cxgbe/t4_tom: Read the chip's DDP page sizes and save them in a
[freebsd-src.git] / sys / i386 / linux / linux_sysvec.c
blob49fafdc17fe082906d045ca93456f955db612c2d
1 /*-
2 * Copyright (c) 1994-1996 Søren Schmidt
3 * All rights reserved.
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
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.
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
32 #include <sys/param.h>
33 #include <sys/systm.h>
34 #include <sys/exec.h>
35 #include <sys/fcntl.h>
36 #include <sys/imgact.h>
37 #include <sys/imgact_aout.h>
38 #include <sys/imgact_elf.h>
39 #include <sys/kernel.h>
40 #include <sys/lock.h>
41 #include <sys/malloc.h>
42 #include <sys/module.h>
43 #include <sys/mutex.h>
44 #include <sys/proc.h>
45 #include <sys/signalvar.h>
46 #include <sys/syscallsubr.h>
47 #include <sys/sysctl.h>
48 #include <sys/sysent.h>
49 #include <sys/sysproto.h>
50 #include <sys/vnode.h>
51 #include <sys/eventhandler.h>
53 #include <vm/vm.h>
54 #include <vm/pmap.h>
55 #include <vm/vm_extern.h>
56 #include <vm/vm_map.h>
57 #include <vm/vm_object.h>
58 #include <vm/vm_page.h>
59 #include <vm/vm_param.h>
61 #include <machine/cpu.h>
62 #include <machine/cputypes.h>
63 #include <machine/md_var.h>
64 #include <machine/pcb.h>
66 #include <i386/linux/linux.h>
67 #include <i386/linux/linux_proto.h>
68 #include <compat/linux/linux_emul.h>
69 #include <compat/linux/linux_futex.h>
70 #include <compat/linux/linux_ioctl.h>
71 #include <compat/linux/linux_mib.h>
72 #include <compat/linux/linux_misc.h>
73 #include <compat/linux/linux_signal.h>
74 #include <compat/linux/linux_util.h>
75 #include <compat/linux/linux_vdso.h>
77 MODULE_VERSION(linux, 1);
79 #if BYTE_ORDER == LITTLE_ENDIAN
80 #define SHELLMAGIC 0x2123 /* #! */
81 #else
82 #define SHELLMAGIC 0x2321
83 #endif
85 #if defined(DEBUG)
86 SYSCTL_PROC(_compat_linux, OID_AUTO, debug,
87 CTLTYPE_STRING | CTLFLAG_RW,
88 0, 0, linux_sysctl_debug, "A",
89 "Linux debugging control");
90 #endif
93 * Allow the sendsig functions to use the ldebug() facility
94 * even though they are not syscalls themselves. Map them
95 * to syscall 0. This is slightly less bogus than using
96 * ldebug(sigreturn).
98 #define LINUX_SYS_linux_rt_sendsig 0
99 #define LINUX_SYS_linux_sendsig 0
101 #define LINUX_PS_STRINGS (LINUX_USRSTACK - sizeof(struct ps_strings))
103 static int linux_szsigcode;
104 static vm_object_t linux_shared_page_obj;
105 static char *linux_shared_page_mapping;
106 extern char _binary_linux_locore_o_start;
107 extern char _binary_linux_locore_o_end;
109 extern struct sysent linux_sysent[LINUX_SYS_MAXSYSCALL];
111 SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler);
113 static int linux_fixup(register_t **stack_base,
114 struct image_params *iparams);
115 static int elf_linux_fixup(register_t **stack_base,
116 struct image_params *iparams);
117 static void linux_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask);
118 static void exec_linux_setregs(struct thread *td,
119 struct image_params *imgp, u_long stack);
120 static register_t *linux_copyout_strings(struct image_params *imgp);
121 static boolean_t linux_trans_osrel(const Elf_Note *note, int32_t *osrel);
122 static void linux_vdso_install(void *param);
123 static void linux_vdso_deinstall(void *param);
125 static int linux_szplatform;
126 const char *linux_kplatform;
128 static eventhandler_tag linux_exit_tag;
129 static eventhandler_tag linux_exec_tag;
130 static eventhandler_tag linux_thread_dtor_tag;
133 * Linux syscalls return negative errno's, we do positive and map them
134 * Reference:
135 * FreeBSD: src/sys/sys/errno.h
136 * Linux: linux-2.6.17.8/include/asm-generic/errno-base.h
137 * linux-2.6.17.8/include/asm-generic/errno.h
139 static int bsd_to_linux_errno[ELAST + 1] = {
140 -0, -1, -2, -3, -4, -5, -6, -7, -8, -9,
141 -10, -35, -12, -13, -14, -15, -16, -17, -18, -19,
142 -20, -21, -22, -23, -24, -25, -26, -27, -28, -29,
143 -30, -31, -32, -33, -34, -11,-115,-114, -88, -89,
144 -90, -91, -92, -93, -94, -95, -96, -97, -98, -99,
145 -100,-101,-102,-103,-104,-105,-106,-107,-108,-109,
146 -110,-111, -40, -36,-112,-113, -39, -11, -87,-122,
147 -116, -66, -6, -6, -6, -6, -6, -37, -38, -9,
148 -6, -6, -43, -42, -75,-125, -84, -95, -16, -74,
149 -72, -67, -71
152 #define LINUX_T_UNKNOWN 255
153 static int _bsd_to_linux_trapcode[] = {
154 LINUX_T_UNKNOWN, /* 0 */
155 6, /* 1 T_PRIVINFLT */
156 LINUX_T_UNKNOWN, /* 2 */
157 3, /* 3 T_BPTFLT */
158 LINUX_T_UNKNOWN, /* 4 */
159 LINUX_T_UNKNOWN, /* 5 */
160 16, /* 6 T_ARITHTRAP */
161 254, /* 7 T_ASTFLT */
162 LINUX_T_UNKNOWN, /* 8 */
163 13, /* 9 T_PROTFLT */
164 1, /* 10 T_TRCTRAP */
165 LINUX_T_UNKNOWN, /* 11 */
166 14, /* 12 T_PAGEFLT */
167 LINUX_T_UNKNOWN, /* 13 */
168 17, /* 14 T_ALIGNFLT */
169 LINUX_T_UNKNOWN, /* 15 */
170 LINUX_T_UNKNOWN, /* 16 */
171 LINUX_T_UNKNOWN, /* 17 */
172 0, /* 18 T_DIVIDE */
173 2, /* 19 T_NMI */
174 4, /* 20 T_OFLOW */
175 5, /* 21 T_BOUND */
176 7, /* 22 T_DNA */
177 8, /* 23 T_DOUBLEFLT */
178 9, /* 24 T_FPOPFLT */
179 10, /* 25 T_TSSFLT */
180 11, /* 26 T_SEGNPFLT */
181 12, /* 27 T_STKFLT */
182 18, /* 28 T_MCHK */
183 19, /* 29 T_XMMFLT */
184 15 /* 30 T_RESERVED */
186 #define bsd_to_linux_trapcode(code) \
187 ((code)<nitems(_bsd_to_linux_trapcode)? \
188 _bsd_to_linux_trapcode[(code)]: \
189 LINUX_T_UNKNOWN)
191 LINUX_VDSO_SYM_INTPTR(linux_sigcode);
192 LINUX_VDSO_SYM_INTPTR(linux_rt_sigcode);
193 LINUX_VDSO_SYM_INTPTR(linux_vsyscall);
196 * If FreeBSD & Linux have a difference of opinion about what a trap
197 * means, deal with it here.
199 * MPSAFE
201 static int
202 translate_traps(int signal, int trap_code)
204 if (signal != SIGBUS)
205 return (signal);
206 switch (trap_code) {
207 case T_PROTFLT:
208 case T_TSSFLT:
209 case T_DOUBLEFLT:
210 case T_PAGEFLT:
211 return (SIGSEGV);
212 default:
213 return (signal);
217 static int
218 linux_fixup(register_t **stack_base, struct image_params *imgp)
220 register_t *argv, *envp;
222 argv = *stack_base;
223 envp = *stack_base + (imgp->args->argc + 1);
224 (*stack_base)--;
225 suword(*stack_base, (intptr_t)(void *)envp);
226 (*stack_base)--;
227 suword(*stack_base, (intptr_t)(void *)argv);
228 (*stack_base)--;
229 suword(*stack_base, imgp->args->argc);
230 return (0);
233 static int
234 elf_linux_fixup(register_t **stack_base, struct image_params *imgp)
236 struct proc *p;
237 Elf32_Auxargs *args;
238 Elf32_Addr *uplatform;
239 struct ps_strings *arginfo;
240 register_t *pos;
241 int issetugid;
243 KASSERT(curthread->td_proc == imgp->proc,
244 ("unsafe elf_linux_fixup(), should be curproc"));
246 p = imgp->proc;
247 issetugid = imgp->proc->p_flag & P_SUGID ? 1 : 0;
248 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
249 uplatform = (Elf32_Addr *)((caddr_t)arginfo - linux_szplatform);
250 args = (Elf32_Auxargs *)imgp->auxargs;
251 pos = *stack_base + (imgp->args->argc + imgp->args->envc + 2);
253 AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO_EHDR,
254 imgp->proc->p_sysent->sv_shared_page_base);
255 AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO, linux_vsyscall);
256 AUXARGS_ENTRY(pos, LINUX_AT_HWCAP, cpu_feature);
259 * Do not export AT_CLKTCK when emulating Linux kernel prior to 2.4.0,
260 * as it has appeared in the 2.4.0-rc7 first time.
261 * Being exported, AT_CLKTCK is returned by sysconf(_SC_CLK_TCK),
262 * glibc falls back to the hard-coded CLK_TCK value when aux entry
263 * is not present.
264 * Also see linux_times() implementation.
266 if (linux_kernver(curthread) >= LINUX_KERNVER_2004000)
267 AUXARGS_ENTRY(pos, LINUX_AT_CLKTCK, stclohz);
268 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
269 AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
270 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
271 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
272 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
273 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
274 AUXARGS_ENTRY(pos, AT_BASE, args->base);
275 AUXARGS_ENTRY(pos, LINUX_AT_SECURE, issetugid);
276 AUXARGS_ENTRY(pos, AT_UID, imgp->proc->p_ucred->cr_ruid);
277 AUXARGS_ENTRY(pos, AT_EUID, imgp->proc->p_ucred->cr_svuid);
278 AUXARGS_ENTRY(pos, AT_GID, imgp->proc->p_ucred->cr_rgid);
279 AUXARGS_ENTRY(pos, AT_EGID, imgp->proc->p_ucred->cr_svgid);
280 AUXARGS_ENTRY(pos, LINUX_AT_PLATFORM, PTROUT(uplatform));
281 AUXARGS_ENTRY(pos, LINUX_AT_RANDOM, imgp->canary);
282 if (imgp->execpathp != 0)
283 AUXARGS_ENTRY(pos, LINUX_AT_EXECFN, imgp->execpathp);
284 if (args->execfd != -1)
285 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
286 AUXARGS_ENTRY(pos, AT_NULL, 0);
288 free(imgp->auxargs, M_TEMP);
289 imgp->auxargs = NULL;
291 (*stack_base)--;
292 suword(*stack_base, (register_t)imgp->args->argc);
293 return (0);
297 * Copied from kern/kern_exec.c
299 static register_t *
300 linux_copyout_strings(struct image_params *imgp)
302 int argc, envc;
303 char **vectp;
304 char *stringp, *destp;
305 register_t *stack_base;
306 struct ps_strings *arginfo;
307 char canary[LINUX_AT_RANDOM_LEN];
308 size_t execpath_len;
309 struct proc *p;
312 * Calculate string base and vector table pointers.
314 p = imgp->proc;
315 if (imgp->execpath != NULL && imgp->auxargs != NULL)
316 execpath_len = strlen(imgp->execpath) + 1;
317 else
318 execpath_len = 0;
319 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
320 destp = (caddr_t)arginfo - SPARE_USRSPACE - linux_szplatform -
321 roundup(sizeof(canary), sizeof(char *)) -
322 roundup(execpath_len, sizeof(char *)) -
323 roundup(ARG_MAX - imgp->args->stringspace, sizeof(char *));
326 * install LINUX_PLATFORM
328 copyout(linux_kplatform, ((caddr_t)arginfo - linux_szplatform),
329 linux_szplatform);
331 if (execpath_len != 0) {
332 imgp->execpathp = (uintptr_t)arginfo -
333 linux_szplatform - execpath_len;
334 copyout(imgp->execpath, (void *)imgp->execpathp, execpath_len);
338 * Prepare the canary for SSP.
340 arc4rand(canary, sizeof(canary), 0);
341 imgp->canary = (uintptr_t)arginfo - linux_szplatform -
342 roundup(execpath_len, sizeof(char *)) -
343 roundup(sizeof(canary), sizeof(char *));
344 copyout(canary, (void *)imgp->canary, sizeof(canary));
347 * If we have a valid auxargs ptr, prepare some room
348 * on the stack.
350 if (imgp->auxargs) {
352 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
353 * lower compatibility.
355 imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
356 (LINUX_AT_COUNT * 2);
358 * The '+ 2' is for the null pointers at the end of each of
359 * the arg and env vector sets,and imgp->auxarg_size is room
360 * for argument of Runtime loader.
362 vectp = (char **)(destp - (imgp->args->argc +
363 imgp->args->envc + 2 + imgp->auxarg_size) * sizeof(char *));
364 } else {
366 * The '+ 2' is for the null pointers at the end of each of
367 * the arg and env vector sets
369 vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc + 2) *
370 sizeof(char *));
374 * vectp also becomes our initial stack base
376 stack_base = (register_t *)vectp;
378 stringp = imgp->args->begin_argv;
379 argc = imgp->args->argc;
380 envc = imgp->args->envc;
383 * Copy out strings - arguments and environment.
385 copyout(stringp, destp, ARG_MAX - imgp->args->stringspace);
388 * Fill in "ps_strings" struct for ps, w, etc.
390 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
391 suword(&arginfo->ps_nargvstr, argc);
394 * Fill in argument portion of vector table.
396 for (; argc > 0; --argc) {
397 suword(vectp++, (long)(intptr_t)destp);
398 while (*stringp++ != 0)
399 destp++;
400 destp++;
403 /* a null vector table pointer separates the argp's from the envp's */
404 suword(vectp++, 0);
406 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
407 suword(&arginfo->ps_nenvstr, envc);
410 * Fill in environment portion of vector table.
412 for (; envc > 0; --envc) {
413 suword(vectp++, (long)(intptr_t)destp);
414 while (*stringp++ != 0)
415 destp++;
416 destp++;
419 /* end of vector table is a null pointer */
420 suword(vectp, 0);
422 return (stack_base);
425 static void
426 linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
428 struct thread *td = curthread;
429 struct proc *p = td->td_proc;
430 struct sigacts *psp;
431 struct trapframe *regs;
432 struct l_rt_sigframe *fp, frame;
433 int sig, code;
434 int oonstack;
436 sig = ksi->ksi_signo;
437 code = ksi->ksi_code;
438 PROC_LOCK_ASSERT(p, MA_OWNED);
439 psp = p->p_sigacts;
440 mtx_assert(&psp->ps_mtx, MA_OWNED);
441 regs = td->td_frame;
442 oonstack = sigonstack(regs->tf_esp);
444 #ifdef DEBUG
445 if (ldebug(rt_sendsig))
446 printf(ARGS(rt_sendsig, "%p, %d, %p, %u"),
447 catcher, sig, (void*)mask, code);
448 #endif
450 * Allocate space for the signal handler context.
452 if ((td->td_pflags & TDP_ALTSTACK) && !oonstack &&
453 SIGISMEMBER(psp->ps_sigonstack, sig)) {
454 fp = (struct l_rt_sigframe *)((uintptr_t)td->td_sigstk.ss_sp +
455 td->td_sigstk.ss_size - sizeof(struct l_rt_sigframe));
456 } else
457 fp = (struct l_rt_sigframe *)regs->tf_esp - 1;
458 mtx_unlock(&psp->ps_mtx);
461 * Build the argument list for the signal handler.
463 sig = bsd_to_linux_signal(sig);
465 bzero(&frame, sizeof(frame));
467 frame.sf_handler = catcher;
468 frame.sf_sig = sig;
469 frame.sf_siginfo = &fp->sf_si;
470 frame.sf_ucontext = &fp->sf_sc;
472 /* Fill in POSIX parts */
473 ksiginfo_to_lsiginfo(ksi, &frame.sf_si, sig);
476 * Build the signal context to be used by sigreturn.
478 frame.sf_sc.uc_flags = 0; /* XXX ??? */
479 frame.sf_sc.uc_link = NULL; /* XXX ??? */
481 frame.sf_sc.uc_stack.ss_sp = td->td_sigstk.ss_sp;
482 frame.sf_sc.uc_stack.ss_size = td->td_sigstk.ss_size;
483 frame.sf_sc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK)
484 ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE;
485 PROC_UNLOCK(p);
487 bsd_to_linux_sigset(mask, &frame.sf_sc.uc_sigmask);
489 frame.sf_sc.uc_mcontext.sc_mask = frame.sf_sc.uc_sigmask.__mask;
490 frame.sf_sc.uc_mcontext.sc_gs = rgs();
491 frame.sf_sc.uc_mcontext.sc_fs = regs->tf_fs;
492 frame.sf_sc.uc_mcontext.sc_es = regs->tf_es;
493 frame.sf_sc.uc_mcontext.sc_ds = regs->tf_ds;
494 frame.sf_sc.uc_mcontext.sc_edi = regs->tf_edi;
495 frame.sf_sc.uc_mcontext.sc_esi = regs->tf_esi;
496 frame.sf_sc.uc_mcontext.sc_ebp = regs->tf_ebp;
497 frame.sf_sc.uc_mcontext.sc_ebx = regs->tf_ebx;
498 frame.sf_sc.uc_mcontext.sc_esp = regs->tf_esp;
499 frame.sf_sc.uc_mcontext.sc_edx = regs->tf_edx;
500 frame.sf_sc.uc_mcontext.sc_ecx = regs->tf_ecx;
501 frame.sf_sc.uc_mcontext.sc_eax = regs->tf_eax;
502 frame.sf_sc.uc_mcontext.sc_eip = regs->tf_eip;
503 frame.sf_sc.uc_mcontext.sc_cs = regs->tf_cs;
504 frame.sf_sc.uc_mcontext.sc_eflags = regs->tf_eflags;
505 frame.sf_sc.uc_mcontext.sc_esp_at_signal = regs->tf_esp;
506 frame.sf_sc.uc_mcontext.sc_ss = regs->tf_ss;
507 frame.sf_sc.uc_mcontext.sc_err = regs->tf_err;
508 frame.sf_sc.uc_mcontext.sc_cr2 = (register_t)ksi->ksi_addr;
509 frame.sf_sc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code);
511 #ifdef DEBUG
512 if (ldebug(rt_sendsig))
513 printf(LMSG("rt_sendsig flags: 0x%x, sp: %p, ss: 0x%x, mask: 0x%x"),
514 frame.sf_sc.uc_stack.ss_flags, td->td_sigstk.ss_sp,
515 td->td_sigstk.ss_size, frame.sf_sc.uc_mcontext.sc_mask);
516 #endif
518 if (copyout(&frame, fp, sizeof(frame)) != 0) {
520 * Process has trashed its stack; give it an illegal
521 * instruction to halt it in its tracks.
523 #ifdef DEBUG
524 if (ldebug(rt_sendsig))
525 printf(LMSG("rt_sendsig: bad stack %p, oonstack=%x"),
526 fp, oonstack);
527 #endif
528 PROC_LOCK(p);
529 sigexit(td, SIGILL);
533 * Build context to run handler in.
535 regs->tf_esp = (int)fp;
536 regs->tf_eip = linux_rt_sigcode;
537 regs->tf_eflags &= ~(PSL_T | PSL_VM | PSL_D);
538 regs->tf_cs = _ucodesel;
539 regs->tf_ds = _udatasel;
540 regs->tf_es = _udatasel;
541 regs->tf_fs = _udatasel;
542 regs->tf_ss = _udatasel;
543 PROC_LOCK(p);
544 mtx_lock(&psp->ps_mtx);
549 * Send an interrupt to process.
551 * Stack is set up to allow sigcode stored
552 * in u. to call routine, followed by kcall
553 * to sigreturn routine below. After sigreturn
554 * resets the signal mask, the stack, and the
555 * frame pointer, it returns to the user
556 * specified pc, psl.
558 static void
559 linux_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
561 struct thread *td = curthread;
562 struct proc *p = td->td_proc;
563 struct sigacts *psp;
564 struct trapframe *regs;
565 struct l_sigframe *fp, frame;
566 l_sigset_t lmask;
567 int sig, code;
568 int oonstack;
570 PROC_LOCK_ASSERT(p, MA_OWNED);
571 psp = p->p_sigacts;
572 sig = ksi->ksi_signo;
573 code = ksi->ksi_code;
574 mtx_assert(&psp->ps_mtx, MA_OWNED);
575 if (SIGISMEMBER(psp->ps_siginfo, sig)) {
576 /* Signal handler installed with SA_SIGINFO. */
577 linux_rt_sendsig(catcher, ksi, mask);
578 return;
580 regs = td->td_frame;
581 oonstack = sigonstack(regs->tf_esp);
583 #ifdef DEBUG
584 if (ldebug(sendsig))
585 printf(ARGS(sendsig, "%p, %d, %p, %u"),
586 catcher, sig, (void*)mask, code);
587 #endif
590 * Allocate space for the signal handler context.
592 if ((td->td_pflags & TDP_ALTSTACK) && !oonstack &&
593 SIGISMEMBER(psp->ps_sigonstack, sig)) {
594 fp = (struct l_sigframe *)((uintptr_t)td->td_sigstk.ss_sp +
595 td->td_sigstk.ss_size - sizeof(struct l_sigframe));
596 } else
597 fp = (struct l_sigframe *)regs->tf_esp - 1;
598 mtx_unlock(&psp->ps_mtx);
599 PROC_UNLOCK(p);
602 * Build the argument list for the signal handler.
604 sig = bsd_to_linux_signal(sig);
606 bzero(&frame, sizeof(frame));
608 frame.sf_handler = catcher;
609 frame.sf_sig = sig;
611 bsd_to_linux_sigset(mask, &lmask);
614 * Build the signal context to be used by sigreturn.
616 frame.sf_sc.sc_mask = lmask.__mask;
617 frame.sf_sc.sc_gs = rgs();
618 frame.sf_sc.sc_fs = regs->tf_fs;
619 frame.sf_sc.sc_es = regs->tf_es;
620 frame.sf_sc.sc_ds = regs->tf_ds;
621 frame.sf_sc.sc_edi = regs->tf_edi;
622 frame.sf_sc.sc_esi = regs->tf_esi;
623 frame.sf_sc.sc_ebp = regs->tf_ebp;
624 frame.sf_sc.sc_ebx = regs->tf_ebx;
625 frame.sf_sc.sc_esp = regs->tf_esp;
626 frame.sf_sc.sc_edx = regs->tf_edx;
627 frame.sf_sc.sc_ecx = regs->tf_ecx;
628 frame.sf_sc.sc_eax = regs->tf_eax;
629 frame.sf_sc.sc_eip = regs->tf_eip;
630 frame.sf_sc.sc_cs = regs->tf_cs;
631 frame.sf_sc.sc_eflags = regs->tf_eflags;
632 frame.sf_sc.sc_esp_at_signal = regs->tf_esp;
633 frame.sf_sc.sc_ss = regs->tf_ss;
634 frame.sf_sc.sc_err = regs->tf_err;
635 frame.sf_sc.sc_cr2 = (register_t)ksi->ksi_addr;
636 frame.sf_sc.sc_trapno = bsd_to_linux_trapcode(ksi->ksi_trapno);
638 frame.sf_extramask[0] = lmask.__mask;
640 if (copyout(&frame, fp, sizeof(frame)) != 0) {
642 * Process has trashed its stack; give it an illegal
643 * instruction to halt it in its tracks.
645 PROC_LOCK(p);
646 sigexit(td, SIGILL);
650 * Build context to run handler in.
652 regs->tf_esp = (int)fp;
653 regs->tf_eip = linux_sigcode;
654 regs->tf_eflags &= ~(PSL_T | PSL_VM | PSL_D);
655 regs->tf_cs = _ucodesel;
656 regs->tf_ds = _udatasel;
657 regs->tf_es = _udatasel;
658 regs->tf_fs = _udatasel;
659 regs->tf_ss = _udatasel;
660 PROC_LOCK(p);
661 mtx_lock(&psp->ps_mtx);
665 * System call to cleanup state after a signal
666 * has been taken. Reset signal mask and
667 * stack state from context left by sendsig (above).
668 * Return to previous pc and psl as specified by
669 * context left by sendsig. Check carefully to
670 * make sure that the user has not modified the
671 * psl to gain improper privileges or to cause
672 * a machine fault.
675 linux_sigreturn(struct thread *td, struct linux_sigreturn_args *args)
677 struct l_sigframe frame;
678 struct trapframe *regs;
679 l_sigset_t lmask;
680 sigset_t bmask;
681 int eflags;
682 ksiginfo_t ksi;
684 regs = td->td_frame;
686 #ifdef DEBUG
687 if (ldebug(sigreturn))
688 printf(ARGS(sigreturn, "%p"), (void *)args->sfp);
689 #endif
691 * The trampoline code hands us the sigframe.
692 * It is unsafe to keep track of it ourselves, in the event that a
693 * program jumps out of a signal handler.
695 if (copyin(args->sfp, &frame, sizeof(frame)) != 0)
696 return (EFAULT);
699 * Check for security violations.
701 #define EFLAGS_SECURE(ef, oef) ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
702 eflags = frame.sf_sc.sc_eflags;
703 if (!EFLAGS_SECURE(eflags, regs->tf_eflags))
704 return (EINVAL);
707 * Don't allow users to load a valid privileged %cs. Let the
708 * hardware check for invalid selectors, excess privilege in
709 * other selectors, invalid %eip's and invalid %esp's.
711 #define CS_SECURE(cs) (ISPL(cs) == SEL_UPL)
712 if (!CS_SECURE(frame.sf_sc.sc_cs)) {
713 ksiginfo_init_trap(&ksi);
714 ksi.ksi_signo = SIGBUS;
715 ksi.ksi_code = BUS_OBJERR;
716 ksi.ksi_trapno = T_PROTFLT;
717 ksi.ksi_addr = (void *)regs->tf_eip;
718 trapsignal(td, &ksi);
719 return (EINVAL);
722 lmask.__mask = frame.sf_sc.sc_mask;
723 linux_to_bsd_sigset(&lmask, &bmask);
724 kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0);
727 * Restore signal context.
729 /* %gs was restored by the trampoline. */
730 regs->tf_fs = frame.sf_sc.sc_fs;
731 regs->tf_es = frame.sf_sc.sc_es;
732 regs->tf_ds = frame.sf_sc.sc_ds;
733 regs->tf_edi = frame.sf_sc.sc_edi;
734 regs->tf_esi = frame.sf_sc.sc_esi;
735 regs->tf_ebp = frame.sf_sc.sc_ebp;
736 regs->tf_ebx = frame.sf_sc.sc_ebx;
737 regs->tf_edx = frame.sf_sc.sc_edx;
738 regs->tf_ecx = frame.sf_sc.sc_ecx;
739 regs->tf_eax = frame.sf_sc.sc_eax;
740 regs->tf_eip = frame.sf_sc.sc_eip;
741 regs->tf_cs = frame.sf_sc.sc_cs;
742 regs->tf_eflags = eflags;
743 regs->tf_esp = frame.sf_sc.sc_esp_at_signal;
744 regs->tf_ss = frame.sf_sc.sc_ss;
746 return (EJUSTRETURN);
750 * System call to cleanup state after a signal
751 * has been taken. Reset signal mask and
752 * stack state from context left by rt_sendsig (above).
753 * Return to previous pc and psl as specified by
754 * context left by sendsig. Check carefully to
755 * make sure that the user has not modified the
756 * psl to gain improper privileges or to cause
757 * a machine fault.
760 linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args)
762 struct l_ucontext uc;
763 struct l_sigcontext *context;
764 sigset_t bmask;
765 l_stack_t *lss;
766 stack_t ss;
767 struct trapframe *regs;
768 int eflags;
769 ksiginfo_t ksi;
771 regs = td->td_frame;
773 #ifdef DEBUG
774 if (ldebug(rt_sigreturn))
775 printf(ARGS(rt_sigreturn, "%p"), (void *)args->ucp);
776 #endif
778 * The trampoline code hands us the ucontext.
779 * It is unsafe to keep track of it ourselves, in the event that a
780 * program jumps out of a signal handler.
782 if (copyin(args->ucp, &uc, sizeof(uc)) != 0)
783 return (EFAULT);
785 context = &uc.uc_mcontext;
788 * Check for security violations.
790 #define EFLAGS_SECURE(ef, oef) ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
791 eflags = context->sc_eflags;
792 if (!EFLAGS_SECURE(eflags, regs->tf_eflags))
793 return (EINVAL);
796 * Don't allow users to load a valid privileged %cs. Let the
797 * hardware check for invalid selectors, excess privilege in
798 * other selectors, invalid %eip's and invalid %esp's.
800 #define CS_SECURE(cs) (ISPL(cs) == SEL_UPL)
801 if (!CS_SECURE(context->sc_cs)) {
802 ksiginfo_init_trap(&ksi);
803 ksi.ksi_signo = SIGBUS;
804 ksi.ksi_code = BUS_OBJERR;
805 ksi.ksi_trapno = T_PROTFLT;
806 ksi.ksi_addr = (void *)regs->tf_eip;
807 trapsignal(td, &ksi);
808 return (EINVAL);
811 linux_to_bsd_sigset(&uc.uc_sigmask, &bmask);
812 kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0);
815 * Restore signal context
817 /* %gs was restored by the trampoline. */
818 regs->tf_fs = context->sc_fs;
819 regs->tf_es = context->sc_es;
820 regs->tf_ds = context->sc_ds;
821 regs->tf_edi = context->sc_edi;
822 regs->tf_esi = context->sc_esi;
823 regs->tf_ebp = context->sc_ebp;
824 regs->tf_ebx = context->sc_ebx;
825 regs->tf_edx = context->sc_edx;
826 regs->tf_ecx = context->sc_ecx;
827 regs->tf_eax = context->sc_eax;
828 regs->tf_eip = context->sc_eip;
829 regs->tf_cs = context->sc_cs;
830 regs->tf_eflags = eflags;
831 regs->tf_esp = context->sc_esp_at_signal;
832 regs->tf_ss = context->sc_ss;
835 * call sigaltstack & ignore results..
837 lss = &uc.uc_stack;
838 ss.ss_sp = lss->ss_sp;
839 ss.ss_size = lss->ss_size;
840 ss.ss_flags = linux_to_bsd_sigaltstack(lss->ss_flags);
842 #ifdef DEBUG
843 if (ldebug(rt_sigreturn))
844 printf(LMSG("rt_sigret flags: 0x%x, sp: %p, ss: 0x%x, mask: 0x%x"),
845 ss.ss_flags, ss.ss_sp, ss.ss_size, context->sc_mask);
846 #endif
847 (void)kern_sigaltstack(td, &ss, NULL);
849 return (EJUSTRETURN);
852 static int
853 linux_fetch_syscall_args(struct thread *td, struct syscall_args *sa)
855 struct proc *p;
856 struct trapframe *frame;
858 p = td->td_proc;
859 frame = td->td_frame;
861 sa->code = frame->tf_eax;
862 sa->args[0] = frame->tf_ebx;
863 sa->args[1] = frame->tf_ecx;
864 sa->args[2] = frame->tf_edx;
865 sa->args[3] = frame->tf_esi;
866 sa->args[4] = frame->tf_edi;
867 sa->args[5] = frame->tf_ebp; /* Unconfirmed */
869 if (sa->code >= p->p_sysent->sv_size)
870 /* nosys */
871 sa->callp = &p->p_sysent->sv_table[p->p_sysent->sv_size - 1];
872 else
873 sa->callp = &p->p_sysent->sv_table[sa->code];
874 sa->narg = sa->callp->sy_narg;
876 td->td_retval[0] = 0;
877 td->td_retval[1] = frame->tf_edx;
879 return (0);
883 * If a linux binary is exec'ing something, try this image activator
884 * first. We override standard shell script execution in order to
885 * be able to modify the interpreter path. We only do this if a linux
886 * binary is doing the exec, so we do not create an EXEC module for it.
888 static int exec_linux_imgact_try(struct image_params *iparams);
890 static int
891 exec_linux_imgact_try(struct image_params *imgp)
893 const char *head = (const char *)imgp->image_header;
894 char *rpath;
895 int error = -1;
898 * The interpreter for shell scripts run from a linux binary needs
899 * to be located in /compat/linux if possible in order to recursively
900 * maintain linux path emulation.
902 if (((const short *)head)[0] == SHELLMAGIC) {
904 * Run our normal shell image activator. If it succeeds attempt
905 * to use the alternate path for the interpreter. If an alternate
906 * path is found, use our stringspace to store it.
908 if ((error = exec_shell_imgact(imgp)) == 0) {
909 linux_emul_convpath(FIRST_THREAD_IN_PROC(imgp->proc),
910 imgp->interpreter_name, UIO_SYSSPACE, &rpath, 0, AT_FDCWD);
911 if (rpath != NULL)
912 imgp->args->fname_buf =
913 imgp->interpreter_name = rpath;
916 return (error);
920 * exec_setregs may initialize some registers differently than Linux
921 * does, thus potentially confusing Linux binaries. If necessary, we
922 * override the exec_setregs default(s) here.
924 static void
925 exec_linux_setregs(struct thread *td, struct image_params *imgp, u_long stack)
927 struct pcb *pcb = td->td_pcb;
929 exec_setregs(td, imgp, stack);
931 /* Linux sets %gs to 0, we default to _udatasel */
932 pcb->pcb_gs = 0;
933 load_gs(0);
935 pcb->pcb_initial_npxcw = __LINUX_NPXCW__;
938 static void
939 linux_get_machine(const char **dst)
942 switch (cpu_class) {
943 case CPUCLASS_686:
944 *dst = "i686";
945 break;
946 case CPUCLASS_586:
947 *dst = "i586";
948 break;
949 case CPUCLASS_486:
950 *dst = "i486";
951 break;
952 default:
953 *dst = "i386";
957 struct sysentvec linux_sysvec = {
958 .sv_size = LINUX_SYS_MAXSYSCALL,
959 .sv_table = linux_sysent,
960 .sv_mask = 0,
961 .sv_errsize = ELAST + 1,
962 .sv_errtbl = bsd_to_linux_errno,
963 .sv_transtrap = translate_traps,
964 .sv_fixup = linux_fixup,
965 .sv_sendsig = linux_sendsig,
966 .sv_sigcode = &_binary_linux_locore_o_start,
967 .sv_szsigcode = &linux_szsigcode,
968 .sv_name = "Linux a.out",
969 .sv_coredump = NULL,
970 .sv_imgact_try = exec_linux_imgact_try,
971 .sv_minsigstksz = LINUX_MINSIGSTKSZ,
972 .sv_pagesize = PAGE_SIZE,
973 .sv_minuser = VM_MIN_ADDRESS,
974 .sv_maxuser = VM_MAXUSER_ADDRESS,
975 .sv_usrstack = LINUX_USRSTACK,
976 .sv_psstrings = PS_STRINGS,
977 .sv_stackprot = VM_PROT_ALL,
978 .sv_copyout_strings = exec_copyout_strings,
979 .sv_setregs = exec_linux_setregs,
980 .sv_fixlimit = NULL,
981 .sv_maxssiz = NULL,
982 .sv_flags = SV_ABI_LINUX | SV_AOUT | SV_IA32 | SV_ILP32,
983 .sv_set_syscall_retval = cpu_set_syscall_retval,
984 .sv_fetch_syscall_args = linux_fetch_syscall_args,
985 .sv_syscallnames = NULL,
986 .sv_shared_page_base = LINUX_SHAREDPAGE,
987 .sv_shared_page_len = PAGE_SIZE,
988 .sv_schedtail = linux_schedtail,
989 .sv_thread_detach = linux_thread_detach,
990 .sv_trap = NULL,
992 INIT_SYSENTVEC(aout_sysvec, &linux_sysvec);
994 struct sysentvec elf_linux_sysvec = {
995 .sv_size = LINUX_SYS_MAXSYSCALL,
996 .sv_table = linux_sysent,
997 .sv_mask = 0,
998 .sv_errsize = ELAST + 1,
999 .sv_errtbl = bsd_to_linux_errno,
1000 .sv_transtrap = translate_traps,
1001 .sv_fixup = elf_linux_fixup,
1002 .sv_sendsig = linux_sendsig,
1003 .sv_sigcode = &_binary_linux_locore_o_start,
1004 .sv_szsigcode = &linux_szsigcode,
1005 .sv_name = "Linux ELF",
1006 .sv_coredump = elf32_coredump,
1007 .sv_imgact_try = exec_linux_imgact_try,
1008 .sv_minsigstksz = LINUX_MINSIGSTKSZ,
1009 .sv_pagesize = PAGE_SIZE,
1010 .sv_minuser = VM_MIN_ADDRESS,
1011 .sv_maxuser = VM_MAXUSER_ADDRESS,
1012 .sv_usrstack = LINUX_USRSTACK,
1013 .sv_psstrings = LINUX_PS_STRINGS,
1014 .sv_stackprot = VM_PROT_ALL,
1015 .sv_copyout_strings = linux_copyout_strings,
1016 .sv_setregs = exec_linux_setregs,
1017 .sv_fixlimit = NULL,
1018 .sv_maxssiz = NULL,
1019 .sv_flags = SV_ABI_LINUX | SV_IA32 | SV_ILP32 | SV_SHP,
1020 .sv_set_syscall_retval = cpu_set_syscall_retval,
1021 .sv_fetch_syscall_args = linux_fetch_syscall_args,
1022 .sv_syscallnames = NULL,
1023 .sv_shared_page_base = LINUX_SHAREDPAGE,
1024 .sv_shared_page_len = PAGE_SIZE,
1025 .sv_schedtail = linux_schedtail,
1026 .sv_thread_detach = linux_thread_detach,
1027 .sv_trap = NULL,
1030 static void
1031 linux_vdso_install(void *param)
1034 linux_szsigcode = (&_binary_linux_locore_o_end -
1035 &_binary_linux_locore_o_start);
1037 if (linux_szsigcode > elf_linux_sysvec.sv_shared_page_len)
1038 panic("Linux invalid vdso size\n");
1040 __elfN(linux_vdso_fixup)(&elf_linux_sysvec);
1042 linux_shared_page_obj = __elfN(linux_shared_page_init)
1043 (&linux_shared_page_mapping);
1045 __elfN(linux_vdso_reloc)(&elf_linux_sysvec, LINUX_SHAREDPAGE);
1047 bcopy(elf_linux_sysvec.sv_sigcode, linux_shared_page_mapping,
1048 linux_szsigcode);
1049 elf_linux_sysvec.sv_shared_page_obj = linux_shared_page_obj;
1051 SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC, SI_ORDER_ANY,
1052 (sysinit_cfunc_t)linux_vdso_install, NULL);
1054 static void
1055 linux_vdso_deinstall(void *param)
1058 __elfN(linux_shared_page_fini)(linux_shared_page_obj);
1060 SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST,
1061 (sysinit_cfunc_t)linux_vdso_deinstall, NULL);
1063 static char GNU_ABI_VENDOR[] = "GNU";
1064 static int GNULINUX_ABI_DESC = 0;
1066 static boolean_t
1067 linux_trans_osrel(const Elf_Note *note, int32_t *osrel)
1069 const Elf32_Word *desc;
1070 uintptr_t p;
1072 p = (uintptr_t)(note + 1);
1073 p += roundup2(note->n_namesz, sizeof(Elf32_Addr));
1075 desc = (const Elf32_Word *)p;
1076 if (desc[0] != GNULINUX_ABI_DESC)
1077 return (FALSE);
1080 * For linux we encode osrel as follows (see linux_mib.c):
1081 * VVVMMMIII (version, major, minor), see linux_mib.c.
1083 *osrel = desc[1] * 1000000 + desc[2] * 1000 + desc[3];
1085 return (TRUE);
1088 static Elf_Brandnote linux_brandnote = {
1089 .hdr.n_namesz = sizeof(GNU_ABI_VENDOR),
1090 .hdr.n_descsz = 16, /* XXX at least 16 */
1091 .hdr.n_type = 1,
1092 .vendor = GNU_ABI_VENDOR,
1093 .flags = BN_TRANSLATE_OSREL,
1094 .trans_osrel = linux_trans_osrel
1097 static Elf32_Brandinfo linux_brand = {
1098 .brand = ELFOSABI_LINUX,
1099 .machine = EM_386,
1100 .compat_3_brand = "Linux",
1101 .emul_path = "/compat/linux",
1102 .interp_path = "/lib/ld-linux.so.1",
1103 .sysvec = &elf_linux_sysvec,
1104 .interp_newpath = NULL,
1105 .brand_note = &linux_brandnote,
1106 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE
1109 static Elf32_Brandinfo linux_glibc2brand = {
1110 .brand = ELFOSABI_LINUX,
1111 .machine = EM_386,
1112 .compat_3_brand = "Linux",
1113 .emul_path = "/compat/linux",
1114 .interp_path = "/lib/ld-linux.so.2",
1115 .sysvec = &elf_linux_sysvec,
1116 .interp_newpath = NULL,
1117 .brand_note = &linux_brandnote,
1118 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE
1121 Elf32_Brandinfo *linux_brandlist[] = {
1122 &linux_brand,
1123 &linux_glibc2brand,
1124 NULL
1127 static int
1128 linux_elf_modevent(module_t mod, int type, void *data)
1130 Elf32_Brandinfo **brandinfo;
1131 int error;
1132 struct linux_ioctl_handler **lihp;
1134 error = 0;
1136 switch(type) {
1137 case MOD_LOAD:
1138 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
1139 ++brandinfo)
1140 if (elf32_insert_brand_entry(*brandinfo) < 0)
1141 error = EINVAL;
1142 if (error == 0) {
1143 SET_FOREACH(lihp, linux_ioctl_handler_set)
1144 linux_ioctl_register_handler(*lihp);
1145 LIST_INIT(&futex_list);
1146 mtx_init(&futex_mtx, "ftllk", NULL, MTX_DEF);
1147 linux_exit_tag = EVENTHANDLER_REGISTER(process_exit, linux_proc_exit,
1148 NULL, 1000);
1149 linux_exec_tag = EVENTHANDLER_REGISTER(process_exec, linux_proc_exec,
1150 NULL, 1000);
1151 linux_thread_dtor_tag = EVENTHANDLER_REGISTER(thread_dtor,
1152 linux_thread_dtor, NULL, EVENTHANDLER_PRI_ANY);
1153 linux_get_machine(&linux_kplatform);
1154 linux_szplatform = roundup(strlen(linux_kplatform) + 1,
1155 sizeof(char *));
1156 linux_osd_jail_register();
1157 stclohz = (stathz ? stathz : hz);
1158 if (bootverbose)
1159 printf("Linux ELF exec handler installed\n");
1160 } else
1161 printf("cannot insert Linux ELF brand handler\n");
1162 break;
1163 case MOD_UNLOAD:
1164 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
1165 ++brandinfo)
1166 if (elf32_brand_inuse(*brandinfo))
1167 error = EBUSY;
1168 if (error == 0) {
1169 for (brandinfo = &linux_brandlist[0];
1170 *brandinfo != NULL; ++brandinfo)
1171 if (elf32_remove_brand_entry(*brandinfo) < 0)
1172 error = EINVAL;
1174 if (error == 0) {
1175 SET_FOREACH(lihp, linux_ioctl_handler_set)
1176 linux_ioctl_unregister_handler(*lihp);
1177 mtx_destroy(&futex_mtx);
1178 EVENTHANDLER_DEREGISTER(process_exit, linux_exit_tag);
1179 EVENTHANDLER_DEREGISTER(process_exec, linux_exec_tag);
1180 EVENTHANDLER_DEREGISTER(thread_dtor, linux_thread_dtor_tag);
1181 linux_osd_jail_deregister();
1182 if (bootverbose)
1183 printf("Linux ELF exec handler removed\n");
1184 } else
1185 printf("Could not deinstall ELF interpreter entry\n");
1186 break;
1187 default:
1188 return (EOPNOTSUPP);
1190 return (error);
1193 static moduledata_t linux_elf_mod = {
1194 "linuxelf",
1195 linux_elf_modevent,
1199 DECLARE_MODULE_TIED(linuxelf, linux_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY);
1200 FEATURE(linux, "Linux 32bit support");