]>
Commit | Line | Data |
---|---|---|
a998dd70 | 1 | /* flock.cc. NT specific implementation of advisory file locking. |
f3a1e23e | 2 | |
3c53eaeb | 3 | Copyright 2003, 2008, 2009, 2010, 2011, 2012 Red Hat, Inc. |
f3a1e23e CV |
4 | |
5 | This file is part of Cygwin. | |
6 | ||
7 | This software is a copyrighted work licensed under the terms of the | |
8 | Cygwin license. Please consult the file "CYGWIN_LICENSE" for | |
9 | details. */ | |
10 | ||
a998dd70 CV |
11 | /* The basic mechanism as well as the datastructures used in the below |
12 | implementation are taken from the FreeBSD repository on 2008-03-18. | |
13 | The essential code of the lf_XXX functions has been taken from the | |
14 | module src/sys/kern/kern_lockf.c. It has been adapted to use NT | |
15 | global namespace subdirs and event objects for synchronization | |
025c1fac CF |
16 | purposes. |
17 | ||
a998dd70 CV |
18 | So, the following copyright applies to most of the code in the lf_XXX |
19 | functions. | |
20 | ||
21 | * Copyright (c) 1982, 1986, 1989, 1993 | |
22 | * The Regents of the University of California. All rights reserved. | |
23 | * | |
24 | * This code is derived from software contributed to Berkeley by | |
25 | * Scooter Morris at Genentech Inc. | |
26 | * | |
27 | * Redistribution and use in source and binary forms, with or without | |
28 | * modification, are permitted provided that the following conditions | |
29 | * are met: | |
30 | * 1. Redistributions of source code must retain the above copyright | |
31 | * notice, this list of conditions and the following disclaimer. | |
32 | * 2. Redistributions in binary form must reproduce the above copyright | |
33 | * notice, this list of conditions and the following disclaimer in the | |
34 | * documentation and/or other materials provided with the distribution. | |
35 | * 4. Neither the name of the University nor the names of its contributors | |
36 | * may be used to endorse or promote products derived from this software | |
37 | * without specific prior written permission. | |
38 | * | |
39 | * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND | |
40 | * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE | |
41 | * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE | |
42 | * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE | |
43 | * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL | |
44 | * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS | |
45 | * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) | |
46 | * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT | |
47 | * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY | |
48 | * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF | |
49 | * SUCH DAMAGE. | |
50 | * | |
51 | * @(#)ufs_lockf.c 8.3 (Berkeley) 1/6/94 | |
025c1fac | 52 | */ |
a998dd70 CV |
53 | |
54 | /* | |
55 | * The flock() function is based upon source taken from the Red Hat | |
56 | * implementation used in their imap-2002d SRPM. | |
57 | * | |
58 | * $RH: flock.c,v 1.2 2000/08/23 17:07:00 nalin Exp $ | |
59 | */ | |
60 | ||
fabfb1a1 | 61 | /* The lockf function is based upon FreeBSD sources with the following |
a998dd70 | 62 | * copyright. |
8c0b8bc3 CV |
63 | */ |
64 | /* | |
a998dd70 CV |
65 | * Copyright (c) 1997 The NetBSD Foundation, Inc. |
66 | * All rights reserved. | |
67 | * | |
68 | * This code is derived from software contributed to The NetBSD Foundation | |
69 | * by Klaus Klein. | |
70 | * | |
71 | * Redistribution and use in source and binary forms, with or without | |
72 | * modification, are permitted provided that the following conditions | |
73 | * are met: | |
74 | * 1. Redistributions of source code must retain the above copyright | |
75 | * notice, this list of conditions and the following disclaimer. | |
76 | * 2. Redistributions in binary form must reproduce the above copyright | |
77 | * notice, this list of conditions and the following disclaimer in the | |
78 | * documentation and/or other materials provided with the distribution. | |
a998dd70 CV |
79 | * |
80 | * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS | |
81 | * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED | |
82 | * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR | |
83 | * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS | |
84 | * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR | |
85 | * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF | |
86 | * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS | |
87 | * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN | |
88 | * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) | |
89 | * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE | |
90 | * POSSIBILITY OF SUCH DAMAGE. | |
8c0b8bc3 | 91 | */ |
a998dd70 | 92 | |
e2a39e2e | 93 | #include "winsup.h" |
a998dd70 | 94 | #include <assert.h> |
f3a1e23e | 95 | #include <sys/file.h> |
f3a1e23e | 96 | #include <unistd.h> |
a998dd70 CV |
97 | #include <stdlib.h> |
98 | #include "cygerrno.h" | |
a998dd70 | 99 | #include "security.h" |
db5ae618 | 100 | #include "shared_info.h" |
a998dd70 CV |
101 | #include "path.h" |
102 | #include "fhandler.h" | |
103 | #include "dtable.h" | |
104 | #include "cygheap.h" | |
a998dd70 CV |
105 | #include "pinfo.h" |
106 | #include "sigproc.h" | |
107 | #include "cygtls.h" | |
30bbc55e | 108 | #include "tls_pbuf.h" |
f4ec8743 | 109 | #include "miscfuncs.h" |
a998dd70 CV |
110 | #include "ntdll.h" |
111 | #include <sys/queue.h> | |
112 | #include <wchar.h> | |
113 | ||
a998dd70 CV |
114 | #define F_WAIT 0x10 /* Wait until lock is granted */ |
115 | #define F_FLOCK 0x20 /* Use flock(2) semantics for lock */ | |
116 | #define F_POSIX 0x40 /* Use POSIX semantics for lock */ | |
117 | ||
118 | #ifndef OFF_MAX | |
119 | #define OFF_MAX LLONG_MAX | |
120 | #endif | |
121 | ||
122 | static NO_COPY muto lockf_guard; | |
123 | ||
124 | #define INODE_LIST_LOCK() (lockf_guard.init ("lockf_guard")->acquire ()) | |
125 | #define INODE_LIST_UNLOCK() (lockf_guard.release ()) | |
126 | ||
dbf576fd CV |
127 | #define LOCK_DIR_NAME_FMT L"flock-%08x-%016X" |
128 | #define LOCK_DIR_NAME_LEN 31 /* Length of the resulting name */ | |
129 | #define LOCK_DIR_NAME_DEV_OFF 6 /* Offset to device number */ | |
130 | #define LOCK_DIR_NAME_INO_OFF 15 /* Offset to inode number */ | |
131 | ||
132 | /* Don't change format without also changing lockf_t::from_obj_name! */ | |
133 | #define LOCK_OBJ_NAME_FMT L"%02x-%01x-%016X-%016X-%016X-%08x-%04x" | |
134 | #define LOCK_OBJ_NAME_LEN 69 /* Length of the resulting name */ | |
a998dd70 | 135 | |
db5ae618 CV |
136 | #define FLOCK_INODE_DIR_ACCESS (DIRECTORY_QUERY \ |
137 | | DIRECTORY_TRAVERSE \ | |
138 | | DIRECTORY_CREATE_OBJECT \ | |
139 | | READ_CONTROL) | |
140 | ||
db5ae618 CV |
141 | #define FLOCK_EVENT_ACCESS (EVENT_QUERY_STATE \ |
142 | | SYNCHRONIZE \ | |
143 | | READ_CONTROL) | |
144 | ||
a998dd70 | 145 | /* This function takes the own process security descriptor DACL and adds |
b9e2579c CV |
146 | SYNCHRONIZE permissions for everyone. This allows all processes |
147 | to wait for this process to die when blocking in a F_SETLKW on a lock | |
148 | which is hold by this process. */ | |
a998dd70 CV |
149 | static void |
150 | allow_others_to_sync () | |
151 | { | |
152 | static NO_COPY bool done; | |
153 | ||
154 | if (done) | |
155 | return; | |
156 | ||
157 | NTSTATUS status; | |
158 | PACL dacl; | |
159 | LPVOID ace; | |
160 | ULONG len; | |
161 | ||
30bbc55e CV |
162 | /* Get this process DACL. We use a rather small stack buffer here which |
163 | should be more than sufficient for process ACLs. Can't use tls functions | |
164 | at this point because this gets called during initialization when the tls | |
165 | is not really available. */ | |
4e8f539f | 166 | #define MAX_PROCESS_SD_SIZE 3072 |
ca7f6a04 | 167 | PISECURITY_DESCRIPTOR sd = (PISECURITY_DESCRIPTOR) alloca (MAX_PROCESS_SD_SIZE); |
f16706de CV |
168 | status = NtQuerySecurityObject (NtCurrentProcess (), |
169 | DACL_SECURITY_INFORMATION, sd, | |
4e8f539f | 170 | MAX_PROCESS_SD_SIZE, &len); |
a998dd70 CV |
171 | if (!NT_SUCCESS (status)) |
172 | { | |
61522196 | 173 | debug_printf ("NtQuerySecurityObject: %y", status); |
a998dd70 CV |
174 | return; |
175 | } | |
30bbc55e | 176 | /* Create a valid dacl pointer and set its size to be as big as |
a998dd70 CV |
177 | there's room in the temporary buffer. Note that the descriptor |
178 | is in self-relative format. */ | |
b4ad7197 CV |
179 | BOOLEAN present, defaulted; |
180 | RtlGetDaclSecurityDescriptor (sd, &present, &dacl, &defaulted); | |
56bed670 | 181 | if (!present) /* If so, dacl has undefined value. */ |
b4ad7197 CV |
182 | { |
183 | dacl = (PACL) (sd + 1); | |
184 | RtlCreateAcl (dacl, MAX_PROCESS_SD_SIZE - sizeof *sd, ACL_REVISION); | |
185 | } | |
56bed670 CV |
186 | else if (dacl == NULL) /* Everyone has all access anyway */ |
187 | { | |
188 | done = true; | |
189 | return; | |
190 | } | |
b4ad7197 CV |
191 | else |
192 | { | |
193 | dacl->AclSize = MAX_PROCESS_SD_SIZE - ((PBYTE) dacl - (PBYTE) sd); | |
194 | } | |
a998dd70 | 195 | /* Allow everyone to SYNCHRONIZE with this process. */ |
5735d5f6 CV |
196 | status = RtlAddAccessAllowedAce (dacl, ACL_REVISION, SYNCHRONIZE, |
197 | well_known_world_sid); | |
198 | if (!NT_SUCCESS (status)) | |
a998dd70 | 199 | { |
61522196 | 200 | debug_printf ("RtlAddAccessAllowedAce: %y", status); |
a998dd70 CV |
201 | return; |
202 | } | |
203 | /* Set the size of the DACL correctly. */ | |
1754539e CV |
204 | status = RtlFirstFreeAce (dacl, &ace); |
205 | if (!NT_SUCCESS (status)) | |
a998dd70 | 206 | { |
61522196 | 207 | debug_printf ("RtlFirstFreeAce: %y", status); |
a998dd70 CV |
208 | return; |
209 | } | |
210 | dacl->AclSize = (char *) ace - (char *) dacl; | |
211 | /* Write the DACL back. */ | |
f16706de | 212 | status = NtSetSecurityObject (NtCurrentProcess (), DACL_SECURITY_INFORMATION, sd); |
a998dd70 CV |
213 | if (!NT_SUCCESS (status)) |
214 | { | |
61522196 | 215 | debug_printf ("NtSetSecurityObject: %y", status); |
a998dd70 CV |
216 | return; |
217 | } | |
218 | done = true; | |
219 | } | |
220 | ||
636c94d8 CV |
221 | /* Get the handle count of an object. */ |
222 | static ULONG | |
223 | get_obj_handle_count (HANDLE h) | |
224 | { | |
225 | OBJECT_BASIC_INFORMATION obi; | |
226 | NTSTATUS status; | |
227 | ULONG hdl_cnt = 0; | |
228 | ||
229 | status = NtQueryObject (h, ObjectBasicInformation, &obi, sizeof obi, NULL); | |
230 | if (!NT_SUCCESS (status)) | |
61522196 | 231 | debug_printf ("NtQueryObject: %y", status); |
636c94d8 CV |
232 | else |
233 | hdl_cnt = obi.HandleCount; | |
234 | return hdl_cnt; | |
235 | } | |
236 | ||
f4ec8743 CV |
237 | /* Helper struct to construct a local OBJECT_ATTRIBUTES on the stack. */ |
238 | struct lockfattr_t | |
239 | { | |
240 | OBJECT_ATTRIBUTES attr; | |
241 | UNICODE_STRING uname; | |
242 | WCHAR name[LOCK_OBJ_NAME_LEN + 1]; | |
243 | }; | |
244 | ||
a998dd70 CV |
245 | /* Per lock class. */ |
246 | class lockf_t | |
247 | { | |
248 | public: | |
dbf576fd CV |
249 | uint16_t lf_flags; /* Semantics: F_POSIX, F_FLOCK, F_WAIT */ |
250 | uint16_t lf_type; /* Lock type: F_RDLCK, F_WRLCK */ | |
61522196 CV |
251 | off_t lf_start; /* Byte # of the start of the lock */ |
252 | off_t lf_end; /* Byte # of the end of the lock (-1=EOF) */ | |
253 | int64_t lf_id; /* Cygwin PID for POSIX locks, a unique id per | |
636c94d8 | 254 | file table entry for BSD flock locks. */ |
f4ec8743 | 255 | DWORD lf_wid; /* Win PID of the resource holding the lock */ |
2e560a09 CV |
256 | uint16_t lf_ver; /* Version number of the lock. If a released |
257 | lock event yet exists because another process | |
258 | is still waiting for it, we use the version | |
259 | field to distinguish old from new locks. */ | |
a998dd70 CV |
260 | class lockf_t **lf_head; /* Back pointer to the head of the lockf_t list */ |
261 | class inode_t *lf_inode; /* Back pointer to the inode_t */ | |
262 | class lockf_t *lf_next; /* Pointer to the next lock on this inode_t */ | |
f4ec8743 | 263 | HANDLE lf_obj; /* Handle to the lock event object. */ |
a998dd70 CV |
264 | |
265 | lockf_t () | |
266 | : lf_flags (0), lf_type (0), lf_start (0), lf_end (0), lf_id (0), | |
2e560a09 | 267 | lf_wid (0), lf_ver (0), lf_head (NULL), lf_inode (NULL), |
636c94d8 | 268 | lf_next (NULL), lf_obj (NULL) |
a998dd70 | 269 | {} |
2e560a09 | 270 | lockf_t (class inode_t *node, class lockf_t **head, |
61522196 | 271 | short flags, short type, off_t start, off_t end, |
2e560a09 | 272 | long long id, DWORD wid, uint16_t ver) |
a998dd70 | 273 | : lf_flags (flags), lf_type (type), lf_start (start), lf_end (end), |
2e560a09 | 274 | lf_id (id), lf_wid (wid), lf_ver (ver), lf_head (head), lf_inode (node), |
a998dd70 CV |
275 | lf_next (NULL), lf_obj (NULL) |
276 | {} | |
277 | ~lockf_t (); | |
278 | ||
dbf576fd CV |
279 | bool from_obj_name (class inode_t *node, class lockf_t **head, |
280 | const wchar_t *name); | |
281 | ||
31390e4c CV |
282 | /* Used to create all locks list in a given TLS buffer. */ |
283 | void *operator new (size_t size, void *p) | |
284 | { return p; } | |
285 | /* Used to store own lock list in the cygheap. */ | |
a998dd70 | 286 | void *operator new (size_t size) |
31390e4c CV |
287 | { return cmalloc (HEAP_FHANDLER, sizeof (lockf_t)); } |
288 | /* Never call on node->i_all_lf! */ | |
a998dd70 CV |
289 | void operator delete (void *p) |
290 | { cfree (p); } | |
291 | ||
f4ec8743 CV |
292 | POBJECT_ATTRIBUTES create_lock_obj_attr (lockfattr_t *attr, |
293 | ULONG flags); | |
294 | ||
a998dd70 | 295 | void create_lock_obj (); |
636c94d8 | 296 | bool open_lock_obj (); |
f4ec8743 | 297 | void close_lock_obj () { NtClose (lf_obj); lf_obj = NULL; } |
636c94d8 | 298 | void del_lock_obj (HANDLE fhdl, bool signal = false); |
a998dd70 CV |
299 | }; |
300 | ||
301 | /* Per inode_t class */ | |
302 | class inode_t | |
303 | { | |
304 | friend class lockf_t; | |
305 | ||
306 | public: | |
2e560a09 CV |
307 | LIST_ENTRY (inode_t) i_next; |
308 | lockf_t *i_lockf; /* List of locks of this process. */ | |
309 | lockf_t *i_all_lf; /* Temp list of all locks for this file. */ | |
a998dd70 | 310 | |
61522196 CV |
311 | dev_t i_dev; /* Device ID */ |
312 | ino_t i_ino; /* inode number */ | |
a998dd70 | 313 | |
636c94d8 | 314 | private: |
2e560a09 CV |
315 | HANDLE i_dir; |
316 | HANDLE i_mtx; | |
317 | uint32_t i_cnt; /* # of threads referencing this instance. */ | |
318 | ||
a998dd70 | 319 | public: |
61522196 | 320 | inode_t (dev_t dev, ino_t ino); |
a998dd70 CV |
321 | ~inode_t (); |
322 | ||
323 | void *operator new (size_t size) | |
31390e4c | 324 | { return cmalloc (HEAP_FHANDLER, sizeof (inode_t)); } |
a998dd70 CV |
325 | void operator delete (void *p) |
326 | { cfree (p); } | |
327 | ||
61522196 | 328 | static inode_t *get (dev_t dev, ino_t ino, |
dbf576fd | 329 | bool create_if_missing, bool lock); |
a998dd70 CV |
330 | |
331 | void LOCK () { WaitForSingleObject (i_mtx, INFINITE); } | |
332 | void UNLOCK () { ReleaseMutex (i_mtx); } | |
333 | ||
dbf576fd CV |
334 | void use () { ++i_cnt; } |
335 | void unuse () { if (i_cnt > 0) --i_cnt; } | |
336 | bool inuse () { return i_cnt > 0; } | |
2e560a09 CV |
337 | void notused () { i_cnt = 0; } |
338 | ||
52a5f8cc | 339 | void unlock_and_remove_if_unused (); |
a74869c0 | 340 | |
31390e4c | 341 | lockf_t *get_all_locks_list (); |
a998dd70 | 342 | |
636c94d8 | 343 | bool del_my_locks (long long id, HANDLE fhdl); |
a998dd70 CV |
344 | }; |
345 | ||
636c94d8 | 346 | inode_t::~inode_t () |
a998dd70 | 347 | { |
636c94d8 CV |
348 | lockf_t *lock, *n_lock; |
349 | for (lock = i_lockf; lock && (n_lock = lock->lf_next, 1); lock = n_lock) | |
350 | delete lock; | |
351 | NtClose (i_mtx); | |
352 | NtClose (i_dir); | |
a998dd70 CV |
353 | } |
354 | ||
2e560a09 | 355 | void |
52a5f8cc | 356 | inode_t::unlock_and_remove_if_unused () |
2e560a09 CV |
357 | { |
358 | UNLOCK (); | |
359 | INODE_LIST_LOCK (); | |
360 | unuse (); | |
361 | if (i_lockf == NULL && !inuse ()) | |
362 | { | |
363 | LIST_REMOVE (this, i_next); | |
364 | delete this; | |
365 | } | |
366 | INODE_LIST_UNLOCK (); | |
367 | } | |
368 | ||
636c94d8 CV |
369 | bool |
370 | inode_t::del_my_locks (long long id, HANDLE fhdl) | |
a998dd70 | 371 | { |
636c94d8 CV |
372 | lockf_t *lock, *n_lock; |
373 | lockf_t **prev = &i_lockf; | |
636c94d8 CV |
374 | for (lock = *prev; lock && (n_lock = lock->lf_next, 1); lock = n_lock) |
375 | { | |
376 | if (lock->lf_flags & F_POSIX) | |
377 | { | |
378 | /* Delete all POSIX locks. */ | |
379 | *prev = n_lock; | |
3c53eaeb CV |
380 | /* When called during fork, the POSIX lock must get deleted but |
381 | *not* signalled. The lock is still active and locked in the | |
382 | parent. So in case of fork, we call close_lock_obj explicitely, | |
383 | since del_lock_obj is called from the destructor. */ | |
384 | if (!id) | |
385 | lock->close_lock_obj (); | |
636c94d8 CV |
386 | delete lock; |
387 | } | |
388 | else if (id && lock->lf_id == id) | |
389 | { | |
390 | int cnt = 0; | |
391 | cygheap_fdenum cfd (true); | |
392 | while (cfd.next () >= 0) | |
393 | if (cfd->get_unique_id () == lock->lf_id && ++cnt > 1) | |
394 | break; | |
395 | /* Delete BSD flock lock when no other fd in this process references | |
396 | it anymore. */ | |
397 | if (cnt <= 1) | |
398 | { | |
399 | *prev = n_lock; | |
400 | lock->del_lock_obj (fhdl); | |
401 | delete lock; | |
402 | } | |
403 | } | |
404 | else | |
405 | prev = &lock->lf_next; | |
406 | } | |
407 | return i_lockf == NULL; | |
408 | } | |
a998dd70 | 409 | |
636c94d8 CV |
410 | /* Used to delete the locks on a file hold by this process. Called from |
411 | close(2) and fixup_after_fork, as well as from fixup_after_exec in | |
412 | case the close_on_exec flag is set. The whole inode is deleted as | |
413 | soon as no lock exists on it anymore. */ | |
414 | void | |
4a77aea0 | 415 | fhandler_base::del_my_locks (del_lock_called_from from) |
636c94d8 | 416 | { |
dbf576fd | 417 | inode_t *node = inode_t::get (get_dev (), get_ino (), false, true); |
636c94d8 CV |
418 | if (node) |
419 | { | |
4a77aea0 CV |
420 | /* When we're called from fixup_after_exec, the fhandler is a |
421 | close-on-exec fhandler. In this case our io handle is already | |
422 | invalid. We can't use it to test for the object reference count. | |
423 | However, that shouldn't be necessary for the following reason. | |
424 | After exec, there are no threads in the current process waiting for | |
425 | the lock. So, either we're the only process accessing the file table | |
426 | entry and there are no threads which require signalling, or we have | |
427 | a parent process still accessing the file object and signalling the | |
428 | lock event would be premature. */ | |
2e560a09 CV |
429 | node->del_my_locks (from == after_fork ? 0 : get_unique_id (), |
430 | from == after_exec ? NULL : get_handle ()); | |
52a5f8cc | 431 | node->unlock_and_remove_if_unused (); |
636c94d8 | 432 | } |
a998dd70 CV |
433 | } |
434 | ||
435 | /* Called in an execed child. The exec'ed process must allow SYNCHRONIZE | |
436 | access to everyone if at least one inode exists. | |
636c94d8 CV |
437 | The lock owner's Windows PID changed and all POSIX lock event objects |
438 | have to be relabeled so that waiting processes know which process to | |
439 | wait on. If the node has been abandoned due to close_on_exec on the | |
440 | referencing fhandlers, remove the inode entirely. */ | |
a998dd70 CV |
441 | void |
442 | fixup_lockf_after_exec () | |
443 | { | |
636c94d8 | 444 | inode_t *node, *next_node; |
a998dd70 CV |
445 | |
446 | INODE_LIST_LOCK (); | |
447 | if (LIST_FIRST (&cygheap->inode_list)) | |
448 | allow_others_to_sync (); | |
636c94d8 | 449 | LIST_FOREACH_SAFE (node, &cygheap->inode_list, i_next, next_node) |
a998dd70 | 450 | { |
2e560a09 | 451 | node->notused (); |
636c94d8 CV |
452 | int cnt = 0; |
453 | cygheap_fdenum cfd (true); | |
454 | while (cfd.next () >= 0) | |
455 | if (cfd->get_dev () == node->i_dev | |
456 | && cfd->get_ino () == node->i_ino | |
457 | && ++cnt > 1) | |
458 | break; | |
459 | if (cnt == 0) | |
a998dd70 | 460 | { |
636c94d8 CV |
461 | LIST_REMOVE (node, i_next); |
462 | delete node; | |
463 | } | |
464 | else | |
465 | { | |
466 | node->LOCK (); | |
467 | for (lockf_t *lock = node->i_lockf; lock; lock = lock->lf_next) | |
468 | if (lock->lf_flags & F_POSIX) | |
469 | { | |
470 | lock->del_lock_obj (NULL); | |
471 | lock->lf_wid = myself->dwProcessId; | |
2e560a09 | 472 | lock->lf_ver = 0; |
636c94d8 CV |
473 | lock->create_lock_obj (); |
474 | } | |
475 | node->UNLOCK (); | |
a998dd70 | 476 | } |
a998dd70 | 477 | } |
a998dd70 CV |
478 | INODE_LIST_UNLOCK (); |
479 | } | |
480 | ||
481 | /* static method to return a pointer to the inode_t structure for a specific | |
482 | file. The file is specified by the device and inode_t number. If inode_t | |
483 | doesn't exist, create it. */ | |
484 | inode_t * | |
61522196 | 485 | inode_t::get (dev_t dev, ino_t ino, bool create_if_missing, bool lock) |
a998dd70 CV |
486 | { |
487 | inode_t *node; | |
488 | ||
489 | INODE_LIST_LOCK (); | |
490 | LIST_FOREACH (node, &cygheap->inode_list, i_next) | |
491 | if (node->i_dev == dev && node->i_ino == ino) | |
492 | break; | |
636c94d8 | 493 | if (!node && create_if_missing) |
a998dd70 CV |
494 | { |
495 | node = new inode_t (dev, ino); | |
31390e4c CV |
496 | if (node) |
497 | LIST_INSERT_HEAD (&cygheap->inode_list, node, i_next); | |
a998dd70 | 498 | } |
636c94d8 | 499 | if (node) |
2e560a09 | 500 | node->use (); |
a998dd70 | 501 | INODE_LIST_UNLOCK (); |
dbf576fd | 502 | if (node && lock) |
2e560a09 | 503 | node->LOCK (); |
a998dd70 CV |
504 | return node; |
505 | } | |
506 | ||
61522196 | 507 | inode_t::inode_t (dev_t dev, ino_t ino) |
2e560a09 | 508 | : i_lockf (NULL), i_all_lf (NULL), i_dev (dev), i_ino (ino), i_cnt (0L) |
a998dd70 CV |
509 | { |
510 | HANDLE parent_dir; | |
db5ae618 | 511 | WCHAR name[48]; |
a998dd70 CV |
512 | UNICODE_STRING uname; |
513 | OBJECT_ATTRIBUTES attr; | |
514 | NTSTATUS status; | |
515 | ||
db5ae618 | 516 | parent_dir = get_shared_parent_dir (); |
a998dd70 CV |
517 | /* Create a subdir which is named after the device and inode_t numbers |
518 | of the given file, in hex notation. */ | |
dbf576fd | 519 | int len = __small_swprintf (name, LOCK_DIR_NAME_FMT, dev, ino); |
a998dd70 | 520 | RtlInitCountedUnicodeString (&uname, name, len * sizeof (WCHAR)); |
5cb524ee CV |
521 | InitializeObjectAttributes (&attr, &uname, OBJ_INHERIT | OBJ_OPENIF, |
522 | parent_dir, everyone_sd (FLOCK_INODE_DIR_ACCESS)); | |
523 | status = NtCreateDirectoryObject (&i_dir, FLOCK_INODE_DIR_ACCESS, &attr); | |
a998dd70 | 524 | if (!NT_SUCCESS (status)) |
61522196 | 525 | api_fatal ("NtCreateDirectoryObject(inode): %y", status); |
a998dd70 CV |
526 | /* Create a mutex object in the file specific dir, which is used for |
527 | access synchronization on the dir and its objects. */ | |
8deb4118 | 528 | InitializeObjectAttributes (&attr, &ro_u_mtx, OBJ_INHERIT | OBJ_OPENIF, i_dir, |
abbde487 CV |
529 | everyone_sd (CYG_MUTANT_ACCESS)); |
530 | status = NtCreateMutant (&i_mtx, CYG_MUTANT_ACCESS, &attr, FALSE); | |
a998dd70 | 531 | if (!NT_SUCCESS (status)) |
61522196 | 532 | api_fatal ("NtCreateMutant(inode): %y", status); |
a998dd70 CV |
533 | } |
534 | ||
a998dd70 CV |
535 | /* Enumerate all lock event objects for this file and create a lockf_t |
536 | list in the i_all_lf member. This list is searched in lf_getblock | |
537 | for locks which potentially block our lock request. */ | |
31390e4c CV |
538 | |
539 | /* Number of lockf_t structs which fit in the temporary buffer. */ | |
540 | #define MAX_LOCKF_CNT ((intptr_t)((NT_MAX_PATH * sizeof (WCHAR)) \ | |
541 | / sizeof (lockf_t))) | |
542 | ||
dbf576fd CV |
543 | bool |
544 | lockf_t::from_obj_name (inode_t *node, lockf_t **head, const wchar_t *name) | |
545 | { | |
546 | wchar_t *endptr; | |
547 | ||
548 | /* "%02x-%01x-%016X-%016X-%016X-%08x-%04x", | |
549 | lf_flags, lf_type, lf_start, lf_end, lf_id, lf_wid, lf_ver */ | |
550 | lf_flags = wcstol (name, &endptr, 16); | |
551 | if ((lf_flags & ~(F_FLOCK | F_POSIX)) != 0 | |
552 | || ((lf_flags & (F_FLOCK | F_POSIX)) == (F_FLOCK | F_POSIX))) | |
553 | return false; | |
554 | lf_type = wcstol (endptr + 1, &endptr, 16); | |
555 | if ((lf_type != F_RDLCK && lf_type != F_WRLCK) || !endptr || *endptr != L'-') | |
556 | return false; | |
61522196 | 557 | lf_start = (off_t) wcstoull (endptr + 1, &endptr, 16); |
dbf576fd CV |
558 | if (lf_start < 0 || !endptr || *endptr != L'-') |
559 | return false; | |
61522196 | 560 | lf_end = (off_t) wcstoull (endptr + 1, &endptr, 16); |
dbf576fd CV |
561 | if (lf_end < -1LL |
562 | || (lf_end > 0 && lf_end < lf_start) | |
563 | || !endptr || *endptr != L'-') | |
564 | return false; | |
565 | lf_id = wcstoll (endptr + 1, &endptr, 16); | |
566 | if (!endptr || *endptr != L'-' | |
567 | || ((lf_flags & F_POSIX) && (lf_id < 1 || lf_id > ULONG_MAX))) | |
568 | return false; | |
569 | lf_wid = wcstoul (endptr + 1, &endptr, 16); | |
570 | if (!endptr || *endptr != L'-') | |
571 | return false; | |
572 | lf_ver = wcstoul (endptr + 1, &endptr, 16); | |
573 | if (endptr && *endptr != L'\0') | |
574 | return false; | |
575 | lf_head = head; | |
576 | lf_inode = node; | |
577 | lf_next = NULL; | |
578 | lf_obj = NULL; | |
579 | return true; | |
580 | } | |
581 | ||
31390e4c | 582 | lockf_t * |
a998dd70 CV |
583 | inode_t::get_all_locks_list () |
584 | { | |
585 | struct fdbi | |
586 | { | |
587 | DIRECTORY_BASIC_INFORMATION dbi; | |
588 | WCHAR buf[2][NAME_MAX + 1]; | |
589 | } f; | |
590 | ULONG context; | |
591 | NTSTATUS status; | |
dbf576fd | 592 | lockf_t newlock, *lock = i_all_lf; |
a998dd70 | 593 | |
a998dd70 CV |
594 | for (BOOLEAN restart = TRUE; |
595 | NT_SUCCESS (status = NtQueryDirectoryObject (i_dir, &f, sizeof f, TRUE, | |
596 | restart, &context, NULL)); | |
597 | restart = FALSE) | |
598 | { | |
599 | if (f.dbi.ObjectName.Length != LOCK_OBJ_NAME_LEN * sizeof (WCHAR)) | |
600 | continue; | |
dbf576fd CV |
601 | f.dbi.ObjectName.Buffer[LOCK_OBJ_NAME_LEN] = L'\0'; |
602 | if (!newlock.from_obj_name (this, &i_all_lf, f.dbi.ObjectName.Buffer)) | |
025c1fac | 603 | continue; |
31390e4c | 604 | if (lock - i_all_lf >= MAX_LOCKF_CNT) |
025c1fac | 605 | { |
31390e4c CV |
606 | system_printf ("Warning, can't handle more than %d locks per file.", |
607 | MAX_LOCKF_CNT); | |
608 | break; | |
609 | } | |
610 | if (lock > i_all_lf) | |
025c1fac | 611 | lock[-1].lf_next = lock; |
dbf576fd | 612 | new (lock++) lockf_t (newlock); |
a998dd70 | 613 | } |
31390e4c CV |
614 | /* If no lock has been found, return NULL. */ |
615 | if (lock == i_all_lf) | |
616 | return NULL; | |
617 | return i_all_lf; | |
a998dd70 CV |
618 | } |
619 | ||
f4ec8743 CV |
620 | /* Create the lock object name. The name is constructed from the lock |
621 | properties which identify it uniquely, all values in hex. */ | |
622 | POBJECT_ATTRIBUTES | |
623 | lockf_t::create_lock_obj_attr (lockfattr_t *attr, ULONG flags) | |
624 | { | |
dbf576fd | 625 | __small_swprintf (attr->name, LOCK_OBJ_NAME_FMT, |
f4ec8743 | 626 | lf_flags & (F_POSIX | F_FLOCK), lf_type, lf_start, lf_end, |
2e560a09 | 627 | lf_id, lf_wid, lf_ver); |
f4ec8743 CV |
628 | RtlInitCountedUnicodeString (&attr->uname, attr->name, |
629 | LOCK_OBJ_NAME_LEN * sizeof (WCHAR)); | |
630 | InitializeObjectAttributes (&attr->attr, &attr->uname, flags, lf_inode->i_dir, | |
631 | everyone_sd (FLOCK_EVENT_ACCESS)); | |
632 | return &attr->attr; | |
633 | } | |
634 | ||
dbf576fd CV |
635 | DWORD WINAPI |
636 | create_lock_in_parent (PVOID param) | |
637 | { | |
638 | HANDLE lf_obj; | |
639 | ULONG size; | |
640 | OBJECT_NAME_INFORMATION *ntfn; | |
641 | NTSTATUS status; | |
642 | wchar_t *lockname, *inodename, *endptr; | |
61522196 CV |
643 | dev_t dev; |
644 | ino_t ino; | |
dbf576fd CV |
645 | inode_t *node; |
646 | lockf_t newlock, *lock; | |
647 | int cnt; | |
648 | ||
649 | /* param is the handle to the lock object, created by caller. */ | |
650 | lf_obj = (HANDLE) param; | |
651 | /* Fetch object path from handle. Typically the length of the path | |
652 | is 146 characters, starting with | |
653 | "\BaseNamedObject\cygwin-1S5-<16-hex-digits>\..." */ | |
654 | size = sizeof (OBJECT_NAME_INFORMATION) + 256 * sizeof (WCHAR); | |
655 | ntfn = (OBJECT_NAME_INFORMATION *) alloca (size); | |
656 | memset (ntfn, 0, size); | |
657 | status = NtQueryObject (lf_obj, ObjectNameInformation, ntfn, size, &size); | |
658 | if (!NT_SUCCESS (status)) | |
659 | goto err; | |
660 | ntfn->Name.Buffer[ntfn->Name.Length / sizeof (WCHAR)] = L'\0'; | |
661 | /* Sanity check so that we don't peek into unchartered territory. */ | |
662 | if (ntfn->Name.Length < LOCK_OBJ_NAME_LEN + LOCK_DIR_NAME_LEN + 1) | |
663 | goto err; | |
664 | /* The names have fixed size, so we know where the substrings start. */ | |
665 | lockname = ntfn->Name.Buffer + ntfn->Name.Length / sizeof (WCHAR) | |
666 | - LOCK_OBJ_NAME_LEN; | |
667 | inodename = lockname - LOCK_DIR_NAME_LEN - 1; | |
668 | dev = wcstoul (inodename + LOCK_DIR_NAME_DEV_OFF, &endptr, 16); | |
669 | if (*endptr != L'-') | |
670 | goto err; | |
671 | ino = wcstoull (inodename + LOCK_DIR_NAME_INO_OFF, &endptr, 16); | |
672 | if (*endptr != L'\\') | |
673 | goto err; | |
674 | if (!newlock.from_obj_name (NULL, NULL, lockname)) | |
675 | goto err; | |
676 | /* Check if we have an open file handle with the same unique id. */ | |
677 | { | |
678 | cnt = 0; | |
679 | cygheap_fdenum cfd (true); | |
680 | while (cfd.next () >= 0) | |
681 | if (cfd->get_unique_id () == newlock.lf_id && ++cnt > 0) | |
682 | break; | |
683 | } | |
684 | /* If not, close handle and return. */ | |
685 | if (!cnt) | |
686 | { | |
687 | NtClose (lf_obj); | |
688 | return 0; | |
689 | } | |
690 | /* otherwise generate inode from directory name... */ | |
691 | node = inode_t::get (dev, ino, true, false); | |
692 | /* ...and generate lock from object name. */ | |
693 | lock = new lockf_t (newlock); | |
694 | lock->lf_inode = node; | |
695 | lock->lf_head = &node->i_lockf; | |
696 | lock->lf_next = node->i_lockf; | |
697 | lock->lf_obj = lf_obj; | |
698 | node->i_lockf = lock; | |
699 | node->unuse (); | |
700 | return 0; | |
701 | ||
702 | err: | |
703 | system_printf ("Adding <%S> lock failed", &ntfn->Name); | |
704 | NtClose (lf_obj); | |
705 | return 1; | |
706 | } | |
707 | ||
708 | DWORD WINAPI | |
709 | delete_lock_in_parent (PVOID param) | |
710 | { | |
711 | inode_t *node; | |
712 | lockf_t *lock, **prev; | |
713 | ||
714 | /* Scan list of all inodes, and reap stale BSD lock if lf_id matches. | |
715 | Remove inode if empty. */ | |
716 | INODE_LIST_LOCK (); | |
717 | LIST_FOREACH (node, &cygheap->inode_list, i_next) | |
718 | if (!node->inuse ()) | |
719 | { | |
720 | for (prev = &node->i_lockf, lock = *prev; lock; lock = *prev) | |
721 | { | |
722 | if ((lock->lf_flags & F_FLOCK) && IsEventSignalled (lock->lf_obj)) | |
723 | { | |
724 | *prev = lock->lf_next; | |
725 | delete lock; | |
726 | } | |
727 | else | |
728 | prev = &lock->lf_next; | |
729 | } | |
730 | if (node->i_lockf == NULL) | |
731 | { | |
732 | LIST_REMOVE (node, i_next); | |
733 | delete node; | |
734 | } | |
735 | } | |
736 | INODE_LIST_UNLOCK (); | |
737 | return 0; | |
738 | } | |
739 | ||
a998dd70 | 740 | /* Create the lock event object in the file's subdir in the NT global |
f4ec8743 | 741 | namespace. */ |
a998dd70 CV |
742 | void |
743 | lockf_t::create_lock_obj () | |
744 | { | |
f4ec8743 | 745 | lockfattr_t attr; |
a998dd70 CV |
746 | NTSTATUS status; |
747 | ||
f4ec8743 CV |
748 | do |
749 | { | |
750 | status = NtCreateEvent (&lf_obj, CYG_EVENT_ACCESS, | |
751 | create_lock_obj_attr (&attr, OBJ_INHERIT), | |
752 | NotificationEvent, FALSE); | |
753 | if (!NT_SUCCESS (status)) | |
754 | { | |
2e560a09 | 755 | if (status != STATUS_OBJECT_NAME_COLLISION) |
61522196 | 756 | api_fatal ("NtCreateEvent(lock): %y", status); |
2e560a09 CV |
757 | /* If we get a STATUS_OBJECT_NAME_COLLISION, the event still exists |
758 | because some other process is waiting for it in lf_setlock. | |
759 | If so, check the event's signal state. If we can't open it, it | |
760 | has been closed in the meantime, so just try again. If we can | |
761 | open it and the object is not signalled, it's surely a bug in the | |
762 | code somewhere. Otherwise, close the event and retry to create | |
763 | a new event with another name. */ | |
f4ec8743 CV |
764 | if (open_lock_obj ()) |
765 | { | |
766 | if (!IsEventSignalled (lf_obj)) | |
61522196 | 767 | api_fatal ("NtCreateEvent(lock): %y", status); |
f4ec8743 | 768 | close_lock_obj (); |
2e560a09 CV |
769 | /* Increment the lf_ver field until we have no collision. */ |
770 | ++lf_ver; | |
f4ec8743 CV |
771 | } |
772 | } | |
773 | } | |
774 | while (!NT_SUCCESS (status)); | |
dbf576fd CV |
775 | /* For BSD locks, notify the parent process. */ |
776 | if (lf_flags & F_FLOCK) | |
777 | { | |
778 | HANDLE parent_proc, parent_thread, parent_lf_obj; | |
779 | ||
780 | pinfo p (myself->ppid); | |
781 | if (!p) /* No access or not a Cygwin parent. */ | |
782 | return; | |
783 | ||
784 | parent_proc = OpenProcess (PROCESS_DUP_HANDLE | |
785 | | PROCESS_CREATE_THREAD | |
786 | | PROCESS_QUERY_INFORMATION | |
787 | | PROCESS_VM_OPERATION | |
788 | | PROCESS_VM_WRITE | |
789 | | PROCESS_VM_READ, | |
790 | FALSE, p->dwProcessId); | |
791 | if (!parent_proc) | |
792 | { | |
793 | debug_printf ("OpenProcess (%u): %E", p->dwProcessId); | |
794 | return; | |
795 | } | |
796 | if (!DuplicateHandle (GetCurrentProcess (), lf_obj, parent_proc, | |
797 | &parent_lf_obj, TRUE, 0, DUPLICATE_SAME_ACCESS)) | |
798 | debug_printf ("DuplicateHandle (lf_obj): %E"); | |
799 | else | |
800 | { | |
801 | parent_thread = CreateRemoteThread (parent_proc, NULL, 256 * 1024, | |
802 | create_lock_in_parent, | |
803 | parent_lf_obj, | |
804 | STACK_SIZE_PARAM_IS_A_RESERVATION, | |
805 | NULL); | |
806 | if (!parent_thread) | |
807 | { | |
808 | debug_printf ("CreateRemoteThread: %E"); | |
809 | /* If thread didn't get started, close object handle in parent, | |
810 | otherwise suffer handle leaks. */ | |
811 | DuplicateHandle (parent_proc, parent_lf_obj, parent_proc, | |
812 | NULL, 0, FALSE, DUPLICATE_CLOSE_SOURCE); | |
813 | } | |
814 | else | |
815 | { | |
816 | /* Must wait to avoid race conditions. */ | |
817 | WaitForSingleObject (parent_thread, INFINITE); | |
818 | CloseHandle (parent_thread); | |
819 | } | |
820 | } | |
821 | CloseHandle (parent_proc); | |
822 | } | |
a998dd70 CV |
823 | } |
824 | ||
825 | /* Open a lock event object for SYNCHRONIZE access (to wait for it). */ | |
636c94d8 CV |
826 | bool |
827 | lockf_t::open_lock_obj () | |
a998dd70 | 828 | { |
f4ec8743 | 829 | lockfattr_t attr; |
a998dd70 | 830 | NTSTATUS status; |
a998dd70 | 831 | |
f4ec8743 CV |
832 | status = NtOpenEvent (&lf_obj, FLOCK_EVENT_ACCESS, |
833 | create_lock_obj_attr (&attr, 0)); | |
a998dd70 | 834 | if (!NT_SUCCESS (status)) |
c46d7be9 CV |
835 | { |
836 | SetLastError (RtlNtStatusToDosError (status)); | |
636c94d8 | 837 | lf_obj = NULL; /* Paranoia... */ |
a998dd70 | 838 | } |
636c94d8 | 839 | return lf_obj != NULL; |
a998dd70 CV |
840 | } |
841 | ||
f4ec8743 CV |
842 | /* Delete a lock event handle. The important thing here is to signal it |
843 | before closing the handle. This way all threads waiting for this lock | |
844 | can wake up. */ | |
a998dd70 | 845 | void |
636c94d8 | 846 | lockf_t::del_lock_obj (HANDLE fhdl, bool signal) |
a998dd70 CV |
847 | { |
848 | if (lf_obj) | |
849 | { | |
636c94d8 CV |
850 | /* Only signal the event if it's either a POSIX lock, or, in case of |
851 | BSD flock locks, if it's an explicit unlock or if the calling fhandler | |
852 | holds the last reference to the file table entry. The file table | |
853 | entry in UNIX terms is equivalent to the FILE_OBJECT in Windows NT | |
854 | terms. It's what the handle/descriptor references when calling | |
855 | CreateFile/open. Calling DuplicateHandle/dup only creates a new | |
856 | handle/descriptor to the same FILE_OBJECT/file table entry. */ | |
857 | if ((lf_flags & F_POSIX) || signal | |
858 | || (fhdl && get_obj_handle_count (fhdl) <= 1)) | |
dbf576fd CV |
859 | { |
860 | NtSetEvent (lf_obj, NULL); | |
861 | /* For BSD locks, notify the parent process. */ | |
862 | if (lf_flags & F_FLOCK) | |
863 | { | |
864 | HANDLE parent_proc, parent_thread; | |
865 | ||
866 | pinfo p (myself->ppid); | |
867 | if (p && (parent_proc = OpenProcess (PROCESS_CREATE_THREAD | |
868 | | PROCESS_QUERY_INFORMATION | |
869 | | PROCESS_VM_OPERATION | |
870 | | PROCESS_VM_WRITE | |
871 | | PROCESS_VM_READ, | |
872 | FALSE, p->dwProcessId))) | |
873 | { | |
874 | parent_thread = CreateRemoteThread (parent_proc, NULL, | |
875 | 256 * 1024, delete_lock_in_parent, | |
876 | NULL, | |
877 | STACK_SIZE_PARAM_IS_A_RESERVATION, | |
878 | NULL); | |
879 | if (parent_thread) | |
880 | { | |
881 | /* Must wait to avoid race conditions. */ | |
882 | WaitForSingleObject (parent_thread, INFINITE); | |
883 | CloseHandle (parent_thread); | |
884 | } | |
885 | CloseHandle (parent_proc); | |
886 | } | |
887 | } | |
888 | } | |
f4ec8743 | 889 | close_lock_obj (); |
a998dd70 CV |
890 | } |
891 | } | |
892 | ||
893 | lockf_t::~lockf_t () | |
894 | { | |
636c94d8 | 895 | del_lock_obj (NULL); |
a998dd70 CV |
896 | } |
897 | ||
898 | /* | |
899 | * This variable controls the maximum number of processes that will | |
900 | * be checked in doing deadlock detection. | |
901 | */ | |
636c94d8 | 902 | #ifndef __CYGWIN__ |
a998dd70 CV |
903 | #define MAXDEPTH 50 |
904 | static int maxlockdepth = MAXDEPTH; | |
905 | #endif | |
906 | ||
907 | #define NOLOCKF (struct lockf_t *)0 | |
908 | #define SELF 0x1 | |
909 | #define OTHERS 0x2 | |
636c94d8 | 910 | static int lf_clearlock (lockf_t *, lockf_t **, HANDLE); |
a998dd70 | 911 | static int lf_findoverlap (lockf_t *, lockf_t *, int, lockf_t ***, lockf_t **); |
31390e4c | 912 | static lockf_t *lf_getblock (lockf_t *, inode_t *node); |
61522196 | 913 | static int lf_getlock (lockf_t *, inode_t *, struct flock *); |
636c94d8 | 914 | static int lf_setlock (lockf_t *, inode_t *, lockf_t **, HANDLE); |
a998dd70 | 915 | static void lf_split (lockf_t *, lockf_t *, lockf_t **); |
636c94d8 | 916 | static void lf_wakelock (lockf_t *, HANDLE); |
a998dd70 CV |
917 | |
918 | int | |
61522196 | 919 | fhandler_disk_file::lock (int a_op, struct flock *fl) |
a998dd70 | 920 | { |
61522196 | 921 | off_t start, end, oadd; |
a998dd70 | 922 | int error = 0; |
a998dd70 CV |
923 | |
924 | short a_flags = fl->l_type & (F_POSIX | F_FLOCK); | |
925 | short type = fl->l_type & (F_RDLCK | F_WRLCK | F_UNLCK); | |
025c1fac | 926 | |
a998dd70 CV |
927 | if (!a_flags) |
928 | a_flags = F_POSIX; /* default */ | |
929 | if (a_op == F_SETLKW) | |
930 | { | |
931 | a_op = F_SETLK; | |
932 | a_flags |= F_WAIT; | |
933 | } | |
934 | if (a_op == F_SETLK) | |
935 | switch (type) | |
936 | { | |
937 | case F_UNLCK: | |
938 | a_op = F_UNLCK; | |
939 | break; | |
940 | case F_RDLCK: | |
fc69f1ae CV |
941 | /* flock semantics don't specify a requirement that the file has |
942 | been opened with a specific open mode, in contrast to POSIX locks | |
943 | which require that a file is opened for reading to place a read | |
944 | lock and opened for writing to place a write lock. */ | |
945 | if ((a_flags & F_POSIX) && !(get_access () & GENERIC_READ)) | |
a998dd70 CV |
946 | { |
947 | set_errno (EBADF); | |
948 | return -1; | |
949 | } | |
950 | break; | |
951 | case F_WRLCK: | |
fc69f1ae CV |
952 | /* See above comment. */ |
953 | if ((a_flags & F_POSIX) && !(get_access () & GENERIC_WRITE)) | |
a998dd70 CV |
954 | { |
955 | set_errno (EBADF); | |
956 | return -1; | |
957 | } | |
958 | break; | |
959 | default: | |
025c1fac | 960 | set_errno (EINVAL); |
a998dd70 CV |
961 | return -1; |
962 | } | |
963 | ||
a998dd70 CV |
964 | /* |
965 | * Convert the flock structure into a start and end. | |
966 | */ | |
967 | switch (fl->l_whence) | |
968 | { | |
969 | case SEEK_SET: | |
970 | start = fl->l_start; | |
971 | break; | |
972 | ||
973 | case SEEK_CUR: | |
974 | if ((start = lseek (0, SEEK_CUR)) == ILLEGAL_SEEK) | |
975 | return -1; | |
976 | break; | |
977 | ||
978 | case SEEK_END: | |
636c94d8 CV |
979 | { |
980 | NTSTATUS status; | |
981 | IO_STATUS_BLOCK io; | |
982 | FILE_STANDARD_INFORMATION fsi; | |
983 | ||
984 | status = NtQueryInformationFile (get_handle (), &io, &fsi, sizeof fsi, | |
985 | FileStandardInformation); | |
986 | if (!NT_SUCCESS (status)) | |
987 | { | |
988 | __seterrno_from_nt_status (status); | |
989 | return -1; | |
990 | } | |
991 | if (fl->l_start > 0 && fsi.EndOfFile.QuadPart > OFF_MAX - fl->l_start) | |
992 | { | |
993 | set_errno (EOVERFLOW); | |
994 | return -1; | |
995 | } | |
996 | start = fsi.EndOfFile.QuadPart + fl->l_start; | |
997 | } | |
a998dd70 CV |
998 | break; |
999 | ||
1000 | default: | |
1001 | return (EINVAL); | |
1002 | } | |
1003 | if (start < 0) | |
1004 | { | |
1005 | set_errno (EINVAL); | |
1006 | return -1; | |
1007 | } | |
1008 | if (fl->l_len < 0) | |
1009 | { | |
1010 | if (start == 0) | |
1011 | { | |
1012 | set_errno (EINVAL); | |
1013 | return -1; | |
1014 | } | |
1015 | end = start - 1; | |
1016 | start += fl->l_len; | |
1017 | if (start < 0) | |
1018 | { | |
1019 | set_errno (EINVAL); | |
1020 | return -1; | |
1021 | } | |
1022 | } | |
1023 | else if (fl->l_len == 0) | |
1024 | end = -1; | |
1025 | else | |
1026 | { | |
1027 | oadd = fl->l_len - 1; | |
1028 | if (oadd > OFF_MAX - start) | |
1029 | { | |
1030 | set_errno (EOVERFLOW); | |
1031 | return -1; | |
1032 | } | |
1033 | end = start + oadd; | |
1034 | } | |
1035 | ||
f4ec8743 CV |
1036 | restart: /* Entry point after a restartable signal came in. */ |
1037 | ||
dbf576fd | 1038 | inode_t *node = inode_t::get (get_dev (), get_ino (), true, true); |
a998dd70 CV |
1039 | if (!node) |
1040 | { | |
636c94d8 CV |
1041 | set_errno (ENOLCK); |
1042 | return -1; | |
a998dd70 | 1043 | } |
636c94d8 | 1044 | |
fabfb1a1 CV |
1045 | /* Unlock the fd table which has been locked in fcntl_worker/lock_worker, |
1046 | otherwise a blocking F_SETLKW never wakes up on a signal. */ | |
a998dd70 CV |
1047 | cygheap->fdtab.unlock (); |
1048 | ||
1049 | lockf_t **head = &node->i_lockf; | |
1050 | ||
9fa63f7a | 1051 | #if 0 |
a998dd70 CV |
1052 | /* |
1053 | * Avoid the common case of unlocking when inode_t has no locks. | |
9fa63f7a CV |
1054 | * |
1055 | * This shortcut is invalid for Cygwin because the above inode_t::get | |
1056 | * call returns with an empty lock list if this process has no locks | |
1057 | * on the file yet. | |
a998dd70 CV |
1058 | */ |
1059 | if (*head == NULL) | |
1060 | { | |
1061 | if (a_op != F_SETLK) | |
1062 | { | |
636c94d8 | 1063 | node->UNLOCK (); |
a998dd70 CV |
1064 | fl->l_type = F_UNLCK; |
1065 | return 0; | |
1066 | } | |
1067 | } | |
9fa63f7a | 1068 | #endif |
a998dd70 CV |
1069 | /* |
1070 | * Allocate a spare structure in case we have to split. | |
1071 | */ | |
1072 | lockf_t *clean = NULL; | |
1073 | if (a_op == F_SETLK || a_op == F_UNLCK) | |
31390e4c CV |
1074 | { |
1075 | clean = new lockf_t (); | |
1076 | if (!clean) | |
025c1fac | 1077 | { |
52a5f8cc | 1078 | node->unlock_and_remove_if_unused (); |
31390e4c CV |
1079 | set_errno (ENOLCK); |
1080 | return -1; | |
1081 | } | |
1082 | } | |
a998dd70 CV |
1083 | /* |
1084 | * Create the lockf_t structure | |
1085 | */ | |
1086 | lockf_t *lock = new lockf_t (node, head, a_flags, type, start, end, | |
636c94d8 CV |
1087 | (a_flags & F_FLOCK) ? get_unique_id () |
1088 | : getpid (), | |
2e560a09 | 1089 | myself->dwProcessId, 0); |
31390e4c CV |
1090 | if (!lock) |
1091 | { | |
52a5f8cc | 1092 | node->unlock_and_remove_if_unused (); |
31390e4c CV |
1093 | set_errno (ENOLCK); |
1094 | return -1; | |
1095 | } | |
a998dd70 | 1096 | |
a998dd70 CV |
1097 | switch (a_op) |
1098 | { | |
1099 | case F_SETLK: | |
636c94d8 | 1100 | error = lf_setlock (lock, node, &clean, get_handle ()); |
a998dd70 CV |
1101 | break; |
1102 | ||
1103 | case F_UNLCK: | |
636c94d8 | 1104 | error = lf_clearlock (lock, &clean, get_handle ()); |
a998dd70 | 1105 | lock->lf_next = clean; |
025c1fac | 1106 | clean = lock; |
a998dd70 | 1107 | break; |
025c1fac | 1108 | |
a998dd70 CV |
1109 | case F_GETLK: |
1110 | error = lf_getlock (lock, node, fl); | |
1111 | lock->lf_next = clean; | |
1112 | clean = lock; | |
1113 | break; | |
1114 | ||
1115 | default: | |
1116 | lock->lf_next = clean; | |
1117 | clean = lock; | |
1118 | error = EINVAL; | |
1119 | break; | |
1120 | } | |
1121 | for (lock = clean; lock != NULL; ) | |
1122 | { | |
f4ec8743 | 1123 | lockf_t *n = lock->lf_next; |
636c94d8 | 1124 | lock->del_lock_obj (get_handle (), a_op == F_UNLCK); |
a998dd70 CV |
1125 | delete lock; |
1126 | lock = n; | |
1127 | } | |
52a5f8cc | 1128 | node->unlock_and_remove_if_unused (); |
f4ec8743 | 1129 | switch (error) |
a998dd70 | 1130 | { |
f4ec8743 CV |
1131 | case 0: /* All is well. */ |
1132 | need_fork_fixup (true); | |
1133 | return 0; | |
1134 | case EINTR: /* Signal came in. */ | |
1135 | if (_my_tls.call_signal_handler ()) | |
1136 | goto restart; | |
1137 | break; | |
1138 | case ECANCELED: /* The thread has been sent a cancellation request. */ | |
1139 | pthread::static_cancel_self (); | |
1140 | /*NOTREACHED*/ | |
1141 | default: | |
1142 | break; | |
a998dd70 | 1143 | } |
f4ec8743 CV |
1144 | set_errno (error); |
1145 | return -1; | |
a998dd70 CV |
1146 | } |
1147 | ||
1148 | /* | |
1149 | * Set a byte-range lock. | |
1150 | */ | |
1151 | static int | |
636c94d8 | 1152 | lf_setlock (lockf_t *lock, inode_t *node, lockf_t **clean, HANDLE fhdl) |
025c1fac | 1153 | { |
a998dd70 CV |
1154 | lockf_t *block; |
1155 | lockf_t **head = lock->lf_head; | |
1156 | lockf_t **prev, *overlap; | |
1157 | int ovcase, priority, old_prio, needtolink; | |
30bbc55e | 1158 | tmp_pathbuf tp; |
a998dd70 CV |
1159 | |
1160 | /* | |
1161 | * Set the priority | |
1162 | */ | |
1163 | priority = old_prio = GetThreadPriority (GetCurrentThread ()); | |
1164 | if (lock->lf_type == F_WRLCK && priority <= THREAD_PRIORITY_ABOVE_NORMAL) | |
1165 | priority = THREAD_PRIORITY_HIGHEST; | |
1166 | /* | |
1167 | * Scan lock list for this file looking for locks that would block us. | |
1168 | */ | |
31390e4c | 1169 | /* Create temporary space for the all locks list. */ |
30bbc55e | 1170 | node->i_all_lf = (lockf_t *) (void *) tp.w_get (); |
a998dd70 CV |
1171 | while ((block = lf_getblock(lock, node))) |
1172 | { | |
636c94d8 CV |
1173 | HANDLE obj = block->lf_obj; |
1174 | block->lf_obj = NULL; | |
1175 | ||
a998dd70 CV |
1176 | /* |
1177 | * Free the structure and return if nonblocking. | |
1178 | */ | |
1179 | if ((lock->lf_flags & F_WAIT) == 0) | |
1180 | { | |
2e560a09 | 1181 | NtClose (obj); |
a998dd70 CV |
1182 | lock->lf_next = *clean; |
1183 | *clean = lock; | |
1184 | return EAGAIN; | |
1185 | } | |
1186 | /* | |
1187 | * We are blocked. Since flock style locks cover | |
1188 | * the whole file, there is no chance for deadlock. | |
1189 | * For byte-range locks we must check for deadlock. | |
1190 | * | |
1191 | * Deadlock detection is done by looking through the | |
1192 | * wait channels to see if there are any cycles that | |
1193 | * involve us. MAXDEPTH is set just to make sure we | |
1194 | * do not go off into neverland. | |
1195 | */ | |
1196 | /* FIXME: We check the handle count of all the lock event objects | |
025c1fac | 1197 | this process holds. If it's > 1, another process is |
a998dd70 CV |
1198 | waiting for one of our locks. This method isn't overly |
1199 | intelligent. If it turns out to be too dumb, we might | |
1200 | have to remove it or to find another method. */ | |
2e560a09 CV |
1201 | if (lock->lf_flags & F_POSIX) |
1202 | for (lockf_t *lk = node->i_lockf; lk; lk = lk->lf_next) | |
1203 | if ((lk->lf_flags & F_POSIX) && get_obj_handle_count (lk->lf_obj) > 1) | |
1204 | { | |
636c94d8 | 1205 | NtClose (obj); |
2e560a09 CV |
1206 | return EDEADLK; |
1207 | } | |
a998dd70 CV |
1208 | |
1209 | /* | |
1210 | * For flock type locks, we must first remove | |
1211 | * any shared locks that we hold before we sleep | |
1212 | * waiting for an exclusive lock. | |
1213 | */ | |
1214 | if ((lock->lf_flags & F_FLOCK) && lock->lf_type == F_WRLCK) | |
1215 | { | |
1216 | lock->lf_type = F_UNLCK; | |
636c94d8 | 1217 | (void) lf_clearlock (lock, clean, fhdl); |
a998dd70 CV |
1218 | lock->lf_type = F_WRLCK; |
1219 | } | |
1220 | ||
1221 | /* | |
1222 | * Add our lock to the blocked list and sleep until we're free. | |
1223 | * Remember who blocked us (for deadlock detection). | |
1224 | */ | |
636c94d8 | 1225 | /* Cygwin: No locked list. See deadlock recognition above. */ |
a998dd70 | 1226 | |
2e560a09 | 1227 | node->UNLOCK (); |
1804be04 | 1228 | |
2e560a09 CV |
1229 | /* Create list of objects to wait for. */ |
1230 | HANDLE w4[4] = { obj, NULL, NULL, NULL }; | |
1231 | DWORD wait_count = 1; | |
1804be04 | 1232 | |
5eb802f8 | 1233 | DWORD timeout; |
2e560a09 | 1234 | HANDLE proc = NULL; |
636c94d8 | 1235 | if (lock->lf_flags & F_POSIX) |
025c1fac | 1236 | { |
2e560a09 | 1237 | proc = OpenProcess (SYNCHRONIZE, FALSE, block->lf_wid); |
636c94d8 | 1238 | if (!proc) |
5eb802f8 | 1239 | timeout = 0L; |
2e560a09 | 1240 | else |
5eb802f8 CV |
1241 | { |
1242 | w4[wait_count++] = proc; | |
1243 | timeout = INFINITE; | |
1244 | } | |
636c94d8 | 1245 | } |
5eb802f8 CV |
1246 | else |
1247 | timeout = 100L; | |
1248 | ||
2e560a09 | 1249 | DWORD WAIT_SIGNAL_ARRIVED = WAIT_OBJECT_0 + wait_count; |
44aa2292 | 1250 | set_signal_arrived here (w4[wait_count++]); |
1804be04 | 1251 | |
2e560a09 CV |
1252 | DWORD WAIT_THREAD_CANCELED = WAIT_TIMEOUT + 1; |
1253 | HANDLE cancel_event = pthread::get_cancel_event (); | |
1254 | if (cancel_event) | |
1255 | { | |
1256 | WAIT_THREAD_CANCELED = WAIT_OBJECT_0 + wait_count; | |
1257 | w4[wait_count++] = cancel_event; | |
b9e2579c | 1258 | } |
2e560a09 CV |
1259 | |
1260 | /* Wait for the blocking object and, for POSIX locks, its holding process. | |
1b23b30b | 1261 | Unfortunately, since BSD flock locks are not attached to a specific |
2e560a09 CV |
1262 | process, we can't recognize an abandoned lock by sync'ing with the |
1263 | creator process. We have to make sure the event object is in a | |
1264 | signalled state, or that it has gone away. The latter we can only | |
1265 | recognize by retrying to fetch the block list, so we must not wait | |
5eb802f8 CV |
1266 | infinitely. For POSIX locks, if the process has already exited, |
1267 | just check if a signal or a thread cancel request arrived. */ | |
2e560a09 | 1268 | SetThreadPriority (GetCurrentThread (), priority); |
5eb802f8 | 1269 | DWORD ret = WaitForMultipleObjects (wait_count, w4, FALSE, timeout); |
636c94d8 | 1270 | SetThreadPriority (GetCurrentThread (), old_prio); |
2e560a09 CV |
1271 | if (proc) |
1272 | CloseHandle (proc); | |
1273 | node->LOCK (); | |
a8c8f19a CV |
1274 | /* Never close lock object handle outside of node lock! */ |
1275 | NtClose (obj); | |
f4ec8743 | 1276 | if (ret == WAIT_SIGNAL_ARRIVED) |
a998dd70 | 1277 | { |
a998dd70 | 1278 | /* A signal came in. */ |
f4ec8743 CV |
1279 | lock->lf_next = *clean; |
1280 | *clean = lock; | |
1281 | return EINTR; | |
1804be04 CV |
1282 | } |
1283 | else if (ret == WAIT_THREAD_CANCELED) | |
1284 | { | |
1285 | /* The thread has been sent a cancellation request. */ | |
f4ec8743 CV |
1286 | lock->lf_next = *clean; |
1287 | *clean = lock; | |
1288 | return ECANCELED; | |
1804be04 CV |
1289 | } |
1290 | else | |
f4ec8743 CV |
1291 | /* The lock object has been set to signalled or ... |
1292 | for POSIX locks, the process holding the lock has exited, or ... | |
2e560a09 | 1293 | just a timeout. Just retry. */ |
f4ec8743 | 1294 | continue; |
a998dd70 | 1295 | } |
a998dd70 CV |
1296 | allow_others_to_sync (); |
1297 | /* | |
1298 | * No blocks!! Add the lock. Note that we will | |
1299 | * downgrade or upgrade any overlapping locks this | |
1300 | * process already owns. | |
1301 | * | |
1302 | * Handle any locks that overlap. | |
1303 | */ | |
1304 | prev = head; | |
1305 | block = *head; | |
1306 | needtolink = 1; | |
1307 | for (;;) | |
1308 | { | |
1309 | ovcase = lf_findoverlap (block, lock, SELF, &prev, &overlap); | |
1310 | if (ovcase) | |
1311 | block = overlap->lf_next; | |
1312 | /* | |
1313 | * Six cases: | |
1314 | * 0) no overlap | |
1315 | * 1) overlap == lock | |
1316 | * 2) overlap contains lock | |
1317 | * 3) lock contains overlap | |
1318 | * 4) overlap starts before lock | |
1319 | * 5) overlap ends after lock | |
1320 | */ | |
1321 | switch (ovcase) | |
1322 | { | |
025c1fac | 1323 | case 0: /* no overlap */ |
a998dd70 CV |
1324 | if (needtolink) |
1325 | { | |
1326 | *prev = lock; | |
1327 | lock->lf_next = overlap; | |
1328 | lock->create_lock_obj (); | |
025c1fac CF |
1329 | } |
1330 | break; | |
a998dd70 | 1331 | |
025c1fac | 1332 | case 1: /* overlap == lock */ |
a998dd70 CV |
1333 | /* |
1334 | * If downgrading lock, others may be | |
1335 | * able to acquire it. | |
1336 | * Cygwin: Always wake lock. | |
1337 | */ | |
636c94d8 | 1338 | lf_wakelock (overlap, fhdl); |
a998dd70 CV |
1339 | overlap->lf_type = lock->lf_type; |
1340 | overlap->create_lock_obj (); | |
1341 | lock->lf_next = *clean; | |
1342 | *clean = lock; | |
1343 | break; | |
1344 | ||
025c1fac | 1345 | case 2: /* overlap contains lock */ |
a998dd70 CV |
1346 | /* |
1347 | * Check for common starting point and different types. | |
1348 | */ | |
1349 | if (overlap->lf_type == lock->lf_type) | |
1350 | { | |
1351 | lock->lf_next = *clean; | |
1352 | *clean = lock; | |
1353 | break; | |
1354 | } | |
1355 | if (overlap->lf_start == lock->lf_start) | |
1356 | { | |
1357 | *prev = lock; | |
1358 | lock->lf_next = overlap; | |
1359 | overlap->lf_start = lock->lf_end + 1; | |
1360 | } | |
1361 | else | |
1362 | lf_split (overlap, lock, clean); | |
636c94d8 | 1363 | lf_wakelock (overlap, fhdl); |
a998dd70 CV |
1364 | overlap->create_lock_obj (); |
1365 | lock->create_lock_obj (); | |
1366 | if (lock->lf_next && !lock->lf_next->lf_obj) | |
1367 | lock->lf_next->create_lock_obj (); | |
1368 | break; | |
1369 | ||
025c1fac | 1370 | case 3: /* lock contains overlap */ |
a998dd70 CV |
1371 | /* |
1372 | * If downgrading lock, others may be able to | |
1373 | * acquire it, otherwise take the list. | |
1374 | * Cygwin: Always wake old lock and create new lock. | |
1375 | */ | |
636c94d8 | 1376 | lf_wakelock (overlap, fhdl); |
a998dd70 CV |
1377 | /* |
1378 | * Add the new lock if necessary and delete the overlap. | |
1379 | */ | |
1380 | if (needtolink) | |
1381 | { | |
1382 | *prev = lock; | |
1383 | lock->lf_next = overlap->lf_next; | |
1384 | prev = &lock->lf_next; | |
1385 | lock->create_lock_obj (); | |
1386 | needtolink = 0; | |
1387 | } | |
1388 | else | |
1389 | *prev = overlap->lf_next; | |
1390 | overlap->lf_next = *clean; | |
1391 | *clean = overlap; | |
1392 | continue; | |
1393 | ||
025c1fac | 1394 | case 4: /* overlap starts before lock */ |
a998dd70 CV |
1395 | /* |
1396 | * Add lock after overlap on the list. | |
1397 | */ | |
1398 | lock->lf_next = overlap->lf_next; | |
1399 | overlap->lf_next = lock; | |
1400 | overlap->lf_end = lock->lf_start - 1; | |
1401 | prev = &lock->lf_next; | |
636c94d8 | 1402 | lf_wakelock (overlap, fhdl); |
a998dd70 CV |
1403 | overlap->create_lock_obj (); |
1404 | lock->create_lock_obj (); | |
1405 | needtolink = 0; | |
1406 | continue; | |
1407 | ||
025c1fac | 1408 | case 5: /* overlap ends after lock */ |
a998dd70 CV |
1409 | /* |
1410 | * Add the new lock before overlap. | |
1411 | */ | |
1412 | if (needtolink) { | |
1413 | *prev = lock; | |
1414 | lock->lf_next = overlap; | |
1415 | } | |
1416 | overlap->lf_start = lock->lf_end + 1; | |
636c94d8 | 1417 | lf_wakelock (overlap, fhdl); |
a998dd70 CV |
1418 | lock->create_lock_obj (); |
1419 | overlap->create_lock_obj (); | |
1420 | break; | |
1421 | } | |
1422 | break; | |
1423 | } | |
1424 | return 0; | |
1425 | } | |
1426 | ||
1427 | /* | |
1428 | * Remove a byte-range lock on an inode_t. | |
1429 | * | |
1430 | * Generally, find the lock (or an overlap to that lock) | |
1431 | * and remove it (or shrink it), then wakeup anyone we can. | |
1432 | */ | |
1433 | static int | |
636c94d8 | 1434 | lf_clearlock (lockf_t *unlock, lockf_t **clean, HANDLE fhdl) |
a998dd70 CV |
1435 | { |
1436 | lockf_t **head = unlock->lf_head; | |
1437 | lockf_t *lf = *head; | |
1438 | lockf_t *overlap, **prev; | |
1439 | int ovcase; | |
1440 | ||
1441 | if (lf == NOLOCKF) | |
1442 | return 0; | |
1443 | prev = head; | |
1444 | while ((ovcase = lf_findoverlap (lf, unlock, SELF, &prev, &overlap))) | |
1445 | { | |
1446 | /* | |
1447 | * Wakeup the list of locks to be retried. | |
1448 | */ | |
636c94d8 | 1449 | lf_wakelock (overlap, fhdl); |
a998dd70 CV |
1450 | |
1451 | switch (ovcase) | |
1452 | { | |
025c1fac | 1453 | case 1: /* overlap == lock */ |
a998dd70 CV |
1454 | *prev = overlap->lf_next; |
1455 | overlap->lf_next = *clean; | |
1456 | *clean = overlap; | |
1457 | break; | |
1458 | ||
025c1fac | 1459 | case 2: /* overlap contains lock: split it */ |
a998dd70 CV |
1460 | if (overlap->lf_start == unlock->lf_start) |
1461 | { | |
1462 | overlap->lf_start = unlock->lf_end + 1; | |
1463 | overlap->create_lock_obj (); | |
1464 | break; | |
1465 | } | |
1466 | lf_split (overlap, unlock, clean); | |
1467 | overlap->lf_next = unlock->lf_next; | |
1468 | overlap->create_lock_obj (); | |
1469 | if (overlap->lf_next && !overlap->lf_next->lf_obj) | |
1470 | overlap->lf_next->create_lock_obj (); | |
1471 | break; | |
1472 | ||
025c1fac | 1473 | case 3: /* lock contains overlap */ |
a998dd70 CV |
1474 | *prev = overlap->lf_next; |
1475 | lf = overlap->lf_next; | |
1476 | overlap->lf_next = *clean; | |
1477 | *clean = overlap; | |
1478 | continue; | |
1479 | ||
025c1fac CF |
1480 | case 4: /* overlap starts before lock */ |
1481 | overlap->lf_end = unlock->lf_start - 1; | |
1482 | prev = &overlap->lf_next; | |
1483 | lf = overlap->lf_next; | |
a998dd70 | 1484 | overlap->create_lock_obj (); |
025c1fac | 1485 | continue; |
a998dd70 | 1486 | |
025c1fac CF |
1487 | case 5: /* overlap ends after lock */ |
1488 | overlap->lf_start = unlock->lf_end + 1; | |
a998dd70 | 1489 | overlap->create_lock_obj (); |
025c1fac | 1490 | break; |
a998dd70 CV |
1491 | } |
1492 | break; | |
1493 | } | |
1494 | return 0; | |
1495 | } | |
1496 | ||
1497 | /* | |
1498 | * Check whether there is a blocking lock, | |
1499 | * and if so return its process identifier. | |
1500 | */ | |
1501 | static int | |
61522196 | 1502 | lf_getlock (lockf_t *lock, inode_t *node, struct flock *fl) |
a998dd70 CV |
1503 | { |
1504 | lockf_t *block; | |
30bbc55e | 1505 | tmp_pathbuf tp; |
a998dd70 | 1506 | |
31390e4c | 1507 | /* Create temporary space for the all locks list. */ |
30bbc55e | 1508 | node->i_all_lf = (lockf_t *) (void * ) tp.w_get (); |
a998dd70 CV |
1509 | if ((block = lf_getblock (lock, node))) |
1510 | { | |
636c94d8 | 1511 | if (block->lf_obj) |
f4ec8743 | 1512 | block->close_lock_obj (); |
a998dd70 CV |
1513 | fl->l_type = block->lf_type; |
1514 | fl->l_whence = SEEK_SET; | |
1515 | fl->l_start = block->lf_start; | |
1516 | if (block->lf_end == -1) | |
1517 | fl->l_len = 0; | |
1518 | else | |
1519 | fl->l_len = block->lf_end - block->lf_start + 1; | |
1520 | if (block->lf_flags & F_POSIX) | |
636c94d8 | 1521 | fl->l_pid = (pid_t) block->lf_id; |
a998dd70 CV |
1522 | else |
1523 | fl->l_pid = -1; | |
1524 | } | |
1525 | else | |
1526 | fl->l_type = F_UNLCK; | |
a998dd70 CV |
1527 | return 0; |
1528 | } | |
1529 | ||
1530 | /* | |
1531 | * Walk the list of locks for an inode_t and | |
1532 | * return the first blocking lock. | |
1533 | */ | |
1534 | static lockf_t * | |
1535 | lf_getblock (lockf_t *lock, inode_t *node) | |
025c1fac | 1536 | { |
31390e4c CV |
1537 | lockf_t **prev, *overlap; |
1538 | lockf_t *lf = node->get_all_locks_list (); | |
a998dd70 CV |
1539 | int ovcase; |
1540 | ||
1541 | prev = lock->lf_head; | |
1542 | while ((ovcase = lf_findoverlap (lf, lock, OTHERS, &prev, &overlap))) | |
1543 | { | |
1544 | /* | |
1545 | * We've found an overlap, see if it blocks us | |
1546 | */ | |
1547 | if ((lock->lf_type == F_WRLCK || overlap->lf_type == F_WRLCK)) | |
636c94d8 CV |
1548 | { |
1549 | /* Open the event object for synchronization. */ | |
f4ec8743 CV |
1550 | if (overlap->open_lock_obj ()) |
1551 | { | |
5eb802f8 CV |
1552 | /* Check if the event object is signalled. If so, the overlap |
1553 | doesn't actually exist anymore. There are just a few open | |
1554 | handles left. */ | |
f4ec8743 CV |
1555 | if (!IsEventSignalled (overlap->lf_obj)) |
1556 | return overlap; | |
1557 | overlap->close_lock_obj (); | |
1558 | } | |
636c94d8 | 1559 | } |
a998dd70 CV |
1560 | /* |
1561 | * Nope, point to the next one on the list and | |
1562 | * see if it blocks us | |
1563 | */ | |
1564 | lf = overlap->lf_next; | |
1565 | } | |
1566 | return NOLOCKF; | |
1567 | } | |
1568 | ||
1569 | /* | |
1570 | * Walk the list of locks for an inode_t to | |
1571 | * find an overlapping lock (if any). | |
1572 | * | |
1573 | * NOTE: this returns only the FIRST overlapping lock. There | |
1574 | * may be more than one. | |
025c1fac | 1575 | */ |
a998dd70 CV |
1576 | static int |
1577 | lf_findoverlap (lockf_t *lf, lockf_t *lock, int type, lockf_t ***prev, | |
1578 | lockf_t **overlap) | |
1579 | { | |
61522196 | 1580 | off_t start, end; |
a998dd70 CV |
1581 | |
1582 | *overlap = lf; | |
1583 | if (lf == NOLOCKF) | |
1584 | return 0; | |
1585 | ||
1586 | start = lock->lf_start; | |
1587 | end = lock->lf_end; | |
1588 | while (lf != NOLOCKF) | |
1589 | { | |
a74869c0 CV |
1590 | if (((type & SELF) && lf->lf_id != lock->lf_id) |
1591 | || ((type & OTHERS) && lf->lf_id == lock->lf_id) | |
636c94d8 | 1592 | /* As on Linux: POSIX locks and BSD flock locks don't interact. */ |
072030c3 CV |
1593 | || (lf->lf_flags & (F_POSIX | F_FLOCK)) |
1594 | != (lock->lf_flags & (F_POSIX | F_FLOCK))) | |
a998dd70 CV |
1595 | { |
1596 | *prev = &lf->lf_next; | |
1597 | *overlap = lf = lf->lf_next; | |
1598 | continue; | |
025c1fac | 1599 | } |
a998dd70 CV |
1600 | /* |
1601 | * OK, check for overlap | |
1602 | * | |
1603 | * Six cases: | |
1604 | * 0) no overlap | |
1605 | * 1) overlap == lock | |
1606 | * 2) overlap contains lock | |
1607 | * 3) lock contains overlap | |
1608 | * 4) overlap starts before lock | |
1609 | * 5) overlap ends after lock | |
1610 | */ | |
1611 | if ((lf->lf_end != -1 && start > lf->lf_end) || | |
1612 | (end != -1 && lf->lf_start > end)) | |
1613 | { | |
1614 | /* Case 0 */ | |
1615 | if ((type & SELF) && end != -1 && lf->lf_start > end) | |
1616 | return 0; | |
1617 | *prev = &lf->lf_next; | |
1618 | *overlap = lf = lf->lf_next; | |
1619 | continue; | |
025c1fac | 1620 | } |
a998dd70 CV |
1621 | if ((lf->lf_start == start) && (lf->lf_end == end)) |
1622 | { | |
1623 | /* Case 1 */ | |
1624 | return 1; | |
1625 | } | |
1626 | if ((lf->lf_start <= start) && (end != -1) && | |
1627 | ((lf->lf_end >= end) || (lf->lf_end == -1))) | |
1628 | { | |
1629 | /* Case 2 */ | |
1630 | return 2; | |
1631 | } | |
1632 | if (start <= lf->lf_start && (end == -1 || | |
1633 | (lf->lf_end != -1 && end >= lf->lf_end))) | |
1634 | { | |
1635 | /* Case 3 */ | |
1636 | return 3; | |
1637 | } | |
1638 | if ((lf->lf_start < start) && | |
1639 | ((lf->lf_end >= start) || (lf->lf_end == -1))) | |
1640 | { | |
1641 | /* Case 4 */ | |
1642 | return 4; | |
1643 | } | |
1644 | if ((lf->lf_start > start) && (end != -1) && | |
1645 | ((lf->lf_end > end) || (lf->lf_end == -1))) | |
1646 | { | |
1647 | /* Case 5 */ | |
1648 | return 5; | |
1649 | } | |
1650 | api_fatal ("lf_findoverlap: default\n"); | |
1651 | } | |
1652 | return 0; | |
1653 | } | |
1654 | ||
1655 | /* | |
1656 | * Split a lock and a contained region into | |
1657 | * two or three locks as necessary. | |
1658 | */ | |
1659 | static void | |
1660 | lf_split (lockf_t *lock1, lockf_t *lock2, lockf_t **split) | |
1661 | { | |
1662 | lockf_t *splitlock; | |
1663 | ||
025c1fac | 1664 | /* |
a998dd70 CV |
1665 | * Check to see if spliting into only two pieces. |
1666 | */ | |
1667 | if (lock1->lf_start == lock2->lf_start) | |
1668 | { | |
1669 | lock1->lf_start = lock2->lf_end + 1; | |
1670 | lock2->lf_next = lock1; | |
1671 | return; | |
1672 | } | |
1673 | if (lock1->lf_end == lock2->lf_end) | |
1674 | { | |
1675 | lock1->lf_end = lock2->lf_start - 1; | |
1676 | lock2->lf_next = lock1->lf_next; | |
1677 | lock1->lf_next = lock2; | |
1678 | return; | |
1679 | } | |
1680 | /* | |
1681 | * Make a new lock consisting of the last part of | |
1682 | * the encompassing lock. We use the preallocated | |
1683 | * splitlock so we don't have to block. | |
1684 | */ | |
1685 | splitlock = *split; | |
1686 | assert (splitlock != NULL); | |
1687 | *split = splitlock->lf_next; | |
1688 | memcpy (splitlock, lock1, sizeof *splitlock); | |
1689 | /* We have to unset the obj HANDLE here which has been copied by the | |
1690 | above memcpy, so that the calling function recognizes the new object. | |
1691 | See post-lf_split handling in lf_setlock and lf_clearlock. */ | |
1692 | splitlock->lf_obj = NULL; | |
1693 | splitlock->lf_start = lock2->lf_end + 1; | |
1694 | lock1->lf_end = lock2->lf_start - 1; | |
1695 | /* | |
1696 | * OK, now link it in | |
1697 | */ | |
1698 | splitlock->lf_next = lock1->lf_next; | |
1699 | lock2->lf_next = splitlock; | |
1700 | lock1->lf_next = lock2; | |
1701 | } | |
1702 | ||
1703 | /* | |
1704 | * Wakeup a blocklist | |
1705 | * Cygwin: Just signal the lock which gets removed. This unblocks | |
1706 | * all threads waiting for this lock. | |
1707 | */ | |
1708 | static void | |
636c94d8 | 1709 | lf_wakelock (lockf_t *listhead, HANDLE fhdl) |
a998dd70 | 1710 | { |
636c94d8 | 1711 | listhead->del_lock_obj (fhdl, true); |
a998dd70 | 1712 | } |
f3a1e23e | 1713 | |
fabfb1a1 | 1714 | extern "C" int |
f3a1e23e CV |
1715 | flock (int fd, int operation) |
1716 | { | |
fabfb1a1 CV |
1717 | int res = -1; |
1718 | int cmd; | |
61522196 | 1719 | struct flock fl = { 0, SEEK_SET, 0, 0, 0 }; |
fabfb1a1 CV |
1720 | |
1721 | myfault efault; | |
1722 | if (efault.faulted (EFAULT)) | |
1723 | return -1; | |
1724 | ||
1725 | cygheap_fdget cfd (fd, true); | |
1726 | if (cfd < 0) | |
1727 | goto done; | |
1728 | ||
1729 | cmd = (operation & LOCK_NB) ? F_SETLK : F_SETLKW; | |
f3a1e23e CV |
1730 | switch (operation & (~LOCK_NB)) |
1731 | { | |
1732 | case LOCK_EX: | |
fabfb1a1 | 1733 | fl.l_type = F_WRLCK | F_FLOCK; |
f3a1e23e CV |
1734 | break; |
1735 | case LOCK_SH: | |
fabfb1a1 | 1736 | fl.l_type = F_RDLCK | F_FLOCK; |
f3a1e23e CV |
1737 | break; |
1738 | case LOCK_UN: | |
fabfb1a1 | 1739 | fl.l_type = F_UNLCK | F_FLOCK; |
f3a1e23e CV |
1740 | break; |
1741 | default: | |
e2a39e2e | 1742 | set_errno (EINVAL); |
fabfb1a1 | 1743 | goto done; |
f3a1e23e | 1744 | } |
fabfb1a1 | 1745 | res = cfd->lock (cmd, &fl); |
7b9e380f | 1746 | if ((res == -1) && ((get_errno () == EAGAIN) || (get_errno () == EACCES))) |
fabfb1a1 CV |
1747 | set_errno (EWOULDBLOCK); |
1748 | done: | |
b9aa8149 | 1749 | syscall_printf ("%R = flock(%d, %d)", res, fd, operation); |
fabfb1a1 | 1750 | return res; |
f3a1e23e CV |
1751 | } |
1752 | ||
a998dd70 | 1753 | extern "C" int |
61522196 | 1754 | lockf (int filedes, int function, off_t size) |
f3a1e23e | 1755 | { |
fabfb1a1 | 1756 | int res = -1; |
a998dd70 | 1757 | int cmd; |
61522196 | 1758 | struct flock fl; |
fabfb1a1 | 1759 | |
1804be04 CV |
1760 | pthread_testcancel (); |
1761 | ||
fabfb1a1 CV |
1762 | myfault efault; |
1763 | if (efault.faulted (EFAULT)) | |
1764 | return -1; | |
1765 | ||
1766 | cygheap_fdget cfd (filedes, true); | |
1767 | if (cfd < 0) | |
1768 | goto done; | |
a998dd70 CV |
1769 | |
1770 | fl.l_start = 0; | |
1771 | fl.l_len = size; | |
1772 | fl.l_whence = SEEK_CUR; | |
1773 | ||
1774 | switch (function) | |
1775 | { | |
1776 | case F_ULOCK: | |
1777 | cmd = F_SETLK; | |
1778 | fl.l_type = F_UNLCK; | |
1779 | break; | |
1780 | case F_LOCK: | |
1781 | cmd = F_SETLKW; | |
1782 | fl.l_type = F_WRLCK; | |
1783 | break; | |
1784 | case F_TLOCK: | |
1785 | cmd = F_SETLK; | |
1786 | fl.l_type = F_WRLCK; | |
1787 | break; | |
1788 | case F_TEST: | |
1789 | fl.l_type = F_WRLCK; | |
fabfb1a1 CV |
1790 | if (cfd->lock (F_GETLK, &fl) == -1) |
1791 | goto done; | |
a998dd70 | 1792 | if (fl.l_type == F_UNLCK || fl.l_pid == getpid ()) |
fabfb1a1 CV |
1793 | res = 0; |
1794 | else | |
1795 | errno = EAGAIN; | |
1796 | goto done; | |
a998dd70 CV |
1797 | /* NOTREACHED */ |
1798 | default: | |
1799 | errno = EINVAL; | |
fabfb1a1 | 1800 | goto done; |
a998dd70 CV |
1801 | /* NOTREACHED */ |
1802 | } | |
fabfb1a1 CV |
1803 | res = cfd->lock (cmd, &fl); |
1804 | done: | |
b9aa8149 | 1805 | syscall_printf ("%R = lockf(%d, %d, %D)", res, filedes, function, size); |
fabfb1a1 | 1806 | return res; |
f3a1e23e | 1807 | } |