Whamcloud - gitweb
LU-8056 llite: POSIX_ACL_XATTR_{ACCESS,DEFAULT} removed in 4.5
[fs/lustre-release.git] / lustre / llite / xattr.c
index 0957d09..461a0ba 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.
  * GPL HEADER END
  */
 /*
- * Copyright  2008 Sun Microsystems, Inc. All rights reserved
+ * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  * Use is subject to license terms.
+ *
+ * Copyright (c) 2011, 2015, Intel Corporation.
  */
 /*
  * This file is part of Lustre, http://www.lustre.org/
 #include <linux/fs.h>
 #include <linux/sched.h>
 #include <linux/mm.h>
-#include <linux/smp_lock.h>
+#include <linux/xattr.h>
+#include <linux/selinux.h>
 
 #define DEBUG_SUBSYSTEM S_LLITE
 
 #include <obd_support.h>
-#include <lustre_lite.h>
 #include <lustre_dlm.h>
 #include <lustre_ver.h>
-//#include <lustre_mdc.h>
-#include <lustre_acl.h>
+#include <lustre_eacl.h>
 
 #include "llite_internal.h"
 
+/* xattr related to IMA(Integrity Measurement Architecture) */
+#ifndef XATTR_NAME_IMA
+#define XATTR_NAME_IMA         "security.ima"
+#endif
+#ifndef XATTR_NAME_EVM
+#define XATTR_NAME_EVM         "security.evm"
+#endif
+
+#ifndef XATTR_NAME_POSIX_ACL_ACCESS
+# define XATTR_NAME_POSIX_ACL_ACCESS POSIX_ACL_XATTR_ACCESS
+#endif
+
+#ifndef XATTR_NAME_POSIX_ACL_DEFAULT
+# define XATTR_NAME_POSIX_ACL_DEFAULT POSIX_ACL_XATTR_DEFAULT
+#endif
+
 #define XATTR_USER_T            (1)
 #define XATTR_TRUSTED_T         (2)
 #define XATTR_SECURITY_T        (3)
 static
 int get_xattr_type(const char *name)
 {
-        if (!strcmp(name, POSIX_ACL_XATTR_ACCESS))
+       if (!strcmp(name, XATTR_NAME_POSIX_ACL_ACCESS))
                 return XATTR_ACL_ACCESS_T;
 
-        if (!strcmp(name, POSIX_ACL_XATTR_DEFAULT))
+       if (!strcmp(name, XATTR_NAME_POSIX_ACL_DEFAULT))
                 return XATTR_ACL_DEFAULT_T;
 
         if (!strncmp(name, XATTR_USER_PREFIX,
@@ -106,84 +121,56 @@ 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_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;
 
+       /*FIXME: enable IMA when the conditions are ready */
+       if (strncmp(name, XATTR_NAME_IMA,
+                   sizeof(XATTR_NAME_IMA)) == 0 ||
+           strncmp(name, XATTR_NAME_EVM,
+                   sizeof(XATTR_NAME_EVM)) == 0)
+               return -EOPNOTSUPP;
+
         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) ||
-            (xattr_type == XATTR_LUSTRE_T && strcmp(name, "lustre.lov") == 0))
+       if ((xattr_type == XATTR_ACL_ACCESS_T ||
+            xattr_type == XATTR_ACL_DEFAULT_T) &&
+#ifdef HAVE_INODE_OWNER_OR_CAPABLE
+           !inode_owner_or_capable(inode))
+#else
+           !is_owner_or_cap(inode))
+#endif
+               return -EPERM;
+
+       /* b10667: ignore lustre special xattr for now */
+       if (strcmp(name, XATTR_NAME_HSM) == 0 ||
+               (xattr_type == XATTR_TRUSTED_T &&
+               strcmp(name, XATTR_NAME_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);
 
-#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) {
-                        size = lustre_posix_acl_xattr_filter(
-                                                (posix_acl_xattr_header *)value,
-                                                size, &new_value);
-                        if (unlikely(size < 0))
-                                RETURN(size);
-
-                        pv = (const char *)new_value;
-                } else
-                        RETURN(-EOPNOTSUPP);
-
-                valid |= rce_ops2valid(rce->rce_ops);
-        }
-#endif
-        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
+        /* 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);
+
+       rc = md_setxattr(sbi->ll_md_exp, ll_inode2fid(inode), valid, name, pv,
+                        size, 0, flags, ll_i2suppgid(inode), &req);
         if (rc) {
                 if (rc == -EOPNOTSUPP && xattr_type == XATTR_USER_T) {
                         LCONSOLE_INFO("Disabling user_xattr feature because "
@@ -197,6 +184,37 @@ int ll_setxattr_common(struct inode *inode, const char *name,
         RETURN(0);
 }
 
+static int get_hsm_state(struct inode *inode, __u32 *hus_states)
+{
+       struct md_op_data *op_data;
+       struct hsm_user_state *hus;
+       int rc;
+
+       OBD_ALLOC_PTR(hus);
+       if (hus == NULL)
+               return -ENOMEM;
+
+       op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
+                                    LUSTRE_OPC_ANY, hus);
+       if (!IS_ERR(op_data)) {
+               rc = obd_iocontrol(LL_IOC_HSM_STATE_GET, ll_i2mdexp(inode),
+                                  sizeof(*op_data), op_data, NULL);
+               if (rc == 0)
+                       *hus_states = hus->hus_states;
+               else
+                       CDEBUG(D_VFSTRACE, "obd_iocontrol failed. rc = %d\n",
+                              rc);
+
+               ll_finish_md_op_data(op_data);
+       } else {
+               rc = PTR_ERR(op_data);
+               CDEBUG(D_VFSTRACE, "Could not prepare the opdata. rc = %d\n",
+                      rc);
+       }
+       OBD_FREE_PTR(hus);
+       return rc;
+}
+
 int ll_setxattr(struct dentry *dentry, const char *name,
                 const void *value, size_t size, int flags)
 {
@@ -205,29 +223,61 @@ int ll_setxattr(struct dentry *dentry, const char *name,
         LASSERT(inode);
         LASSERT(name);
 
-        CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), xattr %s\n",
-               inode->i_ino, inode->i_generation, inode, name);
+       CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), xattr %s\n",
+              PFID(ll_inode2fid(inode)), inode, name);
 
         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_SETXATTR, 1);
 
-        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;
+       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;
+
+               /* 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;
+               /* Avoid anyone directly setting the RELEASED flag. */
+               if (lump != NULL &&
+                       (lump->lmm_pattern & LOV_PATTERN_F_RELEASED)) {
+                       /* Only if we have a released flag check if the file
+                       * was indeed archived. */
+                       __u32 state = HS_NONE;
+                       rc = get_hsm_state(inode, &state);
+                       if (rc != 0)
+                               RETURN(rc);
+                       if (!(state & HS_ARCHIVED)) {
+                               CDEBUG(D_VFSTRACE,
+                                       "hus_states state = %x, pattern = %x\n",
+                                       state, lump->lmm_pattern);
+                               /* Here the state is: real file is not
+                                * archived but user is requesting to set
+                                * the RELEASED flag so we mask off the
+                                * released flag from the request */
+                               lump->lmm_pattern ^= LOV_PATTERN_F_RELEASED;
+                       }
+               }
+
+               if (lump != NULL && S_ISREG(inode->i_mode)) {
+                       struct file     f;
+                       __u64           it_flags = FMODE_WRITE;
+                       int             lum_size;
+
+                       lum_size = ll_lov_user_md_size(lump);
+                       if (lum_size < 0 || size < lum_size)
+                               return 0; /* b=10667: ignore error */
+
+                       memset(&f, 0, sizeof(f)); /* f.f_flags is used below */
+                       f.f_path.dentry = dentry;
+                       rc = ll_lov_setstripe_ea_info(inode, &f, it_flags, lump,
+                                                     lum_size);
+                       /* b=10667: rc always be 0 here for now */
+                       rc = 0;
                 } else if (S_ISDIR(inode->i_mode)) {
                         rc = ll_dir_setstripe(inode, lump, 0);
                 }
@@ -249,15 +299,14 @@ int ll_removexattr(struct dentry *dentry, const char *name)
         LASSERT(inode);
         LASSERT(name);
 
-        CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), xattr %s\n",
-               inode->i_ino, inode->i_generation, inode, name);
+       CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), xattr %s\n",
+              PFID(ll_inode2fid(inode)), inode, name);
 
         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_REMOVEXATTR, 1);
         return ll_setxattr_common(inode, name, NULL, 0, 0,
                                   OBD_MD_FLXATTRRM);
 }
 
-static
 int ll_getxattr_common(struct inode *inode, const char *name,
                        void *buffer, size_t size, __u64 valid)
 {
@@ -266,12 +315,11 @@ int ll_getxattr_common(struct inode *inode, const char *name,
         struct mdt_body *body;
         int xattr_type, rc;
         void *xdata;
-        struct obd_capa *oc;
-        struct rmtacl_ctl_entry *rce = NULL;
+       struct ll_inode_info *lli = ll_i2info(inode);
         ENTRY;
 
-        CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n",
-               inode->i_ino, inode->i_generation, inode);
+       CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
+              PFID(ll_inode2fid(inode)), inode);
 
         /* listxattr have slightly different behavior from of ext3:
          * without 'user_xattr' ext3 will list all xattr names but
@@ -287,38 +335,32 @@ int ll_getxattr_common(struct inode *inode, const char *name,
         if (rc)
                 RETURN(rc);
 
-#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);
-        }
-#endif
+        /* 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
         /* 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 &&
-            !(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
-                struct ll_inode_info *lli = ll_i2info(inode);
-                struct posix_acl *acl;
+       if (xattr_type == XATTR_ACL_ACCESS_T) {
+               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);
         }
@@ -327,170 +369,220 @@ int ll_getxattr_common(struct inode *inode, const char *name,
 #endif
 
 do_getxattr:
-        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) {
-                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);
-        }
-
-        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);
-        }
-
-        /* 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);
-
-#ifdef CONFIG_FS_POSIX_ACL
-        if (body->eadatasize >= 0 && rce && rce->rce_ops == RMT_LSETFACL) {
-                ext_acl_xattr_header *acl;
-
-                acl = lustre_posix_acl_xattr_2ext((posix_acl_xattr_header *)xdata,
-                                                  body->eadatasize);
-                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);
-                }
-        }
-
-        if (xattr_type == XATTR_ACL_ACCESS_T && !body->eadatasize)
-                GOTO(out, rc = -ENODATA);
-#endif
-        LASSERT(buffer);
-        memcpy(buffer, xdata, body->eadatasize);
-        rc = body->eadatasize;
-        EXIT;
+       if (sbi->ll_xattr_cache_enabled &&
+           xattr_type != XATTR_ACL_ACCESS_T &&
+           (xattr_type != XATTR_SECURITY_T ||
+               strcmp(name, "security.selinux") != 0)) {
+               rc = ll_xattr_cache_get(inode, name, buffer, size, valid);
+               if (rc == -EAGAIN)
+                       goto getxattr_nocache;
+               if (rc < 0)
+                       GOTO(out_xattr, rc);
+
+               /* Add "system.posix_acl_access" to the list */
+               if (lli->lli_posix_acl != NULL && valid & OBD_MD_FLXATTRLS) {
+                       if (size == 0) {
+                               rc += sizeof(XATTR_NAME_ACL_ACCESS);
+                       } else if (size - rc >= sizeof(XATTR_NAME_ACL_ACCESS)) {
+                               memcpy(buffer + rc, XATTR_NAME_ACL_ACCESS,
+                                      sizeof(XATTR_NAME_ACL_ACCESS));
+                               rc += sizeof(XATTR_NAME_ACL_ACCESS);
+                       } else {
+                               GOTO(out_xattr, rc = -ERANGE);
+                       }
+               }
+       } else {
+getxattr_nocache:
+               rc = md_getxattr(sbi->ll_md_exp, ll_inode2fid(inode),
+                               valid, name, NULL, 0, size, 0, &req);
+
+               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->mbo_eadatasize);
+
+               if (size < body->mbo_eadatasize) {
+                       CERROR("server bug: replied size %u > %u\n",
+                               body->mbo_eadatasize, (int)size);
+                       GOTO(out, rc = -ERANGE);
+               }
+
+               if (body->mbo_eadatasize == 0)
+                       GOTO(out, rc = -ENODATA);
+
+               /* do not need swab xattr data */
+               xdata = req_capsule_server_sized_get(&req->rq_pill, &RMF_EADATA,
+                                                       body->mbo_eadatasize);
+               if (!xdata)
+                       GOTO(out, rc = -EFAULT);
+
+               memcpy(buffer, xdata, body->mbo_eadatasize);
+               rc = body->mbo_eadatasize;
+       }
+
+       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;
 }
 
-ssize_t ll_getxattr(struct dentry *dentry, const char *name,
-                    void *buffer, size_t size)
+static ssize_t ll_getxattr_lov(struct inode *inode, void *buf, size_t buf_size)
 {
-        struct inode *inode = dentry->d_inode;
-
-        LASSERT(inode);
-        LASSERT(name);
-
-        CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), xattr %s\n",
-               inode->i_ino, inode->i_generation, inode, name);
-
-        ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_GETXATTR, 1);
-
-        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_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);
-                } else {
-                        rc = -ENODATA;
-                }
-
-                if (rc < 0)
-                       GOTO(out, rc);
-                if (size == 0)
-                       GOTO(out, rc = lmmsize);
-
-                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);
-
-                rc = lmmsize;
-out:
-                ptlrpc_req_finished(request);
-                return(rc);
-        }
-
-        return ll_getxattr_common(inode, name, buffer, size, OBD_MD_FLXATTR);
+       ssize_t rc;
+
+       if (S_ISREG(inode->i_mode)) {
+               struct cl_object *obj = ll_i2info(inode)->lli_clob;
+               struct lu_env *env;
+               struct cl_layout cl = {
+                       .cl_buf.lb_buf = buf,
+                       .cl_buf.lb_len = buf_size,
+               };
+               __u16 refcheck;
+
+               if (obj == NULL)
+                       RETURN(-ENODATA);
+
+               env = cl_env_get(&refcheck);
+               if (IS_ERR(env))
+                       RETURN(PTR_ERR(env));
+
+               rc = cl_object_layout_get(env, obj, &cl);
+               if (rc < 0)
+                       GOTO(out_env, rc);
+
+               if (cl.cl_size == 0)
+                       GOTO(out_env, rc = -ENODATA);
+
+               rc = cl.cl_size;
+
+               if (buf_size == 0)
+                       GOTO(out_env, rc);
+
+               LASSERT(buf != NULL && rc <= buf_size);
+
+               /* Do not return layout gen for getxattr() since
+                * otherwise it would confuse tar --xattr by
+                * recognizing layout gen as stripe offset when the
+                * file is restored. See LU-2809. */
+               ((struct lov_mds_md *)buf)->lmm_layout_gen = 0;
+out_env:
+               cl_env_put(env, &refcheck);
+
+               RETURN(rc);
+       } else if (S_ISDIR(inode->i_mode)) {
+               struct lov_mds_md *lmm = NULL;
+               int lmm_size = 0;
+               struct ptlrpc_request *req = NULL;
+
+               rc = ll_dir_getstripe(inode, (void **)&lmm, &lmm_size,
+                                     &req, 0);
+               if (rc < 0)
+                       GOTO(out_req, rc);
+
+               if (buf_size == 0)
+                       GOTO(out_req, rc = lmm_size);
+
+               if (buf_size < lmm_size)
+                       GOTO(out_req, rc = -ERANGE);
+
+               memcpy(buf, lmm, lmm_size);
+               GOTO(out_req, rc = lmm_size);
+out_req:
+               if (req != NULL)
+                       ptlrpc_req_finished(req);
+
+               return rc;
+       } else {
+               RETURN(-ENODATA);
+       }
 }
 
-ssize_t ll_listxattr(struct dentry *dentry, char *buffer, size_t size)
+ssize_t ll_getxattr(struct dentry *dentry, const char *name, void *buf,
+                   size_t buf_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);
+       struct inode *inode = dentry->d_inode;
 
-        ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_LISTXATTR, 1);
+       LASSERT(inode);
+       LASSERT(name);
 
-        rc = ll_getxattr_common(inode, NULL, buffer, size, OBD_MD_FLXATTRLS);
+       CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), xattr %s\n",
+              PFID(ll_inode2fid(inode)), inode, name);
 
-        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);
-        }
+       ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_GETXATTR, 1);
 
-        if (rc2 < 0) {
-                GOTO(out, rc2 = 0);
-        } else {
-                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 (buffer && (rc + total_len) <= size) {
-                        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;
+       if (strcmp(name, XATTR_LUSTRE_LOV) == 0 ||
+           strcmp(name, XATTR_NAME_LOV) == 0)
+               return ll_getxattr_lov(inode, buf, buf_size);
+       else
+               return ll_getxattr_common(inode, name, buf, buf_size,
+                                         OBD_MD_FLXATTR);
+}
 
-        return rc;
+ssize_t ll_listxattr(struct dentry *dentry, char *buf, size_t buf_size)
+{
+       struct inode *inode = dentry->d_inode;
+       struct ll_sb_info *sbi = ll_i2sbi(inode);
+       char *xattr_name;
+       ssize_t rc, rc2;
+       size_t len, rem;
+
+       CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
+              PFID(ll_inode2fid(inode)), inode);
+
+       ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_LISTXATTR, 1);
+
+       rc = ll_getxattr_common(inode, NULL, buf, buf_size, OBD_MD_FLXATTRLS);
+       if (rc < 0)
+               RETURN(rc);
+
+       /* If we're being called to get the size of the xattr list
+        * (buf_size == 0) then just assume that a lustre.lov xattr
+        * exists. */
+       if (buf_size == 0)
+               RETURN(rc + sizeof(XATTR_LUSTRE_LOV));
+
+       xattr_name = buf;
+       rem = rc;
+
+       while (rem > 0) {
+               len = strnlen(xattr_name, rem - 1) + 1;
+               rem -= len;
+               if (xattr_type_filter(sbi, get_xattr_type(xattr_name)) == 0) {
+                       /* Skip OK xattr type, leave it in buffer. */
+                       xattr_name += len;
+                       continue;
+               }
+
+               /* Move up remaining xattrs in buffer removing the
+                * xattr that is not OK. */
+               memmove(xattr_name, xattr_name + len, rem);
+               rc -= len;
+       }
+
+       rc2 = ll_getxattr_lov(inode, NULL, 0);
+       if (rc2 == -ENODATA)
+               RETURN(rc);
+
+       if (rc2 < 0)
+               RETURN(rc2);
+
+       if (buf_size < rc + sizeof(XATTR_LUSTRE_LOV))
+               RETURN(-ERANGE);
+
+       memcpy(buf + rc, XATTR_LUSTRE_LOV, sizeof(XATTR_LUSTRE_LOV));
+
+       RETURN(rc + sizeof(XATTR_LUSTRE_LOV));
 }