Whamcloud - gitweb
LU-3744 llite: use 64bits flags in ll_lov_setea()
[fs/lustre-release.git] / lustre / llite / xattr.c
index 63322d6..1b5b821 100644 (file)
@@ -1,61 +1,61 @@
-/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
- * vim:expandtab:shiftwidth=8:tabstop=8:
+/*
+ * GPL HEADER START
  *
- *  Copyright (c) 2004 - 2005 Cluster File Systems, Inc.
+ * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
- *   This file is part of Lustre, http://www.lustre.org.
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 only,
+ * as published by the Free Software Foundation.
  *
- *   Lustre is free software; you can redistribute it and/or
- *   modify it under the terms of version 2 of the GNU General Public
- *   License as published by the Free Software Foundation.
+ * This program is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License version 2 for more details (a copy is included
+ * in the LICENSE file that accompanied this code).
  *
- *   Lustre is distributed in the hope that it will be useful,
- *   but WITHOUT ANY WARRANTY; without even the implied warranty of
- *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
- *   GNU General Public License for more details.
+ * You should have received a copy of the GNU General Public License
+ * version 2 along with this program; If not, see
+ * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  *
- *   You should have received a copy of the GNU General Public License
- *   along with Lustre; if not, write to the Free Software
- *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+ * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
+ * CA 95054 USA or visit www.sun.com if you need additional information or
+ * have any questions.
+ *
+ * GPL HEADER END
+ */
+/*
+ * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Use is subject to license terms.
+ *
+ * Copyright (c) 2011, 2013, Intel Corporation.
+ */
+/*
+ * This file is part of Lustre, http://www.lustre.org/
+ * Lustre is a trademark of Sun Microsystems, Inc.
  */
 
 #include <linux/fs.h>
 #include <linux/sched.h>
 #include <linux/mm.h>
-#include <linux/smp_lock.h>
+#include <linux/selinux.h>
 
 #define DEBUG_SUBSYSTEM S_LLITE
 
 #include <obd_support.h>
 #include <lustre_lite.h>
 #include <lustre_dlm.h>
-#include <linux/lustre_version.h>
-
-#ifndef POSIX_ACL_XATTR_ACCESS
-#ifndef XATTR_NAME_ACL_ACCESS
-#define XATTR_NAME_ACL_ACCESS   "system.posix_acl_access"
-#endif
-#define POSIX_ACL_XATTR_ACCESS XATTR_NAME_ACL_ACCESS
-#endif
-#ifndef POSIX_ACL_XATTR_DEFAULT
-#ifndef XATTR_NAME_ACL_DEFAULT
-#define XATTR_NAME_ACL_DEFAULT  "system.posix_acl_default"
-#endif
-#define POSIX_ACL_XATTR_DEFAULT XATTR_NAME_ACL_DEFAULT
-#endif
+#include <lustre_ver.h>
+#include <lustre_eacl.h>
 
 #include "llite_internal.h"
 
-#define XATTR_USER_PREFIX       "user."
-#define XATTR_TRUSTED_PREFIX    "trusted."
-#define XATTR_SECURITY_PREFIX   "security."
-
 #define XATTR_USER_T            (1)
 #define XATTR_TRUSTED_T         (2)
 #define XATTR_SECURITY_T        (3)
 #define XATTR_ACL_ACCESS_T      (4)
 #define XATTR_ACL_DEFAULT_T     (5)
-#define XATTR_OTHER_T           (6)
+#define XATTR_LUSTRE_T          (6)
+#define XATTR_OTHER_T           (7)
 
 static
 int get_xattr_type(const char *name)
@@ -78,6 +78,10 @@ int get_xattr_type(const char *name)
                      sizeof(XATTR_SECURITY_PREFIX) - 1))
                 return XATTR_SECURITY_T;
 
+        if (!strncmp(name, XATTR_LUSTRE_PREFIX,
+                     sizeof(XATTR_LUSTRE_PREFIX) - 1))
+                return XATTR_LUSTRE_T;
+
         return XATTR_OTHER_T;
 }
 
@@ -86,12 +90,12 @@ int xattr_type_filter(struct ll_sb_info *sbi, int xattr_type)
 {
         if ((xattr_type == XATTR_ACL_ACCESS_T ||
              xattr_type == XATTR_ACL_DEFAULT_T) &&
-            !(sbi->ll_flags & LL_SBI_ACL))
+           !(sbi->ll_flags & LL_SBI_ACL))
                 return -EOPNOTSUPP;
 
         if (xattr_type == XATTR_USER_T && !(sbi->ll_flags & LL_SBI_USER_XATTR))
                 return -EOPNOTSUPP;
-        if (xattr_type == XATTR_TRUSTED_T && !capable(CAP_SYS_ADMIN))
+        if (xattr_type == XATTR_TRUSTED_T && !cfs_capable(CFS_CAP_SYS_ADMIN))
                 return -EPERM;
         if (xattr_type == XATTR_OTHER_T)
                 return -EOPNOTSUPP;
@@ -101,33 +105,105 @@ int xattr_type_filter(struct ll_sb_info *sbi, int xattr_type)
 
 static
 int ll_setxattr_common(struct inode *inode, const char *name,
-                       const void *value, size_t size,
-                       int flags, __u64 valid)
+                      const void *value, size_t size,
+                      int flags, __u64 valid)
 {
-        struct ll_sb_info *sbi = ll_i2sbi(inode);
-        struct ptlrpc_request *req;
-        struct ll_fid fid;
+       struct ll_sb_info *sbi = ll_i2sbi(inode);
+       struct ptlrpc_request *req = NULL;
         int xattr_type, rc;
+        struct obd_capa *oc;
+        posix_acl_xattr_header *new_value = NULL;
+        struct rmtacl_ctl_entry *rce = NULL;
+        ext_acl_xattr_header *acl = NULL;
+        const char *pv = value;
         ENTRY;
 
-        lprocfs_counter_incr(sbi->ll_stats, LPROC_LL_SETXATTR);
-
         xattr_type = get_xattr_type(name);
         rc = xattr_type_filter(sbi, xattr_type);
         if (rc)
                 RETURN(rc);
 
         /* b10667: ignore lustre special xattr for now */
-        if (xattr_type == XATTR_TRUSTED_T && strcmp(name, "trusted.lov") == 0)
+        if ((xattr_type == XATTR_TRUSTED_T && strcmp(name, "trusted.lov") == 0) ||
+            (xattr_type == XATTR_LUSTRE_T && strcmp(name, "lustre.lov") == 0))
+                RETURN(0);
+
+        /* b15587: ignore security.capability xattr for now */
+        if ((xattr_type == XATTR_SECURITY_T &&
+            strcmp(name, "security.capability") == 0))
                 RETURN(0);
 
-        ll_inode2fid(&fid, inode);
-        rc = mdc_setxattr(sbi->ll_mdc_exp, &fid, valid,
-                          name, value, size, 0, flags, &req);
+        /* LU-549:  Disable security.selinux when selinux is disabled */
+        if (xattr_type == XATTR_SECURITY_T && !selinux_is_enabled() &&
+            strcmp(name, "security.selinux") == 0)
+                RETURN(-EOPNOTSUPP);
+
+#ifdef CONFIG_FS_POSIX_ACL
+        if (sbi->ll_flags & LL_SBI_RMT_CLIENT &&
+            (xattr_type == XATTR_ACL_ACCESS_T ||
+            xattr_type == XATTR_ACL_DEFAULT_T)) {
+                rce = rct_search(&sbi->ll_rct, cfs_curproc_pid());
+                if (rce == NULL ||
+                    (rce->rce_ops != RMT_LSETFACL &&
+                    rce->rce_ops != RMT_RSETFACL))
+                        RETURN(-EOPNOTSUPP);
+
+                if (rce->rce_ops == RMT_LSETFACL) {
+                        struct eacl_entry *ee;
+
+                        ee = et_search_del(&sbi->ll_et, cfs_curproc_pid(),
+                                           ll_inode2fid(inode), xattr_type);
+                        LASSERT(ee != NULL);
+                        if (valid & OBD_MD_FLXATTR) {
+                                acl = lustre_acl_xattr_merge2ext(
+                                                (posix_acl_xattr_header *)value,
+                                                size, ee->ee_acl);
+                                if (IS_ERR(acl)) {
+                                        ee_free(ee);
+                                        RETURN(PTR_ERR(acl));
+                                }
+                                size =  CFS_ACL_XATTR_SIZE(\
+                                                le32_to_cpu(acl->a_count), \
+                                                ext_acl_xattr);
+                                pv = (const char *)acl;
+                        }
+                        ee_free(ee);
+                } else if (rce->rce_ops == RMT_RSETFACL) {
+                       int acl_size = lustre_posix_acl_xattr_filter(
+                                               (posix_acl_xattr_header *)value,
+                                               size, &new_value);
+                       if (unlikely(acl_size < 0))
+                               RETURN(acl_size);
+                       size = acl_size;
+
+                        pv = (const char *)new_value;
+                } else
+                        RETURN(-EOPNOTSUPP);
+
+                valid |= rce_ops2valid(rce->rce_ops);
+        }
+#endif
+       if (sbi->ll_xattr_cache_enabled &&
+           (rce == NULL || rce->rce_ops == RMT_LSETFACL)) {
+               rc = ll_xattr_cache_update(inode, name, pv, size, valid, flags);
+       } else {
+               oc = ll_mdscapa_get(inode);
+               rc = md_setxattr(sbi->ll_md_exp, ll_inode2fid(inode), oc,
+                               valid, name, pv, size, 0, flags,
+                               ll_i2suppgid(inode), &req);
+               capa_put(oc);
+       }
+
+#ifdef CONFIG_FS_POSIX_ACL
+        if (new_value != NULL)
+                lustre_posix_acl_xattr_free(new_value, size);
+        if (acl != NULL)
+                lustre_ext_acl_xattr_free(acl);
+#endif
         if (rc) {
                 if (rc == -EOPNOTSUPP && xattr_type == XATTR_USER_T) {
                         LCONSOLE_INFO("Disabling user_xattr feature because "
-                                      "it is not supported on the server\n"); 
+                                      "it is not supported on the server\n");
                         sbi->ll_flags &= ~LL_SBI_USER_XATTR;
                 }
                 RETURN(rc);
@@ -148,28 +224,43 @@ int ll_setxattr(struct dentry *dentry, const char *name,
         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), xattr %s\n",
                inode->i_ino, inode->i_generation, inode, name);
 
-        ll_vfs_ops_tally(ll_i2sbi(inode), VFS_OPS_SETXATTR);
+        ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_SETXATTR, 1);
 
-        if (strncmp(name, XATTR_TRUSTED_PREFIX, 8) == 0 &&
-            strcmp(name + 8, "lov") == 0) {
+        if ((strncmp(name, XATTR_TRUSTED_PREFIX,
+                     sizeof(XATTR_TRUSTED_PREFIX) - 1) == 0 &&
+             strcmp(name + sizeof(XATTR_TRUSTED_PREFIX) - 1, "lov") == 0) ||
+            (strncmp(name, XATTR_LUSTRE_PREFIX,
+                     sizeof(XATTR_LUSTRE_PREFIX) - 1) == 0 &&
+             strcmp(name + sizeof(XATTR_LUSTRE_PREFIX) - 1, "lov") == 0)) {
                 struct lov_user_md *lump = (struct lov_user_md *)value;
                 int rc = 0;
 
-                if (S_ISREG(inode->i_mode)) {
-                        struct file f;
-                        int flags = FMODE_WRITE;
-                        
-                        f.f_dentry = dentry;
-                        rc = ll_lov_setstripe_ea_info(inode, &f, flags, 
-                                                      lump, sizeof(*lump));
-                        /* b10667: rc always be 0 here for now */
-                        rc = 0;
+                /* Attributes that are saved via getxattr will always have
+                 * the stripe_offset as 0.  Instead, the MDS should be
+                 * allowed to pick the starting OST index.   b=17846 */
+                if (lump != NULL && lump->lmm_stripe_offset == 0)
+                        lump->lmm_stripe_offset = -1;
+
+               if (lump != NULL && S_ISREG(inode->i_mode)) {
+                       struct file     f;
+                       __u64           it_flags = FMODE_WRITE;
+                       int lum_size = (lump->lmm_magic == LOV_USER_MAGIC_V1) ?
+                               sizeof(*lump) : sizeof(struct lov_user_md_v3);
+
+                       f.f_dentry = dentry;
+                       rc = ll_lov_setstripe_ea_info(inode, &f, it_flags, lump,
+                                                     lum_size);
+                       /* b10667: rc always be 0 here for now */
+                       rc = 0;
                 } else if (S_ISDIR(inode->i_mode)) {
-                        rc = ll_dir_setstripe(inode, lump);
+                        rc = ll_dir_setstripe(inode, lump, 0);
                 }
-                
+
                 return rc;
-        }
+
+        } else if (strcmp(name, XATTR_NAME_LMA) == 0 ||
+                   strcmp(name, XATTR_NAME_LINK) == 0)
+                return 0;
 
         return ll_setxattr_common(inode, name, value, size, flags,
                                   OBD_MD_FLXATTR);
@@ -185,7 +276,7 @@ int ll_removexattr(struct dentry *dentry, const char *name)
         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), xattr %s\n",
                inode->i_ino, inode->i_generation, inode, name);
 
-        ll_vfs_ops_tally(ll_i2sbi(inode), VFS_OPS_REMOVEXATTR);
+        ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_REMOVEXATTR, 1);
         return ll_setxattr_common(inode, name, NULL, 0, 0,
                                   OBD_MD_FLXATTRRM);
 }
@@ -196,17 +287,16 @@ int ll_getxattr_common(struct inode *inode, const char *name,
 {
         struct ll_sb_info *sbi = ll_i2sbi(inode);
         struct ptlrpc_request *req = NULL;
-        struct mds_body *body;
-        struct ll_fid fid;
-        void *xdata;
+        struct mdt_body *body;
         int xattr_type, rc;
+        void *xdata;
+        struct obd_capa *oc;
+        struct rmtacl_ctl_entry *rce = NULL;
         ENTRY;
 
         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n",
                inode->i_ino, inode->i_generation, inode);
 
-        lprocfs_counter_incr(sbi->ll_stats, LPROC_LL_GETXATTR);
-
         /* listxattr have slightly different behavior from of ext3:
          * without 'user_xattr' ext3 will list all xattr names but
          * filtered out "^user..*"; we list them all for simplicity.
@@ -221,77 +311,125 @@ int ll_getxattr_common(struct inode *inode, const char *name,
         if (rc)
                 RETURN(rc);
 
+        /* b15587: ignore security.capability xattr for now */
+        if ((xattr_type == XATTR_SECURITY_T &&
+            strcmp(name, "security.capability") == 0))
+                RETURN(-ENODATA);
+
+        /* LU-549:  Disable security.selinux when selinux is disabled */
+        if (xattr_type == XATTR_SECURITY_T && !selinux_is_enabled() &&
+            strcmp(name, "security.selinux") == 0)
+                RETURN(-EOPNOTSUPP);
+
+#ifdef CONFIG_FS_POSIX_ACL
+        if (sbi->ll_flags & LL_SBI_RMT_CLIENT &&
+            (xattr_type == XATTR_ACL_ACCESS_T ||
+            xattr_type == XATTR_ACL_DEFAULT_T)) {
+                rce = rct_search(&sbi->ll_rct, cfs_curproc_pid());
+                if (rce == NULL ||
+                    (rce->rce_ops != RMT_LSETFACL &&
+                    rce->rce_ops != RMT_LGETFACL &&
+                    rce->rce_ops != RMT_RSETFACL &&
+                    rce->rce_ops != RMT_RGETFACL))
+                        RETURN(-EOPNOTSUPP);
+        }
+
         /* posix acl is under protection of LOOKUP lock. when calling to this,
          * we just have path resolution to the target inode, so we have great
          * chance that cached ACL is uptodate.
          */
-#ifdef CONFIG_FS_POSIX_ACL
-        if (xattr_type == XATTR_ACL_ACCESS_T) {
+        if (xattr_type == XATTR_ACL_ACCESS_T &&
+            !(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
                 struct ll_inode_info *lli = ll_i2info(inode);
                 struct posix_acl *acl;
 
-                spin_lock(&lli->lli_lock);
-                acl = posix_acl_dup(lli->lli_posix_acl);
-                spin_unlock(&lli->lli_lock);
+               spin_lock(&lli->lli_lock);
+               acl = posix_acl_dup(lli->lli_posix_acl);
+               spin_unlock(&lli->lli_lock);
 
                 if (!acl)
                         RETURN(-ENODATA);
 
-                rc = posix_acl_to_xattr(acl, buffer, size);
+                rc = posix_acl_to_xattr(&init_user_ns, acl, buffer, size);
                 posix_acl_release(acl);
                 RETURN(rc);
         }
+        if (xattr_type == XATTR_ACL_DEFAULT_T && !S_ISDIR(inode->i_mode))
+                RETURN(-ENODATA);
 #endif
 
 do_getxattr:
-        ll_inode2fid(&fid, inode);
-        rc = mdc_getxattr(sbi->ll_mdc_exp, &fid, valid, name, NULL, 0, size,
-                          &req);
-        if (rc) {
-                if (rc == -EOPNOTSUPP && xattr_type == XATTR_USER_T) {
-                        LCONSOLE_INFO("Disabling user_xattr feature because "
-                                      "it is not supported on the server\n"); 
-                        sbi->ll_flags &= ~LL_SBI_USER_XATTR;
-                }
-                RETURN(rc);
-        }
+       if (sbi->ll_xattr_cache_enabled && (rce == NULL ||
+                                           rce->rce_ops == RMT_LGETFACL ||
+                                           rce->rce_ops == RMT_LSETFACL)) {
+               rc = ll_xattr_cache_get(inode, name, buffer, size, valid);
+               if (rc < 0)
+                       GOTO(out_xattr, rc);
+       } else {
+               oc = ll_mdscapa_get(inode);
+               rc = md_getxattr(sbi->ll_md_exp, ll_inode2fid(inode), oc,
+                               valid | (rce ? rce_ops2valid(rce->rce_ops) : 0),
+                               name, NULL, 0, size, 0, &req);
+               capa_put(oc);
+
+               if (rc < 0)
+                       GOTO(out_xattr, rc);
+
+               body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
+               LASSERT(body);
+
+               /* only detect the xattr size */
+               if (size == 0)
+                       GOTO(out, rc = body->eadatasize);
+
+               if (size < body->eadatasize) {
+                       CERROR("server bug: replied size %u > %u\n",
+                               body->eadatasize, (int)size);
+                       GOTO(out, rc = -ERANGE);
+               }
+
+               if (body->eadatasize == 0)
+                       GOTO(out, rc = -ENODATA);
+
+               /* do not need swab xattr data */
+               xdata = req_capsule_server_sized_get(&req->rq_pill, &RMF_EADATA,
+                                                       body->eadatasize);
+               if (!xdata)
+                       GOTO(out, rc = -EFAULT);
+
+               memcpy(buffer, xdata, body->eadatasize);
+               rc = body->eadatasize;
+       }
 
-        body = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF, sizeof(*body));
-        LASSERT(body);
-        LASSERT_REPSWABBED(req, REPLY_REC_OFF);
-
-        /* only detect the xattr size */
-        if (size == 0)
-                GOTO(out, rc = body->eadatasize);
-
-        if (size < body->eadatasize) {
-                CERROR("server bug: replied size %u > %u\n",
-                       body->eadatasize, (int)size);
-                GOTO(out, rc = -ERANGE);
-        }
-
-        if (lustre_msg_bufcount(req->rq_repmsg) < 3) {
-                CERROR("reply bufcount %u\n",
-                       lustre_msg_bufcount(req->rq_repmsg));
-                GOTO(out, rc = -EFAULT);
-        }
-
-        /* do not need swab xattr data */
-        LASSERT_REPSWAB(req, REPLY_REC_OFF + 1);
-        xdata = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF + 1,
-                               body->eadatasize);
-        if (!xdata) {
-                CERROR("can't extract: %u : %u\n", body->eadatasize,
-                       lustre_msg_buflen(req->rq_repmsg, REPLY_REC_OFF + 1));
-                GOTO(out, rc = -EFAULT);
+#ifdef CONFIG_FS_POSIX_ACL
+       if (rce && rce->rce_ops == RMT_LSETFACL) {
+               ext_acl_xattr_header *acl;
+
+               acl = lustre_posix_acl_xattr_2ext((posix_acl_xattr_header *)buffer,
+                                               rc);
+                if (IS_ERR(acl))
+                        GOTO(out, rc = PTR_ERR(acl));
+
+                rc = ee_add(&sbi->ll_et, cfs_curproc_pid(), ll_inode2fid(inode),
+                            xattr_type, acl);
+                if (unlikely(rc < 0)) {
+                        lustre_ext_acl_xattr_free(acl);
+                        GOTO(out, rc);
+                }
         }
-
-        LASSERT(buffer);
-        memcpy(buffer, xdata, body->eadatasize);
-        rc = body->eadatasize;
+#endif
+       EXIT;
+
+out_xattr:
+       if (rc == -EOPNOTSUPP && xattr_type == XATTR_USER_T) {
+               LCONSOLE_INFO("%s: disabling user_xattr feature because "
+                               "it is not supported on the server: rc = %d\n",
+                               ll_get_fsname(inode->i_sb, NULL, 0), rc);
+               sbi->ll_flags &= ~LL_SBI_USER_XATTR;
+       }
 out:
         ptlrpc_req_finished(req);
-        RETURN(rc);
+        return rc;
 }
 
 ssize_t ll_getxattr(struct dentry *dentry, const char *name,
@@ -305,41 +443,77 @@ ssize_t ll_getxattr(struct dentry *dentry, const char *name,
         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), xattr %s\n",
                inode->i_ino, inode->i_generation, inode, name);
 
-        ll_vfs_ops_tally(ll_i2sbi(inode), VFS_OPS_GETXATTR);
+        ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_GETXATTR, 1);
 
-         if (strncmp(name, XATTR_TRUSTED_PREFIX, 8) == 0 &&
-             strcmp(name + 8, "lov") == 0) {
-                 struct lov_user_md *lump;
-                 struct lov_mds_md *lmm = NULL;
-                 struct ptlrpc_request *request = NULL;
-                 int rc = 0, lmmsize;
+        if ((strncmp(name, XATTR_TRUSTED_PREFIX,
+                     sizeof(XATTR_TRUSTED_PREFIX) - 1) == 0 &&
+             strcmp(name + sizeof(XATTR_TRUSTED_PREFIX) - 1, "lov") == 0) ||
+            (strncmp(name, XATTR_LUSTRE_PREFIX,
+                     sizeof(XATTR_LUSTRE_PREFIX) - 1) == 0 &&
+             strcmp(name + sizeof(XATTR_LUSTRE_PREFIX) - 1, "lov") == 0)) {
+               struct lov_stripe_md *lsm;
+                struct lov_user_md *lump;
+                struct lov_mds_md *lmm = NULL;
+                struct ptlrpc_request *request = NULL;
+                int rc = 0, lmmsize = 0;
 
-                 if (S_ISREG(inode->i_mode)) {
-                         rc = ll_lov_getstripe_ea_info(dentry->d_parent->d_inode, 
-                                                       dentry->d_name.name, &lmm, 
-                                                       &lmmsize, &request);
-                 } else if (S_ISDIR(inode->i_mode)) {
-                         rc = ll_dir_getstripe(inode, &lmm, &lmmsize, &request);
-                 }
+                if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
+                        return -ENODATA;
 
-                 if (rc < 0)
+                if (size == 0 && S_ISDIR(inode->i_mode)) {
+                        /* XXX directory EA is fix for now, optimize to save
+                         * RPC transfer */
+                        GOTO(out, rc = sizeof(struct lov_user_md));
+                }
+
+               lsm = ccc_inode_lsm_get(inode);
+               if (lsm == NULL) {
+                        if (S_ISDIR(inode->i_mode)) {
+                                rc = ll_dir_getstripe(inode, &lmm,
+                                                      &lmmsize, &request);
+                        } else {
+                                rc = -ENODATA;
+                        }
+                } else {
+                        /* LSM is present already after lookup/getattr call.
+                         * we need to grab layout lock once it is implemented */
+                       rc = obd_packmd(ll_i2dtexp(inode), &lmm, lsm);
+                       lmmsize = rc;
+               }
+               ccc_inode_lsm_put(inode, lsm);
+
+                if (rc < 0)
+                       GOTO(out, rc);
+
+                if (size == 0) {
+                        /* used to call ll_get_max_mdsize() forward to get
+                         * the maximum buffer size, while some apps (such as
+                         * rsync 3.0.x) care much about the exact xattr value
+                         * size */
+                        rc = lmmsize;
                         GOTO(out, rc);
-                 if (size == 0)
-                        GOTO(out, rc = lmmsize);
+                }
 
-                 if (size < lmmsize) {
-                         CERROR("server bug: replied size %u > %u\n",
-                                lmmsize, (int)size);
-                         GOTO(out, rc = -ERANGE);
-                 }
+                if (size < lmmsize) {
+                        CERROR("server bug: replied size %d > %d for %s (%s)\n",
+                               lmmsize, (int)size, dentry->d_name.name, name);
+                        GOTO(out, rc = -ERANGE);
+                }
 
-                 lump = (struct lov_user_md *)buffer;
-                 memcpy(lump, lmm, lmmsize);
+                lump = (struct lov_user_md *)buffer;
+                memcpy(lump, lmm, lmmsize);
+               /* do not return layout gen for getxattr otherwise it would
+                * confuse tar --xattr by recognizing layout gen as stripe
+                * offset when the file is restored. See LU-2809. */
+               lump->lmm_layout_gen = 0;
 
-                 rc = lmmsize;
+                rc = lmmsize;
 out:
-                 ptlrpc_req_finished(request);
-                 return(rc);
+                if (request)
+                        ptlrpc_req_finished(request);
+                else if (lmm)
+                        obd_free_diskmd(ll_i2dtexp(inode), &lmm);
+                return(rc);
         }
 
         return ll_getxattr_common(inode, name, buffer, size, OBD_MD_FLXATTR);
@@ -349,51 +523,74 @@ ssize_t ll_listxattr(struct dentry *dentry, char *buffer, size_t size)
 {
         struct inode *inode = dentry->d_inode;
         int rc = 0, rc2 = 0;
+        struct lov_mds_md *lmm = NULL;
+        struct ptlrpc_request *request = NULL;
+        int lmmsize;
 
         LASSERT(inode);
 
         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n",
                inode->i_ino, inode->i_generation, inode);
 
-        ll_vfs_ops_tally(ll_i2sbi(inode), VFS_OPS_LISTXATTR);
+        ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_LISTXATTR, 1);
 
         rc = ll_getxattr_common(inode, NULL, buffer, size, OBD_MD_FLXATTRLS);
+        if (rc < 0)
+                GOTO(out, rc);
+
+       if (buffer != NULL) {
+               struct ll_sb_info *sbi = ll_i2sbi(inode);
+               char *xattr_name = buffer;
+               int xlen, rem = rc;
+
+               while (rem > 0) {
+                       xlen = strnlen(xattr_name, rem - 1) + 1;
+                       rem -= xlen;
+                       if (xattr_type_filter(sbi,
+                                       get_xattr_type(xattr_name)) == 0) {
+                               /* skip OK xattr type
+                                * leave it in buffer
+                                */
+                               xattr_name += xlen;
+                               continue;
+                       }
+                       /* move up remaining xattrs in buffer
+                        * removing the xattr that is not OK
+                        */
+                       memmove(xattr_name, xattr_name + xlen, rem);
+                       rc -= xlen;
+               }
+       }
+       if (S_ISREG(inode->i_mode)) {
+               if (!ll_i2info(inode)->lli_has_smd)
+                        rc2 = -1;
+        } else if (S_ISDIR(inode->i_mode)) {
+                rc2 = ll_dir_getstripe(inode, &lmm, &lmmsize, &request);
+        }
 
-        if (!capable(CAP_SYS_ADMIN)) {
-                struct lov_mds_md *lmm = NULL;
-                struct ptlrpc_request *request = NULL;
-                int lmmsize;
-
-                if (S_ISREG(inode->i_mode)) {
-                        struct ll_inode_info *lli = ll_i2info(inode);
-                        struct lov_stripe_md *lsm = NULL;
-                        lsm = lli->lli_smd;
-                        if (lsm == NULL)
-                                rc2 = -1; 
-                } else if (S_ISDIR(inode->i_mode)) {
-                        rc2 = ll_dir_getstripe(inode, &lmm, &lmmsize, &request);
-                }
-
-                if (rc2 < 0) {
-                        GOTO(out, rc2 = 0);
-                } else {
-                        const int prefix_len = sizeof(XATTR_TRUSTED_PREFIX)-1;
-                        const size_t name_len   = sizeof("lov") - 1;
-                        const size_t total_len  = prefix_len + name_len + 1;
-
-                        if (buffer && (rc + total_len) <= size) {
-                                buffer += rc;
-                                memcpy(buffer,XATTR_TRUSTED_PREFIX, prefix_len);
-                                memcpy(buffer+prefix_len, "lov", name_len);
-                                buffer[prefix_len + name_len] = '\0';
-                        }
-                        rc2 = total_len;
-                }
+        if (rc2 < 0) {
+                GOTO(out, rc2 = 0);
+        } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)) {
+                const int prefix_len = sizeof(XATTR_LUSTRE_PREFIX) - 1;
+                const size_t name_len   = sizeof("lov") - 1;
+                const size_t total_len  = prefix_len + name_len + 1;
+
+               if (((rc + total_len) > size) && (buffer != NULL)) {
+                       ptlrpc_req_finished(request);
+                       return -ERANGE;
+               }
+
+               if (buffer != NULL) {
+                       buffer += rc;
+                       memcpy(buffer, XATTR_LUSTRE_PREFIX, prefix_len);
+                       memcpy(buffer + prefix_len, "lov", name_len);
+                       buffer[prefix_len + name_len] = '\0';
+               }
+               rc2 = total_len;
+       }
 out:
-                ptlrpc_req_finished(request);
-                rc = rc + rc2;
-        }
-        
+        ptlrpc_req_finished(request);
+        rc = rc + rc2;
+
         return rc;
 }
-