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) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2013, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/include/lustre_lib.h
38 * Basic Lustre library routines.
49 #include <libcfs/libcfs.h>
50 #include <lustre/lustre_idl.h>
51 #include <lustre_ver.h>
52 #include <lustre_cfg.h>
53 #include <linux/lustre_lib.h>
57 struct ptlrpc_request;
61 #include <lustre_ha.h>
62 #include <lustre_net.h>
64 #ifdef HAVE_SERVER_SUPPORT
65 void target_client_add_cb(struct obd_device *obd, __u64 transno, void *cb_data,
67 int target_handle_connect(struct ptlrpc_request *req);
68 int target_handle_disconnect(struct ptlrpc_request *req);
69 void target_destroy_export(struct obd_export *exp);
70 int target_handle_ping(struct ptlrpc_request *req);
71 void target_committed_to_req(struct ptlrpc_request *req);
72 void target_cancel_recovery_timer(struct obd_device *obd);
73 void target_stop_recovery_thread(struct obd_device *obd);
74 void target_cleanup_recovery(struct obd_device *obd);
75 int target_queue_recovery_request(struct ptlrpc_request *req,
76 struct obd_device *obd);
77 int target_bulk_io(struct obd_export *exp, struct ptlrpc_bulk_desc *desc,
78 struct l_wait_info *lwi);
81 int target_pack_pool_reply(struct ptlrpc_request *req);
82 int do_set_info_async(struct obd_import *imp,
83 int opcode, int version,
84 obd_count keylen, void *key,
85 obd_count vallen, void *val,
86 struct ptlrpc_request_set *set);
88 #define OBD_RECOVERY_MAX_TIME (obd_timeout * 18) /* b13079 */
90 void target_send_reply(struct ptlrpc_request *req, int rc, int fail_id);
94 int client_sanobd_setup(struct obd_device *obddev, struct lustre_cfg* lcfg);
95 struct client_obd *client_conn2cli(struct lustre_handle *conn);
98 struct obd_client_handle {
99 struct lustre_handle och_fh;
100 struct lu_fid och_fid;
101 struct md_open_data *och_mod;
102 struct lustre_handle och_lease_handle; /* open lock for lease */
106 #define OBD_CLIENT_HANDLE_MAGIC 0xd15ea5ed
109 void statfs_pack(struct obd_statfs *osfs, struct kstatfs *sfs);
110 void statfs_unpack(struct kstatfs *sfs, struct obd_statfs *osfs);
112 /* Until such time as we get_info the per-stripe maximum from the OST,
113 * we define this to be 2T - 4k, which is the ext3 maxbytes. */
114 #define LUSTRE_STRIPE_MAXBYTES 0x1fffffff000ULL
116 /* Special values for remove LOV EA from disk */
117 #define LOVEA_DELETE_VALUES(size, count, offset) (size == 0 && count == 0 && \
118 offset == (typeof(offset))(-1))
120 #define LMVEA_DELETE_VALUES(count, offset) ((count) == 0 && \
121 (offset) == (typeof(offset))(-1))
122 /* #define POISON_BULK 0 */
125 * l_wait_event is a flexible sleeping function, permitting simple caller
126 * configuration of interrupt and timeout sensitivity along with actions to
127 * be performed in the event of either exception.
129 * The first form of usage looks like this:
131 * struct l_wait_info lwi = LWI_TIMEOUT_INTR(timeout, timeout_handler,
132 * intr_handler, callback_data);
133 * rc = l_wait_event(waitq, condition, &lwi);
135 * l_wait_event() makes the current process wait on 'waitq' until 'condition'
136 * is TRUE or a "killable" signal (SIGTERM, SIKGILL, SIGINT) is pending. It
137 * returns 0 to signify 'condition' is TRUE, but if a signal wakes it before
138 * 'condition' becomes true, it optionally calls the specified 'intr_handler'
139 * if not NULL, and returns -EINTR.
141 * If a non-zero timeout is specified, signals are ignored until the timeout
142 * has expired. At this time, if 'timeout_handler' is not NULL it is called.
143 * If it returns FALSE l_wait_event() continues to wait as described above with
144 * signals enabled. Otherwise it returns -ETIMEDOUT.
146 * LWI_INTR(intr_handler, callback_data) is shorthand for
147 * LWI_TIMEOUT_INTR(0, NULL, intr_handler, callback_data)
149 * The second form of usage looks like this:
151 * struct l_wait_info lwi = LWI_TIMEOUT(timeout, timeout_handler);
152 * rc = l_wait_event(waitq, condition, &lwi);
154 * This form is the same as the first except that it COMPLETELY IGNORES
155 * SIGNALS. The caller must therefore beware that if 'timeout' is zero, or if
156 * 'timeout_handler' is not NULL and returns FALSE, then the ONLY thing that
157 * can unblock the current process is 'condition' becoming TRUE.
159 * Another form of usage is:
160 * struct l_wait_info lwi = LWI_TIMEOUT_INTERVAL(timeout, interval,
162 * rc = l_wait_event(waitq, condition, &lwi);
163 * This is the same as previous case, but condition is checked once every
164 * 'interval' jiffies (if non-zero).
166 * Subtle synchronization point: this macro does *not* necessary takes
167 * wait-queue spin-lock before returning, and, hence, following idiom is safe
168 * ONLY when caller provides some external locking:
172 * l_wait_event(&obj->wq, ....); (1)
174 * wake_up(&obj->wq): (2)
175 * spin_lock(&q->lock); (2.1)
176 * __wake_up_common(q, ...); (2.2)
177 * spin_unlock(&q->lock, flags); (2.3)
179 * OBD_FREE_PTR(obj); (3)
181 * As l_wait_event() may "short-cut" execution and return without taking
182 * wait-queue spin-lock, some additional synchronization is necessary to
183 * guarantee that step (3) can begin only after (2.3) finishes.
185 * XXX nikita: some ptlrpc daemon threads have races of that sort.
188 static inline int back_to_sleep(void *arg)
193 #define LWI_ON_SIGNAL_NOOP ((void (*)(void *))(-1))
196 cfs_duration_t lwi_timeout;
197 cfs_duration_t lwi_interval;
199 int (*lwi_on_timeout)(void *);
200 void (*lwi_on_signal)(void *);
204 /* NB: LWI_TIMEOUT ignores signals completely */
205 #define LWI_TIMEOUT(time, cb, data) \
206 ((struct l_wait_info) { \
207 .lwi_timeout = time, \
208 .lwi_on_timeout = cb, \
209 .lwi_cb_data = data, \
211 .lwi_allow_intr = 0 \
214 #define LWI_TIMEOUT_INTERVAL(time, interval, cb, data) \
215 ((struct l_wait_info) { \
216 .lwi_timeout = time, \
217 .lwi_on_timeout = cb, \
218 .lwi_cb_data = data, \
219 .lwi_interval = interval, \
220 .lwi_allow_intr = 0 \
223 #define LWI_TIMEOUT_INTR(time, time_cb, sig_cb, data) \
224 ((struct l_wait_info) { \
225 .lwi_timeout = time, \
226 .lwi_on_timeout = time_cb, \
227 .lwi_on_signal = sig_cb, \
228 .lwi_cb_data = data, \
230 .lwi_allow_intr = 0 \
233 #define LWI_TIMEOUT_INTR_ALL(time, time_cb, sig_cb, data) \
234 ((struct l_wait_info) { \
235 .lwi_timeout = time, \
236 .lwi_on_timeout = time_cb, \
237 .lwi_on_signal = sig_cb, \
238 .lwi_cb_data = data, \
240 .lwi_allow_intr = 1 \
243 #define LWI_INTR(cb, data) LWI_TIMEOUT_INTR(0, NULL, cb, data)
247 * wait for @condition to become true, but no longer than timeout, specified
250 #define __l_wait_event(wq, condition, info, ret, l_add_wait) \
252 wait_queue_t __wait; \
253 cfs_duration_t __timeout = info->lwi_timeout; \
254 sigset_t __blocked; \
255 int __allow_intr = info->lwi_allow_intr; \
261 init_waitqueue_entry_current(&__wait); \
262 l_add_wait(&wq, &__wait); \
264 /* Block all signals (just the non-fatal ones if no timeout). */ \
265 if (info->lwi_on_signal != NULL && (__timeout == 0 || __allow_intr)) \
266 __blocked = cfs_block_sigsinv(LUSTRE_FATAL_SIGS); \
268 __blocked = cfs_block_sigsinv(0); \
273 __wstate = info->lwi_on_signal != NULL && \
274 (__timeout == 0 || __allow_intr) ? \
275 TASK_INTERRUPTIBLE : TASK_UNINTERRUPTIBLE; \
277 set_current_state(TASK_INTERRUPTIBLE); \
282 if (__timeout == 0) { \
283 waitq_wait(&__wait, __wstate); \
285 cfs_duration_t interval = info->lwi_interval? \
286 min_t(cfs_duration_t, \
287 info->lwi_interval,__timeout):\
289 cfs_duration_t remaining = waitq_timedwait(&__wait, \
292 __timeout = cfs_time_sub(__timeout, \
293 cfs_time_sub(interval, remaining));\
294 if (__timeout == 0) { \
295 if (info->lwi_on_timeout == NULL || \
296 info->lwi_on_timeout(info->lwi_cb_data)) { \
300 /* Take signals after the timeout expires. */ \
301 if (info->lwi_on_signal != NULL) \
302 (void)cfs_block_sigsinv(LUSTRE_FATAL_SIGS);\
308 if (cfs_signal_pending()) { \
309 if (info->lwi_on_signal != NULL && \
310 (__timeout == 0 || __allow_intr)) { \
311 if (info->lwi_on_signal != LWI_ON_SIGNAL_NOOP) \
312 info->lwi_on_signal(info->lwi_cb_data);\
316 /* We have to do this here because some signals */ \
317 /* are not blockable - ie from strace(1). */ \
318 /* In these cases we want to schedule_timeout() */ \
319 /* again, because we don't want that to return */ \
320 /* -EINTR when the RPC actually succeeded. */ \
321 /* the recalc_sigpending() below will deliver the */ \
322 /* signal properly. */ \
323 cfs_clear_sigpending(); \
327 cfs_restore_sigs(__blocked); \
329 set_current_state(TASK_RUNNING); \
330 remove_wait_queue(&wq, &__wait); \
335 #define l_wait_event(wq, condition, info) \
338 struct l_wait_info *__info = (info); \
340 __l_wait_event(wq, condition, __info, \
341 __ret, add_wait_queue); \
345 #define l_wait_event_exclusive(wq, condition, info) \
348 struct l_wait_info *__info = (info); \
350 __l_wait_event(wq, condition, __info, \
351 __ret, add_wait_queue_exclusive); \
355 #define l_wait_event_exclusive_head(wq, condition, info) \
358 struct l_wait_info *__info = (info); \
360 __l_wait_event(wq, condition, __info, \
361 __ret, add_wait_queue_exclusive_head); \
365 #define l_wait_condition(wq, condition) \
367 struct l_wait_info lwi = { 0 }; \
368 l_wait_event(wq, condition, &lwi); \
371 #define l_wait_condition_exclusive(wq, condition) \
373 struct l_wait_info lwi = { 0 }; \
374 l_wait_event_exclusive(wq, condition, &lwi); \
377 #define l_wait_condition_exclusive_head(wq, condition) \
379 struct l_wait_info lwi = { 0 }; \
380 l_wait_event_exclusive_head(wq, condition, &lwi); \
385 #endif /* _LUSTRE_LIB_H */