1 /*-
2 * Copyright (c) 2000 Marcel Moolenaar
3 * All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer
10 * in this position and unchanged.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 * notice, this list of conditions and the following disclaimer in the
13 * documentation and/or other materials provided with the distribution.
14 * 3. The name of the author may not be used to endorse or promote products
15 * derived from this software without specific prior written permission.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
18 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
19 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
20 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
21 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
22 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
23 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
24 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
25 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
26 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27 */
28
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD: releng/8.0/sys/i386/linux/linux_machdep.c 198951 2009-11-05 16:16:56Z jhb $");
31
32 #include <sys/param.h>
33 #include <sys/systm.h>
34 #include <sys/file.h>
35 #include <sys/fcntl.h>
36 #include <sys/imgact.h>
37 #include <sys/lock.h>
38 #include <sys/malloc.h>
39 #include <sys/mman.h>
40 #include <sys/mutex.h>
41 #include <sys/sx.h>
42 #include <sys/priv.h>
43 #include <sys/proc.h>
44 #include <sys/queue.h>
45 #include <sys/resource.h>
46 #include <sys/resourcevar.h>
47 #include <sys/signalvar.h>
48 #include <sys/syscallsubr.h>
49 #include <sys/sysproto.h>
50 #include <sys/unistd.h>
51 #include <sys/wait.h>
52 #include <sys/sched.h>
53
54 #include <machine/frame.h>
55 #include <machine/psl.h>
56 #include <machine/segments.h>
57 #include <machine/sysarch.h>
58
59 #include <vm/vm.h>
60 #include <vm/pmap.h>
61 #include <vm/vm_map.h>
62
63 #include <i386/linux/linux.h>
64 #include <i386/linux/linux_proto.h>
65 #include <compat/linux/linux_ipc.h>
66 #include <compat/linux/linux_signal.h>
67 #include <compat/linux/linux_util.h>
68 #include <compat/linux/linux_emul.h>
69
70 #include <i386/include/pcb.h> /* needed for pcb definition in linux_set_thread_area */
71
72 #include "opt_posix.h"
73
74 extern struct sysentvec elf32_freebsd_sysvec; /* defined in i386/i386/elf_machdep.c */
75
76 struct l_descriptor {
77 l_uint entry_number;
78 l_ulong base_addr;
79 l_uint limit;
80 l_uint seg_32bit:1;
81 l_uint contents:2;
82 l_uint read_exec_only:1;
83 l_uint limit_in_pages:1;
84 l_uint seg_not_present:1;
85 l_uint useable:1;
86 };
87
88 struct l_old_select_argv {
89 l_int nfds;
90 l_fd_set *readfds;
91 l_fd_set *writefds;
92 l_fd_set *exceptfds;
93 struct l_timeval *timeout;
94 };
95
96 static int linux_mmap_common(struct thread *td, l_uintptr_t addr,
97 l_size_t len, l_int prot, l_int flags, l_int fd,
98 l_loff_t pos);
99
100 int
101 linux_to_bsd_sigaltstack(int lsa)
102 {
103 int bsa = 0;
104
105 if (lsa & LINUX_SS_DISABLE)
106 bsa |= SS_DISABLE;
107 if (lsa & LINUX_SS_ONSTACK)
108 bsa |= SS_ONSTACK;
109 return (bsa);
110 }
111
112 int
113 bsd_to_linux_sigaltstack(int bsa)
114 {
115 int lsa = 0;
116
117 if (bsa & SS_DISABLE)
118 lsa |= LINUX_SS_DISABLE;
119 if (bsa & SS_ONSTACK)
120 lsa |= LINUX_SS_ONSTACK;
121 return (lsa);
122 }
123
124 int
125 linux_execve(struct thread *td, struct linux_execve_args *args)
126 {
127 int error;
128 char *newpath;
129 struct image_args eargs;
130
131 LCONVPATHEXIST(td, args->path, &newpath);
132
133 #ifdef DEBUG
134 if (ldebug(execve))
135 printf(ARGS(execve, "%s"), newpath);
136 #endif
137
138 error = exec_copyin_args(&eargs, newpath, UIO_SYSSPACE,
139 args->argp, args->envp);
140 free(newpath, M_TEMP);
141 if (error == 0)
142 error = kern_execve(td, &eargs, NULL);
143 if (error == 0)
144 /* linux process can exec fbsd one, dont attempt
145 * to create emuldata for such process using
146 * linux_proc_init, this leads to a panic on KASSERT
147 * because such process has p->p_emuldata == NULL
148 */
149 if (td->td_proc->p_sysent == &elf_linux_sysvec)
150 error = linux_proc_init(td, 0, 0);
151 return (error);
152 }
153
154 struct l_ipc_kludge {
155 struct l_msgbuf *msgp;
156 l_long msgtyp;
157 };
158
159 int
160 linux_ipc(struct thread *td, struct linux_ipc_args *args)
161 {
162
163 switch (args->what & 0xFFFF) {
164 case LINUX_SEMOP: {
165 struct linux_semop_args a;
166
167 a.semid = args->arg1;
168 a.tsops = args->ptr;
169 a.nsops = args->arg2;
170 return (linux_semop(td, &a));
171 }
172 case LINUX_SEMGET: {
173 struct linux_semget_args a;
174
175 a.key = args->arg1;
176 a.nsems = args->arg2;
177 a.semflg = args->arg3;
178 return (linux_semget(td, &a));
179 }
180 case LINUX_SEMCTL: {
181 struct linux_semctl_args a;
182 int error;
183
184 a.semid = args->arg1;
185 a.semnum = args->arg2;
186 a.cmd = args->arg3;
187 error = copyin(args->ptr, &a.arg, sizeof(a.arg));
188 if (error)
189 return (error);
190 return (linux_semctl(td, &a));
191 }
192 case LINUX_MSGSND: {
193 struct linux_msgsnd_args a;
194
195 a.msqid = args->arg1;
196 a.msgp = args->ptr;
197 a.msgsz = args->arg2;
198 a.msgflg = args->arg3;
199 return (linux_msgsnd(td, &a));
200 }
201 case LINUX_MSGRCV: {
202 struct linux_msgrcv_args a;
203
204 a.msqid = args->arg1;
205 a.msgsz = args->arg2;
206 a.msgflg = args->arg3;
207 if ((args->what >> 16) == 0) {
208 struct l_ipc_kludge tmp;
209 int error;
210
211 if (args->ptr == NULL)
212 return (EINVAL);
213 error = copyin(args->ptr, &tmp, sizeof(tmp));
214 if (error)
215 return (error);
216 a.msgp = tmp.msgp;
217 a.msgtyp = tmp.msgtyp;
218 } else {
219 a.msgp = args->ptr;
220 a.msgtyp = args->arg5;
221 }
222 return (linux_msgrcv(td, &a));
223 }
224 case LINUX_MSGGET: {
225 struct linux_msgget_args a;
226
227 a.key = args->arg1;
228 a.msgflg = args->arg2;
229 return (linux_msgget(td, &a));
230 }
231 case LINUX_MSGCTL: {
232 struct linux_msgctl_args a;
233
234 a.msqid = args->arg1;
235 a.cmd = args->arg2;
236 a.buf = args->ptr;
237 return (linux_msgctl(td, &a));
238 }
239 case LINUX_SHMAT: {
240 struct linux_shmat_args a;
241
242 a.shmid = args->arg1;
243 a.shmaddr = args->ptr;
244 a.shmflg = args->arg2;
245 a.raddr = (l_ulong *)args->arg3;
246 return (linux_shmat(td, &a));
247 }
248 case LINUX_SHMDT: {
249 struct linux_shmdt_args a;
250
251 a.shmaddr = args->ptr;
252 return (linux_shmdt(td, &a));
253 }
254 case LINUX_SHMGET: {
255 struct linux_shmget_args a;
256
257 a.key = args->arg1;
258 a.size = args->arg2;
259 a.shmflg = args->arg3;
260 return (linux_shmget(td, &a));
261 }
262 case LINUX_SHMCTL: {
263 struct linux_shmctl_args a;
264
265 a.shmid = args->arg1;
266 a.cmd = args->arg2;
267 a.buf = args->ptr;
268 return (linux_shmctl(td, &a));
269 }
270 default:
271 break;
272 }
273
274 return (EINVAL);
275 }
276
277 int
278 linux_old_select(struct thread *td, struct linux_old_select_args *args)
279 {
280 struct l_old_select_argv linux_args;
281 struct linux_select_args newsel;
282 int error;
283
284 #ifdef DEBUG
285 if (ldebug(old_select))
286 printf(ARGS(old_select, "%p"), args->ptr);
287 #endif
288
289 error = copyin(args->ptr, &linux_args, sizeof(linux_args));
290 if (error)
291 return (error);
292
293 newsel.nfds = linux_args.nfds;
294 newsel.readfds = linux_args.readfds;
295 newsel.writefds = linux_args.writefds;
296 newsel.exceptfds = linux_args.exceptfds;
297 newsel.timeout = linux_args.timeout;
298 return (linux_select(td, &newsel));
299 }
300
301 int
302 linux_fork(struct thread *td, struct linux_fork_args *args)
303 {
304 int error;
305 struct proc *p2;
306 struct thread *td2;
307
308 #ifdef DEBUG
309 if (ldebug(fork))
310 printf(ARGS(fork, ""));
311 #endif
312
313 if ((error = fork1(td, RFFDG | RFPROC | RFSTOPPED, 0, &p2)) != 0)
314 return (error);
315
316 if (error == 0) {
317 td->td_retval[0] = p2->p_pid;
318 td->td_retval[1] = 0;
319 }
320
321 if (td->td_retval[1] == 1)
322 td->td_retval[0] = 0;
323 error = linux_proc_init(td, td->td_retval[0], 0);
324 if (error)
325 return (error);
326
327 td2 = FIRST_THREAD_IN_PROC(p2);
328
329 /*
330 * Make this runnable after we are finished with it.
331 */
332 thread_lock(td2);
333 TD_SET_CAN_RUN(td2);
334 sched_add(td2, SRQ_BORING);
335 thread_unlock(td2);
336
337 return (0);
338 }
339
340 int
341 linux_vfork(struct thread *td, struct linux_vfork_args *args)
342 {
343 int error;
344 struct proc *p2;
345 struct thread *td2;
346
347 #ifdef DEBUG
348 if (ldebug(vfork))
349 printf(ARGS(vfork, ""));
350 #endif
351
352 /* exclude RFPPWAIT */
353 if ((error = fork1(td, RFFDG | RFPROC | RFMEM | RFSTOPPED, 0, &p2)) != 0)
354 return (error);
355 if (error == 0) {
356 td->td_retval[0] = p2->p_pid;
357 td->td_retval[1] = 0;
358 }
359 /* Are we the child? */
360 if (td->td_retval[1] == 1)
361 td->td_retval[0] = 0;
362 error = linux_proc_init(td, td->td_retval[0], 0);
363 if (error)
364 return (error);
365
366 PROC_LOCK(p2);
367 p2->p_flag |= P_PPWAIT;
368 PROC_UNLOCK(p2);
369
370 td2 = FIRST_THREAD_IN_PROC(p2);
371
372 /*
373 * Make this runnable after we are finished with it.
374 */
375 thread_lock(td2);
376 TD_SET_CAN_RUN(td2);
377 sched_add(td2, SRQ_BORING);
378 thread_unlock(td2);
379
380 /* wait for the children to exit, ie. emulate vfork */
381 PROC_LOCK(p2);
382 while (p2->p_flag & P_PPWAIT)
383 cv_wait(&p2->p_pwait, &p2->p_mtx);
384 PROC_UNLOCK(p2);
385
386 return (0);
387 }
388
389 int
390 linux_clone(struct thread *td, struct linux_clone_args *args)
391 {
392 int error, ff = RFPROC | RFSTOPPED;
393 struct proc *p2;
394 struct thread *td2;
395 int exit_signal;
396 struct linux_emuldata *em;
397
398 #ifdef DEBUG
399 if (ldebug(clone)) {
400 printf(ARGS(clone, "flags %x, stack %x, parent tid: %x, child tid: %x"),
401 (unsigned int)args->flags, (unsigned int)args->stack,
402 (unsigned int)args->parent_tidptr, (unsigned int)args->child_tidptr);
403 }
404 #endif
405
406 exit_signal = args->flags & 0x000000ff;
407 if (LINUX_SIG_VALID(exit_signal)) {
408 if (exit_signal <= LINUX_SIGTBLSZ)
409 exit_signal =
410 linux_to_bsd_signal[_SIG_IDX(exit_signal)];
411 } else if (exit_signal != 0)
412 return (EINVAL);
413
414 if (args->flags & LINUX_CLONE_VM)
415 ff |= RFMEM;
416 if (args->flags & LINUX_CLONE_SIGHAND)
417 ff |= RFSIGSHARE;
418 /*
419 * XXX: in linux sharing of fs info (chroot/cwd/umask)
420 * and open files is independant. in fbsd its in one
421 * structure but in reality it doesn't cause any problems
422 * because both of these flags are usually set together.
423 */
424 if (!(args->flags & (LINUX_CLONE_FILES | LINUX_CLONE_FS)))
425 ff |= RFFDG;
426
427 /*
428 * Attempt to detect when linux_clone(2) is used for creating
429 * kernel threads. Unfortunately despite the existence of the
430 * CLONE_THREAD flag, version of linuxthreads package used in
431 * most popular distros as of beginning of 2005 doesn't make
432 * any use of it. Therefore, this detection relies on
433 * empirical observation that linuxthreads sets certain
434 * combination of flags, so that we can make more or less
435 * precise detection and notify the FreeBSD kernel that several
436 * processes are in fact part of the same threading group, so
437 * that special treatment is necessary for signal delivery
438 * between those processes and fd locking.
439 */
440 if ((args->flags & 0xffffff00) == LINUX_THREADING_FLAGS)
441 ff |= RFTHREAD;
442
443 if (args->flags & LINUX_CLONE_PARENT_SETTID)
444 if (args->parent_tidptr == NULL)
445 return (EINVAL);
446
447 error = fork1(td, ff, 0, &p2);
448 if (error)
449 return (error);
450
451 if (args->flags & (LINUX_CLONE_PARENT | LINUX_CLONE_THREAD)) {
452 sx_xlock(&proctree_lock);
453 PROC_LOCK(p2);
454 proc_reparent(p2, td->td_proc->p_pptr);
455 PROC_UNLOCK(p2);
456 sx_xunlock(&proctree_lock);
457 }
458
459 /* create the emuldata */
460 error = linux_proc_init(td, p2->p_pid, args->flags);
461 /* reference it - no need to check this */
462 em = em_find(p2, EMUL_DOLOCK);
463 KASSERT(em != NULL, ("clone: emuldata not found.\n"));
464 /* and adjust it */
465
466 if (args->flags & LINUX_CLONE_THREAD) {
467 /* XXX: linux mangles pgrp and pptr somehow
468 * I think it might be this but I am not sure.
469 */
470 #ifdef notyet
471 PROC_LOCK(p2);
472 p2->p_pgrp = td->td_proc->p_pgrp;
473 PROC_UNLOCK(p2);
474 #endif
475 exit_signal = 0;
476 }
477
478 if (args->flags & LINUX_CLONE_CHILD_SETTID)
479 em->child_set_tid = args->child_tidptr;
480 else
481 em->child_set_tid = NULL;
482
483 if (args->flags & LINUX_CLONE_CHILD_CLEARTID)
484 em->child_clear_tid = args->child_tidptr;
485 else
486 em->child_clear_tid = NULL;
487
488 EMUL_UNLOCK(&emul_lock);
489
490 if (args->flags & LINUX_CLONE_PARENT_SETTID) {
491 error = copyout(&p2->p_pid, args->parent_tidptr, sizeof(p2->p_pid));
492 if (error)
493 printf(LMSG("copyout failed!"));
494 }
495
496 PROC_LOCK(p2);
497 p2->p_sigparent = exit_signal;
498 PROC_UNLOCK(p2);
499 td2 = FIRST_THREAD_IN_PROC(p2);
500 /*
501 * in a case of stack = NULL we are supposed to COW calling process stack
502 * this is what normal fork() does so we just keep the tf_esp arg intact
503 */
504 if (args->stack)
505 td2->td_frame->tf_esp = (unsigned int)args->stack;
506
507 if (args->flags & LINUX_CLONE_SETTLS) {
508 struct l_user_desc info;
509 int idx;
510 int a[2];
511 struct segment_descriptor sd;
512
513 error = copyin((void *)td->td_frame->tf_esi, &info, sizeof(struct l_user_desc));
514 if (error) {
515 printf(LMSG("copyin failed!"));
516 } else {
517
518 idx = info.entry_number;
519
520 /*
521 * looks like we're getting the idx we returned
522 * in the set_thread_area() syscall
523 */
524 if (idx != 6 && idx != 3) {
525 printf(LMSG("resetting idx!"));
526 idx = 3;
527 }
528
529 /* this doesnt happen in practice */
530 if (idx == 6) {
531 /* we might copy out the entry_number as 3 */
532 info.entry_number = 3;
533 error = copyout(&info, (void *) td->td_frame->tf_esi, sizeof(struct l_user_desc));
534 if (error)
535 printf(LMSG("copyout failed!"));
536 }
537
538 a[0] = LINUX_LDT_entry_a(&info);
539 a[1] = LINUX_LDT_entry_b(&info);
540
541 memcpy(&sd, &a, sizeof(a));
542 #ifdef DEBUG
543 if (ldebug(clone))
544 printf("Segment created in clone with CLONE_SETTLS: lobase: %x, hibase: %x, lolimit: %x, hilimit: %x, type: %i, dpl: %i, p: %i, xx: %i, def32: %i, gran: %i\n", sd.sd_lobase,
545 sd.sd_hibase,
546 sd.sd_lolimit,
547 sd.sd_hilimit,
548 sd.sd_type,
549 sd.sd_dpl,
550 sd.sd_p,
551 sd.sd_xx,
552 sd.sd_def32,
553 sd.sd_gran);
554 #endif
555
556 /* set %gs */
557 td2->td_pcb->pcb_gsd = sd;
558 td2->td_pcb->pcb_gs = GSEL(GUGS_SEL, SEL_UPL);
559 }
560 }
561
562 #ifdef DEBUG
563 if (ldebug(clone))
564 printf(LMSG("clone: successful rfork to %ld, stack %p sig = %d"),
565 (long)p2->p_pid, args->stack, exit_signal);
566 #endif
567 if (args->flags & LINUX_CLONE_VFORK) {
568 PROC_LOCK(p2);
569 p2->p_flag |= P_PPWAIT;
570 PROC_UNLOCK(p2);
571 }
572
573 /*
574 * Make this runnable after we are finished with it.
575 */
576 thread_lock(td2);
577 TD_SET_CAN_RUN(td2);
578 sched_add(td2, SRQ_BORING);
579 thread_unlock(td2);
580
581 td->td_retval[0] = p2->p_pid;
582 td->td_retval[1] = 0;
583
584 if (args->flags & LINUX_CLONE_VFORK) {
585 /* wait for the children to exit, ie. emulate vfork */
586 PROC_LOCK(p2);
587 while (p2->p_flag & P_PPWAIT)
588 cv_wait(&p2->p_pwait, &p2->p_mtx);
589 PROC_UNLOCK(p2);
590 }
591
592 return (0);
593 }
594
595 #define STACK_SIZE (2 * 1024 * 1024)
596 #define GUARD_SIZE (4 * PAGE_SIZE)
597
598 int
599 linux_mmap2(struct thread *td, struct linux_mmap2_args *args)
600 {
601
602 #ifdef DEBUG
603 if (ldebug(mmap2))
604 printf(ARGS(mmap2, "%p, %d, %d, 0x%08x, %d, %d"),
605 (void *)args->addr, args->len, args->prot,
606 args->flags, args->fd, args->pgoff);
607 #endif
608
609 return (linux_mmap_common(td, args->addr, args->len, args->prot,
610 args->flags, args->fd, (uint64_t)(uint32_t)args->pgoff *
611 PAGE_SIZE));
612 }
613
614 int
615 linux_mmap(struct thread *td, struct linux_mmap_args *args)
616 {
617 int error;
618 struct l_mmap_argv linux_args;
619
620 error = copyin(args->ptr, &linux_args, sizeof(linux_args));
621 if (error)
622 return (error);
623
624 #ifdef DEBUG
625 if (ldebug(mmap))
626 printf(ARGS(mmap, "%p, %d, %d, 0x%08x, %d, %d"),
627 (void *)linux_args.addr, linux_args.len, linux_args.prot,
628 linux_args.flags, linux_args.fd, linux_args.pgoff);
629 #endif
630
631 return (linux_mmap_common(td, linux_args.addr, linux_args.len,
632 linux_args.prot, linux_args.flags, linux_args.fd,
633 (uint32_t)linux_args.pgoff));
634 }
635
636 static int
637 linux_mmap_common(struct thread *td, l_uintptr_t addr, l_size_t len, l_int prot,
638 l_int flags, l_int fd, l_loff_t pos)
639 {
640 struct proc *p = td->td_proc;
641 struct mmap_args /* {
642 caddr_t addr;
643 size_t len;
644 int prot;
645 int flags;
646 int fd;
647 long pad;
648 off_t pos;
649 } */ bsd_args;
650 int error;
651 struct file *fp;
652
653 error = 0;
654 bsd_args.flags = 0;
655 fp = NULL;
656
657 /*
658 * Linux mmap(2):
659 * You must specify exactly one of MAP_SHARED and MAP_PRIVATE
660 */
661 if (!((flags & LINUX_MAP_SHARED) ^ (flags & LINUX_MAP_PRIVATE)))
662 return (EINVAL);
663
664 if (flags & LINUX_MAP_SHARED)
665 bsd_args.flags |= MAP_SHARED;
666 if (flags & LINUX_MAP_PRIVATE)
667 bsd_args.flags |= MAP_PRIVATE;
668 if (flags & LINUX_MAP_FIXED)
669 bsd_args.flags |= MAP_FIXED;
670 if (flags & LINUX_MAP_ANON)
671 bsd_args.flags |= MAP_ANON;
672 else
673 bsd_args.flags |= MAP_NOSYNC;
674 if (flags & LINUX_MAP_GROWSDOWN)
675 bsd_args.flags |= MAP_STACK;
676
677 /*
678 * PROT_READ, PROT_WRITE, or PROT_EXEC implies PROT_READ and PROT_EXEC
679 * on Linux/i386. We do this to ensure maximum compatibility.
680 * Linux/ia64 does the same in i386 emulation mode.
681 */
682 bsd_args.prot = prot;
683 if (bsd_args.prot & (PROT_READ | PROT_WRITE | PROT_EXEC))
684 bsd_args.prot |= PROT_READ | PROT_EXEC;
685
686 /* Linux does not check file descriptor when MAP_ANONYMOUS is set. */
687 bsd_args.fd = (bsd_args.flags & MAP_ANON) ? -1 : fd;
688 if (bsd_args.fd != -1) {
689 /*
690 * Linux follows Solaris mmap(2) description:
691 * The file descriptor fildes is opened with
692 * read permission, regardless of the
693 * protection options specified.
694 */
695
696 if ((error = fget(td, bsd_args.fd, &fp)) != 0)
697 return (error);
698 if (fp->f_type != DTYPE_VNODE) {
699 fdrop(fp, td);
700 return (EINVAL);
701 }
702
703 /* Linux mmap() just fails for O_WRONLY files */
704 if (!(fp->f_flag & FREAD)) {
705 fdrop(fp, td);
706 return (EACCES);
707 }
708
709 fdrop(fp, td);
710 }
711
712 if (flags & LINUX_MAP_GROWSDOWN) {
713 /*
714 * The Linux MAP_GROWSDOWN option does not limit auto
715 * growth of the region. Linux mmap with this option
716 * takes as addr the inital BOS, and as len, the initial
717 * region size. It can then grow down from addr without
718 * limit. However, linux threads has an implicit internal
719 * limit to stack size of STACK_SIZE. Its just not
720 * enforced explicitly in linux. But, here we impose
721 * a limit of (STACK_SIZE - GUARD_SIZE) on the stack
722 * region, since we can do this with our mmap.
723 *
724 * Our mmap with MAP_STACK takes addr as the maximum
725 * downsize limit on BOS, and as len the max size of
726 * the region. It them maps the top SGROWSIZ bytes,
727 * and auto grows the region down, up to the limit
728 * in addr.
729 *
730 * If we don't use the MAP_STACK option, the effect
731 * of this code is to allocate a stack region of a
732 * fixed size of (STACK_SIZE - GUARD_SIZE).
733 */
734
735 if ((caddr_t)PTRIN(addr) + len > p->p_vmspace->vm_maxsaddr) {
736 /*
737 * Some linux apps will attempt to mmap
738 * thread stacks near the top of their
739 * address space. If their TOS is greater
740 * than vm_maxsaddr, vm_map_growstack()
741 * will confuse the thread stack with the
742 * process stack and deliver a SEGV if they
743 * attempt to grow the thread stack past their
744 * current stacksize rlimit. To avoid this,
745 * adjust vm_maxsaddr upwards to reflect
746 * the current stacksize rlimit rather
747 * than the maximum possible stacksize.
748 * It would be better to adjust the
749 * mmap'ed region, but some apps do not check
750 * mmap's return value.
751 */
752 PROC_LOCK(p);
753 p->p_vmspace->vm_maxsaddr = (char *)USRSTACK -
754 lim_cur(p, RLIMIT_STACK);
755 PROC_UNLOCK(p);
756 }
757
758 /*
759 * This gives us our maximum stack size and a new BOS.
760 * If we're using VM_STACK, then mmap will just map
761 * the top SGROWSIZ bytes, and let the stack grow down
762 * to the limit at BOS. If we're not using VM_STACK
763 * we map the full stack, since we don't have a way
764 * to autogrow it.
765 */
766 if (len > STACK_SIZE - GUARD_SIZE) {
767 bsd_args.addr = (caddr_t)PTRIN(addr);
768 bsd_args.len = len;
769 } else {
770 bsd_args.addr = (caddr_t)PTRIN(addr) -
771 (STACK_SIZE - GUARD_SIZE - len);
772 bsd_args.len = STACK_SIZE - GUARD_SIZE;
773 }
774 } else {
775 bsd_args.addr = (caddr_t)PTRIN(addr);
776 bsd_args.len = len;
777 }
778 bsd_args.pos = pos;
779
780 #ifdef DEBUG
781 if (ldebug(mmap))
782 printf("-> %s(%p, %d, %d, 0x%08x, %d, 0x%x)\n",
783 __func__,
784 (void *)bsd_args.addr, bsd_args.len, bsd_args.prot,
785 bsd_args.flags, bsd_args.fd, (int)bsd_args.pos);
786 #endif
787 error = mmap(td, &bsd_args);
788 #ifdef DEBUG
789 if (ldebug(mmap))
790 printf("-> %s() return: 0x%x (0x%08x)\n",
791 __func__, error, (u_int)td->td_retval[0]);
792 #endif
793 return (error);
794 }
795
796 int
797 linux_mprotect(struct thread *td, struct linux_mprotect_args *uap)
798 {
799 struct mprotect_args bsd_args;
800
801 bsd_args.addr = uap->addr;
802 bsd_args.len = uap->len;
803 bsd_args.prot = uap->prot;
804 if (bsd_args.prot & (PROT_READ | PROT_WRITE | PROT_EXEC))
805 bsd_args.prot |= PROT_READ | PROT_EXEC;
806 return (mprotect(td, &bsd_args));
807 }
808
809 int
810 linux_pipe(struct thread *td, struct linux_pipe_args *args)
811 {
812 int error;
813 int fildes[2];
814
815 #ifdef DEBUG
816 if (ldebug(pipe))
817 printf(ARGS(pipe, "*"));
818 #endif
819
820 error = kern_pipe(td, fildes);
821 if (error)
822 return (error);
823
824 /* XXX: Close descriptors on error. */
825 return (copyout(fildes, args->pipefds, sizeof fildes));
826 }
827
828 int
829 linux_ioperm(struct thread *td, struct linux_ioperm_args *args)
830 {
831 int error;
832 struct i386_ioperm_args iia;
833
834 iia.start = args->start;
835 iia.length = args->length;
836 iia.enable = args->enable;
837 error = i386_set_ioperm(td, &iia);
838 return (error);
839 }
840
841 int
842 linux_iopl(struct thread *td, struct linux_iopl_args *args)
843 {
844 int error;
845
846 if (args->level < 0 || args->level > 3)
847 return (EINVAL);
848 if ((error = priv_check(td, PRIV_IO)) != 0)
849 return (error);
850 if ((error = securelevel_gt(td->td_ucred, 0)) != 0)
851 return (error);
852 td->td_frame->tf_eflags = (td->td_frame->tf_eflags & ~PSL_IOPL) |
853 (args->level * (PSL_IOPL / 3));
854 return (0);
855 }
856
857 int
858 linux_modify_ldt(struct thread *td, struct linux_modify_ldt_args *uap)
859 {
860 int error;
861 struct i386_ldt_args ldt;
862 struct l_descriptor ld;
863 union descriptor desc;
864 int size, written;
865
866 switch (uap->func) {
867 case 0x00: /* read_ldt */
868 ldt.start = 0;
869 ldt.descs = uap->ptr;
870 ldt.num = uap->bytecount / sizeof(union descriptor);
871 error = i386_get_ldt(td, &ldt);
872 td->td_retval[0] *= sizeof(union descriptor);
873 break;
874 case 0x02: /* read_default_ldt = 0 */
875 size = 5*sizeof(struct l_desc_struct);
876 if (size > uap->bytecount)
877 size = uap->bytecount;
878 for (written = error = 0; written < size && error == 0; written++)
879 error = subyte((char *)uap->ptr + written, 0);
880 td->td_retval[0] = written;
881 break;
882 case 0x01: /* write_ldt */
883 case 0x11: /* write_ldt */
884 if (uap->bytecount != sizeof(ld))
885 return (EINVAL);
886
887 error = copyin(uap->ptr, &ld, sizeof(ld));
888 if (error)
889 return (error);
890
891 ldt.start = ld.entry_number;
892 ldt.descs = &desc;
893 ldt.num = 1;
894 desc.sd.sd_lolimit = (ld.limit & 0x0000ffff);
895 desc.sd.sd_hilimit = (ld.limit & 0x000f0000) >> 16;
896 desc.sd.sd_lobase = (ld.base_addr & 0x00ffffff);
897 desc.sd.sd_hibase = (ld.base_addr & 0xff000000) >> 24;
898 desc.sd.sd_type = SDT_MEMRO | ((ld.read_exec_only ^ 1) << 1) |
899 (ld.contents << 2);
900 desc.sd.sd_dpl = 3;
901 desc.sd.sd_p = (ld.seg_not_present ^ 1);
902 desc.sd.sd_xx = 0;
903 desc.sd.sd_def32 = ld.seg_32bit;
904 desc.sd.sd_gran = ld.limit_in_pages;
905 error = i386_set_ldt(td, &ldt, &desc);
906 break;
907 default:
908 error = ENOSYS;
909 break;
910 }
911
912 if (error == EOPNOTSUPP) {
913 printf("linux: modify_ldt needs kernel option USER_LDT\n");
914 error = ENOSYS;
915 }
916
917 return (error);
918 }
919
920 int
921 linux_sigaction(struct thread *td, struct linux_sigaction_args *args)
922 {
923 l_osigaction_t osa;
924 l_sigaction_t act, oact;
925 int error;
926
927 #ifdef DEBUG
928 if (ldebug(sigaction))
929 printf(ARGS(sigaction, "%d, %p, %p"),
930 args->sig, (void *)args->nsa, (void *)args->osa);
931 #endif
932
933 if (args->nsa != NULL) {
934 error = copyin(args->nsa, &osa, sizeof(l_osigaction_t));
935 if (error)
936 return (error);
937 act.lsa_handler = osa.lsa_handler;
938 act.lsa_flags = osa.lsa_flags;
939 act.lsa_restorer = osa.lsa_restorer;
940 LINUX_SIGEMPTYSET(act.lsa_mask);
941 act.lsa_mask.__bits[0] = osa.lsa_mask;
942 }
943
944 error = linux_do_sigaction(td, args->sig, args->nsa ? &act : NULL,
945 args->osa ? &oact : NULL);
946
947 if (args->osa != NULL && !error) {
948 osa.lsa_handler = oact.lsa_handler;
949 osa.lsa_flags = oact.lsa_flags;
950 osa.lsa_restorer = oact.lsa_restorer;
951 osa.lsa_mask = oact.lsa_mask.__bits[0];
952 error = copyout(&osa, args->osa, sizeof(l_osigaction_t));
953 }
954
955 return (error);
956 }
957
958 /*
959 * Linux has two extra args, restart and oldmask. We dont use these,
960 * but it seems that "restart" is actually a context pointer that
961 * enables the signal to happen with a different register set.
962 */
963 int
964 linux_sigsuspend(struct thread *td, struct linux_sigsuspend_args *args)
965 {
966 sigset_t sigmask;
967 l_sigset_t mask;
968
969 #ifdef DEBUG
970 if (ldebug(sigsuspend))
971 printf(ARGS(sigsuspend, "%08lx"), (unsigned long)args->mask);
972 #endif
973
974 LINUX_SIGEMPTYSET(mask);
975 mask.__bits[0] = args->mask;
976 linux_to_bsd_sigset(&mask, &sigmask);
977 return (kern_sigsuspend(td, sigmask));
978 }
979
980 int
981 linux_rt_sigsuspend(struct thread *td, struct linux_rt_sigsuspend_args *uap)
982 {
983 l_sigset_t lmask;
984 sigset_t sigmask;
985 int error;
986
987 #ifdef DEBUG
988 if (ldebug(rt_sigsuspend))
989 printf(ARGS(rt_sigsuspend, "%p, %d"),
990 (void *)uap->newset, uap->sigsetsize);
991 #endif
992
993 if (uap->sigsetsize != sizeof(l_sigset_t))
994 return (EINVAL);
995
996 error = copyin(uap->newset, &lmask, sizeof(l_sigset_t));
997 if (error)
998 return (error);
999
1000 linux_to_bsd_sigset(&lmask, &sigmask);
1001 return (kern_sigsuspend(td, sigmask));
1002 }
1003
1004 int
1005 linux_pause(struct thread *td, struct linux_pause_args *args)
1006 {
1007 struct proc *p = td->td_proc;
1008 sigset_t sigmask;
1009
1010 #ifdef DEBUG
1011 if (ldebug(pause))
1012 printf(ARGS(pause, ""));
1013 #endif
1014
1015 PROC_LOCK(p);
1016 sigmask = td->td_sigmask;
1017 PROC_UNLOCK(p);
1018 return (kern_sigsuspend(td, sigmask));
1019 }
1020
1021 int
1022 linux_sigaltstack(struct thread *td, struct linux_sigaltstack_args *uap)
1023 {
1024 stack_t ss, oss;
1025 l_stack_t lss;
1026 int error;
1027
1028 #ifdef DEBUG
1029 if (ldebug(sigaltstack))
1030 printf(ARGS(sigaltstack, "%p, %p"), uap->uss, uap->uoss);
1031 #endif
1032
1033 if (uap->uss != NULL) {
1034 error = copyin(uap->uss, &lss, sizeof(l_stack_t));
1035 if (error)
1036 return (error);
1037
1038 ss.ss_sp = lss.ss_sp;
1039 ss.ss_size = lss.ss_size;
1040 ss.ss_flags = linux_to_bsd_sigaltstack(lss.ss_flags);
1041 }
1042 error = kern_sigaltstack(td, (uap->uss != NULL) ? &ss : NULL,
1043 (uap->uoss != NULL) ? &oss : NULL);
1044 if (!error && uap->uoss != NULL) {
1045 lss.ss_sp = oss.ss_sp;
1046 lss.ss_size = oss.ss_size;
1047 lss.ss_flags = bsd_to_linux_sigaltstack(oss.ss_flags);
1048 error = copyout(&lss, uap->uoss, sizeof(l_stack_t));
1049 }
1050
1051 return (error);
1052 }
1053
1054 int
1055 linux_ftruncate64(struct thread *td, struct linux_ftruncate64_args *args)
1056 {
1057 struct ftruncate_args sa;
1058
1059 #ifdef DEBUG
1060 if (ldebug(ftruncate64))
1061 printf(ARGS(ftruncate64, "%u, %jd"), args->fd,
1062 (intmax_t)args->length);
1063 #endif
1064
1065 sa.fd = args->fd;
1066 sa.length = args->length;
1067 return ftruncate(td, &sa);
1068 }
1069
1070 int
1071 linux_set_thread_area(struct thread *td, struct linux_set_thread_area_args *args)
1072 {
1073 struct l_user_desc info;
1074 int error;
1075 int idx;
1076 int a[2];
1077 struct segment_descriptor sd;
1078
1079 error = copyin(args->desc, &info, sizeof(struct l_user_desc));
1080 if (error)
1081 return (error);
1082
1083 #ifdef DEBUG
1084 if (ldebug(set_thread_area))
1085 printf(ARGS(set_thread_area, "%i, %x, %x, %i, %i, %i, %i, %i, %i\n"),
1086 info.entry_number,
1087 info.base_addr,
1088 info.limit,
1089 info.seg_32bit,
1090 info.contents,
1091 info.read_exec_only,
1092 info.limit_in_pages,
1093 info.seg_not_present,
1094 info.useable);
1095 #endif
1096
1097 idx = info.entry_number;
1098 /*
1099 * Semantics of linux version: every thread in the system has array of
1100 * 3 tls descriptors. 1st is GLIBC TLS, 2nd is WINE, 3rd unknown. This
1101 * syscall loads one of the selected tls decriptors with a value and
1102 * also loads GDT descriptors 6, 7 and 8 with the content of the
1103 * per-thread descriptors.
1104 *
1105 * Semantics of fbsd version: I think we can ignore that linux has 3
1106 * per-thread descriptors and use just the 1st one. The tls_array[]
1107 * is used only in set/get-thread_area() syscalls and for loading the
1108 * GDT descriptors. In fbsd we use just one GDT descriptor for TLS so
1109 * we will load just one.
1110 *
1111 * XXX: this doesn't work when a user space process tries to use more
1112 * than 1 TLS segment. Comment in the linux sources says wine might do
1113 * this.
1114 */
1115
1116 /*
1117 * we support just GLIBC TLS now
1118 * we should let 3 proceed as well because we use this segment so
1119 * if code does two subsequent calls it should succeed
1120 */
1121 if (idx != 6 && idx != -1 && idx != 3)
1122 return (EINVAL);
1123
1124 /*
1125 * we have to copy out the GDT entry we use
1126 * FreeBSD uses GDT entry #3 for storing %gs so load that
1127 *
1128 * XXX: what if a user space program doesn't check this value and tries
1129 * to use 6, 7 or 8?
1130 */
1131 idx = info.entry_number = 3;
1132 error = copyout(&info, args->desc, sizeof(struct l_user_desc));
1133 if (error)
1134 return (error);
1135
1136 if (LINUX_LDT_empty(&info)) {
1137 a[0] = 0;
1138 a[1] = 0;
1139 } else {
1140 a[0] = LINUX_LDT_entry_a(&info);
1141 a[1] = LINUX_LDT_entry_b(&info);
1142 }
1143
1144 memcpy(&sd, &a, sizeof(a));
1145 #ifdef DEBUG
1146 if (ldebug(set_thread_area))
1147 printf("Segment created in set_thread_area: lobase: %x, hibase: %x, lolimit: %x, hilimit: %x, type: %i, dpl: %i, p: %i, xx: %i, def32: %i, gran: %i\n", sd.sd_lobase,
1148 sd.sd_hibase,
1149 sd.sd_lolimit,
1150 sd.sd_hilimit,
1151 sd.sd_type,
1152 sd.sd_dpl,
1153 sd.sd_p,
1154 sd.sd_xx,
1155 sd.sd_def32,
1156 sd.sd_gran);
1157 #endif
1158
1159 /* this is taken from i386 version of cpu_set_user_tls() */
1160 critical_enter();
1161 /* set %gs */
1162 td->td_pcb->pcb_gsd = sd;
1163 PCPU_GET(fsgs_gdt)[1] = sd;
1164 load_gs(GSEL(GUGS_SEL, SEL_UPL));
1165 critical_exit();
1166
1167 return (0);
1168 }
1169
1170 int
1171 linux_get_thread_area(struct thread *td, struct linux_get_thread_area_args *args)
1172 {
1173
1174 struct l_user_desc info;
1175 int error;
1176 int idx;
1177 struct l_desc_struct desc;
1178 struct segment_descriptor sd;
1179
1180 #ifdef DEBUG
1181 if (ldebug(get_thread_area))
1182 printf(ARGS(get_thread_area, "%p"), args->desc);
1183 #endif
1184
1185 error = copyin(args->desc, &info, sizeof(struct l_user_desc));
1186 if (error)
1187 return (error);
1188
1189 idx = info.entry_number;
1190 /* XXX: I am not sure if we want 3 to be allowed too. */
1191 if (idx != 6 && idx != 3)
1192 return (EINVAL);
1193
1194 idx = 3;
1195
1196 memset(&info, 0, sizeof(info));
1197
1198 sd = PCPU_GET(fsgs_gdt)[1];
1199
1200 memcpy(&desc, &sd, sizeof(desc));
1201
1202 info.entry_number = idx;
1203 info.base_addr = LINUX_GET_BASE(&desc);
1204 info.limit = LINUX_GET_LIMIT(&desc);
1205 info.seg_32bit = LINUX_GET_32BIT(&desc);
1206 info.contents = LINUX_GET_CONTENTS(&desc);
1207 info.read_exec_only = !LINUX_GET_WRITABLE(&desc);
1208 info.limit_in_pages = LINUX_GET_LIMIT_PAGES(&desc);
1209 info.seg_not_present = !LINUX_GET_PRESENT(&desc);
1210 info.useable = LINUX_GET_USEABLE(&desc);
1211
1212 error = copyout(&info, args->desc, sizeof(struct l_user_desc));
1213 if (error)
1214 return (EFAULT);
1215
1216 return (0);
1217 }
1218
1219 /* copied from kern/kern_time.c */
1220 int
1221 linux_timer_create(struct thread *td, struct linux_timer_create_args *args)
1222 {
1223 return ktimer_create(td, (struct ktimer_create_args *) args);
1224 }
1225
1226 int
1227 linux_timer_settime(struct thread *td, struct linux_timer_settime_args *args)
1228 {
1229 return ktimer_settime(td, (struct ktimer_settime_args *) args);
1230 }
1231
1232 int
1233 linux_timer_gettime(struct thread *td, struct linux_timer_gettime_args *args)
1234 {
1235 return ktimer_gettime(td, (struct ktimer_gettime_args *) args);
1236 }
1237
1238 int
1239 linux_timer_getoverrun(struct thread *td, struct linux_timer_getoverrun_args *args)
1240 {
1241 return ktimer_getoverrun(td, (struct ktimer_getoverrun_args *) args);
1242 }
1243
1244 int
1245 linux_timer_delete(struct thread *td, struct linux_timer_delete_args *args)
1246 {
1247 return ktimer_delete(td, (struct ktimer_delete_args *) args);
1248 }
1249
1250 /* XXX: this wont work with module - convert it */
1251 int
1252 linux_mq_open(struct thread *td, struct linux_mq_open_args *args)
1253 {
1254 #ifdef P1003_1B_MQUEUE
1255 return kmq_open(td, (struct kmq_open_args *) args);
1256 #else
1257 return (ENOSYS);
1258 #endif
1259 }
1260
1261 int
1262 linux_mq_unlink(struct thread *td, struct linux_mq_unlink_args *args)
1263 {
1264 #ifdef P1003_1B_MQUEUE
1265 return kmq_unlink(td, (struct kmq_unlink_args *) args);
1266 #else
1267 return (ENOSYS);
1268 #endif
1269 }
1270
1271 int
1272 linux_mq_timedsend(struct thread *td, struct linux_mq_timedsend_args *args)
1273 {
1274 #ifdef P1003_1B_MQUEUE
1275 return kmq_timedsend(td, (struct kmq_timedsend_args *) args);
1276 #else
1277 return (ENOSYS);
1278 #endif
1279 }
1280
1281 int
1282 linux_mq_timedreceive(struct thread *td, struct linux_mq_timedreceive_args *args)
1283 {
1284 #ifdef P1003_1B_MQUEUE
1285 return kmq_timedreceive(td, (struct kmq_timedreceive_args *) args);
1286 #else
1287 return (ENOSYS);
1288 #endif
1289 }
1290
1291 int
1292 linux_mq_notify(struct thread *td, struct linux_mq_notify_args *args)
1293 {
1294 #ifdef P1003_1B_MQUEUE
1295 return kmq_notify(td, (struct kmq_notify_args *) args);
1296 #else
1297 return (ENOSYS);
1298 #endif
1299 }
1300
1301 int
1302 linux_mq_getsetattr(struct thread *td, struct linux_mq_getsetattr_args *args)
1303 {
1304 #ifdef P1003_1B_MQUEUE
1305 return kmq_setattr(td, (struct kmq_setattr_args *) args);
1306 #else
1307 return (ENOSYS);
1308 #endif
1309 }
1310
Cache object: dd7f8e04d557673911d34ce1c641816c
|