-to-date in the meantime; and for write, the page don't have
 * dirty bit marked. \see cl_io_submit_rw()
 *
 * Whenever a cached page is added to a newly constructed req, its
 * cl_page_operations::cpo_make_ready() layer methods are called. At that
 * moment, page state is atomically changed from cl_page_state::CPS_CACHED to
 * cl_page_state::CPS_PAGEOUT, and cl_page::cp_req is set to
 * req. cl_page_operations::cpo_make_ready() method at the particular layer
 * might return -EAGAIN to indicate that this page is not eligible for the
 * transfer right now.
 *
 * FUTURE
 *
 * Plan is to divide transfers into "priority bands" (indicated when
 * submitting cl_page_list, and queuing a page for the opportunistic transfer)
 * and allow glueing of cached pages to immediate transfers only within single
 * band. This would make high priority transfers (like lock cancellation or
 * memory pressure induced write-out) really high priority.
 *
 */

/**
 * Per-transfer attributes.
 */
struct cl_req_attr {
        /** Generic attributes for the server consumption. */
        struct obdo     *cra_oa;
        /** Capability. */
        struct obd_capa *cra_capa;
        /** Jobid */
        char             cra_jobid[LUSTRE_JOBID_SIZE];
};

/**
 * Transfer request operations definable at every layer.
 *
 * Concurrency: transfer formation engine synchronizes calls to all transfer
 * methods.
 */
struct cl_req_operations {
        /**
         * Invoked top-to-bottom by cl_req_prep() when transfer formation is
         * complete (all pages are added).
         *
         * \see osc_req_prep()
         */
        int  (*cro_prep)(const struct lu_env *env,
                         const struct cl_req_slice *slice);
        /**
         * Called top-to-bottom to fill in \a oa fields. This is called twice
         * with different flags, see bug 10150 and osc_build_req().
         *
         * \param obj an object from cl_req which attributes are to be set in
         *            \a oa.
         *
         * \param oa struct obdo where attributes are placed
         *
         * \param flags \a oa fields to be filled.
         */
        void (*cro_attr_set)(const struct lu_env *env,
                             const struct cl_req_slice *slice,
                             const struct cl_object *obj,
                             struct cl_req_attr *attr, u64 flags);
        /**
         * Called top-to-bottom from cl_req_completion() to notify layers that
         * transfer completed. Has to free all state allocated by
         * cl_device_operations::cdo_req_init().
         */
        void (*cro_completion)(const struct lu_env *env,
                               const struct cl_req_slice *slice, int ioret);
};

/**
 * A per-object state that (potentially multi-object) transfer request keeps.
 */
struct cl_req_obj {
        /** object itself */
        struct cl_object   *ro_obj;
        /** reference to cl_req_obj::ro_obj. For debugging. */
        struct lu_ref_link  ro_obj_ref;
        /* something else? Number of pages for a given object? */
};

/**
 * Transfer request.
 *
 * Transfer requests are not reference counted, because IO sub-system owns
 * them exclusively and knows when to free them.
 *
 * Life cycle.
 *
 * cl_req is created by cl_req_alloc() that calls
 * cl_device_operations::cdo_req_init() device methods to allocate per-req
 * state in every layer.
 *
 * Then pages are added (cl_req_page_add()), req keeps track of all objects it
 * contains pages for.
 *
 * Once all pages were collected, cl_page_operations::cpo_prep() method is
 * called top-to-bottom. At that point layers can modify req, let it pass, or
 * deny it completely. This is to support things like SNS that have transfer
 * ordering requirements invisible to the individual req-formation engine.
 *
 * On transfer completion (or transfer timeout, or failure to initiate the
 * transfer of an allocated req), cl_req_operations::cro_completion() method
 * is called, after execution of cl_page_operations::cpo_completion() of all
 * req's pages.
 */
struct cl_req {
        enum cl_req_type        crq_type;
        /** A list of pages being transfered */
        struct list_head        crq_pages;
        /** Number of pages in cl_req::crq_pages */
        unsigned                crq_nrpages;
        /** An array of objects which pages are in ->crq_pages */
        struct cl_req_obj       *crq_o;
        /** Number of elements in cl_req::crq_objs[] */
        unsigned                crq_nrobjs;
        struct list_head        crq_layers;
};

/**
 * Per-layer state for request.
 */
struct cl_req_slice {
        struct cl_req                   *crs_req;
        struct cl_device                *crs_dev;
        struct list_head                 crs_linkage;
        const struct cl_req_operations  *crs_ops;
};

/* @} cl_req */

enum cache_stats_item {
        /** how many cache lookups were performed */
        CS_lookup = 0,
        /** how many times cache lookup resulted in a hit */
        CS_hit,
        /** how many entities are in the cache right now */
        CS_total,
        /** how many entities in the cache are actively used (and cannot be
         * evicted) right now */
        CS_busy,
        /** how many entities were created at all */
        CS_create,
        CS_NR
};

#define CS_NAMES { "lookup", "hit", "total", "busy", "create" }

/**
 * Stats for a generic cache (similar to inode, lu_object, etc. caches).
 */
struct cache_stats {
        const char      *cs_name;
        atomic_t        cs_stats[CS_NR];
};

/** These are not exported so far */
void cache_stats_init (struct cache_stats *cs, const char *name);

/**
 * Client-side site. This represents particular client stack. "Global"
 * variables should (directly or indirectly) be added here to allow multiple
 * clients to co-exist in the single address space.
 */
struct cl_site {
        struct lu_site          cs_lu;
        /**
         * Statistical counters. Atomics do not scale, something better like
         * per-cpu counters is needed.
         *
         * These are exported as /proc/fs/lustre/llite/.../site
         *
         * When interpreting keep in mind that both sub-locks (and sub-pages)
         * and top-locks (and top-pages) are accounted here.
         */
        struct cache_stats      cs_pages;
        atomic_t                cs_pages_state[CPS_NR];
};

int  cl_site_init(struct cl_site *s, struct cl_device *top);
void cl_site_fini(struct cl_site *s);
void cl_stack_fini(const struct lu_env *env, struct cl_device *cl);

/**
 * Output client site statistical counters into a buffer. Suitable for
 * ll_rd_*()-style functions.
 */
int cl_site_stats_print(const struct cl_site *site, struct seq_file *m);

/**
 * \name helpers
 *
 * Type conversion and accessory functions.
 */
/** @{ */

static inline struct cl_site *lu2cl_site(const struct lu_site *site)
{
        return container_of(site, struct cl_site, cs_lu);
}

static inline struct cl_device *lu2cl_dev(const struct lu_device *d)
{
        LASSERT(d == NULL || IS_ERR(d) || lu_device_is_cl(d));
        return container_of0(d, struct cl_device, cd_lu_dev);
}

static inline struct lu_device *cl2lu_dev(struct cl_device *d)
{
        return &d->cd_lu_dev;
}

static inline struct cl_object *lu2cl(const struct lu_object *o)
{
        LASSERT(o == NULL || IS_ERR(o) || lu_device_is_cl(o->lo_dev));
        return container_of0(o, struct cl_object, co_lu);
}

static inline const struct cl_object_conf *
lu2cl_conf(const struct lu_object_conf *conf)
{
        return container_of0(conf, struct cl_object_conf, coc_lu);
}

static inline struct cl_object *cl_object_next(const struct cl_object *obj)
{
        return obj ? lu2cl(lu_object_next(&obj->co_lu)) : NULL;
}

static inline struct cl_object_header *luh2coh(const struct lu_object_header *h)
{
        return container_of0(h, struct cl_object_header, coh_lu);
}

static inline struct cl_site *cl_object_site(const struct cl_object *obj)
{
        return lu2cl_site(obj->co_lu.lo_dev->ld_site);
}

static inline
struct cl_object_header *cl_object_header(const struct cl_object *obj)
{
        return luh2coh(obj->co_lu.lo_header);
}

static inline int cl_device_init(struct cl_device *d, struct lu_device_type *t)
{
        return lu_device_init(&d->cd_lu_dev, t);
}

static inline void cl_device_fini(struct cl_device *d)
{
        lu_device_fini(&d->cd_lu_dev);
}

void cl_page_slice_add(struct cl_page *page, struct cl_page_slice *slice,
                       struct cl_object *obj, pgoff_t index,
                       const struct cl_page_operations *ops);
void cl_lock_slice_add(struct cl_lock *lock, struct cl_lock_slice *slice,
                       struct cl_object *obj,
                       const struct cl_lock_operations *ops);
void cl_io_slice_add(struct cl_io *io, struct cl_io_slice *slice,
                     struct cl_object *obj, const struct cl_io_operations *ops);
void cl_req_slice_add(struct cl_req *req, struct cl_req_slice *slice,
                      struct cl_device *dev,
                      const struct cl_req_operations *ops);
/** @} helpers */

/** \defgroup cl_object cl_object
 * @{ */
struct cl_object *cl_object_top (struct cl_object *o);
struct cl_object *cl_object_find(const struct lu_env *env, struct cl_device *cd,
                                 const struct lu_fid *fid,
                                 const struct cl_object_conf *c);

int  cl_object_header_init(struct cl_object_header *h);
void cl_object_header_fini(struct cl_object_header *h);
void cl_object_put        (const struct lu_env *env, struct cl_object *o);
void cl_object_get        (struct cl_object *o);
void cl_object_attr_lock  (struct cl_object *o);
void cl_object_attr_unlock(struct cl_object *o);
int  cl_object_attr_get   (const struct lu_env *env, struct cl_object *obj,
                           struct cl_attr *attr);
int  cl_object_attr_update(const struct lu_env *env, struct cl_object *obj,
                           const struct cl_attr *attr, unsigned valid);
int  cl_object_glimpse    (const struct lu_env *env, struct cl_object *obj,
                           struct ost_lvb *lvb);
int  cl_conf_set          (const struct lu_env *env, struct cl_object *obj,
                           const struct cl_object_conf *conf);
int  cl_object_prune      (const struct lu_env *env, struct cl_object *obj);
void cl_object_kill       (const struct lu_env *env, struct cl_object *obj);
int cl_object_getstripe(const struct lu_env *env, struct cl_object *obj,
                        struct lov_user_md __user *lum);
int cl_object_find_cbdata(const struct lu_env *env, struct cl_object *obj,
                          ldlm_iterator_t iter, void *data);

/**
 * Returns true, iff \a o0 and \a o1 are slices of the same object.
 */
static inline int cl_object_same(struct cl_object *o0, struct cl_object *o1)
{
        return cl_object_header(o0) == cl_object_header(o1);
}

static inline void cl_object_page_init(struct cl_object *clob, int size)
{
        clob->co_slice_off = cl_object_header(clob)->coh_page_bufsize;
        cl_object_header(clob)->coh_page_bufsize += cfs_size_round(size);
        WARN_ON(cl_object_header(clob)->coh_page_bufsize > 512);
}

static inline void *cl_object_page_slice(struct cl_object *clob,
                                         struct cl_page *page)
{
        return (void *)((char *)page + clob->co_slice_off);
}

/**
 * Return refcount of cl_object.
 */
static inline int cl_object_refc(struct cl_object *clob)
{
        struct lu_object_header *header = clob->co_lu.lo_header;
        return atomic_read(&header->loh_ref);
}

/** @} cl_object */

/** \defgroup cl_page cl_page
 * @{ */
enum {
        CLP_GANG_OKAY = 0,
        CLP_GANG_RESCHED,
        CLP_GANG_AGAIN,
        CLP_GANG_ABORT
};
/* callback of cl_page_gang_lookup() */

struct cl_page *cl_page_find        (const struct lu_env *env,
                                     struct cl_object *obj,
                                     pgoff_t idx, struct page *vmpage,
                                     enum cl_page_type type);
struct cl_page *cl_page_alloc       (const struct lu_env *env,
                                     struct cl_object *o, pgoff_t ind,
                                     struct page *vmpage,
                                     enum cl_page_type type);
void            cl_page_get         (struct cl_page *page);
void            cl_page_put         (const struct lu_env *env,
                                     struct cl_page *page);
void            cl_page_print       (const struct lu_env *env, void *cookie,
                                     lu_printer_t printer,
                                     const struct cl_page *pg);
void            cl_page_header_print(const struct lu_env *env, void *cookie,
                                     lu_printer_t printer,
                                     const struct cl_page *pg);
struct cl_page *cl_vmpage_page      (struct page *vmpage, struct cl_object *obj);
struct cl_page *cl_page_top         (struct cl_page *page);

const struct cl_page_slice *cl_page_at(const struct cl_page *page,
                                       const struct lu_device_type *dtype);

/**
 * \name ownership
 *
 * Functions dealing with the ownership of page by io.
 */
/** @{ */

int  cl_page_own        (const struct lu_env *env,
                         struct cl_io *io, struct cl_page *page);
int  cl_page_own_try    (const struct lu_env *env,
                         struct cl_io *io, struct cl_page *page);
void cl_page_assume     (const struct lu_env *env,
                         struct cl_io *io, struct cl_page *page);
void cl_page_unassume   (const struct lu_env *env,
                         struct cl_io *io, struct cl_page *pg);
void cl_page_disown     (const struct lu_env *env,
                         struct cl_io *io, struct cl_page *page);
int  cl_page_is_owned   (const struct cl_page *pg, const struct cl_io *io);

/** @} ownership */

/**
 * \name transfer
 *
 * Functions dealing with the preparation of a page for a transfer, and
 * tracking transfer state.
 */
/** @{ */
int  cl_page_prep       (const struct lu_env *env, struct cl_io *io,
                         struct cl_page *pg, enum cl_req_type crt);
void cl_page_completion (const struct lu_env *env,
                         struct cl_page *pg, enum cl_req_type crt, int ioret);
int  cl_page_make_ready (const struct lu_env *env, struct cl_page *pg,
                         enum cl_req_type crt);
int  cl_page_cache_add  (const struct lu_env *env, struct cl_io *io,
                         struct cl_page *pg, enum cl_req_type crt);
void cl_page_clip       (const struct lu_env *env, struct cl_page *pg,
                         int from, int to);
int  cl_page_cancel     (const struct lu_env *env, struct cl_page *page);
int  cl_page_flush      (const struct lu_env *env, struct cl_io *io,
                         struct cl_page *pg);

/** @} transfer */


/**
 * \name helper routines
 * Functions to discard, delete and export a cl_page.
 */
/** @{ */
void    cl_page_discard(const struct lu_env *env, struct cl_io *io,
                        struct cl_page *pg);
void    cl_page_delete(const struct lu_env *env, struct cl_page *pg);
int     cl_page_is_vmlocked(const struct lu_env *env,
                            const struct cl_page *pg);
void    cl_page_export(const struct lu_env *env,
                       struct cl_page *pg, int uptodate);
loff_t  cl_offset(const struct cl_object *obj, pgoff_t idx);
pgoff_t cl_index(const struct cl_object *obj, loff_t offset);
size_t  cl_page_size(const struct cl_object *obj);

void cl_lock_print(const struct lu_env *env, void *cookie,
                   lu_printer_t printer, const struct cl_lock *lock);
void cl_lock_descr_print(const struct lu_env *env, void *cookie,
                         lu_printer_t printer,
                         const struct cl_lock_descr *descr);
/* @} helper */

/**
 * Data structure managing a client's cached pages. A count of
 * "unstable" pages is maintained, and an LRU of clean pages is
 * maintained. "unstable" pages are pages pinned by the ptlrpc
 * layer for recovery purposes.
 */
struct cl_client_cache {
        /**
         * # of client cache refcount
         * # of users (OSCs) + 2 (held by llite and lov)
         */
        atomic_t                ccc_users;
        /**
         * # of threads are doing shrinking
         */
        unsigned int            ccc_lru_shrinkers;
        /**
         * # of LRU entries available
         */
        atomic_long_t           ccc_lru_left;
        /**
         * List of entities(OSCs) for this LRU cache
         */
        struct list_head        ccc_lru;
        /**
         * Max # of LRU entries
         */
        unsigned long           ccc_lru_max;
        /**
         * Lock to protect ccc_lru list
         */
        spinlock_t              ccc_lru_lock;
        /**
         * Set if unstable check is enabled
         */
        unsigned int            ccc_unstable_check:1;
        /**
         * # of unstable pages for this mount point
         */
        atomic_long_t           ccc_unstable_nr;
        /**
         * Waitq for awaiting unstable pages to reach zero.
         * Used at umounting time and signaled on BRW commit
         */
        wait_queue_head_t       ccc_unstable_waitq;
};
/**
 * cl_cache functions
 */
struct cl_client_cache *cl_cache_init(unsigned long lru_page_max);
void cl_cache_incref(struct cl_client_cache *cache);
void cl_cache_decref(struct cl_client_cache *cache);

/** @} cl_page */

/** \defgroup cl_lock cl_lock
 * @{ */
int cl_lock_request(const struct lu_env *env, struct cl_io *io,
                    struct cl_lock *lock);
int cl_lock_init(const struct lu_env *env, struct cl_lock *lock,
                 const struct cl_io *io);
void cl_lock_fini(const struct lu_env *env, struct cl_lock *lock);
const struct cl_lock_slice *cl_lock_at(const struct cl_lock *lock,
                                       const struct lu_device_type *dtype);
void cl_lock_release(const struct lu_env *env, struct cl_lock *lock);

int cl_lock_enqueue(const struct lu_env *env, struct cl_io *io,
                    struct cl_lock *lock, struct cl_sync_io *anchor);
void cl_lock_cancel(const struct lu_env *env, struct cl_lock *lock);

/** @} cl_lock */

/** \defgroup cl_io cl_io
 * @{ */

int   cl_io_init         (const struct lu_env *env, struct cl_io *io,
                          enum cl_io_type iot, struct cl_object *obj);
int   cl_io_sub_init     (const struct lu_env *env, struct cl_io *io,
                          enum cl_io_type iot, struct cl_object *obj);
int   cl_io_rw_init      (const struct lu_env *env, struct cl_io *io,
                          enum cl_io_type iot, loff_t pos, size_t count);
int   cl_io_loop         (const struct lu_env *env, struct cl_io *io);

void  cl_io_fini         (const struct lu_env *env, struct cl_io *io);
int   cl_io_iter_init    (const struct lu_env *env, struct cl_io *io);
void  cl_io_iter_fini    (const struct lu_env *env, struct cl_io *io);
int   cl_io_lock         (const struct lu_env *env, struct cl_io *io);
void  cl_io_unlock       (const struct lu_env *env, struct cl_io *io);
int   cl_io_start        (const struct lu_env *env, struct cl_io *io);
void  cl_io_end          (const struct lu_env *env, struct cl_io *io);
int   cl_io_lock_add     (const struct lu_env *env, struct cl_io *io,
                          struct cl_io_lock_link *link);
int   cl_io_lock_alloc_add(const struct lu_env *env, struct cl_io *io,
                           struct cl_lock_descr *descr);
int   cl_io_submit_rw    (const struct lu_env *env, struct cl_io *io,
                          enum cl_req_type iot, struct cl_2queue *queue);
int   cl_io_submit_sync  (const struct lu_env *env, struct cl_io *io,
                          enum cl_req_type iot, struct cl_2queue *queue,
                          long timeout);
int   cl_io_commit_async (const struct lu_env *env, struct cl_io *io,
                          struct cl_page_list *queue, int from, int to,
                          cl_commit_cbt cb);
int   cl_io_read_ahead   (const struct lu_env *env, struct cl_io *io,
                          pgoff_t start, struct cl_read_ahead *ra);
void  cl_io_rw_advance   (const struct lu_env *env, struct cl_io *io,
                          size_t nob);
int   cl_io_cancel       (const struct lu_env *env, struct cl_io *io,
                          struct cl_page_list *queue);
int   cl_io_is_going     (const struct lu_env *env);

/**
 * True, iff \a io is an O_APPEND write(2).
 */
static inline int cl_io_is_append(const struct cl_io *io)
{
        return io->ci_type == CIT_WRITE && io->u.ci_wr.wr_append;
}

static inline int cl_io_is_sync_write(const struct cl_io *io)
{
        return io->ci_type == CIT_WRITE && io->u.ci_wr.wr_sync;
}

static inline int cl_io_is_mkwrite(const struct cl_io *io)
{
        return io->ci_type == CIT_FAULT && io->u.ci_fault.ft_mkwrite;
}

/**
 * True, iff \a io is a truncate(2).
 */
static inline int cl_io_is_trunc(const struct cl_io *io)
{
        return io->ci_type == CIT_SETATTR &&
                (io->u.ci_setattr.sa_valid & ATTR_SIZE);
}

struct cl_io *cl_io_top(struct cl_io *io);

void cl_io_print(const struct lu_env *env, void *cookie,
                 lu_printer_t printer, const struct cl_io *io);

#define CL_IO_SLICE_CLEAN(foo_io, base)                                 \
do {                                                                    \
        typeof(foo_io) __foo_io = (foo_io);                             \
                                                                        \
        CLASSERT(offsetof(typeof(*__foo_io), base) == 0);               \
        memset(&__foo_io->base + 1, 0,                                  \
               (sizeof *__foo_io) - sizeof __foo_io->base);             \
} while (0)

/** @} cl_io */

/** \defgroup cl_page_list cl_page_list
 * @{ */

/**
 * Last page in the page list.
 */
static inline struct cl_page *cl_page_list_last(struct cl_page_list *plist)
{
        LASSERT(plist->pl_nr > 0);
        return list_entry(plist->pl_pages.prev, struct cl_page, cp_batch);
}

static inline struct cl_page *cl_page_list_first(struct cl_page_list *plist)
{
        LASSERT(plist->pl_nr > 0);
        return list_entry(plist->pl_pages.next, struct cl_page, cp_batch);
}

/**
 * Iterate over pages in a page list.
 */
#define cl_page_list_for_each(page, list)                               \
        list_for_each_entry((page), &(list)->pl_pages, cp_batch)

/**
 * Iterate over pages in a page list, taking possible removals into account.
 */
#define cl_page_list_for_each_safe(page, temp, list)                    \
        list_for_each_entry_safe((page), (temp), &(list)->pl_pages, cp_batch)

void cl_page_list_init   (struct cl_page_list *plist);
void cl_page_list_add    (struct cl_page_list *plist, struct cl_page *page);
void cl_page_list_move   (struct cl_page_list *dst, struct cl_page_list *src,
                          struct cl_page *page);
void cl_page_list_move_head(struct cl_page_list *dst, struct cl_page_list *src,
                          struct cl_page *page);
void cl_page_list_splice (struct cl_page_list *list,
                          struct cl_page_list *head);
void cl_page_list_del    (const struct lu_env *env,
                          struct cl_page_list *plist, struct cl_page *page);
void cl_page_list_disown (const struct lu_env *env,
                          struct cl_io *io, struct cl_page_list *plist);
int  cl_page_list_own    (const struct lu_env *env,
                          struct cl_io *io, struct cl_page_list *plist);
void cl_page_list_assume (const struct lu_env *env,
                          struct cl_io *io, struct cl_page_list *plist);
void cl_page_list_discard(const struct lu_env *env,
                          struct cl_io *io, struct cl_page_list *plist);
void cl_page_list_fini   (const struct lu_env *env, struct cl_page_list *plist);

void cl_2queue_init     (struct cl_2queue *queue);
void cl_2queue_add      (struct cl_2queue *queue, struct cl_page *page);
void cl_2queue_disown   (const struct lu_env *env,
                         struct cl_io *io, struct cl_2queue *queue);
void cl_2queue_assume   (const struct lu_env *env,
                         struct cl_io *io, struct cl_2queue *queue);
void cl_2queue_discard  (const struct lu_env *env,
                         struct cl_io *io, struct cl_2queue *queue);
void cl_2queue_fini     (const struct lu_env *env, struct cl_2queue *queue);
void cl_2queue_init_page(struct cl_2queue *queue, struct cl_page *page);

/** @} cl_page_list */

/** \defgroup cl_req cl_req
 * @{ */
struct cl_req *cl_req_alloc(const struct lu_env *env, struct cl_page *page,
                            enum cl_req_type crt, int nr_objects);

void cl_req_page_add  (const struct lu_env *env, struct cl_req *req,
                       struct cl_page *page);
void cl_req_page_done (const struct lu_env *env, struct cl_page *page);
int  cl_req_prep      (const struct lu_env *env, struct cl_req *req);
void cl_req_attr_set(const struct lu_env *env, struct cl_req *req,
                     struct cl_req_attr *attr, u64 flags);
void cl_req_completion(const struct lu_env *env, struct cl_req *req, int ioret);

/** \defgroup cl_sync_io cl_sync_io
 * @{ */

/**
 * Anchor for synchronous transfer. This is allocated on a stack by thread
 * doing synchronous transfer, and a pointer to this structure is set up in
 * every page submitted for transfer. Transfer completion routine updates
 * anchor and wakes up waiting thread when transfer is complete.
 */
struct cl_sync_io {
        /** number of pages yet to be transferred. */
        atomic_t                csi_sync_nr;
        /** error code. */
        int                     csi_sync_rc;
        /** barrier of destroy this structure */
        atomic_t                csi_barrier;
        /** completion to be signaled when transfer is complete. */
        wait_queue_head_t       csi_waitq;
        /** callback to invoke when this IO is finished */
        void                    (*csi_end_io)(const struct lu_env *,
                                              struct cl_sync_io *);
};

void cl_sync_io_init(struct cl_sync_io *anchor, int nr,
                     void (*end)(const struct lu_env *, struct cl_sync_io *));
int  cl_sync_io_wait(const struct lu_env *env, struct cl_sync_io *anchor,
                     long timeout);
void cl_sync_io_note(const struct lu_env *env, struct cl_sync_io *anchor,
                     int ioret);
void cl_sync_io_end(const struct lu_env *env, struct cl_sync_io *anchor);

/** @} cl_sync_io */

/** @} cl_req */

/** \defgroup cl_env cl_env
 *
 * lu_env handling for a client.
 *
 * lu_env is an environment within which lustre code executes. Its major part
 * is lu_context---a fast memory allocation mechanism that is used to conserve
 * precious kernel stack space. Originally lu_env was designed for a server,
 * where
 *
 *     - there is a (mostly) fixed number of threads, and
 *
 *     - call chains have no non-lustre portions inserted between lustre code.
 *
 * On a client both these assumtpion fails, because every user thread can
 * potentially execute lustre code as part of a system call, and lustre calls
 * into VFS or MM that call back into lustre.
 *
 * To deal with that, cl_env wrapper functions implement the following
 * optimizations:
 *
 *     - allocation and destruction of environment is amortized by caching no
 *     longer used environments instead of destroying them;
 *
 *     - there is a notion of "current" environment, attached to the kernel
 *     data structure representing current thread Top-level lustre code
 *     allocates an environment and makes it current, then calls into
 *     non-lustre code, that in turn calls lustre back. Low-level lustre
 *     code thus called can fetch environment created by the top-level code
 *     and reuse it, avoiding additional environment allocation.
 *       Right now, three interfaces can attach the cl_env to running thread:
 *       - cl_env_get
 *       - cl_env_implant
 *       - cl_env_reexit(cl_env_reenter had to be called priorly)
 *
 * \see lu_env, lu_context, lu_context_key
 * @{ */

struct cl_env_nest {
        int   cen_refcheck;
        void *cen_cookie;
};

struct lu_env *cl_env_peek       (int *refcheck);
struct lu_env *cl_env_get        (int *refcheck);
struct lu_env *cl_env_alloc      (int *refcheck, __u32 tags);
struct lu_env *cl_env_nested_get (struct cl_env_nest *nest);
void           cl_env_put        (struct lu_env *env, int *refcheck);
void           cl_env_nested_put (struct cl_env_nest *nest, struct lu_env *env);
void          *cl_env_reenter    (void);
void           cl_env_reexit     (void *cookie);
void           cl_env_implant    (struct lu_env *env, int *refcheck);
void           cl_env_unplant    (struct lu_env *env, int *refcheck);
unsigned       cl_env_cache_purge(unsigned nr);
struct lu_env *cl_env_percpu_get (void);
void           cl_env_percpu_put (struct lu_env *env);

/** @} cl_env */

/*
 * Misc
 */
void cl_attr2lvb(struct ost_lvb *lvb, const struct cl_attr *attr);
void cl_lvb2attr(struct cl_attr *attr, const struct ost_lvb *lvb);

struct cl_device *cl_type_setup(const struct lu_env *env, struct lu_site *site,
                                struct lu_device_type *ldt,
                                struct lu_device *next);
/** @} clio */

int cl_global_init(void);
void cl_global_fini(void);

#endif /* _LINUX_CL_OBJECT_H */