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, Whamcloud, Inc.
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);
64 __u64 obd_max_pages = 0;
65 __u64 obd_max_alloc = 0;
70 /* The following are visible and mutable through /proc/sys/lustre/. */
71 unsigned int obd_debug_peer_on_timeout;
72 EXPORT_SYMBOL(obd_debug_peer_on_timeout);
73 unsigned int obd_dump_on_timeout;
74 EXPORT_SYMBOL(obd_dump_on_timeout);
75 unsigned int obd_dump_on_eviction;
76 EXPORT_SYMBOL(obd_dump_on_eviction);
77 unsigned int obd_max_dirty_pages = 256;
78 EXPORT_SYMBOL(obd_max_dirty_pages);
79 cfs_atomic_t obd_dirty_pages;
80 EXPORT_SYMBOL(obd_dirty_pages);
81 unsigned int obd_timeout = OBD_TIMEOUT_DEFAULT; /* seconds */
82 EXPORT_SYMBOL(obd_timeout);
83 unsigned int ldlm_timeout = LDLM_TIMEOUT_DEFAULT; /* seconds */
84 EXPORT_SYMBOL(ldlm_timeout);
85 unsigned int obd_timeout_set;
86 EXPORT_SYMBOL(obd_timeout_set);
87 unsigned int ldlm_timeout_set;
88 EXPORT_SYMBOL(ldlm_timeout_set);
89 /* Adaptive timeout defs here instead of ptlrpc module for /proc/sys/ access */
90 unsigned int at_min = 0;
91 EXPORT_SYMBOL(at_min);
92 unsigned int at_max = 600;
93 EXPORT_SYMBOL(at_max);
94 unsigned int at_history = 600;
95 EXPORT_SYMBOL(at_history);
96 int at_early_margin = 5;
97 EXPORT_SYMBOL(at_early_margin);
99 EXPORT_SYMBOL(at_extra);
101 cfs_atomic_t obd_dirty_transit_pages;
102 EXPORT_SYMBOL(obd_dirty_transit_pages);
104 char obd_jobid_var[JOBSTATS_JOBID_VAR_MAX_LEN + 1] = JOBSTATS_DISABLE;
105 EXPORT_SYMBOL(obd_jobid_var);
107 /* Get jobid of current process by reading the environment variable
108 * stored in between the "env_start" & "env_end" of task struct.
111 * It's better to cache the jobid for later use if there is any
112 * efficient way, the cl_env code probably could be reused for this
115 * If some job scheduler doesn't store jobid in the "env_start/end",
116 * then an upcall could be issued here to get the jobid by utilizing
117 * the userspace tools/api. Then, the jobid must be cached.
119 int lustre_get_jobid(char *jobid)
121 int jobid_len = JOBSTATS_JOBID_SIZE;
125 memset(jobid, 0, JOBSTATS_JOBID_SIZE);
126 /* Jobstats isn't enabled */
127 if (strcmp(obd_jobid_var, JOBSTATS_DISABLE) == 0)
130 /* Use process name + fsuid as jobid */
131 if (strcmp(obd_jobid_var, JOBSTATS_PROCNAME_UID) == 0) {
132 snprintf(jobid, JOBSTATS_JOBID_SIZE, "%s.%u",
133 cfs_curproc_comm(), cfs_curproc_fsuid());
137 rc = cfs_get_environ(obd_jobid_var, jobid, &jobid_len);
139 if (rc == -EOVERFLOW) {
140 /* For the PBS_JOBID and LOADL_STEP_ID keys (which are
141 * variable length strings instead of just numbers), it
142 * might make sense to keep the unique parts for JobID,
143 * instead of just returning an error. That means a
144 * larger temp buffer for cfs_get_environ(), then
145 * truncating the string at some separator to fit into
146 * the specified jobid_len. Fix later if needed. */
148 if (unlikely(!printed)) {
149 LCONSOLE_ERROR_MSG(0x16b, "%s value too large "
150 "for JobID buffer (%d)\n",
151 obd_jobid_var, jobid_len);
155 CDEBUG((rc == -ENOENT || rc == -EINVAL ||
156 rc == -EDEADLK) ? D_INFO : D_ERROR,
157 "Get jobid for (%s) failed: rc = %d\n",
163 EXPORT_SYMBOL(lustre_get_jobid);
165 static inline void obd_data2conn(struct lustre_handle *conn,
166 struct obd_ioctl_data *data)
168 memset(conn, 0, sizeof *conn);
169 conn->cookie = data->ioc_cookie;
172 static inline void obd_conn2data(struct obd_ioctl_data *data,
173 struct lustre_handle *conn)
175 data->ioc_cookie = conn->cookie;
178 int class_resolve_dev_name(__u32 len, const char *name)
185 CERROR("No name passed,!\n");
186 GOTO(out, rc = -EINVAL);
188 if (name[len - 1] != 0) {
189 CERROR("Name not nul terminated!\n");
190 GOTO(out, rc = -EINVAL);
193 CDEBUG(D_IOCTL, "device name %s\n", name);
194 dev = class_name2dev(name);
196 CDEBUG(D_IOCTL, "No device for name %s!\n", name);
197 GOTO(out, rc = -EINVAL);
200 CDEBUG(D_IOCTL, "device name %s, dev %d\n", name, dev);
207 int class_handle_ioctl(unsigned int cmd, unsigned long arg)
210 struct obd_ioctl_data *data;
211 struct libcfs_debug_ioctl_data *debug_data;
212 struct obd_device *obd = NULL;
213 int err = 0, len = 0;
216 /* only for debugging */
217 if (cmd == LIBCFS_IOC_DEBUG_MASK) {
218 debug_data = (struct libcfs_debug_ioctl_data*)arg;
219 libcfs_subsystem_debug = debug_data->subs;
220 libcfs_debug = debug_data->debug;
224 CDEBUG(D_IOCTL, "cmd = %x\n", cmd);
225 if (obd_ioctl_getdata(&buf, &len, (void *)arg)) {
226 CERROR("OBD ioctl: data error\n");
229 data = (struct obd_ioctl_data *)buf;
232 case OBD_IOC_PROCESS_CFG: {
233 struct lustre_cfg *lcfg;
235 if (!data->ioc_plen1 || !data->ioc_pbuf1) {
236 CERROR("No config buffer passed!\n");
237 GOTO(out, err = -EINVAL);
239 OBD_ALLOC(lcfg, data->ioc_plen1);
241 GOTO(out, err = -ENOMEM);
242 err = cfs_copy_from_user(lcfg, data->ioc_pbuf1,
245 err = lustre_cfg_sanity_check(lcfg, data->ioc_plen1);
247 err = class_process_config(lcfg);
249 OBD_FREE(lcfg, data->ioc_plen1);
253 case OBD_GET_VERSION:
254 if (!data->ioc_inlbuf1) {
255 CERROR("No buffer passed in ioctl\n");
256 GOTO(out, err = -EINVAL);
259 if (strlen(BUILD_VERSION) + 1 > data->ioc_inllen1) {
260 CERROR("ioctl buffer too small to hold version\n");
261 GOTO(out, err = -EINVAL);
264 memcpy(data->ioc_bulk, BUILD_VERSION,
265 strlen(BUILD_VERSION) + 1);
267 err = obd_ioctl_popdata((void *)arg, data, len);
272 case OBD_IOC_NAME2DEV: {
273 /* Resolve a device name. This does not change the
274 * currently selected device.
278 dev = class_resolve_dev_name(data->ioc_inllen1,
282 GOTO(out, err = -EINVAL);
284 err = obd_ioctl_popdata((void *)arg, data, sizeof(*data));
290 case OBD_IOC_UUID2DEV: {
291 /* Resolve a device uuid. This does not change the
292 * currently selected device.
295 struct obd_uuid uuid;
297 if (!data->ioc_inllen1 || !data->ioc_inlbuf1) {
298 CERROR("No UUID passed!\n");
299 GOTO(out, err = -EINVAL);
301 if (data->ioc_inlbuf1[data->ioc_inllen1 - 1] != 0) {
302 CERROR("UUID not NUL terminated!\n");
303 GOTO(out, err = -EINVAL);
306 CDEBUG(D_IOCTL, "device name %s\n", data->ioc_inlbuf1);
307 obd_str2uuid(&uuid, data->ioc_inlbuf1);
308 dev = class_uuid2dev(&uuid);
311 CDEBUG(D_IOCTL, "No device for UUID %s!\n",
313 GOTO(out, err = -EINVAL);
316 CDEBUG(D_IOCTL, "device name %s, dev %d\n", data->ioc_inlbuf1,
318 err = obd_ioctl_popdata((void *)arg, data, sizeof(*data));
324 case OBD_IOC_CLOSE_UUID: {
325 CDEBUG(D_IOCTL, "closing all connections to uuid %s (NOOP)\n",
330 case OBD_IOC_GETDEVICE: {
331 int index = data->ioc_count;
334 if (!data->ioc_inlbuf1) {
335 CERROR("No buffer passed in ioctl\n");
336 GOTO(out, err = -EINVAL);
338 if (data->ioc_inllen1 < 128) {
339 CERROR("ioctl buffer too small to hold version\n");
340 GOTO(out, err = -EINVAL);
343 obd = class_num2obd(index);
345 GOTO(out, err = -ENOENT);
347 if (obd->obd_stopping)
349 else if (obd->obd_set_up)
351 else if (obd->obd_attached)
355 str = (char *)data->ioc_bulk;
356 snprintf(str, len - sizeof(*data), "%3d %s %s %s %s %d",
357 (int)index, status, obd->obd_type->typ_name,
358 obd->obd_name, obd->obd_uuid.uuid,
359 cfs_atomic_read(&obd->obd_refcount));
360 err = obd_ioctl_popdata((void *)arg, data, len);
367 if (data->ioc_dev == OBD_DEV_BY_DEVNAME) {
368 if (data->ioc_inllen4 <= 0 || data->ioc_inlbuf4 == NULL)
369 GOTO(out, err = -EINVAL);
370 if (strnlen(data->ioc_inlbuf4, MAX_OBD_NAME) >= MAX_OBD_NAME)
371 GOTO(out, err = -EINVAL);
372 obd = class_name2obd(data->ioc_inlbuf4);
373 } else if (data->ioc_dev < class_devno_max()) {
374 obd = class_num2obd(data->ioc_dev);
376 CERROR("OBD ioctl: No device\n");
377 GOTO(out, err = -EINVAL);
381 CERROR("OBD ioctl : No Device %d\n", data->ioc_dev);
382 GOTO(out, err = -EINVAL);
384 LASSERT(obd->obd_magic == OBD_DEVICE_MAGIC);
386 if (!obd->obd_set_up || obd->obd_stopping) {
387 CERROR("OBD ioctl: device not setup %d \n", data->ioc_dev);
388 GOTO(out, err = -EINVAL);
392 case OBD_IOC_NO_TRANSNO: {
393 if (!obd->obd_attached) {
394 CERROR("Device %d not attached\n", obd->obd_minor);
395 GOTO(out, err = -ENODEV);
397 CDEBUG(D_HA, "%s: disabling committed-transno notification\n",
399 obd->obd_no_transno = 1;
404 err = obd_iocontrol(cmd, obd->obd_self_export, len, data, NULL);
408 err = obd_ioctl_popdata((void *)arg, data, len);
417 obd_ioctl_freedata(buf, len);
419 } /* class_handle_ioctl */
422 extern cfs_psdev_t obd_psdev;
424 void *obd_psdev = NULL;
427 #define OBD_INIT_CHECK
428 #ifdef OBD_INIT_CHECK
429 int obd_init_checks(void)
431 __u64 u64val, div64val;
435 CDEBUG(D_INFO, "LPU64=%s, LPD64=%s, LPX64=%s\n", LPU64, LPD64, LPX64);
437 CDEBUG(D_INFO, "OBD_OBJECT_EOF = "LPX64"\n", (__u64)OBD_OBJECT_EOF);
439 u64val = OBD_OBJECT_EOF;
440 CDEBUG(D_INFO, "u64val OBD_OBJECT_EOF = "LPX64"\n", u64val);
441 if (u64val != OBD_OBJECT_EOF) {
442 CERROR("__u64 "LPX64"(%d) != 0xffffffffffffffff\n",
443 u64val, (int)sizeof(u64val));
446 len = snprintf(buf, sizeof(buf), LPX64, u64val);
448 CWARN("LPX64 wrong length! strlen(%s)=%d != 18\n", buf, len);
452 div64val = OBD_OBJECT_EOF;
453 CDEBUG(D_INFO, "u64val OBD_OBJECT_EOF = "LPX64"\n", u64val);
454 if (u64val != OBD_OBJECT_EOF) {
455 CERROR("__u64 "LPX64"(%d) != 0xffffffffffffffff\n",
456 u64val, (int)sizeof(u64val));
459 if (u64val >> 8 != OBD_OBJECT_EOF >> 8) {
460 CERROR("__u64 "LPX64"(%d) != 0xffffffffffffffff\n",
461 u64val, (int)sizeof(u64val));
464 if (do_div(div64val, 256) != (u64val & 255)) {
465 CERROR("do_div("LPX64",256) != "LPU64"\n", u64val, u64val &255);
468 if (u64val >> 8 != div64val) {
469 CERROR("do_div("LPX64",256) "LPU64" != "LPU64"\n",
470 u64val, div64val, u64val >> 8);
473 len = snprintf(buf, sizeof(buf), LPX64, u64val);
475 CWARN("LPX64 wrong length! strlen(%s)=%d != 18\n", buf, len);
478 len = snprintf(buf, sizeof(buf), LPU64, u64val);
480 CWARN("LPU64 wrong length! strlen(%s)=%d != 20\n", buf, len);
483 len = snprintf(buf, sizeof(buf), LPD64, u64val);
485 CWARN("LPD64 wrong length! strlen(%s)=%d != 2\n", buf, len);
488 if ((u64val & ~CFS_PAGE_MASK) >= CFS_PAGE_SIZE) {
489 CWARN("mask failed: u64val "LPU64" >= "LPU64"\n", u64val,
490 (__u64)CFS_PAGE_SIZE);
497 #define obd_init_checks() do {} while(0)
500 extern spinlock_t obd_types_lock;
501 extern int class_procfs_init(void);
502 extern int class_procfs_clean(void);
505 static int __init init_obdclass(void)
507 int init_obdclass(void)
512 int lustre_register_fs(void);
514 for (i = CAPA_SITE_CLIENT; i < CAPA_SITE_MAX; i++)
515 CFS_INIT_LIST_HEAD(&capa_list[i]);
518 LCONSOLE_INFO("Lustre: Build Version: "BUILD_VERSION"\n");
520 spin_lock_init(&obd_types_lock);
521 obd_zombie_impexp_init();
523 obd_memory = lprocfs_alloc_stats(OBD_STATS_NUM,
524 LPROCFS_STATS_FLAG_NONE |
525 LPROCFS_STATS_FLAG_IRQ_SAFE);
526 if (obd_memory == NULL) {
527 CERROR("kmalloc of 'obd_memory' failed\n");
531 lprocfs_counter_init(obd_memory, OBD_MEMORY_STAT,
532 LPROCFS_CNTR_AVGMINMAX,
534 lprocfs_counter_init(obd_memory, OBD_MEMORY_PAGES_STAT,
535 LPROCFS_CNTR_AVGMINMAX,
536 "pagesused", "pages");
538 err = obd_init_checks();
539 if (err == -EOVERFLOW)
542 class_init_uuidlist();
543 err = class_handle_init();
547 CFS_INIT_LIST_HEAD(&obd_types);
549 err = cfs_psdev_register(&obd_psdev);
551 CERROR("cannot register %d err %d\n", OBD_DEV_MINOR, err);
555 /* This struct is already zeroed for us (static global) */
556 for (i = 0; i < class_devno_max(); i++)
559 /* Default the dirty page cache cap to 1/2 of system memory.
560 * For clients with less memory, a larger fraction is needed
561 * for other purposes (mostly for BGL). */
562 if (cfs_num_physpages <= 512 << (20 - CFS_PAGE_SHIFT))
563 obd_max_dirty_pages = cfs_num_physpages / 4;
565 obd_max_dirty_pages = cfs_num_physpages / 2;
567 err = obd_init_caches();
571 err = class_procfs_init();
576 err = lu_global_init();
580 err = llog_info_init();
585 err = lustre_register_fs();
591 /* liblustre doesn't call cleanup_obdclass, apparently. we carry on in this
592 * ifdef to the end of the file to cover module and versioning goo.*/
594 static void cleanup_obdclass(void)
597 int lustre_unregister_fs(void);
598 __u64 memory_leaked, pages_leaked;
599 __u64 memory_max, pages_max;
602 lustre_unregister_fs();
604 cfs_psdev_deregister(&obd_psdev);
605 for (i = 0; i < class_devno_max(); i++) {
606 struct obd_device *obd = class_num2obd(i);
607 if (obd && obd->obd_set_up &&
608 OBT(obd) && OBP(obd, detach)) {
609 /* XXX should this call generic detach otherwise? */
610 LASSERT(obd->obd_magic == OBD_DEVICE_MAGIC);
611 OBP(obd, detach)(obd);
617 obd_cleanup_caches();
620 class_procfs_clean();
622 class_handle_cleanup();
623 class_exit_uuidlist();
624 obd_zombie_impexp_stop();
626 memory_leaked = obd_memory_sum();
627 pages_leaked = obd_pages_sum();
629 memory_max = obd_memory_max();
630 pages_max = obd_pages_max();
632 lprocfs_free_stats(&obd_memory);
633 CDEBUG((memory_leaked) ? D_ERROR : D_INFO,
634 "obd_memory max: "LPU64", leaked: "LPU64"\n",
635 memory_max, memory_leaked);
636 CDEBUG((pages_leaked) ? D_ERROR : D_INFO,
637 "obd_memory_pages max: "LPU64", leaked: "LPU64"\n",
638 pages_max, pages_leaked);
643 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
644 MODULE_DESCRIPTION("Lustre Class Driver Build Version: " BUILD_VERSION);
645 MODULE_LICENSE("GPL");
647 cfs_module(obdclass, LUSTRE_VERSION_STRING, init_obdclass, cleanup_obdclass);