Whamcloud - gitweb
LU-7394 obd: remove HAVE_SELINUX_IS_ENABLED
[fs/lustre-release.git] / lustre / obdclass / obd_mount_server.c
index d67963b..4452f05 100644 (file)
@@ -27,7 +27,7 @@
  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  * Use is subject to license terms.
  *
- * Copyright (c) 2013, Intel Corporation.
+ * Copyright (c) 2013, 2015, Intel Corporation.
  */
 /*
  * This file is part of Lustre, http://www.lustre.org/
@@ -46,6 +46,9 @@
 #define PRINT_CMD CDEBUG
 #define PRINT_MASK (D_SUPER | D_CONFIG)
 
+#include <linux/types.h>
+#include <linux/selinux.h>
+#include <linux/statfs.h>
 #include <obd.h>
 #include <obd_class.h>
 #include <lustre/lustre_user.h>
 #ifdef HAVE_KERNEL_LOCKED
 #include <linux/smp_lock.h>
 #endif
-#ifdef HAVE_SELINUX_IS_ENABLED
-#include <linux/selinux.h>
-#endif
 
 /*********** mount lookup *********/
 
-DEFINE_MUTEX(lustre_mount_info_lock);
-static CFS_LIST_HEAD(server_mount_info_list);
+static DEFINE_MUTEX(lustre_mount_info_lock);
+static struct list_head server_mount_info_list =
+       LIST_HEAD_INIT(server_mount_info_list);
 
 static struct lustre_mount_info *server_find_mount(const char *name)
 {
-       cfs_list_t *tmp;
+       struct list_head *tmp;
        struct lustre_mount_info *lmi;
        ENTRY;
 
-       cfs_list_for_each(tmp, &server_mount_info_list) {
-               lmi = cfs_list_entry(tmp, struct lustre_mount_info,
-                                    lmi_list_chain);
+       list_for_each(tmp, &server_mount_info_list) {
+               lmi = list_entry(tmp, struct lustre_mount_info,
+                                lmi_list_chain);
                if (strcmp(name, lmi->lmi_name) == 0)
                        RETURN(lmi);
        }
@@ -113,7 +114,7 @@ static int server_register_mount(const char *name, struct super_block *sb)
        }
        lmi->lmi_name = name_cp;
        lmi->lmi_sb = sb;
-       cfs_list_add(&lmi->lmi_list_chain, &server_mount_info_list);
+       list_add(&lmi->lmi_list_chain, &server_mount_info_list);
 
        mutex_unlock(&lustre_mount_info_lock);
 
@@ -139,7 +140,7 @@ static int server_deregister_mount(const char *name)
        CDEBUG(D_MOUNT, "deregister mount %p from %s\n", lmi->lmi_sb, name);
 
        OBD_FREE(lmi->lmi_name, strlen(lmi->lmi_name) + 1);
-       cfs_list_del(&lmi->lmi_list_chain);
+       list_del(&lmi->lmi_list_chain);
        OBD_FREE(lmi, sizeof(*lmi));
        mutex_unlock(&lustre_mount_info_lock);
 
@@ -173,28 +174,19 @@ struct lustre_mount_info *server_get_mount(const char *name)
 }
 EXPORT_SYMBOL(server_get_mount);
 
-/*
- * Used by mdt to get mount_info from obdname.
- * There are no blocking when using the mount_info.
- * Do not use server_get_mount for this purpose.
+/**
+ * server_put_mount: to be called from obd_cleanup methods
+ * @name:      obd name
+ * @dereg_mnt: 0 or 1 depending on whether the mount is to be deregistered or
+ * not
+ *
+ * The caller decides whether server_deregister_mount() needs to be called or
+ * not. Calling of server_deregister_mount() does not depend on refcounting on
+ * lsi because we could have say the mgs and mds on the same node and we
+ * unmount the mds, then the ref on the lsi would still be non-zero but we
+ * would still want to deregister the mds mount.
  */
-struct lustre_mount_info *server_get_mount_2(const char *name)
-{
-       struct lustre_mount_info *lmi;
-       ENTRY;
-
-       mutex_lock(&lustre_mount_info_lock);
-       lmi = server_find_mount(name);
-       mutex_unlock(&lustre_mount_info_lock);
-       if (!lmi)
-               CERROR("Can't find mount for %s\n", name);
-
-       RETURN(lmi);
-}
-EXPORT_SYMBOL(server_get_mount_2);
-
-/* to be called from obd_cleanup methods */
-int server_put_mount(const char *name)
+int server_put_mount(const char *name, bool dereg_mnt)
 {
        struct lustre_mount_info *lmi;
        struct lustre_sb_info *lsi;
@@ -216,21 +208,14 @@ int server_put_mount(const char *name)
                CDEBUG(D_MOUNT, "Last put of mount %p from %s\n",
                       lmi->lmi_sb, name);
 
-       /* this obd should never need the mount again */
-       server_deregister_mount(name);
+       if (dereg_mnt)
+               /* this obd should never need the mount again */
+               server_deregister_mount(name);
 
        RETURN(0);
 }
 EXPORT_SYMBOL(server_put_mount);
 
-/* Corresponding to server_get_mount_2 */
-int server_put_mount_2(const char *name, struct vfsmount *mnt)
-{
-       ENTRY;
-       RETURN(0);
-}
-EXPORT_SYMBOL(server_put_mount_2);
-
 /* Set up a MGS to serve startup logs */
 static int server_start_mgs(struct super_block *sb)
 {
@@ -255,8 +240,8 @@ static int server_start_mgs(struct super_block *sb)
 
        if (!rc) {
                rc = lustre_start_simple(LUSTRE_MGS_OBDNAME, LUSTRE_MGS_NAME,
-                                        LUSTRE_MGS_OBDNAME, 0, 0,
-                                        lsi->lsi_osd_obdname, 0);
+                                        LUSTRE_MGS_OBDNAME, NULL, NULL,
+                                        lsi->lsi_osd_obdname, NULL);
                /* server_deregister_mount() is not called previously, for lsi
                 * and other stuff can't be freed cleanly when mgs calls
                 * server_put_mount() in error handling case (see b=17758),
@@ -278,8 +263,14 @@ static int server_stop_mgs(struct super_block *sb)
 {
        struct obd_device *obd;
        int rc;
+       struct lustre_mount_info *lmi;
        ENTRY;
 
+       /* Do not stop MGS if this device is not the running MGT */
+       lmi = server_find_mount(LUSTRE_MGS_OBDNAME);
+       if (lmi != NULL && lmi->lmi_sb != sb)
+               RETURN(0);
+
        CDEBUG(D_MOUNT, "Stop MGS service %s\n", LUSTRE_MGS_OBDNAME);
 
        /* There better be only one MGS */
@@ -390,8 +381,9 @@ cleanup:
 }
 EXPORT_SYMBOL(tgt_name2lwp_name);
 
-static CFS_LIST_HEAD(lwp_register_list);
-DEFINE_MUTEX(lwp_register_list_lock);
+static struct list_head lwp_register_list =
+       LIST_HEAD_INIT(lwp_register_list);
+static DEFINE_MUTEX(lwp_register_list_lock);
 
 int lustre_register_lwp_item(const char *lwpname, struct obd_export **exp,
                             register_lwp_cb cb_func, void *cb_data)
@@ -429,8 +421,8 @@ int lustre_register_lwp_item(const char *lwpname, struct obd_export **exp,
        lri->lri_exp = exp;
        lri->lri_cb_func = cb_func;
        lri->lri_cb_data = cb_data;
-       CFS_INIT_LIST_HEAD(&lri->lri_list);
-       cfs_list_add(&lri->lri_list, &lwp_register_list);
+       INIT_LIST_HEAD(&lri->lri_list);
+       list_add(&lri->lri_list, &lwp_register_list);
 
        if (*exp != NULL && cb_func != NULL)
                cb_func(cb_data);
@@ -445,11 +437,11 @@ void lustre_deregister_lwp_item(struct obd_export **exp)
        struct lwp_register_item *lri, *tmp;
 
        mutex_lock(&lwp_register_list_lock);
-       cfs_list_for_each_entry_safe(lri, tmp, &lwp_register_list, lri_list) {
+       list_for_each_entry_safe(lri, tmp, &lwp_register_list, lri_list) {
                if (exp == lri->lri_exp) {
                        if (*exp)
                                class_export_put(*exp);
-                       cfs_list_del(&lri->lri_list);
+                       list_del(&lri->lri_list);
                        OBD_FREE_PTR(lri);
                        break;
                }
@@ -468,7 +460,7 @@ struct obd_export *lustre_find_lwp_by_index(const char *dev, __u32 idx)
        char                      lwp_name[24];
        int                       rc;
 
-       lmi = server_get_mount_2(dev);
+       lmi = server_get_mount(dev);
        if (lmi == NULL)
                return NULL;
 
@@ -477,7 +469,7 @@ struct obd_export *lustre_find_lwp_by_index(const char *dev, __u32 idx)
        if (rc != 0) {
                CERROR("%s: failed to get fsname: rc = %d\n",
                       lsi->lsi_svname, rc);
-               return NULL;
+               goto err_lmi;
        }
 
        snprintf(lwp_name, sizeof(lwp_name), "%s-MDT%04x", fsname, idx);
@@ -492,17 +484,20 @@ struct obd_export *lustre_find_lwp_by_index(const char *dev, __u32 idx)
        }
        spin_unlock(&lsi->lsi_lwp_lock);
 
+err_lmi:
+       server_put_mount(dev, false);
+
        return exp;
 }
 EXPORT_SYMBOL(lustre_find_lwp_by_index);
 
-static void lustre_notify_lwp_list(struct obd_export *exp)
+void lustre_notify_lwp_list(struct obd_export *exp)
 {
        struct lwp_register_item *lri, *tmp;
        LASSERT(exp != NULL);
 
        mutex_lock(&lwp_register_list_lock);
-       cfs_list_for_each_entry_safe(lri, tmp, &lwp_register_list, lri_list) {
+       list_for_each_entry_safe(lri, tmp, &lwp_register_list, lri_list) {
                if (strcmp(exp->exp_obd->obd_name, lri->lri_name))
                        continue;
                if (*lri->lri_exp != NULL)
@@ -513,6 +508,7 @@ static void lustre_notify_lwp_list(struct obd_export *exp)
        }
        mutex_unlock(&lwp_register_list_lock);
 }
+EXPORT_SYMBOL(lustre_notify_lwp_list);
 
 static int lustre_lwp_connect(struct obd_device *lwp)
 {
@@ -543,7 +539,8 @@ static int lustre_lwp_connect(struct obd_device *lwp)
        data->ocd_connect_flags |= OBD_CONNECT_MDS_MDS | OBD_CONNECT_FID |
                OBD_CONNECT_AT | OBD_CONNECT_LRU_RESIZE |
                OBD_CONNECT_FULL20 | OBD_CONNECT_LVB_TYPE |
-               OBD_CONNECT_LIGHTWEIGHT | OBD_CONNECT_LFSCK;
+               OBD_CONNECT_LIGHTWEIGHT | OBD_CONNECT_LFSCK |
+               OBD_CONNECT_BULK_MBITS;
        OBD_ALLOC_PTR(uuid);
        if (uuid == NULL)
                GOTO(out, rc = -ENOMEM);
@@ -563,7 +560,6 @@ static int lustre_lwp_connect(struct obd_device *lwp)
                if (unlikely(lwp->obd_lwp_export != NULL))
                        class_export_put(lwp->obd_lwp_export);
                lwp->obd_lwp_export = class_export_get(exp);
-               lustre_notify_lwp_list(exp);
        }
 
        GOTO(out, rc);
@@ -619,7 +615,7 @@ static int lustre_lwp_setup(struct lustre_cfg *lcfg, struct lustre_sb_info *lsi,
        sprintf(lwpuuid, "%s_UUID", lwpname);
        rc = lustre_start_simple(lwpname, LUSTRE_LWP_NAME,
                                 lwpuuid, lustre_cfg_string(lcfg, 1),
-                                0, 0, 0);
+                                NULL, NULL, NULL);
        if (rc) {
                CERROR("%s: setup up failed: rc %d\n", lwpname, rc);
                GOTO(out, rc);
@@ -711,16 +707,18 @@ static int lustre_lwp_add_conn(struct lustre_cfg *cfg,
                                   lustre_cfg_string(cfg, 1));
 
        lcfg = lustre_cfg_new(LCFG_ADD_CONN, bufs);
-
+       if (lcfg == NULL)
+               GOTO(out_cfg, rc = -ENOMEM);
        rc = class_add_conn(lwp, lcfg);
        if (rc)
                CERROR("%s: can't add conn: rc = %d\n", lwpname, rc);
 
-out:
-       if (bufs != NULL)
-               OBD_FREE_PTR(bufs);
        if (lcfg != NULL)
                lustre_cfg_free(lcfg);
+out_cfg:
+       if (bufs != NULL)
+               OBD_FREE_PTR(bufs);
+out:
        if (lwpname != NULL)
                OBD_FREE(lwpname, MTI_NAME_MAXLEN);
        RETURN(rc);
@@ -839,8 +837,11 @@ static int client_lwp_config_process(const struct lu_env *env,
                        *ptr = '-';
                        break;
                }
-
                *ptr = '-';
+
+               if (IS_MDT(lsi) && idx != 0)
+                       break;
+
                rc = lustre_lwp_add_conn(lcfg, lsi, idx);
                break;
        }
@@ -882,7 +883,7 @@ static int lustre_disconnect_lwp(struct super_block *sb)
                /* end log first */
                cfg->cfg_instance = sb;
                rc = lustre_end_log(sb, logname, cfg);
-               if (rc != 0)
+               if (rc != 0 && rc != -ENOENT)
                        GOTO(out, rc);
 
                lsi->lsi_lwp_started = 0;
@@ -991,8 +992,8 @@ static int lustre_start_lwp(struct super_block *sb)
        cfg->cfg_callback = client_lwp_config_process;
        cfg->cfg_instance = sb;
        rc = lustre_process_log(sb, logname, cfg);
-       if (rc == 0)
-               lsi->lsi_lwp_started = 1;
+       /* need to remove config llog from mgc */
+       lsi->lsi_lwp_started = 1;
 
        GOTO(out, rc);
 
@@ -1004,7 +1005,7 @@ out:
        return rc;
 }
 
-DEFINE_MUTEX(server_start_lock);
+static DEFINE_MUTEX(server_start_lock);
 
 /* Stop MDS/OSS if nobody is using them */
 static int server_stop_servers(int lsiflags)
@@ -1052,7 +1053,6 @@ int server_mti_print(const char *title, struct mgs_target_info *mti)
                  mti->mti_config_ver, mti->mti_flags);
        return 0;
 }
-EXPORT_SYMBOL(server_mti_print);
 
 /* Generate data for registration */
 static int server_lsi2mti(struct lustre_sb_info *lsi,
@@ -1169,8 +1169,8 @@ static int server_register_target(struct lustre_sb_info *lsi)
                                "rc = %d. Is the MGS running?\n",
                                lsi->lsi_svname, rc);
                } else {
-                       CERROR("%s: error registering with the MGS: rc = %d "
-                              "(not fatal)\n", lsi->lsi_svname, rc);
+                       CDEBUG(D_HA, "%s: error registering with the MGS: "
+                              "rc = %d (not fatal)\n", lsi->lsi_svname, rc);
                        /* reset the error code for non-fatal error. */
                        rc = 0;
                }
@@ -1250,7 +1250,7 @@ static int server_start_targets(struct super_block *sb)
                        rc = lustre_start_simple(LUSTRE_MDS_OBDNAME,
                                                 LUSTRE_MDS_NAME,
                                                 LUSTRE_MDS_OBDNAME"_uuid",
-                                                0, 0, 0, 0);
+                                                NULL, NULL, NULL, NULL);
                        if (rc) {
                                mutex_unlock(&server_start_lock);
                                CERROR("failed to start MDS: %d\n", rc);
@@ -1269,7 +1269,7 @@ static int server_start_targets(struct super_block *sb)
                        rc = lustre_start_simple(LUSTRE_OSS_OBDNAME,
                                                 LUSTRE_OSS_NAME,
                                                 LUSTRE_OSS_OBDNAME"_uuid",
-                                                0, 0, 0, 0);
+                                                NULL, NULL, NULL, NULL);
                        if (rc) {
                                mutex_unlock(&server_start_lock);
                                CERROR("failed to start OSS: %d\n", rc);
@@ -1373,6 +1373,8 @@ out_stop_service:
 
 static int lsi_prepare(struct lustre_sb_info *lsi)
 {
+       const char *osd_type;
+       const char *fstype;
        __u32 index;
        int rc;
        ENTRY;
@@ -1386,28 +1388,26 @@ static int lsi_prepare(struct lustre_sb_info *lsi)
                RETURN(-EINVAL);
        }
 
-       if (strlen(lsi->lsi_lmd->lmd_profile) >= sizeof(lsi->lsi_svname))
-               RETURN(-ENAMETOOLONG);
-
-       strcpy(lsi->lsi_svname, lsi->lsi_lmd->lmd_profile);
-
        /* Determine osd type */
-       if (lsi->lsi_lmd->lmd_osd_type != NULL) {
-               if (strlen(lsi->lsi_lmd->lmd_osd_type) >=
-                   sizeof(lsi->lsi_osd_type))
-                       RETURN(-ENAMETOOLONG);
-
-               strcpy(lsi->lsi_osd_type, lsi->lsi_lmd->lmd_osd_type);
+       if (lsi->lsi_lmd->lmd_osd_type == NULL) {
+               osd_type = LUSTRE_OSD_LDISKFS_NAME;
+               fstype = "ldiskfs";
        } else {
-               strcpy(lsi->lsi_osd_type, LUSTRE_OSD_LDISKFS_NAME);
+               osd_type = lsi->lsi_lmd->lmd_osd_type;
+               fstype = lsi->lsi_lmd->lmd_osd_type;
        }
 
+       if (strlen(lsi->lsi_lmd->lmd_profile) >= sizeof(lsi->lsi_svname) ||
+           strlen(osd_type) >= sizeof(lsi->lsi_osd_type) ||
+           strlen(fstype) >= sizeof(lsi->lsi_fstype))
+               RETURN(-ENAMETOOLONG);
+
+       strlcpy(lsi->lsi_svname, lsi->lsi_lmd->lmd_profile,
+               sizeof(lsi->lsi_svname));
+       strlcpy(lsi->lsi_osd_type, osd_type, sizeof(lsi->lsi_osd_type));
        /* XXX: a temp. solution for components using ldiskfs
         *      to be removed in one of the subsequent patches */
-       if (!strcmp(lsi->lsi_lmd->lmd_osd_type, "osd-ldiskfs"))
-               strcpy(lsi->lsi_fstype, "ldiskfs");
-       else
-               strcpy(lsi->lsi_fstype, lsi->lsi_lmd->lmd_osd_type);
+       strlcpy(lsi->lsi_fstype, fstype, sizeof(lsi->lsi_fstype));
 
        /* Determine server type */
        rc = server_name2index(lsi->lsi_svname, &index, NULL);
@@ -1486,9 +1486,13 @@ static void server_put_super(struct super_block *sb)
                   If there are any setup/cleanup errors, save the lov
                   name for safety cleanup later. */
                lprof = class_get_profile(lsi->lsi_svname);
-               if (lprof && lprof->lp_dt) {
-                       OBD_ALLOC(extraname, strlen(lprof->lp_dt) + 1);
-                       strcpy(extraname, lprof->lp_dt);
+               if (lprof != NULL) {
+                       if (lprof->lp_dt != NULL) {
+                               OBD_ALLOC(extraname, strlen(lprof->lp_dt) + 1);
+                               strncpy(extraname, lprof->lp_dt,
+                                       strlen(lprof->lp_dt) + 1);
+                       }
+                       class_put_profile(lprof);
                }
 
                obd = class_name2obd(lsi->lsi_svname);
@@ -1624,7 +1628,7 @@ static ssize_t lustre_listxattr(struct dentry *d_entry, char *name,
        return -EOPNOTSUPP;
 }
 
-const struct inode_operations server_inode_operations = {
+static const struct inode_operations server_inode_operations = {
        .setxattr       = lustre_setxattr,
        .getxattr       = lustre_getxattr,
        .listxattr      = lustre_listxattr,
@@ -1635,7 +1639,7 @@ const struct inode_operations server_inode_operations = {
 
 static int server_fill_super_common(struct super_block *sb)
 {
-       struct inode *root = 0;
+       struct inode *root = NULL;
        ENTRY;
 
        CDEBUG(D_MOUNT, "Server sb, dev=%d\n", (int)sb->s_dev);
@@ -1696,10 +1700,20 @@ static int osd_start(struct lustre_sb_info *lsi, unsigned long mflags)
                        GOTO(out, rc);
                obd = class_name2obd(lsi->lsi_osd_obdname);
                LASSERT(obd);
+       } else {
+               CDEBUG(D_MOUNT, "%s already started\n", lsi->lsi_osd_obdname);
+               /* but continue setup to allow special case of MDT and internal
+                * MGT being started separately. */
+               if (!((IS_MGS(lsi) && (lsi->lsi_lmd->lmd_flags &
+                                     LMD_FLG_NOMGS)) ||
+                    (IS_MDT(lsi) && (lsi->lsi_lmd->lmd_flags &
+                                     LMD_FLG_NOSVC))))
+                       RETURN(-EALREADY);
        }
 
        rc = obd_connect(NULL, &lsi->lsi_osd_exp,
                         obd, &obd->obd_uuid, NULL, NULL);
+
        if (rc) {
                obd->obd_force = 1;
                class_manual_cleanup(obd);
@@ -1732,6 +1746,9 @@ int server_fill_super(struct super_block *sb)
        int rc;
        ENTRY;
 
+       /* to simulate target mount race */
+       OBD_RACE(OBD_FAIL_TGT_MOUNT_RACE);
+
        rc = lsi_prepare(lsi);
        if (rc)
                RETURN(rc);
@@ -1861,4 +1878,3 @@ void server_calc_timeout(struct lustre_sb_info *lsi, struct obd_device *obd)
        obd->obd_recovery_time_hard = hard;
        obd->obd_recovery_ir_factor = factor;
 }
-EXPORT_SYMBOL(server_calc_timeout);