/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
* vim:expandtab:shiftwidth=8:tabstop=8:
*
- * Lustre Light Super operations
+ * Lustre Light common routines
*
- * Copyright (c) 2002, 2003 Cluster File Systems, Inc.
+ * Copyright (c) 2002-2004 Cluster File Systems, Inc.
*
* This file is part of Lustre, http://www.lustre.org.
*
* Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
*/
-#define DEBUG_SUBSYSTEM S_LLITE
-
#include <stdlib.h>
#include <string.h>
-#include <error.h>
#include <assert.h>
+#include <signal.h>
#include <sys/types.h>
+#include <sys/stat.h>
#include <sys/queue.h>
#include <sysio.h>
+#ifdef HAVE_XTIO_H
+#include <xtio.h>
+#endif
#include <fs.h>
#include <mount.h>
#include <inode.h>
+#ifdef HAVE_FILE_H
#include <file.h>
+#endif
-#include <netinet/in.h>
-#include <sys/socket.h>
-#include <arpa/inet.h>
-
-#include <portals/api-support.h> /* needed for ptpctl.h */
-#include <portals/ptlctl.h> /* needed for parse_dump */
+/* both sys/queue.h (libsysio require it) and portals/lists.h have definition
+ * of 'LIST_HEAD'. undef it to suppress warnings
+ */
+#undef LIST_HEAD
+#include <liblustre.h>
+#include <lnet/lnetctl.h> /* needed for parse_dump */
+#include "lutil.h"
#include "llite_lib.h"
-
-ptl_handle_ni_t tcpnal_ni;
-struct task_struct *current;
-struct obd_class_user_state ocus;
-
-/* portals interfaces */
-ptl_handle_ni_t *
-kportal_get_ni (int nal)
+static int lllib_init(void)
{
- return &tcpnal_ni;
-}
+ if (liblustre_init_current("liblustre") ||
+ init_lib_portals() ||
+ init_obdclass() ||
+ ptlrpc_init() ||
+ mgc_init() ||
+ lmv_init() ||
+ mdc_init() ||
+ lov_init() ||
+ osc_init())
+ return -1;
-inline void
-kportal_put_ni (int nal)
-{
- return;
+ return _sysio_fssw_register("lustre", &llu_fssw_ops);
}
-struct ldlm_namespace;
-struct ldlm_res_id;
-struct obd_import;
-
-extern int ldlm_cli_cancel_unused(struct ldlm_namespace *ns, struct ldlm_res_id *res_id, int flags);
-extern int ldlm_namespace_cleanup(struct ldlm_namespace *ns, int local_only);
-extern int ldlm_replay_locks(struct obd_import *imp);
-
-void *inter_module_get(char *arg)
+int liblustre_process_log(struct config_llog_instance *cfg,
+ char *mgsnid, char *profile,
+ int allow_recov)
{
- if (!strcmp(arg, "tcpnal_ni"))
- return &tcpnal_ni;
- else if (!strcmp(arg, "ldlm_cli_cancel_unused"))
- return ldlm_cli_cancel_unused;
- else if (!strcmp(arg, "ldlm_namespace_cleanup"))
- return ldlm_namespace_cleanup;
- else if (!strcmp(arg, "ldlm_replay_locks"))
- return ldlm_replay_locks;
- else
- return NULL;
-}
+ struct lustre_cfg_bufs bufs;
+ struct lustre_cfg *lcfg;
+ char *peer = "MGS_UUID";
+ struct obd_device *obd;
+ struct lustre_handle mgc_conn = {0, };
+ struct obd_export *exp;
+ char *name = "mgc_dev";
+ class_uuid_t uuid;
+ struct obd_uuid mgc_uuid;
+ struct llog_ctxt *ctxt;
+ lnet_nid_t nid = 0;
+ char *mdsnid;
+ int err, rc = 0;
+ struct obd_connect_data *ocd = NULL;
+ ENTRY;
+
+ ll_generate_random_uuid(uuid);
+ class_uuid_unparse(uuid, &mgc_uuid);
+
+ nid = libcfs_str2nid(mgsnid);
+ if (nid == LNET_NID_ANY) {
+ CERROR("Can't parse NID %s\n", mgsnid);
+ RETURN(-EINVAL);
+ }
-void init_current(char *comm)
-{
- current = malloc(sizeof(*current));
- current->fs = malloc(sizeof(*current->fs));
- current->fs->umask = umask(0777);
- umask(current->fs->umask);
- strncpy(current->comm, comm, sizeof(current->comm));
- current->pid = getpid();
- current->fsuid = 0;
- current->fsgid = 0;
- current->cap_effective = 0;
- memset(¤t->pending, 0, sizeof(current->pending));
-}
+ lustre_cfg_bufs_reset(&bufs, NULL);
+ lustre_cfg_bufs_set_string(&bufs, 1, peer);
+ lcfg = lustre_cfg_new(LCFG_ADD_UUID, &bufs);
+ lcfg->lcfg_nid = nid;
+ rc = class_process_config(lcfg);
+ lustre_cfg_free(lcfg);
+ if (rc < 0)
+ GOTO(out, rc);
+
+ lustre_cfg_bufs_reset(&bufs, name);
+ lustre_cfg_bufs_set_string(&bufs, 1, LUSTRE_MGC_NAME);
+ lustre_cfg_bufs_set_string(&bufs, 2, mgc_uuid.uuid);
+ lcfg = lustre_cfg_new(LCFG_ATTACH, &bufs);
+ rc = class_process_config(lcfg);
+ lustre_cfg_free(lcfg);
+ if (rc < 0)
+ GOTO(out_del_uuid, rc);
+
+ lustre_cfg_bufs_reset(&bufs, name);
+ lustre_cfg_bufs_set_string(&bufs, 1, LUSTRE_MGS_OBDNAME);
+ lustre_cfg_bufs_set_string(&bufs, 2, peer);
+ lcfg = lustre_cfg_new(LCFG_SETUP, &bufs);
+ rc = class_process_config(lcfg);
+ lustre_cfg_free(lcfg);
+ if (rc < 0)
+ GOTO(out_detach, rc);
+
+ while ((mdsnid = strsep(&mgsnid, ","))) {
+ nid = libcfs_str2nid(mdsnid);
+ lustre_cfg_bufs_reset(&bufs, NULL);
+ lustre_cfg_bufs_set_string(&bufs, 1, libcfs_nid2str(nid));
+ lcfg = lustre_cfg_new(LCFG_ADD_UUID, &bufs);
+ lcfg->lcfg_nid = nid;
+ rc = class_process_config(lcfg);
+ lustre_cfg_free(lcfg);
+ if (rc) {
+ CERROR("Add uuid for %s failed %d\n",
+ libcfs_nid2str(nid), rc);
+ continue;
+ }
-ptl_nid_t tcpnal_mynid;
+ lustre_cfg_bufs_reset(&bufs, name);
+ lustre_cfg_bufs_set_string(&bufs, 1, libcfs_nid2str(nid));
+ lcfg = lustre_cfg_new(LCFG_ADD_CONN, &bufs);
+ lcfg->lcfg_nid = nid;
+ rc = class_process_config(lcfg);
+ lustre_cfg_free(lcfg);
+ if (rc) {
+ CERROR("Add conn for %s failed %d\n",
+ libcfs_nid2str(nid), rc);
+ continue;
+ }
+ }
-int init_lib_portals()
-{
- int rc;
-
- PtlInit();
- rc = PtlNIInit(procbridge_interface, 0, 0, 0, &tcpnal_ni);
- if (rc != 0) {
- CERROR("ksocknal: PtlNIInit failed: error %d\n", rc);
- PtlFini();
- RETURN (rc);
+ obd = class_name2obd(name);
+ if (obd == NULL)
+ GOTO(out_cleanup, rc = -EINVAL);
+
+ OBD_ALLOC(ocd, sizeof(*ocd));
+ if (ocd == NULL)
+ GOTO(out_cleanup, rc = -ENOMEM);
+
+ ocd->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_FID;
+#ifdef LIBLUSTRE_POSIX_ACL
+ ocd->ocd_connect_flags |= OBD_CONNECT_ACL;
+#endif
+ ocd->ocd_version = LUSTRE_VERSION_CODE;
+
+ rc = obd_connect(NULL, &mgc_conn, obd, &mgc_uuid, ocd, NULL);
+ if (rc) {
+ CERROR("cannot connect to %s at %s: rc = %d\n",
+ LUSTRE_MGS_OBDNAME, mgsnid, rc);
+ GOTO(out_cleanup, rc);
}
- PtlNIDebug(tcpnal_ni, ~0);
- return rc;
-}
-extern int class_handle_ioctl(struct obd_class_user_state *ocus, unsigned int cmd, unsigned long arg);
+ exp = class_conn2export(&mgc_conn);
+
+ ctxt = llog_get_context(exp->exp_obd, LLOG_CONFIG_REPL_CTXT);
+ cfg->cfg_flags |= CFG_F_COMPAT146;
+ rc = class_config_parse_llog(ctxt, profile, cfg);
+ llog_ctxt_put(ctxt);
+ if (rc) {
+ CERROR("class_config_parse_llog failed: rc = %d\n", rc);
+ }
-struct mount_option_s mount_option = {NULL, NULL};
+ /* We don't so much care about errors in cleaning up the config llog
+ * connection, as we have already read the config by this point. */
+ err = obd_disconnect(exp);
+ if (err)
+ CERROR("obd_disconnect failed: rc = %d\n", err);
+
+out_cleanup:
+ if (ocd)
+ OBD_FREE(ocd, sizeof(*ocd));
+
+ lustre_cfg_bufs_reset(&bufs, name);
+ lcfg = lustre_cfg_new(LCFG_CLEANUP, &bufs);
+ err = class_process_config(lcfg);
+ lustre_cfg_free(lcfg);
+ if (err)
+ CERROR("md_cleanup failed: rc = %d\n", err);
+
+out_detach:
+ lustre_cfg_bufs_reset(&bufs, name);
+ lcfg = lustre_cfg_new(LCFG_DETACH, &bufs);
+ err = class_process_config(lcfg);
+ lustre_cfg_free(lcfg);
+ if (err)
+ CERROR("md_detach failed: rc = %d\n", err);
+
+out_del_uuid:
+ lustre_cfg_bufs_reset(&bufs, name);
+ lustre_cfg_bufs_set_string(&bufs, 1, peer);
+ lcfg = lustre_cfg_new(LCFG_DEL_UUID, &bufs);
+ err = class_process_config(lcfg);
+ if (err)
+ CERROR("del MDC UUID failed: rc = %d\n", err);
+ lustre_cfg_free(lcfg);
+out:
+
+ RETURN(rc);
+}
-/* FIXME simple arg parser FIXME */
-void parse_mount_options(void *arg)
+/* parse host:/fsname string */
+int ll_parse_mount_target(const char *target, char **mgsnid,
+ char **fsname)
{
- char *buf = NULL;
- struct obd_ioctl_data *data;
- char *ptr, *comma, *eq, **tgt, *v;
- int len;
+ static char buf[256];
+ char *s;
- if (obd_ioctl_getdata(&buf, &len, arg)) {
- CERROR("OBD ioctl: data error\n");
- return;
- }
- data = (struct obd_ioctl_data *)buf;
- ptr = data->ioc_inlbuf1;
- printf("mount option: %s\n", ptr);
-
- while (ptr) {
- eq = strchr(ptr, '=');
- if (!eq)
- return;
-
- *eq = 0;
- if (!strcmp("osc", ptr))
- tgt = &mount_option.osc_uuid;
- else if (!strcmp("mdc", ptr))
- tgt = &mount_option.mdc_uuid;
- else {
- printf("Unknown mount option %s\n", ptr);
- return;
- }
+ buf[255] = 0;
+ strncpy(buf, target, 255);
- v = eq + 1;
- comma = strchr(v, ',');
- if (comma) {
- *comma = 0;
- ptr = comma + 1;
- } else
- ptr = NULL;
+ if ((s = strchr(buf, ':'))) {
+ *mgsnid = buf;
+ *s = '\0';
- *tgt = malloc(strlen(v)+1);
- strcpy(*tgt, v);
+ while (*++s == '/')
+ ;
+ sprintf(s + strlen(s), "-client");
+ *fsname = s;
+
+ return 0;
}
- if (buf)
- obd_ioctl_freedata(buf, len);
+ return -1;
}
-int lib_ioctl(int dev_id, int opc, void * ptr)
+/*
+ * early liblustre init
+ * called from C startup in catamount apps, before main()
+ *
+ * The following is a skeleton sysio startup sequence,
+ * as implemented in C startup (skipping error handling).
+ * In this framework none of these calls need be made here
+ * or in the apps themselves. The NAMESPACE_STRING specifying
+ * the initial set of fs ops (creates, mounts, etc.) is passed
+ * as an environment variable.
+ *
+ * _sysio_init();
+ * _sysio_incore_init();
+ * _sysio_native_init();
+ * _sysio_lustre_init();
+ * _sysio_boot(NAMESPACE_STRING);
+ *
+ * the name _sysio_lustre_init() follows the naming convention
+ * established in other fs drivers from libsysio:
+ * _sysio_incore_init(), _sysio_native_init()
+ *
+ * _sysio_lustre_init() must be called before _sysio_boot()
+ * to enable libsysio's processing of namespace init strings containing
+ * lustre filesystem operations
+ */
+int _sysio_lustre_init(void)
{
- int rc;
+ int err;
+ char *envstr;
+#ifndef INIT_SYSIO
+ extern void __liblustre_cleanup_(void);
+#endif
+
+ liblustre_init_random();
+
+ err = lllib_init();
+ if (err) {
+ perror("init llite driver");
+ return err;
+ }
- if (dev_id == OBD_DEV_ID) {
- struct obd_ioctl_data *ioc = ptr;
+ envstr = getenv("LIBLUSTRE_TIMEOUT");
+ if (envstr != NULL) {
+ obd_timeout = (unsigned int)strtol(envstr, NULL, 0);
+ printf("LibLustre: obd timeout=%u seconds\n",
+ obd_timeout);
+ }
- if (opc == OBD_IOC_MOUNTOPT) {
- parse_mount_options(ptr);
- return 0;
- }
+ /* debug peer on timeout? */
+ envstr = getenv("LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT");
+ if (envstr != NULL) {
+ obd_debug_peer_on_timeout =
+ (unsigned int)strtol(envstr, NULL, 0);
+ printf("LibLustre: debug peer on timeout=%d\n",
+ obd_debug_peer_on_timeout ? 0 : 1);
+ }
- rc = class_handle_ioctl(&ocus, opc, (unsigned long)ptr);
+#ifndef INIT_SYSIO
+ (void)atexit(__liblustre_cleanup_);
+#endif
+ return err;
+}
- /* you _may_ need to call obd_ioctl_unpack or some
- other verification function if you want to use ioc
- directly here */
- printf ("processing ioctl cmd: %x buf len: %d, rc %d\n",
- opc, ioc->ioc_len, rc);
+extern int _sysio_native_init();
- if (rc)
- return rc;
- }
- return (0);
-}
+char *lustre_path = NULL;
-int lllib_init(char *arg)
+void __liblustre_setup_(void)
{
- tcpnal_mynid = ntohl(inet_addr(arg));
- INIT_LIST_HEAD(&ocus.ocus_conns);
+ char *target = NULL;
+ char *lustre_driver = "lustre";
+ unsigned mntflgs = 0;
+ int err;
+
+ lustre_path = getenv("LIBLUSTRE_MOUNT_POINT");
+ if (!lustre_path) {
+ lustre_path = "/mnt/lustre";
+ }
- init_current("dummy");
- if (init_obdclass() ||
- init_lib_portals() ||
- ptlrpc_init() ||
- ldlm_init() ||
- mdc_init() ||
- lov_init() ||
- osc_init())
- return -1;
+ /* mount target */
+ target = getenv("LIBLUSTRE_MOUNT_TARGET");
+ if (!target) {
+ printf("LibLustre: no mount target specified\n");
+ exit(1);
+ }
- if (parse_dump("/tmp/DUMP_FILE", lib_ioctl))
- return -1;
+ CDEBUG(D_CONFIG, "LibLustre: mount point %s, target %s\n",
+ lustre_path, target);
+
+#ifdef INIT_SYSIO
+ /* initialize libsysio & mount rootfs */
+ if (_sysio_init()) {
+ perror("init sysio");
+ exit(1);
+ }
+ _sysio_native_init();
+
+ err = _sysio_mount_root("/", "native", mntflgs, NULL);
+ if (err) {
+ fprintf(stderr, "sysio mount failed: %s\n", strerror(errno));
+ exit(1);
+ }
+
+ if (_sysio_lustre_init())
+ exit(1);
+#endif /* INIT_SYSIO */
- return _sysio_fssw_register("llite", &llu_fssw_ops);
+ err = mount(target, lustre_path, lustre_driver, mntflgs, NULL);
+ if (err) {
+ fprintf(stderr, "Lustre mount failed: %s\n", strerror(errno));
+ exit(1);
+ }
}
-/* FIXME */
-void generate_random_uuid(unsigned char uuid_out[16])
+void __liblustre_cleanup_(void)
{
- int *arr = (int*)uuid_out;
- int i;
+#ifndef INIT_SYSIO
+ /* guard against being called multiple times */
+ static int cleaned = 0;
- for (i = 0; i < sizeof(uuid_out)/sizeof(int); i++)
- arr[i] = rand();
+ if (cleaned)
+ return;
+ cleaned++;
+#endif
+
+ /* user app might chdir to a lustre directory, and leave busy pnode
+ * during finaly libsysio cleanup. here we chdir back to "/".
+ * but it can't fix the situation that liblustre is mounted
+ * at "/".
+ */
+ chdir("/");
+#if 0
+ umount(lustre_path);
+#endif
+ /* we can't call umount here, because libsysio will not cleanup
+ * opening files for us. _sysio_shutdown() will cleanup fds at
+ * first but which will also close the sockets we need for umount
+ * liblutre. this dilema lead to another hack in
+ * libsysio/src/file_hack.c FIXME
+ */
+#ifdef INIT_SYSIO
+ _sysio_shutdown();
+ cleanup_lib_portals();
+ LNetFini();
+#endif
}
-