4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 1999, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2012, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
37 #define DEBUG_SUBSYSTEM S_CLASS
39 # include <liblustre.h>
41 # include <asm/atomic.h>
44 #include <obd_support.h>
45 #include <obd_class.h>
46 #include <lnet/lnetctl.h>
47 #include <lustre_debug.h>
48 #include <lprocfs_status.h>
49 #include <lustre/lustre_build_version.h>
50 #include <libcfs/list.h>
51 #include <cl_object.h>
52 #ifdef HAVE_SERVER_SUPPORT
53 # include <dt_object.h>
54 # include <md_object.h>
55 #endif /* HAVE_SERVER_SUPPORT */
56 #include "llog_internal.h"
59 /* liblustre workaround */
60 cfs_atomic_t libcfs_kmemory = {0};
63 struct obd_device *obd_devs[MAX_OBD_DEVICES];
64 EXPORT_SYMBOL(obd_devs);
66 DEFINE_RWLOCK(obd_dev_lock);
68 __u64 obd_max_pages = 0;
69 __u64 obd_max_alloc = 0;
74 DEFINE_SPINLOCK(obd_updatemax_lock);
76 /* The following are visible and mutable through /proc/sys/lustre/. */
77 unsigned int obd_alloc_fail_rate = 0;
78 EXPORT_SYMBOL(obd_alloc_fail_rate);
79 unsigned int obd_debug_peer_on_timeout;
80 EXPORT_SYMBOL(obd_debug_peer_on_timeout);
81 unsigned int obd_dump_on_timeout;
82 EXPORT_SYMBOL(obd_dump_on_timeout);
83 unsigned int obd_dump_on_eviction;
84 EXPORT_SYMBOL(obd_dump_on_eviction);
85 unsigned int obd_max_dirty_pages = 256;
86 EXPORT_SYMBOL(obd_max_dirty_pages);
87 cfs_atomic_t obd_dirty_pages;
88 EXPORT_SYMBOL(obd_dirty_pages);
89 unsigned int obd_timeout = OBD_TIMEOUT_DEFAULT; /* seconds */
90 EXPORT_SYMBOL(obd_timeout);
91 unsigned int ldlm_timeout = LDLM_TIMEOUT_DEFAULT; /* seconds */
92 EXPORT_SYMBOL(ldlm_timeout);
93 unsigned int obd_timeout_set;
94 EXPORT_SYMBOL(obd_timeout_set);
95 unsigned int ldlm_timeout_set;
96 EXPORT_SYMBOL(ldlm_timeout_set);
97 /* Adaptive timeout defs here instead of ptlrpc module for /proc/sys/ access */
98 unsigned int at_min = 0;
99 EXPORT_SYMBOL(at_min);
100 unsigned int at_max = 600;
101 EXPORT_SYMBOL(at_max);
102 unsigned int at_history = 600;
103 EXPORT_SYMBOL(at_history);
104 int at_early_margin = 5;
105 EXPORT_SYMBOL(at_early_margin);
107 EXPORT_SYMBOL(at_extra);
109 cfs_atomic_t obd_dirty_transit_pages;
110 EXPORT_SYMBOL(obd_dirty_transit_pages);
112 char obd_jobid_var[JOBSTATS_JOBID_VAR_MAX_LEN + 1] = JOBSTATS_DISABLE;
113 EXPORT_SYMBOL(obd_jobid_var);
115 /* Get jobid of current process by reading the environment variable
116 * stored in between the "env_start" & "env_end" of task struct.
119 * It's better to cache the jobid for later use if there is any
120 * efficient way, the cl_env code probably could be reused for this
123 * If some job scheduler doesn't store jobid in the "env_start/end",
124 * then an upcall could be issued here to get the jobid by utilizing
125 * the userspace tools/api. Then, the jobid must be cached.
127 int lustre_get_jobid(char *jobid)
129 int jobid_len = JOBSTATS_JOBID_SIZE;
133 memset(jobid, 0, JOBSTATS_JOBID_SIZE);
134 /* Jobstats isn't enabled */
135 if (strcmp(obd_jobid_var, JOBSTATS_DISABLE) == 0)
138 /* Use process name + fsuid as jobid */
139 if (strcmp(obd_jobid_var, JOBSTATS_PROCNAME_UID) == 0) {
140 snprintf(jobid, JOBSTATS_JOBID_SIZE, "%s.%u",
141 cfs_curproc_comm(), cfs_curproc_fsuid());
145 rc = cfs_get_environ(obd_jobid_var, jobid, &jobid_len);
147 if (rc == -EOVERFLOW) {
148 /* For the PBS_JOBID and LOADL_STEP_ID keys (which are
149 * variable length strings instead of just numbers), it
150 * might make sense to keep the unique parts for JobID,
151 * instead of just returning an error. That means a
152 * larger temp buffer for cfs_get_environ(), then
153 * truncating the string at some separator to fit into
154 * the specified jobid_len. Fix later if needed. */
156 if (unlikely(!printed)) {
157 LCONSOLE_ERROR_MSG(0x16b, "%s value too large "
158 "for JobID buffer (%d)\n",
159 obd_jobid_var, jobid_len);
163 CDEBUG((rc == -ENOENT || rc == -EINVAL ||
164 rc == -EDEADLK) ? D_INFO : D_ERROR,
165 "Get jobid for (%s) failed: rc = %d\n",
171 EXPORT_SYMBOL(lustre_get_jobid);
173 int obd_alloc_fail(const void *ptr, const char *name, const char *type,
174 size_t size, const char *file, int line)
177 (cfs_rand() & OBD_ALLOC_FAIL_MASK) < obd_alloc_fail_rate) {
178 CERROR("%s%salloc of %s ("LPU64" bytes) failed at %s:%d\n",
179 ptr ? "force " :"", type, name, (__u64)size, file,
181 CERROR(LPU64" total bytes and "LPU64" total pages "
182 "("LPU64" bytes) allocated by Lustre, "
183 "%d total bytes by LNET\n",
185 obd_pages_sum() << PAGE_CACHE_SHIFT,
187 cfs_atomic_read(&libcfs_kmemory));
192 EXPORT_SYMBOL(obd_alloc_fail);
194 static inline void obd_data2conn(struct lustre_handle *conn,
195 struct obd_ioctl_data *data)
197 memset(conn, 0, sizeof *conn);
198 conn->cookie = data->ioc_cookie;
201 static inline void obd_conn2data(struct obd_ioctl_data *data,
202 struct lustre_handle *conn)
204 data->ioc_cookie = conn->cookie;
207 int class_resolve_dev_name(__u32 len, const char *name)
214 CERROR("No name passed,!\n");
215 GOTO(out, rc = -EINVAL);
217 if (name[len - 1] != 0) {
218 CERROR("Name not nul terminated!\n");
219 GOTO(out, rc = -EINVAL);
222 CDEBUG(D_IOCTL, "device name %s\n", name);
223 dev = class_name2dev(name);
225 CDEBUG(D_IOCTL, "No device for name %s!\n", name);
226 GOTO(out, rc = -EINVAL);
229 CDEBUG(D_IOCTL, "device name %s, dev %d\n", name, dev);
236 int class_handle_ioctl(unsigned int cmd, unsigned long arg)
239 struct obd_ioctl_data *data;
240 struct libcfs_debug_ioctl_data *debug_data;
241 struct obd_device *obd = NULL;
242 int err = 0, len = 0;
245 /* only for debugging */
246 if (cmd == LIBCFS_IOC_DEBUG_MASK) {
247 debug_data = (struct libcfs_debug_ioctl_data*)arg;
248 libcfs_subsystem_debug = debug_data->subs;
249 libcfs_debug = debug_data->debug;
253 CDEBUG(D_IOCTL, "cmd = %x\n", cmd);
254 if (obd_ioctl_getdata(&buf, &len, (void *)arg)) {
255 CERROR("OBD ioctl: data error\n");
258 data = (struct obd_ioctl_data *)buf;
261 case OBD_IOC_PROCESS_CFG: {
262 struct lustre_cfg *lcfg;
264 if (!data->ioc_plen1 || !data->ioc_pbuf1) {
265 CERROR("No config buffer passed!\n");
266 GOTO(out, err = -EINVAL);
268 OBD_ALLOC(lcfg, data->ioc_plen1);
270 GOTO(out, err = -ENOMEM);
271 err = copy_from_user(lcfg, data->ioc_pbuf1,
274 err = lustre_cfg_sanity_check(lcfg, data->ioc_plen1);
276 err = class_process_config(lcfg);
278 OBD_FREE(lcfg, data->ioc_plen1);
282 case OBD_GET_VERSION:
283 if (!data->ioc_inlbuf1) {
284 CERROR("No buffer passed in ioctl\n");
285 GOTO(out, err = -EINVAL);
288 if (strlen(BUILD_VERSION) + 1 > data->ioc_inllen1) {
289 CERROR("ioctl buffer too small to hold version\n");
290 GOTO(out, err = -EINVAL);
293 memcpy(data->ioc_bulk, BUILD_VERSION,
294 strlen(BUILD_VERSION) + 1);
296 err = obd_ioctl_popdata((void *)arg, data, len);
301 case OBD_IOC_NAME2DEV: {
302 /* Resolve a device name. This does not change the
303 * currently selected device.
307 dev = class_resolve_dev_name(data->ioc_inllen1,
311 GOTO(out, err = -EINVAL);
313 err = obd_ioctl_popdata((void *)arg, data, sizeof(*data));
319 case OBD_IOC_UUID2DEV: {
320 /* Resolve a device uuid. This does not change the
321 * currently selected device.
324 struct obd_uuid uuid;
326 if (!data->ioc_inllen1 || !data->ioc_inlbuf1) {
327 CERROR("No UUID passed!\n");
328 GOTO(out, err = -EINVAL);
330 if (data->ioc_inlbuf1[data->ioc_inllen1 - 1] != 0) {
331 CERROR("UUID not NUL terminated!\n");
332 GOTO(out, err = -EINVAL);
335 CDEBUG(D_IOCTL, "device name %s\n", data->ioc_inlbuf1);
336 obd_str2uuid(&uuid, data->ioc_inlbuf1);
337 dev = class_uuid2dev(&uuid);
340 CDEBUG(D_IOCTL, "No device for UUID %s!\n",
342 GOTO(out, err = -EINVAL);
345 CDEBUG(D_IOCTL, "device name %s, dev %d\n", data->ioc_inlbuf1,
347 err = obd_ioctl_popdata((void *)arg, data, sizeof(*data));
353 case OBD_IOC_CLOSE_UUID: {
354 CDEBUG(D_IOCTL, "closing all connections to uuid %s (NOOP)\n",
359 case OBD_IOC_GETDEVICE: {
360 int index = data->ioc_count;
363 if (!data->ioc_inlbuf1) {
364 CERROR("No buffer passed in ioctl\n");
365 GOTO(out, err = -EINVAL);
367 if (data->ioc_inllen1 < 128) {
368 CERROR("ioctl buffer too small to hold version\n");
369 GOTO(out, err = -EINVAL);
372 obd = class_num2obd(index);
374 GOTO(out, err = -ENOENT);
376 if (obd->obd_stopping)
378 else if (obd->obd_set_up)
380 else if (obd->obd_attached)
384 str = (char *)data->ioc_bulk;
385 snprintf(str, len - sizeof(*data), "%3d %s %s %s %s %d",
386 (int)index, status, obd->obd_type->typ_name,
387 obd->obd_name, obd->obd_uuid.uuid,
388 cfs_atomic_read(&obd->obd_refcount));
389 err = obd_ioctl_popdata((void *)arg, data, len);
396 if (data->ioc_dev == OBD_DEV_BY_DEVNAME) {
397 if (data->ioc_inllen4 <= 0 || data->ioc_inlbuf4 == NULL)
398 GOTO(out, err = -EINVAL);
399 if (strnlen(data->ioc_inlbuf4, MAX_OBD_NAME) >= MAX_OBD_NAME)
400 GOTO(out, err = -EINVAL);
401 obd = class_name2obd(data->ioc_inlbuf4);
402 } else if (data->ioc_dev < class_devno_max()) {
403 obd = class_num2obd(data->ioc_dev);
405 CERROR("OBD ioctl: No device\n");
406 GOTO(out, err = -EINVAL);
410 CERROR("OBD ioctl : No Device %d\n", data->ioc_dev);
411 GOTO(out, err = -EINVAL);
413 LASSERT(obd->obd_magic == OBD_DEVICE_MAGIC);
415 if (!obd->obd_set_up || obd->obd_stopping) {
416 CERROR("OBD ioctl: device not setup %d \n", data->ioc_dev);
417 GOTO(out, err = -EINVAL);
421 case OBD_IOC_NO_TRANSNO: {
422 if (!obd->obd_attached) {
423 CERROR("Device %d not attached\n", obd->obd_minor);
424 GOTO(out, err = -ENODEV);
426 CDEBUG(D_HA, "%s: disabling committed-transno notification\n",
428 obd->obd_no_transno = 1;
433 err = obd_iocontrol(cmd, obd->obd_self_export, len, data, NULL);
437 err = obd_ioctl_popdata((void *)arg, data, len);
446 obd_ioctl_freedata(buf, len);
448 } /* class_handle_ioctl */
451 extern cfs_psdev_t obd_psdev;
453 void *obd_psdev = NULL;
456 #define OBD_INIT_CHECK
457 #ifdef OBD_INIT_CHECK
458 int obd_init_checks(void)
460 __u64 u64val, div64val;
464 CDEBUG(D_INFO, "LPU64=%s, LPD64=%s, LPX64=%s\n", LPU64, LPD64, LPX64);
466 CDEBUG(D_INFO, "OBD_OBJECT_EOF = "LPX64"\n", (__u64)OBD_OBJECT_EOF);
468 u64val = OBD_OBJECT_EOF;
469 CDEBUG(D_INFO, "u64val OBD_OBJECT_EOF = "LPX64"\n", u64val);
470 if (u64val != OBD_OBJECT_EOF) {
471 CERROR("__u64 "LPX64"(%d) != 0xffffffffffffffff\n",
472 u64val, (int)sizeof(u64val));
475 len = snprintf(buf, sizeof(buf), LPX64, u64val);
477 CWARN("LPX64 wrong length! strlen(%s)=%d != 18\n", buf, len);
481 div64val = OBD_OBJECT_EOF;
482 CDEBUG(D_INFO, "u64val OBD_OBJECT_EOF = "LPX64"\n", u64val);
483 if (u64val != OBD_OBJECT_EOF) {
484 CERROR("__u64 "LPX64"(%d) != 0xffffffffffffffff\n",
485 u64val, (int)sizeof(u64val));
488 if (u64val >> 8 != OBD_OBJECT_EOF >> 8) {
489 CERROR("__u64 "LPX64"(%d) != 0xffffffffffffffff\n",
490 u64val, (int)sizeof(u64val));
493 if (do_div(div64val, 256) != (u64val & 255)) {
494 CERROR("do_div("LPX64",256) != "LPU64"\n", u64val, u64val &255);
497 if (u64val >> 8 != div64val) {
498 CERROR("do_div("LPX64",256) "LPU64" != "LPU64"\n",
499 u64val, div64val, u64val >> 8);
502 len = snprintf(buf, sizeof(buf), LPX64, u64val);
504 CWARN("LPX64 wrong length! strlen(%s)=%d != 18\n", buf, len);
507 len = snprintf(buf, sizeof(buf), LPU64, u64val);
509 CWARN("LPU64 wrong length! strlen(%s)=%d != 20\n", buf, len);
512 len = snprintf(buf, sizeof(buf), LPD64, u64val);
514 CWARN("LPD64 wrong length! strlen(%s)=%d != 2\n", buf, len);
517 if ((u64val & ~CFS_PAGE_MASK) >= PAGE_CACHE_SIZE) {
518 CWARN("mask failed: u64val "LPU64" >= "LPU64"\n", u64val,
519 (__u64)PAGE_CACHE_SIZE);
526 #define obd_init_checks() do {} while(0)
529 extern spinlock_t obd_types_lock;
530 extern int class_procfs_init(void);
531 extern int class_procfs_clean(void);
534 static int __init init_obdclass(void)
536 int init_obdclass(void)
541 int lustre_register_fs(void);
543 for (i = CAPA_SITE_CLIENT; i < CAPA_SITE_MAX; i++)
544 CFS_INIT_LIST_HEAD(&capa_list[i]);
547 LCONSOLE_INFO("Lustre: Build Version: "BUILD_VERSION"\n");
549 spin_lock_init(&obd_types_lock);
550 obd_zombie_impexp_init();
552 obd_memory = lprocfs_alloc_stats(OBD_STATS_NUM,
553 LPROCFS_STATS_FLAG_NONE |
554 LPROCFS_STATS_FLAG_IRQ_SAFE);
555 if (obd_memory == NULL) {
556 CERROR("kmalloc of 'obd_memory' failed\n");
560 lprocfs_counter_init(obd_memory, OBD_MEMORY_STAT,
561 LPROCFS_CNTR_AVGMINMAX,
563 lprocfs_counter_init(obd_memory, OBD_MEMORY_PAGES_STAT,
564 LPROCFS_CNTR_AVGMINMAX,
565 "pagesused", "pages");
567 err = obd_init_checks();
568 if (err == -EOVERFLOW)
571 class_init_uuidlist();
572 err = class_handle_init();
576 CFS_INIT_LIST_HEAD(&obd_types);
578 err = cfs_psdev_register(&obd_psdev);
580 CERROR("cannot register %d err %d\n", OBD_DEV_MINOR, err);
584 /* This struct is already zeroed for us (static global) */
585 for (i = 0; i < class_devno_max(); i++)
588 /* Default the dirty page cache cap to 1/2 of system memory.
589 * For clients with less memory, a larger fraction is needed
590 * for other purposes (mostly for BGL). */
591 if (num_physpages <= 512 << (20 - PAGE_CACHE_SHIFT))
592 obd_max_dirty_pages = num_physpages / 4;
594 obd_max_dirty_pages = num_physpages / 2;
596 err = obd_init_caches();
600 err = class_procfs_init();
605 err = lu_global_init();
609 err = cl_global_init();
613 #if defined(__KERNEL__) && defined(HAVE_SERVER_SUPPORT)
614 err = dt_global_init();
618 err = lu_ucred_global_init();
623 err = llog_info_init();
628 err = lustre_register_fs();
634 void obd_update_maxusage(void)
638 max1 = obd_pages_sum();
639 max2 = obd_memory_sum();
641 spin_lock(&obd_updatemax_lock);
642 if (max1 > obd_max_pages)
643 obd_max_pages = max1;
644 if (max2 > obd_max_alloc)
645 obd_max_alloc = max2;
646 spin_unlock(&obd_updatemax_lock);
648 EXPORT_SYMBOL(obd_update_maxusage);
651 __u64 obd_memory_max(void)
655 spin_lock(&obd_updatemax_lock);
657 spin_unlock(&obd_updatemax_lock);
661 EXPORT_SYMBOL(obd_memory_max);
663 __u64 obd_pages_max(void)
667 spin_lock(&obd_updatemax_lock);
669 spin_unlock(&obd_updatemax_lock);
673 EXPORT_SYMBOL(obd_pages_max);
676 /* liblustre doesn't call cleanup_obdclass, apparently. we carry on in this
677 * ifdef to the end of the file to cover module and versioning goo.*/
679 static void cleanup_obdclass(void)
682 int lustre_unregister_fs(void);
683 __u64 memory_leaked, pages_leaked;
684 __u64 memory_max, pages_max;
687 lustre_unregister_fs();
689 cfs_psdev_deregister(&obd_psdev);
690 for (i = 0; i < class_devno_max(); i++) {
691 struct obd_device *obd = class_num2obd(i);
692 if (obd && obd->obd_set_up &&
693 OBT(obd) && OBP(obd, detach)) {
694 /* XXX should this call generic detach otherwise? */
695 LASSERT(obd->obd_magic == OBD_DEVICE_MAGIC);
696 OBP(obd, detach)(obd);
700 #ifdef HAVE_SERVER_SUPPORT
701 lu_ucred_global_fini();
707 obd_cleanup_caches();
710 class_procfs_clean();
712 class_handle_cleanup();
713 class_exit_uuidlist();
714 obd_zombie_impexp_stop();
716 memory_leaked = obd_memory_sum();
717 pages_leaked = obd_pages_sum();
719 memory_max = obd_memory_max();
720 pages_max = obd_pages_max();
722 lprocfs_free_stats(&obd_memory);
723 CDEBUG((memory_leaked) ? D_ERROR : D_INFO,
724 "obd_memory max: "LPU64", leaked: "LPU64"\n",
725 memory_max, memory_leaked);
726 CDEBUG((pages_leaked) ? D_ERROR : D_INFO,
727 "obd_memory_pages max: "LPU64", leaked: "LPU64"\n",
728 pages_max, pages_leaked);
733 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
734 MODULE_DESCRIPTION("Lustre Class Driver Build Version: " BUILD_VERSION);
735 MODULE_LICENSE("GPL");
737 cfs_module(obdclass, LUSTRE_VERSION_STRING, init_obdclass, cleanup_obdclass);