/*
* Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
+ *
+ * Copyright (c) 2011 Whamcloud, Inc.
+ *
*/
/*
* This file is part of Lustre, http://www.lustre.org/
#include <lustre_disk.h>
#include <lustre_param.h>
-static int (*client_fill_super)(struct super_block *sb) = NULL;
+static int (*client_fill_super)(struct super_block *sb,
+ struct vfsmount *mnt) = NULL;
static void (*kill_super_cb)(struct super_block *sb) = NULL;
/*********** mount lookup *********/
#endif
static int ldd_parse(struct lvfs_run_ctxt *mount_ctxt,
- struct lustre_disk_data *ldd)
+ struct lustre_disk_data *ldd)
{
struct lvfs_run_ctxt saved;
struct file *file;
push_ctxt(&saved, mount_ctxt, NULL);
- file = filp_open(MOUNT_DATA_FILE, O_RDWR, 0644);
+ file = filp_open(MOUNT_DATA_FILE, O_RDWR|O_SYNC, 0644);
if (IS_ERR(file)) {
rc = PTR_ERR(file);
CERROR("cannot open %s: rc = %d\n", MOUNT_DATA_FILE, rc);
struct config_llog_instance *cfg)
{
struct lustre_cfg *lcfg;
- struct lustre_cfg_bufs bufs;
+ struct lustre_cfg_bufs *bufs;
struct lustre_sb_info *lsi = s2lsi(sb);
struct obd_device *mgc = lsi->lsi_mgc;
int rc;
LASSERT(mgc);
LASSERT(cfg);
+ OBD_ALLOC_PTR(bufs);
+ if (bufs == NULL)
+ RETURN(-ENOMEM);
+
/* mgc_process_config */
- lustre_cfg_bufs_reset(&bufs, mgc->obd_name);
- lustre_cfg_bufs_set_string(&bufs, 1, logname);
- lustre_cfg_bufs_set(&bufs, 2, cfg, sizeof(*cfg));
- lustre_cfg_bufs_set(&bufs, 3, &sb, sizeof(sb));
- lcfg = lustre_cfg_new(LCFG_LOG_START, &bufs);
+ lustre_cfg_bufs_reset(bufs, mgc->obd_name);
+ lustre_cfg_bufs_set_string(bufs, 1, logname);
+ lustre_cfg_bufs_set(bufs, 2, cfg, sizeof(*cfg));
+ lustre_cfg_bufs_set(bufs, 3, &sb, sizeof(sb));
+ lcfg = lustre_cfg_new(LCFG_LOG_START, bufs);
rc = obd_process_config(mgc, sizeof(*lcfg), lcfg);
lustre_cfg_free(lcfg);
+ OBD_FREE_PTR(bufs);
+
if (rc == -EINVAL)
LCONSOLE_ERROR_MSG(0x15b, "%s: The configuration from log '%s'"
"failed from the MGS (%d). Make sure this "
struct obd_uuid *uuid;
class_uuid_t uuidc;
lnet_nid_t nid;
- char *mgcname, *niduuid, *mgssec;
+ char *mgcname = NULL, *niduuid = NULL, *mgssec = NULL;
char *ptr;
int recov_bk;
int rc = 0, i = 0, j, len;
RETURN(-EINVAL);
}
+ cfs_mutex_down(&mgc_start_lock);
+
len = strlen(LUSTRE_MGC_OBDNAME) + strlen(libcfs_nid2str(nid)) + 1;
OBD_ALLOC(mgcname, len);
OBD_ALLOC(niduuid, len + 2);
mgssec = lsi->lsi_lmd->lmd_mgssec ? lsi->lsi_lmd->lmd_mgssec : "";
- cfs_mutex_down(&mgc_start_lock);
+ OBD_ALLOC_PTR(data);
+ if (data == NULL)
+ GOTO(out_free, rc = -ENOMEM);
obd = class_name2obd(mgcname);
if (obd && !obd->obd_stopping) {
/* Re-using an existing MGC */
cfs_atomic_inc(&obd->u.cli.cl_mgc_refcount);
+ /* IR compatibility check, only for clients */
+ if (lmd_is_client(lsi->lsi_lmd)) {
+ int has_ir;
+ int vallen = sizeof(*data);
+ __u32 *flags = &lsi->lsi_lmd->lmd_flags;
+
+ rc = obd_get_info(obd->obd_self_export,
+ strlen(KEY_CONN_DATA), KEY_CONN_DATA,
+ &vallen, data, NULL);
+ LASSERT(rc == 0);
+ has_ir = OCD_HAS_FLAG(data, IMP_RECOV);
+ if (has_ir ^ !(*flags & LMD_FLG_NOIR)) {
+ /* LMD_FLG_NOIR is for test purpose only */
+ LCONSOLE_WARN(
+ "Trying to mount a client with IR setting "
+ "not compatible with current mgc. "
+ "Force to use current mgc setting that is "
+ "IR %s.\n",
+ has_ir ? "enabled" : "disabled");
+ if (has_ir)
+ *flags &= ~LMD_FLG_NOIR;
+ else
+ *flags |= LMD_FLG_NOIR;
+ }
+ }
+
recov_bk = 0;
/* If we are restarting the MGS, don't try to keep the MGC's
old connection, or registration will fail. */
/* nonfatal */
CWARN("can't set %s %d\n", KEY_INIT_RECOV_BACKUP, rc);
/* We connect to the MGS at setup, and don't disconnect until cleanup */
- OBD_ALLOC_PTR(data);
- if (data == NULL)
- GOTO(out, rc = -ENOMEM);
data->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_FID |
- OBD_CONNECT_AT | OBD_CONNECT_FULL20;
+ OBD_CONNECT_AT | OBD_CONNECT_FULL20 |
+ OBD_CONNECT_IMP_RECOV;
+ if (lmd_is_client(lsi->lsi_lmd) &&
+ lsi->lsi_lmd->lmd_flags & LMD_FLG_NOIR)
+ data->ocd_connect_flags &= ~OBD_CONNECT_IMP_RECOV;
data->ocd_version = LUSTRE_VERSION_CODE;
rc = obd_connect(NULL, &exp, obd, &(obd->obd_uuid), data, NULL);
- OBD_FREE_PTR(data);
if (rc) {
CERROR("connect failed %d\n", rc);
GOTO(out, rc);
out_free:
cfs_mutex_up(&mgc_start_lock);
+ if (data)
+ OBD_FREE_PTR(data);
if (mgcname)
OBD_FREE(mgcname, len);
if (niduuid)
if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
continue;
- if (class_find_param(ldd->ldd_params,
- PARAM_NETWORK, NULL) == 0 &&
- !class_match_net(ldd->ldd_params, id.nid)) {
- /* can't match specified network */
+ /* server use --servicenode param, only allow specified
+ * nids be registered */
+ if ((ldd->ldd_flags & LDD_F_NO_PRIMNODE) != 0 &&
+ class_match_nid(ldd->ldd_params,
+ PARAM_FAILNODE, id.nid) < 1)
+ continue;
+
+ /* match specified network */
+ if (!class_match_net(ldd->ldd_params,
+ PARAM_NETWORK, LNET_NIDNET(id.nid)))
continue;
- }
mti->mti_nids[mti->mti_nid_count] = id.nid;
mti->mti_nid_count++;
struct obd_device *mgc = lsi->lsi_mgc;
struct lustre_disk_data *ldd = lsi->lsi_ldd;
struct mgs_target_info *mti = NULL;
+ bool writeconf;
int rc;
ENTRY;
libcfs_nid2str(mti->mti_nids[0]), mti->mti_stripe_index,
mti->mti_flags);
+ /* if write_conf is true, the registration must succeed */
+ writeconf = !!(ldd->ldd_flags & (LDD_F_NEED_INDEX | LDD_F_UPDATE));
+ mti->mti_flags |= LDD_F_OPC_REG;
+
/* Register the target */
/* FIXME use mgc_process_config instead */
rc = obd_set_info_async(mgc->u.cli.cl_mgc_mgsexp,
sizeof(KEY_REGISTER_TARGET), KEY_REGISTER_TARGET,
sizeof(*mti), mti, NULL);
- if (rc)
+ if (rc) {
+ if (mti->mti_flags & LDD_F_ERROR) {
+ LCONSOLE_ERROR_MSG(0x160,
+ "The MGS is refusing to allow this "
+ "server (%s) to start. Please see messages"
+ " on the MGS node.\n", ldd->ldd_svname);
+ } else if (writeconf) {
+ LCONSOLE_ERROR_MSG(0x15f,
+ "Communication to the MGS return error %d. "
+ "Is the MGS running?\n", rc);
+ } else {
+ CERROR("Cannot talk to the MGS: %d, not fatal\n", rc);
+ /* reset the error code for non-fatal error. */
+ rc = 0;
+ }
GOTO(out, rc);
+ }
/* Always update our flags */
- ldd->ldd_flags = mti->mti_flags & ~LDD_F_REWRITE_LDD;
+ ldd->ldd_flags = mti->mti_flags & LDD_F_ONDISK_MASK;
/* If this flag is set, it means the MGS wants us to change our
on-disk data. (So far this means just the index.) */
RETURN(rc);
}
+/**
+ * Notify the MGS that this target is ready.
+ * Used by IR - if the MGS receives this message, it will notify clients.
+ */
+static int server_notify_target(struct super_block *sb, struct obd_device *obd)
+{
+ struct lustre_sb_info *lsi = s2lsi(sb);
+ struct obd_device *mgc = lsi->lsi_mgc;
+ struct mgs_target_info *mti = NULL;
+ int rc;
+ ENTRY;
+
+ LASSERT(mgc);
+
+ if (!(lsi->lsi_flags & LSI_SERVER))
+ RETURN(-EINVAL);
+
+ OBD_ALLOC_PTR(mti);
+ if (!mti)
+ RETURN(-ENOMEM);
+ rc = server_sb2mti(sb, mti);
+ if (rc)
+ GOTO(out, rc);
+
+ mti->mti_instance = obd->u.obt.obt_instance;
+ mti->mti_flags |= LDD_F_OPC_READY;
+
+ /* FIXME use mgc_process_config instead */
+ rc = obd_set_info_async(mgc->u.cli.cl_mgc_mgsexp,
+ sizeof(KEY_REGISTER_TARGET),
+ KEY_REGISTER_TARGET,
+ sizeof(*mti), mti, NULL);
+
+ /* Imperative recovery: if the mgs informs us to use IR? */
+ if (!rc && !(mti->mti_flags & LDD_F_ERROR) &&
+ (mti->mti_flags & LDD_F_IR_CAPABLE))
+ lsi->lsi_flags |= LSI_IR_CAPABLE;
+
+out:
+ if (mti)
+ OBD_FREE_PTR(mti);
+ RETURN(rc);
+
+}
+
/** Start server targets: MDTs and OSTs
*/
static int server_start_targets(struct super_block *sb, struct vfsmount *mnt)
#endif
/* If we're an OST, make sure the global OSS is running */
- if (lsi->lsi_ldd->ldd_flags & LDD_F_SV_TYPE_OST) {
+ if (IS_OST(lsi->lsi_ldd)) {
/* make sure OSS is started */
cfs_mutex_down(&server_start_lock);
obd = class_name2obd(LUSTRE_OSS_OBDNAME);
/* Register with MGS */
rc = server_register_target(sb);
- if (rc && (lsi->lsi_ldd->ldd_flags &
- (LDD_F_NEED_INDEX | LDD_F_UPDATE | LDD_F_UPGRADE14))){
- CERROR("Required registration failed for %s: %d\n",
- lsi->lsi_ldd->ldd_svname, rc);
- if (rc == -EIO) {
- LCONSOLE_ERROR_MSG(0x15f, "Communication error with "
- "the MGS. Is the MGS running?\n");
- }
- GOTO(out_mgc, rc);
- }
- if (rc == -EINVAL) {
- LCONSOLE_ERROR_MSG(0x160, "The MGS is refusing to allow this "
- "server (%s) to start. Please see messages"
- " on the MGS node.\n",
- lsi->lsi_ldd->ldd_svname);
- GOTO(out_mgc, rc);
- }
- /* non-fatal error of registeration with MGS */
if (rc)
- CDEBUG(D_MOUNT, "Cannot register with MGS: %d\n", rc);
+ GOTO(out_mgc, rc);
/* Let the target look up the mount using the target's name
(we can't pass the sb or mnt through class_process_config.) */
obd->obd_self_export, 0, NULL, NULL);
}
+ server_notify_target(sb, obd);
+
/* log has been fully processed */
obd_notify(obd, NULL, OBD_NOTIFY_CONFIG, (void *)CONFIG_LOG);
+
+ /* calculate recovery timeout, do it after lustre_process_log */
+ server_calc_timeout(lsi, obd);
}
RETURN(rc);
waited);
/* Cannot use l_event_wait() for an interruptible sleep. */
waited += 3;
- blocked = l_w_e_set_sigs(sigmask(SIGKILL));
+ blocked = cfs_block_sigsinv(sigmask(SIGKILL));
cfs_waitq_wait_event_interruptible_timeout(
waitq,
(atomic_read(&mnt->mnt_count) == 1),
cfs_block_sigs(blocked);
if (rc < 0) {
LCONSOLE_EMERG("Danger: interrupted umount %s with "
- "%d refs!\n",
- mnt->mnt_devname,
+ "%d refs!\n", mnt->mnt_devname,
atomic_read(&mnt->mnt_count));
break;
}
return rc;
}
+/*
+ * Calculate timeout value for a target.
+ */
+void server_calc_timeout(struct lustre_sb_info *lsi, struct obd_device *obd)
+{
+ struct lustre_mount_data *lmd;
+ int soft = 0;
+ int hard = 0;
+ int factor = 0;
+ bool has_ir = !!(lsi->lsi_flags & LSI_IR_CAPABLE);
+ int min = OBD_RECOVERY_TIME_MIN;
+
+ LASSERT(lsi->lsi_flags & LSI_SERVER);
+
+ lmd = lsi->lsi_lmd;
+ if (lmd) {
+ soft = lmd->lmd_recovery_time_soft;
+ hard = lmd->lmd_recovery_time_hard;
+ has_ir = has_ir && !(lmd->lmd_flags & LMD_FLG_NOIR);
+ obd->obd_no_ir = !has_ir;
+ }
+
+ if (soft == 0)
+ soft = OBD_RECOVERY_TIME_SOFT;
+ if (hard == 0)
+ hard = OBD_RECOVERY_TIME_HARD;
+
+ /* target may have ir_factor configured. */
+ factor = OBD_IR_FACTOR_DEFAULT;
+ if (obd->obd_recovery_ir_factor)
+ factor = obd->obd_recovery_ir_factor;
+
+ if (has_ir) {
+ int new_soft = soft;
+ int new_hard = hard;
+
+ /* adjust timeout value by imperative recovery */
+
+ new_soft = (soft * factor) / OBD_IR_FACTOR_MAX;
+ new_hard = (hard * factor) / OBD_IR_FACTOR_MAX;
+
+ /* make sure the timeout is not too short */
+ new_soft = max(min, new_soft);
+ new_hard = max(new_soft, new_hard);
+
+ LCONSOLE_INFO("%s: Imperative Recovery enabled, recovery "
+ "window shrunk from %d-%d down to %d-%d\n",
+ obd->obd_name, soft, hard, new_soft, new_hard);
+
+ soft = new_soft;
+ hard = new_hard;
+ }
+
+ /* we're done */
+ obd->obd_recovery_timeout = soft;
+ obd->obd_recovery_time_hard = hard;
+ obd->obd_recovery_ir_factor = factor;
+}
+EXPORT_SYMBOL(server_calc_timeout);
+
/*************** mount common betweeen server and client ***************/
/* Common umount */
s1 = options;
while (*s1) {
int clear = 0;
- int time_min = 2 * (CONNECTION_SWITCH_MAX +
- 2 * INITIAL_CONNECT_TIMEOUT);
+ int time_min = OBD_RECOVERY_TIME_MIN;
/* Skip whitespace and extra commas */
while (*s1 == ' ' || *s1 == ',')
lmd->lmd_recovery_time_hard = max_t(int,
simple_strtoul(s1 + 19, NULL, 10), time_min);
clear++;
+ } else if (strncmp(s1, "noir", 4) == 0) {
+ lmd->lmd_flags |= LMD_FLG_NOIR; /* test purpose only. */
+ clear++;
} else if (strncmp(s1, "nosvc", 5) == 0) {
lmd->lmd_flags |= LMD_FLG_NOSVC;
clear++;
s1 = strstr(devname, ":/");
if (s1) {
++s1;
- lmd->lmd_flags = LMD_FLG_CLIENT;
+ lmd->lmd_flags |= LMD_FLG_CLIENT;
/* Remove leading /s from fsname */
while (*++s1 == '/') ;
/* Freed in lustre_free_lsi */
RETURN(-EINVAL);
}
+struct lustre_mount_data2 {
+ void *lmd2_data;
+ struct vfsmount *lmd2_mnt;
+};
/** This is the entry point for the mount call into Lustre.
* This is called when a server or client is mounted,
int lustre_fill_super(struct super_block *sb, void *data, int silent)
{
struct lustre_mount_data *lmd;
+ struct lustre_mount_data2 *lmd2 = data;
struct lustre_sb_info *lsi;
int rc;
ENTRY;
cfs_lockdep_off();
/* Figure out the lmd from the mount options */
- if (lmd_parse((char *)data, lmd)) {
+ if (lmd_parse((char *)(lmd2->lmd2_data), lmd)) {
lustre_put_lsi(sb);
GOTO(out, rc = -EINVAL);
}
}
/* Connect and start */
/* (should always be ll_fill_super) */
- rc = (*client_fill_super)(sb);
+ rc = (*client_fill_super)(sb, lmd2->lmd2_mnt);
/* c_f_s will call lustre_common_put_super on failure */
}
} else {
/* We can't call ll_fill_super by name because it lives in a module that
must be loaded after this one. */
-void lustre_register_client_fill_super(int (*cfs)(struct super_block *sb))
+void lustre_register_client_fill_super(int (*cfs)(struct super_block *sb,
+ struct vfsmount *mnt))
{
client_fill_super = cfs;
}
/***************** FS registration ******************/
#if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,18))
-struct super_block * lustre_get_sb(struct file_system_type *fs_type,
- int flags, const char *devname, void * data)
+struct super_block * lustre_get_sb(struct file_system_type *fs_type, int flags,
+ const char *devname, void * data)
{
return get_sb_nodev(fs_type, flags, data, lustre_fill_super);
}
#else
-int lustre_get_sb(struct file_system_type *fs_type,
- int flags, const char *devname, void * data,
- struct vfsmount *mnt)
+int lustre_get_sb(struct file_system_type *fs_type, int flags,
+ const char *devname, void * data, struct vfsmount *mnt)
{
- return get_sb_nodev(fs_type, flags, data, lustre_fill_super, mnt);
+ struct lustre_mount_data2 lmd2 = {data, mnt};
+
+ return get_sb_nodev(fs_type, flags, &lmd2, lustre_fill_super, mnt);
}
#endif
git://git.whamcloud.com / fs / lustre-release.git / blobdiff