Whamcloud - gitweb
LU-1014 mountconf: Revert "MGS should process parameter config"
[fs/lustre-release.git] / lustre / obdclass / obd_mount.c
index 8c50cfb..890d409 100644 (file)
@@ -1,6 +1,4 @@
-/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
- * vim:expandtab:shiftwidth=8:tabstop=8:
- *
+/*
  * GPL HEADER START
  *
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
@@ -29,8 +27,7 @@
  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  * Use is subject to license terms.
  *
- * Copyright (c) 2011 Whamcloud, Inc.
- *
+ * Copyright (c) 2011, 2012, Whamcloud, Inc.
  */
 /*
  * This file is part of Lustre, http://www.lustre.org/
@@ -65,7 +62,7 @@ static void (*kill_super_cb)(struct super_block *sb) = NULL;
 
 /*********** mount lookup *********/
 
-CFS_DECLARE_MUTEX(lustre_mount_info_lock);
+CFS_DEFINE_MUTEX(lustre_mount_info_lock);
 static CFS_LIST_HEAD(server_mount_info_list);
 
 static struct lustre_mount_info *server_find_mount(const char *name)
@@ -106,10 +103,10 @@ static int server_register_mount(const char *name, struct super_block *sb,
         }
         strcpy(name_cp, name);
 
-        cfs_down(&lustre_mount_info_lock);
+        cfs_mutex_lock(&lustre_mount_info_lock);
 
         if (server_find_mount(name)) {
-                cfs_up(&lustre_mount_info_lock);
+                cfs_mutex_unlock(&lustre_mount_info_lock);
                 OBD_FREE(lmi, sizeof(*lmi));
                 OBD_FREE(name_cp, strlen(name) + 1);
                 CERROR("Already registered %s\n", name);
@@ -120,10 +117,10 @@ static int server_register_mount(const char *name, struct super_block *sb,
         lmi->lmi_mnt = mnt;
         cfs_list_add(&lmi->lmi_list_chain, &server_mount_info_list);
 
-        cfs_up(&lustre_mount_info_lock);
+        cfs_mutex_unlock(&lustre_mount_info_lock);
 
         CDEBUG(D_MOUNT, "reg_mnt %p from %s, vfscount=%d\n",
-               lmi->lmi_mnt, name, cfs_atomic_read(&lmi->lmi_mnt->mnt_count));
+               lmi->lmi_mnt, name, mnt_get_count(lmi->lmi_mnt));
 
         RETURN(0);
 }
@@ -134,21 +131,21 @@ static int server_deregister_mount(const char *name)
         struct lustre_mount_info *lmi;
         ENTRY;
 
-        cfs_down(&lustre_mount_info_lock);
+        cfs_mutex_lock(&lustre_mount_info_lock);
         lmi = server_find_mount(name);
         if (!lmi) {
-                cfs_up(&lustre_mount_info_lock);
+                cfs_mutex_unlock(&lustre_mount_info_lock);
                 CERROR("%s not registered\n", name);
                 RETURN(-ENOENT);
         }
 
         CDEBUG(D_MOUNT, "dereg_mnt %p from %s, vfscount=%d\n",
-               lmi->lmi_mnt, name, cfs_atomic_read(&lmi->lmi_mnt->mnt_count));
+               lmi->lmi_mnt, name, mnt_get_count(lmi->lmi_mnt));
 
         OBD_FREE(lmi->lmi_name, strlen(lmi->lmi_name) + 1);
         cfs_list_del(&lmi->lmi_list_chain);
         OBD_FREE(lmi, sizeof(*lmi));
-        cfs_up(&lustre_mount_info_lock);
+        cfs_mutex_unlock(&lustre_mount_info_lock);
 
         RETURN(0);
 }
@@ -162,9 +159,9 @@ struct lustre_mount_info *server_get_mount(const char *name)
         struct lustre_sb_info *lsi;
         ENTRY;
 
-        cfs_down(&lustre_mount_info_lock);
+        cfs_mutex_lock(&lustre_mount_info_lock);
         lmi = server_find_mount(name);
-        cfs_up(&lustre_mount_info_lock);
+        cfs_mutex_unlock(&lustre_mount_info_lock);
         if (!lmi) {
                 CERROR("Can't find mount for %s\n", name);
                 RETURN(NULL);
@@ -175,7 +172,7 @@ struct lustre_mount_info *server_get_mount(const char *name)
 
         CDEBUG(D_MOUNT, "get_mnt %p from %s, refs=%d, vfscount=%d\n",
                lmi->lmi_mnt, name, cfs_atomic_read(&lsi->lsi_mounts),
-               cfs_atomic_read(&lmi->lmi_mnt->mnt_count));
+               mnt_get_count(lmi->lmi_mnt));
 
         RETURN(lmi);
 }
@@ -190,9 +187,9 @@ struct lustre_mount_info *server_get_mount_2(const char *name)
         struct lustre_mount_info *lmi;
         ENTRY;
 
-        cfs_down(&lustre_mount_info_lock);
+        cfs_mutex_lock(&lustre_mount_info_lock);
         lmi = server_find_mount(name);
-        cfs_up(&lustre_mount_info_lock);
+        cfs_mutex_unlock(&lustre_mount_info_lock);
         if (!lmi)
                 CERROR("Can't find mount for %s\n", name);
 
@@ -201,13 +198,18 @@ struct lustre_mount_info *server_get_mount_2(const char *name)
 
 static void unlock_mntput(struct vfsmount *mnt)
 {
-        if (kernel_locked()) {
-                cfs_unlock_kernel();
-                mntput(mnt);
-                cfs_lock_kernel();
-        } else {
-                mntput(mnt);
-        }
+#ifdef HAVE_KERNEL_LOCKED
+       /* for kernel < 2.6.37 */
+       if (kernel_locked()) {
+               unlock_kernel();
+               mntput(mnt);
+               lock_kernel();
+       } else {
+               mntput(mnt);
+       }
+#else
+       mntput(mnt);
+#endif
 }
 
 static int lustre_put_lsi(struct super_block *sb);
@@ -217,15 +219,15 @@ int server_put_mount(const char *name, struct vfsmount *mnt)
 {
         struct lustre_mount_info *lmi;
         struct lustre_sb_info *lsi;
-        int count = atomic_read(&mnt->mnt_count) - 1;
+        int count = mnt_get_count(mnt) - 1;
         ENTRY;
 
         /* This might be the last one, can't deref after this */
         unlock_mntput(mnt);
 
-        cfs_down(&lustre_mount_info_lock);
+        cfs_mutex_lock(&lustre_mount_info_lock);
         lmi = server_find_mount(name);
-        cfs_up(&lustre_mount_info_lock);
+        cfs_mutex_unlock(&lustre_mount_info_lock);
         if (!lmi) {
                 CERROR("Can't find mount for %s\n", name);
                 RETURN(-ENOENT);
@@ -581,7 +583,7 @@ static int server_stop_mgs(struct super_block *sb)
         RETURN(rc);
 }
 
-CFS_DECLARE_MUTEX(mgc_start_lock);
+CFS_DEFINE_MUTEX(mgc_start_lock);
 
 /** Set up a mgc obd to process startup logs
  *
@@ -634,7 +636,7 @@ static int lustre_start_mgc(struct super_block *sb)
                 RETURN(-EINVAL);
         }
 
-        cfs_mutex_down(&mgc_start_lock);
+        cfs_mutex_lock(&mgc_start_lock);
 
         len = strlen(LUSTRE_MGC_OBDNAME) + strlen(libcfs_nid2str(nid)) + 1;
         OBD_ALLOC(mgcname, len);
@@ -651,7 +653,7 @@ static int lustre_start_mgc(struct super_block *sb)
 
         obd = class_name2obd(mgcname);
         if (obd && !obd->obd_stopping) {
-                rc = obd_set_info_async(obd->obd_self_export,
+                rc = obd_set_info_async(NULL, obd->obd_self_export,
                                         strlen(KEY_MGSSEC), KEY_MGSSEC,
                                         strlen(mgssec), mgssec, NULL);
                 if (rc)
@@ -666,7 +668,7 @@ static int lustre_start_mgc(struct super_block *sb)
                         int vallen = sizeof(*data);
                         __u32 *flags = &lsi->lsi_lmd->lmd_flags;
 
-                        rc = obd_get_info(obd->obd_self_export,
+                        rc = obd_get_info(NULL, obd->obd_self_export,
                                           strlen(KEY_CONN_DATA), KEY_CONN_DATA,
                                           &vallen, data, NULL);
                         LASSERT(rc == 0);
@@ -700,7 +702,7 @@ static int lustre_start_mgc(struct super_block *sb)
                    if at all possible. */
                 recov_bk++;
                 CDEBUG(D_MOUNT, "%s: Set MGC reconnect %d\n", mgcname,recov_bk);
-                rc = obd_set_info_async(obd->obd_self_export,
+                rc = obd_set_info_async(NULL, obd->obd_self_export,
                                         sizeof(KEY_INIT_RECOV_BACKUP),
                                         KEY_INIT_RECOV_BACKUP,
                                         sizeof(recov_bk), &recov_bk, NULL);
@@ -795,7 +797,7 @@ static int lustre_start_mgc(struct super_block *sb)
                 GOTO(out_free, rc = -ENOTCONN);
         }
 
-        rc = obd_set_info_async(obd->obd_self_export,
+        rc = obd_set_info_async(NULL, obd->obd_self_export,
                                 strlen(KEY_MGSSEC), KEY_MGSSEC,
                                 strlen(mgssec), mgssec, NULL);
         if (rc)
@@ -807,7 +809,7 @@ static int lustre_start_mgc(struct super_block *sb)
 
         /* Try all connections, but only once. */
         recov_bk = 1;
-        rc = obd_set_info_async(obd->obd_self_export,
+        rc = obd_set_info_async(NULL, obd->obd_self_export,
                                 sizeof(KEY_INIT_RECOV_BACKUP),
                                 KEY_INIT_RECOV_BACKUP,
                                 sizeof(recov_bk), &recov_bk, NULL);
@@ -835,7 +837,7 @@ out:
            to the same mgc.*/
         lsi->lsi_mgc = obd;
 out_free:
-        cfs_mutex_up(&mgc_start_lock);
+        cfs_mutex_unlock(&mgc_start_lock);
 
         if (data)
                 OBD_FREE_PTR(data);
@@ -861,7 +863,7 @@ static int lustre_stop_mgc(struct super_block *sb)
                 RETURN(-ENOENT);
         lsi->lsi_mgc = NULL;
 
-        cfs_mutex_down(&mgc_start_lock);
+        cfs_mutex_lock(&mgc_start_lock);
         LASSERT(cfs_atomic_read(&obd->u.cli.cl_mgc_refcount) > 0);
         if (!cfs_atomic_dec_and_test(&obd->u.cli.cl_mgc_refcount)) {
                 /* This is not fatal, every client that stops
@@ -913,7 +915,7 @@ out:
                 OBD_FREE(niduuid, len);
 
         /* class_import_put will get rid of the additional connections */
-        cfs_mutex_up(&mgc_start_lock);
+        cfs_mutex_unlock(&mgc_start_lock);
         RETURN(rc);
 }
 
@@ -928,7 +930,7 @@ static int server_mgc_set_fs(struct obd_device *mgc, struct super_block *sb)
         CDEBUG(D_MOUNT, "Set mgc disk for %s\n", lsi->lsi_lmd->lmd_dev);
 
         /* cl_mgc_sem in mgc insures we sleep if the mgc_fs is busy */
-        rc = obd_set_info_async(mgc->obd_self_export,
+        rc = obd_set_info_async(NULL, mgc->obd_self_export,
                                 sizeof(KEY_SET_FS), KEY_SET_FS,
                                 sizeof(*sb), sb, NULL);
         if (rc) {
@@ -945,13 +947,13 @@ static int server_mgc_clear_fs(struct obd_device *mgc)
 
         CDEBUG(D_MOUNT, "Unassign mgc disk\n");
 
-        rc = obd_set_info_async(mgc->obd_self_export,
+        rc = obd_set_info_async(NULL, mgc->obd_self_export,
                                 sizeof(KEY_CLEAR_FS), KEY_CLEAR_FS,
                                 0, NULL, NULL);
         RETURN(rc);
 }
 
-CFS_DECLARE_MUTEX(server_start_lock);
+CFS_DEFINE_MUTEX(server_start_lock);
 
 /* Stop MDS/OSS if nobody is using them */
 static int server_stop_servers(int lddflags, int lsiflags)
@@ -961,7 +963,7 @@ static int server_stop_servers(int lddflags, int lsiflags)
         int rc = 0;
         ENTRY;
 
-        cfs_mutex_down(&server_start_lock);
+        cfs_mutex_lock(&server_start_lock);
 
         /* Either an MDT or an OST or neither  */
         /* if this was an MDT, and there are no more MDT's, clean up the MDS */
@@ -985,7 +987,7 @@ static int server_stop_servers(int lddflags, int lsiflags)
                         rc = err;
         }
 
-        cfs_mutex_up(&server_start_lock);
+        cfs_mutex_unlock(&server_start_lock);
 
         RETURN(rc);
 }
@@ -1093,7 +1095,7 @@ int server_register_target(struct super_block *sb)
 
         /* Register the target */
         /* FIXME use mgc_process_config instead */
-        rc = obd_set_info_async(mgc->u.cli.cl_mgc_mgsexp,
+        rc = obd_set_info_async(NULL, mgc->u.cli.cl_mgc_mgsexp,
                                 sizeof(KEY_REGISTER_TARGET), KEY_REGISTER_TARGET,
                                 sizeof(*mti), mti, NULL);
         if (rc) {
@@ -1176,7 +1178,7 @@ static int server_notify_target(struct super_block *sb, struct obd_device *obd)
         mti->mti_flags |= LDD_F_OPC_READY;
 
         /* FIXME use mgc_process_config instead */
-        rc = obd_set_info_async(mgc->u.cli.cl_mgc_mgsexp,
+        rc = obd_set_info_async(NULL, mgc->u.cli.cl_mgc_mgsexp,
                                 sizeof(KEY_REGISTER_TARGET),
                                 KEY_REGISTER_TARGET,
                                 sizeof(*mti), mti, NULL);
@@ -1209,7 +1211,7 @@ static int server_start_targets(struct super_block *sb, struct vfsmount *mnt)
         /* If we're an MDT, make sure the global MDS is running */
         if (lsi->lsi_ldd->ldd_flags & LDD_F_SV_TYPE_MDT) {
                 /* make sure the MDS is started */
-                cfs_mutex_down(&server_start_lock);
+                cfs_mutex_lock(&server_start_lock);
                 obd = class_name2obd(LUSTRE_MDS_OBDNAME);
                 if (!obd) {
                         rc = lustre_start_simple(LUSTRE_MDS_OBDNAME,
@@ -1218,19 +1220,19 @@ static int server_start_targets(struct super_block *sb, struct vfsmount *mnt)
                                                  LUSTRE_MDS_OBDNAME"_uuid",
                                                  0, 0);
                         if (rc) {
-                                cfs_mutex_up(&server_start_lock);
+                                cfs_mutex_unlock(&server_start_lock);
                                 CERROR("failed to start MDS: %d\n", rc);
                                 RETURN(rc);
                         }
                 }
-                cfs_mutex_up(&server_start_lock);
+                cfs_mutex_unlock(&server_start_lock);
         }
 #endif
 
         /* If we're an OST, make sure the global OSS is running */
         if (IS_OST(lsi->lsi_ldd)) {
                 /* make sure OSS is started */
-                cfs_mutex_down(&server_start_lock);
+                cfs_mutex_lock(&server_start_lock);
                 obd = class_name2obd(LUSTRE_OSS_OBDNAME);
                 if (!obd) {
                         rc = lustre_start_simple(LUSTRE_OSS_OBDNAME,
@@ -1238,12 +1240,12 @@ static int server_start_targets(struct super_block *sb, struct vfsmount *mnt)
                                                  LUSTRE_OSS_OBDNAME"_uuid",
                                                  0, 0);
                         if (rc) {
-                                cfs_mutex_up(&server_start_lock);
+                                cfs_mutex_unlock(&server_start_lock);
                                 CERROR("failed to start OSS: %d\n", rc);
                                 RETURN(rc);
                         }
                 }
-                cfs_mutex_up(&server_start_lock);
+                cfs_mutex_unlock(&server_start_lock);
         }
 
         /* Set the mgc fs to our server disk.  This allows the MGC to
@@ -1406,6 +1408,7 @@ static struct vfsmount *server_kernel_mount(struct super_block *sb)
         struct lustre_disk_data *ldd;
         struct lustre_mount_data *lmd = lsi->lsi_lmd;
         struct vfsmount *mnt;
+        struct file_system_type *type;
         char *options = NULL;
         unsigned long page, s_flags;
         struct page *__page;
@@ -1437,7 +1440,13 @@ static struct vfsmount *server_kernel_mount(struct super_block *sb)
 
         /* Pre-mount ldiskfs to read the MOUNT_DATA_FILE */
         CDEBUG(D_MOUNT, "Pre-mount ldiskfs %s\n", lmd->lmd_dev);
-        mnt = ll_kern_mount("ldiskfs", s_flags, lmd->lmd_dev, (void *)options);
+        type = get_fs_type("ldiskfs");
+        if (!type) {
+                CERROR("premount failed: cannot find ldiskfs module\n");
+                GOTO(out_free, rc = -ENODEV);
+        }
+        mnt = vfs_kern_mount(type, s_flags, lmd->lmd_dev, (void *)options);
+        cfs_module_put(type->owner);
         if (IS_ERR(mnt)) {
                 rc = PTR_ERR(mnt);
                 CERROR("premount %s:%#lx ldiskfs failed: %d "
@@ -1483,11 +1492,16 @@ static struct vfsmount *server_kernel_mount(struct super_block *sb)
 
         CDEBUG(D_MOUNT, "kern_mount: %s %s %s\n",
                MT_STR(ldd), lmd->lmd_dev, options);
-        mnt = ll_kern_mount(MT_STR(ldd), s_flags, lmd->lmd_dev,
-                            (void *)options);
+        type = get_fs_type(MT_STR(ldd));
+        if (!type) {
+                CERROR("get_fs_type failed\n");
+                GOTO(out_free, rc = -ENODEV);
+        }
+        mnt = vfs_kern_mount(type, s_flags, lmd->lmd_dev, (void *)options);
+        cfs_module_put(type->owner);
         if (IS_ERR(mnt)) {
                 rc = PTR_ERR(mnt);
-                CERROR("ll_kern_mount failed: rc = %d\n", rc);
+                CERROR("vfs_kern_mount failed: rc = %d\n", rc);
                 GOTO(out_free, rc);
         }
 
@@ -1520,25 +1534,25 @@ static void server_wait_finished(struct vfsmount *mnt)
 
        cfs_waitq_init(&waitq);
 
-       while (atomic_read(&mnt->mnt_count) > 1) {
+       while (mnt_get_count(mnt) > 1) {
                if (waited && (waited % 30 == 0))
                        LCONSOLE_WARN("Mount still busy with %d refs after "
                                       "%d secs.\n",
-                                      atomic_read(&mnt->mnt_count),
+                                      mnt_get_count(mnt),
                                       waited);
                /* Cannot use l_event_wait() for an interruptible sleep. */
                waited += 3;
                blocked = cfs_block_sigsinv(sigmask(SIGKILL));
                cfs_waitq_wait_event_interruptible_timeout(
                        waitq,
-                       (atomic_read(&mnt->mnt_count) == 1),
+                       (mnt_get_count(mnt) == 1),
                        cfs_time_seconds(3),
                        rc);
-               cfs_block_sigs(blocked);
+              cfs_restore_sigs(blocked);
                if (rc < 0) {
                        LCONSOLE_EMERG("Danger: interrupted umount %s with "
                                       "%d refs!\n", mnt->mnt_devname,
-                                      atomic_read(&mnt->mnt_count));
+                                      mnt_get_count(mnt));
                        break;
                }
 
@@ -1725,7 +1739,7 @@ static int server_fill_super_common(struct super_block *sb)
         sb->s_blocksize = 4096;
         sb->s_blocksize_bits = log2(sb->s_blocksize);
         sb->s_magic = LUSTRE_SUPER_MAGIC;
-        sb->s_maxbytes = 0; //PAGE_CACHE_MAXBYTES;
+        sb->s_maxbytes = 0; /* we don't allow file IO on server mountpoints */
         sb->s_flags |= MS_RDONLY;
         sb->s_op = &server_ops;
 
@@ -2150,6 +2164,9 @@ static int lmd_parse(char *options, struct lustre_mount_data *lmd)
                 } else if (strncmp(s1, "nomgs", 5) == 0) {
                         lmd->lmd_flags |= LMD_FLG_NOMGS;
                         clear++;
+               } else if (strncmp(s1, "noscrub", 7) == 0) {
+                       lmd->lmd_flags |= LMD_FLG_NOSCRUB;
+                       clear++;
                 } else if (strncmp(s1, "writeconf", 9) == 0) {
                         lmd->lmd_flags |= LMD_FLG_WRITECONF;
                         clear++;
@@ -2267,6 +2284,11 @@ int lustre_fill_super(struct super_block *sb, void *data, int silent)
          */
         cfs_lockdep_off();
 
+        /*
+         * LU-639: the obd cleanup of last mount may not finish yet, wait here.
+         */
+        obd_zombie_barrier();
+
         /* Figure out the lmd from the mount options */
         if (lmd_parse((char *)(lmd2->lmd2_data), lmd)) {
                 lustre_put_lsi(sb);