#define PRINT_MASK (D_SUPER | D_CONFIG)
#include <linux/types.h>
+#ifdef HAVE_LINUX_SELINUX_IS_ENABLED
#include <linux/selinux.h>
+#endif
#include <linux/statfs.h>
#include <linux/version.h>
-#ifdef HAVE_KERNEL_LOCKED
-#include <linux/smp_lock.h>
-#endif
+#include <linux/delay.h>
#include <llog_swab.h>
#include <lustre_disk.h>
OBD_FREE(lmi, sizeof(*lmi));
mutex_unlock(&lustre_mount_info_lock);
+ OBD_RACE(OBD_FAIL_MDS_LLOG_UMOUNT_RACE);
RETURN(0);
}
RETURN(-ENOMEM);
}
memcpy(uuid->uuid, lwpname, strlen(lwpname));
- *exp = cfs_hash_lookup(lwp->obd_uuid_hash, uuid);
+ *exp = obd_uuid_lookup(lwp, uuid);
OBD_FREE_PTR(uuid);
}
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));
+ schedule_timeout_interruptible(cfs_time_seconds(1));
}
lustre_put_lwp_item(lri);
}
}
snprintf(lwp_name, sizeof(lwp_name), "%s-MDT%04x", fsname, idx);
- spin_lock(&lsi->lsi_lwp_lock);
+ mutex_lock(&lsi->lsi_lwp_mutex);
list_for_each_entry(lwp, &lsi->lsi_lwp_list, obd_lwp_list) {
char *ptr = strstr(lwp->obd_name, lwp_name);
break;
}
}
- spin_unlock(&lsi->lsi_lwp_lock);
+ mutex_unlock(&lsi->lsi_lwp_mutex);
err_lmi:
server_put_mount(dev, false);
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);
+ mutex_lock(&lsi->lsi_lwp_mutex);
list_add_tail(&obd->obd_lwp_list, &lsi->lsi_lwp_list);
- spin_unlock(&lsi->lsi_lwp_lock);
+ mutex_unlock(&lsi->lsi_lwp_mutex);
} else {
CERROR("%s: connect failed: rc = %d\n", lwpname, rc);
}
if (!is_mdc_device(devname))
break;
+ if (!(cfg->cfg_flags & CFG_F_MARKER)) {
+ CDEBUG(D_CONFIG, "Skipping add_conn for %s, rec %d\n",
+ devname, rec->lrh_index);
+ break;
+ }
+
+ /* add_conn should follow by add_uuid. This
+ * guarantee lwp device was created
+ */
+ if (!(cfg->cfg_flags & CFG_F_SKIP)) {
+ CWARN("Error at config for %s rec %d, add_conn should follow by add_uuid\n",
+ devname, rec->lrh_index);
+ break;
+ }
ptr = strrchr(devname, '-');
if (ptr == NULL)
break;
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);
if (bufs == NULL)
GOTO(out, rc = -ENOMEM);
+ mutex_lock(&lsi->lsi_lwp_mutex);
list_for_each_entry(lwp, &lsi->lsi_lwp_list, obd_lwp_list) {
struct lustre_cfg *lcfg;
lustre_cfg_bufs_set_string(bufs, 1, NULL);
OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount,
bufs->lcfg_buflen));
- if (!lcfg)
- GOTO(out, rc = -ENOMEM);
+ if (!lcfg) {
+ rc = -ENOMEM;
+ break;
+ }
lustre_cfg_init(lcfg, LCFG_CLEANUP, bufs);
/* Disconnect import first. NULL is passed for the '@env',
rc1 = rc;
}
}
+ mutex_unlock(&lsi->lsi_lwp_mutex);
GOTO(out, rc);
int rc1 = 0;
ENTRY;
+ mutex_lock(&lsi->lsi_lwp_mutex);
while (!list_empty(&lsi->lsi_lwp_list)) {
lwp = list_entry(lsi->lsi_lwp_list.next, struct obd_device,
obd_lwp_list);
list_del_init(&lwp->obd_lwp_list);
lwp->obd_force = 1;
+ mutex_unlock(&lsi->lsi_lwp_mutex);
+
rc = class_manual_cleanup(lwp);
if (rc != 0) {
CERROR("%s: fail to stop LWP: rc = %d\n",
lwp->obd_name, rc);
rc1 = rc;
}
+ mutex_lock(&lsi->lsi_lwp_mutex);
}
+ mutex_unlock(&lsi->lsi_lwp_mutex);
RETURN(rc1 != 0 ? rc1 : rc);
}
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;
struct obd_device *obd = NULL;
struct obd_type *type = NULL;
int rc = 0;
+ bool type_last;
ENTRY;
mutex_lock(&server_start_lock);
/* if this was an MDT, and there are no more MDT's, clean up the MDS */
if (lsiflags & LDD_F_SV_TYPE_MDT) {
obd = class_name2obd(LUSTRE_MDS_OBDNAME);
- if (obd != NULL)
- type = class_search_type(LUSTRE_MDT_NAME);
- }
-
+ type = class_search_type(LUSTRE_MDT_NAME);
+ } else if (lsiflags & LDD_F_SV_TYPE_OST) {
/* if this was an OST, and there are no more OST's, clean up the OSS */
- if (lsiflags & LDD_F_SV_TYPE_OST) {
obd = class_name2obd(LUSTRE_OSS_OBDNAME);
- if (obd != NULL)
- type = class_search_type(LUSTRE_OST_NAME);
+ type = class_search_type(LUSTRE_OST_NAME);
}
- if (obd != NULL && (type == NULL || type->typ_refcnt == 0)) {
+ /* server_stop_servers is a pair of server_start_targets
+ * Here we put type which was taken at server_start_targets.
+ * If type is NULL then there is a wrong logic around type or
+ * type reference. */
+ LASSERTF(type, "Server flags %d, obd %s\n", lsiflags,
+ obd ? obd->obd_name : "NULL");
+
+ type_last = (atomic_read(&type->typ_refcnt) == 1);
+
+ class_put_type(type);
+ if (obd != NULL && type_last) {
obd->obd_force = 1;
/* obd_fail doesn't mean much on a server obd */
rc = class_manual_cleanup(obd);
}
+ /* put reference taken by class_search_type */
+ kobject_put(&type->typ_kobj);
+
mutex_unlock(&server_start_lock);
RETURN(rc);
mti->mti_nid_count = 0;
while (LNetGetId(i++, &id) != -ENOENT) {
- if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
+ if (id.nid == LNET_NID_LO_0)
continue;
/* server use --servicenode param, only allow specified
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(
- msecs_to_jiffies(MSEC_PER_SEC) * 2);
- set_current_state(TASK_RUNNING);
+ schedule_timeout_interruptible(
+ cfs_time_seconds(2));
if (!signal_pending(current))
goto again;
}
struct config_llog_instance cfg;
struct lu_env mgc_env;
struct lu_device *dev;
+ char *name_service, *obd_name_service = NULL;
+ struct obd_type *type = NULL;
int rc;
ENTRY;
CDEBUG(D_MOUNT, "starting target %s\n", lsi->lsi_svname);
+ LASSERTF(IS_MDT(lsi) || IS_OST(lsi), "designed for MDT or OST only\n");
+
if (IS_MDT(lsi)) {
- /* make sure the MDS is started */
- mutex_lock(&server_start_lock);
- obd = class_name2obd(LUSTRE_MDS_OBDNAME);
- if (!obd) {
- rc = lustre_start_simple(LUSTRE_MDS_OBDNAME,
- LUSTRE_MDS_NAME,
- LUSTRE_MDS_OBDNAME"_uuid",
- NULL, NULL, NULL, NULL);
- if (rc) {
- mutex_unlock(&server_start_lock);
- CERROR("failed to start MDS: %d\n", rc);
- RETURN(rc);
- }
- }
- mutex_unlock(&server_start_lock);
+ obd_name_service = LUSTRE_MDS_OBDNAME;
+ name_service = LUSTRE_MDS_NAME;
+ } else {
+ obd_name_service = LUSTRE_OSS_OBDNAME;
+ name_service = LUSTRE_OSS_NAME;
}
- /* If we're an OST, make sure the global OSS is running */
- if (IS_OST(lsi)) {
- /* make sure OSS is started */
- mutex_lock(&server_start_lock);
- obd = class_name2obd(LUSTRE_OSS_OBDNAME);
- if (!obd) {
- rc = lustre_start_simple(LUSTRE_OSS_OBDNAME,
- LUSTRE_OSS_NAME,
- LUSTRE_OSS_OBDNAME"_uuid",
- NULL, NULL, NULL, NULL);
- if (rc) {
- mutex_unlock(&server_start_lock);
- CERROR("failed to start OSS: %d\n", rc);
- RETURN(rc);
- }
+ /* make sure MDS/OSS is started */
+ mutex_lock(&server_start_lock);
+ obd = class_name2obd(obd_name_service);
+ if (!obd) {
+ rc = lustre_start_simple(obd_name_service, name_service,
+ (IS_MDT(lsi) ?
+ LUSTRE_MDS_OBDNAME"_uuid" :
+ LUSTRE_OSS_OBDNAME"_uuid"),
+ NULL, NULL, NULL, NULL);
+ if (rc) {
+ mutex_unlock(&server_start_lock);
+ CERROR("failed to start %s: %d\n",
+ obd_name_service, rc);
+ RETURN(rc);
}
+ }
+ /* hold a type reference and put it at server_stop_servers */
+ type = class_get_type(IS_MDT(lsi) ?
+ LUSTRE_MDT_NAME : LUSTRE_OST_NAME);
+ if (!type) {
mutex_unlock(&server_start_lock);
+ GOTO(out_stop_service, rc = -ENODEV);
+ }
+ lsi->lsi_server_started = 1;
+ mutex_unlock(&server_start_lock);
+ if (OBD_FAIL_PRECHECK(OBD_FAIL_OBD_STOP_MDS_RACE) &&
+ IS_MDT(lsi)) {
+ OBD_RACE(OBD_FAIL_OBD_STOP_MDS_RACE);
+ msleep(2 * MSEC_PER_SEC);
}
rc = lu_env_init(&mgc_env, LCT_MG_THREAD);
/* abort recovery only on the complete stack:
* many devices can be involved */
- if ((lsi->lsi_lmd->lmd_flags & LMD_FLG_ABORT_RECOV) &&
+ if ((lsi->lsi_lmd->lmd_flags &
+ (LMD_FLG_ABORT_RECOV | LMD_FLG_ABORT_RECOV_MDT)) &&
(OBP(obd, iocontrol))) {
+ struct obd_ioctl_data karg = {
+ .ioc_type = lsi->lsi_lmd->lmd_flags,
+ };
+
obd_iocontrol(OBD_IOC_ABORT_RECOVERY, obd->obd_self_export, 0,
- NULL, NULL);
+ &karg, NULL);
}
out_mgc:
out_env:
lu_env_fini(&mgc_env);
out_stop_service:
- if (rc != 0)
- server_stop_servers(lsi->lsi_flags);
+ /* in case of error upper function call
+ * server_put_super->server_stop_servers()
+ */
RETURN(rc);
}
*/
lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_WRITECONF) ?
LDD_F_WRITECONF : 0;
+ lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_NO_LOCAL_LOGS) ?
+ LDD_F_NO_LOCAL_LOGS : 0;
lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_VIRGIN) ?
LDD_F_VIRGIN : 0;
lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_UPDATE) ?
char *tmpname, *extraname = NULL;
int tmpname_sz;
int lsiflags = lsi->lsi_flags;
+ bool stop_servers = lsi->lsi_server_started;
ENTRY;
LASSERT(IS_SERVER(lsi));
* to .put_super, so we better make sure we clean up! */
obd->obd_force = 1;
class_manual_cleanup(obd);
+ if (OBD_FAIL_PRECHECK(OBD_FAIL_OBD_STOP_MDS_RACE)) {
+ int idx;
+ server_name2index(lsi->lsi_svname, &idx, NULL);
+ /* sleeping for MDT0001 */
+ if (idx == 1)
+ OBD_RACE(OBD_FAIL_OBD_STOP_MDS_RACE);
+ }
} else {
CERROR("no obd %s\n", lsi->lsi_svname);
server_deregister_mount(lsi->lsi_svname);
/* Stop the servers (MDS, OSS) if no longer needed. We must wait
until the target is really gone so that our type refcount check
is right. */
- server_stop_servers(lsiflags);
+ if (stop_servers)
+ server_stop_servers(lsiflags);
/* In case of startup or cleanup err, stop related obds */
if (extraname) {
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);
/** The operations we support directly on the superblock:
* mount, umount, and df.
*/
-static struct super_operations server_ops = {
+static const struct super_operations server_ops = {
.put_super = server_put_super,
.umount_begin = server_umount_begin, /* umount -f */
.statfs = server_statfs,
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,
.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
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);
/* 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);
server_put_super(sb);
return rc;
}
+EXPORT_SYMBOL(server_fill_super);
/*
* Calculate timeout value for a target.
git://git.whamcloud.com / fs / lustre-release.git / blobdiff