Whamcloud - gitweb
LU-1302 llog: structures changes, llog_thread_info
[fs/lustre-release.git] / lustre / obdclass / class_obd.c
index 243063d..1e935f4 100644 (file)
@@ -1,42 +1,49 @@
-/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
- * vim:expandtab:shiftwidth=8:tabstop=8:
+/*
+ * GPL HEADER START
  *
- * Object Devices Class Driver
+ * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
- *  Copyright (C) 2001-2003 Cluster File Systems, Inc.
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 only,
+ * as published by the Free Software Foundation.
  *
- *   This file is part of the Lustre file system, http://www.lustre.org
- *   Lustre is a trademark of Cluster File Systems, Inc.
+ * This program is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License version 2 for more details (a copy is included
+ * in the LICENSE file that accompanied this code).
  *
- *   You may have signed or agreed to another license before downloading
- *   this software.  If so, you are bound by the terms and conditions
- *   of that agreement, and the following does not apply to you.  See the
- *   LICENSE file included with this distribution for more information.
+ * You should have received a copy of the GNU General Public License
+ * version 2 along with this program; If not, see
+ * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  *
- *   If you did not agree to a different license, then this copy of Lustre
- *   is open source software; you can redistribute it and/or modify it
- *   under the terms of version 2 of the GNU General Public License as
- *   published by the Free Software Foundation.
+ * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
+ * CA 95054 USA or visit www.sun.com if you need additional information or
+ * have any questions.
  *
- *   In either case, Lustre is distributed in the hope that it will be
- *   useful, but WITHOUT ANY WARRANTY; without even the implied warranty
- *   of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
- *   license text for more details.
+ * GPL HEADER END
+ */
+/*
+ * Copyright (c) 1999, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Use is subject to license terms.
  *
- * These are the only exported functions, they provide some generic
- * infrastructure for managing object devices
+ * Copyright (c) 2011, Whamcloud, Inc.
+ */
+/*
+ * This file is part of Lustre, http://www.lustre.org/
+ * Lustre is a trademark of Sun Microsystems, Inc.
  */
 
 #define DEBUG_SUBSYSTEM S_CLASS
-#ifndef EXPORT_SYMTAB
-# define EXPORT_SYMTAB
-#endif
 #ifndef __KERNEL__
 # include <liblustre.h>
+#else
+# include <asm/atomic.h>
 #endif
 
 #include <obd_support.h>
 #include <obd_class.h>
+#include <lnet/lnetctl.h>
 #include <lustre_debug.h>
 #include <lprocfs_status.h>
 #include <lustre/lustre_build_version.h>
 
 #ifndef __KERNEL__
 /* liblustre workaround */
-atomic_t libcfs_kmemory = {0};
+cfs_atomic_t libcfs_kmemory = {0};
 #endif
 
 struct obd_device *obd_devs[MAX_OBD_DEVICES];
-struct list_head obd_types;
-spinlock_t obd_dev_lock = SPIN_LOCK_UNLOCKED;
+EXPORT_SYMBOL(obd_devs);
+cfs_list_t obd_types;
+DEFINE_RWLOCK(obd_dev_lock);
 
 #ifndef __KERNEL__
 __u64 obd_max_pages = 0;
@@ -61,47 +69,87 @@ __u64 obd_pages;
 
 /* The following are visible and mutable through /proc/sys/lustre/. */
 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;
-unsigned int obd_timeout = OBD_TIMEOUT_DEFAULT; /* seconds */
-unsigned int ldlm_timeout = LDLM_TIMEOUT_DEFAULT; /* seconds */
-unsigned int obd_health_check_timeout = HEALTH_CHECK_TIMEOUT_DEFAULT; /* seconds */
+EXPORT_SYMBOL(obd_dump_on_eviction);
 unsigned int obd_max_dirty_pages = 256;
-atomic_t obd_dirty_pages;
-
-cfs_waitq_t obd_race_waitq;
-int obd_race_state;
-
-#ifdef __KERNEL__
-unsigned int obd_print_fail_loc(void)
-{
-        CWARN("obd_fail_loc = %x\n", obd_fail_loc);
-        return obd_fail_loc;
-}
-
-void obd_set_fail_loc(unsigned int fl)
-{
-        obd_fail_loc = fl;
-}
-
-/*  opening /dev/obd */
-static int obd_class_open(unsigned long flags, void *args)
-{
-        ENTRY;
-
-        PORTAL_MODULE_USE;
-        RETURN(0);
-}
-
-/*  closing /dev/obd */
-static int obd_class_release(unsigned long flags, void *args)
+EXPORT_SYMBOL(obd_max_dirty_pages);
+cfs_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_transit_pages;
+EXPORT_SYMBOL(obd_dirty_transit_pages);
+
+char obd_jobid_var[JOBSTATS_JOBID_VAR_MAX_LEN + 1] = JOBSTATS_DISABLE;
+EXPORT_SYMBOL(obd_jobid_var);
+
+/* 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)
 {
-        ENTRY;
+#ifdef __KERNEL__
+       int jobid_len = JOBSTATS_JOBID_SIZE;
+#endif
+       int ret = 0;
+       ENTRY;
+
+       memset(jobid, 0, JOBSTATS_JOBID_SIZE);
+       /* Jobstats isn't enabled */
+       if (!memcmp(obd_jobid_var, JOBSTATS_DISABLE,
+                   strlen(JOBSTATS_DISABLE))) {
+               RETURN(0);
+       }
+
+       /* Use process name + fsuid as jobid */
+       if (!memcmp(obd_jobid_var, JOBSTATS_PROCNAME_UID,
+                   strlen(JOBSTATS_PROCNAME_UID))) {
+               snprintf(jobid, JOBSTATS_JOBID_SIZE, "%s_%u",
+                        cfs_curproc_comm(), cfs_curproc_fsuid());
+               RETURN(0);
+       }
 
-        PORTAL_MODULE_UNUSE;
-        RETURN(0);
-}
+#ifdef __KERNEL__
+       ret = cfs_get_environ(obd_jobid_var, jobid, &jobid_len);
+       if (ret) {
+               CDEBUG((ret != -ENOENT && ret != -EINVAL && ret != -EDEADLK) ?
+                      D_ERROR : D_INFO, "Get jobid for (%s) failed: rc = %d\n",
+                      obd_jobid_var, ret);
+       }
 #endif
+       RETURN(ret);
+}
+EXPORT_SYMBOL(lustre_get_jobid);
 
 static inline void obd_data2conn(struct lustre_handle *conn,
                                  struct obd_ioctl_data *data)
@@ -165,7 +213,7 @@ int class_handle_ioctl(unsigned int cmd, unsigned long arg)
         CDEBUG(D_IOCTL, "cmd = %x\n", cmd);
         if (obd_ioctl_getdata(&buf, &len, (void *)arg)) {
                 CERROR("OBD ioctl: data error\n");
-                GOTO(out, err = -EINVAL);
+                RETURN(-EINVAL);
         }
         data = (struct obd_ioctl_data *)buf;
 
@@ -180,7 +228,8 @@ int class_handle_ioctl(unsigned int cmd, unsigned long arg)
                 OBD_ALLOC(lcfg, data->ioc_plen1);
                 if (lcfg == NULL)
                         GOTO(out, err = -ENOMEM);
-                err = copy_from_user(lcfg, data->ioc_pbuf1, data->ioc_plen1);
+                err = cfs_copy_from_user(lcfg, data->ioc_pbuf1,
+                                         data->ioc_plen1);
                 if (!err)
                         err = lustre_cfg_sanity_check(lcfg, data->ioc_plen1);
                 if (!err)
@@ -279,7 +328,7 @@ int class_handle_ioctl(unsigned int cmd, unsigned long arg)
                         CERROR("ioctl buffer too small to hold version\n");
                         GOTO(out, err = -EINVAL);
                 }
-                                
+
                 obd = class_num2obd(index);
                 if (!obd)
                         GOTO(out, err = -ENOENT);
@@ -296,7 +345,7 @@ int class_handle_ioctl(unsigned int cmd, unsigned long arg)
                 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,
-                         atomic_read(&obd->obd_refcount));
+                         cfs_atomic_read(&obd->obd_refcount));
                 err = obd_ioctl_popdata((void *)arg, data, len);
 
                 GOTO(out, err = 0);
@@ -304,12 +353,19 @@ int class_handle_ioctl(unsigned int cmd, unsigned long arg)
 
         }
 
-        if (data->ioc_dev >= class_devno_max()) {
+        if (data->ioc_dev == OBD_DEV_BY_DEVNAME) {
+                if (data->ioc_inllen4 <= 0 || data->ioc_inlbuf4 == NULL)
+                        GOTO(out, err = -EINVAL);
+                if (strnlen(data->ioc_inlbuf4, MAX_OBD_NAME) >= MAX_OBD_NAME)
+                        GOTO(out, err = -EINVAL);
+                obd = class_name2obd(data->ioc_inlbuf4);
+        } else if (data->ioc_dev < class_devno_max()) {
+                obd = class_num2obd(data->ioc_dev);
+        } else {
                 CERROR("OBD ioctl: No device\n");
                 GOTO(out, err = -EINVAL);
         }
 
-        obd = class_num2obd(data->ioc_dev);
         if (obd == NULL) {
                 CERROR("OBD ioctl : No Device %d\n", data->ioc_dev);
                 GOTO(out, err = -EINVAL);
@@ -351,96 +407,12 @@ int class_handle_ioctl(unsigned int cmd, unsigned long arg)
         RETURN(err);
 } /* class_handle_ioctl */
 
-
-
-#define OBD_MINOR 241
 #ifdef __KERNEL__
-/* to control /dev/obd */
-static int obd_class_ioctl (struct cfs_psdev_file *pfile, unsigned long cmd,
-                            void *arg)
-{
-        return class_handle_ioctl(cmd, (unsigned long)arg);
-}
-
-/* declare character device */
-struct cfs_psdev_ops obd_psdev_ops = {
-        /* .p_open    = */ obd_class_open,      /* open */
-        /* .p_close   = */ obd_class_release,   /* release */
-        /* .p_read    = */ NULL,
-        /* .p_write   = */ NULL,
-        /* .p_ioctl   = */ obd_class_ioctl     /* ioctl */
-};
-
 extern cfs_psdev_t obd_psdev;
 #else
 void *obd_psdev = NULL;
 #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_health_check_timeout);
-EXPORT_SYMBOL(obd_max_dirty_pages);
-EXPORT_SYMBOL(obd_dirty_pages);
-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_name2dev);
-EXPORT_SYMBOL(class_name2obd);
-EXPORT_SYMBOL(class_uuid2dev);
-EXPORT_SYMBOL(class_uuid2obd);
-EXPORT_SYMBOL(class_find_client_obd);
-EXPORT_SYMBOL(class_find_client_notype);
-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)
@@ -449,8 +421,7 @@ int obd_init_checks(void)
         char buf[64];
         int len, ret = 0;
 
-        CDEBUG(D_INFO, "LPU64=%s, LPD64=%s, LPX64=%s, LPSZ=%s, LPSSZ=%s\n",
-               LPU64, LPD64, LPX64, LPSZ, LPSSZ);
+        CDEBUG(D_INFO, "LPU64=%s, LPD64=%s, LPX64=%s\n", LPU64, LPD64, LPX64);
 
         CDEBUG(D_INFO, "OBD_OBJECT_EOF = "LPX64"\n", (__u64)OBD_OBJECT_EOF);
 
@@ -504,8 +475,8 @@ int obd_init_checks(void)
                 ret = -EINVAL;
         }
         if ((u64val & ~CFS_PAGE_MASK) >= CFS_PAGE_SIZE) {
-                CWARN("mask failed: u64val "LPU64" >= %lu\n", u64val,
-                      CFS_PAGE_SIZE);
+                CWARN("mask failed: u64val "LPU64" >= "LPU64"\n", u64val,
+                      (__u64)CFS_PAGE_SIZE);
                 ret = -EINVAL;
         }
 
@@ -515,7 +486,7 @@ int obd_init_checks(void)
 #define obd_init_checks() do {} while(0)
 #endif
 
-extern spinlock_t obd_types_lock;
+extern cfs_spinlock_t obd_types_lock;
 extern int class_procfs_init(void);
 extern int class_procfs_clean(void);
 
@@ -529,34 +500,28 @@ int init_obdclass(void)
 #ifdef __KERNEL__
         int lustre_register_fs(void);
 
-        printk(KERN_INFO "Lustre: OBD class driver, info@clusterfs.com\n");
-        printk(KERN_INFO "        Lustre Version: "LUSTRE_VERSION_STRING"\n");
-        printk(KERN_INFO "        Build Version: "BUILD_VERSION"\n");
-
         for (i = CAPA_SITE_CLIENT; i < CAPA_SITE_MAX; i++)
-                INIT_LIST_HEAD(&capa_list[i]);
-#else
-        CDEBUG(D_INFO, "Lustre: OBD class driver, info@clusterfs.com\n");
-        CDEBUG(D_INFO, "        Lustre Version: "LUSTRE_VERSION_STRING"\n");
-        CDEBUG(D_INFO, "        Build Version: "BUILD_VERSION"\n");
+                CFS_INIT_LIST_HEAD(&capa_list[i]);
 #endif
 
-        spin_lock_init(&obd_types_lock);
-        cfs_waitq_init(&obd_race_waitq);
+        LCONSOLE_INFO("Lustre: Build Version: "BUILD_VERSION"\n");
+
+        cfs_spin_lock_init(&obd_types_lock);
         obd_zombie_impexp_init();
 #ifdef LPROCFS
-        obd_memory = lprocfs_alloc_stats(OBD_STATS_NUM, 
-                                         LPROCFS_STATS_FLAG_PERCPU);
+        obd_memory = lprocfs_alloc_stats(OBD_STATS_NUM,
+                                        LPROCFS_STATS_FLAG_NONE |
+                                        LPROCFS_STATS_FLAG_IRQ_SAFE);
         if (obd_memory == NULL) {
                 CERROR("kmalloc of 'obd_memory' failed\n");
                 RETURN(-ENOMEM);
         }
+
         lprocfs_counter_init(obd_memory, OBD_MEMORY_STAT,
-                             LPROCFS_CNTR_AVGMINMAX, 
+                             LPROCFS_CNTR_AVGMINMAX,
                              "memused", "bytes");
         lprocfs_counter_init(obd_memory, OBD_MEMORY_PAGES_STAT,
-                             LPROCFS_CNTR_AVGMINMAX, 
+                             LPROCFS_CNTR_AVGMINMAX,
                              "pagesused", "pages");
 #endif
         err = obd_init_checks();
@@ -568,32 +533,44 @@ int init_obdclass(void)
         if (err)
                 return err;
 
-        spin_lock_init(&obd_dev_lock);
-        INIT_LIST_HEAD(&obd_types);
+        CFS_INIT_LIST_HEAD(&obd_types);
 
         err = cfs_psdev_register(&obd_psdev);
         if (err) {
-                CERROR("cannot register %d err %d\n", OBD_MINOR, err);
+                CERROR("cannot register %d err %d\n", OBD_DEV_MINOR, err);
                 return err;
         }
 
-        /* This struct is already zerod for us (static global) */
+        /* 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 */
-        obd_max_dirty_pages = num_physpages / 2;
+        /* 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;
 
         err = obd_init_caches();
         if (err)
                 return err;
 #ifdef __KERNEL__
-        err = lu_global_init();
+        err = class_procfs_init();
         if (err)
                 return err;
-        err = class_procfs_init();
+#endif
+
+        err = lu_global_init();
         if (err)
                 return err;
+
+       err = llog_info_init();
+       if (err)
+               return err;
+
+#ifdef __KERNEL__
         err = lustre_register_fs();
 #endif
 
@@ -623,6 +600,7 @@ static void cleanup_obdclass(void)
                         OBP(obd, detach)(obd);
                 }
         }
+       llog_info_fini();
         lu_global_fini();
 
         obd_cleanup_caches();
@@ -632,27 +610,26 @@ static void cleanup_obdclass(void)
 
         class_handle_cleanup();
         class_exit_uuidlist();
+        obd_zombie_impexp_stop();
 
         memory_leaked = obd_memory_sum();
         pages_leaked = obd_pages_sum();
-        
+
         memory_max = obd_memory_max();
         pages_max = obd_pages_max();
 
         lprocfs_free_stats(&obd_memory);
-        if (memory_leaked > 0) {
-                CWARN("Memory leaks detected (max "LPU64", leaked "LPD64")\n",
-                      memory_max, memory_leaked);
-        }
-        if (pages_leaked > 0) {
-                CWARN("Page leaks detected (max "LPU64", leaked "LPU64")\n",
-                      pages_max, pages_leaked);
-        }
+        CDEBUG((memory_leaked) ? D_ERROR : D_INFO,
+               "obd_memory max: "LPU64", leaked: "LPU64"\n",
+               memory_max, memory_leaked);
+        CDEBUG((pages_leaked) ? D_ERROR : D_INFO,
+               "obd_memory_pages max: "LPU64", leaked: "LPU64"\n",
+               pages_max, pages_leaked);
+
         EXIT;
 }
 
-MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
+MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
 MODULE_DESCRIPTION("Lustre Class Driver Build Version: " BUILD_VERSION);
 MODULE_LICENSE("GPL");