1 /* Copyright (C) 2002-2023 Free Software Foundation, Inc.
2 This file is part of the GNU C Library.
3
4 The GNU C Library is free software; you can redistribute it and/or
5 modify it under the terms of the GNU Lesser General Public
6 License as published by the Free Software Foundation; either
7 version 2.1 of the License, or (at your option) any later version.
8
9 The GNU C Library is distributed in the hope that it will be useful,
10 but WITHOUT ANY WARRANTY; without even the implied warranty of
11 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
12 Lesser General Public License for more details.
13
14 You should have received a copy of the GNU Lesser General Public
15 License along with the GNU C Library; if not, see
16 <https://www.gnu.org/licenses/>. */
17
18 #ifndef _DESCR_H
19 #define _DESCR_H 1
20
21 #include <limits.h>
22 #include <sched.h>
23 #include <setjmp.h>
24 #include <stdbool.h>
25 #include <sys/types.h>
26 #include <hp-timing.h>
27 #include <list_t.h>
28 #include <lowlevellock.h>
29 #include <pthreaddef.h>
30 #include <dl-sysdep.h>
31 #include <thread_db.h>
32 #include <tls.h>
33 #include <unwind.h>
34 #include <bits/types/res_state.h>
35 #include <kernel-features.h>
36 #include <tls-internal-struct.h>
37 #include <internal-sigset.h>
38
39 #ifndef TCB_ALIGNMENT
40 # define TCB_ALIGNMENT 32
41 #elif TCB_ALIGNMENT < 32
42 # error TCB_ALIGNMENT must be at least 32
43 #endif
44
45
46 /* We keep thread specific data in a special data structure, a two-level
47 array. The top-level array contains pointers to dynamically allocated
48 arrays of a certain number of data pointers. So we can implement a
49 sparse array. Each dynamic second-level array has
50 PTHREAD_KEY_2NDLEVEL_SIZE
51 entries. This value shouldn't be too large. */
52 #define PTHREAD_KEY_2NDLEVEL_SIZE 32
53
54 /* We need to address PTHREAD_KEYS_MAX key with PTHREAD_KEY_2NDLEVEL_SIZE
55 keys in each subarray. */
56 #define PTHREAD_KEY_1STLEVEL_SIZE \
57 ((PTHREAD_KEYS_MAX + PTHREAD_KEY_2NDLEVEL_SIZE - 1) \
58 / PTHREAD_KEY_2NDLEVEL_SIZE)
59
60
61
62
63 /* Internal version of the buffer to store cancellation handler
64 information. */
65 struct pthread_unwind_buf
66 {
67 struct
68 {
69 __jmp_buf jmp_buf;
70 int mask_was_saved;
71 } cancel_jmp_buf[1];
72
73 union
74 {
75 /* This is the placeholder of the public version. */
76 void *pad[4];
77
78 struct
79 {
80 /* Pointer to the previous cleanup buffer. */
81 struct pthread_unwind_buf *prev;
82
83 /* Backward compatibility: state of the old-style cleanup
84 handler at the time of the previous new-style cleanup handler
85 installment. */
86 struct _pthread_cleanup_buffer *cleanup;
87
88 /* Cancellation type before the push call. */
89 int canceltype;
90 } data;
91 } priv;
92 };
93
94
95 /* Opcodes and data types for communication with the signal handler to
96 change user/group IDs. */
97 struct xid_command
98 {
99 int syscall_no;
100 /* Enforce zero-extension for the pointer argument in
101
102 int setgroups (size_t size, const gid_t *list);
103
104 The kernel XID arguments are unsigned and do not require sign
105 extension. */
106 unsigned long int id[3];
107 volatile int cntr;
108 volatile int error; /* -1: no call yet, 0: success seen, >0: error seen. */
109 };
110
111
112 /* Data structure used by the kernel to find robust futexes. */
113 struct robust_list_head
114 {
115 void *list;
116 long int futex_offset;
117 void *list_op_pending;
118 };
119
120
121 /* Data structure used to handle thread priority protection. */
122 struct priority_protection_data
123 {
124 int priomax;
125 unsigned int priomap[];
126 };
127
128
129 /* Thread descriptor data structure. */
130 struct pthread
131 {
132 union
133 {
134 #if !TLS_DTV_AT_TP
135 /* This overlaps the TCB as used for TLS without threads (see tls.h). */
136 tcbhead_t header;
137 #else
138 struct
139 {
140 /* multiple_threads is enabled either when the process has spawned at
141 least one thread or when a single-threaded process cancels itself.
142 This enables additional code to introduce locking before doing some
143 compare_and_exchange operations and also enable cancellation points.
144 The concepts of multiple threads and cancellation points ideally
145 should be separate, since it is not necessary for multiple threads to
146 have been created for cancellation points to be enabled, as is the
147 case is when single-threaded process cancels itself.
148
149 Since enabling multiple_threads enables additional code in
150 cancellation points and compare_and_exchange operations, there is a
151 potential for an unneeded performance hit when it is enabled in a
152 single-threaded, self-canceling process. This is OK though, since a
153 single-threaded process will enable async cancellation only when it
154 looks to cancel itself and is hence going to end anyway. */
155 int multiple_threads;
156 int gscope_flag;
157 } header;
158 #endif
159
160 /* This extra padding has no special purpose, and this structure layout
161 is private and subject to change without affecting the official ABI.
162 We just have it here in case it might be convenient for some
163 implementation-specific instrumentation hack or suchlike. */
164 void *__padding[24];
165 };
166
167 /* This descriptor's link on the GL (dl_stack_used) or
168 GL (dl_stack_user) list. */
169 list_t list;
170
171 /* Thread ID - which is also a 'is this thread descriptor (and
172 therefore stack) used' flag. */
173 pid_t tid;
174
175 /* List of robust mutexes the thread is holding. */
176 #if __PTHREAD_MUTEX_HAVE_PREV
177 void *robust_prev;
178 struct robust_list_head robust_head;
179
180 /* The list above is strange. It is basically a double linked list
181 but the pointer to the next/previous element of the list points
182 in the middle of the object, the __next element. Whenever
183 casting to __pthread_list_t we need to adjust the pointer
184 first.
185 These operations are effectively concurrent code in that the thread
186 can get killed at any point in time and the kernel takes over. Thus,
187 the __next elements are a kind of concurrent list and we need to
188 enforce using compiler barriers that the individual operations happen
189 in such a way that the kernel always sees a consistent list. The
190 backward links (ie, the __prev elements) are not used by the kernel.
191 FIXME We should use relaxed MO atomic operations here and signal fences
192 because this kind of concurrency is similar to synchronizing with a
193 signal handler. */
194 # define QUEUE_PTR_ADJUST (offsetof (__pthread_list_t, __next))
195
196 # define ENQUEUE_MUTEX_BOTH(mutex, val) \
197 do { \
198 __pthread_list_t *next = (__pthread_list_t *) \
199 ((((uintptr_t) THREAD_GETMEM (THREAD_SELF, robust_head.list)) & ~1ul) \
200 - QUEUE_PTR_ADJUST); \
201 next->__prev = (void *) &mutex->__data.__list.__next; \
202 mutex->__data.__list.__next = THREAD_GETMEM (THREAD_SELF, \
203 robust_head.list); \
204 mutex->__data.__list.__prev = (void *) &THREAD_SELF->robust_head; \
205 /* Ensure that the new list entry is ready before we insert it. */ \
206 __asm ("" ::: "memory"); \
207 THREAD_SETMEM (THREAD_SELF, robust_head.list, \
208 (void *) (((uintptr_t) &mutex->__data.__list.__next) \
209 | val)); \
210 } while (0)
211 # define DEQUEUE_MUTEX(mutex) \
212 do { \
213 __pthread_list_t *next = (__pthread_list_t *) \
214 ((char *) (((uintptr_t) mutex->__data.__list.__next) & ~1ul) \
215 - QUEUE_PTR_ADJUST); \
216 next->__prev = mutex->__data.__list.__prev; \
217 __pthread_list_t *prev = (__pthread_list_t *) \
218 ((char *) (((uintptr_t) mutex->__data.__list.__prev) & ~1ul) \
219 - QUEUE_PTR_ADJUST); \
220 prev->__next = mutex->__data.__list.__next; \
221 /* Ensure that we remove the entry from the list before we change the \
222 __next pointer of the entry, which is read by the kernel. */ \
223 __asm ("" ::: "memory"); \
224 mutex->__data.__list.__prev = NULL; \
225 mutex->__data.__list.__next = NULL; \
226 } while (0)
227 #else
228 union
229 {
230 __pthread_slist_t robust_list;
231 struct robust_list_head robust_head;
232 };
233
234 # define ENQUEUE_MUTEX_BOTH(mutex, val) \
235 do { \
236 mutex->__data.__list.__next \
237 = THREAD_GETMEM (THREAD_SELF, robust_list.__next); \
238 /* Ensure that the new list entry is ready before we insert it. */ \
239 __asm ("" ::: "memory"); \
240 THREAD_SETMEM (THREAD_SELF, robust_list.__next, \
241 (void *) (((uintptr_t) &mutex->__data.__list) | val)); \
242 } while (0)
243 # define DEQUEUE_MUTEX(mutex) \
244 do { \
245 __pthread_slist_t *runp = (__pthread_slist_t *) \
246 (((uintptr_t) THREAD_GETMEM (THREAD_SELF, robust_list.__next)) & ~1ul); \
247 if (runp == &mutex->__data.__list) \
248 THREAD_SETMEM (THREAD_SELF, robust_list.__next, runp->__next); \
249 else \
250 { \
251 __pthread_slist_t *next = (__pthread_slist_t *) \
252 (((uintptr_t) runp->__next) & ~1ul); \
253 while (next != &mutex->__data.__list) \
254 { \
255 runp = next; \
256 next = (__pthread_slist_t *) (((uintptr_t) runp->__next) & ~1ul); \
257 } \
258 \
259 runp->__next = next->__next; \
260 /* Ensure that we remove the entry from the list before we change the \
261 __next pointer of the entry, which is read by the kernel. */ \
262 __asm ("" ::: "memory"); \
263 mutex->__data.__list.__next = NULL; \
264 } \
265 } while (0)
266 #endif
267 #define ENQUEUE_MUTEX(mutex) ENQUEUE_MUTEX_BOTH (mutex, 0)
268 #define ENQUEUE_MUTEX_PI(mutex) ENQUEUE_MUTEX_BOTH (mutex, 1)
269
270 /* List of cleanup buffers. */
271 struct _pthread_cleanup_buffer *cleanup;
272
273 /* Unwind information. */
274 struct pthread_unwind_buf *cleanup_jmp_buf;
275 #define HAVE_CLEANUP_JMP_BUF
276
277 /* Flags determining processing of cancellation. */
278 int cancelhandling;
279 /* Bit set if cancellation is disabled. */
280 #define CANCELSTATE_BIT 0
281 #define CANCELSTATE_BITMASK (1 << CANCELSTATE_BIT)
282 /* Bit set if asynchronous cancellation mode is selected. */
283 #define CANCELTYPE_BIT 1
284 #define CANCELTYPE_BITMASK (1 << CANCELTYPE_BIT)
285 /* Bit set if canceling has been initiated. */
286 #define CANCELING_BIT 2
287 #define CANCELING_BITMASK (1 << CANCELING_BIT)
288 /* Bit set if canceled. */
289 #define CANCELED_BIT 3
290 #define CANCELED_BITMASK (1 << CANCELED_BIT)
291 /* Bit set if thread is exiting. */
292 #define EXITING_BIT 4
293 #define EXITING_BITMASK (1 << EXITING_BIT)
294 /* Bit set if thread terminated and TCB is freed. */
295 #define TERMINATED_BIT 5
296 #define TERMINATED_BITMASK (1 << TERMINATED_BIT)
297 /* Bit set if thread is supposed to change XID. */
298 #define SETXID_BIT 6
299 #define SETXID_BITMASK (1 << SETXID_BIT)
300
301 /* Flags. Including those copied from the thread attribute. */
302 int flags;
303
304 /* We allocate one block of references here. This should be enough
305 to avoid allocating any memory dynamically for most applications. */
306 struct pthread_key_data
307 {
308 /* Sequence number. We use uintptr_t to not require padding on
309 32- and 64-bit machines. On 64-bit machines it helps to avoid
310 wrapping, too. */
311 uintptr_t seq;
312
313 /* Data pointer. */
314 void *data;
315 } specific_1stblock[PTHREAD_KEY_2NDLEVEL_SIZE];
316
317 /* Two-level array for the thread-specific data. */
318 struct pthread_key_data *specific[PTHREAD_KEY_1STLEVEL_SIZE];
319
320 /* Flag which is set when specific data is set. */
321 bool specific_used;
322
323 /* True if events must be reported. */
324 bool report_events;
325
326 /* True if the user provided the stack. */
327 bool user_stack;
328
329 /* True if thread must stop at startup time. */
330 bool stopped_start;
331
332 /* Indicate that a thread creation setup has failed (for instance the
333 scheduler or affinity). */
334 int setup_failed;
335
336 /* Lock to synchronize access to the descriptor. */
337 int lock;
338
339 /* Lock for synchronizing setxid calls. */
340 unsigned int setxid_futex;
341
342 /* If the thread waits to join another one the ID of the latter is
343 stored here.
344
345 In case a thread is detached this field contains a pointer of the
346 TCB if the thread itself. This is something which cannot happen
347 in normal operation. */
348 struct pthread *joinid;
349 /* Check whether a thread is detached. */
350 #define IS_DETACHED(pd) ((pd)->joinid == (pd))
351
352 /* The result of the thread function. */
353 void *result;
354
355 /* Scheduling parameters for the new thread. */
356 struct sched_param schedparam;
357 int schedpolicy;
358
359 /* Start position of the code to be executed and the argument passed
360 to the function. */
361 void *(*start_routine) (void *);
362 void *arg;
363
364 /* Debug state. */
365 td_eventbuf_t eventbuf;
366 /* Next descriptor with a pending event. */
367 struct pthread *nextevent;
368
369 /* Machine-specific unwind info. */
370 struct _Unwind_Exception exc;
371
372 /* If nonzero, pointer to the area allocated for the stack and guard. */
373 void *stackblock;
374 /* Size of the stackblock area including the guard. */
375 size_t stackblock_size;
376 /* Size of the included guard area. */
377 size_t guardsize;
378 /* This is what the user specified and what we will report. */
379 size_t reported_guardsize;
380
381 /* Thread Priority Protection data. */
382 struct priority_protection_data *tpp;
383
384 /* Resolver state. */
385 struct __res_state res;
386
387 /* Signal mask for the new thread. Used during thread startup to
388 restore the signal mask. (Threads are launched with all signals
389 masked.) */
390 internal_sigset_t sigmask;
391
392 /* Used by the exception handling implementation in the dynamic loader. */
393 struct rtld_catch *rtld_catch;
394
395 /* Indicates whether is a C11 thread created by thrd_creat. */
396 bool c11;
397
398 /* Used in __pthread_kill_internal to detected a thread that has
399 exited or is about to exit. exit_lock must only be acquired
400 after blocking signals. */
401 bool exiting;
402 int exit_lock; /* A low-level lock (for use with __libc_lock_init etc). */
403
404 /* Used on strsignal. */
405 struct tls_internal_t tls_state;
406
407 /* rseq area registered with the kernel. Use a custom definition
408 here to isolate from kernel struct rseq changes. The
409 implementation of sched_getcpu needs acccess to the cpu_id field;
410 the other fields are unused and not included here. */
411 union
412 {
413 struct
414 {
415 uint32_t cpu_id_start;
416 uint32_t cpu_id;
417 };
418 char pad[32]; /* Original rseq area size. */
419 } rseq_area __attribute__ ((aligned (32)));
420
421 /* Amount of end padding, if any, in this structure.
422 This definition relies on rseq_area being last. */
423 #define PTHREAD_STRUCT_END_PADDING \
424 (sizeof (struct pthread) - offsetof (struct pthread, rseq_area) \
425 + sizeof ((struct pthread) {}.rseq_area))
426 } __attribute ((aligned (TCB_ALIGNMENT)));
427
428 static inline bool
429 cancel_enabled_and_canceled (int value)
430 {
431 return (value & (CANCELSTATE_BITMASK | CANCELED_BITMASK | EXITING_BITMASK
432 | TERMINATED_BITMASK))
433 == CANCELED_BITMASK;
434 }
435
436 static inline bool
437 cancel_enabled_and_canceled_and_async (int value)
438 {
439 return ((value) & (CANCELSTATE_BITMASK | CANCELTYPE_BITMASK | CANCELED_BITMASK
440 | EXITING_BITMASK | TERMINATED_BITMASK))
441 == (CANCELTYPE_BITMASK | CANCELED_BITMASK);
442 }
443
444 /* This yields the pointer that TLS support code calls the thread pointer. */
445 #if TLS_TCB_AT_TP
446 # define TLS_TPADJ(pd) (pd)
447 #elif TLS_DTV_AT_TP
448 # define TLS_TPADJ(pd) ((struct pthread *)((char *) (pd) + TLS_PRE_TCB_SIZE))
449 #endif
450
451 #endif /* descr.h */