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