xref: /illumos-gate/usr/src/uts/common/fs/smbsrv/smb_oplock.c (revision abdf5d9abf528d6c318fd8533e09bc3cac1f228b)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 /*
26  * SMB Locking library functions.
27  *
28  * You will notice that the functions in this file exit the lock of the session
29  * and reenter it before returning. They even assume that the lock has been
30  * entered in READER mode. The reason for that is a potential deadlock that may
31  * occur when an oplock needs to be broken and the function
32  * smb_session_break_oplock() is called. It should be noticed that the mutex of
33  * the smb node, the oplock of which needs to be broken, is also exited before
34  * calling smb_session_break_oplock(). The reason for that is the same: avoiding
35  * a deadlock. That complexity is due to the fact that the lock of the session
36  * is held during the treatment of a request. That complexity will go away when
37  * that is not the case anymore.
38  */
39 
40 #include <smbsrv/smb_incl.h>
41 #include <smbsrv/smb_fsops.h>
42 #include <inet/tcp.h>
43 
44 static void smb_oplock_wait(smb_node_t *);
45 
46 /*
47  *	Magic		0xFF 'S' 'M' 'B'
48  *	smb_com 	a byte, the "first" command
49  *	Error		a 4-byte union, ignored in a request
50  *	smb_flg		a one byte set of eight flags
51  *	smb_flg2	a two byte set of 16 flags
52  *	.		twelve reserved bytes, have a role
53  *			in connectionless transports (IPX, UDP?)
54  *	smb_tid		a 16-bit tree ID, a mount point sorta,
55  *			0xFFFF is this command does not have
56  *			or require a tree context
57  *	smb_pid		a 16-bit process ID
58  *	smb_uid		a 16-bit user ID, specific to this "session"
59  *			and mapped to a system (bona-fide) UID
60  *	smb_mid		a 16-bit multiplex ID, used to differentiate
61  *			multiple simultaneous requests from the same
62  *			process (pid) (ref RPC "xid")
63  *
64  * SMB_COM_LOCKING_ANDX allows both locking and/or unlocking of file range(s).
65  *
66  *  Client Request                     Description
67  *  ================================== =================================
68  *
69  *  UCHAR WordCount;                   Count of parameter words = 8
70  *  UCHAR AndXCommand;                 Secondary (X) command;  0xFF = none
71  *  UCHAR AndXReserved;                Reserved (must be 0)
72  *  USHORT AndXOffset;                 Offset to next command WordCount
73  *  USHORT Fid;                        File handle
74  *  UCHAR LockType;                    See LockType table below
75  *  UCHAR OplockLevel;                 The new oplock level
76  *  ULONG Timeout;                     Milliseconds to wait for unlock
77  *  USHORT NumberOfUnlocks;            Num. unlock range structs following
78  *  USHORT NumberOfLocks;              Num. lock range structs following
79  *  USHORT ByteCount;                  Count of data bytes
80  *  LOCKING_ANDX_RANGE Unlocks[];      Unlock ranges
81  *  LOCKING_ANDX_RANGE Locks[];        Lock ranges
82  *
83  *  LockType Flag Name            Value Description
84  *  ============================  ===== ================================
85  *
86  *  LOCKING_ANDX_SHARED_LOCK      0x01  Read-only lock
87  *  LOCKING_ANDX_OPLOCK_RELEASE   0x02  Oplock break notification
88  *  LOCKING_ANDX_CHANGE_LOCKTYPE  0x04  Change lock type
89  *  LOCKING_ANDX_CANCEL_LOCK      0x08  Cancel outstanding request
90  *  LOCKING_ANDX_LARGE_FILES      0x10  Large file locking format
91  *
92  *  LOCKING_ANDX_RANGE Format
93  *  =====================================================================
94  *
95  *  USHORT Pid;                        PID of process "owning" lock
96  *  ULONG Offset;                      Offset to bytes to [un]lock
97  *  ULONG Length;                      Number of bytes to [un]lock
98  *
99  *  Large File LOCKING_ANDX_RANGE Format
100  *  =====================================================================
101  *
102  *  USHORT Pid;                        PID of process "owning" lock
103  *  USHORT Pad;                        Pad to DWORD align (mbz)
104  *  ULONG OffsetHigh;                  Offset to bytes to [un]lock
105  *                                      (high)
106  *  ULONG OffsetLow;                   Offset to bytes to [un]lock (low)
107  *  ULONG LengthHigh;                  Number of bytes to [un]lock
108  *                                      (high)
109  *  ULONG LengthLow;                   Number of bytes to [un]lock (low)
110  *
111  *  Server Response                    Description
112  *  ================================== =================================
113  *
114  *  UCHAR WordCount;                   Count of parameter words = 2
115  *  UCHAR AndXCommand;                 Secondary (X) command;  0xFF =
116  *                                      none
117  *  UCHAR AndXReserved;                Reserved (must be 0)
118  *  USHORT AndXOffset;                 Offset to next command WordCount
119  *  USHORT ByteCount;                  Count of data bytes = 0
120  *
121  */
122 
123 /*
124  * smb_oplock_acquire
125  *
126  * Attempt to acquire an oplock. Note that the oplock granted may be
127  * none, i.e. the oplock was not granted. The result of the acquisition is
128  * provided in ol->ol_level.
129  *
130  * Grant an oplock to the requestor if this session is the only one
131  * that has the file open, regardless of the number of instances of
132  * the file opened by this session.
133  *
134  * However, if there is no oplock on this file and there is already
135  * at least one open, we will not grant an oplock, even if the only
136  * existing opens are from the same client.  This is "server discretion."
137  *
138  * An oplock may need to be broken in order for one to be granted, and
139  * depending on what action is taken by the other client (unlock or close),
140  * an oplock may or may not be granted.  (The breaking of an oplock is
141  * done earlier in the calling path.)
142  */
143 void
144 smb_oplock_acquire(smb_node_t *node, smb_ofile_t *of, open_param_t *op)
145 {
146 	smb_session_t	*session;
147 	smb_oplock_t	*ol;
148 	clock_t		time;
149 
150 	SMB_NODE_VALID(node);
151 	SMB_OFILE_VALID(of);
152 
153 	ASSERT(node == SMB_OFILE_GET_NODE(of));
154 
155 	session = SMB_OFILE_GET_SESSION(of);
156 
157 	if (!smb_session_oplocks_enable(session) ||
158 	    smb_tree_has_feature(SMB_OFILE_GET_TREE(of), SMB_TREE_NO_OPLOCKS)) {
159 		op->op_oplock_level = SMB_OPLOCK_NONE;
160 		return;
161 	}
162 
163 	ol = &node->n_oplock;
164 	time = MSEC_TO_TICK(smb_oplock_timeout) + ddi_get_lbolt();
165 
166 	mutex_enter(&node->n_mutex);
167 
168 	switch (node->n_state) {
169 	case SMB_NODE_STATE_OPLOCK_GRANTED:
170 		if (SMB_SESSION_GET_ID(session) == ol->ol_sess_id) {
171 			mutex_exit(&node->n_mutex);
172 			return;
173 		}
174 		break;
175 	case SMB_NODE_STATE_AVAILABLE:
176 	case SMB_NODE_STATE_OPLOCK_BREAKING:
177 		break;
178 	default:
179 		SMB_PANIC();
180 	}
181 
182 	for (;;) {
183 		int	rc;
184 
185 		smb_oplock_wait(node);
186 
187 		if (node->n_state == SMB_NODE_STATE_AVAILABLE) {
188 			if ((op->op_oplock_level == SMB_OPLOCK_LEVEL_II) ||
189 			    (op->op_oplock_level == SMB_OPLOCK_NONE) ||
190 			    (node->n_open_count > 1)) {
191 				mutex_exit(&node->n_mutex);
192 				op->op_oplock_level = SMB_OPLOCK_NONE;
193 				return;
194 			}
195 			ol->ol_ofile = of;
196 			ol->ol_sess_id = SMB_SESSION_GET_ID(session);
197 			ol->ol_level = op->op_oplock_level;
198 			ol->ol_xthread = curthread;
199 			node->n_state = SMB_NODE_STATE_OPLOCK_GRANTED;
200 			mutex_exit(&node->n_mutex);
201 			if (smb_fsop_oplock_install(node, of->f_mode) == 0) {
202 				smb_ofile_set_oplock_granted(of);
203 				return;
204 			}
205 			mutex_enter(&node->n_mutex);
206 			ASSERT(node->n_state == SMB_NODE_STATE_OPLOCK_GRANTED);
207 			node->n_state = SMB_NODE_STATE_AVAILABLE;
208 			ol->ol_xthread = NULL;
209 			op->op_oplock_level = SMB_OPLOCK_NONE;
210 			cv_broadcast(&ol->ol_cv);
211 			break;
212 		}
213 
214 		if (node->n_state == SMB_NODE_STATE_OPLOCK_GRANTED) {
215 			if (SMB_SESSION_GET_ID(session) == ol->ol_sess_id)
216 				break;
217 			node->n_state = SMB_NODE_STATE_OPLOCK_BREAKING;
218 			mutex_exit(&node->n_mutex);
219 			smb_session_oplock_break(
220 			    SMB_OFILE_GET_SESSION(ol->ol_ofile), ol->ol_ofile);
221 			mutex_enter(&node->n_mutex);
222 			continue;
223 		}
224 
225 		ASSERT(node->n_state == SMB_NODE_STATE_OPLOCK_BREAKING);
226 
227 		rc = cv_timedwait(&ol->ol_cv, &node->n_mutex, time);
228 
229 		if (rc == -1) {
230 			/*
231 			 * Oplock release timed out.
232 			 */
233 			if (node->n_state == SMB_NODE_STATE_OPLOCK_BREAKING) {
234 				node->n_state = SMB_NODE_STATE_AVAILABLE;
235 				ol->ol_xthread = curthread;
236 				mutex_exit(&node->n_mutex);
237 				smb_fsop_oplock_uninstall(node);
238 				mutex_enter(&node->n_mutex);
239 				ol->ol_xthread = NULL;
240 				cv_broadcast(&ol->ol_cv);
241 			}
242 		}
243 	}
244 	mutex_exit(&node->n_mutex);
245 }
246 
247 /*
248  * smb_oplock_break
249  *
250  * The oplock break may succeed for multiple reasons: file close, oplock
251  * release, holder connection dropped, requesting client disconnect etc.
252  *
253  * Returns:
254  *
255  *	B_TRUE	The oplock is broken.
256  *	B_FALSE	The oplock is being broken. This is returned if nowait is set
257  *		to B_TRUE;
258  */
259 boolean_t
260 smb_oplock_break(smb_node_t *node, smb_session_t *session, boolean_t nowait)
261 {
262 	smb_oplock_t	*ol;
263 	clock_t		time;
264 
265 	SMB_NODE_VALID(node);
266 	ol = &node->n_oplock;
267 	time = MSEC_TO_TICK(smb_oplock_timeout) + ddi_get_lbolt();
268 
269 	if (session != NULL) {
270 		mutex_enter(&node->n_mutex);
271 		if (SMB_SESSION_GET_ID(session) == ol->ol_sess_id) {
272 			mutex_exit(&node->n_mutex);
273 			return (B_TRUE);
274 		}
275 	} else {
276 		mutex_enter(&node->n_mutex);
277 	}
278 
279 	for (;;) {
280 		int	rc;
281 
282 		smb_oplock_wait(node);
283 
284 		if (node->n_state == SMB_NODE_STATE_AVAILABLE) {
285 			mutex_exit(&node->n_mutex);
286 			return (B_TRUE);
287 		}
288 
289 		if (node->n_state == SMB_NODE_STATE_OPLOCK_GRANTED) {
290 			node->n_state = SMB_NODE_STATE_OPLOCK_BREAKING;
291 			mutex_exit(&node->n_mutex);
292 			smb_session_oplock_break(
293 			    SMB_OFILE_GET_SESSION(ol->ol_ofile), ol->ol_ofile);
294 			mutex_enter(&node->n_mutex);
295 			continue;
296 		}
297 
298 		ASSERT(node->n_state == SMB_NODE_STATE_OPLOCK_BREAKING);
299 		if (nowait) {
300 			mutex_exit(&node->n_mutex);
301 			return (B_FALSE);
302 		}
303 		rc = cv_timedwait(&ol->ol_cv, &node->n_mutex, time);
304 		if (rc == -1) {
305 			/*
306 			 * Oplock release timed out.
307 			 */
308 			if (node->n_state == SMB_NODE_STATE_OPLOCK_BREAKING) {
309 				node->n_state = SMB_NODE_STATE_AVAILABLE;
310 				ol->ol_xthread = curthread;
311 				mutex_exit(&node->n_mutex);
312 				smb_fsop_oplock_uninstall(node);
313 				mutex_enter(&node->n_mutex);
314 				ol->ol_xthread = NULL;
315 				cv_broadcast(&ol->ol_cv);
316 				break;
317 			}
318 		}
319 	}
320 	mutex_exit(&node->n_mutex);
321 	return (B_TRUE);
322 }
323 
324 /*
325  * smb_oplock_release
326  *
327  * This function releases the oplock on the node passed in. If other threads
328  * were waiting for the oplock to be released they are signaled.
329  */
330 void
331 smb_oplock_release(smb_node_t *node, smb_ofile_t *of)
332 {
333 	smb_oplock_t	*ol;
334 
335 	SMB_NODE_VALID(node);
336 	ol = &node->n_oplock;
337 
338 	mutex_enter(&node->n_mutex);
339 	smb_oplock_wait(node);
340 	switch (node->n_state) {
341 	case SMB_NODE_STATE_AVAILABLE:
342 		break;
343 
344 	case SMB_NODE_STATE_OPLOCK_GRANTED:
345 	case SMB_NODE_STATE_OPLOCK_BREAKING:
346 		if (ol->ol_ofile == of) {
347 			node->n_state = SMB_NODE_STATE_AVAILABLE;
348 			ol->ol_xthread = curthread;
349 			mutex_exit(&node->n_mutex);
350 			smb_fsop_oplock_uninstall(node);
351 			mutex_enter(&node->n_mutex);
352 			ol->ol_xthread = NULL;
353 			cv_broadcast(&ol->ol_cv);
354 		}
355 		break;
356 
357 	default:
358 		SMB_PANIC();
359 	}
360 	mutex_exit(&node->n_mutex);
361 }
362 
363 /*
364  * smb_oplock_conflict
365  *
366  * The two checks on "session" and "op" are primarily for the open path.
367  * Other SMB functions may call smb_oplock_conflict() with a session
368  * pointer so as to do the session check.
369  */
370 boolean_t
371 smb_oplock_conflict(smb_node_t *node, smb_session_t *session, open_param_t *op)
372 {
373 	boolean_t	rb;
374 
375 	SMB_NODE_VALID(node);
376 	SMB_SESSION_VALID(session);
377 
378 	mutex_enter(&node->n_mutex);
379 	smb_oplock_wait(node);
380 	switch (node->n_state) {
381 	case SMB_NODE_STATE_AVAILABLE:
382 		rb = B_FALSE;
383 		break;
384 
385 	case SMB_NODE_STATE_OPLOCK_GRANTED:
386 	case SMB_NODE_STATE_OPLOCK_BREAKING:
387 		if (SMB_SESSION_GET_ID(session) == node->n_oplock.ol_sess_id) {
388 			rb = B_FALSE;
389 			break;
390 		}
391 
392 		if (op != NULL) {
393 			if (((op->desired_access & ~(FILE_READ_ATTRIBUTES |
394 			    FILE_WRITE_ATTRIBUTES | SYNCHRONIZE)) == 0) &&
395 			    (op->create_disposition != FILE_SUPERSEDE) &&
396 			    (op->create_disposition != FILE_OVERWRITE)) {
397 				/* Attributs only */
398 				rb = B_FALSE;
399 				break;
400 			}
401 		}
402 		rb = B_TRUE;
403 		break;
404 
405 	default:
406 		SMB_PANIC();
407 	}
408 	mutex_exit(&node->n_mutex);
409 	return (rb);
410 }
411 
412 /*
413  * smb_oplock_broadcast
414  *
415  * The the calling thread has the pointer to its context stored in ol_thread
416  * it resets that field. If any other thread is waiting for that field to
417  * turn to NULL it is signaled.
418  *
419  * Returns:
420  *	B_TRUE	Oplock unlocked
421  *	B_FALSE	Oplock still locked
422  */
423 boolean_t
424 smb_oplock_broadcast(smb_node_t *node)
425 {
426 	smb_oplock_t	*ol;
427 	boolean_t	rb;
428 
429 	SMB_NODE_VALID(node);
430 	ol = &node->n_oplock;
431 	rb = B_FALSE;
432 
433 	mutex_enter(&node->n_mutex);
434 	if ((ol->ol_xthread != NULL) && (ol->ol_xthread == curthread)) {
435 		ol->ol_xthread = NULL;
436 		cv_broadcast(&ol->ol_cv);
437 		rb = B_TRUE;
438 	}
439 	mutex_exit(&node->n_mutex);
440 	return (rb);
441 }
442 
443 /*
444  * smb_oplock_wait
445  *
446  * The mutex of the node must have been entered before calling this function.
447  * If the field ol_xthread is not NULL and doesn't contain the pointer to the
448  * context of the calling thread, the caller will sleep until that field is
449  * reset (set to NULL).
450  */
451 static void
452 smb_oplock_wait(smb_node_t *node)
453 {
454 	smb_oplock_t	*ol = &node->n_oplock;
455 
456 	if ((ol->ol_xthread != NULL) && (ol->ol_xthread != curthread)) {
457 		ASSERT(!MUTEX_HELD(&ol->ol_ofile->f_mutex));
458 		while (ol->ol_xthread != NULL)
459 			cv_wait(&ol->ol_cv, &node->n_mutex);
460 	}
461 }
462