1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * Copyright (C) 2001 Cluster File Systems, Inc. <braam@clusterfs.com>
6 * This file is part of Lustre, http://www.lustre.org.
8 * Lustre is free software; you can redistribute it and/or
9 * modify it under the terms of version 2 of the GNU General Public
10 * License as published by the Free Software Foundation.
12 * Lustre is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
17 * You should have received a copy of the GNU General Public License
18 * along with Lustre; if not, write to the Free Software
19 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
21 * Basic Lustre library routines.
32 # include <sys/types.h>
34 # include <asm/semaphore.h>
35 # include <linux/rwsem.h>
36 # include <linux/sched.h>
37 # include <linux/signal.h>
38 # include <linux/types.h>
40 #include <linux/portals_lib.h>
41 #include <linux/kp30.h> /* XXX just for LASSERT! */
42 #include <linux/lustre_idl.h>
43 #include <linux/lustre_cfg.h>
46 /* x86_64 has 64bit longs and defines u64 as long long */
47 #if BITS_PER_LONG > 32 && !defined(__x86_64__)
51 #define LP_POISON ((void *)0x5a5a5a5a5a5a5a5a)
56 #define LP_POISON ((void *)0x5a5a5a5a)
61 struct ptlrpc_request;
65 #include <linux/lustre_ha.h>
66 #include <linux/lustre_net.h>
67 #include <linux/lustre_compat25.h>
68 #include <linux/lvfs.h>
70 int target_handle_connect(struct ptlrpc_request *req, svc_handler_t handler);
71 int target_handle_disconnect(struct ptlrpc_request *req);
72 void target_destroy_export(struct obd_export *exp);
73 int target_handle_reconnect(struct lustre_handle *conn, struct obd_export *exp,
74 struct obd_uuid *cluuid);
75 int target_handle_ping(struct ptlrpc_request *req);
76 void target_cancel_recovery_timer(struct obd_device *obd);
78 #define OBD_RECOVERY_TIMEOUT (obd_timeout * 5 * HZ / 2) /* *waves hands* */
79 void target_start_recovery_timer(struct obd_device *obd, svc_handler_t handler);
80 void target_abort_recovery(void *data);
81 int target_queue_recovery_request(struct ptlrpc_request *req,
82 struct obd_device *obd);
83 int target_queue_final_reply(struct ptlrpc_request *req, int rc);
84 void target_send_reply(struct ptlrpc_request *req, int rc, int fail_id);
88 int client_sanobd_setup(struct obd_device *obddev, obd_count len, void *buf);
89 struct client_obd *client_conn2cli(struct lustre_handle *conn);
92 struct obd_client_handle {
93 struct lustre_handle och_fh;
94 struct llog_cookie och_cookie;
95 struct mdc_open_data *och_mod;
98 #define OBD_CLIENT_HANDLE_MAGIC 0xd15ea5ed
101 void statfs_pack(struct obd_statfs *osfs, struct kstatfs *sfs);
102 void statfs_unpack(struct kstatfs *sfs, struct obd_statfs *osfs);
107 struct task_struct *l_owner;
108 struct semaphore l_sem;
112 void l_lock_init(struct lustre_lock *);
113 void l_lock(struct lustre_lock *);
114 void l_unlock(struct lustre_lock *);
115 int l_has_lock(struct lustre_lock *);
118 #include <linux/portals_lib.h>
123 #define OBD_IOCTL_VERSION 0x00010004
125 struct obd_ioctl_data {
127 uint32_t ioc_version;
133 struct obdo ioc_obdo1;
134 struct obdo ioc_obdo2;
139 uint32_t ioc_command;
145 /* buffers the kernel will treat as user pointers */
151 /* inline buffers for various arguments */
152 uint32_t ioc_inllen1;
154 uint32_t ioc_inllen2;
156 uint32_t ioc_inllen3;
158 uint32_t ioc_inllen4;
164 struct obd_ioctl_hdr {
166 uint32_t ioc_version;
169 static inline int obd_ioctl_packlen(struct obd_ioctl_data *data)
171 int len = size_round(sizeof(struct obd_ioctl_data));
172 len += size_round(data->ioc_inllen1);
173 len += size_round(data->ioc_inllen2);
174 len += size_round(data->ioc_inllen3);
175 len += size_round(data->ioc_inllen4);
180 static inline int obd_ioctl_is_invalid(struct obd_ioctl_data *data)
182 if (data->ioc_len > (1<<30)) {
183 printk("LustreError: OBD ioctl: ioc_len larger than 1<<30\n");
186 if (data->ioc_inllen1 > (1<<30)) {
187 printk("LustreError: OBD ioctl: ioc_inllen1 larger than 1<<30\n");
190 if (data->ioc_inllen2 > (1<<30)) {
191 printk("LustreError: OBD ioctl: ioc_inllen2 larger than 1<<30\n");
194 if (data->ioc_inllen3 > (1<<30)) {
195 printk("LustreError: OBD ioctl: ioc_inllen3 larger than 1<<30\n");
198 if (data->ioc_inllen4 > (1<<30)) {
199 printk("LustreError: OBD ioctl: ioc_inllen4 larger than 1<<30\n");
202 if (data->ioc_inlbuf1 && !data->ioc_inllen1) {
203 printk("LustreError: OBD ioctl: inlbuf1 pointer but 0 length\n");
206 if (data->ioc_inlbuf2 && !data->ioc_inllen2) {
207 printk("LustreError: OBD ioctl: inlbuf2 pointer but 0 length\n");
210 if (data->ioc_inlbuf3 && !data->ioc_inllen3) {
211 printk("LustreError: OBD ioctl: inlbuf3 pointer but 0 length\n");
214 if (data->ioc_inlbuf4 && !data->ioc_inllen4) {
215 printk("LustreError: OBD ioctl: inlbuf4 pointer but 0 length\n");
218 if (data->ioc_pbuf1 && !data->ioc_plen1) {
219 printk("LustreError: OBD ioctl: pbuf1 pointer but 0 length\n");
222 if (data->ioc_pbuf2 && !data->ioc_plen2) {
223 printk("LustreError: OBD ioctl: pbuf2 pointer but 0 length\n");
226 if (data->ioc_plen1 && !data->ioc_pbuf1) {
227 printk("LustreError: OBD ioctl: plen1 set but NULL pointer\n");
230 if (data->ioc_plen2 && !data->ioc_pbuf2) {
231 printk("LustreError: OBD ioctl: plen2 set but NULL pointer\n");
234 if (obd_ioctl_packlen(data) != data->ioc_len) {
235 printk("LustreError: OBD ioctl: packlen exceeds ioc_len (%d != %d)\n",
236 obd_ioctl_packlen(data), data->ioc_len);
243 static inline int obd_ioctl_pack(struct obd_ioctl_data *data, char **pbuf,
247 struct obd_ioctl_data *overlay;
248 data->ioc_len = obd_ioctl_packlen(data);
249 data->ioc_version = OBD_IOCTL_VERSION;
251 if (*pbuf && data->ioc_len > max)
254 *pbuf = malloc(data->ioc_len);
258 overlay = (struct obd_ioctl_data *)*pbuf;
259 memcpy(*pbuf, data, sizeof(*data));
261 ptr = overlay->ioc_bulk;
262 if (data->ioc_inlbuf1)
263 LOGL(data->ioc_inlbuf1, data->ioc_inllen1, ptr);
264 if (data->ioc_inlbuf2)
265 LOGL(data->ioc_inlbuf2, data->ioc_inllen2, ptr);
266 if (data->ioc_inlbuf3)
267 LOGL(data->ioc_inlbuf3, data->ioc_inllen3, ptr);
268 if (data->ioc_inlbuf4)
269 LOGL(data->ioc_inlbuf4, data->ioc_inllen4, ptr);
270 if (obd_ioctl_is_invalid(overlay))
276 static inline int obd_ioctl_unpack(struct obd_ioctl_data *data, char *pbuf,
280 struct obd_ioctl_data *overlay;
284 overlay = (struct obd_ioctl_data *)pbuf;
286 /* Preserve the caller's buffer pointers */
287 overlay->ioc_inlbuf1 = data->ioc_inlbuf1;
288 overlay->ioc_inlbuf2 = data->ioc_inlbuf2;
289 overlay->ioc_inlbuf3 = data->ioc_inlbuf3;
290 overlay->ioc_inlbuf4 = data->ioc_inlbuf4;
292 memcpy(data, pbuf, sizeof(*data));
294 ptr = overlay->ioc_bulk;
295 if (data->ioc_inlbuf1)
296 LOGU(data->ioc_inlbuf1, data->ioc_inllen1, ptr);
297 if (data->ioc_inlbuf2)
298 LOGU(data->ioc_inlbuf2, data->ioc_inllen2, ptr);
299 if (data->ioc_inlbuf3)
300 LOGU(data->ioc_inlbuf3, data->ioc_inllen3, ptr);
301 if (data->ioc_inlbuf4)
302 LOGU(data->ioc_inlbuf4, data->ioc_inllen4, ptr);
308 #include <linux/obd_support.h>
310 /* buffer MUST be at least the size of obd_ioctl_hdr */
311 static inline int obd_ioctl_getdata(char **buf, int *len, void *arg)
313 struct obd_ioctl_hdr hdr;
314 struct obd_ioctl_data *data;
319 err = copy_from_user(&hdr, (void *)arg, sizeof(hdr));
325 if (hdr.ioc_version != OBD_IOCTL_VERSION) {
326 CERROR("Version mismatch kernel vs application\n");
330 if (hdr.ioc_len > OBD_MAX_IOCTL_BUFFER) {
331 CERROR("User buffer len %d exceeds %d max buffer\n",
332 hdr.ioc_len, OBD_MAX_IOCTL_BUFFER);
336 if (hdr.ioc_len < sizeof(struct obd_ioctl_data)) {
337 printk("LustreError: OBD: user buffer too small for ioctl\n");
341 /* XXX allocate this more intelligently, using kmalloc when
343 OBD_VMALLOC(*buf, hdr.ioc_len);
345 CERROR("Cannot allocate control buffer of len %d\n",
350 data = (struct obd_ioctl_data *)*buf;
352 err = copy_from_user(*buf, (void *)arg, hdr.ioc_len);
358 if (obd_ioctl_is_invalid(data)) {
359 CERROR("ioctl not correctly formatted\n");
363 if (data->ioc_inllen1) {
364 data->ioc_inlbuf1 = &data->ioc_bulk[0];
365 offset += size_round(data->ioc_inllen1);
368 if (data->ioc_inllen2) {
369 data->ioc_inlbuf2 = &data->ioc_bulk[0] + offset;
370 offset += size_round(data->ioc_inllen2);
373 if (data->ioc_inllen3) {
374 data->ioc_inlbuf3 = &data->ioc_bulk[0] + offset;
375 offset += size_round(data->ioc_inllen3);
378 if (data->ioc_inllen4) {
379 data->ioc_inlbuf4 = &data->ioc_bulk[0] + offset;
386 static inline void obd_ioctl_freedata(char *buf, int len)
395 #define OBD_IOC_CREATE _IOR ('f', 101, long)
396 #define OBD_IOC_DESTROY _IOW ('f', 104, long)
397 #define OBD_IOC_PREALLOCATE _IOWR('f', 105, long)
399 #define OBD_IOC_SETATTR _IOW ('f', 107, long)
400 #define OBD_IOC_GETATTR _IOR ('f', 108, long)
401 #define OBD_IOC_READ _IOWR('f', 109, long)
402 #define OBD_IOC_WRITE _IOWR('f', 110, long)
405 #define OBD_IOC_STATFS _IOWR('f', 113, long)
406 #define OBD_IOC_SYNC _IOR ('f', 114, long)
407 #define OBD_IOC_READ2 _IOWR('f', 115, long)
408 #define OBD_IOC_FORMAT _IOWR('f', 116, long)
409 #define OBD_IOC_PARTITION _IOWR('f', 117, long)
410 #define OBD_IOC_COPY _IOWR('f', 120, long)
411 #define OBD_IOC_MIGR _IOWR('f', 121, long)
412 #define OBD_IOC_PUNCH _IOWR('f', 122, long)
414 #define OBD_IOC_MODULE_DEBUG _IOWR('f', 124, long)
415 #define OBD_IOC_BRW_READ _IOWR('f', 125, long)
416 #define OBD_IOC_BRW_WRITE _IOWR('f', 126, long)
417 #define OBD_IOC_NAME2DEV _IOWR('f', 127, long)
418 #define OBD_IOC_UUID2DEV _IOWR('f', 130, long)
420 #define OBD_IOC_LOV_GET_CONFIG _IOWR('f', 132, long)
421 #define OBD_IOC_CLIENT_RECOVER _IOW ('f', 133, long)
423 #define OBD_IOC_PING _IOWR('f', 135, long)
425 #define OBD_IOC_DEC_FS_USE_COUNT _IO ('f', 139 )
426 #define OBD_IOC_NO_TRANSNO _IOW ('f', 140, long)
427 #define OBD_IOC_SET_READONLY _IOW ('f', 141, long)
428 #define OBD_IOC_ABORT_RECOVERY _IOR ('f', 142, long)
430 #define OBD_GET_VERSION _IOWR ('f', 144, long)
432 #define OBD_IOC_CLOSE_UUID _IOWR ('f', 147, long)
434 #define OBD_IOC_LOV_SETSTRIPE _IOW ('f', 154, long)
435 #define OBD_IOC_LOV_GETSTRIPE _IOW ('f', 155, long)
436 #define OBD_IOC_LOV_SETEA _IOW ('f', 156, long)
438 #define OBD_IOC_MOUNTOPT _IOWR('f', 170, long)
440 #define OBD_IOC_RECORD _IOWR('f', 180, long)
441 #define OBD_IOC_ENDRECORD _IOWR('f', 181, long)
442 #define OBD_IOC_PARSE _IOWR('f', 182, long)
443 #define OBD_IOC_DORECORD _IOWR('f', 183, long)
444 #define OBD_IOC_PROCESS_CFG _IOWR('f', 184, long)
445 #define OBD_IOC_DUMP_LOG _IOWR('f', 185, long)
446 #define OBD_IOC_CLEAR_LOG _IOWR('f', 186, long)
448 #define OBD_IOC_CATLOGLIST _IOWR('f', 190, long)
449 #define OBD_IOC_LLOG_INFO _IOWR('f', 191, long)
450 #define OBD_IOC_LLOG_PRINT _IOWR('f', 192, long)
451 #define OBD_IOC_LLOG_CANCEL _IOWR('f', 193, long)
452 #define OBD_IOC_LLOG_REMOVE _IOWR('f', 194, long)
453 #define OBD_IOC_LLOG_CHECK _IOWR('f', 195, long)
454 #define OBD_IOC_LLOG_CATINFO _IOWR('f', 196, long)
456 #define ECHO_IOC_GET_STRIPE _IOWR('f', 200, long)
457 #define ECHO_IOC_SET_STRIPE _IOWR('f', 201, long)
458 #define ECHO_IOC_ENQUEUE _IOWR('f', 202, long)
459 #define ECHO_IOC_CANCEL _IOWR('f', 203, long)
461 /* XXX _IOWR('f', 250, long) has been defined in
462 * portals/include/linux/kp30.h for debug, don't use it
465 /* Until such time as we get_info the per-stripe maximum from the OST,
466 * we define this to be 2T - 4k, which is the ext3 maxbytes. */
467 #define LUSTRE_STRIPE_MAXBYTES 0x1fffffff000ULL
469 #define CHECKSUM_BULK 0
470 #define POISON_BULK 0
473 static inline void ost_checksum(obd_count *cksum, void *addr, int len)
475 unsigned char *ptr = (unsigned char *)addr;
478 /* very stupid, but means I don't have to think about byte order */
482 *cksum = (*cksum << 2) + sum;
486 static inline int ll_insecure_random_int(void)
491 return (int)(t.tv_usec);
493 return (int)(get_cycles() >> 2);
498 * l_wait_event is a flexible sleeping function, permitting simple caller
499 * configuration of interrupt and timeout sensitivity along with actions to
500 * be performed in the event of either exception.
502 * Common usage looks like this:
504 * struct l_wait_info lwi = LWI_TIMEOUT_INTR(timeout, timeout_handler,
505 * intr_handler, callback_data);
506 * rc = l_wait_event(waitq, condition, &lwi);
508 * (LWI_TIMEOUT and LWI_INTR macros are available for timeout- and
509 * interrupt-only variants, respectively.)
511 * If a timeout is specified, the timeout_handler will be invoked in the event
512 * that the timeout expires before the process is awakened. (Note that any
513 * waking of the process will restart the timeout, even if the condition is
514 * not satisfied and the process immediately returns to sleep. This might be
515 * considered a bug.) If the timeout_handler returns non-zero, l_wait_event
516 * will return -ETIMEDOUT and the caller will continue. If the handler returns
517 * zero instead, the process will go back to sleep until it is awakened by the
518 * waitq or some similar mechanism, or an interrupt occurs (if the caller has
519 * asked for interrupts to be detected). The timeout will only fire once, so
520 * callers should take care that a timeout_handler which returns zero will take
521 * future steps to awaken the process. N.B. that these steps must include
522 * making the provided condition become true.
524 * If the interrupt flag (lwi_signals) is non-zero, then the process will be
525 * interruptible, and will be awakened by any "killable" signal (SIGTERM,
526 * SIGKILL or SIGINT). If a timeout is also specified, then the process will
527 * only become interruptible _after_ the timeout has expired, though it can be
528 * awakened by a signal that was delivered before the timeout and is still
529 * pending when the timeout expires. If a timeout is not specified, the process
530 * will be interruptible at all times during l_wait_event.
535 int (*lwi_on_timeout)(void *);
537 void (*lwi_on_signal)(void *);
541 #define LWI_TIMEOUT(time, cb, data) \
542 ((struct l_wait_info) { \
544 lwi_on_timeout: cb, \
548 #define LWI_INTR(cb, data) \
549 ((struct l_wait_info) { \
555 #define LWI_TIMEOUT_INTR(time, time_cb, sig_cb, data) \
556 ((struct l_wait_info) { \
558 lwi_on_timeout: time_cb, \
560 lwi_on_signal: sig_cb, \
564 #define LUSTRE_FATAL_SIGS (sigmask(SIGKILL) | sigmask(SIGINT) | \
565 sigmask(SIGTERM) | sigmask(SIGQUIT) | \
569 static inline sigset_t l_w_e_set_sigs(int sigs)
572 unsigned long irqflags;
574 SIGNAL_MASK_LOCK(current, irqflags);
575 old = current->blocked;
576 siginitsetinv(¤t->blocked, sigs);
578 SIGNAL_MASK_UNLOCK(current, irqflags);
583 #define __l_wait_event(wq, condition, info, ret, excl) \
585 wait_queue_t __wait; \
586 int __timed_out = 0; \
587 unsigned long irqflags; \
590 init_waitqueue_entry(&__wait, current); \
592 add_wait_queue_exclusive(&wq, &__wait); \
594 add_wait_queue(&wq, &__wait); \
596 /* Block all signals (just the non-fatal ones if no timeout). */ \
597 if (info->lwi_signals && !info->lwi_timeout) \
598 blocked = l_w_e_set_sigs(LUSTRE_FATAL_SIGS); \
600 blocked = l_w_e_set_sigs(0); \
603 set_current_state(TASK_INTERRUPTIBLE); \
606 if (signal_pending(current)) { \
607 if (info->lwi_on_signal) \
608 info->lwi_on_signal(info->lwi_cb_data); \
612 if (info->lwi_timeout && !__timed_out) { \
613 if (schedule_timeout(info->lwi_timeout) == 0) { \
615 if (!info->lwi_on_timeout || \
616 info->lwi_on_timeout(info->lwi_cb_data)) { \
620 /* We'll take signals after a timeout. */ \
621 if (info->lwi_signals) \
622 (void)l_w_e_set_sigs(LUSTRE_FATAL_SIGS); \
629 SIGNAL_MASK_LOCK(current, irqflags); \
630 current->blocked = blocked; \
632 SIGNAL_MASK_UNLOCK(current, irqflags); \
634 current->state = TASK_RUNNING; \
635 remove_wait_queue(&wq, &__wait); \
638 #else /* !__KERNEL__ */
639 #define __l_wait_event(wq, condition, info, ret, excl) \
641 int __timed_out = 0; \
646 if (liblustre_wait_event(info->lwi_timeout)) \
648 if (info->lwi_timeout && !__timed_out) { \
650 if (info->lwi_on_timeout == NULL || \
651 info->lwi_on_timeout(info->lwi_cb_data)) { \
659 #endif /* __KERNEL__ */
661 #define l_wait_event(wq, condition, info) \
664 struct l_wait_info *__info = (info); \
666 __l_wait_event(wq, condition, __info, __ret, 0); \
670 #define l_wait_event_exclusive(wq, condition, info) \
673 struct l_wait_info *__info = (info); \
675 __l_wait_event(wq, condition, __info, __ret, 1); \
679 #endif /* _LUSTRE_LIB_H */
681 #define LMD_MAGIC 0xbdacbdac
683 #define lmd_bad_magic(LMDP) \
685 struct lustre_mount_data *_lmd__ = (LMDP); \
688 CERROR("Missing mount data: " \
689 "check that /sbin/mount.lustre is installed.\n");\
691 } else if (_lmd__->lmd_magic != LMD_MAGIC) { \
692 CERROR("Invalid mount data (%#x != %#x): " \
693 "check that /sbin/mount.lustre is installed\n", \
694 _lmd__->lmd_magic, LMD_MAGIC); \