Whamcloud - gitweb
LU-12043 llite: move tunable params to sysfs_memparse()
[fs/lustre-release.git] / lustre / llite / llite_lib.c
index 0ffbd6e..ce55b7f 100644 (file)
 #include <linux/version.h>
 #include <linux/mm.h>
 #include <linux/user_namespace.h>
-#ifdef HAVE_UIDGID_HEADER
-# include <linux/uidgid.h>
-#endif
+#include <linux/delay.h>
+#include <linux/uidgid.h>
 #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>
@@ -110,7 +113,7 @@ static struct ll_sb_info *ll_init_sbi(void)
                GOTO(out_destroy_ra, rc = -ENOMEM);
 
        sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
-                                          SBI_DEFAULT_READAHEAD_MAX);
+                                                   SBI_DEFAULT_READ_AHEAD_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;
@@ -151,7 +154,7 @@ 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;
@@ -186,8 +189,7 @@ static void ll_free_sbi(struct super_block *sb)
        EXIT;
 }
 
-static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
-                                    struct vfsmount *mnt)
+static int client_common_fill_super(struct super_block *sb, char *md, char *dt)
 {
        struct inode *root = NULL;
        struct ll_sb_info *sbi = ll_s2sbi(sb);
@@ -251,6 +253,7 @@ 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;
@@ -259,7 +262,7 @@ 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
@@ -269,29 +272,29 @@ 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)
@@ -307,16 +310,16 @@ 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;
 
@@ -378,28 +381,28 @@ 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;
@@ -460,7 +463,8 @@ 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;
@@ -512,7 +516,7 @@ static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
        /* Don't change value if it was specified in the config log */
        if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages == -1) {
                sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
-                       max_t(unsigned long, SBI_DEFAULT_READAHEAD_WHOLE_MAX,
+                       max_t(unsigned long, SBI_DEFAULT_READ_AHEAD_WHOLE_MAX,
                              (data->ocd_brw_size >> PAGE_SHIFT));
                if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages >
                    sbi->ll_ra_info.ra_max_pages_per_file)
@@ -550,9 +554,7 @@ 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
@@ -596,7 +598,7 @@ 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;
@@ -604,7 +606,8 @@ 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);
        }
 
@@ -628,9 +631,6 @@ 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;
 
@@ -795,8 +795,8 @@ 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);
@@ -805,13 +805,11 @@ 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));
-               }
+               while (atomic_read(&sbi->ll_sa_running) > 0)
+                       schedule_timeout_uninterruptible(
+                               cfs_time_seconds(1) >> 3);
        }
 
        EXIT;
@@ -838,38 +836,38 @@ 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;
@@ -984,7 +982,6 @@ 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);
@@ -994,7 +991,7 @@ void ll_lli_init(struct ll_inode_info *lli)
        } else {
                mutex_init(&lli->lli_size_mutex);
                lli->lli_symlink_name = NULL;
-               init_rwsem(&lli->lli_trunc_sem);
+               ll_trunc_sem_init(&lli->lli_trunc_sem);
                range_lock_tree_init(&lli->lli_write_tree);
                init_rwsem(&lli->lli_glimpse_sem);
                lli->lli_glimpse_time = ktime_set(0, 0);
@@ -1007,6 +1004,11 @@ 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;
+               lli->lli_pcc_dsflags = PCC_DATASET_INVALID;
+               lli->lli_pcc_generation = 0;
+               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));
@@ -1048,7 +1050,7 @@ static int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
 }
 #endif /* !HAVE_SUPER_SETUP_BDI_NAME */
 
-int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
+int ll_fill_super(struct super_block *sb)
 {
        struct  lustre_profile *lprof = NULL;
        struct  lustre_sb_info *lsi = s2lsi(sb);
@@ -1057,7 +1059,7 @@ int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
        char    *profilenm = get_profile_name(sb);
        struct config_llog_instance *cfg;
        /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
-       const int instlen = 16 + 2;
+       const int instlen = LUSTRE_MAXINSTANCE + 2;
        unsigned long cfg_instance = ll_get_cfg_instance(sb);
        char name[MAX_STRING_SIZE];
        int md_len = 0;
@@ -1072,8 +1074,6 @@ int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
        CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
               profilenm, cfg_instance, sb);
 
-       try_module_get(THIS_MODULE);
-
        OBD_ALLOC_PTR(cfg);
        if (cfg == NULL)
                GOTO(out_free_cfg, err = -ENOMEM);
@@ -1087,10 +1087,9 @@ int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
        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);
@@ -1170,7 +1169,7 @@ int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
        snprintf(md, md_len - 1, "%s-%016lx", lprof->lp_md, cfg_instance);
 
        /* connections, registrations, sb setup */
-       err = client_common_fill_super(sb, md, dt, mnt);
+       err = client_common_fill_super(sb, md, dt);
        if (err < 0)
                GOTO(out_free_md, err);
 
@@ -1224,49 +1223,48 @@ void ll_put_super(struct super_block *sb)
        params_cfg.cfg_instance = cfg_instance;
        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;
-        }
+       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,
-                       atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
-                       &lwi);
+               rc = l_wait_event_abortable(
+                       sbi->ll_cache->ccc_unstable_waitq,
+                       atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0);
        }
 
        ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
-       if (force == 0 && rc != -EINTR)
+       if (force == 0 && rc != -ERESTARTSYS)
                LASSERTF(ccc_count == 0, "count: %li\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) {
-                next = 0;
-                while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
-                                                     &next)) != NULL) {
-                        obd->obd_force = force;
-                }
-        }
+       /* 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) {
+               next = 0;
+               while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
+                                                    &next)) != NULL) {
+                       obd->obd_force = force;
+               }
+       }
 
        if (sbi->ll_client_common_fill_super_succeeded) {
                /* Only if client_common_fill_super succeeded */
                client_common_put_super(sb);
        }
 
-        next = 0;
-        while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
-                class_manual_cleanup(obd);
-        }
+       next = 0;
+       while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)))
+               class_manual_cleanup(obd);
 
-        if (sbi->ll_flags & LL_SBI_VERBOSE)
-                LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
+       if (sbi->ll_flags & LL_SBI_VERBOSE)
+               LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
 
-        if (profilenm)
-                class_del_profile(profilenm);
+       if (profilenm)
+               class_del_profile(profilenm);
 
 #ifndef HAVE_SUPER_SETUP_BDI_NAME
        if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
@@ -1275,8 +1273,8 @@ void ll_put_super(struct super_block *sb)
        }
 #endif
 
-        ll_free_sbi(sb);
-        lsi->lsi_llsbi = NULL;
+       ll_free_sbi(sb);
+       lsi->lsi_llsbi = NULL;
 out_no_sbi:
        lustre_common_put_super(sb);
 
@@ -1475,6 +1473,7 @@ static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
 {
        struct ll_inode_info *lli = ll_i2info(inode);
        struct lmv_stripe_md *lsm = md->lmv;
+       struct cl_attr  *attr;
        int rc = 0;
 
        ENTRY;
@@ -1498,69 +1497,63 @@ static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
         * normally dir layout doesn't change, only take read lock to check
         * that to avoid blocking other MD operations.
         */
-       if (lli->lli_lsm_md)
-               down_read(&lli->lli_lsm_sem);
-       else
-               down_write(&lli->lli_lsm_sem);
+       down_read(&lli->lli_lsm_sem);
 
-       /*
-        * if dir layout mismatch, check whether version is increased, which
-        * means layout is changed, this happens in dir migration and lfsck.
+       /* some current lookup initialized lsm, and unchanged */
+       if (lli->lli_lsm_md && lsm_md_eq(lli->lli_lsm_md, lsm))
+               GOTO(unlock, rc = 0);
+
+       /* if dir layout doesn't match, check whether version is increased,
+        * which means layout is changed, this happens in dir split/merge and
+        * lfsck.
         *
         * foreign LMV should not change.
         */
-       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,
-                              PFID(&lli->lli_fid));
-                       lsm_md_dump(D_ERROR, lli->lli_lsm_md);
-                       lsm_md_dump(D_ERROR, lsm);
-                       GOTO(unlock, rc = -EINVAL);
-               }
+       if (lli->lli_lsm_md && 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, PFID(&lli->lli_fid));
+               lsm_md_dump(D_ERROR, lli->lli_lsm_md);
+               lsm_md_dump(D_ERROR, lsm);
+               GOTO(unlock, rc = -EINVAL);
+       }
 
-               /* layout changed, switch to write lock */
-               up_read(&lli->lli_lsm_sem);
-               down_write(&lli->lli_lsm_sem);
-               ll_dir_clear_lsm_md(inode);
+       up_read(&lli->lli_lsm_sem);
+       down_write(&lli->lli_lsm_sem);
+       /* clear existing lsm */
+       if (lli->lli_lsm_md) {
+               lmv_free_memmd(lli->lli_lsm_md);
+               lli->lli_lsm_md = NULL;
        }
 
-       /* set directory layout */
-       if (!lli->lli_lsm_md) {
-               struct cl_attr  *attr;
+       rc = ll_init_lsm_md(inode, md);
+       up_write(&lli->lli_lsm_sem);
 
-               rc = ll_init_lsm_md(inode, md);
-               up_write(&lli->lli_lsm_sem);
-               if (rc != 0)
-                       RETURN(rc);
+       if (rc)
+               RETURN(rc);
 
-               /* set md->lmv to NULL, so the following free lustre_md
-                * will not free this lsm */
-               md->lmv = NULL;
+       /* set md->lmv to NULL, so the following free lustre_md will not free
+        * this lsm.
+        */
+       md->lmv = NULL;
 
-               /*
-                * md_merge_attr() may take long, since lsm is already set,
-                * switch to read lock.
-                */
-               down_read(&lli->lli_lsm_sem);
+       /* md_merge_attr() may take long, since lsm is already set, switch to
+        * read lock.
+        */
+       down_read(&lli->lli_lsm_sem);
 
-               if (!lmv_dir_striped(lli->lli_lsm_md))
-                       GOTO(unlock, rc);
+       if (!lmv_dir_striped(lli->lli_lsm_md))
+               GOTO(unlock, rc = 0);
 
-               OBD_ALLOC_PTR(attr);
-               if (attr == NULL)
-                       GOTO(unlock, rc = -ENOMEM);
-
-               /* validate the lsm */
-               rc = md_merge_attr(ll_i2mdexp(inode), lsm, attr,
-                                  ll_md_blocking_ast);
-               if (rc != 0) {
-                       OBD_FREE_PTR(attr);
-                       GOTO(unlock, rc);
-               }
+       OBD_ALLOC_PTR(attr);
+       if (!attr)
+               GOTO(unlock, rc = -ENOMEM);
 
+       /* validate the lsm */
+       rc = md_merge_attr(ll_i2mdexp(inode), lli->lli_lsm_md, attr,
+                          ll_md_blocking_ast);
+       if (!rc) {
                if (md->body->mbo_valid & OBD_MD_FLNLINK)
                        md->body->mbo_nlink = attr->cat_nlink;
                if (md->body->mbo_valid & OBD_MD_FLSIZE)
@@ -1571,13 +1564,14 @@ static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
                        md->body->mbo_ctime = attr->cat_ctime;
                if (md->body->mbo_valid & OBD_MD_FLMTIME)
                        md->body->mbo_mtime = attr->cat_mtime;
-
-               OBD_FREE_PTR(attr);
        }
+
+       OBD_FREE_PTR(attr);
+       GOTO(unlock, rc);
 unlock:
        up_read(&lli->lli_lsm_sem);
 
-       RETURN(rc);
+       return rc;
 }
 
 void ll_clear_inode(struct inode *inode)
@@ -1620,7 +1614,7 @@ 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);
@@ -1722,6 +1716,7 @@ 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;
@@ -1780,11 +1775,8 @@ int ll_setattr_raw(struct dentry *dentry, struct iattr *attr,
                       (s64)attr->ia_mtime.tv_sec, (s64)attr->ia_ctime.tv_sec,
                       ktime_get_real_seconds());
 
-       if (S_ISREG(inode->i_mode)) {
-               if (attr->ia_valid & ATTR_SIZE)
-                       inode_dio_write_done(inode);
+       if (S_ISREG(inode->i_mode))
                inode_unlock(inode);
-       }
 
        /* We always do an MDS RPC, even if we're only changing the size;
         * only the MDS knows whether truncate() should fail with -ETXTBUSY */
@@ -1803,7 +1795,7 @@ int ll_setattr_raw(struct dentry *dentry, struct iattr *attr,
        }
 
        if (attr->ia_valid & ATTR_FILE) {
-               struct ll_file_data *fd = LUSTRE_FPRIVATE(attr->ia_file);
+               struct ll_file_data *fd = attr->ia_file->private_data;
 
                if (fd->fd_lease_och)
                        op_data->op_bias |= MDS_TRUNC_KEEP_LEASE;
@@ -1894,8 +1886,10 @@ 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;
 }
@@ -1941,6 +1935,9 @@ int ll_statfs_internal(struct ll_sb_info *sbi, struct obd_statfs *osfs,
        ENTRY;
        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)
                RETURN(rc);
@@ -1953,9 +1950,6 @@ 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);
@@ -1983,15 +1977,16 @@ 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);
@@ -2014,11 +2009,15 @@ 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;
 }
 
@@ -2068,7 +2067,7 @@ 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)
@@ -2232,6 +2231,8 @@ 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) {
@@ -2256,9 +2257,9 @@ 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, "
@@ -2266,9 +2267,7 @@ 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;
@@ -2385,8 +2384,7 @@ void ll_umount_begin(struct super_block *sb)
        struct ll_sb_info *sbi = ll_s2sbi(sb);
        struct obd_device *obd;
        struct obd_ioctl_data *ioc_data;
-       struct l_wait_info lwi;
-       wait_queue_head_t waitq;
+       int cnt;
        ENTRY;
 
        CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
@@ -2401,16 +2399,16 @@ 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);
@@ -2425,44 +2423,46 @@ void ll_umount_begin(struct super_block *sb)
         * and then continue.  For now, we just periodically checking for vfs
         * to decrement mnt_cnt and hope to finish it within 10sec.
         */
-       init_waitqueue_head(&waitq);
-       lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(10),
-                                  cfs_time_seconds(1), NULL, NULL);
-       l_wait_event(waitq, may_umount(sbi->ll_mnt.mnt), &lwi);
+       cnt = 10;
+       while (cnt > 0 &&
+              !may_umount(sbi->ll_mnt.mnt)) {
+               ssleep(1);
+               cnt -= 1;
+       }
 
        EXIT;
 }
 
 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;
 }
 
 /**
@@ -2555,7 +2555,7 @@ 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;
@@ -2605,8 +2605,10 @@ out:
        /* cleanup will be done if necessary */
        md_free_lustre_md(sbi->ll_md_exp, &md);
 
-       if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
+       if (rc != 0 && it != NULL && it->it_op & IT_OPEN) {
+               ll_intent_drop_lock(it);
                ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
+       }
 
        return rc;
 }
@@ -2754,36 +2756,32 @@ void ll_finish_md_op_data(struct md_op_data *op_data)
         OBD_FREE_PTR(op_data);
 }
 
-#ifdef HAVE_SUPEROPS_USE_DENTRY
 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
-#else
-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));
+       LASSERT(seq && dentry);
        sbi = ll_s2sbi(dentry->d_sb);
-#else
-       LASSERT((seq != NULL) && (vfs != NULL));
-       sbi = ll_s2sbi(vfs->mnt_sb);
-#endif
 
-        if (sbi->ll_flags & LL_SBI_NOLCK)
-                seq_puts(seq, ",nolock");
+       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_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");
@@ -2791,7 +2789,7 @@ 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);
 }
 
 /**
@@ -2848,6 +2846,7 @@ 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,
@@ -2876,12 +2875,12 @@ 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);
        }
 
@@ -2900,7 +2899,7 @@ 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 {
@@ -2921,7 +2920,7 @@ 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);
 }
 
 /**
@@ -3013,13 +3012,8 @@ 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);