1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21
22 /*
23 * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
24 */
25
26 #include <sys/errno.h>
27 #include <sys/exec.h>
28 #include <sys/file.h>
29 #include <sys/kmem.h>
30 #include <sys/modctl.h>
31 #include <sys/model.h>
32 #include <sys/proc.h>
33 #include <sys/syscall.h>
34 #include <sys/systm.h>
35 #include <sys/thread.h>
36 #include <sys/cmn_err.h>
37 #include <sys/archsystm.h>
38 #include <sys/pathname.h>
39 #include <sys/sunddi.h>
40
41 #include <sys/machbrand.h>
42 #include <sys/brand.h>
43 #include "s10_brand.h"
44
45 char *s10_emulation_table = NULL;
46
47 void s10_init_brand_data(zone_t *);
48 void s10_free_brand_data(zone_t *);
49 void s10_setbrand(proc_t *);
50 int s10_getattr(zone_t *, int, void *, size_t *);
51 int s10_setattr(zone_t *, int, void *, size_t);
52 int s10_brandsys(int, int64_t *, uintptr_t, uintptr_t, uintptr_t,
53 uintptr_t, uintptr_t, uintptr_t);
54 void s10_copy_procdata(proc_t *, proc_t *);
55 void s10_proc_exit(struct proc *, klwp_t *);
56 void s10_exec();
57 int s10_initlwp(klwp_t *);
58 void s10_forklwp(klwp_t *, klwp_t *);
59 void s10_freelwp(klwp_t *);
60 void s10_lwpexit(klwp_t *);
61 int s10_elfexec(vnode_t *, execa_t *, uarg_t *, intpdata_t *, int,
62 long *, int, caddr_t, cred_t *, int);
63 void s10_sigset_native_to_s10(sigset_t *);
64 void s10_sigset_s10_to_native(sigset_t *);
65
66 /* s10 brand */
67 struct brand_ops s10_brops = {
68 s10_init_brand_data,
69 s10_free_brand_data,
70 s10_brandsys,
71 s10_setbrand,
72 s10_getattr,
73 s10_setattr,
74 s10_copy_procdata,
75 s10_proc_exit,
76 s10_exec,
77 lwp_setrval,
78 s10_initlwp,
79 s10_forklwp,
80 s10_freelwp,
81 s10_lwpexit,
82 s10_elfexec,
83 s10_sigset_native_to_s10,
84 s10_sigset_s10_to_native,
85 S10_NSIG,
86 };
87
88 #ifdef sparc
89
90 struct brand_mach_ops s10_mops = {
91 s10_brand_syscall_callback,
92 s10_brand_syscall32_callback
93 };
94
95 #else /* sparc */
96
97 #ifdef __amd64
98
99 struct brand_mach_ops s10_mops = {
100 s10_brand_sysenter_callback,
101 NULL,
102 s10_brand_int91_callback,
103 s10_brand_syscall_callback,
104 s10_brand_syscall32_callback,
105 NULL
106 };
107
108 #else /* ! __amd64 */
109
110 struct brand_mach_ops s10_mops = {
111 s10_brand_sysenter_callback,
112 NULL,
113 NULL,
114 s10_brand_syscall_callback,
115 NULL,
116 NULL
117 };
118 #endif /* __amd64 */
119
120 #endif /* _sparc */
121
122 struct brand s10_brand = {
123 BRAND_VER_1,
124 "solaris10",
125 &s10_brops,
126 &s10_mops
127 };
128
129 static struct modlbrand modlbrand = {
130 &mod_brandops, /* type of module */
131 "Solaris 10 Brand", /* description of module */
132 &s10_brand /* driver ops */
133 };
134
135 static struct modlinkage modlinkage = {
136 MODREV_1, (void *)&modlbrand, NULL
137 };
138
139 void
140 s10_setbrand(proc_t *p)
141 {
142 brand_solaris_setbrand(p, &s10_brand);
143 }
144
145 /*ARGSUSED*/
146 int
147 s10_getattr(zone_t *zone, int attr, void *buf, size_t *bufsize)
148 {
149 ASSERT(zone->zone_brand == &s10_brand);
150 if (attr == S10_EMUL_BITMAP) {
151 if (buf == NULL || *bufsize != sizeof (s10_emul_bitmap_t))
152 return (EINVAL);
153 if (copyout(((s10_zone_data_t *)zone->zone_brand_data)->
154 emul_bitmap, buf, sizeof (s10_emul_bitmap_t)) != 0)
155 return (EFAULT);
156 return (0);
157 }
158
159 return (EINVAL);
160 }
161
162 int
163 s10_setattr(zone_t *zone, int attr, void *buf, size_t bufsize)
164 {
165 ASSERT(zone->zone_brand == &s10_brand);
166 if (attr == S10_EMUL_BITMAP) {
167 if (buf == NULL || bufsize != sizeof (s10_emul_bitmap_t))
168 return (EINVAL);
169 if (copyin(buf, ((s10_zone_data_t *)zone->zone_brand_data)->
170 emul_bitmap, sizeof (s10_emul_bitmap_t)) != 0)
171 return (EFAULT);
172 return (0);
173 }
174
175 return (EINVAL);
176 }
177
178 #ifdef __amd64
179 /*
180 * The Nevada kernel clears %fs for threads in 64-bit x86 processes but S10's
181 * libc expects %fs to be nonzero. This causes some committed
182 * libc/libthread interfaces (e.g., thr_main()) to fail, which impacts several
183 * libraries, including libdoor. This function sets the specified LWP's %fs
184 * register to the legacy S10 selector value (LWPFS_SEL).
185 *
186 * The best solution to the aforementioned problem is backporting CRs
187 * 6467491 to Solaris 10 so that 64-bit x86 Solaris 10 processes
188 * would accept zero for %fs. Backporting the CRs is a requirement for running
189 * S10 Containers in PV domUs because 64-bit Xen clears %fsbase when %fs is
190 * nonzero. Such behavior breaks 64-bit processes because Xen has to fetch the
191 * FS segments' base addresses from the LWPs' GDTs, which are only capable of
192 * 32-bit addressing.
193 */
194 /*ARGSUSED*/
195 static void
196 s10_amd64_correct_fsreg(klwp_t *l)
197 {
198 if (lwp_getdatamodel(l) == DATAMODEL_NATIVE) {
199 kpreempt_disable();
200 l->lwp_pcb.pcb_fs = LWPFS_SEL;
201 l->lwp_pcb.pcb_rupdate = 1;
202 lwptot(l)->t_post_sys = 1; /* Guarantee update_sregs() */
203 kpreempt_enable();
204 }
205 }
206 #endif /* __amd64 */
207
208 /*
209 * Native processes are started with the native ld.so.1 as the command. This
210 * brand op is invoked by s10_npreload to fix up the command and arguments
211 * so that apps like pgrep or ps see the expected command strings.
212 */
213 int
214 s10_native(void *cmd, void *args)
215 {
216 struct user *up = PTOU(curproc);
217 char cmd_buf[MAXCOMLEN + 1];
218 char arg_buf[PSARGSZ];
219
220 if (copyin(cmd, &cmd_buf, sizeof (cmd_buf)) != 0)
221 return (EFAULT);
222 if (copyin(args, &arg_buf, sizeof (arg_buf)) != 0)
223 return (EFAULT);
224
225 /*
226 * Make sure that the process' interpreter is the native dynamic linker.
227 * Convention dictates that native processes executing within solaris10-
228 * branded zones are interpreted by the native dynamic linker (the
229 * process and its arguments are specified as arguments to the dynamic
230 * linker). If this convention is violated (i.e.,
231 * brandsys(B_S10_NATIVE, ...) is invoked by a process that shouldn't be
232 * native), then do nothing and silently indicate success.
233 */
234 if (strcmp(up->u_comm, S10_LINKER_NAME) != 0)
235 return (0);
236
237 /*
238 * The sizeof has an extra value for the trailing '\0' so this covers
239 * the appended " " in the following strcmps.
240 */
241 if (strncmp(up->u_psargs, BRAND_NATIVE_LINKER64 " ",
242 sizeof (BRAND_NATIVE_LINKER64)) != 0 &&
243 strncmp(up->u_psargs, BRAND_NATIVE_LINKER32 " ",
244 sizeof (BRAND_NATIVE_LINKER32)) != 0)
245 return (0);
246
247 mutex_enter(&curproc->p_lock);
248 (void) strlcpy(up->u_comm, cmd_buf, sizeof (up->u_comm));
249 (void) strlcpy(up->u_psargs, arg_buf, sizeof (up->u_psargs));
250 mutex_exit(&curproc->p_lock);
251
252 return (0);
253 }
254
255 /*ARGSUSED*/
256 int
257 s10_brandsys(int cmd, int64_t *rval, uintptr_t arg1, uintptr_t arg2,
258 uintptr_t arg3, uintptr_t arg4, uintptr_t arg5, uintptr_t arg6)
259 {
260 proc_t *p = curproc;
261 int res;
262
263 *rval = 0;
264
265 if (cmd == B_S10_NATIVE)
266 return (s10_native((void *)arg1, (void *)arg2));
267
268 res = brand_solaris_cmd(cmd, arg1, arg2, arg3, &s10_brand, S10_VERSION);
269 if (res >= 0)
270 return (res);
271
272 switch ((cmd)) {
273 case B_S10_PIDINFO:
274 /*
275 * The s10 brand needs to be able to get the pid of the
276 * current process and the pid of the zone's init, and it
277 * needs to do this on every process startup. Early in
278 * brand startup, we can't call getpid() because calls to
279 * getpid() represent a magical signal to some old-skool
280 * debuggers. By merging all of this into one call, we
281 * make this quite a bit cheaper and easier to handle in
282 * the brand module.
283 */
284 if (copyout(&p->p_pid, (void *)arg1, sizeof (pid_t)) != 0)
285 return (EFAULT);
286 if (copyout(&p->p_zone->zone_proc_initpid, (void *)arg2,
287 sizeof (pid_t)) != 0)
288 return (EFAULT);
289 return (0);
290
291 case B_S10_ISFDXATTRDIR: {
292 /*
293 * This subcommand enables the userland brand emulation library
294 * to determine whether a file descriptor refers to an extended
295 * file attributes directory. There is no standard syscall or
296 * libc function that can make such a determination.
297 */
298 file_t *dir_filep;
299
300 dir_filep = getf((int)arg1);
301 if (dir_filep == NULL)
302 return (EBADF);
303 ASSERT(dir_filep->f_vnode != NULL);
304 *rval = IS_XATTRDIR(dir_filep->f_vnode);
305 releasef((int)arg1);
306 return (0);
307 }
308
309 #ifdef __amd64
310 case B_S10_FSREGCORRECTION:
311 /*
312 * This subcommand exists so that the SYS_lwp_private and
313 * SYS_lwp_create syscalls can manually set the current thread's
314 * %fs register to the legacy S10 selector value for 64-bit x86
315 * processes.
316 */
317 s10_amd64_correct_fsreg(ttolwp(curthread));
318 return (0);
319 #endif /* __amd64 */
320 }
321
322 return (EINVAL);
323 }
324
325 void
326 s10_copy_procdata(proc_t *child, proc_t *parent)
327 {
328 brand_solaris_copy_procdata(child, parent, &s10_brand);
329 }
330
331 void
332 s10_proc_exit(struct proc *p, klwp_t *l)
333 {
334 brand_solaris_proc_exit(p, l, &s10_brand);
335 }
336
337 void
338 s10_exec()
339 {
340 brand_solaris_exec(&s10_brand);
341 }
342
343 int
344 s10_initlwp(klwp_t *l)
345 {
346 return (brand_solaris_initlwp(l, &s10_brand));
347 }
348
349 void
350 s10_forklwp(klwp_t *p, klwp_t *c)
351 {
352 brand_solaris_forklwp(p, c, &s10_brand);
353
354 #ifdef __amd64
355 /*
356 * Only correct the child's %fs register if the parent's %fs register
357 * is LWPFS_SEL. If the parent's %fs register is zero, then the Solaris
358 * 10 environment that we're emulating uses a version of libc that
359 * works when %fs is zero (i.e., it contains backports of CRs 6467491
360 * and 6501650).
361 */
362 if (p->lwp_pcb.pcb_fs == LWPFS_SEL)
363 s10_amd64_correct_fsreg(c);
364 #endif /* __amd64 */
365 }
366
367 void
368 s10_freelwp(klwp_t *l)
369 {
370 brand_solaris_freelwp(l, &s10_brand);
371 }
372
373 void
374 s10_lwpexit(klwp_t *l)
375 {
376 brand_solaris_lwpexit(l, &s10_brand);
377 }
378
379 void
380 s10_free_brand_data(zone_t *zone)
381 {
382 kmem_free(zone->zone_brand_data, sizeof (s10_zone_data_t));
383 }
384
385 void
386 s10_init_brand_data(zone_t *zone)
387 {
388 ASSERT(zone->zone_brand == &s10_brand);
389 ASSERT(zone->zone_brand_data == NULL);
390 zone->zone_brand_data = kmem_zalloc(sizeof (s10_zone_data_t), KM_SLEEP);
391 }
392
393 int
394 s10_elfexec(vnode_t *vp, execa_t *uap, uarg_t *args, intpdata_t *idatap,
395 int level, long *execsz, int setid, caddr_t exec_file, cred_t *cred,
396 int brand_action)
397 {
398 return (brand_solaris_elfexec(vp, uap, args, idatap, level, execsz,
399 setid, exec_file, cred, brand_action, &s10_brand, S10_BRANDNAME,
400 S10_LIB, S10_LIB32, S10_LINKER, S10_LINKER32));
401 }
402
403 void
404 s10_sigset_native_to_s10(sigset_t *set)
405 {
406 int nativesig;
407 int s10sig;
408 sigset_t s10set;
409
410 /*
411 * Shortcut: we know the first 32 signals are the same in both
412 * s10 and native Solaris. Just assign the first word.
413 */
414 s10set.__sigbits[0] = set->__sigbits[0];
415 s10set.__sigbits[1] = 0;
416 s10set.__sigbits[2] = 0;
417 s10set.__sigbits[3] = 0;
418
419 /*
420 * Copy the remainder of the initial set of common signals.
421 */
422 for (nativesig = 33; nativesig < S10_SIGRTMIN; nativesig++)
423 if (sigismember(set, nativesig))
424 sigaddset(&s10set, nativesig);
425
426 /*
427 * Convert any native RT signals to their S10 values.
428 */
429 for (nativesig = _SIGRTMIN, s10sig = S10_SIGRTMIN;
430 nativesig <= _SIGRTMAX && s10sig <= S10_SIGRTMAX;
431 nativesig++, s10sig++) {
432 if (sigismember(set, nativesig))
433 sigaddset(&s10set, s10sig);
434 }
435
436 *set = s10set;
437 }
438
439 void
440 s10_sigset_s10_to_native(sigset_t *set)
441 {
442 int s10sig;
443 int nativesig;
444 sigset_t nativeset;
445
446 /*
447 * Shortcut: we know the first 32 signals are the same in both
448 * s10 and native Solaris. Just assign the first word.
449 */
450 nativeset.__sigbits[0] = set->__sigbits[0];
451 nativeset.__sigbits[1] = 0;
452 nativeset.__sigbits[2] = 0;
453 nativeset.__sigbits[3] = 0;
454
455 /*
456 * Copy the remainder of the initial set of common signals.
457 */
458 for (s10sig = 33; s10sig < S10_SIGRTMIN; s10sig++)
459 if (sigismember(set, s10sig))
460 sigaddset(&nativeset, s10sig);
461
462 /*
463 * Convert any S10 RT signals to their native values.
464 */
465 for (s10sig = S10_SIGRTMIN, nativesig = _SIGRTMIN;
466 s10sig <= S10_SIGRTMAX && nativesig <= _SIGRTMAX;
467 s10sig++, nativesig++) {
468 if (sigismember(set, s10sig))
469 sigaddset(&nativeset, nativesig);
470 }
471
472 *set = nativeset;
473 }
474
475 int
476 _init(void)
477 {
478 int err;
479
480 /*
481 * Set up the table indicating which system calls we want to
482 * interpose on. We should probably build this automatically from
483 * a list of system calls that is shared with the user-space
484 * library.
485 */
486 s10_emulation_table = kmem_zalloc(NSYSCALL, KM_SLEEP);
487 s10_emulation_table[S10_SYS_forkall] = 1; /* 2 */
488 s10_emulation_table[S10_SYS_open] = 1; /* 5 */
489 s10_emulation_table[S10_SYS_wait] = 1; /* 7 */
490 s10_emulation_table[S10_SYS_creat] = 1; /* 8 */
491 s10_emulation_table[S10_SYS_link] = 1; /* 9 */
492 s10_emulation_table[S10_SYS_unlink] = 1; /* 10 */
493 s10_emulation_table[S10_SYS_exec] = 1; /* 11 */
494 s10_emulation_table[S10_SYS_mknod] = 1; /* 14 */
495 s10_emulation_table[S10_SYS_chmod] = 1; /* 15 */
496 s10_emulation_table[S10_SYS_chown] = 1; /* 16 */
497 s10_emulation_table[S10_SYS_stat] = 1; /* 18 */
498 s10_emulation_table[S10_SYS_umount] = 1; /* 22 */
499 s10_emulation_table[S10_SYS_fstat] = 1; /* 28 */
500 s10_emulation_table[S10_SYS_utime] = 1; /* 30 */
501 s10_emulation_table[S10_SYS_access] = 1; /* 33 */
502 s10_emulation_table[SYS_kill] = 1; /* 37 */
503 s10_emulation_table[S10_SYS_dup] = 1; /* 41 */
504 s10_emulation_table[SYS_ioctl] = 1; /* 54 */
505 s10_emulation_table[SYS_execve] = 1; /* 59 */
506 s10_emulation_table[SYS_acctctl] = 1; /* 71 */
507 s10_emulation_table[S10_SYS_issetugid] = 1; /* 75 */
508 s10_emulation_table[S10_SYS_fsat] = 1; /* 76 */
509 s10_emulation_table[S10_SYS_rmdir] = 1; /* 79 */
510 s10_emulation_table[S10_SYS_mkdir] = 1; /* 80 */
511 s10_emulation_table[SYS_getdents] = 1; /* 81 */
512 s10_emulation_table[S10_SYS_poll] = 1; /* 87 */
513 s10_emulation_table[S10_SYS_lstat] = 1; /* 88 */
514 s10_emulation_table[S10_SYS_symlink] = 1; /* 89 */
515 s10_emulation_table[S10_SYS_readlink] = 1; /* 90 */
516 s10_emulation_table[S10_SYS_fchmod] = 1; /* 93 */
517 s10_emulation_table[S10_SYS_fchown] = 1; /* 94 */
518 s10_emulation_table[SYS_sigprocmask] = 1; /* 95 */
519 s10_emulation_table[SYS_sigsuspend] = 1; /* 96 */
520 s10_emulation_table[SYS_sigaction] = 1; /* 98 */
521 s10_emulation_table[SYS_sigpending] = 1; /* 99 */
522 s10_emulation_table[SYS_waitid] = 1; /* 107 */
523 s10_emulation_table[SYS_sigsendsys] = 1; /* 108 */
524 #if defined(__x86)
525 s10_emulation_table[S10_SYS_xstat] = 1; /* 123 */
526 s10_emulation_table[S10_SYS_lxstat] = 1; /* 124 */
527 s10_emulation_table[S10_SYS_fxstat] = 1; /* 125 */
528 s10_emulation_table[S10_SYS_xmknod] = 1; /* 126 */
529 #endif
530 s10_emulation_table[S10_SYS_lchown] = 1; /* 130 */
531 s10_emulation_table[S10_SYS_rename] = 1; /* 134 */
532 s10_emulation_table[SYS_uname] = 1; /* 135 */
533 s10_emulation_table[SYS_sysconfig] = 1; /* 137 */
534 s10_emulation_table[SYS_systeminfo] = 1; /* 139 */
535 s10_emulation_table[S10_SYS_fork1] = 1; /* 143 */
536 s10_emulation_table[SYS_sigtimedwait] = 1; /* 144 */
537 s10_emulation_table[S10_SYS_lwp_sema_wait] = 1; /* 147 */
538 s10_emulation_table[S10_SYS_utimes] = 1; /* 154 */
539 s10_emulation_table[SYS_lwp_create] = 1; /* 159 */
540 s10_emulation_table[SYS_lwp_kill] = 1; /* 163 */
541 s10_emulation_table[SYS_lwp_sigmask] = 1; /* 165 */
542 #if defined(__amd64)
543 s10_emulation_table[SYS_lwp_private] = 1; /* 166 */
544 #endif /* __amd64 */
545 s10_emulation_table[S10_SYS_lwp_mutex_lock] = 1; /* 169 */
546 s10_emulation_table[SYS_pwrite] = 1; /* 174 */
547 s10_emulation_table[SYS_acl] = 1; /* 185 */
548 s10_emulation_table[SYS_auditsys] = 1; /* 186 */
549 s10_emulation_table[SYS_sigqueue] = 1; /* 190 */
550 s10_emulation_table[SYS_facl] = 1; /* 200 */
551 s10_emulation_table[SYS_signotify] = 1; /* 205 */
552 s10_emulation_table[SYS_lwp_mutex_timedlock] = 1; /* 210 */
553 s10_emulation_table[SYS_getdents64] = 1; /* 213 */
554 s10_emulation_table[S10_SYS_stat64] = 1; /* 215 */
555 s10_emulation_table[S10_SYS_lstat64] = 1; /* 216 */
556 s10_emulation_table[S10_SYS_fstat64] = 1; /* 217 */
557 s10_emulation_table[SYS_pwrite64] = 1; /* 223 */
558 s10_emulation_table[S10_SYS_creat64] = 1; /* 224 */
559 s10_emulation_table[S10_SYS_open64] = 1; /* 225 */
560 s10_emulation_table[SYS_zone] = 1; /* 227 */
561 s10_emulation_table[SYS_lwp_mutex_trylock] = 1; /* 251 */
562
563 err = mod_install(&modlinkage);
564 if (err) {
565 cmn_err(CE_WARN, "Couldn't install brand module");
566 kmem_free(s10_emulation_table, NSYSCALL);
567 }
568
569 return (err);
570 }
571
572 int
573 _info(struct modinfo *modinfop)
574 {
575 return (mod_info(&modlinkage, modinfop));
576 }
577
578 int
579 _fini(void)
580 {
581 return (brand_solaris_fini(&s10_emulation_table, &modlinkage,
582 &s10_brand));
583 }