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_unstable_pages;
88 EXPORT_SYMBOL(obd_unstable_pages);
89 cfs_atomic_t obd_dirty_pages;
90 EXPORT_SYMBOL(obd_dirty_pages);
91 unsigned int obd_timeout = OBD_TIMEOUT_DEFAULT; /* seconds */
92 EXPORT_SYMBOL(obd_timeout);
93 unsigned int ldlm_timeout = LDLM_TIMEOUT_DEFAULT; /* seconds */
94 EXPORT_SYMBOL(ldlm_timeout);
95 unsigned int obd_timeout_set;
96 EXPORT_SYMBOL(obd_timeout_set);
97 unsigned int ldlm_timeout_set;
98 EXPORT_SYMBOL(ldlm_timeout_set);
99 /* Adaptive timeout defs here instead of ptlrpc module for /proc/sys/ access */
100 unsigned int at_min = 0;
101 EXPORT_SYMBOL(at_min);
102 unsigned int at_max = 600;
103 EXPORT_SYMBOL(at_max);
104 unsigned int at_history = 600;
105 EXPORT_SYMBOL(at_history);
106 int at_early_margin = 5;
107 EXPORT_SYMBOL(at_early_margin);
109 EXPORT_SYMBOL(at_extra);
111 cfs_atomic_t obd_dirty_transit_pages;
112 EXPORT_SYMBOL(obd_dirty_transit_pages);
114 char obd_jobid_var[JOBSTATS_JOBID_VAR_MAX_LEN + 1] = JOBSTATS_DISABLE;
115 EXPORT_SYMBOL(obd_jobid_var);
117 /* Get jobid of current process by reading the environment variable
118 * stored in between the "env_start" & "env_end" of task struct.
121 * It's better to cache the jobid for later use if there is any
122 * efficient way, the cl_env code probably could be reused for this
125 * If some job scheduler doesn't store jobid in the "env_start/end",
126 * then an upcall could be issued here to get the jobid by utilizing
127 * the userspace tools/api. Then, the jobid must be cached.
129 int lustre_get_jobid(char *jobid)
131 int jobid_len = JOBSTATS_JOBID_SIZE;
135 memset(jobid, 0, JOBSTATS_JOBID_SIZE);
136 /* Jobstats isn't enabled */
137 if (strcmp(obd_jobid_var, JOBSTATS_DISABLE) == 0)
140 /* Use process name + fsuid as jobid */
141 if (strcmp(obd_jobid_var, JOBSTATS_PROCNAME_UID) == 0) {
142 snprintf(jobid, JOBSTATS_JOBID_SIZE, "%s.%u",
143 cfs_curproc_comm(), cfs_curproc_fsuid());
147 rc = cfs_get_environ(obd_jobid_var, jobid, &jobid_len);
149 if (rc == -EOVERFLOW) {
150 /* For the PBS_JOBID and LOADL_STEP_ID keys (which are
151 * variable length strings instead of just numbers), it
152 * might make sense to keep the unique parts for JobID,
153 * instead of just returning an error. That means a
154 * larger temp buffer for cfs_get_environ(), then
155 * truncating the string at some separator to fit into
156 * the specified jobid_len. Fix later if needed. */
158 if (unlikely(!printed)) {
159 LCONSOLE_ERROR_MSG(0x16b, "%s value too large "
160 "for JobID buffer (%d)\n",
161 obd_jobid_var, jobid_len);
165 CDEBUG((rc == -ENOENT || rc == -EINVAL ||
166 rc == -EDEADLK) ? D_INFO : D_ERROR,
167 "Get jobid for (%s) failed: rc = %d\n",
173 EXPORT_SYMBOL(lustre_get_jobid);
175 int obd_alloc_fail(const void *ptr, const char *name, const char *type,
176 size_t size, const char *file, int line)
179 (cfs_rand() & OBD_ALLOC_FAIL_MASK) < obd_alloc_fail_rate) {
180 CERROR("%s%salloc of %s ("LPU64" bytes) failed at %s:%d\n",
181 ptr ? "force " :"", type, name, (__u64)size, file,
183 CERROR(LPU64" total bytes and "LPU64" total pages "
184 "("LPU64" bytes) allocated by Lustre, "
185 "%d total bytes by LNET\n",
187 obd_pages_sum() << CFS_PAGE_SHIFT,
189 cfs_atomic_read(&libcfs_kmemory));
194 EXPORT_SYMBOL(obd_alloc_fail);
196 static inline void obd_data2conn(struct lustre_handle *conn,
197 struct obd_ioctl_data *data)
199 memset(conn, 0, sizeof *conn);
200 conn->cookie = data->ioc_cookie;
203 static inline void obd_conn2data(struct obd_ioctl_data *data,
204 struct lustre_handle *conn)
206 data->ioc_cookie = conn->cookie;
209 int class_resolve_dev_name(__u32 len, const char *name)
216 CERROR("No name passed,!\n");
217 GOTO(out, rc = -EINVAL);
219 if (name[len - 1] != 0) {
220 CERROR("Name not nul terminated!\n");
221 GOTO(out, rc = -EINVAL);
224 CDEBUG(D_IOCTL, "device name %s\n", name);
225 dev = class_name2dev(name);
227 CDEBUG(D_IOCTL, "No device for name %s!\n", name);
228 GOTO(out, rc = -EINVAL);
231 CDEBUG(D_IOCTL, "device name %s, dev %d\n", name, dev);
238 int class_handle_ioctl(unsigned int cmd, unsigned long arg)
241 struct obd_ioctl_data *data;
242 struct libcfs_debug_ioctl_data *debug_data;
243 struct obd_device *obd = NULL;
244 int err = 0, len = 0;
247 /* only for debugging */
248 if (cmd == LIBCFS_IOC_DEBUG_MASK) {
249 debug_data = (struct libcfs_debug_ioctl_data*)arg;
250 libcfs_subsystem_debug = debug_data->subs;
251 libcfs_debug = debug_data->debug;
255 CDEBUG(D_IOCTL, "cmd = %x\n", cmd);
256 if (obd_ioctl_getdata(&buf, &len, (void *)arg)) {
257 CERROR("OBD ioctl: data error\n");
260 data = (struct obd_ioctl_data *)buf;
263 case OBD_IOC_PROCESS_CFG: {
264 struct lustre_cfg *lcfg;
266 if (!data->ioc_plen1 || !data->ioc_pbuf1) {
267 CERROR("No config buffer passed!\n");
268 GOTO(out, err = -EINVAL);
270 OBD_ALLOC(lcfg, data->ioc_plen1);
272 GOTO(out, err = -ENOMEM);
273 err = cfs_copy_from_user(lcfg, data->ioc_pbuf1,
276 err = lustre_cfg_sanity_check(lcfg, data->ioc_plen1);
278 err = class_process_config(lcfg);
280 OBD_FREE(lcfg, data->ioc_plen1);
284 case OBD_GET_VERSION:
285 if (!data->ioc_inlbuf1) {
286 CERROR("No buffer passed in ioctl\n");
287 GOTO(out, err = -EINVAL);
290 if (strlen(BUILD_VERSION) + 1 > data->ioc_inllen1) {
291 CERROR("ioctl buffer too small to hold version\n");
292 GOTO(out, err = -EINVAL);
295 memcpy(data->ioc_bulk, BUILD_VERSION,
296 strlen(BUILD_VERSION) + 1);
298 err = obd_ioctl_popdata((void *)arg, data, len);
303 case OBD_IOC_NAME2DEV: {
304 /* Resolve a device name. This does not change the
305 * currently selected device.
309 dev = class_resolve_dev_name(data->ioc_inllen1,
313 GOTO(out, err = -EINVAL);
315 err = obd_ioctl_popdata((void *)arg, data, sizeof(*data));
321 case OBD_IOC_UUID2DEV: {
322 /* Resolve a device uuid. This does not change the
323 * currently selected device.
326 struct obd_uuid uuid;
328 if (!data->ioc_inllen1 || !data->ioc_inlbuf1) {
329 CERROR("No UUID passed!\n");
330 GOTO(out, err = -EINVAL);
332 if (data->ioc_inlbuf1[data->ioc_inllen1 - 1] != 0) {
333 CERROR("UUID not NUL terminated!\n");
334 GOTO(out, err = -EINVAL);
337 CDEBUG(D_IOCTL, "device name %s\n", data->ioc_inlbuf1);
338 obd_str2uuid(&uuid, data->ioc_inlbuf1);
339 dev = class_uuid2dev(&uuid);
342 CDEBUG(D_IOCTL, "No device for UUID %s!\n",
344 GOTO(out, err = -EINVAL);
347 CDEBUG(D_IOCTL, "device name %s, dev %d\n", data->ioc_inlbuf1,
349 err = obd_ioctl_popdata((void *)arg, data, sizeof(*data));
355 case OBD_IOC_CLOSE_UUID: {
356 CDEBUG(D_IOCTL, "closing all connections to uuid %s (NOOP)\n",
361 case OBD_IOC_GETDEVICE: {
362 int index = data->ioc_count;
365 if (!data->ioc_inlbuf1) {
366 CERROR("No buffer passed in ioctl\n");
367 GOTO(out, err = -EINVAL);
369 if (data->ioc_inllen1 < 128) {
370 CERROR("ioctl buffer too small to hold version\n");
371 GOTO(out, err = -EINVAL);
374 obd = class_num2obd(index);
376 GOTO(out, err = -ENOENT);
378 if (obd->obd_stopping)
380 else if (obd->obd_set_up)
382 else if (obd->obd_attached)
386 str = (char *)data->ioc_bulk;
387 snprintf(str, len - sizeof(*data), "%3d %s %s %s %s %d",
388 (int)index, status, obd->obd_type->typ_name,
389 obd->obd_name, obd->obd_uuid.uuid,
390 cfs_atomic_read(&obd->obd_refcount));
391 err = obd_ioctl_popdata((void *)arg, data, len);
398 if (data->ioc_dev == OBD_DEV_BY_DEVNAME) {
399 if (data->ioc_inllen4 <= 0 || data->ioc_inlbuf4 == NULL)
400 GOTO(out, err = -EINVAL);
401 if (strnlen(data->ioc_inlbuf4, MAX_OBD_NAME) >= MAX_OBD_NAME)
402 GOTO(out, err = -EINVAL);
403 obd = class_name2obd(data->ioc_inlbuf4);
404 } else if (data->ioc_dev < class_devno_max()) {
405 obd = class_num2obd(data->ioc_dev);
407 CERROR("OBD ioctl: No device\n");
408 GOTO(out, err = -EINVAL);
412 CERROR("OBD ioctl : No Device %d\n", data->ioc_dev);
413 GOTO(out, err = -EINVAL);
415 LASSERT(obd->obd_magic == OBD_DEVICE_MAGIC);
417 if (!obd->obd_set_up || obd->obd_stopping) {
418 CERROR("OBD ioctl: device not setup %d \n", data->ioc_dev);
419 GOTO(out, err = -EINVAL);
423 case OBD_IOC_NO_TRANSNO: {
424 if (!obd->obd_attached) {
425 CERROR("Device %d not attached\n", obd->obd_minor);
426 GOTO(out, err = -ENODEV);
428 CDEBUG(D_HA, "%s: disabling committed-transno notification\n",
430 obd->obd_no_transno = 1;
435 err = obd_iocontrol(cmd, obd->obd_self_export, len, data, NULL);
439 err = obd_ioctl_popdata((void *)arg, data, len);
448 obd_ioctl_freedata(buf, len);
450 } /* class_handle_ioctl */
453 extern cfs_psdev_t obd_psdev;
455 void *obd_psdev = NULL;
458 #define OBD_INIT_CHECK
459 #ifdef OBD_INIT_CHECK
460 int obd_init_checks(void)
462 __u64 u64val, div64val;
466 CDEBUG(D_INFO, "LPU64=%s, LPD64=%s, LPX64=%s\n", LPU64, LPD64, LPX64);
468 CDEBUG(D_INFO, "OBD_OBJECT_EOF = "LPX64"\n", (__u64)OBD_OBJECT_EOF);
470 u64val = OBD_OBJECT_EOF;
471 CDEBUG(D_INFO, "u64val OBD_OBJECT_EOF = "LPX64"\n", u64val);
472 if (u64val != OBD_OBJECT_EOF) {
473 CERROR("__u64 "LPX64"(%d) != 0xffffffffffffffff\n",
474 u64val, (int)sizeof(u64val));
477 len = snprintf(buf, sizeof(buf), LPX64, u64val);
479 CWARN("LPX64 wrong length! strlen(%s)=%d != 18\n", buf, len);
483 div64val = OBD_OBJECT_EOF;
484 CDEBUG(D_INFO, "u64val OBD_OBJECT_EOF = "LPX64"\n", u64val);
485 if (u64val != OBD_OBJECT_EOF) {
486 CERROR("__u64 "LPX64"(%d) != 0xffffffffffffffff\n",
487 u64val, (int)sizeof(u64val));
490 if (u64val >> 8 != OBD_OBJECT_EOF >> 8) {
491 CERROR("__u64 "LPX64"(%d) != 0xffffffffffffffff\n",
492 u64val, (int)sizeof(u64val));
495 if (do_div(div64val, 256) != (u64val & 255)) {
496 CERROR("do_div("LPX64",256) != "LPU64"\n", u64val, u64val &255);
499 if (u64val >> 8 != div64val) {
500 CERROR("do_div("LPX64",256) "LPU64" != "LPU64"\n",
501 u64val, div64val, u64val >> 8);
504 len = snprintf(buf, sizeof(buf), LPX64, u64val);
506 CWARN("LPX64 wrong length! strlen(%s)=%d != 18\n", buf, len);
509 len = snprintf(buf, sizeof(buf), LPU64, u64val);
511 CWARN("LPU64 wrong length! strlen(%s)=%d != 20\n", buf, len);
514 len = snprintf(buf, sizeof(buf), LPD64, u64val);
516 CWARN("LPD64 wrong length! strlen(%s)=%d != 2\n", buf, len);
519 if ((u64val & ~CFS_PAGE_MASK) >= CFS_PAGE_SIZE) {
520 CWARN("mask failed: u64val "LPU64" >= "LPU64"\n", u64val,
521 (__u64)CFS_PAGE_SIZE);
528 #define obd_init_checks() do {} while(0)
531 extern spinlock_t obd_types_lock;
532 extern int class_procfs_init(void);
533 extern int class_procfs_clean(void);
536 static int __init init_obdclass(void)
538 int init_obdclass(void)
543 int lustre_register_fs(void);
545 for (i = CAPA_SITE_CLIENT; i < CAPA_SITE_MAX; i++)
546 CFS_INIT_LIST_HEAD(&capa_list[i]);
549 LCONSOLE_INFO("Lustre: Build Version: "BUILD_VERSION"\n");
551 spin_lock_init(&obd_types_lock);
552 obd_zombie_impexp_init();
554 obd_memory = lprocfs_alloc_stats(OBD_STATS_NUM,
555 LPROCFS_STATS_FLAG_NONE |
556 LPROCFS_STATS_FLAG_IRQ_SAFE);
557 if (obd_memory == NULL) {
558 CERROR("kmalloc of 'obd_memory' failed\n");
562 lprocfs_counter_init(obd_memory, OBD_MEMORY_STAT,
563 LPROCFS_CNTR_AVGMINMAX,
565 lprocfs_counter_init(obd_memory, OBD_MEMORY_PAGES_STAT,
566 LPROCFS_CNTR_AVGMINMAX,
567 "pagesused", "pages");
569 err = obd_init_checks();
570 if (err == -EOVERFLOW)
573 class_init_uuidlist();
574 err = class_handle_init();
578 CFS_INIT_LIST_HEAD(&obd_types);
580 err = cfs_psdev_register(&obd_psdev);
582 CERROR("cannot register %d err %d\n", OBD_DEV_MINOR, err);
586 /* This struct is already zeroed for us (static global) */
587 for (i = 0; i < class_devno_max(); i++)
590 /* Default the dirty page cache cap to 1/2 of system memory.
591 * For clients with less memory, a larger fraction is needed
592 * for other purposes (mostly for BGL). */
593 if (cfs_num_physpages <= 512 << (20 - CFS_PAGE_SHIFT))
594 obd_max_dirty_pages = cfs_num_physpages / 4;
596 obd_max_dirty_pages = cfs_num_physpages / 2;
598 err = obd_init_caches();
602 err = class_procfs_init();
607 err = lu_global_init();
611 err = cl_global_init();
615 #if defined(__KERNEL__) && defined(HAVE_SERVER_SUPPORT)
616 err = dt_global_init();
620 err = lu_ucred_global_init();
624 err = llo_global_init();
629 err = llog_info_init();
634 err = lustre_register_fs();
640 void obd_update_maxusage(void)
644 max1 = obd_pages_sum();
645 max2 = obd_memory_sum();
647 spin_lock(&obd_updatemax_lock);
648 if (max1 > obd_max_pages)
649 obd_max_pages = max1;
650 if (max2 > obd_max_alloc)
651 obd_max_alloc = max2;
652 spin_unlock(&obd_updatemax_lock);
654 EXPORT_SYMBOL(obd_update_maxusage);
657 __u64 obd_memory_max(void)
661 spin_lock(&obd_updatemax_lock);
663 spin_unlock(&obd_updatemax_lock);
667 EXPORT_SYMBOL(obd_memory_max);
669 __u64 obd_pages_max(void)
673 spin_lock(&obd_updatemax_lock);
675 spin_unlock(&obd_updatemax_lock);
679 EXPORT_SYMBOL(obd_pages_max);
682 /* liblustre doesn't call cleanup_obdclass, apparently. we carry on in this
683 * ifdef to the end of the file to cover module and versioning goo.*/
685 static void cleanup_obdclass(void)
688 int lustre_unregister_fs(void);
689 __u64 memory_leaked, pages_leaked;
690 __u64 memory_max, pages_max;
693 lustre_unregister_fs();
695 cfs_psdev_deregister(&obd_psdev);
696 for (i = 0; i < class_devno_max(); i++) {
697 struct obd_device *obd = class_num2obd(i);
698 if (obd && obd->obd_set_up &&
699 OBT(obd) && OBP(obd, detach)) {
700 /* XXX should this call generic detach otherwise? */
701 LASSERT(obd->obd_magic == OBD_DEVICE_MAGIC);
702 OBP(obd, detach)(obd);
706 #ifdef HAVE_SERVER_SUPPORT
708 lu_ucred_global_fini();
714 obd_cleanup_caches();
717 class_procfs_clean();
719 class_handle_cleanup();
720 class_exit_uuidlist();
721 obd_zombie_impexp_stop();
723 memory_leaked = obd_memory_sum();
724 pages_leaked = obd_pages_sum();
726 memory_max = obd_memory_max();
727 pages_max = obd_pages_max();
729 lprocfs_free_stats(&obd_memory);
730 CDEBUG((memory_leaked) ? D_ERROR : D_INFO,
731 "obd_memory max: "LPU64", leaked: "LPU64"\n",
732 memory_max, memory_leaked);
733 CDEBUG((pages_leaked) ? D_ERROR : D_INFO,
734 "obd_memory_pages max: "LPU64", leaked: "LPU64"\n",
735 pages_max, pages_leaked);
740 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
741 MODULE_DESCRIPTION("Lustre Class Driver Build Version: " BUILD_VERSION);
742 MODULE_LICENSE("GPL");
744 cfs_module(obdclass, LUSTRE_VERSION_STRING, init_obdclass, cleanup_obdclass);