Whamcloud - gitweb
LU-5100 llite: set dir LOV xattr length variable
[fs/lustre-release.git] / lustre / llite / xattr.c
index aaa7631..effdd01 100644 (file)
@@ -27,7 +27,7 @@
  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  * Use is subject to license terms.
  *
- * Copyright (c) 2011, 2012, Whamcloud, Inc.
+ * Copyright (c) 2011, 2013, 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>
-#ifdef HAVE_SELINUX_IS_ENABLED
 #include <linux/selinux.h>
-#endif
 
 #define DEBUG_SUBSYSTEM S_LLITE
 
@@ -48,7 +45,7 @@
 #include <lustre_lite.h>
 #include <lustre_dlm.h>
 #include <lustre_ver.h>
-#include <lustre_acl.h>
+#include <lustre_eacl.h>
 
 #include "llite_internal.h"
 
@@ -96,8 +93,6 @@ int xattr_type_filter(struct ll_sb_info *sbi, int xattr_type)
            !(sbi->ll_flags & LL_SBI_ACL))
                 return -EOPNOTSUPP;
 
-        if (xattr_type == XATTR_SECURITY_T && !selinux_is_enabled())
-                return -EOPNOTSUPP;
         if (xattr_type == XATTR_USER_T && !(sbi->ll_flags & LL_SBI_USER_XATTR))
                 return -EOPNOTSUPP;
         if (xattr_type == XATTR_TRUSTED_T && !cfs_capable(CFS_CAP_SYS_ADMIN))
@@ -110,11 +105,11 @@ 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;
@@ -128,6 +123,15 @@ int ll_setxattr_common(struct inode *inode, const char *name,
         if (rc)
                 RETURN(rc);
 
+       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 ((xattr_type == XATTR_TRUSTED_T && strcmp(name, "trusted.lov") == 0) ||
             (xattr_type == XATTR_LUSTRE_T && strcmp(name, "lustre.lov") == 0))
@@ -138,22 +142,27 @@ int ll_setxattr_common(struct inode *inode, const char *name,
             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);
+        /* 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);
 
-                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);
+#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, current_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, current_pid(),
+                                          ll_inode2fid(inode), xattr_type);
+                       LASSERT(ee != NULL);
                         if (valid & OBD_MD_FLXATTR) {
                                 acl = lustre_acl_xattr_merge2ext(
                                                 (posix_acl_xattr_header *)value,
@@ -169,11 +178,12 @@ int ll_setxattr_common(struct inode *inode, const char *name,
                         }
                         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);
+                       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
@@ -182,11 +192,11 @@ int ll_setxattr_common(struct inode *inode, const char *name,
                 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);
+       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);
@@ -214,8 +224,8 @@ 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);
 
@@ -234,15 +244,17 @@ int ll_setxattr(struct dentry *dentry, const char *name,
                 if (lump != NULL && lump->lmm_stripe_offset == 0)
                         lump->lmm_stripe_offset = -1;
 
-                if (lump != NULL && 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 (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, 0);
                 }
@@ -264,8 +276,8 @@ 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,
@@ -283,10 +295,11 @@ int ll_getxattr_common(struct inode *inode, const char *name,
         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
@@ -307,18 +320,23 @@ int ll_getxattr_common(struct inode *inode, const char *name,
             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);
-        }
+       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, current_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
@@ -326,17 +344,17 @@ int ll_getxattr_common(struct inode *inode, const char *name,
          */
         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;
 
-                cfs_spin_lock(&lli->lli_lock);
-                acl = posix_acl_dup(lli->lli_posix_acl);
-                cfs_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);
         }
@@ -345,68 +363,87 @@ 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 (sbi->ll_xattr_cache_enabled && xattr_type != XATTR_ACL_ACCESS_T) {
+               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:
+               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 (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);
 
-        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);
 
-        /* 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;
+       }
 
 #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 (rce != NULL && rce->rce_ops == RMT_LSETFACL) {
+               ext_acl_xattr_header *acl;
+
+               acl = lustre_posix_acl_xattr_2ext(buffer, rc);
+               if (IS_ERR(acl))
+                       GOTO(out, rc = PTR_ERR(acl));
+
+               rc = ee_add(&sbi->ll_et, current_pid(), ll_inode2fid(inode),
+                           xattr_type, acl);
+               if (unlikely(rc < 0)) {
+                       lustre_ext_acl_xattr_free(acl);
+                       GOTO(out, rc);
+               }
+       }
 #endif
-
-        if (body->eadatasize == 0) {
-                rc = -ENODATA;
-        } else {
-                LASSERT(buffer);
-                memcpy(buffer, xdata, body->eadatasize);
-                rc = body->eadatasize;
-        }
-        EXIT;
+       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;
@@ -420,8 +457,8 @@ ssize_t ll_getxattr(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_GETXATTR, 1);
 
@@ -431,6 +468,7 @@ ssize_t ll_getxattr(struct dentry *dentry, const char *name,
             (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;
@@ -439,26 +477,21 @@ ssize_t ll_getxattr(struct dentry *dentry, const char *name,
                 if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
                         return -ENODATA;
 
-                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));
-                }
-
-                if (!ll_i2info(inode)->lli_smd) {
-                        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,
-                                        ll_i2info(inode)->lli_smd);
-                        lmmsize = rc;
-                }
+               lsm = ccc_inode_lsm_get(inode);
+               if (lsm == NULL) {
+                       if (S_ISDIR(inode->i_mode)) {
+                               rc = ll_dir_getstripe(inode, (void **)&lmm,
+                                                     &lmmsize, &request, 0);
+                       } 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);
@@ -480,6 +513,10 @@ ssize_t ll_getxattr(struct dentry *dentry, const char *name,
 
                 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;
 out:
@@ -503,8 +540,8 @@ ssize_t ll_listxattr(struct dentry *dentry, char *buffer, size_t size)
 
         LASSERT(inode);
 
-        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);
 
         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_LISTXATTR, 1);
 
@@ -535,12 +572,13 @@ ssize_t ll_listxattr(struct dentry *dentry, char *buffer, size_t size)
                        rc -= xlen;
                }
        }
-        if (S_ISREG(inode->i_mode)) {
-                if (ll_i2info(inode)->lli_smd == NULL)
-                        rc2 = -1;
-        } else if (S_ISDIR(inode->i_mode)) {
-                rc2 = ll_dir_getstripe(inode, &lmm, &lmmsize, &request);
-        }
+       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, (void **)&lmm, &lmmsize, &request,
+                                      0);
+       }
 
         if (rc2 < 0) {
                 GOTO(out, rc2 = 0);
@@ -549,7 +587,12 @@ ssize_t ll_listxattr(struct dentry *dentry, char *buffer, size_t size)
                 const size_t name_len   = sizeof("lov") - 1;
                 const size_t total_len  = prefix_len + name_len + 1;
 
-               if (buffer && (rc + total_len) <= size) {
+               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);