Whamcloud - gitweb
LU-12809 llite: statfs to use NODELAY with MDS
[fs/lustre-release.git] / lustre / llite / llite_lib.c
index 7f696c8..83e9490 100644 (file)
 #include <linux/security.h>
 
 #include <uapi/linux/lustre/lustre_ioctl.h>
+#ifdef HAVE_UAPI_LINUX_MOUNT_H
+#include <uapi/linux/mount.h>
+#endif
+
 #include <lustre_ha.h>
 #include <lustre_dlm.h>
 #include <lprocfs_status.h>
@@ -73,32 +77,46 @@ static struct ll_sb_info *ll_init_sbi(void)
        unsigned long pages;
        unsigned long lru_page_max;
        struct sysinfo si;
+       int rc;
        int i;
+
        ENTRY;
 
        OBD_ALLOC_PTR(sbi);
        if (sbi == NULL)
-               RETURN(NULL);
+               RETURN(ERR_PTR(-ENOMEM));
+
+       rc = pcc_super_init(&sbi->ll_pcc_super);
+       if (rc < 0)
+               GOTO(out_sbi, rc);
 
        spin_lock_init(&sbi->ll_lock);
        mutex_init(&sbi->ll_lco.lco_lock);
        spin_lock_init(&sbi->ll_pp_extent_lock);
        spin_lock_init(&sbi->ll_process_lock);
         sbi->ll_rw_stats_on = 0;
+       sbi->ll_statfs_max_age = OBD_STATFS_CACHE_SECONDS;
 
         si_meminfo(&si);
         pages = si.totalram - si.totalhigh;
        lru_page_max = pages / 2;
 
+       sbi->ll_ra_info.ra_async_max_active = 0;
+       sbi->ll_ra_info.ll_readahead_wq =
+               alloc_workqueue("ll-readahead-wq", WQ_UNBOUND,
+                               sbi->ll_ra_info.ra_async_max_active);
+       if (!sbi->ll_ra_info.ll_readahead_wq)
+               GOTO(out_pcc, rc = -ENOMEM);
+
        /* initialize ll_cache data */
        sbi->ll_cache = cl_cache_init(lru_page_max);
-       if (sbi->ll_cache == NULL) {
-               OBD_FREE(sbi, sizeof(*sbi));
-               RETURN(NULL);
-       }
+       if (sbi->ll_cache == NULL)
+               GOTO(out_destroy_ra, rc = -ENOMEM);
 
        sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
                                           SBI_DEFAULT_READAHEAD_MAX);
+       sbi->ll_ra_info.ra_async_pages_per_file_threshold =
+                               sbi->ll_ra_info.ra_max_pages_per_file;
        sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
        sbi->ll_ra_info.ra_max_read_ahead_whole_pages = -1;
 
@@ -137,13 +155,19 @@ static struct ll_sb_info *ll_init_sbi(void)
        sbi->ll_squash.rsi_uid = 0;
        sbi->ll_squash.rsi_gid = 0;
        INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
-       init_rwsem(&sbi->ll_squash.rsi_sem);
-       pcc_super_init(&sbi->ll_pcc_super);
+       spin_lock_init(&sbi->ll_squash.rsi_lock);
 
        /* Per-filesystem file heat */
        sbi->ll_heat_decay_weight = SBI_DEFAULT_HEAT_DECAY_WEIGHT;
        sbi->ll_heat_period_second = SBI_DEFAULT_HEAT_PERIOD_SECOND;
        RETURN(sbi);
+out_destroy_ra:
+       destroy_workqueue(sbi->ll_ra_info.ll_readahead_wq);
+out_pcc:
+       pcc_super_fini(&sbi->ll_pcc_super);
+out_sbi:
+       OBD_FREE_PTR(sbi);
+       RETURN(ERR_PTR(rc));
 }
 
 static void ll_free_sbi(struct super_block *sb)
@@ -154,6 +178,8 @@ static void ll_free_sbi(struct super_block *sb)
        if (sbi != NULL) {
                if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
                        cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
+               if (sbi->ll_ra_info.ll_readahead_wq)
+                       destroy_workqueue(sbi->ll_ra_info.ll_readahead_wq);
                if (sbi->ll_cache != NULL) {
                        cl_cache_decref(sbi->ll_cache);
                        sbi->ll_cache = NULL;
@@ -229,6 +255,7 @@ static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
                                   OBD_CONNECT2_FLR |
                                   OBD_CONNECT2_LOCK_CONVERT |
                                   OBD_CONNECT2_ARCHIVE_ID_ARRAY |
+                                  OBD_CONNECT2_INC_XID |
                                   OBD_CONNECT2_LSOM |
                                   OBD_CONNECT2_ASYNC_DISCARD |
                                   OBD_CONNECT2_PCC;
@@ -237,7 +264,7 @@ static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 #endif
-#ifdef CONFIG_FS_POSIX_ACL
+#ifdef CONFIG_LUSTRE_FS_POSIX_ACL
        data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK |
                                   OBD_CONNECT_LARGE_ACL;
 #endif
@@ -247,29 +274,29 @@ static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
        if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
                /* flag mdc connection as lightweight, only used for test
                 * purpose, use with care */
-                data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
+               data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
 
-        data->ocd_ibits_known = MDS_INODELOCK_FULL;
-        data->ocd_version = LUSTRE_VERSION_CODE;
+       data->ocd_ibits_known = MDS_INODELOCK_FULL;
+       data->ocd_version = LUSTRE_VERSION_CODE;
 
-        if (sb->s_flags & MS_RDONLY)
-                data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
-        if (sbi->ll_flags & LL_SBI_USER_XATTR)
-                data->ocd_connect_flags |= OBD_CONNECT_XATTR;
+       if (sb->s_flags & SB_RDONLY)
+               data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
+       if (sbi->ll_flags & LL_SBI_USER_XATTR)
+               data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 
-#ifdef MS_NOSEC
+#ifdef SB_NOSEC
        /* Setting this indicates we correctly support S_NOSEC (See kernel
         * commit 9e1f1de02c2275d7172e18dc4e7c2065777611bf)
         */
-       sb->s_flags |= MS_NOSEC;
+       sb->s_flags |= SB_NOSEC;
 #endif
 
-        if (sbi->ll_flags & LL_SBI_FLOCK)
-                sbi->ll_fop = &ll_file_operations_flock;
-        else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
-                sbi->ll_fop = &ll_file_operations;
-        else
-                sbi->ll_fop = &ll_file_operations_noflock;
+       if (sbi->ll_flags & LL_SBI_FLOCK)
+               sbi->ll_fop = &ll_file_operations_flock;
+       else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
+               sbi->ll_fop = &ll_file_operations;
+       else
+               sbi->ll_fop = &ll_file_operations_noflock;
 
        /* always ping even if server suppress_pings */
        if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
@@ -285,16 +312,16 @@ static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
 
        err = obd_connect(NULL, &sbi->ll_md_exp, sbi->ll_md_obd,
                          &sbi->ll_sb_uuid, data, sbi->ll_cache);
-        if (err == -EBUSY) {
-                LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
-                                   "recovery, of which this client is not a "
-                                   "part. Please wait for recovery to complete,"
-                                   " abort, or time out.\n", md);
-                GOTO(out, err);
-        } else if (err) {
-                CERROR("cannot connect to %s: rc = %d\n", md, err);
-                GOTO(out, err);
-        }
+       if (err == -EBUSY) {
+               LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
+                                  "recovery, of which this client is not a "
+                                  "part. Please wait for recovery to complete,"
+                                  " abort, or time out.\n", md);
+               GOTO(out, err);
+       } else if (err) {
+               CERROR("cannot connect to %s: rc = %d\n", md, err);
+               GOTO(out, err);
+       }
 
        sbi->ll_md_exp->exp_connect_data = *data;
 
@@ -310,7 +337,7 @@ static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
         * can make sure the client can be mounted as long as MDT0 is
         * avaible */
        err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
-                       ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
+                       ktime_get_seconds() - sbi->ll_statfs_max_age,
                        OBD_STATFS_FOR_MDT0);
        if (err)
                GOTO(out_md_fid, err);
@@ -356,28 +383,28 @@ static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
        sbi->ll_namelen = osfs->os_namelen;
        sbi->ll_mnt.mnt = current->fs->root.mnt;
 
-        if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
-            !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
-                LCONSOLE_INFO("Disabling user_xattr feature because "
-                              "it is not supported on the server\n");
-                sbi->ll_flags &= ~LL_SBI_USER_XATTR;
-        }
+       if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
+           !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
+               LCONSOLE_INFO("Disabling user_xattr feature because "
+                             "it is not supported on the server\n");
+               sbi->ll_flags &= ~LL_SBI_USER_XATTR;
+       }
 
-        if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
-#ifdef MS_POSIXACL
-                sb->s_flags |= MS_POSIXACL;
+       if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
+#ifdef SB_POSIXACL
+               sb->s_flags |= SB_POSIXACL;
 #endif
-                sbi->ll_flags |= LL_SBI_ACL;
-        } else {
-                LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
-#ifdef MS_POSIXACL
-                sb->s_flags &= ~MS_POSIXACL;
+               sbi->ll_flags |= LL_SBI_ACL;
+       } else {
+               LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
+#ifdef SB_POSIXACL
+               sb->s_flags &= ~SB_POSIXACL;
 #endif
-                sbi->ll_flags &= ~LL_SBI_ACL;
-        }
+               sbi->ll_flags &= ~LL_SBI_ACL;
+       }
 
-        if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
-                sbi->ll_flags |= LL_SBI_64BIT_HASH;
+       if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
+               sbi->ll_flags |= LL_SBI_64BIT_HASH;
 
        if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
                sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
@@ -438,7 +465,8 @@ static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
        data->ocd_connect_flags |= OBD_CONNECT_LOCKAHEAD_OLD;
 #endif
 
-       data->ocd_connect_flags2 = OBD_CONNECT2_LOCKAHEAD;
+       data->ocd_connect_flags2 = OBD_CONNECT2_LOCKAHEAD |
+                                  OBD_CONNECT2_INC_XID;
 
        if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
                data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
@@ -528,9 +556,7 @@ static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
        CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 
        sb->s_op = &lustre_super_operations;
-#ifdef HAVE_XATTR_HANDLER_FLAGS
        sb->s_xattr = ll_xattr_handlers;
-#endif
 #if THREAD_SIZE >= 8192 /*b=17630*/
        sb->s_export_op = &lustre_export_operations;
 #endif
@@ -574,7 +600,7 @@ static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
        ptlrpc_req_finished(request);
 
        if (IS_ERR(root)) {
-#ifdef CONFIG_FS_POSIX_ACL
+#ifdef CONFIG_LUSTRE_FS_POSIX_ACL
                if (lmd.posix_acl) {
                        posix_acl_release(lmd.posix_acl);
                        lmd.posix_acl = NULL;
@@ -582,7 +608,8 @@ static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
 #endif
                err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
                root = NULL;
-               CERROR("lustre_lite: bad iget4 for root\n");
+               CERROR("%s: bad ll_iget() for root: rc = %d\n",
+                      sbi->ll_fsname, err);
                GOTO(out_root, err);
        }
 
@@ -606,9 +633,6 @@ static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
                       sbi->ll_fsname, err);
                GOTO(out_root, err);
        }
-#ifdef HAVE_DCACHE_LOCK
-       sb->s_root->d_op = &ll_d_ops;
-#endif
 
        sbi->ll_sdev_orig = sb->s_dev;
 
@@ -773,8 +797,8 @@ void ll_kill_super(struct super_block *sb)
        struct ll_sb_info *sbi;
        ENTRY;
 
-        /* not init sb ?*/
-       if (!(sb->s_flags & MS_ACTIVE))
+       /* not init sb ?*/
+       if (!(sb->s_flags & SB_ACTIVE))
                return;
 
        sbi = ll_s2sbi(sb);
@@ -783,12 +807,11 @@ void ll_kill_super(struct super_block *sb)
         * put_super not affected real removing devices */
        if (sbi) {
                sb->s_dev = sbi->ll_sdev_orig;
-               sbi->ll_umounting = 1;
 
                /* wait running statahead threads to quit */
                while (atomic_read(&sbi->ll_sa_running) > 0) {
                        set_current_state(TASK_UNINTERRUPTIBLE);
-                       schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC >> 3));
+                       schedule_timeout(cfs_time_seconds(1) >> 3);
                }
        }
 
@@ -816,38 +839,38 @@ static int ll_options(char *options, struct ll_sb_info *sbi)
 
        CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 
-        while (*s1) {
-                CDEBUG(D_SUPER, "next opt=%s\n", s1);
-                tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
-                if (tmp) {
-                        *flags |= tmp;
-                        goto next;
-                }
-                tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
-                if (tmp) {
-                        *flags |= tmp;
-                        goto next;
-                }
-                tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
-                if (tmp) {
-                        *flags |= tmp;
-                        goto next;
-                }
-                tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
-                if (tmp) {
-                        *flags &= ~tmp;
-                        goto next;
-                }
-                tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
-                if (tmp) {
-                        *flags |= tmp;
-                        goto next;
-                }
-                tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
-                if (tmp) {
-                        *flags &= ~tmp;
-                        goto next;
-                }
+       while (*s1) {
+               CDEBUG(D_SUPER, "next opt=%s\n", s1);
+               tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
+               if (tmp) {
+                       *flags |= tmp;
+                       goto next;
+               }
+               tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
+               if (tmp) {
+                       *flags = (*flags & ~LL_SBI_LOCALFLOCK) | tmp;
+                       goto next;
+               }
+               tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
+               if (tmp) {
+                       *flags = (*flags & ~LL_SBI_FLOCK) | tmp;
+                       goto next;
+               }
+               tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
+               if (tmp) {
+                       *flags &= ~tmp;
+                       goto next;
+               }
+               tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
+               if (tmp) {
+                       *flags |= tmp;
+                       goto next;
+               }
+               tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
+               if (tmp) {
+                       *flags &= ~tmp;
+                       goto next;
+               }
                tmp = ll_set_opt("context", s1, 1);
                if (tmp)
                        goto next;
@@ -962,13 +985,11 @@ void ll_lli_init(struct ll_inode_info *lli)
 
        LASSERT(lli->lli_vfs_inode.i_mode != 0);
        if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
-               mutex_init(&lli->lli_readdir_mutex);
                lli->lli_opendir_key = NULL;
                lli->lli_sai = NULL;
                spin_lock_init(&lli->lli_sa_lock);
                lli->lli_opendir_pid = 0;
                lli->lli_sa_enabled = 0;
-               lli->lli_def_stripe_offset = -1;
                init_rwsem(&lli->lli_lsm_sem);
        } else {
                mutex_init(&lli->lli_size_mutex);
@@ -986,6 +1007,9 @@ void ll_lli_init(struct ll_inode_info *lli)
                mutex_init(&lli->lli_pcc_lock);
                lli->lli_pcc_state = PCC_STATE_FL_NONE;
                lli->lli_pcc_inode = NULL;
+               mutex_init(&lli->lli_group_mutex);
+               lli->lli_group_users = 0;
+               lli->lli_group_gid = 0;
        }
        mutex_init(&lli->lli_layout_mutex);
        memset(lli->lli_jobid, 0, sizeof(lli->lli_jobid));
@@ -1059,17 +1083,16 @@ int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
 
        /* client additional sb info */
        lsi->lsi_llsbi = sbi = ll_init_sbi();
-       if (!sbi)
-               GOTO(out_free_cfg, err = -ENOMEM);
+       if (IS_ERR(sbi))
+               GOTO(out_free_cfg, err = PTR_ERR(sbi));
 
        err = ll_options(lsi->lsi_lmd->lmd_opts, sbi);
        if (err)
                GOTO(out_free_cfg, err);
 
-#ifndef HAVE_DCACHE_LOCK
        /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
        sb->s_d_op = &ll_d_ops;
-#endif
+
        /* UUID handling */
        generate_random_uuid(uuid.b);
        snprintf(sbi->ll_sb_uuid.uuid, UUID_SIZE, "%pU", uuid.b);
@@ -1190,7 +1213,7 @@ void ll_put_super(struct super_block *sb)
        int next, force = 1, rc = 0;
        ENTRY;
 
-       if (!sbi)
+       if (IS_ERR(sbi))
                GOTO(out_no_sbi, 0);
 
        /* Should replace instance_id with something better for ASLR */
@@ -1297,10 +1320,15 @@ void ll_dir_clear_lsm_md(struct inode *inode)
 
        LASSERT(S_ISDIR(inode->i_mode));
 
-       if (lli->lli_lsm_md != NULL) {
+       if (lli->lli_lsm_md) {
                lmv_free_memmd(lli->lli_lsm_md);
                lli->lli_lsm_md = NULL;
        }
+
+       if (lli->lli_default_lsm_md) {
+               lmv_free_memmd(lli->lli_default_lsm_md);
+               lli->lli_default_lsm_md = NULL;
+       }
 }
 
 static struct inode *ll_iget_anon_dir(struct super_block *sb,
@@ -1369,6 +1397,9 @@ static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
               ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid));
        lsm_md_dump(D_INODE, lsm);
 
+       if (!lmv_dir_striped(lsm))
+               goto out;
+
        /* XXX sigh, this lsm_root initialization should be in
         * LMV layer, but it needs ll_iget right now, so we
         * put this here right now. */
@@ -1396,12 +1427,52 @@ static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
                        return rc;
                }
        }
-
+out:
        lli->lli_lsm_md = lsm;
 
        return 0;
 }
 
+static void ll_update_default_lsm_md(struct inode *inode, struct lustre_md *md)
+{
+       struct ll_inode_info *lli = ll_i2info(inode);
+
+       if (!md->default_lmv) {
+               /* clear default lsm */
+               if (lli->lli_default_lsm_md) {
+                       down_write(&lli->lli_lsm_sem);
+                       if (lli->lli_default_lsm_md) {
+                               lmv_free_memmd(lli->lli_default_lsm_md);
+                               lli->lli_default_lsm_md = NULL;
+                       }
+                       up_write(&lli->lli_lsm_sem);
+               }
+       } else if (lli->lli_default_lsm_md) {
+               /* update default lsm if it changes */
+               down_read(&lli->lli_lsm_sem);
+               if (lli->lli_default_lsm_md &&
+                   !lsm_md_eq(lli->lli_default_lsm_md, md->default_lmv)) {
+                       up_read(&lli->lli_lsm_sem);
+                       down_write(&lli->lli_lsm_sem);
+                       if (lli->lli_default_lsm_md)
+                               lmv_free_memmd(lli->lli_default_lsm_md);
+                       lli->lli_default_lsm_md = md->default_lmv;
+                       lsm_md_dump(D_INODE, md->default_lmv);
+                       md->default_lmv = NULL;
+                       up_write(&lli->lli_lsm_sem);
+               } else {
+                       up_read(&lli->lli_lsm_sem);
+               }
+       } else {
+               /* init default lsm */
+               down_write(&lli->lli_lsm_sem);
+               lli->lli_default_lsm_md = md->default_lmv;
+               lsm_md_dump(D_INODE, md->default_lmv);
+               md->default_lmv = NULL;
+               up_write(&lli->lli_lsm_sem);
+       }
+}
+
 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
 {
        struct ll_inode_info *lli = ll_i2info(inode);
@@ -1414,6 +1485,10 @@ static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
        CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
               PFID(ll_inode2fid(inode)));
 
+       /* update default LMV */
+       if (md->default_lmv)
+               ll_update_default_lsm_md(inode, md);
+
        /*
         * no striped information from request, lustre_md from req does not
         * include stripeEA, see ll_md_setattr()
@@ -1436,10 +1511,9 @@ static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
         *
         * foreign LMV should not change.
         */
-       if (lli->lli_lsm_md &&
-           lli->lli_lsm_md->lsm_md_magic != LMV_MAGIC_FOREIGN &&
-          !lsm_md_eq(lli->lli_lsm_md, lsm)) {
-               if (lsm->lsm_md_layout_version <=
+       if (lli->lli_lsm_md && !lsm_md_eq(lli->lli_lsm_md, lsm)) {
+               if (lmv_dir_striped(lli->lli_lsm_md) &&
+                   lsm->lsm_md_layout_version <=
                    lli->lli_lsm_md->lsm_md_layout_version) {
                        CERROR("%s: "DFID" dir layout mismatch:\n",
                               ll_i2sbi(inode)->ll_fsname,
@@ -1459,15 +1533,6 @@ static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
        if (!lli->lli_lsm_md) {
                struct cl_attr  *attr;
 
-               if (lsm->lsm_md_magic == LMV_MAGIC_FOREIGN) {
-                       /* set md->lmv to NULL, so the following free lustre_md
-                        * will not free this lsm */
-                       md->lmv = NULL;
-                       lli->lli_lsm_md = lsm;
-                       up_write(&lli->lli_lsm_sem);
-                       RETURN(0);
-               }
-
                rc = ll_init_lsm_md(inode, md);
                up_write(&lli->lli_lsm_sem);
                if (rc != 0)
@@ -1483,6 +1548,9 @@ static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
                 */
                down_read(&lli->lli_lsm_sem);
 
+               if (!lmv_dir_striped(lli->lli_lsm_md))
+                       GOTO(unlock, rc);
+
                OBD_ALLOC_PTR(attr);
                if (attr == NULL)
                        GOTO(unlock, rc = -ENOMEM);
@@ -1554,7 +1622,7 @@ void ll_clear_inode(struct inode *inode)
 
        ll_xattr_cache_destroy(inode);
 
-#ifdef CONFIG_FS_POSIX_ACL
+#ifdef CONFIG_LUSTRE_FS_POSIX_ACL
        forget_all_cached_acls(inode);
        if (lli->lli_posix_acl) {
                posix_acl_release(lli->lli_posix_acl);
@@ -1656,6 +1724,7 @@ int ll_setattr_raw(struct dentry *dentry, struct iattr *attr,
         struct inode *inode = dentry->d_inode;
         struct ll_inode_info *lli = ll_i2info(inode);
         struct md_op_data *op_data = NULL;
+       ktime_t kstart = ktime_get();
        int rc = 0;
 
        ENTRY;
@@ -1828,8 +1897,10 @@ out:
                inode_has_no_xattr(inode);
        }
 
-       ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
-                       LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
+       if (!rc)
+               ll_stats_ops_tally(ll_i2sbi(inode), attr->ia_valid & ATTR_SIZE ?
+                                       LPROC_LL_TRUNC : LPROC_LL_SETATTR,
+                                  ktime_us_delta(ktime_get(), kstart));
 
        return rc;
 }
@@ -1873,7 +1944,10 @@ int ll_statfs_internal(struct ll_sb_info *sbi, struct obd_statfs *osfs,
        int rc;
 
        ENTRY;
-       max_age = ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS;
+       max_age = ktime_get_seconds() - sbi->ll_statfs_max_age;
+
+       if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
+               flags |= OBD_STATFS_NODELAY;
 
        rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
        if (rc)
@@ -1887,9 +1961,6 @@ int ll_statfs_internal(struct ll_sb_info *sbi, struct obd_statfs *osfs,
        if (osfs->os_state & OS_STATE_SUM)
                GOTO(out, rc);
 
-       if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
-               flags |= OBD_STATFS_NODELAY;
-
        rc = obd_statfs(NULL, sbi->ll_dt_exp, &obd_osfs, max_age, flags);
        if (rc) /* Possibly a filesystem with no OSTs.  Report MDT totals. */
                GOTO(out, rc = 0);
@@ -1917,15 +1988,16 @@ int ll_statfs_internal(struct ll_sb_info *sbi, struct obd_statfs *osfs,
 out:
        RETURN(rc);
 }
+
 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);
+       ktime_t kstart = ktime_get();
        int rc;
 
-       CDEBUG(D_VFSTRACE, "VFS Op: at %llu jiffies\n", get_jiffies_64());
-        ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
+       CDEBUG(D_VFSTRACE, "VFS Op:sb=%s (%p)\n", sb->s_id, sb);
 
        /* Some amount of caching on the client is allowed */
        rc = ll_statfs_internal(ll_s2sbi(sb), &osfs, OBD_STATFS_SUM);
@@ -1948,11 +2020,15 @@ 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_blocks = osfs.os_blocks;
+       sfs->f_bfree = osfs.os_bfree;
+       sfs->f_bavail = osfs.os_bavail;
        sfs->f_fsid.val[0] = (__u32)fsid;
        sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
+
+       ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STATFS,
+                          ktime_us_delta(ktime_get(), kstart));
+
        return 0;
 }
 
@@ -2002,7 +2078,7 @@ int ll_update_inode(struct inode *inode, struct lustre_md *md)
                        return rc;
        }
 
-#ifdef CONFIG_FS_POSIX_ACL
+#ifdef CONFIG_LUSTRE_FS_POSIX_ACL
        if (body->mbo_valid & OBD_MD_FLACL) {
                spin_lock(&lli->lli_lock);
                if (lli->lli_posix_acl)
@@ -2166,6 +2242,8 @@ void ll_delete_inode(struct inode *inode)
        struct ll_inode_info *lli = ll_i2info(inode);
        struct address_space *mapping = &inode->i_data;
        unsigned long nrpages;
+       unsigned long flags;
+
        ENTRY;
 
        if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL) {
@@ -2190,9 +2268,9 @@ void ll_delete_inode(struct inode *inode)
         */
        nrpages = mapping->nrpages;
        if (nrpages) {
-               xa_lock_irq(&mapping->i_pages);
+               xa_lock_irqsave(&mapping->i_pages, flags);
                nrpages = mapping->nrpages;
-               xa_unlock_irq(&mapping->i_pages);
+               xa_unlock_irqrestore(&mapping->i_pages, flags);
        } /* Workaround end */
 
        LASSERTF(nrpages == 0, "%s: inode="DFID"(%p) nrpages=%lu, "
@@ -2200,9 +2278,7 @@ void ll_delete_inode(struct inode *inode)
                 ll_i2sbi(inode)->ll_fsname,
                 PFID(ll_inode2fid(inode)), inode, nrpages);
 
-#ifdef HAVE_SBOPS_EVICT_INODE
        ll_clear_inode(inode);
-#endif
        clear_inode(inode);
 
         EXIT;
@@ -2335,16 +2411,16 @@ void ll_umount_begin(struct super_block *sb)
        }
        obd->obd_force = 1;
 
-        obd = class_exp2obd(sbi->ll_dt_exp);
-        if (obd == NULL) {
+       obd = class_exp2obd(sbi->ll_dt_exp);
+       if (obd == NULL) {
                CERROR("Invalid LOV connection handle %#llx\n",
-                       sbi->ll_dt_exp->exp_handle.h_cookie);
-                EXIT;
-                return;
-        }
-        obd->obd_force = 1;
+                      sbi->ll_dt_exp->exp_handle.h_cookie);
+               EXIT;
+               return;
+       }
+       obd->obd_force = 1;
 
-        OBD_ALLOC_PTR(ioc_data);
+       OBD_ALLOC_PTR(ioc_data);
        if (ioc_data) {
                obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
                              sizeof *ioc_data, ioc_data, NULL);
@@ -2369,34 +2445,34 @@ void ll_umount_begin(struct super_block *sb)
 
 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
 {
-        struct ll_sb_info *sbi = ll_s2sbi(sb);
-        char *profilenm = get_profile_name(sb);
-        int err;
-        __u32 read_only;
-
-        if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
-                read_only = *flags & MS_RDONLY;
-                err = obd_set_info_async(NULL, sbi->ll_md_exp,
-                                         sizeof(KEY_READ_ONLY),
-                                         KEY_READ_ONLY, sizeof(read_only),
-                                         &read_only, NULL);
-                if (err) {
-                        LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
-                                      profilenm, read_only ?
-                                      "read-only" : "read-write", err);
-                        return err;
-                }
+       struct ll_sb_info *sbi = ll_s2sbi(sb);
+       char *profilenm = get_profile_name(sb);
+       int err;
+       __u32 read_only;
+
+       if ((*flags & MS_RDONLY) != (sb->s_flags & SB_RDONLY)) {
+               read_only = *flags & MS_RDONLY;
+               err = obd_set_info_async(NULL, sbi->ll_md_exp,
+                                        sizeof(KEY_READ_ONLY),
+                                        KEY_READ_ONLY, sizeof(read_only),
+                                        &read_only, NULL);
+               if (err) {
+                       LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
+                                     profilenm, read_only ?
+                                     "read-only" : "read-write", err);
+                       return err;
+               }
 
-                if (read_only)
-                        sb->s_flags |= MS_RDONLY;
-                else
-                        sb->s_flags &= ~MS_RDONLY;
+               if (read_only)
+                       sb->s_flags |= SB_RDONLY;
+               else
+                       sb->s_flags &= ~SB_RDONLY;
 
-                if (sbi->ll_flags & LL_SBI_VERBOSE)
-                        LCONSOLE_WARN("Remounted %s %s\n", profilenm,
-                                      read_only ?  "read-only" : "read-write");
-        }
-        return 0;
+               if (sbi->ll_flags & LL_SBI_VERBOSE)
+                       LCONSOLE_WARN("Remounted %s %s\n", profilenm,
+                                     read_only ?  "read-only" : "read-write");
+       }
+       return 0;
 }
 
 /**
@@ -2446,7 +2522,9 @@ int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
 {
        struct ll_sb_info *sbi = NULL;
        struct lustre_md md = { NULL };
+       bool default_lmv_deleted = false;
        int rc;
+
        ENTRY;
 
        LASSERT(*inode || sb);
@@ -2456,6 +2534,15 @@ int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
        if (rc != 0)
                GOTO(out, rc);
 
+       /*
+        * clear default_lmv only if intent_getattr reply doesn't contain it.
+        * but it needs to be done after iget, check this early because
+        * ll_update_lsm_md() may change md.
+        */
+       if (it && (it->it_op & (IT_LOOKUP | IT_GETATTR)) &&
+           S_ISDIR(md.body->mbo_mode) && !md.default_lmv)
+               default_lmv_deleted = true;
+
        if (*inode) {
                rc = ll_update_inode(*inode, &md);
                if (rc != 0)
@@ -2478,7 +2565,7 @@ int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
                                             sbi->ll_flags & LL_SBI_32BIT_API),
                                 &md);
                if (IS_ERR(*inode)) {
-#ifdef CONFIG_FS_POSIX_ACL
+#ifdef CONFIG_LUSTRE_FS_POSIX_ACL
                         if (md.posix_acl) {
                                 posix_acl_release(md.posix_acl);
                                 md.posix_acl = NULL;
@@ -2519,6 +2606,9 @@ int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
                LDLM_LOCK_PUT(lock);
        }
 
+       if (default_lmv_deleted)
+               ll_update_default_lsm_md(*inode, &md);
+
        GOTO(out, rc = 0);
 
 out:
@@ -2595,7 +2685,8 @@ void ll_unlock_md_op_lsm(struct md_op_data *op_data)
 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
                                      struct inode *i1, struct inode *i2,
                                      const char *name, size_t namelen,
-                                     __u32 mode, __u32 opc, void *data)
+                                     __u32 mode, enum md_op_code opc,
+                                     void *data)
 {
        LASSERT(i1 != NULL);
 
@@ -2619,15 +2710,13 @@ struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
 
        ll_i2gids(op_data->op_suppgids, i1, i2);
        op_data->op_fid1 = *ll_inode2fid(i1);
-       op_data->op_default_stripe_offset = -1;
+       op_data->op_code = opc;
 
        if (S_ISDIR(i1->i_mode)) {
                down_read(&ll_i2info(i1)->lli_lsm_sem);
                op_data->op_mea1_sem = &ll_i2info(i1)->lli_lsm_sem;
                op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
-               if (opc == LUSTRE_OPC_MKDIR)
-                       op_data->op_default_stripe_offset =
-                                  ll_i2info(i1)->lli_def_stripe_offset;
+               op_data->op_default_mea1 = ll_i2info(i1)->lli_default_lsm_md;
        }
 
        if (i2) {
@@ -2681,7 +2770,7 @@ int ll_show_options(struct seq_file *seq, struct dentry *dentry)
 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
 #endif
 {
-        struct ll_sb_info *sbi;
+       struct ll_sb_info *sbi;
 
 #ifdef HAVE_SUPEROPS_USE_DENTRY
        LASSERT((seq != NULL) && (dentry != NULL));
@@ -2691,20 +2780,25 @@ int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
        sbi = ll_s2sbi(vfs->mnt_sb);
 #endif
 
-        if (sbi->ll_flags & LL_SBI_NOLCK)
-                seq_puts(seq, ",nolock");
-
-        if (sbi->ll_flags & LL_SBI_FLOCK)
-                seq_puts(seq, ",flock");
+       if (sbi->ll_flags & LL_SBI_NOLCK)
+               seq_puts(seq, ",nolock");
 
-        if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
-                seq_puts(seq, ",localflock");
+       /* "flock" is the default since 2.13, but it wasn't for many years,
+        * so it is still useful to print this to show it is enabled.
+        * Start to print "noflock" so it is now clear when flock is disabled.
+        */
+       if (sbi->ll_flags & LL_SBI_FLOCK)
+               seq_puts(seq, ",flock");
+       else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
+               seq_puts(seq, ",localflock");
+       else
+               seq_puts(seq, ",noflock");
 
-        if (sbi->ll_flags & LL_SBI_USER_XATTR)
-                seq_puts(seq, ",user_xattr");
+       if (sbi->ll_flags & LL_SBI_USER_XATTR)
+               seq_puts(seq, ",user_xattr");
 
-        if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
-                seq_puts(seq, ",lazystatfs");
+       if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
+               seq_puts(seq, ",lazystatfs");
 
        if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
                seq_puts(seq, ",user_fid2path");
@@ -2712,7 +2806,7 @@ int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
        if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
                seq_puts(seq, ",always_ping");
 
-        RETURN(0);
+       RETURN(0);
 }
 
 /**
@@ -2769,6 +2863,7 @@ void ll_dirty_page_discard_warn(struct page *page, int ioret)
                        path = ll_d_path(dentry, buf, PAGE_SIZE);
        }
 
+       /* The below message is checked in recovery-small.sh test_24b */
        CDEBUG(D_WARNING,
               "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
               "(rc %d)\n", ll_i2sbi(inode)->ll_fsname,
@@ -2797,12 +2892,12 @@ ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
        if (lum_size < 0)
                RETURN(lum_size);
 
-       OBD_ALLOC(*kbuf, lum_size);
+       OBD_ALLOC_LARGE(*kbuf, lum_size);
        if (*kbuf == NULL)
                RETURN(-ENOMEM);
 
        if (copy_from_user(*kbuf, md, lum_size) != 0) {
-               OBD_FREE(*kbuf, lum_size);
+               OBD_FREE_LARGE(*kbuf, lum_size);
                RETURN(-EFAULT);
        }
 
@@ -2821,7 +2916,7 @@ void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
        struct lnet_process_id id;
 
        /* Update norootsquash flag */
-       down_write(&squash->rsi_sem);
+       spin_lock(&squash->rsi_lock);
        if (list_empty(&squash->rsi_nosquash_nids))
                sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
        else {
@@ -2842,7 +2937,7 @@ void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
                else
                        sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
        }
-       up_write(&squash->rsi_sem);
+       spin_unlock(&squash->rsi_lock);
 }
 
 /**
@@ -2934,13 +3029,8 @@ int ll_getparent(struct file *file, struct getparent __user *arg)
        if (rc < 0)
                GOTO(ldata_free, rc);
 
-#ifdef HAVE_XATTR_HANDLER_FLAGS
        rc = ll_xattr_list(inode, XATTR_NAME_LINK, XATTR_TRUSTED_T, buf.lb_buf,
                           buf.lb_len, OBD_MD_FLXATTR);
-#else
-       rc = ll_getxattr(file_dentry(file), XATTR_NAME_LINK, buf.lb_buf,
-                        buf.lb_len);
-#endif /* HAVE_XATTR_HANDLER_FLAGS */
        if (rc < 0)
                GOTO(lb_free, rc);