Whamcloud - gitweb
LU-12355 llite: MS_* flags and SB_* flags split
[fs/lustre-release.git] / lustre / obdclass / obd_mount_server.c
index afd51ac..7842e1c 100644 (file)
@@ -23,7 +23,7 @@
  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  * Use is subject to license terms.
  *
- * Copyright (c) 2013, 2016, Intel Corporation.
+ * Copyright (c) 2013, 2017, Intel Corporation.
  */
 /*
  * This file is part of Lustre, http://www.lustre.org/
 #include <linux/smp_lock.h>
 #endif
 
-#include <lustre/lustre_idl.h>
-#include <lustre/lustre_user.h>
-
 #include <llog_swab.h>
 #include <lustre_disk.h>
-#include <uapi/linux/lustre_ioctl.h>
+#include <uapi/linux/lustre/lustre_ioctl.h>
 #include <lustre_log.h>
-#include <uapi/linux/lustre_param.h>
+#include <uapi/linux/lustre/lustre_param.h>
 #include <obd.h>
 #include <obd_class.h>
 
@@ -455,18 +452,34 @@ EXPORT_SYMBOL(lustre_register_lwp_item);
 void lustre_deregister_lwp_item(struct obd_export **exp)
 {
        struct lwp_register_item *lri;
+       bool removed = false;
+       int repeat = 0;
 
        spin_lock(&lwp_register_list_lock);
        list_for_each_entry(lri, &lwp_register_list, lri_list) {
                if (exp == lri->lri_exp) {
                        list_del_init(&lri->lri_list);
-                       spin_unlock(&lwp_register_list_lock);
-
-                       lustre_put_lwp_item(lri);
-                       return;
+                       removed = true;
+                       break;
                }
        }
        spin_unlock(&lwp_register_list_lock);
+
+       if (!removed)
+               return;
+
+       /* See lustre_notify_lwp_list(), in some extreme race conditions,
+        * the notify callback could be still on the fly, we need to wait
+        * for the callback done before moving on to free the data used
+        * by callback. */
+       while (atomic_read(&lri->lri_ref) > 1) {
+               CDEBUG(D_MOUNT, "lri reference count %u, repeat: %d\n",
+                      atomic_read(&lri->lri_ref), repeat);
+               repeat++;
+               set_current_state(TASK_INTERRUPTIBLE);
+               schedule_timeout(cfs_time_seconds(1));
+       }
+       lustre_put_lwp_item(lri);
 }
 EXPORT_SYMBOL(lustre_deregister_lwp_item);
 
@@ -543,7 +556,7 @@ again:
 }
 EXPORT_SYMBOL(lustre_notify_lwp_list);
 
-static int lustre_lwp_connect(struct obd_device *lwp)
+static int lustre_lwp_connect(struct obd_device *lwp, bool is_mdt)
 {
        struct lu_env            env;
        struct lu_context        session_ctx;
@@ -569,11 +582,14 @@ static int lustre_lwp_connect(struct obd_device *lwp)
 
        data->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_INDEX;
        data->ocd_version = LUSTRE_VERSION_CODE;
-       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_BULK_MBITS;
+       data->ocd_connect_flags |= 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_BULK_MBITS;
+
+       if (is_mdt)
+               data->ocd_connect_flags |= OBD_CONNECT_MDS_MDS;
+
        OBD_ALLOC_PTR(uuid);
        if (uuid == NULL)
                GOTO(out, rc = -ENOMEM);
@@ -657,7 +673,7 @@ static int lustre_lwp_setup(struct lustre_cfg *lcfg, struct lustre_sb_info *lsi,
        obd = class_name2obd(lwpname);
        LASSERT(obd != NULL);
 
-       rc = lustre_lwp_connect(obd);
+       rc = lustre_lwp_connect(obd, strstr(lsi->lsi_svname, "-MDT") != NULL);
        if (rc == 0) {
                obd->u.cli.cl_max_mds_easize = MAX_MD_SIZE;
                spin_lock(&lsi->lsi_lwp_lock);
@@ -918,7 +934,7 @@ static int lustre_disconnect_lwp(struct super_block *sb)
                        GOTO(out, rc = -ENOMEM);
 
                /* end log first */
-               cfg->cfg_instance = sb;
+               cfg->cfg_instance = ll_get_cfg_instance(sb);
                rc = lustre_end_log(sb, logname, cfg);
                if (rc != 0 && rc != -ENOENT)
                        GOTO(out, rc);
@@ -1030,7 +1046,7 @@ static int lustre_start_lwp(struct super_block *sb)
                GOTO(out, rc = -ENOMEM);
 
        cfg->cfg_callback = client_lwp_config_process;
-       cfg->cfg_instance = sb;
+       cfg->cfg_instance = ll_get_cfg_instance(sb);
        rc = lustre_process_log(sb, logname, cfg);
        /* need to remove config llog from mgc */
        lsi->lsi_lwp_started = 1;
@@ -1174,6 +1190,7 @@ static int server_register_target(struct lustre_sb_info *lsi)
        struct mgs_target_info *mti = NULL;
        bool writeconf;
        int rc;
+       int tried = 0;
        ENTRY;
 
        LASSERT(mgc);
@@ -1198,6 +1215,7 @@ static int server_register_target(struct lustre_sb_info *lsi)
        writeconf = !!(lsi->lsi_flags & (LDD_F_NEED_INDEX | LDD_F_UPDATE));
        mti->mti_flags |= LDD_F_OPC_REG;
 
+again:
        /* Register the target */
        /* FIXME use mgc_process_config instead */
        rc = obd_set_info_async(NULL, mgc->u.cli.cl_mgc_mgsexp,
@@ -1211,6 +1229,16 @@ static int server_register_target(struct lustre_sb_info *lsi)
                                "to start: rc = %d. Please see messages on "
                                "the MGS.\n", lsi->lsi_svname, rc);
                } else if (writeconf) {
+                       if ((rc == -ESHUTDOWN || rc == -EIO) && ++tried < 5) {
+                               /* The connection with MGS is not established.
+                                * Try again after 2 seconds. Interruptable. */
+                               set_current_state(TASK_INTERRUPTIBLE);
+                               schedule_timeout(cfs_time_seconds(2));
+                               set_current_state(TASK_RUNNING);
+                               if (!signal_pending(current))
+                                       goto again;
+                       }
+
                        LCONSOLE_ERROR_MSG(0x15f,
                                "%s: cannot register this server with the MGS: "
                                "rc = %d. Is the MGS running?\n",
@@ -1648,6 +1676,63 @@ static int server_statfs(struct dentry *dentry, struct kstatfs *buf)
        RETURN(0);
 }
 
+#ifdef HAVE_SUPEROPS_USE_DENTRY
+int server_show_options(struct seq_file *seq, struct dentry *dentry)
+#else
+int server_show_options(struct seq_file *seq, struct vfsmount *vfs)
+#endif
+{
+       struct lustre_sb_info *lsi;
+       struct lustre_mount_data *lmd;
+
+#ifdef HAVE_SUPEROPS_USE_DENTRY
+       LASSERT(seq != NULL && dentry != NULL);
+       lsi = s2lsi(dentry->d_sb);
+#else
+       LASSERT(seq != NULL && vfs != NULL);
+       lsi = s2lsi(vfs->mnt_sb);
+#endif
+
+       lmd = lsi->lsi_lmd;
+       seq_printf(seq, ",svname=%s", lmd->lmd_profile);
+
+       if  (lmd->lmd_flags & LMD_FLG_ABORT_RECOV)
+               seq_puts(seq, ",abort_recov");
+
+       if (lmd->lmd_flags & LMD_FLG_NOIR)
+               seq_puts(seq, ",noir");
+
+       if (lmd->lmd_flags & LMD_FLG_NOSVC)
+               seq_puts(seq, ",nosvc");
+
+       if (lmd->lmd_flags & LMD_FLG_NOMGS)
+               seq_puts(seq, ",nomgs");
+
+       if (lmd->lmd_flags & LMD_FLG_NOSCRUB)
+               seq_puts(seq, ",noscrub");
+       if (lmd->lmd_flags & LMD_FLG_SKIP_LFSCK)
+               seq_puts(seq, ",skip_lfsck");
+
+       if (lmd->lmd_flags & LMD_FLG_DEV_RDONLY)
+               seq_puts(seq, ",rdonly_dev");
+
+       if (lmd->lmd_flags & LMD_FLG_MGS)
+               seq_puts(seq, ",mgs");
+
+       if (lmd->lmd_mgs != NULL)
+               seq_printf(seq, ",mgsnode=%s", lmd->lmd_mgs);
+
+       if (lmd->lmd_osd_type != NULL)
+               seq_printf(seq, ",osd=%s", lmd->lmd_osd_type);
+
+       if (lmd->lmd_opts != NULL) {
+               seq_putc(seq, ',');
+               seq_puts(seq, lmd->lmd_opts);
+       }
+
+       RETURN(0);
+}
+
 /** The operations we support directly on the superblock:
  * mount, umount, and df.
  */
@@ -1655,11 +1740,13 @@ static struct super_operations server_ops = {
        .put_super      = server_put_super,
        .umount_begin   = server_umount_begin, /* umount -f */
        .statfs         = server_statfs,
+       .show_options   = server_show_options,
 };
 
 /*
  * Xattr support for Lustre servers
  */
+#ifdef HAVE_IOP_XATTR
 static ssize_t lustre_getxattr(struct dentry *dentry, const char *name,
                                void *buffer, size_t size)
 {
@@ -1673,6 +1760,7 @@ static int lustre_setxattr(struct dentry *dentry, const char *name,
 {
        return -EOPNOTSUPP;
 }
+#endif
 
 static ssize_t lustre_listxattr(struct dentry *d_entry, char *name,
                                size_t size)
@@ -1680,12 +1768,55 @@ static ssize_t lustre_listxattr(struct dentry *d_entry, char *name,
        return -EOPNOTSUPP;
 }
 
+static bool is_cmd_supported(unsigned int command)
+{
+       switch (command) {
+       case FITRIM:
+               return true;
+       default:
+               return false;
+       }
+
+       return false;
+}
+
+static long server_ioctl(struct file *filp, unsigned int command,
+                        unsigned long arg)
+{
+       struct file active_filp;
+       struct inode *inode = file_inode(filp);
+       struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
+       struct super_block *dd_sb = dt_mnt_sb_get(lsi->lsi_dt_dev);
+       struct inode *active_inode;
+       int err = -EOPNOTSUPP;
+
+       if (IS_ERR(dd_sb) || !is_cmd_supported(command))
+               return err;
+
+       active_inode = igrab(dd_sb->s_root->d_inode);
+       if (!active_inode)
+               return -EACCES;
+
+       active_filp.f_inode = active_inode;
+       if (active_inode->i_fop && active_inode->i_fop->unlocked_ioctl)
+               err = active_inode->i_fop->unlocked_ioctl(&active_filp,
+                                                         command, arg);
+       iput(active_inode);
+       return err;
+}
+
 static const struct inode_operations server_inode_operations = {
+#ifdef HAVE_IOP_XATTR
        .setxattr       = lustre_setxattr,
        .getxattr       = lustre_getxattr,
+#endif
        .listxattr      = lustre_listxattr,
 };
 
+static const struct file_operations server_file_operations = {
+       .unlocked_ioctl = server_ioctl,
+};
+
 #define log2(n) ffz(~(n))
 #define LUSTRE_SUPER_MAGIC 0x0BD00BD1
 
@@ -1700,7 +1831,7 @@ static int server_fill_super_common(struct super_block *sb)
        sb->s_blocksize_bits = log2(sb->s_blocksize);
        sb->s_magic = LUSTRE_SUPER_MAGIC;
        sb->s_maxbytes = 0; /* we don't allow file IO on server mountpoints */
-       sb->s_flags |= MS_RDONLY;
+       sb->s_flags |= SB_RDONLY;
        sb->s_op = &server_ops;
 
        root = new_inode(sb);
@@ -1714,6 +1845,7 @@ static int server_fill_super_common(struct super_block *sb)
        /* apparently we need to be a directory for the mount to finish */
        root->i_mode = S_IFDIR;
        root->i_op = &server_inode_operations;
+       root->i_fop = &server_file_operations;
        sb->s_root = d_make_root(root);
        if (!sb->s_root) {
                CERROR("%s: can't make root dentry\n", sb->s_id);
@@ -1726,10 +1858,10 @@ static int server_fill_super_common(struct super_block *sb)
 static int osd_start(struct lustre_sb_info *lsi, unsigned long mflags)
 {
        struct lustre_mount_data *lmd = lsi->lsi_lmd;
-       struct obd_device        *obd;
-       struct dt_device_param    p;
-       char                      flagstr[16];
-       int                       rc;
+       struct obd_device *obd;
+       struct dt_device_param p;
+       char flagstr[20 + 1 + 10 + 1];
+       int rc;
        ENTRY;
 
        CDEBUG(D_MOUNT,
@@ -1739,7 +1871,7 @@ static int osd_start(struct lustre_sb_info *lsi, unsigned long mflags)
        sprintf(lsi->lsi_osd_obdname, "%s-osd", lsi->lsi_svname);
        strcpy(lsi->lsi_osd_uuid, lsi->lsi_osd_obdname);
        strcat(lsi->lsi_osd_uuid, "_UUID");
-       sprintf(flagstr, "%lu:%lu", mflags, (unsigned long) lmd->lmd_flags);
+       snprintf(flagstr, sizeof(flagstr), "%lu:%u", mflags, lmd->lmd_flags);
 
        obd = class_name2obd(lsi->lsi_osd_obdname);
        if (obd == NULL) {
@@ -1802,8 +1934,10 @@ int server_fill_super(struct super_block *sb)
        OBD_RACE(OBD_FAIL_TGT_MOUNT_RACE);
 
        rc = lsi_prepare(lsi);
-       if (rc)
+       if (rc) {
+               lustre_put_lsi(sb);
                RETURN(rc);
+       }
 
        /* Start low level OSD */
        rc = osd_start(lsi, sb->s_flags);