Whamcloud - gitweb
LU-4198 clio: AIO support for direct IO
[fs/lustre-release.git] / lustre / llite / file.c
index e36bb37..d3cdc6f 100644 (file)
@@ -42,9 +42,7 @@
 #include <linux/file.h>
 #include <linux/sched.h>
 #include <linux/user_namespace.h>
-#ifdef HAVE_UIDGID_HEADER
-# include <linux/uidgid.h>
-#endif
+#include <linux/uidgid.h>
 
 #include <uapi/linux/lustre/lustre_ioctl.h>
 #include <lustre_swab.h>
@@ -163,6 +161,7 @@ static int ll_close_inode_openhandle(struct inode *inode,
                op_data->op_attr_blocks += ((struct inode *)data)->i_blocks;
                op_data->op_attr.ia_valid |= ATTR_SIZE;
                op_data->op_xvalid |= OP_XVALID_BLOCKS;
+               /* fallthrough */
        case MDS_CLOSE_LAYOUT_SPLIT:
        case MDS_CLOSE_LAYOUT_SWAP: {
                struct split_param *sp = data;
@@ -353,7 +352,9 @@ static int ll_md_close(struct inode *inode, struct file *file)
        }
        mutex_unlock(&lli->lli_och_mutex);
 
-       if (!md_lock_match(ll_i2mdexp(inode), flags, ll_inode2fid(inode),
+       /* LU-4398: do not cache write open lock if the file has exec bit */
+       if ((lockmode == LCK_CW && inode->i_mode & S_IXUGO) ||
+           !md_lock_match(ll_i2mdexp(inode), flags, ll_inode2fid(inode),
                           LDLM_IBITS, &policy, lockmode, &lockh))
                rc = ll_md_real_close(inode, fd->fd_omode);
 
@@ -371,19 +372,19 @@ out:
  */
 int ll_file_release(struct inode *inode, struct file *file)
 {
-        struct ll_file_data *fd;
-        struct ll_sb_info *sbi = ll_i2sbi(inode);
-        struct ll_inode_info *lli = ll_i2info(inode);
-        int rc;
-        ENTRY;
+       struct ll_file_data *fd;
+       struct ll_sb_info *sbi = ll_i2sbi(inode);
+       struct ll_inode_info *lli = ll_i2info(inode);
+       ktime_t kstart = ktime_get();
+       int rc;
+
+       ENTRY;
 
        CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
               PFID(ll_inode2fid(inode)), inode);
 
-       if (inode->i_sb->s_root != file_dentry(file))
-                ll_stats_ops_tally(sbi, LPROC_LL_RELEASE, 1);
-        fd = LUSTRE_FPRIVATE(file);
-        LASSERT(fd != NULL);
+       fd = LUSTRE_FPRIVATE(file);
+       LASSERT(fd != NULL);
 
        /* The last ref on @file, maybe not the the owner pid of statahead,
         * because parent and child process can share the same file handle. */
@@ -393,7 +394,7 @@ int ll_file_release(struct inode *inode, struct file *file)
        if (inode->i_sb->s_root == file_dentry(file)) {
                LUSTRE_FPRIVATE(file) = NULL;
                ll_file_data_put(fd);
-               RETURN(0);
+               GOTO(out, rc = 0);
        }
 
        pcc_file_release(inode, file);
@@ -409,6 +410,10 @@ int ll_file_release(struct inode *inode, struct file *file)
        if (CFS_FAIL_TIMEOUT_MS(OBD_FAIL_PTLRPC_DUMP_LOG, cfs_fail_val))
                libcfs_debug_dumplog();
 
+out:
+       if (!rc && inode->i_sb->s_root != file_dentry(file))
+               ll_stats_ops_tally(sbi, LPROC_LL_RELEASE,
+                                  ktime_us_delta(ktime_get(), kstart));
        RETURN(rc);
 }
 
@@ -448,8 +453,8 @@ void ll_dom_finish_open(struct inode *inode, struct ptlrpc_request *req,
        if (obj == NULL)
                RETURN_EXIT;
 
-       if (!req_capsule_has_field(&req->rq_pill, &RMF_NIOBUF_INLINE,
-                                  RCL_SERVER))
+       if (!req_capsule_field_present(&req->rq_pill, &RMF_NIOBUF_INLINE,
+                                      RCL_SERVER))
                RETURN_EXIT;
 
        rnb = req_capsule_server_get(&req->rq_pill, &RMF_NIOBUF_INLINE);
@@ -461,7 +466,7 @@ void ll_dom_finish_open(struct inode *inode, struct ptlrpc_request *req,
         * client PAGE_SIZE to be used on that client, if server's PAGE_SIZE is
         * smaller then offset may be not aligned and that data is just ignored.
         */
-       if (rnb->rnb_offset % PAGE_SIZE)
+       if (rnb->rnb_offset & ~PAGE_MASK)
                RETURN_EXIT;
 
        /* Server returns whole file or just file tail if it fills in reply
@@ -481,9 +486,9 @@ void ll_dom_finish_open(struct inode *inode, struct ptlrpc_request *req,
        data = (char *)rnb + sizeof(*rnb);
 
        lnb.lnb_file_offset = rnb->rnb_offset;
-       start = lnb.lnb_file_offset / PAGE_SIZE;
+       start = lnb.lnb_file_offset >> PAGE_SHIFT;
        index = 0;
-       LASSERT(lnb.lnb_file_offset % PAGE_SIZE == 0);
+       LASSERT((lnb.lnb_file_offset & ~PAGE_MASK) == 0);
        lnb.lnb_page_offset = 0;
        do {
                lnb.lnb_data = data + (index << PAGE_SHIFT);
@@ -690,6 +695,7 @@ int ll_file_open(struct inode *inode, struct file *file)
        struct obd_client_handle **och_p = NULL;
        __u64 *och_usecount = NULL;
        struct ll_file_data *fd;
+       ktime_t kstart = ktime_get();
        int rc = 0;
        ENTRY;
 
@@ -779,7 +785,9 @@ restart:
         } else {
                 LASSERT(*och_usecount == 0);
                if (!it->it_disposition) {
-                       struct ll_dentry_data *ldd = ll_d2d(file->f_path.dentry);
+                       struct dentry *dentry = file_dentry(file);
+                       struct ll_dentry_data *ldd;
+
                         /* 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
@@ -801,18 +809,21 @@ restart:
                         *  lookup path only, since ll_iget_for_nfs always calls
                         *  ll_d_init().
                         */
+                       ldd = ll_d2d(dentry);
                        if (ldd && ldd->lld_nfs_dentry) {
                                ldd->lld_nfs_dentry = 0;
-                               it->it_flags |= MDS_OPEN_LOCK;
+                               if (!filename_is_volatile(dentry->d_name.name,
+                                                         dentry->d_name.len,
+                                                         NULL))
+                                       it->it_flags |= MDS_OPEN_LOCK;
                        }
 
-                        /*
+                       /*
                         * Always specify MDS_OPEN_BY_FID because we don't want
                         * to get file with different fid.
                         */
                        it->it_flags |= MDS_OPEN_BY_FID;
-                       rc = ll_intent_file_open(file_dentry(file), NULL, 0,
-                                                it);
+                       rc = ll_intent_file_open(dentry, NULL, 0, it);
                         if (rc)
                                 GOTO(out_openerr, rc);
 
@@ -847,14 +858,17 @@ restart:
                GOTO(out_och_free, rc);
 
        mutex_unlock(&lli->lli_och_mutex);
-        fd = NULL;
 
-        /* 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))
-                GOTO(out_och_free, rc);
+       /* lockless for direct IO so that it can do IO in parallel */
+       if (file->f_flags & O_DIRECT)
+               fd->fd_flags |= LL_FILE_LOCKLESS_IO;
+       fd = NULL;
 
+       /* 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))
+               GOTO(out_och_free, rc);
        cl_lov_delay_create_clear(&file->f_flags);
        GOTO(out_och_free, rc);
 
@@ -873,9 +887,10 @@ out_openerr:
 
                if (fd != NULL)
                        ll_file_data_put(fd);
-        } else {
-                ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_OPEN, 1);
-        }
+       } else {
+               ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_OPEN,
+                                  ktime_us_delta(ktime_get(), kstart));
+       }
 
 out_nofiledata:
        if (it && it_disposition(it, DISP_ENQ_OPEN_REF)) {
@@ -883,7 +898,7 @@ out_nofiledata:
                it_clear_disposition(it, DISP_ENQ_OPEN_REF);
        }
 
-        return rc;
+       return rc;
 }
 
 static int ll_md_blocking_lease_ast(struct ldlm_lock *lock,
@@ -1063,7 +1078,9 @@ ll_lease_open(struct inode *inode, struct file *file, fmode_t fmode,
                GOTO(out_release_it, rc);
 
        LASSERT(it_disposition(&it, DISP_ENQ_OPEN_REF));
-       ll_och_fill(sbi->ll_md_exp, &it, och);
+       rc = ll_och_fill(sbi->ll_md_exp, &it, och);
+       if (rc)
+               GOTO(out_release_it, rc);
 
        if (!it_disposition(&it, DISP_OPEN_LEASE)) /* old server? */
                GOTO(out_close, rc = -EOPNOTSUPP);
@@ -1278,11 +1295,10 @@ int ll_merge_attr(const struct lu_env *env, struct inode *inode)
         * POSIX. Solving this problem needs to send an RPC to MDT for each
         * read, this will hurt performance.
         */
-       if (inode->i_atime.tv_sec < lli->lli_atime ||
-           lli->lli_update_atime) {
+       if (ll_file_test_and_clear_flag(lli, LLIF_UPDATE_ATIME) ||
+           inode->i_atime.tv_sec < lli->lli_atime)
                inode->i_atime.tv_sec = lli->lli_atime;
-               lli->lli_update_atime = 0;
-       }
+
        inode->i_mtime.tv_sec = lli->lli_mtime;
        inode->i_ctime.tv_sec = lli->lli_ctime;
 
@@ -1378,7 +1394,8 @@ static bool file_is_noatime(const struct file *file)
        return false;
 }
 
-void ll_io_init(struct cl_io *io, struct file *file, enum cl_io_type iot)
+void ll_io_init(struct cl_io *io, struct file *file, enum cl_io_type iot,
+               struct vvp_io_args *args)
 {
        struct inode *inode = file_inode(file);
        struct ll_file_data *fd  = LUSTRE_FPRIVATE(file);
@@ -1391,7 +1408,13 @@ void ll_io_init(struct cl_io *io, struct file *file, enum cl_io_type iot)
                io->u.ci_wr.wr_sync   = !!(file->f_flags & O_SYNC ||
                                           file->f_flags & O_DIRECT ||
                                           IS_SYNC(inode));
+#ifdef HAVE_GENERIC_WRITE_SYNC_2ARGS
+               io->u.ci_wr.wr_sync  |= !!(args &&
+                                          args->via_io_subtype == IO_NORMAL &&
+                                          args->u.normal.via_iocb->ki_flags & IOCB_DSYNC);
+#endif
        }
+
        io->ci_obj = ll_i2info(inode)->lli_clob;
        io->ci_lockreq = CILR_MAYBE;
        if (ll_file_nolock(file)) {
@@ -1455,7 +1478,7 @@ ll_file_io_generic(const struct lu_env *env, struct vvp_io_args *args,
        ssize_t                 result = 0;
        int                     rc = 0;
        unsigned                retried = 0;
-       bool                    restarted = false;
+       unsigned                ignore_lockless = 0;
 
        ENTRY;
 
@@ -1465,7 +1488,8 @@ ll_file_io_generic(const struct lu_env *env, struct vvp_io_args *args,
 
 restart:
        io = vvp_env_thread_io(env);
-       ll_io_init(io, file, iot);
+       ll_io_init(io, file, iot, args);
+       io->ci_ignore_lockless = ignore_lockless;
        io->ci_ndelay_tried = retried;
 
        if (cl_io_rw_init(env, io, iot, *ppos, count) == 0) {
@@ -1538,7 +1562,8 @@ out:
               file->f_path.dentry->d_name.name,
               iot, rc, result, io->ci_need_restart);
 
-       if ((rc == 0 || rc == -ENODATA) && count > 0 && io->ci_need_restart) {
+       if ((rc == 0 || rc == -ENODATA || rc == -ENOLCK) &&
+           count > 0 && io->ci_need_restart) {
                CDEBUG(D_VFSTRACE,
                       "%s: restart %s from %lld, count: %zu, ret: %zd, rc: %d\n",
                       file_dentry(file)->d_name.name,
@@ -1546,7 +1571,7 @@ out:
                       *ppos, count, result, rc);
                /* preserve the tried count for FLR */
                retried = io->ci_ndelay_tried;
-               restarted = true;
+               ignore_lockless = io->ci_ignore_lockless;
                goto restart;
        }
 
@@ -1636,7 +1661,7 @@ ll_do_fast_read(struct kiocb *iocb, struct iov_iter *iter)
        if (result > 0) {
                ll_heat_add(file_inode(iocb->ki_filp), CIT_READ, result);
                ll_stats_ops_tally(ll_i2sbi(file_inode(iocb->ki_filp)),
-                               LPROC_LL_READ_BYTES, result);
+                                  LPROC_LL_READ_BYTES, result);
        }
 
        return result;
@@ -1649,9 +1674,11 @@ static ssize_t ll_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
 {
        struct lu_env *env;
        struct vvp_io_args *args;
+       struct file *file = iocb->ki_filp;
        ssize_t result;
        ssize_t rc2;
        __u16 refcheck;
+       ktime_t kstart = ktime_get();
        bool cached;
 
        if (!iov_iter_count(to))
@@ -1670,9 +1697,9 @@ static ssize_t ll_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
         */
        result = pcc_file_read_iter(iocb, to, &cached);
        if (cached)
-               return result;
+               GOTO(out, result);
 
-       ll_ras_enter(iocb->ki_filp);
+       ll_ras_enter(file, iocb->ki_pos, iov_iter_count(to));
 
        result = ll_do_fast_read(iocb, to);
        if (result < 0 || iov_iter_count(to) == 0)
@@ -1686,7 +1713,7 @@ static ssize_t ll_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
        args->u.normal.via_iter = to;
        args->u.normal.via_iocb = iocb;
 
-       rc2 = ll_file_io_generic(env, args, iocb->ki_filp, CIT_READ,
+       rc2 = ll_file_io_generic(env, args, file, CIT_READ,
                                 &iocb->ki_pos, iov_iter_count(to));
        if (rc2 > 0)
                result += rc2;
@@ -1695,6 +1722,14 @@ static ssize_t ll_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
 
        cl_env_put(env, &refcheck);
 out:
+       if (result > 0) {
+               ll_rw_stats_tally(ll_i2sbi(file_inode(file)), current->pid,
+                                 LUSTRE_FPRIVATE(file), iocb->ki_pos, result,
+                                 READ);
+               ll_stats_ops_tally(ll_i2sbi(file_inode(file)), LPROC_LL_READ,
+                                  ktime_us_delta(ktime_get(), kstart));
+       }
+
        return result;
 }
 
@@ -1765,8 +1800,10 @@ static ssize_t ll_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
        struct vvp_io_args *args;
        struct lu_env *env;
        ssize_t rc_tiny = 0, rc_normal;
+       struct file *file = iocb->ki_filp;
        __u16 refcheck;
        bool cached;
+       ktime_t kstart = ktime_get();
        int result;
 
        ENTRY;
@@ -1786,15 +1823,15 @@ static ssize_t ll_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
         */
        result = pcc_file_write_iter(iocb, from, &cached);
        if (cached && result != -ENOSPC && result != -EDQUOT)
-               return result;
+               GOTO(out, rc_normal = result);
 
        /* NB: we can't do direct IO for tiny writes because they use the page
         * cache, we can't do sync writes because tiny writes can't flush
         * pages, and we can't do append writes because we can't guarantee the
         * required DLM locks are held to protect file size.
         */
-       if (ll_sbi_has_tiny_write(ll_i2sbi(file_inode(iocb->ki_filp))) &&
-           !(iocb->ki_filp->f_flags & (O_DIRECT | O_SYNC | O_APPEND)))
+       if (ll_sbi_has_tiny_write(ll_i2sbi(file_inode(file))) &&
+           !(file->f_flags & (O_DIRECT | O_SYNC | O_APPEND)))
                rc_tiny = ll_do_tiny_write(iocb, from);
 
        /* In case of error, go on and try normal write - Only stop if tiny
@@ -1811,8 +1848,8 @@ static ssize_t ll_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
        args->u.normal.via_iter = from;
        args->u.normal.via_iocb = iocb;
 
-       rc_normal = ll_file_io_generic(env, args, iocb->ki_filp, CIT_WRITE,
-                                   &iocb->ki_pos, iov_iter_count(from));
+       rc_normal = ll_file_io_generic(env, args, file, CIT_WRITE,
+                                      &iocb->ki_pos, iov_iter_count(from));
 
        /* On success, combine bytes written. */
        if (rc_tiny >= 0 && rc_normal > 0)
@@ -1825,6 +1862,14 @@ static ssize_t ll_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
 
        cl_env_put(env, &refcheck);
 out:
+       if (rc_normal > 0) {
+               ll_rw_stats_tally(ll_i2sbi(file_inode(file)), current->pid,
+                                 LUSTRE_FPRIVATE(file), iocb->ki_pos,
+                                 rc_normal, WRITE);
+               ll_stats_ops_tally(ll_i2sbi(file_inode(file)), LPROC_LL_WRITE,
+                                  ktime_us_delta(ktime_get(), kstart));
+       }
+
        RETURN(rc_normal);
 }
 
@@ -1990,7 +2035,7 @@ static ssize_t ll_file_splice_read(struct file *in_file, loff_t *ppos,
        if (cached)
                RETURN(result);
 
-       ll_ras_enter(in_file);
+       ll_ras_enter(in_file, *ppos, count);
 
        env = cl_env_get(&refcheck);
         if (IS_ERR(env))
@@ -2002,6 +2047,11 @@ static ssize_t ll_file_splice_read(struct file *in_file, loff_t *ppos,
 
         result = ll_file_io_generic(env, args, in_file, CIT_READ, ppos, count);
         cl_env_put(env, &refcheck);
+
+       if (result > 0)
+               ll_rw_stats_tally(ll_i2sbi(file_inode(in_file)), current->pid,
+                                 LUSTRE_FPRIVATE(in_file), *ppos, result,
+                                 READ);
         RETURN(result);
 }
 
@@ -2018,7 +2068,7 @@ int ll_lov_setstripe_ea_info(struct inode *inode, struct dentry *dentry,
        if ((__swab32(lum->lmm_magic) & le32_to_cpu(LOV_MAGIC_MASK)) ==
            le32_to_cpu(LOV_MAGIC_MAGIC)) {
                /* this code will only exist for big-endian systems */
-               lustre_swab_lov_user_md(lum);
+               lustre_swab_lov_user_md(lum, 0);
        }
 
        ll_inode_size_lock(inode);
@@ -2101,7 +2151,7 @@ int ll_lov_getstripe_ea_info(struct inode *inode, const char *filename,
                                stripe_count = 0;
                }
 
-               lustre_swab_lov_user_md((struct lov_user_md *)lmm);
+               lustre_swab_lov_user_md((struct lov_user_md *)lmm, 0);
 
                /* if function called for directory - we should
                 * avoid swab not existent lsm objects */
@@ -2204,6 +2254,7 @@ out:
        RETURN(rc);
 }
 
+
 static int
 ll_get_grouplock(struct inode *inode, struct file *file, unsigned long arg)
 {
@@ -2219,18 +2270,28 @@ ll_get_grouplock(struct inode *inode, struct file *file, unsigned long arg)
                RETURN(-EINVAL);
        }
 
-        if (ll_file_nolock(file))
-                RETURN(-EOPNOTSUPP);
+       if (ll_file_nolock(file))
+               RETURN(-EOPNOTSUPP);
+retry:
+       if (file->f_flags & O_NONBLOCK) {
+               if (!mutex_trylock(&lli->lli_group_mutex))
+                       RETURN(-EAGAIN);
+       } else
+               mutex_lock(&lli->lli_group_mutex);
 
-       spin_lock(&lli->lli_lock);
        if (fd->fd_flags & LL_FILE_GROUP_LOCKED) {
                CWARN("group lock already existed with gid %lu\n",
                      fd->fd_grouplock.lg_gid);
-               spin_unlock(&lli->lli_lock);
-               RETURN(-EINVAL);
+               GOTO(out, rc = -EINVAL);
+       }
+       if (arg != lli->lli_group_gid && lli->lli_group_users != 0) {
+               if (file->f_flags & O_NONBLOCK)
+                       GOTO(out, rc = -EAGAIN);
+               mutex_unlock(&lli->lli_group_mutex);
+               wait_var_event(&lli->lli_group_users, !lli->lli_group_users);
+               GOTO(retry, rc = 0);
        }
        LASSERT(fd->fd_grouplock.lg_lock == NULL);
-       spin_unlock(&lli->lli_lock);
 
        /**
         * XXX: group lock needs to protect all OST objects while PFL
@@ -2250,7 +2311,7 @@ ll_get_grouplock(struct inode *inode, struct file *file, unsigned long arg)
 
                env = cl_env_get(&refcheck);
                if (IS_ERR(env))
-                       RETURN(PTR_ERR(env));
+                       GOTO(out, rc = PTR_ERR(env));
 
                rc = cl_object_layout_get(env, obj, &cl);
                if (!rc && cl.cl_is_composite)
@@ -2259,28 +2320,26 @@ ll_get_grouplock(struct inode *inode, struct file *file, unsigned long arg)
 
                cl_env_put(env, &refcheck);
                if (rc)
-                       RETURN(rc);
+                       GOTO(out, rc);
        }
 
        rc = cl_get_grouplock(ll_i2info(inode)->lli_clob,
                              arg, (file->f_flags & O_NONBLOCK), &grouplock);
-       if (rc)
-               RETURN(rc);
 
-       spin_lock(&lli->lli_lock);
-       if (fd->fd_flags & LL_FILE_GROUP_LOCKED) {
-               spin_unlock(&lli->lli_lock);
-               CERROR("another thread just won the race\n");
-               cl_put_grouplock(&grouplock);
-               RETURN(-EINVAL);
-       }
+       if (rc)
+               GOTO(out, rc);
 
        fd->fd_flags |= LL_FILE_GROUP_LOCKED;
        fd->fd_grouplock = grouplock;
-       spin_unlock(&lli->lli_lock);
+       if (lli->lli_group_users == 0)
+               lli->lli_group_gid = grouplock.lg_gid;
+       lli->lli_group_users++;
 
        CDEBUG(D_INFO, "group lock %lu obtained\n", arg);
-       RETURN(0);
+out:
+       mutex_unlock(&lli->lli_group_mutex);
+
+       RETURN(rc);
 }
 
 static int ll_put_grouplock(struct inode *inode, struct file *file,
@@ -2289,32 +2348,40 @@ static int ll_put_grouplock(struct inode *inode, struct file *file,
        struct ll_inode_info   *lli = ll_i2info(inode);
        struct ll_file_data    *fd = LUSTRE_FPRIVATE(file);
        struct ll_grouplock     grouplock;
+       int                     rc;
        ENTRY;
 
-       spin_lock(&lli->lli_lock);
+       mutex_lock(&lli->lli_group_mutex);
        if (!(fd->fd_flags & LL_FILE_GROUP_LOCKED)) {
-               spin_unlock(&lli->lli_lock);
-                CWARN("no group lock held\n");
-                RETURN(-EINVAL);
-        }
+               CWARN("no group lock held\n");
+               GOTO(out, rc = -EINVAL);
+       }
 
        LASSERT(fd->fd_grouplock.lg_lock != NULL);
 
        if (fd->fd_grouplock.lg_gid != arg) {
                CWARN("group lock %lu doesn't match current id %lu\n",
                      arg, fd->fd_grouplock.lg_gid);
-               spin_unlock(&lli->lli_lock);
-               RETURN(-EINVAL);
+               GOTO(out, rc = -EINVAL);
        }
 
        grouplock = fd->fd_grouplock;
        memset(&fd->fd_grouplock, 0, sizeof(fd->fd_grouplock));
        fd->fd_flags &= ~LL_FILE_GROUP_LOCKED;
-       spin_unlock(&lli->lli_lock);
 
        cl_put_grouplock(&grouplock);
+
+       lli->lli_group_users--;
+       if (lli->lli_group_users == 0) {
+               lli->lli_group_gid = 0;
+               wake_up_var(&lli->lli_group_users);
+       }
        CDEBUG(D_INFO, "group lock %lu released\n", arg);
-       RETURN(0);
+       GOTO(out, rc = 0);
+out:
+       mutex_unlock(&lli->lli_group_mutex);
+
+       RETURN(rc);
 }
 
 /**
@@ -2349,7 +2416,9 @@ int ll_release_openhandle(struct dentry *dentry, struct lookup_intent *it)
         if (!och)
                 GOTO(out, rc = -ENOMEM);
 
-       ll_och_fill(ll_i2sbi(inode)->ll_md_exp, it, och);
+       rc = ll_och_fill(ll_i2sbi(inode)->ll_md_exp, it, och);
+       if (rc)
+               GOTO(out, rc);
 
        rc = ll_close_inode_openhandle(inode, och, 0, NULL);
 out:
@@ -2995,6 +3064,7 @@ static int ll_ladvise_sanity(struct inode *inode,
                               ladvise_names[advice], rc);
                        GOTO(out, rc);
                }
+               /* fallthrough */
        case LU_LADVISE_WILLREAD:
        case LU_LADVISE_DONTNEED:
        default:
@@ -3981,6 +4051,7 @@ static loff_t ll_file_seek(struct file *file, loff_t offset, int origin)
 {
        struct inode *inode = file_inode(file);
        loff_t retval, eof = 0;
+       ktime_t kstart = ktime_get();
 
        ENTRY;
        retval = offset + ((origin == SEEK_END) ? i_size_read(inode) :
@@ -3988,7 +4059,6 @@ static loff_t ll_file_seek(struct file *file, loff_t offset, int origin)
        CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), to=%llu=%#llx(%d)\n",
               PFID(ll_inode2fid(inode)), inode, retval, retval,
               origin);
-       ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_LLSEEK, 1);
 
        if (origin == SEEK_END || origin == SEEK_HOLE || origin == SEEK_DATA) {
                retval = ll_glimpse_size(inode);
@@ -3998,7 +4068,10 @@ static loff_t ll_file_seek(struct file *file, loff_t offset, int origin)
        }
 
        retval = ll_generic_file_llseek_size(file, offset, origin,
-                                         ll_file_maxbytes(inode), eof);
+                                            ll_file_maxbytes(inode), eof);
+       if (retval >= 0)
+               ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_LLSEEK,
+                                  ktime_us_delta(ktime_get(), kstart));
        RETURN(retval);
 }
 
@@ -4088,6 +4161,7 @@ int ll_fsync(struct file *file, loff_t start, loff_t end, int datasync)
        struct inode *inode = dentry->d_inode;
        struct ll_inode_info *lli = ll_i2info(inode);
        struct ptlrpc_request *req;
+       ktime_t kstart = ktime_get();
        int rc, err;
 
        ENTRY;
@@ -4096,8 +4170,6 @@ int ll_fsync(struct file *file, loff_t start, loff_t end, int datasync)
               "datasync %d\n",
               PFID(ll_inode2fid(inode)), inode, start, end, datasync);
 
-       ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FSYNC, 1);
-
        /* fsync's caller has already called _fdata{sync,write}, we want
         * that IO to finish before calling the osc and mdc sync methods */
        rc = filemap_write_and_wait_range(inode->i_mapping, start, end);
@@ -4143,6 +4215,10 @@ int ll_fsync(struct file *file, loff_t start, loff_t end, int datasync)
        }
 
        inode_unlock(inode);
+
+       if (!rc)
+               ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FSYNC,
+                                  ktime_us_delta(ktime_get(), kstart));
        RETURN(rc);
 }
 
@@ -4160,6 +4236,7 @@ ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock)
        struct lustre_handle lockh = { 0 };
        union ldlm_policy_data flock = { { 0 } };
        int fl_type = file_lock->fl_type;
+       ktime_t kstart = ktime_get();
        __u64 flags = 0;
        int rc;
        int rc2 = 0;
@@ -4168,23 +4245,22 @@ ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock)
        CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID" file_lock=%p\n",
               PFID(ll_inode2fid(inode)), file_lock);
 
-        ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FLOCK, 1);
-
-        if (file_lock->fl_flags & FL_FLOCK) {
-                LASSERT((cmd == F_SETLKW) || (cmd == F_SETLK));
-                /* flocks are whole-file locks */
-                flock.l_flock.end = OFFSET_MAX;
-                /* For flocks owner is determined by the local file desctiptor*/
-                flock.l_flock.owner = (unsigned long)file_lock->fl_file;
-        } else if (file_lock->fl_flags & FL_POSIX) {
-                flock.l_flock.owner = (unsigned long)file_lock->fl_owner;
-                flock.l_flock.start = file_lock->fl_start;
-                flock.l_flock.end = file_lock->fl_end;
-        } else {
-                RETURN(-EINVAL);
-        }
-        flock.l_flock.pid = file_lock->fl_pid;
+       if (file_lock->fl_flags & FL_FLOCK) {
+               LASSERT((cmd == F_SETLKW) || (cmd == F_SETLK));
+               /* flocks are whole-file locks */
+               flock.l_flock.end = OFFSET_MAX;
+               /* For flocks owner is determined by the local file desctiptor*/
+               flock.l_flock.owner = (unsigned long)file_lock->fl_file;
+       } else if (file_lock->fl_flags & FL_POSIX) {
+               flock.l_flock.owner = (unsigned long)file_lock->fl_owner;
+               flock.l_flock.start = file_lock->fl_start;
+               flock.l_flock.end = file_lock->fl_end;
+       } else {
+               RETURN(-EINVAL);
+       }
+       flock.l_flock.pid = file_lock->fl_pid;
 
+#if defined(HAVE_LM_COMPARE_OWNER) || defined(lm_compare_owner)
        /* Somewhat ugly workaround for svc lockd.
         * lockd installs custom fl_lmops->lm_compare_owner that checks
         * for the fl_owner to be the same (which it always is on local node
@@ -4194,6 +4270,7 @@ ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock)
         * pointer space for current->files are not intersecting */
        if (file_lock->fl_lmops && file_lock->fl_lmops->lm_compare_owner)
                flock.l_flock.owner = (unsigned long)file_lock->fl_pid;
+#endif
 
        switch (fl_type) {
         case F_RDLCK:
@@ -4286,7 +4363,10 @@ ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock)
 
        ll_finish_md_op_data(op_data);
 
-        RETURN(rc);
+       if (!rc)
+               ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FLOCK,
+                                  ktime_us_delta(ktime_get(), kstart));
+       RETURN(rc);
 }
 
 int ll_get_fid_by_name(struct inode *parent, const char *name,
@@ -4626,9 +4706,9 @@ static int ll_inode_revalidate(struct dentry *dentry, enum ldlm_intent_flags op)
         * here to preserve get_cwd functionality on 2.6.
         * Bug 10503 */
        if (!dentry->d_inode->i_nlink) {
-               ll_lock_dcache(inode);
+               spin_lock(&inode->i_lock);
                d_lustre_invalidate(dentry, 0);
-               ll_unlock_dcache(inode);
+               spin_unlock(&inode->i_lock);
        }
 
        ll_lookup_finish_locks(&oit, dentry);
@@ -4672,10 +4752,9 @@ int ll_getattr_dentry(struct dentry *de, struct kstat *stat)
        struct inode *inode = de->d_inode;
        struct ll_sb_info *sbi = ll_i2sbi(inode);
        struct ll_inode_info *lli = ll_i2info(inode);
+       ktime_t kstart = ktime_get();
        int rc;
 
-       ll_stats_ops_tally(sbi, LPROC_LL_GETATTR, 1);
-
        rc = ll_inode_revalidate(de, IT_GETATTR);
        if (rc < 0)
                RETURN(rc);
@@ -4736,7 +4815,10 @@ int ll_getattr_dentry(struct dentry *de, struct kstat *stat)
        stat->size = i_size_read(inode);
        stat->blocks = inode->i_blocks;
 
-        return 0;
+       ll_stats_ops_tally(sbi, LPROC_LL_GETATTR,
+                          ktime_us_delta(ktime_get(), kstart));
+
+       return 0;
 }
 
 #ifdef HAVE_INODEOPS_ENHANCED_GETATTR
@@ -4804,7 +4886,7 @@ struct posix_acl *ll_get_acl(struct inode *inode, int type)
 }
 
 #ifdef HAVE_IOP_SET_ACL
-#ifdef CONFIG_FS_POSIX_ACL
+#ifdef CONFIG_LUSTRE_FS_POSIX_ACL
 int ll_set_acl(struct inode *inode, struct posix_acl *acl, int type)
 {
        struct ll_sb_info *sbi = ll_i2sbi(inode);
@@ -4860,50 +4942,10 @@ out:
                set_cached_acl(inode, type, acl);
        RETURN(rc);
 }
-#endif /* CONFIG_FS_POSIX_ACL */
+#endif /* CONFIG_LUSTRE_FS_POSIX_ACL */
 #endif /* HAVE_IOP_SET_ACL */
 
-#ifndef HAVE_GENERIC_PERMISSION_2ARGS
-static int
-# ifdef HAVE_GENERIC_PERMISSION_4ARGS
-ll_check_acl(struct inode *inode, int mask, unsigned int flags)
-# else
-ll_check_acl(struct inode *inode, int mask)
-# endif
-{
-# ifdef CONFIG_FS_POSIX_ACL
-       struct posix_acl *acl;
-       int rc;
-       ENTRY;
-
-#  ifdef HAVE_GENERIC_PERMISSION_4ARGS
-       if (flags & IPERM_FLAG_RCU)
-               return -ECHILD;
-#  endif
-       acl = ll_get_acl(inode, ACL_TYPE_ACCESS);
-
-       if (!acl)
-               RETURN(-EAGAIN);
-
-       rc = posix_acl_permission(inode, acl, mask);
-       posix_acl_release(acl);
-
-       RETURN(rc);
-# else /* !CONFIG_FS_POSIX_ACL */
-       return -EAGAIN;
-# endif /* CONFIG_FS_POSIX_ACL */
-}
-#endif /* HAVE_GENERIC_PERMISSION_2ARGS */
-
-#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;
        struct ll_sb_info *sbi;
@@ -4912,15 +4954,11 @@ int ll_inode_permission(struct inode *inode, int mask, struct nameidata *nd)
        const struct cred *old_cred = NULL;
        cfs_cap_t cap;
        bool squash_id = false;
+       ktime_t kstart = ktime_get();
        ENTRY;
 
-#ifdef MAY_NOT_BLOCK
        if (mask & MAY_NOT_BLOCK)
                return -ECHILD;
-#elif defined(HAVE_GENERIC_PERMISSION_4ARGS)
-       if (flags & IPERM_FLAG_RCU)
-               return -ECHILD;
-#endif
 
        /* as root inode are NOT getting validated in lookup operation,
         * need to do it before permission check. */
@@ -4962,14 +5000,17 @@ int ll_inode_permission(struct inode *inode, int mask, struct nameidata *nd)
                old_cred = override_creds(cred);
        }
 
-       ll_stats_ops_tally(sbi, LPROC_LL_INODE_PERM, 1);
-       rc = ll_generic_permission(inode, mask, flags, ll_check_acl);
+       rc = generic_permission(inode, mask);
        /* restore current process's credentials and FS capability */
        if (squash_id) {
                revert_creds(old_cred);
                put_cred(cred);
        }
 
+       if (!rc)
+               ll_stats_ops_tally(sbi, LPROC_LL_INODE_PERM,
+                                  ktime_us_delta(ktime_get(), kstart));
+
        RETURN(rc);
 }