[go: nahoru, domu]

1/*
2 * GPL HEADER START
3 *
4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5 *
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.
9 *
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).
15 *
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
19 *
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
22 * have any questions.
23 *
24 * GPL HEADER END
25 */
26/*
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
29 *
30 * Copyright (c) 2011, 2012, Intel Corporation.
31 */
32/*
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
35 *
36 * lustre/include/lustre_lib.h
37 *
38 * Basic Lustre library routines.
39 */
40
41#ifndef _LUSTRE_LIB_H
42#define _LUSTRE_LIB_H
43
44/** \defgroup lib lib
45 *
46 * @{
47 */
48
49#include <linux/sched.h>
50#include <linux/signal.h>
51#include <linux/types.h>
52#include "../../include/linux/libcfs/libcfs.h"
53#include "lustre/lustre_idl.h"
54#include "lustre_ver.h"
55#include "lustre_cfg.h"
56
57/* target.c */
58struct kstatfs;
59struct ptlrpc_request;
60struct obd_export;
61struct lu_target;
62struct l_wait_info;
63#include "lustre_ha.h"
64#include "lustre_net.h"
65
66#define LI_POISON 0x5a5a5a5a
67#if BITS_PER_LONG > 32
68# define LL_POISON 0x5a5a5a5a5a5a5a5aL
69#else
70# define LL_POISON 0x5a5a5a5aL
71#endif
72#define LP_POISON ((void *)LL_POISON)
73
74int target_pack_pool_reply(struct ptlrpc_request *req);
75int do_set_info_async(struct obd_import *imp,
76		      int opcode, int version,
77		      u32 keylen, void *key,
78		      u32 vallen, void *val,
79		      struct ptlrpc_request_set *set);
80
81#define OBD_RECOVERY_MAX_TIME (obd_timeout * 18) /* b13079 */
82#define OBD_MAX_IOCTL_BUFFER CONFIG_LUSTRE_OBD_MAX_IOCTL_BUFFER
83
84void target_send_reply(struct ptlrpc_request *req, int rc, int fail_id);
85
86/* client.c */
87
88int client_sanobd_setup(struct obd_device *obddev, struct lustre_cfg* lcfg);
89struct client_obd *client_conn2cli(struct lustre_handle *conn);
90
91struct md_open_data;
92struct obd_client_handle {
93	struct lustre_handle	 och_fh;
94	struct lu_fid		 och_fid;
95	struct md_open_data	*och_mod;
96	struct lustre_handle	 och_lease_handle; /* open lock for lease */
97	__u32			 och_magic;
98	fmode_t			 och_flags;
99};
100#define OBD_CLIENT_HANDLE_MAGIC 0xd15ea5ed
101
102/* statfs_pack.c */
103void statfs_pack(struct obd_statfs *osfs, struct kstatfs *sfs);
104void statfs_unpack(struct kstatfs *sfs, struct obd_statfs *osfs);
105
106/*
107 * For md echo client
108 */
109enum md_echo_cmd {
110	ECHO_MD_CREATE       = 1, /* Open/Create file on MDT */
111	ECHO_MD_MKDIR	= 2, /* Mkdir on MDT */
112	ECHO_MD_DESTROY      = 3, /* Unlink file on MDT */
113	ECHO_MD_RMDIR	= 4, /* Rmdir on MDT */
114	ECHO_MD_LOOKUP       = 5, /* Lookup on MDT */
115	ECHO_MD_GETATTR      = 6, /* Getattr on MDT */
116	ECHO_MD_SETATTR      = 7, /* Setattr on MDT */
117	ECHO_MD_ALLOC_FID    = 8, /* Get FIDs from MDT */
118};
119
120/*
121 *   OBD IOCTLS
122 */
123#define OBD_IOCTL_VERSION 0x00010004
124
125struct obd_ioctl_data {
126	__u32 ioc_len;
127	__u32 ioc_version;
128
129	union {
130		__u64 ioc_cookie;
131		__u64 ioc_u64_1;
132	};
133	union {
134		__u32 ioc_conn1;
135		__u32 ioc_u32_1;
136	};
137	union {
138		__u32 ioc_conn2;
139		__u32 ioc_u32_2;
140	};
141
142	struct obdo ioc_obdo1;
143	struct obdo ioc_obdo2;
144
145	u64	 ioc_count;
146	u64	 ioc_offset;
147	__u32    ioc_dev;
148	__u32    ioc_command;
149
150	__u64 ioc_nid;
151	__u32 ioc_nal;
152	__u32 ioc_type;
153
154	/* buffers the kernel will treat as user pointers */
155	__u32  ioc_plen1;
156	char  *ioc_pbuf1;
157	__u32  ioc_plen2;
158	char  *ioc_pbuf2;
159
160	/* inline buffers for various arguments */
161	__u32  ioc_inllen1;
162	char  *ioc_inlbuf1;
163	__u32  ioc_inllen2;
164	char  *ioc_inlbuf2;
165	__u32  ioc_inllen3;
166	char  *ioc_inlbuf3;
167	__u32  ioc_inllen4;
168	char  *ioc_inlbuf4;
169
170	char    ioc_bulk[0];
171};
172
173struct obd_ioctl_hdr {
174	__u32 ioc_len;
175	__u32 ioc_version;
176};
177
178static inline int obd_ioctl_packlen(struct obd_ioctl_data *data)
179{
180	int len = cfs_size_round(sizeof(struct obd_ioctl_data));
181	len += cfs_size_round(data->ioc_inllen1);
182	len += cfs_size_round(data->ioc_inllen2);
183	len += cfs_size_round(data->ioc_inllen3);
184	len += cfs_size_round(data->ioc_inllen4);
185	return len;
186}
187
188
189static inline int obd_ioctl_is_invalid(struct obd_ioctl_data *data)
190{
191	if (data->ioc_len > OBD_MAX_IOCTL_BUFFER) {
192		CERROR("OBD ioctl: ioc_len larger than %d\n",
193		       OBD_MAX_IOCTL_BUFFER);
194		return 1;
195	}
196	if (data->ioc_inllen1 > OBD_MAX_IOCTL_BUFFER) {
197		CERROR("OBD ioctl: ioc_inllen1 larger than ioc_len\n");
198		return 1;
199	}
200	if (data->ioc_inllen2 > OBD_MAX_IOCTL_BUFFER) {
201		CERROR("OBD ioctl: ioc_inllen2 larger than ioc_len\n");
202		return 1;
203	}
204	if (data->ioc_inllen3 > OBD_MAX_IOCTL_BUFFER) {
205		CERROR("OBD ioctl: ioc_inllen3 larger than ioc_len\n");
206		return 1;
207	}
208	if (data->ioc_inllen4 > OBD_MAX_IOCTL_BUFFER) {
209		CERROR("OBD ioctl: ioc_inllen4 larger than ioc_len\n");
210		return 1;
211	}
212	if (data->ioc_inlbuf1 && !data->ioc_inllen1) {
213		CERROR("OBD ioctl: inlbuf1 pointer but 0 length\n");
214		return 1;
215	}
216	if (data->ioc_inlbuf2 && !data->ioc_inllen2) {
217		CERROR("OBD ioctl: inlbuf2 pointer but 0 length\n");
218		return 1;
219	}
220	if (data->ioc_inlbuf3 && !data->ioc_inllen3) {
221		CERROR("OBD ioctl: inlbuf3 pointer but 0 length\n");
222		return 1;
223	}
224	if (data->ioc_inlbuf4 && !data->ioc_inllen4) {
225		CERROR("OBD ioctl: inlbuf4 pointer but 0 length\n");
226		return 1;
227	}
228	if (data->ioc_pbuf1 && !data->ioc_plen1) {
229		CERROR("OBD ioctl: pbuf1 pointer but 0 length\n");
230		return 1;
231	}
232	if (data->ioc_pbuf2 && !data->ioc_plen2) {
233		CERROR("OBD ioctl: pbuf2 pointer but 0 length\n");
234		return 1;
235	}
236	if (data->ioc_plen1 && !data->ioc_pbuf1) {
237		CERROR("OBD ioctl: plen1 set but NULL pointer\n");
238		return 1;
239	}
240	if (data->ioc_plen2 && !data->ioc_pbuf2) {
241		CERROR("OBD ioctl: plen2 set but NULL pointer\n");
242		return 1;
243	}
244	if (obd_ioctl_packlen(data) > data->ioc_len) {
245		CERROR("OBD ioctl: packlen exceeds ioc_len (%d > %d)\n",
246		       obd_ioctl_packlen(data), data->ioc_len);
247		return 1;
248	}
249	return 0;
250}
251
252
253#include "obd_support.h"
254
255/* function defined in lustre/obdclass/<platform>/<platform>-module.c */
256int obd_ioctl_getdata(char **buf, int *len, void *arg);
257int obd_ioctl_popdata(void *arg, void *data, int len);
258
259static inline void obd_ioctl_freedata(char *buf, int len)
260{
261	OBD_FREE_LARGE(buf, len);
262	return;
263}
264
265/*
266 * BSD ioctl description:
267 * #define IOC_V1       _IOR(g, n1, long)
268 * #define IOC_V2       _IOW(g, n2, long)
269 *
270 * ioctl(f, IOC_V1, arg);
271 * arg will be treated as a long value,
272 *
273 * ioctl(f, IOC_V2, arg)
274 * arg will be treated as a pointer, bsd will call
275 * copyin(buf, arg, sizeof(long))
276 *
277 * To make BSD ioctl handles argument correctly and simplely,
278 * we change _IOR to _IOWR so BSD will copyin obd_ioctl_data
279 * for us. Does this change affect Linux?  (XXX Liang)
280 */
281#define OBD_IOC_DATA_TYPE long
282
283#define OBD_IOC_CREATE		 _IOWR('f', 101, OBD_IOC_DATA_TYPE)
284#define OBD_IOC_DESTROY		_IOW ('f', 104, OBD_IOC_DATA_TYPE)
285#define OBD_IOC_PREALLOCATE	    _IOWR('f', 105, OBD_IOC_DATA_TYPE)
286
287#define OBD_IOC_SETATTR		_IOW ('f', 107, OBD_IOC_DATA_TYPE)
288#define OBD_IOC_GETATTR		_IOWR ('f', 108, OBD_IOC_DATA_TYPE)
289#define OBD_IOC_READ		   _IOWR('f', 109, OBD_IOC_DATA_TYPE)
290#define OBD_IOC_WRITE		  _IOWR('f', 110, OBD_IOC_DATA_TYPE)
291
292
293#define OBD_IOC_STATFS		 _IOWR('f', 113, OBD_IOC_DATA_TYPE)
294#define OBD_IOC_SYNC		   _IOW ('f', 114, OBD_IOC_DATA_TYPE)
295#define OBD_IOC_READ2		  _IOWR('f', 115, OBD_IOC_DATA_TYPE)
296#define OBD_IOC_FORMAT		 _IOWR('f', 116, OBD_IOC_DATA_TYPE)
297#define OBD_IOC_PARTITION	      _IOWR('f', 117, OBD_IOC_DATA_TYPE)
298#define OBD_IOC_COPY		   _IOWR('f', 120, OBD_IOC_DATA_TYPE)
299#define OBD_IOC_MIGR		   _IOWR('f', 121, OBD_IOC_DATA_TYPE)
300#define OBD_IOC_PUNCH		  _IOWR('f', 122, OBD_IOC_DATA_TYPE)
301
302#define OBD_IOC_MODULE_DEBUG	   _IOWR('f', 124, OBD_IOC_DATA_TYPE)
303#define OBD_IOC_BRW_READ	       _IOWR('f', 125, OBD_IOC_DATA_TYPE)
304#define OBD_IOC_BRW_WRITE	      _IOWR('f', 126, OBD_IOC_DATA_TYPE)
305#define OBD_IOC_NAME2DEV	       _IOWR('f', 127, OBD_IOC_DATA_TYPE)
306#define OBD_IOC_UUID2DEV	       _IOWR('f', 130, OBD_IOC_DATA_TYPE)
307
308#define OBD_IOC_GETNAME		_IOWR('f', 131, OBD_IOC_DATA_TYPE)
309#define OBD_IOC_GETMDNAME	      _IOR('f', 131, char[MAX_OBD_NAME])
310#define OBD_IOC_GETDTNAME	       OBD_IOC_GETNAME
311
312#define OBD_IOC_LOV_GET_CONFIG	 _IOWR('f', 132, OBD_IOC_DATA_TYPE)
313#define OBD_IOC_CLIENT_RECOVER	 _IOW ('f', 133, OBD_IOC_DATA_TYPE)
314#define OBD_IOC_PING_TARGET	    _IOW ('f', 136, OBD_IOC_DATA_TYPE)
315
316#define OBD_IOC_DEC_FS_USE_COUNT       _IO  ('f', 139      )
317#define OBD_IOC_NO_TRANSNO	     _IOW ('f', 140, OBD_IOC_DATA_TYPE)
318#define OBD_IOC_SET_READONLY	   _IOW ('f', 141, OBD_IOC_DATA_TYPE)
319#define OBD_IOC_ABORT_RECOVERY	 _IOR ('f', 142, OBD_IOC_DATA_TYPE)
320
321#define OBD_IOC_ROOT_SQUASH	    _IOWR('f', 143, OBD_IOC_DATA_TYPE)
322
323#define OBD_GET_VERSION		_IOWR ('f', 144, OBD_IOC_DATA_TYPE)
324
325#define OBD_IOC_GSS_SUPPORT	    _IOWR('f', 145, OBD_IOC_DATA_TYPE)
326
327#define OBD_IOC_CLOSE_UUID	     _IOWR ('f', 147, OBD_IOC_DATA_TYPE)
328
329#define OBD_IOC_CHANGELOG_SEND	 _IOW ('f', 148, OBD_IOC_DATA_TYPE)
330#define OBD_IOC_GETDEVICE	      _IOWR ('f', 149, OBD_IOC_DATA_TYPE)
331#define OBD_IOC_FID2PATH	       _IOWR ('f', 150, OBD_IOC_DATA_TYPE)
332/* see also <lustre/lustre_user.h> for ioctls 151-153 */
333/* OBD_IOC_LOV_SETSTRIPE: See also LL_IOC_LOV_SETSTRIPE */
334#define OBD_IOC_LOV_SETSTRIPE	  _IOW ('f', 154, OBD_IOC_DATA_TYPE)
335/* OBD_IOC_LOV_GETSTRIPE: See also LL_IOC_LOV_GETSTRIPE */
336#define OBD_IOC_LOV_GETSTRIPE	  _IOW ('f', 155, OBD_IOC_DATA_TYPE)
337/* OBD_IOC_LOV_SETEA: See also LL_IOC_LOV_SETEA */
338#define OBD_IOC_LOV_SETEA	      _IOW ('f', 156, OBD_IOC_DATA_TYPE)
339/* see <lustre/lustre_user.h> for ioctls 157-159 */
340/* OBD_IOC_QUOTACHECK: See also LL_IOC_QUOTACHECK */
341#define OBD_IOC_QUOTACHECK	     _IOW ('f', 160, int)
342/* OBD_IOC_POLL_QUOTACHECK: See also LL_IOC_POLL_QUOTACHECK */
343#define OBD_IOC_POLL_QUOTACHECK	_IOR ('f', 161, struct if_quotacheck *)
344/* OBD_IOC_QUOTACTL: See also LL_IOC_QUOTACTL */
345#define OBD_IOC_QUOTACTL	       _IOWR('f', 162, struct if_quotactl)
346/* see  also <lustre/lustre_user.h> for ioctls 163-176 */
347#define OBD_IOC_CHANGELOG_REG	  _IOW ('f', 177, struct obd_ioctl_data)
348#define OBD_IOC_CHANGELOG_DEREG	_IOW ('f', 178, struct obd_ioctl_data)
349#define OBD_IOC_CHANGELOG_CLEAR	_IOW ('f', 179, struct obd_ioctl_data)
350#define OBD_IOC_RECORD		 _IOWR('f', 180, OBD_IOC_DATA_TYPE)
351#define OBD_IOC_ENDRECORD	      _IOWR('f', 181, OBD_IOC_DATA_TYPE)
352#define OBD_IOC_PARSE		  _IOWR('f', 182, OBD_IOC_DATA_TYPE)
353#define OBD_IOC_DORECORD	       _IOWR('f', 183, OBD_IOC_DATA_TYPE)
354#define OBD_IOC_PROCESS_CFG	    _IOWR('f', 184, OBD_IOC_DATA_TYPE)
355#define OBD_IOC_DUMP_LOG	       _IOWR('f', 185, OBD_IOC_DATA_TYPE)
356#define OBD_IOC_CLEAR_LOG	      _IOWR('f', 186, OBD_IOC_DATA_TYPE)
357#define OBD_IOC_PARAM		  _IOW ('f', 187, OBD_IOC_DATA_TYPE)
358#define OBD_IOC_POOL		   _IOWR('f', 188, OBD_IOC_DATA_TYPE)
359#define OBD_IOC_REPLACE_NIDS	   _IOWR('f', 189, OBD_IOC_DATA_TYPE)
360
361#define OBD_IOC_CATLOGLIST	     _IOWR('f', 190, OBD_IOC_DATA_TYPE)
362#define OBD_IOC_LLOG_INFO	      _IOWR('f', 191, OBD_IOC_DATA_TYPE)
363#define OBD_IOC_LLOG_PRINT	     _IOWR('f', 192, OBD_IOC_DATA_TYPE)
364#define OBD_IOC_LLOG_CANCEL	    _IOWR('f', 193, OBD_IOC_DATA_TYPE)
365#define OBD_IOC_LLOG_REMOVE	    _IOWR('f', 194, OBD_IOC_DATA_TYPE)
366#define OBD_IOC_LLOG_CHECK	     _IOWR('f', 195, OBD_IOC_DATA_TYPE)
367/* OBD_IOC_LLOG_CATINFO is deprecated */
368#define OBD_IOC_LLOG_CATINFO	   _IOWR('f', 196, OBD_IOC_DATA_TYPE)
369
370#define ECHO_IOC_GET_STRIPE	    _IOWR('f', 200, OBD_IOC_DATA_TYPE)
371#define ECHO_IOC_SET_STRIPE	    _IOWR('f', 201, OBD_IOC_DATA_TYPE)
372#define ECHO_IOC_ENQUEUE	       _IOWR('f', 202, OBD_IOC_DATA_TYPE)
373#define ECHO_IOC_CANCEL		_IOWR('f', 203, OBD_IOC_DATA_TYPE)
374
375#define OBD_IOC_GET_OBJ_VERSION	_IOR('f', 210, OBD_IOC_DATA_TYPE)
376
377/* <lustre/lustre_user.h> defines ioctl number 218-219 */
378#define OBD_IOC_GET_MNTOPT	     _IOW('f', 220, mntopt_t)
379
380#define OBD_IOC_ECHO_MD		_IOR('f', 221, struct obd_ioctl_data)
381#define OBD_IOC_ECHO_ALLOC_SEQ	 _IOWR('f', 222, struct obd_ioctl_data)
382
383#define OBD_IOC_START_LFSCK	       _IOWR('f', 230, OBD_IOC_DATA_TYPE)
384#define OBD_IOC_STOP_LFSCK	       _IOW('f', 231, OBD_IOC_DATA_TYPE)
385#define OBD_IOC_PAUSE_LFSCK	       _IOW('f', 232, OBD_IOC_DATA_TYPE)
386
387/* XXX _IOWR('f', 250, long) has been defined in
388 * libcfs/include/libcfs/libcfs_private.h for debug, don't use it
389 */
390
391/* Until such time as we get_info the per-stripe maximum from the OST,
392 * we define this to be 2T - 4k, which is the ext3 maxbytes. */
393#define LUSTRE_STRIPE_MAXBYTES 0x1fffffff000ULL
394
395/* Special values for remove LOV EA from disk */
396#define LOVEA_DELETE_VALUES(size, count, offset) (size == 0 && count == 0 && \
397						 offset == (typeof(offset))(-1))
398
399/* #define POISON_BULK 0 */
400
401/*
402 * l_wait_event is a flexible sleeping function, permitting simple caller
403 * configuration of interrupt and timeout sensitivity along with actions to
404 * be performed in the event of either exception.
405 *
406 * The first form of usage looks like this:
407 *
408 * struct l_wait_info lwi = LWI_TIMEOUT_INTR(timeout, timeout_handler,
409 *					   intr_handler, callback_data);
410 * rc = l_wait_event(waitq, condition, &lwi);
411 *
412 * l_wait_event() makes the current process wait on 'waitq' until 'condition'
413 * is TRUE or a "killable" signal (SIGTERM, SIKGILL, SIGINT) is pending.  It
414 * returns 0 to signify 'condition' is TRUE, but if a signal wakes it before
415 * 'condition' becomes true, it optionally calls the specified 'intr_handler'
416 * if not NULL, and returns -EINTR.
417 *
418 * If a non-zero timeout is specified, signals are ignored until the timeout
419 * has expired.  At this time, if 'timeout_handler' is not NULL it is called.
420 * If it returns FALSE l_wait_event() continues to wait as described above with
421 * signals enabled.  Otherwise it returns -ETIMEDOUT.
422 *
423 * LWI_INTR(intr_handler, callback_data) is shorthand for
424 * LWI_TIMEOUT_INTR(0, NULL, intr_handler, callback_data)
425 *
426 * The second form of usage looks like this:
427 *
428 * struct l_wait_info lwi = LWI_TIMEOUT(timeout, timeout_handler);
429 * rc = l_wait_event(waitq, condition, &lwi);
430 *
431 * This form is the same as the first except that it COMPLETELY IGNORES
432 * SIGNALS.  The caller must therefore beware that if 'timeout' is zero, or if
433 * 'timeout_handler' is not NULL and returns FALSE, then the ONLY thing that
434 * can unblock the current process is 'condition' becoming TRUE.
435 *
436 * Another form of usage is:
437 * struct l_wait_info lwi = LWI_TIMEOUT_INTERVAL(timeout, interval,
438 *					       timeout_handler);
439 * rc = l_wait_event(waitq, condition, &lwi);
440 * This is the same as previous case, but condition is checked once every
441 * 'interval' jiffies (if non-zero).
442 *
443 * Subtle synchronization point: this macro does *not* necessary takes
444 * wait-queue spin-lock before returning, and, hence, following idiom is safe
445 * ONLY when caller provides some external locking:
446 *
447 *	     Thread1			    Thread2
448 *
449 *   l_wait_event(&obj->wq, ....);				       (1)
450 *
451 *				    wake_up(&obj->wq):		 (2)
452 *					 spin_lock(&q->lock);	  (2.1)
453 *					 __wake_up_common(q, ...);     (2.2)
454 *					 spin_unlock(&q->lock, flags); (2.3)
455 *
456 *   OBD_FREE_PTR(obj);						  (3)
457 *
458 * As l_wait_event() may "short-cut" execution and return without taking
459 * wait-queue spin-lock, some additional synchronization is necessary to
460 * guarantee that step (3) can begin only after (2.3) finishes.
461 *
462 * XXX nikita: some ptlrpc daemon threads have races of that sort.
463 *
464 */
465static inline int back_to_sleep(void *arg)
466{
467	return 0;
468}
469
470#define LWI_ON_SIGNAL_NOOP ((void (*)(void *))(-1))
471
472struct l_wait_info {
473	long lwi_timeout;
474	long lwi_interval;
475	int	    lwi_allow_intr;
476	int  (*lwi_on_timeout)(void *);
477	void (*lwi_on_signal)(void *);
478	void  *lwi_cb_data;
479};
480
481/* NB: LWI_TIMEOUT ignores signals completely */
482#define LWI_TIMEOUT(time, cb, data)	     \
483((struct l_wait_info) {			 \
484	.lwi_timeout    = time,		 \
485	.lwi_on_timeout = cb,		   \
486	.lwi_cb_data    = data,		 \
487	.lwi_interval   = 0,		    \
488	.lwi_allow_intr = 0		     \
489})
490
491#define LWI_TIMEOUT_INTERVAL(time, interval, cb, data)  \
492((struct l_wait_info) {				 \
493	.lwi_timeout    = time,			 \
494	.lwi_on_timeout = cb,			   \
495	.lwi_cb_data    = data,			 \
496	.lwi_interval   = interval,		     \
497	.lwi_allow_intr = 0			     \
498})
499
500#define LWI_TIMEOUT_INTR(time, time_cb, sig_cb, data)   \
501((struct l_wait_info) {				 \
502	.lwi_timeout    = time,			 \
503	.lwi_on_timeout = time_cb,		      \
504	.lwi_on_signal  = sig_cb,		       \
505	.lwi_cb_data    = data,			 \
506	.lwi_interval   = 0,			    \
507	.lwi_allow_intr = 0			     \
508})
509
510#define LWI_TIMEOUT_INTR_ALL(time, time_cb, sig_cb, data)       \
511((struct l_wait_info) {					 \
512	.lwi_timeout    = time,				 \
513	.lwi_on_timeout = time_cb,			      \
514	.lwi_on_signal  = sig_cb,			       \
515	.lwi_cb_data    = data,				 \
516	.lwi_interval   = 0,				    \
517	.lwi_allow_intr = 1				     \
518})
519
520#define LWI_INTR(cb, data)  LWI_TIMEOUT_INTR(0, NULL, cb, data)
521
522#define LUSTRE_FATAL_SIGS (sigmask(SIGKILL) | sigmask(SIGINT) |		\
523			   sigmask(SIGTERM) | sigmask(SIGQUIT) |	\
524			   sigmask(SIGALRM))
525
526
527/*
528 * wait for @condition to become true, but no longer than timeout, specified
529 * by @info.
530 */
531#define __l_wait_event(wq, condition, info, ret, l_add_wait)		   \
532do {									   \
533	wait_queue_t __wait;						 \
534	long __timeout = info->lwi_timeout;			  \
535	sigset_t   __blocked;					      \
536	int   __allow_intr = info->lwi_allow_intr;			     \
537									       \
538	ret = 0;							       \
539	if (condition)							 \
540		break;							 \
541									       \
542	init_waitqueue_entry(&__wait, current);					    \
543	l_add_wait(&wq, &__wait);					      \
544									       \
545	/* Block all signals (just the non-fatal ones if no timeout). */       \
546	if (info->lwi_on_signal != NULL && (__timeout == 0 || __allow_intr))   \
547		__blocked = cfs_block_sigsinv(LUSTRE_FATAL_SIGS);	      \
548	else								   \
549		__blocked = cfs_block_sigsinv(0);			      \
550									       \
551	for (;;) {							     \
552		unsigned       __wstate;				       \
553									       \
554		__wstate = info->lwi_on_signal != NULL &&		      \
555			   (__timeout == 0 || __allow_intr) ?		  \
556			TASK_INTERRUPTIBLE : TASK_UNINTERRUPTIBLE;	       \
557									       \
558		set_current_state(TASK_INTERRUPTIBLE);		 \
559									       \
560		if (condition)						 \
561			break;						 \
562									       \
563		if (__timeout == 0) {					  \
564			schedule();						\
565		} else {						       \
566			long interval = info->lwi_interval?	  \
567					     min_t(long,	     \
568						 info->lwi_interval,__timeout):\
569					     __timeout;			\
570			long remaining = schedule_timeout(interval);\
571			__timeout = cfs_time_sub(__timeout,		    \
572					    cfs_time_sub(interval, remaining));\
573			if (__timeout == 0) {				  \
574				if (info->lwi_on_timeout == NULL ||	    \
575				    info->lwi_on_timeout(info->lwi_cb_data)) { \
576					ret = -ETIMEDOUT;		      \
577					break;				 \
578				}					      \
579				/* Take signals after the timeout expires. */  \
580				if (info->lwi_on_signal != NULL)	       \
581				    (void)cfs_block_sigsinv(LUSTRE_FATAL_SIGS);\
582			}						      \
583		}							      \
584									       \
585		if (condition)						 \
586			break;						 \
587		if (cfs_signal_pending()) {				    \
588			if (info->lwi_on_signal != NULL &&		     \
589			    (__timeout == 0 || __allow_intr)) {		\
590				if (info->lwi_on_signal != LWI_ON_SIGNAL_NOOP) \
591					info->lwi_on_signal(info->lwi_cb_data);\
592				ret = -EINTR;				  \
593				break;					 \
594			}						      \
595			/* We have to do this here because some signals */     \
596			/* are not blockable - ie from strace(1).       */     \
597			/* In these cases we want to schedule_timeout() */     \
598			/* again, because we don't want that to return  */     \
599			/* -EINTR when the RPC actually succeeded.      */     \
600			/* the recalc_sigpending() below will deliver the */     \
601			/* signal properly.			     */     \
602			cfs_clear_sigpending();				\
603		}							      \
604	}								      \
605									       \
606	cfs_restore_sigs(__blocked);					   \
607									       \
608	set_current_state(TASK_RUNNING);			       \
609	remove_wait_queue(&wq, &__wait);					   \
610} while (0)
611
612
613
614#define l_wait_event(wq, condition, info)		       \
615({							      \
616	int		 __ret;			      \
617	struct l_wait_info *__info = (info);		    \
618								\
619	__l_wait_event(wq, condition, __info,		   \
620		       __ret, add_wait_queue);		   \
621	__ret;						  \
622})
623
624#define l_wait_event_exclusive(wq, condition, info)	     \
625({							      \
626	int		 __ret;			      \
627	struct l_wait_info *__info = (info);		    \
628								\
629	__l_wait_event(wq, condition, __info,		   \
630		       __ret, add_wait_queue_exclusive);	 \
631	__ret;						  \
632})
633
634#define l_wait_event_exclusive_head(wq, condition, info)	\
635({							      \
636	int		 __ret;			      \
637	struct l_wait_info *__info = (info);		    \
638								\
639	__l_wait_event(wq, condition, __info,		   \
640		       __ret, add_wait_queue_exclusive_head);    \
641	__ret;						  \
642})
643
644#define l_wait_condition(wq, condition)			 \
645({							      \
646	struct l_wait_info lwi = { 0 };			 \
647	l_wait_event(wq, condition, &lwi);		      \
648})
649
650#define l_wait_condition_exclusive(wq, condition)	       \
651({							      \
652	struct l_wait_info lwi = { 0 };			 \
653	l_wait_event_exclusive(wq, condition, &lwi);	    \
654})
655
656#define l_wait_condition_exclusive_head(wq, condition)	  \
657({							      \
658	struct l_wait_info lwi = { 0 };			 \
659	l_wait_event_exclusive_head(wq, condition, &lwi);       \
660})
661
662#define LIBLUSTRE_CLIENT (0)
663
664/** @} lib */
665
666#endif /* _LUSTRE_LIB_H */
667