Whamcloud - gitweb
LU-2388 statahead: don't statahead if it's stopped
[fs/lustre-release.git] / lustre / llite / llite_internal.h
index 4803907..c8f0f3b 100644 (file)
@@ -1,6 +1,4 @@
-/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
- * vim:expandtab:shiftwidth=8:tabstop=8:
- *
+/*
  * GPL HEADER START
  *
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
@@ -28,9 +26,8 @@
 /*
  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
  * Use is subject to license terms.
- */
-/*
- * Copyright (c) 2011 Whamcloud, Inc.
+ *
+ * Copyright (c) 2011, 2012, Whamcloud, Inc.
  */
 /*
  * This file is part of Lustre, http://www.lustre.org/
 #define VM_FAULT_RETRY 0
 #endif
 
+/* Kernel 3.1 kills LOOKUP_CONTINUE, LOOKUP_PARENT is equivalent to it.
+ * seem kernel commit 49084c3bb2055c401f3493c13edae14d49128ca0 */
+#ifndef LOOKUP_CONTINUE
+#define LOOKUP_CONTINUE LOOKUP_PARENT
+#endif
+
 /** Only used on client-side for indicating the tail of dir hash/offset. */
 #define LL_DIR_END_OFF          0x7fffffffffffffffULL
 #define LL_DIR_END_OFF_32BIT    0x7fffffffUL
 
-#ifndef DCACHE_LUSTRE_INVALID
-#define DCACHE_LUSTRE_INVALID 0x4000000
-#endif
-
 #define LL_IT2STR(it) ((it) ? ldlm_it2str((it)->it_op) : "0")
 #define LUSTRE_FPRIVATE(file) ((file)->private_data)
 
 struct ll_dentry_data {
-        int                      lld_cwd_count;
-        int                      lld_mnt_count;
-        struct obd_client_handle lld_cwd_och;
-        struct obd_client_handle lld_mnt_och;
-        struct lookup_intent    *lld_it;
-        unsigned int             lld_sa_generation;
+       int                             lld_cwd_count;
+       int                             lld_mnt_count;
+       struct obd_client_handle        lld_cwd_och;
+       struct obd_client_handle        lld_mnt_och;
+       struct lookup_intent            *lld_it;
+       unsigned int                    lld_sa_generation;
+       unsigned int                    lld_invalid:1;
+       struct rcu_head                 lld_rcu_head;
 };
 
 #define ll_d2d(de) ((struct ll_dentry_data*)((de)->d_fsdata))
@@ -127,15 +128,15 @@ enum lli_flags {
 };
 
 struct ll_inode_info {
-        __u32                           lli_inode_magic;
-        __u32                           lli_flags;
-        __u64                           lli_ioepoch;
+       __u32                           lli_inode_magic;
+       __u32                           lli_flags;
+       __u64                           lli_ioepoch;
 
-        cfs_spinlock_t                  lli_lock;
-        struct posix_acl               *lli_posix_acl;
+       spinlock_t                      lli_lock;
+       struct posix_acl                *lli_posix_acl;
 
-        cfs_hlist_head_t               *lli_remote_perms;
-        cfs_semaphore_t                 lli_rmtperm_sem;
+       cfs_hlist_head_t                *lli_remote_perms;
+       struct mutex                            lli_rmtperm_mutex;
 
         /* identifying fields for both metadata and data stacks. */
         struct lu_fid                   lli_fid;
@@ -164,22 +165,22 @@ struct ll_inode_info {
         __u64                           lli_open_fd_read_count;
         __u64                           lli_open_fd_write_count;
         __u64                           lli_open_fd_exec_count;
-        /* Protects access to och pointers and their usage counters, also
-         * atomicity of check-update of lli_smd */
-        cfs_semaphore_t                 lli_och_sem;
+        /* Protects access to och pointers and their usage counters */
+       struct mutex                    lli_och_mutex;
 
-        struct inode                    lli_vfs_inode;
+       struct inode                    lli_vfs_inode;
 
-        /* the most recent timestamps obtained from mds */
-        struct ost_lvb                  lli_lvb;
+       /* the most recent timestamps obtained from mds */
+       struct ost_lvb                  lli_lvb;
+       spinlock_t                      lli_agl_lock;
 
-        /* Try to make the d::member and f::member are aligned. Before using
-         * these members, make clear whether it is directory or not. */
-        union {
-                /* for directory */
-                struct {
-                        /* serialize normal readdir and statahead-readdir. */
-                        cfs_semaphore_t                 d_readdir_sem;
+       /* Try to make the d::member and f::member are aligned. Before using
+        * these members, make clear whether it is directory or not. */
+       union {
+               /* for directory */
+               struct {
+                       /* serialize normal readdir and statahead-readdir. */
+                       struct mutex                    d_readdir_mutex;
 
                         /* metadata statahead */
                         /* since parent-child threads can share the same @file
@@ -188,42 +189,54 @@ struct ll_inode_info {
                          * cleanup the dir readahead. */
                         void                           *d_opendir_key;
                         struct ll_statahead_info       *d_sai;
-                        __u64                           d_sa_pos;
                         struct posix_acl               *d_def_acl;
                         /* protect statahead stuff. */
-                        cfs_spinlock_t                  d_sa_lock;
-                        /* "opendir_pid" is the token when lookup/revalid
-                         * -- I am the owner of dir statahead. */
-                        pid_t                           d_opendir_pid;
-                } d;
+                       spinlock_t                      d_sa_lock;
+                       /* "opendir_pid" is the token when lookup/revalid
+                        * -- I am the owner of dir statahead. */
+                       pid_t                           d_opendir_pid;
+               } d;
 
-#define lli_readdir_sem         u.d.d_readdir_sem
+#define lli_readdir_mutex       u.d.d_readdir_mutex
 #define lli_opendir_key         u.d.d_opendir_key
 #define lli_sai                 u.d.d_sai
-#define lli_sa_pos              u.d.d_sa_pos
 #define lli_def_acl             u.d.d_def_acl
 #define lli_sa_lock             u.d.d_sa_lock
 #define lli_opendir_pid         u.d.d_opendir_pid
 
-                /* for non-directory */
-                struct {
-                        cfs_semaphore_t                 f_size_sem;
-                        void                           *f_size_sem_owner;
-                        char                           *f_symlink_name;
-                        __u64                           f_maxbytes;
-                        /*
-                         * cfs_rw_semaphore_t {
-                         *    signed long      count;     // align u.d.d_def_acl
-                         *    cfs_spinlock_t   wait_lock; // align u.d.d_sa_lock
-                         *    struct list_head wait_list;
-                         * }
-                         */
-                        cfs_rw_semaphore_t              f_trunc_sem;
-                        cfs_semaphore_t                 f_write_sem;
-
-                        /* for writepage() only to communicate to fsync */
-                        int                             f_async_rc;
-                        int                             f_write_rc;
+               /* for non-directory */
+               struct {
+                       struct semaphore                f_size_sem;
+                       void                            *f_size_sem_owner;
+                       char                            *f_symlink_name;
+                       __u64                           f_maxbytes;
+                       /*
+                        * struct rw_semaphore {
+                        *    signed long       count;     // align d.d_def_acl
+                        *    spinlock_t        wait_lock; // align d.d_sa_lock
+                        *    struct list_head wait_list;
+                        * }
+                        */
+                       struct rw_semaphore             f_trunc_sem;
+                       struct mutex                    f_write_mutex;
+
+                       struct rw_semaphore             f_glimpse_sem;
+                       cfs_time_t                      f_glimpse_time;
+                       cfs_list_t                      f_agl_list;
+                       __u64                           f_agl_index;
+
+                       /* for writepage() only to communicate to fsync */
+                       int                             f_async_rc;
+
+                       /*
+                        * whenever a process try to read/write the file, the
+                        * jobid of the process will be saved here, and it'll
+                        * be packed into the write PRC when flush later.
+                        *
+                        * so the read/write statistics for jobid will not be
+                        * accurate if the file is shared by different jobs.
+                        */
+                       char                     f_jobid[JOBSTATS_JOBID_SIZE];
                 } f;
 
 #define lli_size_sem            u.f.f_size_sem
@@ -231,11 +244,15 @@ struct ll_inode_info {
 #define lli_symlink_name        u.f.f_symlink_name
 #define lli_maxbytes            u.f.f_maxbytes
 #define lli_trunc_sem           u.f.f_trunc_sem
-#define lli_write_sem           u.f.f_write_sem
-#define lli_async_rc            u.f.f_async_rc
-#define lli_write_rc            u.f.f_write_rc
+#define lli_write_mutex         u.f.f_write_mutex
+#define lli_glimpse_sem        u.f.f_glimpse_sem
+#define lli_glimpse_time       u.f.f_glimpse_time
+#define lli_agl_list           u.f.f_agl_list
+#define lli_agl_index          u.f.f_agl_index
+#define lli_async_rc           u.f.f_async_rc
+#define lli_jobid              u.f.f_jobid
 
-        } u;
+       } u;
 
         /* XXX: For following frequent used members, although they maybe special
          *      used for non-directory object, it is some time-wasting to check
@@ -247,20 +264,24 @@ struct ll_inode_info {
          *      In the future, if more members are added only for directory,
          *      some of the following members can be moved into u.f.
          */
-        struct lov_stripe_md           *lli_smd;
-        struct cl_object               *lli_clob;
+       bool                            lli_has_smd;
+       struct cl_object               *lli_clob;
+
+       /* mutex to request for layout lock exclusively. */
+       struct mutex                    lli_layout_mutex;
+       /* valid only inside LAYOUT ibits lock, protected by lli_layout_mutex */
+       __u32                           lli_layout_gen;
 };
 
 /*
  * Locking to guarantee consistency of non-atomic updates to long long i_size,
- * consistency between file size and KMS, and consistency within
- * ->lli_smd->lsm_oinfo[]'s.
+ * consistency between file size and KMS.
  *
- * Implemented by ->lli_size_sem and ->lsm_sem, nested in that order.
+ * Implemented by ->lli_size_sem and ->lsm_lock, nested in that order.
  */
 
-void ll_inode_size_lock(struct inode *inode, int lock_lsm);
-void ll_inode_size_unlock(struct inode *inode, int unlock_lsm);
+void ll_inode_size_lock(struct inode *inode);
+void ll_inode_size_unlock(struct inode *inode);
 
 // FIXME: replace the name of this with LL_I to conform to kernel stuff
 // static inline struct ll_inode_info *LL_I(struct inode *inode)
@@ -369,6 +390,10 @@ enum stats_track_type {
 #define LL_SBI_SOM_PREVIEW     0x1000 /* SOM preview mount option */
 #define LL_SBI_32BIT_API       0x2000 /* generate 32 bit inodes. */
 #define LL_SBI_64BIT_HASH      0x4000 /* support 64-bits dir hash/offset */
+#define LL_SBI_AGL_ENABLED     0x8000 /* enable agl */
+#define LL_SBI_VERBOSE        0x10000 /* verbose mount/umount */
+#define LL_SBI_LAYOUT_LOCK    0x20000 /* layout lock support */
+#define LL_SBI_USER_FID2PATH  0x40000 /* allow fid2path by unprivileged users */
 
 /* default value for ll_sb_info->contention_time */
 #define SBI_DEFAULT_CONTENTION_SECONDS     60
@@ -383,8 +408,8 @@ struct rmtacl_ctl_entry {
 };
 
 struct rmtacl_ctl_table {
-        cfs_spinlock_t   rct_lock;
-        cfs_list_t       rct_entries[RCE_HASHES];
+       spinlock_t      rct_lock;
+       cfs_list_t      rct_entries[RCE_HASHES];
 };
 
 #define EE_HASHES       32
@@ -398,17 +423,17 @@ struct eacl_entry {
 };
 
 struct eacl_table {
-        cfs_spinlock_t   et_lock;
-        cfs_list_t       et_entries[EE_HASHES];
+       spinlock_t      et_lock;
+       cfs_list_t      et_entries[EE_HASHES];
 };
 
 struct ll_sb_info {
-        cfs_list_t                ll_list;
-        /* this protects pglist and ra_info.  It isn't safe to
-         * grab from interrupt contexts */
-        cfs_spinlock_t            ll_lock;
-        cfs_spinlock_t            ll_pp_extent_lock; /* Lock for pp_extent entries */
-        cfs_spinlock_t            ll_process_lock; /* Lock for ll_rw_process_info */
+       cfs_list_t                ll_list;
+       /* this protects pglist and ra_info.  It isn't safe to
+        * grab from interrupt contexts */
+       spinlock_t                ll_lock;
+       spinlock_t                ll_pp_extent_lock; /* pp_extent entry*/
+       spinlock_t                ll_process_lock; /* ll_rw_process_info */
         struct obd_uuid           ll_sb_uuid;
         struct obd_export        *ll_md_exp;
         struct obd_export        *ll_dt_exp;
@@ -424,8 +449,7 @@ struct ll_sb_info {
 
         struct lprocfs_stats     *ll_stats; /* lprocfs stats counter */
 
-        unsigned long             ll_async_page_max;
-        unsigned long             ll_async_page_count;
+       struct cl_client_cache    ll_cache;
 
         struct lprocfs_stats     *ll_ra_stats;
 
@@ -447,8 +471,8 @@ struct ll_sb_info {
         unsigned int              ll_offset_process_count;
         struct ll_rw_process_info ll_rw_offset_info[LL_OFFSET_HIST_MAX];
         unsigned int              ll_rw_offset_entry_count;
-        enum stats_track_type     ll_stats_track_type;
         int                       ll_stats_track_id;
+        enum stats_track_type     ll_stats_track_type;
         int                       ll_rw_stats_on;
 
         /* metadata stat-ahead */
@@ -457,12 +481,12 @@ struct ll_sb_info {
                                                   * count */
         atomic_t                  ll_sa_wrong;   /* statahead thread stopped for
                                                   * low hit ratio */
+        atomic_t                  ll_agl_total;  /* AGL thread started count */
 
         dev_t                     ll_sdev_orig; /* save s_dev before assign for
                                                  * clustred nfs */
         struct rmtacl_ctl_table   ll_rct;
         struct eacl_table         ll_et;
-        struct vfsmount          *ll_mnt;
 };
 
 #define LL_DEFAULT_MAX_RW_CHUNK      (32 * 1024 * 1024)
@@ -478,7 +502,7 @@ struct ll_ra_read {
  * per file-descriptor read-ahead data.
  */
 struct ll_readahead_state {
-        cfs_spinlock_t  ras_lock;
+       spinlock_t  ras_lock;
         /*
          * index of the last page that read(2) needed and that wasn't in the
          * cache. Used by ras_update() to detect seeks.
@@ -560,25 +584,24 @@ struct ll_readahead_state {
         unsigned long   ras_consecutive_stride_requests;
 };
 
-struct ll_file_dir {
-        __u64 lfd_pos;
-        __u64 lfd_next;
-};
-
 extern cfs_mem_cache_t *ll_file_data_slab;
 struct lustre_handle;
 struct ll_file_data {
         struct ll_readahead_state fd_ras;
         int fd_omode;
         struct ccc_grouplock fd_grouplock;
-        struct ll_file_dir fd_dir;
+       __u64 lfd_pos;
         __u32 fd_flags;
         struct file *fd_file;
+       /* Indicate whether need to report failure when close.
+        * true: failure is known, not report again.
+        * false: unknown failure, should report. */
+       bool fd_write_failed;
 };
 
 struct lov_stripe_md;
 
-extern cfs_spinlock_t inode_lock;
+extern spinlock_t inode_lock;
 
 extern struct proc_dir_entry *proc_lustre_fs_root;
 
@@ -637,9 +660,10 @@ static void lprocfs_llite_init_vars(struct lprocfs_static_vars *lvars)
 void ll_release_page(struct page *page, int remove);
 extern struct file_operations ll_dir_operations;
 extern struct inode_operations ll_dir_inode_operations;
-struct page *ll_get_dir_page(struct file *filp, struct inode *dir, __u64 hash,
+struct page *ll_get_dir_page(struct inode *dir, __u64 hash,
                              struct ll_dir_chain *chain);
-int ll_readdir(struct file *filp, void *cookie, filldir_t filldir);
+int ll_dir_read(struct inode *inode, __u64 *_pos, void *cookie,
+               filldir_t filldir);
 
 int ll_get_mdt_idx(struct inode *inode);
 /* llite/namei.c */
@@ -653,16 +677,16 @@ struct lookup_intent *ll_convert_intent(struct open_intent *oit,
                                         int lookup_flags);
 int ll_lookup_it_finish(struct ptlrpc_request *request,
                         struct lookup_intent *it, void *data);
-struct dentry *ll_find_alias(struct inode *inode, struct dentry *de);
+struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de);
 
 /* llite/rw.c */
 int ll_prepare_write(struct file *, struct page *, unsigned from, unsigned to);
 int ll_commit_write(struct file *, struct page *, unsigned from, unsigned to);
 int ll_writepage(struct page *page, struct writeback_control *wbc);
+int ll_writepages(struct address_space *, struct writeback_control *wbc);
 void ll_removepage(struct page *page);
 int ll_readpage(struct file *file, struct page *page);
 void ll_readahead_init(struct inode *inode, struct ll_readahead_state *ras);
-void ll_truncate(struct inode *inode);
 int ll_file_punch(struct inode *, loff_t, int);
 ssize_t ll_file_lockless_io(struct file *, char *, size_t, loff_t *, int);
 void ll_clear_file_contended(struct inode*);
@@ -681,7 +705,7 @@ extern int ll_inode_revalidate_it(struct dentry *, struct lookup_intent *,
 extern int ll_have_md_lock(struct inode *inode, __u64 *bits,
                            ldlm_mode_t l_req_mode);
 extern ldlm_mode_t ll_take_md_lock(struct inode *inode, __u64 bits,
-                                   struct lustre_handle *lockh);
+                                   struct lustre_handle *lockh, __u64 flags);
 int __ll_inode_revalidate_it(struct dentry *, struct lookup_intent *,
                              __u64 bits);
 int ll_revalidate_nd(struct dentry *dentry, struct nameidata *nd);
@@ -703,7 +727,7 @@ void ll_done_writing_attr(struct inode *inode, struct md_op_data *op_data);
 int ll_som_update(struct inode *inode, struct md_op_data *op_data);
 int ll_inode_getattr(struct inode *inode, struct obdo *obdo,
                      __u64 ioepoch, int sync);
-int ll_md_setattr(struct inode *inode, struct md_op_data *op_data,
+int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data,
                   struct md_open_data **mod);
 void ll_pack_inode2opdata(struct inode *inode, struct md_op_data *op_data,
                           struct lustre_handle *fh);
@@ -714,6 +738,8 @@ int ll_getattr_it(struct vfsmount *mnt, struct dentry *de,
                struct lookup_intent *it, struct kstat *stat);
 int ll_getattr(struct vfsmount *mnt, struct dentry *de, struct kstat *stat);
 struct ll_file_data *ll_file_data_get(void);
+struct posix_acl * ll_get_acl(struct inode *inode, int type);
+
 #ifdef HAVE_GENERIC_PERMISSION_4ARGS
 int ll_inode_permission(struct inode *inode, int mask, unsigned int flags);
 #else
@@ -723,6 +749,7 @@ int ll_inode_permission(struct inode *inode, int mask, struct nameidata *nd);
 int ll_inode_permission(struct inode *inode, int mask);
 # endif
 #endif
+
 int ll_lov_setstripe_ea_info(struct inode *inode, struct file *file,
                              int flags, struct lov_user_md *lum,
                              int lum_size);
@@ -733,25 +760,36 @@ int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
                      int set_default);
 int ll_dir_getstripe(struct inode *inode, struct lov_mds_md **lmm,
                      int *lmm_size, struct ptlrpc_request **request);
+#ifdef HAVE_FILE_FSYNC_4ARGS
+int ll_fsync(struct file *file, loff_t start, loff_t end, int data);
+#elif defined(HAVE_FILE_FSYNC_2ARGS)
+int ll_fsync(struct file *file, int data);
+#else
 int ll_fsync(struct file *file, struct dentry *dentry, int data);
+#endif
 int ll_do_fiemap(struct inode *inode, struct ll_user_fiemap *fiemap,
               int num_bytes);
 int ll_merge_lvb(struct inode *inode);
 int ll_get_grouplock(struct inode *inode, struct file *file, unsigned long arg);
 int ll_put_grouplock(struct inode *inode, struct file *file, unsigned long arg);
-int ll_fid2path(struct obd_export *exp, void *arg);
+int ll_fid2path(struct inode *inode, void *arg);
 
 /* llite/dcache.c */
+
 int ll_dops_init(struct dentry *de, int block, int init_sa);
-extern cfs_spinlock_t ll_lookup_lock;
 extern struct dentry_operations ll_d_ops;
 void ll_intent_drop_lock(struct lookup_intent *);
 void ll_intent_release(struct lookup_intent *);
-int ll_drop_dentry(struct dentry *dentry);
-void ll_unhash_aliases(struct inode *);
+void ll_invalidate_aliases(struct inode *);
 void ll_frob_intent(struct lookup_intent **itp, struct lookup_intent *deft);
 void ll_lookup_finish_locks(struct lookup_intent *it, struct dentry *dentry);
+#ifdef HAVE_D_COMPARE_7ARGS
+int ll_dcompare(const struct dentry *parent, const struct inode *pinode,
+               const struct dentry *dentry, const struct inode *inode,
+               unsigned int len, const char *str, const struct qstr *d_name);
+#else
 int ll_dcompare(struct dentry *parent, struct qstr *d_name, struct qstr *name);
+#endif
 int ll_revalidate_it_finish(struct ptlrpc_request *request,
                             struct lookup_intent *it, struct dentry *de);
 
@@ -765,13 +803,9 @@ void ll_put_super(struct super_block *sb);
 void ll_kill_super(struct super_block *sb);
 struct inode *ll_inode_from_lock(struct ldlm_lock *lock);
 void ll_clear_inode(struct inode *inode);
-int ll_setattr_raw(struct inode *inode, struct iattr *attr);
+int ll_setattr_raw(struct dentry *dentry, struct iattr *attr);
 int ll_setattr(struct dentry *de, struct iattr *attr);
-#ifndef HAVE_STATFS_DENTRY_PARAM
-int ll_statfs(struct super_block *sb, struct kstatfs *sfs);
-#else
 int ll_statfs(struct dentry *de, struct kstatfs *sfs);
-#endif
 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
                        __u64 max_age, __u32 flags);
 void ll_update_inode(struct inode *inode, struct lustre_md *md);
@@ -786,9 +820,14 @@ void ll_umount_begin(struct vfsmount *vfsmnt, int flags);
 void ll_umount_begin(struct super_block *sb);
 #endif
 int ll_remount_fs(struct super_block *sb, int *flags, char *data);
+#ifdef HAVE_SUPEROPS_USE_DENTRY
+int ll_show_options(struct seq_file *seq, struct dentry *dentry);
+#else
 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs);
+#endif
+void ll_dirty_page_discard_warn(cfs_page_t *page, int ioret);
 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
-                  struct super_block *);
+                 struct super_block *);
 void lustre_dump_dentry(struct dentry *, int recur);
 void lustre_dump_inode(struct inode *);
 int ll_obd_statfs(struct inode *inode, void *arg);
@@ -799,6 +838,8 @@ struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
                                       const char *name, int namelen,
                                       int mode, __u32 opc, void *data);
 void ll_finish_md_op_data(struct md_op_data *op_data);
+int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg);
+char *ll_get_fsname(struct super_block *sb, char *buf, int buflen);
 
 /* llite/llite_nfs.c */
 extern struct export_operations lustre_export_operations;
@@ -818,11 +859,11 @@ extern struct inode_operations ll_fast_symlink_inode_operations;
 
 /* llite/llite_close.c */
 struct ll_close_queue {
-        cfs_spinlock_t          lcq_lock;
-        cfs_list_t              lcq_head;
-        cfs_waitq_t             lcq_waitq;
-        cfs_completion_t        lcq_comp;
-        cfs_atomic_t            lcq_stop;
+       spinlock_t              lcq_lock;
+       cfs_list_t              lcq_head;
+       cfs_waitq_t             lcq_waitq;
+       struct completion       lcq_comp;
+       cfs_atomic_t            lcq_stop;
 };
 
 struct ccc_object *cl_inode2ccc(struct inode *inode);
@@ -1012,7 +1053,8 @@ struct ll_lock_tree_node * ll_node_from_inode(struct inode *inode, __u64 start,
                                               __u64 end, ldlm_mode_t mode);
 void policy_from_vma(ldlm_policy_data_t *policy,
                 struct vm_area_struct *vma, unsigned long addr, size_t count);
-struct vm_area_struct *our_vma(unsigned long addr, size_t count);
+struct vm_area_struct *our_vma(struct mm_struct *mm, unsigned long addr,
+                               size_t count);
 
 static inline void ll_invalidate_page(struct page *vmpage)
 {
@@ -1068,9 +1110,10 @@ static inline struct obd_export *ll_i2mdexp(struct inode *inode)
 static inline struct lu_fid *ll_inode2fid(struct inode *inode)
 {
         struct lu_fid *fid;
+
         LASSERT(inode != NULL);
         fid = &ll_i2info(inode)->lli_fid;
-        LASSERT(fid_is_igif(fid) || fid_ver(fid) == 0);
+
         return fid;
 }
 
@@ -1196,21 +1239,36 @@ struct ll_statahead_info {
         unsigned int            sai_skip_hidden;/* skipped hidden dentry count */
         unsigned int            sai_ls_all:1,   /* "ls -al", do stat-ahead for
                                                  * hidden entries */
-                                sai_in_readpage:1;/* statahead is in readdir()*/
+                                sai_in_readpage:1,/* statahead is in readdir()*/
+                                sai_agl_valid:1;/* AGL is valid for the dir */
         cfs_waitq_t             sai_waitq;      /* stat-ahead wait queue */
         struct ptlrpc_thread    sai_thread;     /* stat-ahead thread */
+        struct ptlrpc_thread    sai_agl_thread; /* AGL thread */
         cfs_list_t              sai_entries_sent;     /* entries sent out */
         cfs_list_t              sai_entries_received; /* entries returned */
         cfs_list_t              sai_entries_stated;   /* entries stated */
+        cfs_list_t              sai_entries_agl; /* AGL entries to be sent */
         cfs_list_t              sai_cache[LL_SA_CACHE_SIZE];
-        cfs_spinlock_t          sai_cache_lock[LL_SA_CACHE_SIZE];
-        cfs_atomic_t            sai_cache_count;      /* entry count in cache */
+       spinlock_t              sai_cache_lock[LL_SA_CACHE_SIZE];
+       cfs_atomic_t            sai_cache_count; /* entry count in cache */
 };
 
 int do_statahead_enter(struct inode *dir, struct dentry **dentry,
                        int only_unplug);
 void ll_stop_statahead(struct inode *dir, void *key);
 
+static inline int ll_glimpse_size(struct inode *inode)
+{
+       struct ll_inode_info *lli = ll_i2info(inode);
+       int rc;
+
+       down_read(&lli->lli_glimpse_sem);
+       rc = cl_glimpse_size(inode);
+       lli->lli_glimpse_time = cfs_time_current();
+       up_read(&lli->lli_glimpse_sem);
+       return rc;
+}
+
 static inline void
 ll_statahead_mark(struct inode *dir, struct dentry *dentry)
 {
@@ -1233,13 +1291,17 @@ ll_statahead_enter(struct inode *dir, struct dentry **dentryp, int only_unplug)
         struct ll_dentry_data *ldd;
 
         if (ll_i2sbi(dir)->ll_sa_max == 0)
-                return -ENOTSUPP;
+                return -EAGAIN;
 
         lli = ll_i2info(dir);
         /* not the same process, don't statahead */
         if (lli->lli_opendir_pid != cfs_curproc_pid())
                 return -EAGAIN;
 
+       /* statahead has been stopped */
+       if (lli->lli_opendir_key == NULL)
+               return -EAGAIN;
+
         ldd = ll_d2d(*dentryp);
         /*
          * When stats a dentry, the system trigger more than once "revalidate"
@@ -1316,14 +1378,21 @@ void ll_iocontrol_unregister(void *magic);
 #define cl_inode_mode(inode) ((inode)->i_mode)
 #define cl_i2sbi ll_i2sbi
 
-static inline void cl_isize_lock(struct inode *inode, int lsmlock)
+static inline struct ll_file_data *cl_iattr2fd(struct inode *inode,
+                                               const struct iattr *attr)
 {
-        ll_inode_size_lock(inode, lsmlock);
+        LASSERT(attr->ia_valid & ATTR_FILE);
+        return LUSTRE_FPRIVATE(attr->ia_file);
 }
 
-static inline void cl_isize_unlock(struct inode *inode, int lsmlock)
+static inline void cl_isize_lock(struct inode *inode)
 {
-        ll_inode_size_unlock(inode, lsmlock);
+       ll_inode_size_lock(inode);
+}
+
+static inline void cl_isize_unlock(struct inode *inode)
+{
+       ll_inode_size_unlock(inode);
 }
 
 static inline void cl_isize_write_nolock(struct inode *inode, loff_t kms)
@@ -1334,9 +1403,9 @@ static inline void cl_isize_write_nolock(struct inode *inode, loff_t kms)
 
 static inline void cl_isize_write(struct inode *inode, loff_t kms)
 {
-        ll_inode_size_lock(inode, 0);
-        i_size_write(inode, kms);
-        ll_inode_size_unlock(inode, 0);
+       ll_inode_size_lock(inode);
+       i_size_write(inode, kms);
+       ll_inode_size_unlock(inode);
 }
 
 #define cl_isize_read(inode)             i_size_read(inode)
@@ -1352,6 +1421,9 @@ static inline int cl_merge_lvb(struct inode *inode)
 
 struct obd_capa *cl_capa_lookup(struct inode *inode, enum cl_req_type crt);
 
+int cl_sync_file_range(struct inode *inode, loff_t start, loff_t end,
+                      enum cl_fsync_mode mode);
+
 /** direct write pages */
 struct ll_dio_pages {
         /** page array to be written. we don't support
@@ -1407,31 +1479,66 @@ static inline void ll_set_lock_data(struct obd_export *exp, struct inode *inode,
                 *bits = it->d.lustre.it_lock_bits;
 }
 
-static inline void ll_dentry_rehash(struct dentry *dentry, int locked)
+static inline void ll_lock_dcache(struct inode *inode)
 {
-        if (!locked) {
-                cfs_spin_lock(&ll_lookup_lock);
-                spin_lock(&dcache_lock);
-        }
-        if (d_unhashed(dentry))
-                d_rehash_cond(dentry, 0);
-        if (!locked) {
-                spin_unlock(&dcache_lock);
-                cfs_spin_unlock(&ll_lookup_lock);
-        }
+#ifdef HAVE_DCACHE_LOCK
+       spin_lock(&dcache_lock);
+#else
+       spin_lock(&inode->i_lock);
+#endif
 }
 
-static inline void ll_dentry_reset_flags(struct dentry *dentry, __u64 bits)
+static inline void ll_unlock_dcache(struct inode *inode)
 {
-        if (bits & MDS_INODELOCK_LOOKUP &&
-            dentry->d_flags & DCACHE_LUSTRE_INVALID) {
-                lock_dentry(dentry);
-                dentry->d_flags &= ~DCACHE_LUSTRE_INVALID;
-                unlock_dentry(dentry);
-        }
+#ifdef HAVE_DCACHE_LOCK
+       spin_unlock(&dcache_lock);
+#else
+       spin_unlock(&inode->i_lock);
+#endif
 }
 
-#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2,7,50,0)
+static inline int d_lustre_invalid(const struct dentry *dentry)
+{
+       struct ll_dentry_data *lld = ll_d2d(dentry);
+
+       return (lld == NULL) || lld->lld_invalid;
+}
+
+static inline void __d_lustre_invalidate(struct dentry *dentry)
+{
+       struct ll_dentry_data *lld = ll_d2d(dentry);
+
+       if (lld != NULL)
+               lld->lld_invalid = 1;
+}
+
+/*
+ * Mark dentry INVALID, if dentry refcount is zero (this is normally case for
+ * ll_md_blocking_ast), unhash this dentry, and let dcache to reclaim it later;
+ * else dput() of the last refcount will unhash this dentry and kill it.
+ */
+static inline void d_lustre_invalidate(struct dentry *dentry)
+{
+       CDEBUG(D_DENTRY, "invalidate dentry %.*s (%p) parent %p inode %p "
+              "refc %d\n", dentry->d_name.len, dentry->d_name.name, dentry,
+              dentry->d_parent, dentry->d_inode, d_refcount(dentry));
+
+       spin_lock(&dentry->d_lock);
+       __d_lustre_invalidate(dentry);
+       if (d_refcount(dentry) == 0)
+               __d_drop(dentry);
+       spin_unlock(&dentry->d_lock);
+}
+
+static inline void d_lustre_revalidate(struct dentry *dentry)
+{
+       spin_lock(&dentry->d_lock);
+       LASSERT(ll_d2d(dentry) != NULL);
+       ll_d2d(dentry)->lld_invalid = 0;
+       spin_unlock(&dentry->d_lock);
+}
+
+#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 50, 0)
 /* Compatibility for old (1.8) compiled userspace quota code */
 struct if_quotactl_18 {
         __u32                   qc_cmd;
@@ -1447,6 +1554,9 @@ struct if_quotactl_18 {
 /* End compatibility for old (1.8) compiled userspace quota code */
 #else
 #warning "remove old LL_IOC_QUOTACTL_18 compatibility code"
-#endif /* LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2,7,50,0) */
+#endif /* LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 50, 0) */
+
+int ll_layout_conf(struct inode *inode, const struct cl_object_conf *conf);
+int ll_layout_refresh(struct inode *inode, __u32 *gen);
 
 #endif /* LLITE_INTERNAL_H */