-/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
- * vim:expandtab:shiftwidth=8:tabstop=8:
- *
+/*
* GPL HEADER START
*
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
/*
* Copyright (c) 1999, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
+ *
+ * Copyright (c) 2011, 2012, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
*/
#define DEBUG_SUBSYSTEM S_CLASS
-#ifndef EXPORT_SYMTAB
-# define EXPORT_SYMTAB
-#endif
#ifndef __KERNEL__
# include <liblustre.h>
#else
#include <lprocfs_status.h>
#include <lustre/lustre_build_version.h>
#include <libcfs/list.h>
+#include <cl_object.h>
+#ifdef HAVE_SERVER_SUPPORT
+# include <dt_object.h>
+# include <md_object.h>
+#endif /* HAVE_SERVER_SUPPORT */
#include "llog_internal.h"
#ifndef __KERNEL__
/* liblustre workaround */
-cfs_atomic_t libcfs_kmemory = {0};
+atomic_t libcfs_kmemory = {0};
#endif
struct obd_device *obd_devs[MAX_OBD_DEVICES];
+EXPORT_SYMBOL(obd_devs);
cfs_list_t obd_types;
-cfs_spinlock_t obd_dev_lock = CFS_SPIN_LOCK_UNLOCKED;
+DEFINE_RWLOCK(obd_dev_lock);
-#ifndef __KERNEL__
__u64 obd_max_pages = 0;
__u64 obd_max_alloc = 0;
+#ifndef __KERNEL__
__u64 obd_alloc;
__u64 obd_pages;
#endif
+DEFINE_SPINLOCK(obd_updatemax_lock);
/* The following are visible and mutable through /proc/sys/lustre/. */
+unsigned int obd_alloc_fail_rate = 0;
+EXPORT_SYMBOL(obd_alloc_fail_rate);
unsigned int obd_debug_peer_on_timeout;
+EXPORT_SYMBOL(obd_debug_peer_on_timeout);
unsigned int obd_dump_on_timeout;
+EXPORT_SYMBOL(obd_dump_on_timeout);
unsigned int obd_dump_on_eviction;
+EXPORT_SYMBOL(obd_dump_on_eviction);
unsigned int obd_max_dirty_pages = 256;
+EXPORT_SYMBOL(obd_max_dirty_pages);
+atomic_t obd_unstable_pages;
+EXPORT_SYMBOL(obd_unstable_pages);
+atomic_t obd_dirty_pages;
+EXPORT_SYMBOL(obd_dirty_pages);
unsigned int obd_timeout = OBD_TIMEOUT_DEFAULT; /* seconds */
+EXPORT_SYMBOL(obd_timeout);
unsigned int ldlm_timeout = LDLM_TIMEOUT_DEFAULT; /* seconds */
+EXPORT_SYMBOL(ldlm_timeout);
+unsigned int obd_timeout_set;
+EXPORT_SYMBOL(obd_timeout_set);
+unsigned int ldlm_timeout_set;
+EXPORT_SYMBOL(ldlm_timeout_set);
/* Adaptive timeout defs here instead of ptlrpc module for /proc/sys/ access */
unsigned int at_min = 0;
+EXPORT_SYMBOL(at_min);
unsigned int at_max = 600;
+EXPORT_SYMBOL(at_max);
unsigned int at_history = 600;
+EXPORT_SYMBOL(at_history);
int at_early_margin = 5;
+EXPORT_SYMBOL(at_early_margin);
int at_extra = 30;
+EXPORT_SYMBOL(at_extra);
-cfs_atomic_t obd_dirty_pages;
-cfs_atomic_t obd_dirty_transit_pages;
+atomic_t obd_dirty_transit_pages;
+EXPORT_SYMBOL(obd_dirty_transit_pages);
-cfs_waitq_t obd_race_waitq;
-int obd_race_state;
+char obd_jobid_var[JOBSTATS_JOBID_VAR_MAX_LEN + 1] = JOBSTATS_DISABLE;
+EXPORT_SYMBOL(obd_jobid_var);
-#ifdef __KERNEL__
-unsigned long obd_print_fail_loc(void)
+#ifdef LPROCFS
+struct lprocfs_stats *obd_memory = NULL;
+EXPORT_SYMBOL(obd_memory);
+#endif
+
+/* Get jobid of current process by reading the environment variable
+ * stored in between the "env_start" & "env_end" of task struct.
+ *
+ * TODO:
+ * It's better to cache the jobid for later use if there is any
+ * efficient way, the cl_env code probably could be reused for this
+ * purpose.
+ *
+ * If some job scheduler doesn't store jobid in the "env_start/end",
+ * then an upcall could be issued here to get the jobid by utilizing
+ * the userspace tools/api. Then, the jobid must be cached.
+ */
+int lustre_get_jobid(char *jobid)
{
- CWARN("obd_fail_loc = %lx\n", obd_fail_loc);
- return obd_fail_loc;
+ int jobid_len = JOBSTATS_JOBID_SIZE;
+ int rc = 0;
+ ENTRY;
+
+ memset(jobid, 0, JOBSTATS_JOBID_SIZE);
+ /* Jobstats isn't enabled */
+ if (strcmp(obd_jobid_var, JOBSTATS_DISABLE) == 0)
+ RETURN(0);
+
+ /* Use process name + fsuid as jobid */
+ if (strcmp(obd_jobid_var, JOBSTATS_PROCNAME_UID) == 0) {
+ snprintf(jobid, JOBSTATS_JOBID_SIZE, "%s.%u",
+ current_comm(), current_fsuid());
+ RETURN(0);
+ }
+
+ rc = cfs_get_environ(obd_jobid_var, jobid, &jobid_len);
+ if (rc) {
+ if (rc == -EOVERFLOW) {
+ /* For the PBS_JOBID and LOADL_STEP_ID keys (which are
+ * variable length strings instead of just numbers), it
+ * might make sense to keep the unique parts for JobID,
+ * instead of just returning an error. That means a
+ * larger temp buffer for cfs_get_environ(), then
+ * truncating the string at some separator to fit into
+ * the specified jobid_len. Fix later if needed. */
+ static bool printed;
+ if (unlikely(!printed)) {
+ LCONSOLE_ERROR_MSG(0x16b, "%s value too large "
+ "for JobID buffer (%d)\n",
+ obd_jobid_var, jobid_len);
+ printed = true;
+ }
+ } else {
+ CDEBUG((rc == -ENOENT || rc == -EINVAL ||
+ rc == -EDEADLK) ? D_INFO : D_ERROR,
+ "Get jobid for (%s) failed: rc = %d\n",
+ obd_jobid_var, rc);
+ }
+ }
+ RETURN(rc);
}
-#endif
+EXPORT_SYMBOL(lustre_get_jobid);
+
+int obd_alloc_fail(const void *ptr, const char *name, const char *type,
+ size_t size, const char *file, int line)
+{
+ if (ptr == NULL ||
+ (cfs_rand() & OBD_ALLOC_FAIL_MASK) < obd_alloc_fail_rate) {
+ CERROR("%s%salloc of %s ("LPU64" bytes) failed at %s:%d\n",
+ ptr ? "force " :"", type, name, (__u64)size, file,
+ line);
+ CERROR(LPU64" total bytes and "LPU64" total pages "
+ "("LPU64" bytes) allocated by Lustre, "
+ "%d total bytes by LNET\n",
+ obd_memory_sum(),
+ obd_pages_sum() << PAGE_CACHE_SHIFT,
+ obd_pages_sum(),
+ atomic_read(&libcfs_kmemory));
+ return 1;
+ }
+ return 0;
+}
+EXPORT_SYMBOL(obd_alloc_fail);
static inline void obd_data2conn(struct lustre_handle *conn,
struct obd_ioctl_data *data)
OBD_ALLOC(lcfg, data->ioc_plen1);
if (lcfg == NULL)
GOTO(out, err = -ENOMEM);
- err = cfs_copy_from_user(lcfg, data->ioc_pbuf1,
+ err = copy_from_user(lcfg, data->ioc_pbuf1,
data->ioc_plen1);
if (!err)
err = lustre_cfg_sanity_check(lcfg, data->ioc_plen1);
case OBD_IOC_GETDEVICE: {
int index = data->ioc_count;
- const char *status;
- char *str;
+ char *status, *str;
if (!data->ioc_inlbuf1) {
CERROR("No buffer passed in ioctl\n");
if (!obd)
GOTO(out, err = -ENOENT);
- status = obd_dev_status(obd);
+ if (obd->obd_stopping)
+ status = "ST";
+ else if (obd->obd_set_up)
+ status = "UP";
+ else if (obd->obd_attached)
+ status = "AT";
+ else
+ status = "--";
str = (char *)data->ioc_bulk;
snprintf(str, len - sizeof(*data), "%3d %s %s %s %s %d",
(int)index, status, obd->obd_type->typ_name,
obd->obd_name, obd->obd_uuid.uuid,
- cfs_atomic_read(&obd->obd_refcount));
+ atomic_read(&obd->obd_refcount));
err = obd_ioctl_popdata((void *)arg, data, len);
GOTO(out, err = 0);
if (strnlen(data->ioc_inlbuf4, MAX_OBD_NAME) >= MAX_OBD_NAME)
GOTO(out, err = -EINVAL);
obd = class_name2obd(data->ioc_inlbuf4);
- } else if (obd_minor_valid(data->ioc_dev)) {
- /* XXX = max allocated minor */
+ } else if (data->ioc_dev < class_devno_max()) {
obd = class_num2obd(data->ioc_dev);
} else {
CERROR("OBD ioctl: No device\n");
RETURN(err);
} /* class_handle_ioctl */
-
-
#ifdef __KERNEL__
-extern cfs_psdev_t obd_psdev;
+extern struct miscdevice obd_psdev;
#else
-void *obd_psdev = NULL;
+struct miscdevice obd_psdev;
#endif
-EXPORT_SYMBOL(obd_devs);
-EXPORT_SYMBOL(obd_print_fail_loc);
-EXPORT_SYMBOL(obd_race_waitq);
-EXPORT_SYMBOL(obd_race_state);
-EXPORT_SYMBOL(obd_debug_peer_on_timeout);
-EXPORT_SYMBOL(obd_dump_on_timeout);
-EXPORT_SYMBOL(obd_dump_on_eviction);
-EXPORT_SYMBOL(obd_timeout);
-EXPORT_SYMBOL(ldlm_timeout);
-EXPORT_SYMBOL(obd_max_dirty_pages);
-EXPORT_SYMBOL(obd_dirty_pages);
-EXPORT_SYMBOL(obd_dirty_transit_pages);
-EXPORT_SYMBOL(at_min);
-EXPORT_SYMBOL(at_max);
-EXPORT_SYMBOL(at_extra);
-EXPORT_SYMBOL(at_early_margin);
-EXPORT_SYMBOL(at_history);
-EXPORT_SYMBOL(ptlrpc_put_connection_superhack);
-
-EXPORT_SYMBOL(proc_lustre_root);
-
-EXPORT_SYMBOL(class_register_type);
-EXPORT_SYMBOL(class_unregister_type);
-EXPORT_SYMBOL(class_get_type);
-EXPORT_SYMBOL(class_put_type);
-EXPORT_SYMBOL(class_name2obd);
-EXPORT_SYMBOL(class_uuid2obd);
-EXPORT_SYMBOL(class_find_client_obd);
-EXPORT_SYMBOL(class_devices_in_group);
-EXPORT_SYMBOL(class_conn2export);
-EXPORT_SYMBOL(class_exp2obd);
-EXPORT_SYMBOL(class_conn2obd);
-EXPORT_SYMBOL(class_exp2cliimp);
-EXPORT_SYMBOL(class_conn2cliimp);
-EXPORT_SYMBOL(class_disconnect);
-EXPORT_SYMBOL(class_num2obd);
-
-/* uuid.c */
-EXPORT_SYMBOL(class_uuid_unparse);
-EXPORT_SYMBOL(lustre_uuid_to_peer);
-
-EXPORT_SYMBOL(class_handle_hash);
-EXPORT_SYMBOL(class_handle_unhash);
-EXPORT_SYMBOL(class_handle_hash_back);
-EXPORT_SYMBOL(class_handle2object);
-EXPORT_SYMBOL(class_handle_free_cb);
-
-/* obd_config.c */
-EXPORT_SYMBOL(class_incref);
-EXPORT_SYMBOL(class_decref);
-EXPORT_SYMBOL(class_get_profile);
-EXPORT_SYMBOL(class_del_profile);
-EXPORT_SYMBOL(class_del_profiles);
-EXPORT_SYMBOL(class_process_config);
-EXPORT_SYMBOL(class_process_proc_param);
-EXPORT_SYMBOL(class_config_parse_llog);
-EXPORT_SYMBOL(class_config_dump_llog);
-EXPORT_SYMBOL(class_attach);
-EXPORT_SYMBOL(class_setup);
-EXPORT_SYMBOL(class_cleanup);
-EXPORT_SYMBOL(class_detach);
-EXPORT_SYMBOL(class_manual_cleanup);
-
-/* mea.c */
-EXPORT_SYMBOL(mea_name2idx);
-EXPORT_SYMBOL(raw_name2idx);
-
#define OBD_INIT_CHECK
#ifdef OBD_INIT_CHECK
int obd_init_checks(void)
CWARN("LPD64 wrong length! strlen(%s)=%d != 2\n", buf, len);
ret = -EINVAL;
}
- if ((u64val & ~CFS_PAGE_MASK) >= CFS_PAGE_SIZE) {
+ if ((u64val & ~CFS_PAGE_MASK) >= PAGE_CACHE_SIZE) {
CWARN("mask failed: u64val "LPU64" >= "LPU64"\n", u64val,
- (__u64)CFS_PAGE_SIZE);
+ (__u64)PAGE_CACHE_SIZE);
ret = -EINVAL;
}
#define obd_init_checks() do {} while(0)
#endif
-extern cfs_spinlock_t obd_types_lock;
+extern spinlock_t obd_types_lock;
extern int class_procfs_init(void);
extern int class_procfs_clean(void);
int init_obdclass(void)
#endif
{
- int err;
+ int i, err;
#ifdef __KERNEL__
- int i;
int lustre_register_fs(void);
for (i = CAPA_SITE_CLIENT; i < CAPA_SITE_MAX; i++)
CFS_INIT_LIST_HEAD(&capa_list[i]);
#endif
- LCONSOLE_INFO("OBD class driver, http://www.lustre.org/\n");
- LCONSOLE_INFO(" Lustre Version: "LUSTRE_VERSION_STRING"\n");
- LCONSOLE_INFO(" Build Version: "BUILD_VERSION"\n");
+ LCONSOLE_INFO("Lustre: Build Version: "BUILD_VERSION"\n");
- cfs_spin_lock_init(&obd_types_lock);
- cfs_waitq_init(&obd_race_waitq);
+ spin_lock_init(&obd_types_lock);
obd_zombie_impexp_init();
#ifdef LPROCFS
obd_memory = lprocfs_alloc_stats(OBD_STATS_NUM,
- LPROCFS_STATS_FLAG_NONE);
+ LPROCFS_STATS_FLAG_NONE |
+ LPROCFS_STATS_FLAG_IRQ_SAFE);
if (obd_memory == NULL) {
CERROR("kmalloc of 'obd_memory' failed\n");
RETURN(-ENOMEM);
if (err)
return err;
- cfs_spin_lock_init(&obd_dev_lock);
CFS_INIT_LIST_HEAD(&obd_types);
- err = cfs_psdev_register(&obd_psdev);
- if (err) {
- CERROR("cannot register %d err %d\n", OBD_DEV_MINOR, err);
- return err;
- }
+ err = misc_register(&obd_psdev);
+ if (err) {
+ CERROR("cannot register %d err %d\n", OBD_DEV_MINOR, err);
+ return err;
+ }
+ /* This struct is already zeroed for us (static global) */
+ for (i = 0; i < class_devno_max(); i++)
+ obd_devs[i] = NULL;
/* Default the dirty page cache cap to 1/2 of system memory.
* For clients with less memory, a larger fraction is needed
* for other purposes (mostly for BGL). */
- if (cfs_num_physpages <= 512 << (20 - CFS_PAGE_SHIFT))
- obd_max_dirty_pages = cfs_num_physpages / 4;
- else
- obd_max_dirty_pages = cfs_num_physpages / 2;
+ if (totalram_pages <= 512 << (20 - PAGE_CACHE_SHIFT))
+ obd_max_dirty_pages = totalram_pages / 4;
+ else
+ obd_max_dirty_pages = totalram_pages / 2;
err = obd_init_caches();
if (err)
return err;
#endif
- err = lu_global_init();
- if (err)
- return err;
+ err = lu_global_init();
+ if (err)
+ return err;
+
+ err = lu_capainfo_init();
+ if (err)
+ return err;
+
+ err = cl_global_init();
+ if (err != 0)
+ return err;
+
+#if defined(__KERNEL__) && defined(HAVE_SERVER_SUPPORT)
+ err = dt_global_init();
+ if (err != 0)
+ return err;
+
+ err = lu_ucred_global_init();
+ if (err != 0)
+ return err;
+#endif
+
+ err = llog_info_init();
+ if (err)
+ return err;
#ifdef __KERNEL__
err = lustre_register_fs();
- if (err)
- return err;
#endif
- err = obd_hashes_init();
- if (err)
- return err;
return err;
}
+void obd_update_maxusage(void)
+{
+ __u64 max1, max2;
+
+ max1 = obd_pages_sum();
+ max2 = obd_memory_sum();
+
+ spin_lock(&obd_updatemax_lock);
+ if (max1 > obd_max_pages)
+ obd_max_pages = max1;
+ if (max2 > obd_max_alloc)
+ obd_max_alloc = max2;
+ spin_unlock(&obd_updatemax_lock);
+}
+EXPORT_SYMBOL(obd_update_maxusage);
+
+#ifdef LPROCFS
+__u64 obd_memory_max(void)
+{
+ __u64 ret;
+
+ spin_lock(&obd_updatemax_lock);
+ ret = obd_max_alloc;
+ spin_unlock(&obd_updatemax_lock);
+
+ return ret;
+}
+EXPORT_SYMBOL(obd_memory_max);
+
+__u64 obd_pages_max(void)
+{
+ __u64 ret;
+
+ spin_lock(&obd_updatemax_lock);
+ ret = obd_max_pages;
+ spin_unlock(&obd_updatemax_lock);
+
+ return ret;
+}
+EXPORT_SYMBOL(obd_pages_max);
+#endif
+
/* liblustre doesn't call cleanup_obdclass, apparently. we carry on in this
* ifdef to the end of the file to cover module and versioning goo.*/
#ifdef __KERNEL__
static void cleanup_obdclass(void)
{
+ int i;
int lustre_unregister_fs(void);
__u64 memory_leaked, pages_leaked;
__u64 memory_max, pages_max;
- struct obd_device *obd;
ENTRY;
lustre_unregister_fs();
- cfs_psdev_deregister(&obd_psdev);
- for (obd_devlist_first(&obd);
- obd != NULL;
- obd_devlist_next(&obd)) {
- if (obd->obd_set_up &&
- OBT(obd) && OBP(obd, detach)) {
- /* XXX should this call generic detach otherwise? */
- LASSERT(obd->obd_magic == OBD_DEVICE_MAGIC);
- OBP(obd, detach)(obd);
- }
- }
- obd_hashes_fini();
-
- lu_global_fini();
+ misc_deregister(&obd_psdev);
+ for (i = 0; i < class_devno_max(); i++) {
+ struct obd_device *obd = class_num2obd(i);
+ if (obd && obd->obd_set_up &&
+ OBT(obd) && OBP(obd, detach)) {
+ /* XXX should this call generic detach otherwise? */
+ LASSERT(obd->obd_magic == OBD_DEVICE_MAGIC);
+ OBP(obd, detach)(obd);
+ }
+ }
+ llog_info_fini();
+#ifdef HAVE_SERVER_SUPPORT
+ lu_ucred_global_fini();
+ dt_global_fini();
+#endif
+ cl_global_fini();
+ lu_capainfo_fini();
+ lu_global_fini();
obd_cleanup_caches();
obd_sysctl_clean();