/* -*- 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/queue.h>
#include <inode.h>
#include <file.h>
-#include <netinet/in.h>
-#include <sys/socket.h>
-#include <arpa/inet.h>
+#ifdef REDSTORM
+#define CSTART_INIT
+#endif
-#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 <portals/ptlctl.h>
+#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;
-}
-
-inline void
-kportal_put_ni (int nal)
-{
- return;
-}
+ liblustre_set_nal_nid();
-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);
+ if (liblustre_init_current("dummy") ||
+ init_obdclass() ||
+ init_lib_portals() ||
+ ptlrpc_init() ||
+ mdc_init() ||
+ lov_init() ||
+ osc_init())
+ return -1;
-void *inter_module_get(char *arg)
-{
- 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;
+ return _sysio_fssw_register("llite", &llu_fssw_ops);
}
+
+#ifndef CRAY_PORTALS
+#define LIBLUSTRE_NAL_NAME "tcp"
+#elif defined REDSTORM
+#define LIBLUSTRE_NAL_NAME "cray_qk_ernal"
+#else
+#define LIBLUSTRE_NAL_NAME "cray_pb_ernal"
+#endif
+
+int liblustre_process_log(struct config_llog_instance *cfg, int allow_recov)
+{
+ struct lustre_cfg lcfg;
+ char *peer = "MDS_PEER_UUID";
+ struct obd_device *obd;
+ struct lustre_handle mdc_conn = {0, };
+ struct obd_export *exp;
+ char *name = "mdc_dev";
+ class_uuid_t uuid;
+ struct obd_uuid mdc_uuid;
+ struct llog_ctxt *ctxt;
+ ptl_nid_t nid = 0;
+ int nal, err, rc = 0;
+ ENTRY;
+
+ generate_random_uuid(uuid);
+ class_uuid_unparse(uuid, &mdc_uuid);
+
+ if (ptl_parse_nid(&nid, g_zconf_mdsnid)) {
+ CERROR("Can't parse NID %s\n", g_zconf_mdsnid);
+ 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));
+ nal = ptl_name2nal(LIBLUSTRE_NAL_NAME);
+ if (nal <= 0) {
+ CERROR("Can't parse NAL %s\n", LIBLUSTRE_NAL_NAME);
+ RETURN(-EINVAL);
+ }
+ LCFG_INIT(lcfg, LCFG_ADD_UUID, name);
+ lcfg.lcfg_nid = nid;
+ lcfg.lcfg_inllen1 = strlen(peer) + 1;
+ lcfg.lcfg_inlbuf1 = peer;
+ lcfg.lcfg_nal = nal;
+ err = class_process_config(&lcfg);
+ if (err < 0)
+ GOTO(out, err);
+
+ LCFG_INIT(lcfg, LCFG_ATTACH, name);
+ lcfg.lcfg_inlbuf1 = "mdc";
+ lcfg.lcfg_inllen1 = strlen(lcfg.lcfg_inlbuf1) + 1;
+ lcfg.lcfg_inlbuf2 = mdc_uuid.uuid;
+ lcfg.lcfg_inllen2 = strlen(lcfg.lcfg_inlbuf2) + 1;
+ err = class_process_config(&lcfg);
+ if (err < 0)
+ GOTO(out_del_uuid, err);
+
+ LCFG_INIT(lcfg, LCFG_SETUP, name);
+ lcfg.lcfg_inlbuf1 = g_zconf_mdsname;
+ lcfg.lcfg_inllen1 = strlen(lcfg.lcfg_inlbuf1) + 1;
+ lcfg.lcfg_inlbuf2 = peer;
+ lcfg.lcfg_inllen2 = strlen(lcfg.lcfg_inlbuf2) + 1;
+ err = class_process_config(&lcfg);
+ if (err < 0)
+ GOTO(out_detach, err);
+
+ obd = class_name2obd(name);
+ if (obd == NULL)
+ GOTO(out_cleanup, err = -EINVAL);
+
+ /* Disable initial recovery on this import */
+ err = obd_set_info(obd->obd_self_export,
+ strlen("initial_recov"), "initial_recov",
+ sizeof(allow_recov), &allow_recov);
+
+ err = obd_connect(&mdc_conn, obd, &mdc_uuid, 0);
+ if (err) {
+ CERROR("cannot connect to %s: rc = %d\n",
+ g_zconf_mdsname, err);
+ GOTO(out_cleanup, err);
+ }
+
+ exp = class_conn2export(&mdc_conn);
+
+ ctxt = exp->exp_obd->obd_llog_ctxt[LLOG_CONFIG_REPL_CTXT];
+ rc = class_config_process_llog(ctxt, g_zconf_profile, cfg);
+ if (rc)
+ CERROR("class_config_process_llog failed: rc = %d\n", rc);
+
+ err = obd_disconnect(exp, 0);
+
+out_cleanup:
+ LCFG_INIT(lcfg, LCFG_CLEANUP, name);
+ err = class_process_config(&lcfg);
+ if (err < 0)
+ GOTO(out, err);
+
+out_detach:
+ LCFG_INIT(lcfg, LCFG_DETACH, name);
+ err = class_process_config(&lcfg);
+ if (err < 0)
+ GOTO(out, err);
+
+out_del_uuid:
+ LCFG_INIT(lcfg, LCFG_DEL_UUID, name);
+ lcfg.lcfg_inllen1 = strlen(peer) + 1;
+ lcfg.lcfg_inlbuf1 = peer;
+ err = class_process_config(&lcfg);
+
+out:
+ if (rc == 0)
+ rc = err;
+
+ RETURN(rc);
}
-ptl_nid_t tcpnal_mynid;
-
-int init_lib_portals()
+/* parse host:/mdsname/profile string */
+int ll_parse_mount_target(const char *target, char **mdsnid,
+ char **mdsname, char **profile)
{
- 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);
+ static char buf[256];
+ char *s;
+
+ buf[255] = 0;
+ strncpy(buf, target, 255);
+
+ if ((s = strchr(buf, ':'))) {
+ *mdsnid = buf;
+ *s = '\0';
+
+ while (*++s == '/')
+ ;
+ *mdsname = s;
+ if ((s = strchr(*mdsname, '/'))) {
+ *s = '\0';
+ *profile = s + 1;
+ return 0;
+ }
}
- PtlNIDebug(tcpnal_ni, ~0);
- return rc;
-}
-
-extern int class_handle_ioctl(struct obd_class_user_state *ocus, unsigned int cmd, unsigned long arg);
-struct mount_option_s mount_option = {NULL, NULL};
+ return -1;
+}
-/* FIXME simple arg parser FIXME */
-void parse_mount_options(void *arg)
+/*
+ * 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)
{
- char *buf = NULL;
- struct obd_ioctl_data *data;
- char *ptr, *comma, *eq, **tgt, *v;
- int len;
-
- 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;
- }
+ int err;
- v = eq + 1;
- comma = strchr(v, ',');
- if (comma) {
- *comma = 0;
- ptr = comma + 1;
- } else
- ptr = NULL;
+#if 0
+ portal_debug = -1;
+ portal_subsystem_debug = -1;
+#endif
- *tgt = malloc(strlen(v)+1);
- strcpy(*tgt, v);
- }
+ liblustre_init_random();
- if (buf)
- obd_ioctl_freedata(buf, len);
+ err = lllib_init();
+ if (err) {
+ perror("init llite driver");
+ }
+ return err;
}
-int lib_ioctl(int dev_id, int opc, void * ptr)
-{
- int rc;
+/* env variables */
+#define ENV_LUSTRE_MNTPNT "LIBLUSTRE_MOUNT_POINT"
+#define ENV_LUSTRE_MNTTGT "LIBLUSTRE_MOUNT_TARGET"
+#define ENV_LUSTRE_TIMEOUT "LIBLUSTRE_TIMEOUT"
+#define ENV_LUSTRE_DUMPFILE "LIBLUSTRE_DUMPFILE"
+#define ENV_LUSTRE_DEBUG_MASK "LIBLUSTRE_DEBUG_MASK"
+#define ENV_LUSTRE_DEBUG_SUBSYS "LIBLUSTRE_DEBUG_SUBSYS"
- if (dev_id == OBD_DEV_ID) {
- struct obd_ioctl_data *ioc = ptr;
+extern int _sysio_native_init();
+extern unsigned int obd_timeout;
- if (opc == OBD_IOC_MOUNTOPT) {
- parse_mount_options(ptr);
- return 0;
- }
+static char *lustre_path = NULL;
- rc = class_handle_ioctl(&ocus, opc, (unsigned long)ptr);
+/* global variables */
+char *g_zconf_mdsname = NULL; /* mdsname, for zeroconf */
+char *g_zconf_mdsnid = NULL; /* mdsnid, for zeroconf */
+char *g_zconf_profile = NULL; /* profile, for zeroconf */
- /* 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);
- if (rc)
- return rc;
+void __liblustre_setup_(void)
+{
+ char *target = NULL;
+ char *timeout = NULL;
+ char *debug_mask = NULL;
+ char *debug_subsys = NULL;
+ char *root_driver = "native";
+ char *lustre_driver = "llite";
+ char *root_path = "/";
+ unsigned mntflgs = 0;
+ int err;
+
+ lustre_path = getenv(ENV_LUSTRE_MNTPNT);
+ if (!lustre_path) {
+ lustre_path = "/mnt/lustre";
}
- return (0);
-}
-int lllib_init(char *arg)
-{
- tcpnal_mynid = ntohl(inet_addr(arg));
- INIT_LIST_HEAD(&ocus.ocus_conns);
+ /* mount target */
+ target = getenv(ENV_LUSTRE_MNTTGT);
+ if (!target) {
+ printf("LibLustre: no mount target specified\n");
+ exit(1);
+ }
+ if (ll_parse_mount_target(target,
+ &g_zconf_mdsnid,
+ &g_zconf_mdsname,
+ &g_zconf_profile)) {
+ CERROR("mal-formed target %s \n", target);
+ exit(1);
+ }
+ if (!g_zconf_mdsnid || !g_zconf_mdsname || !g_zconf_profile) {
+ printf("Liblustre: invalid target %s\n", target);
+ exit(1);
+ }
+ printf("LibLustre: mount point %s, target %s\n",
+ lustre_path, target);
+
+ timeout = getenv(ENV_LUSTRE_TIMEOUT);
+ if (timeout) {
+ obd_timeout = (unsigned int) strtol(timeout, NULL, 0);
+ printf("LibLustre: set obd timeout as %u seconds\n",
+ obd_timeout);
+ }
- init_current("dummy");
- if (init_obdclass() ||
- init_lib_portals() ||
- ptlrpc_init() ||
- ldlm_init() ||
- mdc_init() ||
- lov_init() ||
- osc_init())
- return -1;
+ /* debug masks */
+ debug_mask = getenv(ENV_LUSTRE_DEBUG_MASK);
+ if (debug_mask)
+ portal_debug = (unsigned int) strtol(debug_mask, NULL, 0);
+
+ debug_subsys = getenv(ENV_LUSTRE_DEBUG_SUBSYS);
+ if (debug_subsys)
+ portal_subsystem_debug =
+ (unsigned int) strtol(debug_subsys, NULL, 0);
+
+#ifndef CSTART_INIT
+ /* initialize libsysio & mount rootfs */
+ if (_sysio_init()) {
+ perror("init sysio");
+ exit(1);
+ }
+ _sysio_native_init();
- if (parse_dump("/tmp/DUMP_FILE", lib_ioctl))
- return -1;
+ err = _sysio_mount_root(root_path, root_driver, mntflgs, NULL);
+ if (err) {
+ perror(root_driver);
+ exit(1);
+ }
- return _sysio_fssw_register("llite", &llu_fssw_ops);
+ if (_sysio_lustre_init())
+ exit(1);
+#endif
+
+ err = mount("/", lustre_path, lustre_driver, mntflgs, NULL);
+ if (err) {
+ errno = -err;
+ perror(lustre_driver);
+ exit(1);
+ }
}
-/* FIXME */
-void generate_random_uuid(unsigned char uuid_out[16])
+void __liblustre_cleanup_(void)
{
- int *arr = (int*)uuid_out;
- int i;
-
- for (i = 0; i < sizeof(uuid_out)/sizeof(int); i++)
- arr[i] = rand();
+ /* 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 delima lead to another hack in
+ * libsysio/src/file_hack.c FIXME
+ */
+ _sysio_shutdown();
+ cleanup_lib_portals();
+ PtlFini();
}
-