Whamcloud - gitweb
LU-1030 clio: reimplement ll_fsync in clio way
[fs/lustre-release.git] / lustre / llite / file.c
index cb55e3f..5498217 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) 2002, 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/
@@ -203,15 +200,15 @@ int ll_md_real_close(struct inode *inode, int flags)
                 och_usecount = &lli->lli_open_fd_read_count;
         }
 
-        cfs_down(&lli->lli_och_sem);
+        cfs_mutex_lock(&lli->lli_och_mutex);
         if (*och_usecount) { /* There are still users of this handle, so
                                 skip freeing it. */
-                cfs_up(&lli->lli_och_sem);
+                cfs_mutex_unlock(&lli->lli_och_mutex);
                 RETURN(0);
         }
         och=*och_p;
         *och_p = NULL;
-        cfs_up(&lli->lli_och_sem);
+        cfs_mutex_unlock(&lli->lli_och_mutex);
 
         if (och) { /* There might be a race and somebody have freed this och
                       already */
@@ -243,7 +240,7 @@ int ll_md_close(struct obd_export *md_exp, struct inode *inode,
                 struct inode *inode = file->f_dentry->d_inode;
                 ldlm_policy_data_t policy = {.l_inodebits={MDS_INODELOCK_OPEN}};
 
-                cfs_down(&lli->lli_och_sem);
+                cfs_mutex_lock(&lli->lli_och_mutex);
                 if (fd->fd_omode & FMODE_WRITE) {
                         lockmode = LCK_CW;
                         LASSERT(lli->lli_open_fd_write_count);
@@ -257,7 +254,7 @@ int ll_md_close(struct obd_export *md_exp, struct inode *inode,
                         LASSERT(lli->lli_open_fd_read_count);
                         lli->lli_open_fd_read_count--;
                 }
-                cfs_up(&lli->lli_och_sem);
+                cfs_mutex_unlock(&lli->lli_och_mutex);
 
                 if (!md_lock_match(md_exp, flags, ll_inode2fid(inode),
                                    LDLM_IBITS, &policy, lockmode,
@@ -318,7 +315,8 @@ int ll_file_release(struct inode *inode, struct file *file)
         /* The last ref on @file, maybe not the the owner pid of statahead.
          * Different processes can open the same dir, "ll_opendir_key" means:
          * it is me that should stop the statahead thread. */
-        if (lli->lli_opendir_key == fd && lli->lli_opendir_pid != 0)
+        if (S_ISDIR(inode->i_mode) && lli->lli_opendir_key == fd &&
+            lli->lli_opendir_pid != 0)
                 ll_stop_statahead(inode, lli->lli_opendir_key);
 
         if (inode->i_sb->s_root == file->f_dentry) {
@@ -327,9 +325,11 @@ int ll_file_release(struct inode *inode, struct file *file)
                 RETURN(0);
         }
 
-        if (lsm)
-                lov_test_and_clear_async_rc(lsm);
-        lli->lli_async_rc = 0;
+        if (!S_ISDIR(inode->i_mode)) {
+                if (lsm)
+                        lov_test_and_clear_async_rc(lsm);
+                lli->lli_async_rc = 0;
+        }
 
         rc = ll_md_close(sbi->ll_md_exp, inode, file);
 
@@ -481,9 +481,7 @@ int ll_local_open(struct file *file, struct lookup_intent *it,
 
 /* Open a file, and (for the very first open) create objects on the OSTs at
  * this time.  If opened with O_LOV_DELAY_CREATE, then we don't do the object
- * creation or open until ll_lov_setstripe() ioctl is called.  We grab
- * lli_open_sem to ensure no other process will create objects, send the
- * stripe MD to the MDS, or try to destroy the objects if that fails.
+ * creation or open until ll_lov_setstripe() ioctl is called.
  *
  * If we already have the stripe MD locally then we don't request it in
  * md_open(), by passing a lmm_size = 0.
@@ -519,8 +517,8 @@ int ll_file_open(struct inode *inode, struct file *file)
         fd->fd_file = file;
         if (S_ISDIR(inode->i_mode)) {
                 cfs_spin_lock(&lli->lli_sa_lock);
-                if (lli->lli_opendir_key == NULL && lli->lli_opendir_pid == 0) {
-                        LASSERT(lli->lli_sai == NULL);
+                if (lli->lli_opendir_key == NULL && lli->lli_sai == NULL &&
+                    lli->lli_opendir_pid == 0) {
                         lli->lli_opendir_key = fd;
                         lli->lli_opendir_pid = cfs_curproc_pid();
                         opendir_set = 1;
@@ -575,27 +573,25 @@ restart:
                 och_usecount = &lli->lli_open_fd_read_count;
         }
 
-        cfs_down(&lli->lli_och_sem);
+        cfs_mutex_lock(&lli->lli_och_mutex);
         if (*och_p) { /* Open handle is present */
                 if (it_disposition(it, DISP_OPEN_OPEN)) {
                         /* Well, there's extra open request that we do not need,
                            let's close it somehow. This will decref request. */
                         rc = it_open_error(DISP_OPEN_OPEN, it);
                         if (rc) {
-                                cfs_up(&lli->lli_och_sem);
+                                cfs_mutex_unlock(&lli->lli_och_mutex);
                                 GOTO(out_openerr, rc);
                         }
 
                         ll_release_openhandle(file->f_dentry, it);
-                        lprocfs_counter_incr(ll_i2sbi(inode)->ll_stats,
-                                             LPROC_LL_OPEN);
                 }
                 (*och_usecount)++;
 
                 rc = ll_local_open(file, it, fd, NULL);
                 if (rc) {
                         (*och_usecount)--;
-                        cfs_up(&lli->lli_och_sem);
+                        cfs_mutex_unlock(&lli->lli_och_mutex);
                         GOTO(out_openerr, rc);
                 }
         } else {
@@ -604,9 +600,9 @@ restart:
                         /* We cannot just request lock handle now, new ELC code
                            means that one of other OPEN locks for this file
                            could be cancelled, and since blocking ast handler
-                           would attempt to grab och_sem as well, that would
+                           would attempt to grab och_mutex as well, that would
                            result in a deadlock */
-                        cfs_up(&lli->lli_och_sem);
+                        cfs_mutex_unlock(&lli->lli_och_mutex);
                         it->it_create_mode |= M_CHECK_STALE;
                         rc = ll_intent_file_open(file, NULL, 0, it);
                         it->it_create_mode &= ~M_CHECK_STALE;
@@ -632,15 +628,14 @@ restart:
 
                 LASSERT(it_disposition(it, DISP_ENQ_OPEN_REF));
 
-                ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_OPEN, 1);
                 rc = ll_local_open(file, it, fd, *och_p);
                 if (rc)
                         GOTO(out_och_free, rc);
         }
-        cfs_up(&lli->lli_och_sem);
+        cfs_mutex_unlock(&lli->lli_och_mutex);
         fd = NULL;
 
-        /* Must do this outside lli_och_sem lock to prevent deadlock where
+        /* Must do this outside lli_och_mutex lock to prevent deadlock where
            different kind of OPEN lock for this same inode gets cancelled
            by ldlm_cancel_lru */
         if (!S_ISREG(inode->i_mode))
@@ -671,13 +666,15 @@ out_och_free:
                         *och_p = NULL; /* OBD_FREE writes some magic there */
                         (*och_usecount)--;
                 }
-                cfs_up(&lli->lli_och_sem);
+                cfs_mutex_unlock(&lli->lli_och_mutex);
 
 out_openerr:
                 if (opendir_set != 0)
                         ll_stop_statahead(inode, lli->lli_opendir_key);
                 if (fd != NULL)
                         ll_file_data_put(fd);
+        } else {
+                ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_OPEN, 1);
         }
 
         return rc;
@@ -706,7 +703,8 @@ static int ll_lsm_getattr(struct lov_stripe_md *lsm, struct obd_export *exp,
                                OBD_MD_FLSIZE | OBD_MD_FLBLOCKS |
                                OBD_MD_FLBLKSZ | OBD_MD_FLATIME |
                                OBD_MD_FLMTIME | OBD_MD_FLCTIME |
-                               OBD_MD_FLGROUP | OBD_MD_FLEPOCH;
+                               OBD_MD_FLGROUP | OBD_MD_FLEPOCH |
+                               OBD_MD_FLDATAVERSION;
         oinfo.oi_capa = capa;
         if (sync) {
                 oinfo.oi_oa->o_valid |= OBD_MD_FLFLAGS;
@@ -726,7 +724,8 @@ static int ll_lsm_getattr(struct lov_stripe_md *lsm, struct obd_export *exp,
         if (rc == 0)
                 oinfo.oi_oa->o_valid &= (OBD_MD_FLBLOCKS | OBD_MD_FLBLKSZ |
                                          OBD_MD_FLATIME | OBD_MD_FLMTIME |
-                                         OBD_MD_FLCTIME | OBD_MD_FLSIZE);
+                                         OBD_MD_FLCTIME | OBD_MD_FLSIZE |
+                                         OBD_MD_FLDATAVERSION);
         RETURN(rc);
 }
 
@@ -748,7 +747,7 @@ int ll_inode_getattr(struct inode *inode, struct obdo *obdo,
         if (rc == 0) {
                 obdo_refresh_inode(inode, obdo, obdo->o_valid);
                 CDEBUG(D_INODE,
-                       "objid "LPX64" size %Lu, blocks %llu, blksize %lu\n",
+                       "objid "LPX64" size %llu, blocks %llu, blksize %lu\n",
                        lli->lli_smd->lsm_object_id, i_size_read(inode),
                        (unsigned long long)inode->i_blocks,
                        (unsigned long)ll_inode_blksize(inode));
@@ -837,7 +836,7 @@ static ssize_t ll_file_io_generic(const struct lu_env *env,
         if (cl_io_rw_init(env, io, iot, *ppos, count) == 0) {
                 struct vvp_io *vio = vvp_env_io(env);
                 struct ccc_io *cio = ccc_env_io(env);
-                int write_sem_locked = 0;
+                int write_mutex_locked = 0;
 
                 cio->cui_fd  = LUSTRE_FPRIVATE(file);
                 vio->cui_io_subtype = args->via_io_subtype;
@@ -852,9 +851,10 @@ static ssize_t ll_file_io_generic(const struct lu_env *env,
 #endif
                         if ((iot == CIT_WRITE) &&
                             !(cio->cui_fd->fd_flags & LL_FILE_GROUP_LOCKED)) {
-                                if(cfs_down_interruptible(&lli->lli_write_sem))
+                                if (cfs_mutex_lock_interruptible(&lli->
+                                                               lli_write_mutex))
                                         GOTO(out, result = -ERESTARTSYS);
-                                write_sem_locked = 1;
+                                write_mutex_locked = 1;
                         } else if (iot == CIT_READ) {
                                 cfs_down_read(&lli->lli_trunc_sem);
                         }
@@ -872,8 +872,8 @@ static ssize_t ll_file_io_generic(const struct lu_env *env,
                         LBUG();
                 }
                 result = cl_io_loop(env, io);
-                if (write_sem_locked)
-                        cfs_up(&lli->lli_write_sem);
+                if (write_mutex_locked)
+                        cfs_mutex_unlock(&lli->lli_write_mutex);
                 else if (args->via_io_subtype == IO_NORMAL && iot == CIT_READ)
                         cfs_up_read(&lli->lli_trunc_sem);
         } else {
@@ -1247,10 +1247,11 @@ static int ll_lov_recreate(struct inode *inode, obd_id id, obd_seq seq,
         oa->o_nlink = ost_idx;
         oa->o_flags |= OBD_FL_RECREATE_OBJS;
         oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS | OBD_MD_FLGROUP;
-        obdo_from_inode(oa, inode, &ll_i2info(inode)->lli_fid, OBD_MD_FLTYPE |
-                        OBD_MD_FLATIME | OBD_MD_FLMTIME | OBD_MD_FLCTIME);
+        obdo_from_inode(oa, inode, OBD_MD_FLTYPE | OBD_MD_FLATIME |
+                                   OBD_MD_FLMTIME | OBD_MD_FLCTIME);
+        obdo_set_parent_fid(oa, &ll_i2info(inode)->lli_fid);
         memcpy(lsm2, lsm, lsm_size);
-        rc = obd_create(exp, oa, &lsm2, &oti);
+        rc = obd_create(NULL, exp, oa, &lsm2, &oti);
 
         OBD_FREE_LARGE(lsm2, lsm_size);
         GOTO(out, rc);
@@ -1348,8 +1349,8 @@ int ll_lov_getstripe_ea_info(struct inode *inode, const char *filename,
         op_data = ll_prep_md_op_data(NULL, inode, NULL, filename,
                                      strlen(filename), lmmsize,
                                      LUSTRE_OPC_ANY, NULL);
-        if (op_data == NULL)
-                RETURN(-ENOMEM);
+        if (IS_ERR(op_data))
+                RETURN(PTR_ERR(op_data));
 
         op_data->op_valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
         rc = md_getattr_name(sbi->ll_md_exp, op_data, &req);
@@ -1639,8 +1640,8 @@ int ll_do_fiemap(struct inode *inode, struct ll_user_fiemap *fiemap,
         fm_key.oa.o_seq = lsm->lsm_object_seq;
         fm_key.oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
 
-        obdo_from_inode(&fm_key.oa, inode, &ll_i2info(inode)->lli_fid,
-                        OBD_MD_FLSIZE);
+        obdo_from_inode(&fm_key.oa, inode, OBD_MD_FLSIZE);
+        obdo_set_parent_fid(&fm_key.oa, &ll_i2info(inode)->lli_fid);
         /* If filesize is 0, then there would be no objects for mapping */
         if (fm_key.oa.o_size == 0) {
                 fiemap->fm_mapped_extents = 0;
@@ -1649,7 +1650,8 @@ int ll_do_fiemap(struct inode *inode, struct ll_user_fiemap *fiemap,
 
         memcpy(&fm_key.fiemap, fiemap, sizeof(*fiemap));
 
-        rc = obd_get_info(exp, sizeof(fm_key), &fm_key, &vallen, fiemap, lsm);
+        rc = obd_get_info(NULL, exp, sizeof(fm_key), &fm_key, &vallen,
+                          fiemap, lsm);
         if (rc)
                 CERROR("obd_get_info failed: rc = %d\n", rc);
 
@@ -1744,17 +1746,54 @@ error:
         RETURN(rc);
 }
 
-#ifdef HAVE_UNLOCKED_IOCTL
+/*
+ * Read the data_version for inode.
+ *
+ * This value is computed using stripe object version on OST.
+ * Version is computed using server side locking.
+ *
+ * @param extent_lock  Take extent lock. Not needed if a process is already
+ *                     holding the OST object group locks.
+ */
+static int ll_data_version(struct inode *inode, __u64 *data_version,
+                           int extent_lock)
+{
+        struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
+        struct ll_sb_info    *sbi = ll_i2sbi(inode);
+        struct obdo          *obdo = NULL;
+        int                   rc;
+        ENTRY;
+
+        /* If no stripe, we consider version is 0. */
+        if (!lsm) {
+                *data_version = 0;
+                CDEBUG(D_INODE, "No object for inode\n");
+                RETURN(0);
+        }
+
+        OBD_ALLOC_PTR(obdo);
+        if (obdo == NULL)
+                RETURN(-ENOMEM);
+
+        rc = ll_lsm_getattr(lsm, sbi->ll_dt_exp, NULL, obdo, 0, extent_lock);
+        if (!rc) {
+                if (!(obdo->o_valid & OBD_MD_FLDATAVERSION))
+                        rc = -EOPNOTSUPP;
+                else
+                        *data_version = obdo->o_data_version;
+        }
+
+        OBD_FREE_PTR(obdo);
+
+        RETURN(rc);
+}
+
 long ll_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
 {
         struct inode *inode = file->f_dentry->d_inode;
-#else
-int ll_file_ioctl(struct inode *inode, struct file *file, unsigned int cmd,
-                  unsigned long arg)
-{
-#endif
         struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
         int flags;
+
         ENTRY;
 
         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p),cmd=%x\n", inode->i_ino,
@@ -1833,6 +1872,22 @@ int ll_file_ioctl(struct inode *inode, struct file *file, unsigned int cmd,
         }
         case OBD_IOC_FID2PATH:
                 RETURN(ll_fid2path(ll_i2mdexp(inode), (void *)arg));
+        case LL_IOC_DATA_VERSION: {
+                struct ioc_data_version idv;
+                int rc;
+
+                if (cfs_copy_from_user(&idv, (char *)arg, sizeof(idv)))
+                        RETURN(-EFAULT);
+
+                rc = ll_data_version(inode, &idv.idv_version,
+                                     !(idv.idv_flags & LL_DV_NOFLUSH));
+
+                if (rc == 0 &&
+                    cfs_copy_to_user((char *) arg, &idv, sizeof(idv)))
+                        RETURN(-EFAULT);
+
+                RETURN(rc);
+        }
 
         case LL_IOC_GET_MDTIDX: {
                 int mdtidx;
@@ -1846,7 +1901,9 @@ int ll_file_ioctl(struct inode *inode, struct file *file, unsigned int cmd,
 
                 RETURN(0);
         }
-
+        case OBD_IOC_GETDTNAME:
+        case OBD_IOC_GETMDNAME:
+                RETURN(ll_get_obd_name(inode, cmd, arg));
         default: {
                 int err;
 
@@ -1867,18 +1924,15 @@ loff_t ll_file_seek(struct file *file, loff_t offset, int origin)
         ENTRY;
         retval = offset + ((origin == 2) ? i_size_read(inode) :
                            (origin == 1) ? file->f_pos : 0);
-        CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), to=%Lu=%#Lx(%s)\n",
+        CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), to=%llu=%#llx(%s)\n",
                inode->i_ino, inode->i_generation, inode, retval, retval,
                origin == 2 ? "SEEK_END": origin == 1 ? "SEEK_CUR" : "SEEK_SET");
         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_LLSEEK, 1);
 
         if (origin == 2) { /* SEEK_END */
-                int nonblock = 0, rc;
-
-                if (file->f_flags & O_NONBLOCK)
-                        nonblock = LDLM_FL_BLOCK_NOWAIT;
+                int rc;
 
-                rc = cl_glimpse_size(inode);
+                rc = ll_glimpse_size(inode);
                 if (rc != 0)
                         RETURN(rc);
 
@@ -1909,6 +1963,8 @@ int ll_flush(struct file *file)
         struct lov_stripe_md *lsm = lli->lli_smd;
         int rc, err;
 
+        LASSERT(!S_ISDIR(inode->i_mode));
+
         /* the application should know write failure already. */
         if (lli->lli_write_rc)
                 return 0;
@@ -1926,9 +1982,57 @@ int ll_flush(struct file *file)
         return rc ? -EIO : 0;
 }
 
+/**
+ * Called to make sure a portion of file has been written out.
+ * if @local_only is not true, it will send OST_SYNC RPCs to ost.
+ */
+int cl_sync_file_range(struct inode *inode, loff_t start, loff_t end)
+{
+       struct cl_env_nest nest;
+       struct lu_env *env;
+       struct cl_io *io;
+       struct obd_capa *capa = NULL;
+       struct cl_fsync_io *fio;
+       int result;
+       ENTRY;
+
+       env = cl_env_nested_get(&nest);
+       if (IS_ERR(env))
+               RETURN(PTR_ERR(env));
+
+       capa = ll_osscapa_get(inode, CAPA_OPC_OSS_WRITE);
+
+       io = ccc_env_thread_io(env);
+       io->ci_obj = cl_i2info(inode)->lli_clob;
+
+       /* initialize parameters for sync */
+       fio = &io->u.ci_fsync;
+       fio->fi_capa = capa;
+       fio->fi_start = start;
+       fio->fi_end = end;
+       fio->fi_fid = ll_inode2fid(inode);
+
+       if (cl_io_init(env, io, CIT_FSYNC, io->ci_obj) == 0)
+               result = cl_io_loop(env, io);
+       else
+               result = io->ci_result;
+       cl_io_fini(env, io);
+       cl_env_nested_put(&nest, env);
+
+       capa_put(capa);
+
+       RETURN(result);
+}
+
+#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
 {
-        struct inode *inode = dentry->d_inode;
+        struct inode *inode = file->f_dentry->d_inode;
         struct ll_inode_info *lli = ll_i2info(inode);
         struct lov_stripe_md *lsm = lli->lli_smd;
         struct ptlrpc_request *req;
@@ -1945,14 +2049,16 @@ int ll_fsync(struct file *file, struct dentry *dentry, int data)
 
         /* catch async errors that were recorded back when async writeback
          * failed for pages in this mapping. */
-        err = lli->lli_async_rc;
-        lli->lli_async_rc = 0;
-        if (rc == 0)
-                rc = err;
-        if (lsm) {
-                err = lov_test_and_clear_async_rc(lsm);
+        if (!S_ISDIR(inode->i_mode)) {
+                err = lli->lli_async_rc;
+                lli->lli_async_rc = 0;
                 if (rc == 0)
                         rc = err;
+                if (lsm) {
+                        err = lov_test_and_clear_async_rc(lsm);
+                        if (rc == 0)
+                                rc = err;
+                }
         }
 
         oc = ll_mdscapa_get(inode);
@@ -1965,33 +2071,10 @@ int ll_fsync(struct file *file, struct dentry *dentry, int data)
                 ptlrpc_req_finished(req);
 
         if (data && lsm) {
-                struct obd_info *oinfo;
-
-                OBD_ALLOC_PTR(oinfo);
-                if (!oinfo)
-                        RETURN(rc ? rc : -ENOMEM);
-                OBDO_ALLOC(oinfo->oi_oa);
-                if (!oinfo->oi_oa) {
-                        OBD_FREE_PTR(oinfo);
-                        RETURN(rc ? rc : -ENOMEM);
-                }
-                oinfo->oi_oa->o_id = lsm->lsm_object_id;
-                oinfo->oi_oa->o_seq = lsm->lsm_object_seq;
-                oinfo->oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
-                obdo_from_inode(oinfo->oi_oa, inode, &ll_i2info(inode)->lli_fid,
-                                OBD_MD_FLTYPE | OBD_MD_FLATIME |
-                                OBD_MD_FLMTIME | OBD_MD_FLCTIME |
-                                OBD_MD_FLGROUP);
-                oinfo->oi_md = lsm;
-                oinfo->oi_capa = ll_osscapa_get(inode, CAPA_OPC_OSS_WRITE);
-                err = obd_sync_rqset(ll_i2sbi(inode)->ll_dt_exp, oinfo, 0,
-                                     OBD_OBJECT_EOF);
-                capa_put(oinfo->oi_capa);
+               err = cl_sync_file_range(inode, 0, OBD_OBJECT_EOF);
                 if (!rc)
                         rc = err;
-                OBDO_FREE(oinfo->oi_oa);
-                OBD_FREE_PTR(oinfo);
-                lli->lli_write_rc = err < 0 ? : 0;
+                lli->lli_write_rc = rc < 0 ? rc : 0;
         }
 
         RETURN(rc);
@@ -2060,8 +2143,9 @@ int ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock)
                 einfo.ei_mode = LCK_PW;
                 break;
         default:
-                CERROR("unknown fcntl lock type: %d\n", file_lock->fl_type);
-                RETURN (-EINVAL);
+                CDEBUG(D_INFO, "Unknown fcntl lock type: %d\n",
+                        file_lock->fl_type);
+                RETURN (-ENOTSUPP);
         }
 
         switch (cmd) {
@@ -2107,13 +2191,11 @@ int ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock)
 
         if ((file_lock->fl_flags & FL_FLOCK) &&
             (rc == 0 || file_lock->fl_type == F_UNLCK))
-                ll_flock_lock_file_wait(file, file_lock, (cmd == F_SETLKW));
-#ifdef HAVE_F_OP_FLOCK
+                flock_lock_file_wait(file, file_lock);
         if ((file_lock->fl_flags & FL_POSIX) &&
             (rc == 0 || file_lock->fl_type == F_UNLCK) &&
             !(flags & LDLM_FL_TEST_LOCK))
                 posix_lock_file_wait(file, file_lock);
-#endif
 
         RETURN(rc);
 }
@@ -2220,7 +2302,6 @@ int __ll_inode_revalidate_it(struct dentry *dentry, struct lookup_intent *it,
 {
         struct inode *inode = dentry->d_inode;
         struct ptlrpc_request *req = NULL;
-        struct ll_sb_info *sbi;
         struct obd_export *exp;
         int rc = 0;
         ENTRY;
@@ -2229,7 +2310,6 @@ int __ll_inode_revalidate_it(struct dentry *dentry, struct lookup_intent *it,
                 CERROR("REPORT THIS LINE TO PETER\n");
                 RETURN(0);
         }
-        sbi = ll_i2sbi(inode);
 
         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p),name=%s\n",
                inode->i_ino, inode->i_generation, inode, dentry->d_name.name);
@@ -2301,8 +2381,8 @@ int __ll_inode_revalidate_it(struct dentry *dentry, struct lookup_intent *it,
                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
                                              0, ealen, LUSTRE_OPC_ANY,
                                              NULL);
-                if (op_data == NULL)
-                        RETURN(-ENOMEM);
+                if (IS_ERR(op_data))
+                        RETURN(PTR_ERR(op_data));
 
                 op_data->op_valid = valid;
                 /* Once OBD_CONNECT_ATTRFID is not supported, we can't find one
@@ -2339,11 +2419,11 @@ int ll_inode_revalidate_it(struct dentry *dentry, struct lookup_intent *it,
                 RETURN(0);
         }
 
-        /* cl_glimpse_size will prefer locally cached writes if they extend
+        /* ll_glimpse_size will prefer locally cached writes if they extend
          * the file */
 
         if (rc == 0)
-                rc = cl_glimpse_size(inode);
+                rc = ll_glimpse_size(inode);
 
         RETURN(rc);
 }
@@ -2430,8 +2510,12 @@ int ll_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
 #endif
 
 
-static
-int lustre_check_acl(struct inode *inode, int mask)
+static int
+#ifdef HAVE_GENERIC_PERMISSION_4ARGS
+lustre_check_acl(struct inode *inode, int mask, unsigned int flags)
+#else
+lustre_check_acl(struct inode *inode, int mask)
+#endif
 {
 #ifdef CONFIG_FS_POSIX_ACL
         struct ll_inode_info *lli = ll_i2info(inode);
@@ -2439,6 +2523,10 @@ int lustre_check_acl(struct inode *inode, int mask)
         int rc;
         ENTRY;
 
+#ifdef HAVE_GENERIC_PERMISSION_4ARGS
+        if (flags & IPERM_FLAG_RCU)
+                return -ECHILD;
+#endif
         cfs_spin_lock(&lli->lli_lock);
         acl = posix_acl_dup(lli->lli_posix_acl);
         cfs_spin_unlock(&lli->lli_lock);
@@ -2455,11 +2543,14 @@ int lustre_check_acl(struct inode *inode, int mask)
 #endif
 }
 
-#if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,10))
-#ifndef HAVE_INODE_PERMISION_2ARGS
-int ll_inode_permission(struct inode *inode, int mask, struct nameidata *nd)
+#ifdef HAVE_GENERIC_PERMISSION_4ARGS
+int ll_inode_permission(struct inode *inode, int mask, unsigned int flags)
 #else
+# ifdef HAVE_INODE_PERMISION_2ARGS
 int ll_inode_permission(struct inode *inode, int mask)
+# else
+int ll_inode_permission(struct inode *inode, int mask, struct nameidata *nd)
+# endif
 #endif
 {
         int rc = 0;
@@ -2484,61 +2575,10 @@ int ll_inode_permission(struct inode *inode, int mask)
                 return lustre_check_remote_perm(inode, mask);
 
         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_INODE_PERM, 1);
-        rc = generic_permission(inode, mask, lustre_check_acl);
+        rc = ll_generic_permission(inode, mask, flags, lustre_check_acl);
 
         RETURN(rc);
 }
-#else
-int ll_inode_permission(struct inode *inode, int mask, struct nameidata *nd)
-{
-        int mode = inode->i_mode;
-        int rc;
-
-        CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), mask %o\n",
-               inode->i_ino, inode->i_generation, inode, mask);
-
-        if (ll_i2sbi(inode)->ll_flags & LL_SBI_RMT_CLIENT)
-                return lustre_check_remote_perm(inode, mask);
-
-        ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_INODE_PERM, 1);
-
-        if ((mask & MAY_WRITE) && IS_RDONLY(inode) &&
-            (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
-                return -EROFS;
-        if ((mask & MAY_WRITE) && IS_IMMUTABLE(inode))
-                return -EACCES;
-        if (cfs_curproc_fsuid() == inode->i_uid) {
-                mode >>= 6;
-        } else if (1) {
-                if (((mode >> 3) & mask & S_IRWXO) != mask)
-                        goto check_groups;
-                rc = lustre_check_acl(inode, mask);
-                if (rc == -EAGAIN)
-                        goto check_groups;
-                if (rc == -EACCES)
-                        goto check_capabilities;
-                return rc;
-        } else {
-check_groups:
-                if (cfs_curproc_is_in_groups(inode->i_gid))
-                        mode >>= 3;
-        }
-        if ((mode & mask & S_IRWXO) == mask)
-                return 0;
-
-check_capabilities:
-        if (!(mask & MAY_EXEC) ||
-            (inode->i_mode & S_IXUGO) || S_ISDIR(inode->i_mode))
-                if (cfs_capable(CFS_CAP_DAC_OVERRIDE))
-                        return 0;
-
-        if (cfs_capable(CFS_CAP_DAC_READ_SEARCH) && ((mask == MAY_READ) ||
-            (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE))))
-                return 0;
-
-        return -EACCES;
-}
-#endif
 
 #ifdef HAVE_FILE_READV
 #define READ_METHOD readv
@@ -2558,11 +2598,7 @@ struct file_operations ll_file_operations = {
         .READ_METHOD    = READ_FUNCTION,
         .write          = ll_file_write,
         .WRITE_METHOD   = WRITE_FUNCTION,
-#ifdef HAVE_UNLOCKED_IOCTL
         .unlocked_ioctl = ll_file_ioctl,
-#else
-        .ioctl          = ll_file_ioctl,
-#endif
         .open           = ll_file_open,
         .release        = ll_file_release,
         .mmap           = ll_file_mmap,
@@ -2582,11 +2618,7 @@ struct file_operations ll_file_operations_flock = {
         .READ_METHOD    = READ_FUNCTION,
         .write          = ll_file_write,
         .WRITE_METHOD   = WRITE_FUNCTION,
-#ifdef HAVE_UNLOCKED_IOCTL
         .unlocked_ioctl = ll_file_ioctl,
-#else
-        .ioctl          = ll_file_ioctl,
-#endif
         .open           = ll_file_open,
         .release        = ll_file_release,
         .mmap           = ll_file_mmap,
@@ -2599,9 +2631,7 @@ struct file_operations ll_file_operations_flock = {
 #endif
         .fsync          = ll_fsync,
         .flush          = ll_flush,
-#ifdef HAVE_F_OP_FLOCK
         .flock          = ll_file_flock,
-#endif
         .lock           = ll_file_flock
 };
 
@@ -2611,11 +2641,7 @@ struct file_operations ll_file_operations_noflock = {
         .READ_METHOD    = READ_FUNCTION,
         .write          = ll_file_write,
         .WRITE_METHOD   = WRITE_FUNCTION,
-#ifdef HAVE_UNLOCKED_IOCTL
         .unlocked_ioctl = ll_file_ioctl,
-#else
-        .ioctl          = ll_file_ioctl,
-#endif
         .open           = ll_file_open,
         .release        = ll_file_release,
         .mmap           = ll_file_mmap,
@@ -2628,9 +2654,7 @@ struct file_operations ll_file_operations_noflock = {
 #endif
         .fsync          = ll_fsync,
         .flush          = ll_flush,
-#ifdef HAVE_F_OP_FLOCK
         .flock          = ll_file_noflock,
-#endif
         .lock           = ll_file_noflock
 };