Whamcloud - gitweb
LU-2139 osc: Track and limit "unstable" pages
[fs/lustre-release.git] / lustre / llite / llite_lib.c
index c549b78..66e72d5 100644 (file)
@@ -99,13 +99,16 @@ static struct ll_sb_info *ll_init_sbi(void)
                lru_page_max = (pages / 4) * 3;
        }
 
-       /* initialize lru data */
+       /* initialize ll_cache data */
        cfs_atomic_set(&sbi->ll_cache.ccc_users, 0);
        sbi->ll_cache.ccc_lru_max = lru_page_max;
        cfs_atomic_set(&sbi->ll_cache.ccc_lru_left, lru_page_max);
        spin_lock_init(&sbi->ll_cache.ccc_lru_lock);
        CFS_INIT_LIST_HEAD(&sbi->ll_cache.ccc_lru);
 
+       cfs_atomic_set(&sbi->ll_cache.ccc_unstable_nr, 0);
+       init_waitqueue_head(&sbi->ll_cache.ccc_unstable_waitq);
+
         sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
                                            SBI_DEFAULT_READAHEAD_MAX);
         sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
@@ -162,11 +165,6 @@ void ll_free_sbi(struct super_block *sb)
        EXIT;
 }
 
-static struct dentry_operations ll_d_root_ops = {
-        .d_compare = ll_dcompare,
-        .d_revalidate = ll_revalidate_nd,
-};
-
 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
                                     struct vfsmount *mnt)
 {
@@ -200,13 +198,6 @@ static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
                 RETURN(-ENOMEM);
         }
 
-        if (proc_lustre_fs_root) {
-                err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
-                                                  dt, md);
-                if (err < 0)
-                        CERROR("could not register mount in /proc/fs/lustre\n");
-        }
-
         /* indicate the features supported by this client */
         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
                                   OBD_CONNECT_ATTRFID  |
@@ -218,7 +209,9 @@ static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
                                   OBD_CONNECT_FULL20   | OBD_CONNECT_64BITHASH|
                                  OBD_CONNECT_EINPROGRESS |
                                  OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
-                                 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS;
+                                 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS |
+                                 OBD_CONNECT_MAX_EASIZE |
+                                 OBD_CONNECT_FLOCK_DEAD;
 
         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
@@ -394,6 +387,16 @@ static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
                sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
        }
 
+       if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
+               if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
+                       LCONSOLE_INFO("%s: disabling xattr cache due to "
+                                     "unknown maximum xattr size.\n", dt);
+               } else {
+                       sbi->ll_flags |= LL_SBI_XATTR_CACHE;
+                       sbi->ll_xattr_cache_enabled = 1;
+               }
+       }
+
        obd = class_name2obd(dt);
        if (!obd) {
                CERROR("DT %s: not setup or attached\n", dt);
@@ -578,13 +581,8 @@ static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
                        ll_get_fsname(sb, NULL, 0));
                GOTO(out_root, err = -ENOMEM);
        }
-
 #ifdef HAVE_DCACHE_LOCK
-       sb->s_root->d_op = &ll_d_root_ops;
-#else
-       /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
-       d_set_d_op(sb->s_root, &ll_d_root_ops);
-       sb->s_d_op = &ll_d_ops;
+       sb->s_root->d_op = &ll_d_ops;
 #endif
 
         sbi->ll_sdev_orig = sb->s_dev;
@@ -595,15 +593,22 @@ static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
         /* s_dev is also used in lt_compare() to compare two fs, but that is
          * only a node-local comparison. */
         uuid = obd_get_uuid(sbi->ll_md_exp);
-       if (uuid != NULL) {
+       if (uuid != NULL)
                sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
-               get_uuid2fsid(uuid->uuid, strlen(uuid->uuid), &sbi->ll_fsid);
-       }
 
         if (data != NULL)
                 OBD_FREE_PTR(data);
         if (osfs != NULL)
                 OBD_FREE_PTR(osfs);
+       if (proc_lustre_fs_root != NULL) {
+               err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
+                                                 dt, md);
+               if (err < 0) {
+                       CERROR("%s: could not register mount in lprocfs: "
+                              "rc = %d\n", ll_get_fsname(sb, NULL, 0), err);
+                       err = 0;
+               }
+       }
 
         RETURN(err);
 out_root:
@@ -626,7 +631,6 @@ out:
                 OBD_FREE_PTR(data);
         if (osfs != NULL)
                 OBD_FREE_PTR(osfs);
-        lprocfs_unregister_mountpoint(sbi);
         return err;
 }
 
@@ -654,8 +658,10 @@ void ll_dump_inode(struct inode *inode)
        ll_d_hlist_for_each(tmp, &inode->i_dentry)
                dentry_count++;
 
-       CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
-              inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
+       CERROR("%s: inode %p dump: dev=%s fid="DFID
+              " mode=%o count=%u, %d dentries\n",
+              ll_get_fsname(inode->i_sb, NULL, 0), inode,
+              ll_i2mdexp(inode)->exp_obd->obd_name, PFID(ll_inode2fid(inode)),
               inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 }
 
@@ -818,7 +824,7 @@ static int ll_options(char *options, int *flags)
                         *flags &= ~tmp;
                         goto next;
                 }
-#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 5, 50, 0)
+#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 5, 51, 0)
                 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
                 if (tmp) {
                         /* Ignore deprecated mount option.  The client will
@@ -947,6 +953,9 @@ void ll_lli_init(struct ll_inode_info *lli)
        lli->lli_layout_gen = LL_LAYOUT_GEN_NONE;
        lli->lli_clob = NULL;
 
+       init_rwsem(&lli->lli_xattrs_list_rwsem);
+       mutex_init(&lli->lli_xattrs_enq_lock);
+
        LASSERT(lli->lli_vfs_inode.i_mode != 0);
        if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
                mutex_init(&lli->lli_readdir_mutex);
@@ -999,15 +1008,15 @@ int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
         if (cfg == NULL)
                 RETURN(-ENOMEM);
 
-        cfs_module_get();
+       try_module_get(THIS_MODULE);
 
-        /* client additional sb info */
-        lsi->lsi_llsbi = sbi = ll_init_sbi();
-        if (!sbi) {
-                cfs_module_put(THIS_MODULE);
-                OBD_FREE_PTR(cfg);
-                RETURN(-ENOMEM);
-        }
+       /* client additional sb info */
+       lsi->lsi_llsbi = sbi = ll_init_sbi();
+       if (!sbi) {
+               module_put(THIS_MODULE);
+               OBD_FREE_PTR(cfg);
+               RETURN(-ENOMEM);
+       }
 
         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
         if (err)
@@ -1023,6 +1032,10 @@ int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
                GOTO(out_free, err);
 
         sb->s_bdi = &lsi->lsi_bdi;
+#ifndef HAVE_DCACHE_LOCK
+       /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
+       sb->s_d_op = &ll_d_ops;
+#endif
 
         /* Generate a string unique to this super, in case some joker tries
            to mount the same fs at two mount points.
@@ -1080,12 +1093,12 @@ void lu_context_keys_dump(void);
 
 void ll_put_super(struct super_block *sb)
 {
-        struct config_llog_instance cfg;
+       struct config_llog_instance cfg, params_cfg;
         struct obd_device *obd;
         struct lustre_sb_info *lsi = s2lsi(sb);
         struct ll_sb_info *sbi = ll_s2sbi(sb);
         char *profilenm = get_profile_name(sb);
-        int force = 1, next;
+       int ccc_count, next, force = 1, rc = 0;
         ENTRY;
 
         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
@@ -1095,12 +1108,28 @@ void ll_put_super(struct super_block *sb)
         cfg.cfg_instance = sb;
         lustre_end_log(sb, profilenm, &cfg);
 
+       params_cfg.cfg_instance = sb;
+       lustre_end_log(sb, PARAMS_FILENAME, &params_cfg);
+
         if (sbi->ll_md_exp) {
                 obd = class_exp2obd(sbi->ll_md_exp);
                 if (obd)
                         force = obd->obd_force;
         }
 
+       /* Wait for unstable pages to be committed to stable storage */
+       if (force == 0) {
+               struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
+               rc = l_wait_event(sbi->ll_cache.ccc_unstable_waitq,
+                       cfs_atomic_read(&sbi->ll_cache.ccc_unstable_nr) == 0,
+                       &lwi);
+       }
+
+       ccc_count = cfs_atomic_read(&sbi->ll_cache.ccc_unstable_nr);
+       if (force == 0 && rc != -EINTR)
+               LASSERTF(ccc_count == 0, "count: %i\n", ccc_count);
+
+
         /* We need to set force before the lov_disconnect in
            lustre_common_put_super, since l_d cleans up osc's as well. */
         if (force) {
@@ -1135,11 +1164,11 @@ void ll_put_super(struct super_block *sb)
         ll_free_sbi(sb);
         lsi->lsi_llsbi = NULL;
 
-        lustre_common_put_super(sb);
+       lustre_common_put_super(sb);
 
-        cfs_module_put(THIS_MODULE);
+       module_put(THIS_MODULE);
 
-        EXIT;
+       EXIT;
 } /* client_put_super */
 
 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
@@ -1195,8 +1224,8 @@ void ll_clear_inode(struct inode *inode)
         struct ll_sb_info *sbi = ll_i2sbi(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);
 
         if (S_ISDIR(inode->i_mode)) {
                 /* these should have been cleared in ll_file_release */
@@ -1227,6 +1256,8 @@ void ll_clear_inode(struct inode *inode)
                 lli->lli_symlink_name = NULL;
         }
 
+       ll_xattr_cache_destroy(inode);
+
         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
                 LASSERT(lli->lli_posix_acl == NULL);
                 if (lli->lli_remote_perms) {
@@ -1344,8 +1375,9 @@ static int ll_setattr_done_writing(struct inode *inode,
                  * from OSTs and send setattr to back to MDS. */
                 rc = ll_som_update(inode, op_data);
         } else if (rc) {
-                CERROR("inode %lu mdc truncate failed: rc = %d\n",
-                       inode->i_ino, rc);
+               CERROR("%s: inode "DFID" mdc truncate failed: rc = %d\n",
+                      ll_i2sbi(inode)->ll_md_exp->exp_obd->obd_name,
+                      PFID(ll_inode2fid(inode)), rc);
         }
         RETURN(rc);
 }
@@ -1382,21 +1414,24 @@ static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
  * at the same time.
+ *
+ * In case of HSMimport, we only set attr on MDS.
  */
-int ll_setattr_raw(struct dentry *dentry, struct iattr *attr)
+int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
 {
         struct inode *inode = dentry->d_inode;
         struct ll_inode_info *lli = ll_i2info(inode);
         struct md_op_data *op_data = NULL;
         struct md_open_data *mod = NULL;
        bool file_is_released = false;
-        int rc = 0, rc1 = 0;
-        ENTRY;
+       int rc = 0, rc1 = 0;
+       ENTRY;
 
-       CDEBUG(D_VFSTRACE, "%s: setattr inode %p/fid:"DFID" from %llu to %llu, "
-               "valid %x\n", ll_get_fsname(inode->i_sb, NULL, 0), inode,
-               PFID(&lli->lli_fid), i_size_read(inode), attr->ia_size,
-               attr->ia_valid);
+       CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
+              "valid %x, hsm_import %d\n",
+              ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
+              inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
+              hsm_import);
 
        if (attr->ia_valid & ATTR_SIZE) {
                 /* Check new size against VFS/VM file size limit and rlimit */
@@ -1417,12 +1452,12 @@ int ll_setattr_raw(struct dentry *dentry, struct iattr *attr)
                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
         }
 
-        /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
+       /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
        if (attr->ia_valid & TIMES_SET_FLAGS) {
-                if (cfs_curproc_fsuid() != inode->i_uid &&
-                    !cfs_capable(CFS_CAP_FOWNER))
-                        RETURN(-EPERM);
-        }
+               if (current_fsuid() != inode->i_uid &&
+                   !cfs_capable(CFS_CAP_FOWNER))
+                       RETURN(-EPERM);
+       }
 
         /* We mark all of the fields "set" so MDS/OST does not re-set them */
         if (attr->ia_valid & ATTR_CTIME) {
@@ -1445,14 +1480,6 @@ int ll_setattr_raw(struct dentry *dentry, struct iattr *attr)
                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
                        cfs_time_current_sec());
 
-       /* If we are changing file size, file content is modified, flag it. */
-       if (attr->ia_valid & ATTR_SIZE) {
-               attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
-               spin_lock(&lli->lli_lock);
-               lli->lli_flags |= LLIF_DATA_MODIFIED;
-               spin_unlock(&lli->lli_lock);
-       }
-
         /* We always do an MDS RPC, even if we're only changing the size;
          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
 
@@ -1467,13 +1494,6 @@ int ll_setattr_raw(struct dentry *dentry, struct iattr *attr)
                down_write(&lli->lli_trunc_sem);
        }
 
-        memcpy(&op_data->op_attr, attr, sizeof(*attr));
-
-        /* Open epoch for truncate. */
-        if (exp_connect_som(ll_i2mdexp(inode)) &&
-            (attr->ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
-                op_data->op_flags = MF_EPOCH_OPEN;
-
        /* truncate on a released file must failed with -ENODATA,
         * so size must not be set on MDS for released file
         * but other attributes must be set
@@ -1487,27 +1507,38 @@ int ll_setattr_raw(struct dentry *dentry, struct iattr *attr)
                if (lsm && lsm->lsm_pattern & LOV_PATTERN_F_RELEASED)
                        file_is_released = true;
                ccc_inode_lsm_put(inode, lsm);
-       }
 
-       /* clear size attr for released file
-        * we clear the attribute send to MDT in op_data, not the original
-        * received from caller in attr which is used later to
-        * decide return code */
-       if (file_is_released && (attr->ia_valid & ATTR_SIZE))
-               op_data->op_attr.ia_valid &= ~ATTR_SIZE;
+               if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
+                       if (file_is_released) {
+                               rc = ll_layout_restore(inode, 0, attr->ia_size);
+                               if (rc < 0)
+                                       GOTO(out, rc);
 
-        rc = ll_md_setattr(dentry, op_data, &mod);
-        if (rc)
-                GOTO(out, rc);
+                               file_is_released = false;
+                               ll_layout_refresh(inode, &gen);
+                       }
 
-       /* truncate failed, others succeed */
-       if (file_is_released) {
-               if (attr->ia_valid & ATTR_SIZE)
-                       GOTO(out, rc = -ENODATA);
-               else
-                       GOTO(out, rc = 0);
+                       /* If we are changing file size, file content is
+                        * modified, flag it. */
+                       attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
+                       spin_lock(&lli->lli_lock);
+                       lli->lli_flags |= LLIF_DATA_MODIFIED;
+                       spin_unlock(&lli->lli_lock);
+                       op_data->op_bias |= MDS_DATA_MODIFIED;
+               }
        }
 
+       memcpy(&op_data->op_attr, attr, sizeof(*attr));
+
+       /* Open epoch for truncate. */
+       if (exp_connect_som(ll_i2mdexp(inode)) && !hsm_import &&
+           (attr->ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
+               op_data->op_flags = MF_EPOCH_OPEN;
+
+       rc = ll_md_setattr(dentry, op_data, &mod);
+       if (rc)
+               GOTO(out, rc);
+
        /* RPC to MDT is sent, cancel data modification flag */
        if (rc == 0 && (op_data->op_bias & MDS_DATA_MODIFIED)) {
                spin_lock(&lli->lli_lock);
@@ -1515,35 +1546,35 @@ int ll_setattr_raw(struct dentry *dentry, struct iattr *attr)
                spin_unlock(&lli->lli_lock);
        }
 
-        ll_ioepoch_open(lli, op_data->op_ioepoch);
-       if (!S_ISREG(inode->i_mode))
-                GOTO(out, rc = 0);
+       ll_ioepoch_open(lli, op_data->op_ioepoch);
+       if (!S_ISREG(inode->i_mode) || file_is_released)
+               GOTO(out, rc = 0);
 
        if (attr->ia_valid & (ATTR_SIZE |
                              ATTR_ATIME | ATTR_ATIME_SET |
                              ATTR_MTIME | ATTR_MTIME_SET)) {
-                /* For truncate and utimes sending attributes to OSTs, setting
-                 * mtime/atime to the past will be performed under PW [0:EOF]
-                 * extent lock (new_size:EOF for truncate).  It may seem
-                 * excessive to send mtime/atime updates to OSTs when not
-                 * setting times to past, but it is necessary due to possible
-                 * time de-synchronization between MDT inode and OST objects */
-                rc = ll_setattr_ost(inode, attr);
+               /* For truncate and utimes sending attributes to OSTs, setting
+                * mtime/atime to the past will be performed under PW [0:EOF]
+                * extent lock (new_size:EOF for truncate).  It may seem
+                * excessive to send mtime/atime updates to OSTs when not
+                * setting times to past, but it is necessary due to possible
+                * time de-synchronization between MDT inode and OST objects */
+               rc = ll_setattr_ost(inode, attr);
        }
-        EXIT;
+       EXIT;
 out:
-        if (op_data) {
-                if (op_data->op_ioepoch) {
-                        rc1 = ll_setattr_done_writing(inode, op_data, mod);
-                        if (!rc)
-                                rc = rc1;
-                }
-                ll_finish_md_op_data(op_data);
-        }
+       if (op_data) {
+               if (op_data->op_ioepoch) {
+                       rc1 = ll_setattr_done_writing(inode, op_data, mod);
+                       if (!rc)
+                               rc = rc1;
+               }
+               ll_finish_md_op_data(op_data);
+       }
        if (!S_ISDIR(inode->i_mode)) {
                up_write(&lli->lli_trunc_sem);
                mutex_lock(&inode->i_mutex);
-               if (attr->ia_valid & ATTR_SIZE)
+               if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
                        inode_dio_wait(inode);
        }
 
@@ -1555,30 +1586,30 @@ out:
 
 int ll_setattr(struct dentry *de, struct iattr *attr)
 {
-        int mode = de->d_inode->i_mode;
-
-        if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
-                              (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
-                attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
-
-        if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
-                               (ATTR_SIZE|ATTR_MODE)) &&
-            (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
-             (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
-              !(attr->ia_mode & S_ISGID))))
-                attr->ia_valid |= ATTR_FORCE;
-
-        if ((mode & S_ISUID) &&
-            !(attr->ia_mode & S_ISUID) &&
-            !(attr->ia_valid & ATTR_KILL_SUID))
-                attr->ia_valid |= ATTR_KILL_SUID;
-
-        if (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
-            !(attr->ia_mode & S_ISGID) &&
-            !(attr->ia_valid & ATTR_KILL_SGID))
-                attr->ia_valid |= ATTR_KILL_SGID;
-
-        return ll_setattr_raw(de, attr);
+       int mode = de->d_inode->i_mode;
+
+       if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
+                             (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
+               attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
+
+       if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
+                              (ATTR_SIZE|ATTR_MODE)) &&
+           (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
+            (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
+             !(attr->ia_mode & S_ISGID))))
+               attr->ia_valid |= ATTR_FORCE;
+
+       if ((mode & S_ISUID) &&
+           !(attr->ia_mode & S_ISUID) &&
+           !(attr->ia_valid & ATTR_KILL_SUID))
+               attr->ia_valid |= ATTR_KILL_SUID;
+
+       if (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
+           !(attr->ia_mode & S_ISGID) &&
+           !(attr->ia_valid & ATTR_KILL_SGID))
+               attr->ia_valid |= ATTR_KILL_SGID;
+
+       return ll_setattr_raw(de, attr, false);
 }
 
 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
@@ -1634,6 +1665,7 @@ int ll_statfs(struct dentry *de, struct kstatfs *sfs)
 {
        struct super_block *sb = de->d_sb;
        struct obd_statfs osfs;
+       __u64 fsid = huge_encode_dev(sb->s_dev);
        int rc;
 
         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
@@ -1665,7 +1697,8 @@ int ll_statfs(struct dentry *de, struct kstatfs *sfs)
         sfs->f_blocks = osfs.os_blocks;
         sfs->f_bfree = osfs.os_bfree;
         sfs->f_bavail = osfs.os_bavail;
-       sfs->f_fsid = ll_s2sbi(sb)->ll_fsid;
+       sfs->f_fsid.val[0] = (__u32)fsid;
+       sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
        return 0;
 }
 
@@ -1772,9 +1805,9 @@ void ll_update_inode(struct inode *inode, struct lustre_md *md)
                 if (fid_is_sane(&lli->lli_fid)) {
                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
                                  "Trying to change FID "DFID
-                                 " to the "DFID", inode %lu/%u(%p)\n",
-                                 PFID(&lli->lli_fid), PFID(&body->fid1),
-                                 inode->i_ino, inode->i_generation, inode);
+                                " to the "DFID", inode "DFID"(%p)\n",
+                                PFID(&lli->lli_fid), PFID(&body->fid1),
+                                PFID(ll_inode2fid(inode)), inode);
                 } else
                         lli->lli_fid = body->fid1;
         }
@@ -1792,15 +1825,19 @@ void ll_update_inode(struct inode *inode, struct lustre_md *md)
                          * lock on the client and set LLIF_MDS_SIZE_LOCK holding
                          * it. */
                         mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
-                                               &lockh, LDLM_FL_CBPENDING);
+                                              &lockh, LDLM_FL_CBPENDING,
+                                              LCK_CR | LCK_CW |
+                                              LCK_PR | LCK_PW);
                         if (mode) {
                                 if (lli->lli_flags & (LLIF_DONE_WRITING |
                                                       LLIF_EPOCH_PENDING |
                                                       LLIF_SOM_DIRTY)) {
-                                        CERROR("ino %lu flags %u still has "
-                                               "size authority! do not trust "
-                                               "the size got from MDS\n",
-                                               inode->i_ino, lli->lli_flags);
+                                       CERROR("%s: inode "DFID" flags %u still"
+                                              " has size authority! do not "
+                                              "trust the size from MDS\n",
+                                              sbi->ll_md_exp->exp_obd->obd_name,
+                                              PFID(ll_inode2fid(inode)),
+                                              lli->lli_flags);
                                 } else {
                                         /* Use old size assignment to avoid
                                          * deadlock bz14138 & bz14326 */
@@ -1816,8 +1853,10 @@ void ll_update_inode(struct inode *inode, struct lustre_md *md)
                          * deadlock bz14138 & bz14326 */
                         i_size_write(inode, body->size);
 
-                        CDEBUG(D_VFSTRACE, "inode=%lu, updating i_size %llu\n",
-                               inode->i_ino, (unsigned long long)body->size);
+                       CDEBUG(D_VFSTRACE,
+                              "inode="DFID", updating i_size %llu\n",
+                              PFID(ll_inode2fid(inode)),
+                              (unsigned long long)body->size);
                 }
 
                 if (body->valid & OBD_MD_FLBLOCKS)
@@ -1904,12 +1943,12 @@ void ll_delete_inode(struct inode *inode)
 
         /* Workaround for LU-118 */
         if (inode->i_data.nrpages) {
-                TREE_READ_LOCK_IRQ(&inode->i_data);
-                TREE_READ_UNLOCK_IRQ(&inode->i_data);
+               spin_lock_irq(&inode->i_data.tree_lock);
+               spin_unlock_irq(&inode->i_data.tree_lock);
                 LASSERTF(inode->i_data.nrpages == 0,
-                         "inode=%lu/%u(%p) nrpages=%lu, see "
-                         "http://jira.whamcloud.com/browse/LU-118\n",
-                         inode->i_ino, inode->i_generation, inode,
+                        "inode="DFID"(%p) nrpages=%lu, see "
+                        "http://jira.whamcloud.com/browse/LU-118\n",
+                        PFID(ll_inode2fid(inode)), inode,
                          inode->i_data.nrpages);
         }
         /* Workaround end */
@@ -1945,7 +1984,9 @@ int ll_iocontrol(struct inode *inode, struct file *file,
                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
                 ll_finish_md_op_data(op_data);
                 if (rc) {
-                        CERROR("failure %d inode %lu\n", rc, inode->i_ino);
+                       CERROR("%s: failure inode "DFID": rc = %d\n",
+                              sbi->ll_md_exp->exp_obd->obd_name,
+                              PFID(ll_inode2fid(inode)), rc);
                         RETURN(-abs(rc));
                 }
 
@@ -2018,17 +2059,17 @@ int ll_iocontrol(struct inode *inode, struct file *file,
 
 int ll_flush_ctx(struct inode *inode)
 {
-        struct ll_sb_info  *sbi = ll_i2sbi(inode);
+       struct ll_sb_info  *sbi = ll_i2sbi(inode);
 
-        CDEBUG(D_SEC, "flush context for user %d\n", cfs_curproc_uid());
+       CDEBUG(D_SEC, "flush context for user %d\n", current_uid());
 
-        obd_set_info_async(NULL, sbi->ll_md_exp,
-                           sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
-                           0, NULL, NULL);
-        obd_set_info_async(NULL, sbi->ll_dt_exp,
-                           sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
-                           0, NULL, NULL);
-        return 0;
+       obd_set_info_async(NULL, sbi->ll_md_exp,
+                          sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
+                          0, NULL, NULL);
+       obd_set_info_async(NULL, sbi->ll_dt_exp,
+                          sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
+                          0, NULL, NULL);
+       return 0;
 }
 
 /* umount -f client means force down, don't save state */
@@ -2071,12 +2112,12 @@ void ll_umount_begin(struct super_block *sb)
                OBD_FREE_PTR(ioc_data);
        }
 
-        /* Really, we'd like to wait until there are no requests outstanding,
-         * and then continue.  For now, we just invalidate the requests,
-         * schedule() and sleep one second if needed, and hope.
-         */
-        cfs_schedule();
-        EXIT;
+       /* Really, we'd like to wait until there are no requests outstanding,
+        * and then continue.  For now, we just invalidate the requests,
+        * schedule() and sleep one second if needed, and hope.
+        */
+       schedule();
+       EXIT;
 }
 
 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
@@ -2297,8 +2338,8 @@ struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
        op_data->op_namelen = namelen;
        op_data->op_mode = mode;
        op_data->op_mod_time = cfs_time_current_sec();
-       op_data->op_fsuid = cfs_curproc_fsuid();
-       op_data->op_fsgid = cfs_curproc_fsgid();
+       op_data->op_fsuid = current_fsuid();
+       op_data->op_fsgid = current_fsgid();
        op_data->op_cap = cfs_curproc_cap_pack();
        op_data->op_bias = 0;
        op_data->op_cli_flags = 0;
@@ -2437,7 +2478,6 @@ static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
 {
        char *path = NULL;
 
-#ifdef HAVE_FS_STRUCT_USE_PATH
        struct path p;
 
        p.dentry = dentry;
@@ -2445,10 +2485,6 @@ static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
        path_get(&p);
        path = d_path(&p, buf, bufsize);
        path_put(&p);
-#else
-       path = d_path(dentry, current->fs->rootmnt, buf, bufsize);
-#endif
-
        return path;
 }