*
* 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.
+ * http://www.gnu.org/licenses/gpl-2.0.html
*
* 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.
+ * Copyright (c) 2011, 2016, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
* 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
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_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);
-};
+struct cl_req_attr;
/**
* Device in the client stack.
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
*/
loff_t cat_kms;
/** Modification time. Measured in seconds since epoch. */
- time_t cat_mtime;
+ time64_t cat_mtime;
/** Access time. Measured in seconds since epoch. */
- time_t cat_atime;
+ time64_t cat_atime;
/** Change time. Measured in seconds since epoch. */
- time_t cat_ctime;
+ time64_t cat_ctime;
/**
* Blocks allocated to this cl_object on the server file system.
*
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);
- /**
* Get FIEMAP mapping from the object.
*/
int (*coo_fiemap)(const struct lu_env *env, struct cl_object *obj,
* Get maximum size of the object.
*/
loff_t (*coo_maxbytes)(struct cl_object *obj);
+ /**
+ * Set request attributes.
+ */
+ void (*coo_req_attr_set)(const struct lu_env *env,
+ struct cl_object *obj,
+ struct cl_req_attr *attr);
};
/**
*
* - [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;
+ * into an RPC being constructed, and yanks it from the cache;
*
* - [cl_page_state::CPS_FREEING] VM callback is executed to
* evict the page form the memory;
* 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,
+ * construction of read request from cached, presumably not up to date,
* pages.
*
* Underlying VM page is locked for the duration of transfer.
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 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.
* 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. */
/**
* Requested transfer type.
- * \ingroup cl_req
*/
enum cl_req_type {
CRT_READ,
/**
* \name transfer
*
- * Transfer methods. See comment on cl_req for a description of
- * transfer formation and life-cycle.
+ * Transfer methods.
*
* @{
*/
int ioret);
/**
* Called when cached page is about to be added to the
- * cl_req as a part of req formation.
+ * ptlrpc request as a part of req formation.
*
* \return 0 : proceed with this page;
* \return -EAGAIN : skip this page;
/** IO types */
enum cl_io_type {
/** read system call */
- CIT_READ,
+ CIT_READ = 1,
/** write system call */
CIT_WRITE,
/** truncate, utime system calls */
* cl_io_loop() is never called for it.
*/
CIT_MISC,
+ /**
+ * ladvise handling
+ * To give advice about access of a file
+ */
+ CIT_LADVISE,
CIT_OP_NR
};
* 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
+ /* optimal RPC size for this read, by pages */
+ unsigned long cra_rpc_size;
+ /* Release callback. 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 */
*/
CEF_PEEK = 0x00000040,
/**
+ * Lock match only. Used by group lock in I/O as group lock
+ * is known to exist.
+ */
+ CEF_LOCK_MATCH = 0x00000080,
+ /**
* mask of enq_flags.
*/
- CEF_MASK = 0x0000007f,
+ CEF_MASK = 0x000000ff,
};
/**
int crw_nonblock;
};
-
/**
* State for io.
*
/* how many pages were written/discarded */
unsigned int fi_nr_written;
} ci_fsync;
+ struct cl_ladvise_io {
+ __u64 li_start;
+ __u64 li_end;
+ /** file system level fid */
+ struct lu_fid *li_fid;
+ enum lu_ladvise_type li_advice;
+ __u64 li_flags;
+ } ci_ladvise;
} u;
struct cl_2queue ci_queue;
size_t ci_nob;
/** @} 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 {
+ enum cl_req_type cra_type;
+ u64 cra_flags;
+ struct cl_page *cra_page;
/** Generic attributes for the server consumption. */
struct obdo *cra_oa;
/** 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 transferred */
- 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,
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
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);
int cl_object_fiemap(const struct lu_env *env, struct cl_object *obj,
struct ll_fiemap_info_key *fmkey, struct fiemap *fiemap,
size_t *buflen);
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).
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,
/** @} 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);
+void cl_req_attr_set(const struct lu_env *env, struct cl_object *obj,
+ struct cl_req_attr *attr);
/** \defgroup cl_sync_io cl_sync_io
* @{ */
/** @} cl_sync_io */
-/** @} cl_req */
-
/** \defgroup cl_env cl_env
*
* lu_env handling for a client.
* - 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);
+struct lu_env *cl_env_get(__u16 *refcheck);
+struct lu_env *cl_env_alloc(__u16 *refcheck, __u32 tags);
+void cl_env_put(struct lu_env *env, __u16 *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 */