Whamcloud - gitweb
LU-12355 llite: MS_* flags and SB_* flags split
[fs/lustre-release.git] / lustre / llite / llite_lib.c
index 91eb088..5ba34db 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;
 
@@ -137,12 +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);
+       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)
@@ -153,10 +178,13 @@ 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;
                }
+               pcc_super_fini(&sbi->ll_pcc_super);
                OBD_FREE(sbi, sizeof(*sbi));
        }
        EXIT;
@@ -228,7 +256,8 @@ static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
                                   OBD_CONNECT2_LOCK_CONVERT |
                                   OBD_CONNECT2_ARCHIVE_ID_ARRAY |
                                   OBD_CONNECT2_LSOM |
-                                  OBD_CONNECT2_ASYNC_DISCARD;
+                                  OBD_CONNECT2_ASYNC_DISCARD |
+                                  OBD_CONNECT2_PCC;
 
 #ifdef HAVE_LRU_RESIZE_SUPPORT
         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
@@ -244,29 +273,29 @@ 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)
@@ -282,16 +311,16 @@ 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;
 
@@ -307,7 +336,7 @@ 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);
@@ -353,28 +382,28 @@ 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;
@@ -770,8 +799,8 @@ 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);
@@ -785,7 +814,7 @@ void ll_kill_super(struct super_block *sb)
                /* 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));
                }
        }
 
@@ -965,7 +994,6 @@ void ll_lli_init(struct ll_inode_info *lli)
                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);
@@ -980,6 +1008,9 @@ void ll_lli_init(struct ll_inode_info *lli)
                spin_lock_init(&lli->lli_heat_lock);
                obd_heat_clear(lli->lli_heat_instances, OBD_HEAT_COUNT);
                lli->lli_heat_flags = 0;
+               mutex_init(&lli->lli_pcc_lock);
+               lli->lli_pcc_state = PCC_STATE_FL_NONE;
+               lli->lli_pcc_inode = NULL;
        }
        mutex_init(&lli->lli_layout_mutex);
        memset(lli->lli_jobid, 0, sizeof(lli->lli_jobid));
@@ -1053,8 +1084,8 @@ 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)
@@ -1184,7 +1215,7 @@ 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 */
@@ -1291,10 +1322,15 @@ 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,
@@ -1363,6 +1399,9 @@ 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. */
@@ -1390,12 +1429,52 @@ 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);
@@ -1408,6 +1487,10 @@ 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()
@@ -1430,10 +1513,9 @@ 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,
@@ -1453,15 +1535,6 @@ 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)
@@ -1477,6 +1550,9 @@ 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);
@@ -1512,17 +1588,20 @@ void ll_clear_inode(struct inode *inode)
 {
         struct ll_inode_info *lli = ll_i2info(inode);
         struct ll_sb_info *sbi = ll_i2sbi(inode);
+
         ENTRY;
 
        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 */
-                LASSERT(lli->lli_opendir_key == NULL);
-                LASSERT(lli->lli_sai == NULL);
-                LASSERT(lli->lli_opendir_pid == 0);
-        }
+       if (S_ISDIR(inode->i_mode)) {
+               /* these should have been cleared in ll_file_release */
+               LASSERT(lli->lli_opendir_key == NULL);
+               LASSERT(lli->lli_sai == NULL);
+               LASSERT(lli->lli_opendir_pid == 0);
+       } else {
+               pcc_inode_free(inode);
+       }
 
        md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
 
@@ -1648,6 +1727,7 @@ int ll_setattr_raw(struct dentry *dentry, struct iattr *attr,
         struct ll_inode_info *lli = ll_i2info(inode);
         struct md_op_data *op_data = NULL;
        int rc = 0;
+
        ENTRY;
 
        CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
@@ -1746,14 +1826,28 @@ int ll_setattr_raw(struct dentry *dentry, struct iattr *attr,
        if (attr->ia_valid & (ATTR_SIZE | ATTR_ATIME | ATTR_ATIME_SET |
                              ATTR_MTIME | ATTR_MTIME_SET | ATTR_CTIME) ||
            xvalid & OP_XVALID_CTIME_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 = cl_setattr_ost(lli->lli_clob, attr, xvalid, 0);
+               bool cached = false;
+
+               rc = pcc_inode_setattr(inode, attr, &cached);
+               if (cached) {
+                       if (rc) {
+                               CERROR("%s: PCC inode "DFID" setattr failed: "
+                                      "rc = %d\n",
+                                      ll_i2sbi(inode)->ll_fsname,
+                                      PFID(&lli->lli_fid), rc);
+                               GOTO(out, rc);
+                       }
+               } else {
+                       /* 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 = cl_setattr_ost(lli->lli_clob, attr, xvalid, 0);
+               }
        }
 
        /* If the file was restored, it needs to set dirty flag.
@@ -1849,7 +1943,7 @@ 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;
 
        rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
        if (rc)
@@ -1901,7 +1995,7 @@ int ll_statfs(struct dentry *de, struct kstatfs *sfs)
        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);
+       ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STATFS, 1);
 
        /* Some amount of caching on the client is allowed */
        rc = ll_statfs_internal(ll_s2sbi(sb), &osfs, OBD_STATFS_SUM);
@@ -2345,34 +2439,34 @@ 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;
 }
 
 /**
@@ -2422,7 +2516,9 @@ 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);
@@ -2432,6 +2528,15 @@ 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)
@@ -2495,6 +2600,9 @@ 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:
@@ -2571,7 +2679,8 @@ 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);
 
@@ -2595,15 +2704,13 @@ 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) {
@@ -2797,7 +2904,7 @@ 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 {
@@ -2818,7 +2925,7 @@ 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);
 }
 
 /**