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 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 atomic_t obd_unstable_pages;
88 EXPORT_SYMBOL(obd_unstable_pages);
89 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 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);
118 struct lprocfs_stats *obd_memory = NULL;
119 EXPORT_SYMBOL(obd_memory);
122 /* Get jobid of current process by reading the environment variable
123 * stored in between the "env_start" & "env_end" of task struct.
126 * It's better to cache the jobid for later use if there is any
127 * efficient way, the cl_env code probably could be reused for this
130 * If some job scheduler doesn't store jobid in the "env_start/end",
131 * then an upcall could be issued here to get the jobid by utilizing
132 * the userspace tools/api. Then, the jobid must be cached.
134 int lustre_get_jobid(char *jobid)
136 int jobid_len = JOBSTATS_JOBID_SIZE;
140 memset(jobid, 0, JOBSTATS_JOBID_SIZE);
141 /* Jobstats isn't enabled */
142 if (strcmp(obd_jobid_var, JOBSTATS_DISABLE) == 0)
145 /* Use process name + fsuid as jobid */
146 if (strcmp(obd_jobid_var, JOBSTATS_PROCNAME_UID) == 0) {
147 snprintf(jobid, JOBSTATS_JOBID_SIZE, "%s.%u",
149 from_kuid(&init_user_ns, current_fsuid()));
153 rc = cfs_get_environ(obd_jobid_var, jobid, &jobid_len);
155 if (rc == -EOVERFLOW) {
156 /* For the PBS_JOBID and LOADL_STEP_ID keys (which are
157 * variable length strings instead of just numbers), it
158 * might make sense to keep the unique parts for JobID,
159 * instead of just returning an error. That means a
160 * larger temp buffer for cfs_get_environ(), then
161 * truncating the string at some separator to fit into
162 * the specified jobid_len. Fix later if needed. */
164 if (unlikely(!printed)) {
165 LCONSOLE_ERROR_MSG(0x16b, "%s value too large "
166 "for JobID buffer (%d)\n",
167 obd_jobid_var, jobid_len);
171 CDEBUG((rc == -ENOENT || rc == -EINVAL ||
172 rc == -EDEADLK) ? D_INFO : D_ERROR,
173 "Get jobid for (%s) failed: rc = %d\n",
179 EXPORT_SYMBOL(lustre_get_jobid);
181 int obd_alloc_fail(const void *ptr, const char *name, const char *type,
182 size_t size, const char *file, int line)
185 (cfs_rand() & OBD_ALLOC_FAIL_MASK) < obd_alloc_fail_rate) {
186 CERROR("%s%salloc of %s ("LPU64" bytes) failed at %s:%d\n",
187 ptr ? "force " :"", type, name, (__u64)size, file,
189 CERROR(LPU64" total bytes and "LPU64" total pages "
190 "("LPU64" bytes) allocated by Lustre, "
191 "%d total bytes by LNET\n",
193 obd_pages_sum() << PAGE_CACHE_SHIFT,
195 atomic_read(&libcfs_kmemory));
200 EXPORT_SYMBOL(obd_alloc_fail);
202 static inline void obd_data2conn(struct lustre_handle *conn,
203 struct obd_ioctl_data *data)
205 memset(conn, 0, sizeof *conn);
206 conn->cookie = data->ioc_cookie;
209 static inline void obd_conn2data(struct obd_ioctl_data *data,
210 struct lustre_handle *conn)
212 data->ioc_cookie = conn->cookie;
215 int class_resolve_dev_name(__u32 len, const char *name)
222 CERROR("No name passed,!\n");
223 GOTO(out, rc = -EINVAL);
225 if (name[len - 1] != 0) {
226 CERROR("Name not nul terminated!\n");
227 GOTO(out, rc = -EINVAL);
230 CDEBUG(D_IOCTL, "device name %s\n", name);
231 dev = class_name2dev(name);
233 CDEBUG(D_IOCTL, "No device for name %s!\n", name);
234 GOTO(out, rc = -EINVAL);
237 CDEBUG(D_IOCTL, "device name %s, dev %d\n", name, dev);
244 int class_handle_ioctl(unsigned int cmd, unsigned long arg)
247 struct obd_ioctl_data *data;
248 struct libcfs_debug_ioctl_data *debug_data;
249 struct obd_device *obd = NULL;
250 int err = 0, len = 0;
253 /* only for debugging */
254 if (cmd == LIBCFS_IOC_DEBUG_MASK) {
255 debug_data = (struct libcfs_debug_ioctl_data*)arg;
256 libcfs_subsystem_debug = debug_data->subs;
257 libcfs_debug = debug_data->debug;
261 CDEBUG(D_IOCTL, "cmd = %x\n", cmd);
262 if (obd_ioctl_getdata(&buf, &len, (void *)arg)) {
263 CERROR("OBD ioctl: data error\n");
266 data = (struct obd_ioctl_data *)buf;
269 case OBD_IOC_PROCESS_CFG: {
270 struct lustre_cfg *lcfg;
272 if (!data->ioc_plen1 || !data->ioc_pbuf1) {
273 CERROR("No config buffer passed!\n");
274 GOTO(out, err = -EINVAL);
276 OBD_ALLOC(lcfg, data->ioc_plen1);
278 GOTO(out, err = -ENOMEM);
279 err = copy_from_user(lcfg, data->ioc_pbuf1,
282 err = lustre_cfg_sanity_check(lcfg, data->ioc_plen1);
284 err = class_process_config(lcfg);
286 OBD_FREE(lcfg, data->ioc_plen1);
290 case OBD_GET_VERSION:
291 if (!data->ioc_inlbuf1) {
292 CERROR("No buffer passed in ioctl\n");
293 GOTO(out, err = -EINVAL);
296 if (strlen(BUILD_VERSION) + 1 > data->ioc_inllen1) {
297 CERROR("ioctl buffer too small to hold version\n");
298 GOTO(out, err = -EINVAL);
301 memcpy(data->ioc_bulk, BUILD_VERSION,
302 strlen(BUILD_VERSION) + 1);
304 err = obd_ioctl_popdata((void *)arg, data, len);
309 case OBD_IOC_NAME2DEV: {
310 /* Resolve a device name. This does not change the
311 * currently selected device.
315 dev = class_resolve_dev_name(data->ioc_inllen1,
319 GOTO(out, err = -EINVAL);
321 err = obd_ioctl_popdata((void *)arg, data, sizeof(*data));
327 case OBD_IOC_UUID2DEV: {
328 /* Resolve a device uuid. This does not change the
329 * currently selected device.
332 struct obd_uuid uuid;
334 if (!data->ioc_inllen1 || !data->ioc_inlbuf1) {
335 CERROR("No UUID passed!\n");
336 GOTO(out, err = -EINVAL);
338 if (data->ioc_inlbuf1[data->ioc_inllen1 - 1] != 0) {
339 CERROR("UUID not NUL terminated!\n");
340 GOTO(out, err = -EINVAL);
343 CDEBUG(D_IOCTL, "device name %s\n", data->ioc_inlbuf1);
344 obd_str2uuid(&uuid, data->ioc_inlbuf1);
345 dev = class_uuid2dev(&uuid);
348 CDEBUG(D_IOCTL, "No device for UUID %s!\n",
350 GOTO(out, err = -EINVAL);
353 CDEBUG(D_IOCTL, "device name %s, dev %d\n", data->ioc_inlbuf1,
355 err = obd_ioctl_popdata((void *)arg, data, sizeof(*data));
361 case OBD_IOC_CLOSE_UUID: {
362 CDEBUG(D_IOCTL, "closing all connections to uuid %s (NOOP)\n",
367 case OBD_IOC_GETDEVICE: {
368 int index = data->ioc_count;
371 if (!data->ioc_inlbuf1) {
372 CERROR("No buffer passed in ioctl\n");
373 GOTO(out, err = -EINVAL);
375 if (data->ioc_inllen1 < 128) {
376 CERROR("ioctl buffer too small to hold version\n");
377 GOTO(out, err = -EINVAL);
380 obd = class_num2obd(index);
382 GOTO(out, err = -ENOENT);
384 if (obd->obd_stopping)
386 else if (obd->obd_set_up)
388 else if (obd->obd_attached)
392 str = (char *)data->ioc_bulk;
393 snprintf(str, len - sizeof(*data), "%3d %s %s %s %s %d",
394 (int)index, status, obd->obd_type->typ_name,
395 obd->obd_name, obd->obd_uuid.uuid,
396 atomic_read(&obd->obd_refcount));
397 err = obd_ioctl_popdata((void *)arg, data, len);
404 if (data->ioc_dev == OBD_DEV_BY_DEVNAME) {
405 if (data->ioc_inllen4 <= 0 || data->ioc_inlbuf4 == NULL)
406 GOTO(out, err = -EINVAL);
407 if (strnlen(data->ioc_inlbuf4, MAX_OBD_NAME) >= MAX_OBD_NAME)
408 GOTO(out, err = -EINVAL);
409 obd = class_name2obd(data->ioc_inlbuf4);
410 } else if (data->ioc_dev < class_devno_max()) {
411 obd = class_num2obd(data->ioc_dev);
413 CERROR("OBD ioctl: No device\n");
414 GOTO(out, err = -EINVAL);
418 CERROR("OBD ioctl : No Device %d\n", data->ioc_dev);
419 GOTO(out, err = -EINVAL);
421 LASSERT(obd->obd_magic == OBD_DEVICE_MAGIC);
423 if (!obd->obd_set_up || obd->obd_stopping) {
424 CERROR("OBD ioctl: device not setup %d \n", data->ioc_dev);
425 GOTO(out, err = -EINVAL);
429 case OBD_IOC_NO_TRANSNO: {
430 if (!obd->obd_attached) {
431 CERROR("Device %d not attached\n", obd->obd_minor);
432 GOTO(out, err = -ENODEV);
434 CDEBUG(D_HA, "%s: disabling committed-transno notification\n",
436 obd->obd_no_transno = 1;
441 err = obd_iocontrol(cmd, obd->obd_self_export, len, data, NULL);
445 err = obd_ioctl_popdata((void *)arg, data, len);
454 obd_ioctl_freedata(buf, len);
456 } /* class_handle_ioctl */
459 extern struct miscdevice obd_psdev;
461 struct miscdevice obd_psdev;
464 #define OBD_INIT_CHECK
465 #ifdef OBD_INIT_CHECK
466 int obd_init_checks(void)
468 __u64 u64val, div64val;
472 CDEBUG(D_INFO, "LPU64=%s, LPD64=%s, LPX64=%s\n", LPU64, LPD64, LPX64);
474 CDEBUG(D_INFO, "OBD_OBJECT_EOF = "LPX64"\n", (__u64)OBD_OBJECT_EOF);
476 u64val = OBD_OBJECT_EOF;
477 CDEBUG(D_INFO, "u64val OBD_OBJECT_EOF = "LPX64"\n", u64val);
478 if (u64val != OBD_OBJECT_EOF) {
479 CERROR("__u64 "LPX64"(%d) != 0xffffffffffffffff\n",
480 u64val, (int)sizeof(u64val));
483 len = snprintf(buf, sizeof(buf), LPX64, u64val);
485 CWARN("LPX64 wrong length! strlen(%s)=%d != 18\n", buf, len);
489 div64val = OBD_OBJECT_EOF;
490 CDEBUG(D_INFO, "u64val OBD_OBJECT_EOF = "LPX64"\n", u64val);
491 if (u64val != OBD_OBJECT_EOF) {
492 CERROR("__u64 "LPX64"(%d) != 0xffffffffffffffff\n",
493 u64val, (int)sizeof(u64val));
496 if (u64val >> 8 != OBD_OBJECT_EOF >> 8) {
497 CERROR("__u64 "LPX64"(%d) != 0xffffffffffffffff\n",
498 u64val, (int)sizeof(u64val));
501 if (do_div(div64val, 256) != (u64val & 255)) {
502 CERROR("do_div("LPX64",256) != "LPU64"\n", u64val, u64val &255);
505 if (u64val >> 8 != div64val) {
506 CERROR("do_div("LPX64",256) "LPU64" != "LPU64"\n",
507 u64val, div64val, u64val >> 8);
510 len = snprintf(buf, sizeof(buf), LPX64, u64val);
512 CWARN("LPX64 wrong length! strlen(%s)=%d != 18\n", buf, len);
515 len = snprintf(buf, sizeof(buf), LPU64, u64val);
517 CWARN("LPU64 wrong length! strlen(%s)=%d != 20\n", buf, len);
520 len = snprintf(buf, sizeof(buf), LPD64, u64val);
522 CWARN("LPD64 wrong length! strlen(%s)=%d != 2\n", buf, len);
525 if ((u64val & ~CFS_PAGE_MASK) >= PAGE_CACHE_SIZE) {
526 CWARN("mask failed: u64val "LPU64" >= "LPU64"\n", u64val,
527 (__u64)PAGE_CACHE_SIZE);
534 #define obd_init_checks() do {} while(0)
537 extern spinlock_t obd_types_lock;
538 extern int class_procfs_init(void);
539 extern int class_procfs_clean(void);
542 static int __init init_obdclass(void)
544 int init_obdclass(void)
549 int lustre_register_fs(void);
551 for (i = CAPA_SITE_CLIENT; i < CAPA_SITE_MAX; i++)
552 CFS_INIT_LIST_HEAD(&capa_list[i]);
555 LCONSOLE_INFO("Lustre: Build Version: "BUILD_VERSION"\n");
557 spin_lock_init(&obd_types_lock);
558 obd_zombie_impexp_init();
560 obd_memory = lprocfs_alloc_stats(OBD_STATS_NUM,
561 LPROCFS_STATS_FLAG_NONE |
562 LPROCFS_STATS_FLAG_IRQ_SAFE);
563 if (obd_memory == NULL) {
564 CERROR("kmalloc of 'obd_memory' failed\n");
568 lprocfs_counter_init(obd_memory, OBD_MEMORY_STAT,
569 LPROCFS_CNTR_AVGMINMAX,
571 lprocfs_counter_init(obd_memory, OBD_MEMORY_PAGES_STAT,
572 LPROCFS_CNTR_AVGMINMAX,
573 "pagesused", "pages");
575 err = obd_init_checks();
576 if (err == -EOVERFLOW)
579 class_init_uuidlist();
580 err = class_handle_init();
584 CFS_INIT_LIST_HEAD(&obd_types);
586 err = misc_register(&obd_psdev);
588 CERROR("cannot register %d err %d\n", OBD_DEV_MINOR, err);
592 /* This struct is already zeroed for us (static global) */
593 for (i = 0; i < class_devno_max(); i++)
596 /* Default the dirty page cache cap to 1/2 of system memory.
597 * For clients with less memory, a larger fraction is needed
598 * for other purposes (mostly for BGL). */
599 if (totalram_pages <= 512 << (20 - PAGE_CACHE_SHIFT))
600 obd_max_dirty_pages = totalram_pages / 4;
602 obd_max_dirty_pages = totalram_pages / 2;
604 err = obd_init_caches();
608 err = class_procfs_init();
613 err = lu_global_init();
617 err = lu_capainfo_init();
621 err = cl_global_init();
625 #if defined(__KERNEL__) && defined(HAVE_SERVER_SUPPORT)
626 err = dt_global_init();
630 err = lu_ucred_global_init();
635 err = llog_info_init();
640 err = lustre_register_fs();
646 void obd_update_maxusage(void)
650 max1 = obd_pages_sum();
651 max2 = obd_memory_sum();
653 spin_lock(&obd_updatemax_lock);
654 if (max1 > obd_max_pages)
655 obd_max_pages = max1;
656 if (max2 > obd_max_alloc)
657 obd_max_alloc = max2;
658 spin_unlock(&obd_updatemax_lock);
660 EXPORT_SYMBOL(obd_update_maxusage);
663 __u64 obd_memory_max(void)
667 spin_lock(&obd_updatemax_lock);
669 spin_unlock(&obd_updatemax_lock);
673 EXPORT_SYMBOL(obd_memory_max);
675 __u64 obd_pages_max(void)
679 spin_lock(&obd_updatemax_lock);
681 spin_unlock(&obd_updatemax_lock);
685 EXPORT_SYMBOL(obd_pages_max);
688 /* liblustre doesn't call cleanup_obdclass, apparently. we carry on in this
689 * ifdef to the end of the file to cover module and versioning goo.*/
691 static void cleanup_obdclass(void)
694 int lustre_unregister_fs(void);
695 __u64 memory_leaked, pages_leaked;
696 __u64 memory_max, pages_max;
699 lustre_unregister_fs();
701 misc_deregister(&obd_psdev);
702 for (i = 0; i < class_devno_max(); i++) {
703 struct obd_device *obd = class_num2obd(i);
704 if (obd && obd->obd_set_up &&
705 OBT(obd) && OBP(obd, detach)) {
706 /* XXX should this call generic detach otherwise? */
707 LASSERT(obd->obd_magic == OBD_DEVICE_MAGIC);
708 OBP(obd, detach)(obd);
712 #ifdef HAVE_SERVER_SUPPORT
713 lu_ucred_global_fini();
720 obd_cleanup_caches();
723 class_procfs_clean();
725 class_handle_cleanup();
726 class_exit_uuidlist();
727 obd_zombie_impexp_stop();
729 memory_leaked = obd_memory_sum();
730 pages_leaked = obd_pages_sum();
732 memory_max = obd_memory_max();
733 pages_max = obd_pages_max();
735 lprocfs_free_stats(&obd_memory);
736 CDEBUG((memory_leaked) ? D_ERROR : D_INFO,
737 "obd_memory max: "LPU64", leaked: "LPU64"\n",
738 memory_max, memory_leaked);
739 CDEBUG((pages_leaked) ? D_ERROR : D_INFO,
740 "obd_memory_pages max: "LPU64", leaked: "LPU64"\n",
741 pages_max, pages_leaked);
746 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
747 MODULE_DESCRIPTION("Lustre Class Driver Build Version: " BUILD_VERSION);
748 MODULE_LICENSE("GPL");
750 cfs_module(obdclass, LUSTRE_VERSION_STRING, init_obdclass, cleanup_obdclass);