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 "llog_internal.h"
54 /* liblustre workaround */
55 cfs_atomic_t libcfs_kmemory = {0};
58 struct obd_device *obd_devs[MAX_OBD_DEVICES];
59 EXPORT_SYMBOL(obd_devs);
61 DEFINE_RWLOCK(obd_dev_lock);
63 __u64 obd_max_pages = 0;
64 __u64 obd_max_alloc = 0;
69 DEFINE_SPINLOCK(obd_updatemax_lock);
71 /* The following are visible and mutable through /proc/sys/lustre/. */
72 unsigned int obd_alloc_fail_rate = 0;
73 EXPORT_SYMBOL(obd_alloc_fail_rate);
74 unsigned int obd_debug_peer_on_timeout;
75 EXPORT_SYMBOL(obd_debug_peer_on_timeout);
76 unsigned int obd_dump_on_timeout;
77 EXPORT_SYMBOL(obd_dump_on_timeout);
78 unsigned int obd_dump_on_eviction;
79 EXPORT_SYMBOL(obd_dump_on_eviction);
80 unsigned int obd_max_dirty_pages = 256;
81 EXPORT_SYMBOL(obd_max_dirty_pages);
82 cfs_atomic_t obd_dirty_pages;
83 EXPORT_SYMBOL(obd_dirty_pages);
84 unsigned int obd_timeout = OBD_TIMEOUT_DEFAULT; /* seconds */
85 EXPORT_SYMBOL(obd_timeout);
86 unsigned int ldlm_timeout = LDLM_TIMEOUT_DEFAULT; /* seconds */
87 EXPORT_SYMBOL(ldlm_timeout);
88 unsigned int obd_timeout_set;
89 EXPORT_SYMBOL(obd_timeout_set);
90 unsigned int ldlm_timeout_set;
91 EXPORT_SYMBOL(ldlm_timeout_set);
92 /* Adaptive timeout defs here instead of ptlrpc module for /proc/sys/ access */
93 unsigned int at_min = 0;
94 EXPORT_SYMBOL(at_min);
95 unsigned int at_max = 600;
96 EXPORT_SYMBOL(at_max);
97 unsigned int at_history = 600;
98 EXPORT_SYMBOL(at_history);
99 int at_early_margin = 5;
100 EXPORT_SYMBOL(at_early_margin);
102 EXPORT_SYMBOL(at_extra);
104 cfs_atomic_t obd_dirty_transit_pages;
105 EXPORT_SYMBOL(obd_dirty_transit_pages);
107 char obd_jobid_var[JOBSTATS_JOBID_VAR_MAX_LEN + 1] = JOBSTATS_DISABLE;
108 EXPORT_SYMBOL(obd_jobid_var);
110 /* Get jobid of current process by reading the environment variable
111 * stored in between the "env_start" & "env_end" of task struct.
114 * It's better to cache the jobid for later use if there is any
115 * efficient way, the cl_env code probably could be reused for this
118 * If some job scheduler doesn't store jobid in the "env_start/end",
119 * then an upcall could be issued here to get the jobid by utilizing
120 * the userspace tools/api. Then, the jobid must be cached.
122 int lustre_get_jobid(char *jobid)
124 int jobid_len = JOBSTATS_JOBID_SIZE;
128 memset(jobid, 0, JOBSTATS_JOBID_SIZE);
129 /* Jobstats isn't enabled */
130 if (strcmp(obd_jobid_var, JOBSTATS_DISABLE) == 0)
133 /* Use process name + fsuid as jobid */
134 if (strcmp(obd_jobid_var, JOBSTATS_PROCNAME_UID) == 0) {
135 snprintf(jobid, JOBSTATS_JOBID_SIZE, "%s.%u",
136 cfs_curproc_comm(), cfs_curproc_fsuid());
140 rc = cfs_get_environ(obd_jobid_var, jobid, &jobid_len);
142 if (rc == -EOVERFLOW) {
143 /* For the PBS_JOBID and LOADL_STEP_ID keys (which are
144 * variable length strings instead of just numbers), it
145 * might make sense to keep the unique parts for JobID,
146 * instead of just returning an error. That means a
147 * larger temp buffer for cfs_get_environ(), then
148 * truncating the string at some separator to fit into
149 * the specified jobid_len. Fix later if needed. */
151 if (unlikely(!printed)) {
152 LCONSOLE_ERROR_MSG(0x16b, "%s value too large "
153 "for JobID buffer (%d)\n",
154 obd_jobid_var, jobid_len);
158 CDEBUG((rc == -ENOENT || rc == -EINVAL ||
159 rc == -EDEADLK) ? D_INFO : D_ERROR,
160 "Get jobid for (%s) failed: rc = %d\n",
166 EXPORT_SYMBOL(lustre_get_jobid);
168 int obd_alloc_fail(const void *ptr, const char *name, const char *type,
169 size_t size, const char *file, int line)
172 (cfs_rand() & OBD_ALLOC_FAIL_MASK) < obd_alloc_fail_rate) {
173 CERROR("%s%salloc of %s ("LPU64" bytes) failed at %s:%d\n",
174 ptr ? "force " :"", type, name, (__u64)size, file,
176 CERROR(LPU64" total bytes and "LPU64" total pages "
177 "("LPU64" bytes) allocated by Lustre, "
178 "%d total bytes by LNET\n",
180 obd_pages_sum() << CFS_PAGE_SHIFT,
182 cfs_atomic_read(&libcfs_kmemory));
187 EXPORT_SYMBOL(obd_alloc_fail);
189 static inline void obd_data2conn(struct lustre_handle *conn,
190 struct obd_ioctl_data *data)
192 memset(conn, 0, sizeof *conn);
193 conn->cookie = data->ioc_cookie;
196 static inline void obd_conn2data(struct obd_ioctl_data *data,
197 struct lustre_handle *conn)
199 data->ioc_cookie = conn->cookie;
202 int class_resolve_dev_name(__u32 len, const char *name)
209 CERROR("No name passed,!\n");
210 GOTO(out, rc = -EINVAL);
212 if (name[len - 1] != 0) {
213 CERROR("Name not nul terminated!\n");
214 GOTO(out, rc = -EINVAL);
217 CDEBUG(D_IOCTL, "device name %s\n", name);
218 dev = class_name2dev(name);
220 CDEBUG(D_IOCTL, "No device for name %s!\n", name);
221 GOTO(out, rc = -EINVAL);
224 CDEBUG(D_IOCTL, "device name %s, dev %d\n", name, dev);
231 int class_handle_ioctl(unsigned int cmd, unsigned long arg)
234 struct obd_ioctl_data *data;
235 struct libcfs_debug_ioctl_data *debug_data;
236 struct obd_device *obd = NULL;
237 int err = 0, len = 0;
240 /* only for debugging */
241 if (cmd == LIBCFS_IOC_DEBUG_MASK) {
242 debug_data = (struct libcfs_debug_ioctl_data*)arg;
243 libcfs_subsystem_debug = debug_data->subs;
244 libcfs_debug = debug_data->debug;
248 CDEBUG(D_IOCTL, "cmd = %x\n", cmd);
249 if (obd_ioctl_getdata(&buf, &len, (void *)arg)) {
250 CERROR("OBD ioctl: data error\n");
253 data = (struct obd_ioctl_data *)buf;
256 case OBD_IOC_PROCESS_CFG: {
257 struct lustre_cfg *lcfg;
259 if (!data->ioc_plen1 || !data->ioc_pbuf1) {
260 CERROR("No config buffer passed!\n");
261 GOTO(out, err = -EINVAL);
263 OBD_ALLOC(lcfg, data->ioc_plen1);
265 GOTO(out, err = -ENOMEM);
266 err = cfs_copy_from_user(lcfg, data->ioc_pbuf1,
269 err = lustre_cfg_sanity_check(lcfg, data->ioc_plen1);
271 err = class_process_config(lcfg);
273 OBD_FREE(lcfg, data->ioc_plen1);
277 case OBD_GET_VERSION:
278 if (!data->ioc_inlbuf1) {
279 CERROR("No buffer passed in ioctl\n");
280 GOTO(out, err = -EINVAL);
283 if (strlen(BUILD_VERSION) + 1 > data->ioc_inllen1) {
284 CERROR("ioctl buffer too small to hold version\n");
285 GOTO(out, err = -EINVAL);
288 memcpy(data->ioc_bulk, BUILD_VERSION,
289 strlen(BUILD_VERSION) + 1);
291 err = obd_ioctl_popdata((void *)arg, data, len);
296 case OBD_IOC_NAME2DEV: {
297 /* Resolve a device name. This does not change the
298 * currently selected device.
302 dev = class_resolve_dev_name(data->ioc_inllen1,
306 GOTO(out, err = -EINVAL);
308 err = obd_ioctl_popdata((void *)arg, data, sizeof(*data));
314 case OBD_IOC_UUID2DEV: {
315 /* Resolve a device uuid. This does not change the
316 * currently selected device.
319 struct obd_uuid uuid;
321 if (!data->ioc_inllen1 || !data->ioc_inlbuf1) {
322 CERROR("No UUID passed!\n");
323 GOTO(out, err = -EINVAL);
325 if (data->ioc_inlbuf1[data->ioc_inllen1 - 1] != 0) {
326 CERROR("UUID not NUL terminated!\n");
327 GOTO(out, err = -EINVAL);
330 CDEBUG(D_IOCTL, "device name %s\n", data->ioc_inlbuf1);
331 obd_str2uuid(&uuid, data->ioc_inlbuf1);
332 dev = class_uuid2dev(&uuid);
335 CDEBUG(D_IOCTL, "No device for UUID %s!\n",
337 GOTO(out, err = -EINVAL);
340 CDEBUG(D_IOCTL, "device name %s, dev %d\n", data->ioc_inlbuf1,
342 err = obd_ioctl_popdata((void *)arg, data, sizeof(*data));
348 case OBD_IOC_CLOSE_UUID: {
349 CDEBUG(D_IOCTL, "closing all connections to uuid %s (NOOP)\n",
354 case OBD_IOC_GETDEVICE: {
355 int index = data->ioc_count;
358 if (!data->ioc_inlbuf1) {
359 CERROR("No buffer passed in ioctl\n");
360 GOTO(out, err = -EINVAL);
362 if (data->ioc_inllen1 < 128) {
363 CERROR("ioctl buffer too small to hold version\n");
364 GOTO(out, err = -EINVAL);
367 obd = class_num2obd(index);
369 GOTO(out, err = -ENOENT);
371 if (obd->obd_stopping)
373 else if (obd->obd_set_up)
375 else if (obd->obd_attached)
379 str = (char *)data->ioc_bulk;
380 snprintf(str, len - sizeof(*data), "%3d %s %s %s %s %d",
381 (int)index, status, obd->obd_type->typ_name,
382 obd->obd_name, obd->obd_uuid.uuid,
383 cfs_atomic_read(&obd->obd_refcount));
384 err = obd_ioctl_popdata((void *)arg, data, len);
391 if (data->ioc_dev == OBD_DEV_BY_DEVNAME) {
392 if (data->ioc_inllen4 <= 0 || data->ioc_inlbuf4 == NULL)
393 GOTO(out, err = -EINVAL);
394 if (strnlen(data->ioc_inlbuf4, MAX_OBD_NAME) >= MAX_OBD_NAME)
395 GOTO(out, err = -EINVAL);
396 obd = class_name2obd(data->ioc_inlbuf4);
397 } else if (data->ioc_dev < class_devno_max()) {
398 obd = class_num2obd(data->ioc_dev);
400 CERROR("OBD ioctl: No device\n");
401 GOTO(out, err = -EINVAL);
405 CERROR("OBD ioctl : No Device %d\n", data->ioc_dev);
406 GOTO(out, err = -EINVAL);
408 LASSERT(obd->obd_magic == OBD_DEVICE_MAGIC);
410 if (!obd->obd_set_up || obd->obd_stopping) {
411 CERROR("OBD ioctl: device not setup %d \n", data->ioc_dev);
412 GOTO(out, err = -EINVAL);
416 case OBD_IOC_NO_TRANSNO: {
417 if (!obd->obd_attached) {
418 CERROR("Device %d not attached\n", obd->obd_minor);
419 GOTO(out, err = -ENODEV);
421 CDEBUG(D_HA, "%s: disabling committed-transno notification\n",
423 obd->obd_no_transno = 1;
428 err = obd_iocontrol(cmd, obd->obd_self_export, len, data, NULL);
432 err = obd_ioctl_popdata((void *)arg, data, len);
441 obd_ioctl_freedata(buf, len);
443 } /* class_handle_ioctl */
446 extern cfs_psdev_t obd_psdev;
448 void *obd_psdev = NULL;
451 #define OBD_INIT_CHECK
452 #ifdef OBD_INIT_CHECK
453 int obd_init_checks(void)
455 __u64 u64val, div64val;
459 CDEBUG(D_INFO, "LPU64=%s, LPD64=%s, LPX64=%s\n", LPU64, LPD64, LPX64);
461 CDEBUG(D_INFO, "OBD_OBJECT_EOF = "LPX64"\n", (__u64)OBD_OBJECT_EOF);
463 u64val = OBD_OBJECT_EOF;
464 CDEBUG(D_INFO, "u64val OBD_OBJECT_EOF = "LPX64"\n", u64val);
465 if (u64val != OBD_OBJECT_EOF) {
466 CERROR("__u64 "LPX64"(%d) != 0xffffffffffffffff\n",
467 u64val, (int)sizeof(u64val));
470 len = snprintf(buf, sizeof(buf), LPX64, u64val);
472 CWARN("LPX64 wrong length! strlen(%s)=%d != 18\n", buf, len);
476 div64val = 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 if (u64val >> 8 != OBD_OBJECT_EOF >> 8) {
484 CERROR("__u64 "LPX64"(%d) != 0xffffffffffffffff\n",
485 u64val, (int)sizeof(u64val));
488 if (do_div(div64val, 256) != (u64val & 255)) {
489 CERROR("do_div("LPX64",256) != "LPU64"\n", u64val, u64val &255);
492 if (u64val >> 8 != div64val) {
493 CERROR("do_div("LPX64",256) "LPU64" != "LPU64"\n",
494 u64val, div64val, u64val >> 8);
497 len = snprintf(buf, sizeof(buf), LPX64, u64val);
499 CWARN("LPX64 wrong length! strlen(%s)=%d != 18\n", buf, len);
502 len = snprintf(buf, sizeof(buf), LPU64, u64val);
504 CWARN("LPU64 wrong length! strlen(%s)=%d != 20\n", buf, len);
507 len = snprintf(buf, sizeof(buf), LPD64, u64val);
509 CWARN("LPD64 wrong length! strlen(%s)=%d != 2\n", buf, len);
512 if ((u64val & ~CFS_PAGE_MASK) >= CFS_PAGE_SIZE) {
513 CWARN("mask failed: u64val "LPU64" >= "LPU64"\n", u64val,
514 (__u64)CFS_PAGE_SIZE);
521 #define obd_init_checks() do {} while(0)
524 extern spinlock_t obd_types_lock;
525 extern int class_procfs_init(void);
526 extern int class_procfs_clean(void);
529 static int __init init_obdclass(void)
531 int init_obdclass(void)
536 int lustre_register_fs(void);
538 for (i = CAPA_SITE_CLIENT; i < CAPA_SITE_MAX; i++)
539 CFS_INIT_LIST_HEAD(&capa_list[i]);
542 LCONSOLE_INFO("Lustre: Build Version: "BUILD_VERSION"\n");
544 spin_lock_init(&obd_types_lock);
545 obd_zombie_impexp_init();
547 obd_memory = lprocfs_alloc_stats(OBD_STATS_NUM,
548 LPROCFS_STATS_FLAG_NONE |
549 LPROCFS_STATS_FLAG_IRQ_SAFE);
550 if (obd_memory == NULL) {
551 CERROR("kmalloc of 'obd_memory' failed\n");
555 lprocfs_counter_init(obd_memory, OBD_MEMORY_STAT,
556 LPROCFS_CNTR_AVGMINMAX,
558 lprocfs_counter_init(obd_memory, OBD_MEMORY_PAGES_STAT,
559 LPROCFS_CNTR_AVGMINMAX,
560 "pagesused", "pages");
562 err = obd_init_checks();
563 if (err == -EOVERFLOW)
566 class_init_uuidlist();
567 err = class_handle_init();
571 CFS_INIT_LIST_HEAD(&obd_types);
573 err = cfs_psdev_register(&obd_psdev);
575 CERROR("cannot register %d err %d\n", OBD_DEV_MINOR, err);
579 /* This struct is already zeroed for us (static global) */
580 for (i = 0; i < class_devno_max(); i++)
583 /* Default the dirty page cache cap to 1/2 of system memory.
584 * For clients with less memory, a larger fraction is needed
585 * for other purposes (mostly for BGL). */
586 if (cfs_num_physpages <= 512 << (20 - CFS_PAGE_SHIFT))
587 obd_max_dirty_pages = cfs_num_physpages / 4;
589 obd_max_dirty_pages = cfs_num_physpages / 2;
591 err = obd_init_caches();
595 err = class_procfs_init();
600 err = lu_global_init();
604 err = llog_info_init();
609 err = lustre_register_fs();
615 void obd_update_maxusage(void)
619 max1 = obd_pages_sum();
620 max2 = obd_memory_sum();
622 spin_lock(&obd_updatemax_lock);
623 if (max1 > obd_max_pages)
624 obd_max_pages = max1;
625 if (max2 > obd_max_alloc)
626 obd_max_alloc = max2;
627 spin_unlock(&obd_updatemax_lock);
629 EXPORT_SYMBOL(obd_update_maxusage);
632 __u64 obd_memory_max(void)
636 spin_lock(&obd_updatemax_lock);
638 spin_unlock(&obd_updatemax_lock);
642 EXPORT_SYMBOL(obd_memory_max);
644 __u64 obd_pages_max(void)
648 spin_lock(&obd_updatemax_lock);
650 spin_unlock(&obd_updatemax_lock);
654 EXPORT_SYMBOL(obd_pages_max);
657 /* liblustre doesn't call cleanup_obdclass, apparently. we carry on in this
658 * ifdef to the end of the file to cover module and versioning goo.*/
660 static void cleanup_obdclass(void)
663 int lustre_unregister_fs(void);
664 __u64 memory_leaked, pages_leaked;
665 __u64 memory_max, pages_max;
668 lustre_unregister_fs();
670 cfs_psdev_deregister(&obd_psdev);
671 for (i = 0; i < class_devno_max(); i++) {
672 struct obd_device *obd = class_num2obd(i);
673 if (obd && obd->obd_set_up &&
674 OBT(obd) && OBP(obd, detach)) {
675 /* XXX should this call generic detach otherwise? */
676 LASSERT(obd->obd_magic == OBD_DEVICE_MAGIC);
677 OBP(obd, detach)(obd);
683 obd_cleanup_caches();
686 class_procfs_clean();
688 class_handle_cleanup();
689 class_exit_uuidlist();
690 obd_zombie_impexp_stop();
692 memory_leaked = obd_memory_sum();
693 pages_leaked = obd_pages_sum();
695 memory_max = obd_memory_max();
696 pages_max = obd_pages_max();
698 lprocfs_free_stats(&obd_memory);
699 CDEBUG((memory_leaked) ? D_ERROR : D_INFO,
700 "obd_memory max: "LPU64", leaked: "LPU64"\n",
701 memory_max, memory_leaked);
702 CDEBUG((pages_leaked) ? D_ERROR : D_INFO,
703 "obd_memory_pages max: "LPU64", leaked: "LPU64"\n",
704 pages_max, pages_leaked);
709 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
710 MODULE_DESCRIPTION("Lustre Class Driver Build Version: " BUILD_VERSION);
711 MODULE_LICENSE("GPL");
713 cfs_module(obdclass, LUSTRE_VERSION_STRING, init_obdclass, cleanup_obdclass);