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 2016 Toomas Soome <tsoome@me.com>
24 * Copyright (c) 2013, OmniTI Computer Consulting, Inc. All rights reserved.
25 * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
26 */
27
28 #include <errno.h>
29 #include <fcntl.h>
30 #include <dirent.h>
31 #include <stddef.h>
32 #include <stdio.h>
33 #include <stdlib.h>
34 #include <strings.h>
35 #include <unistd.h>
36 #include <thread.h>
37 #include <sys/auxv.h>
38 #include <sys/brand.h>
39 #include <sys/inttypes.h>
40 #include <sys/lwp.h>
41 #include <sys/syscall.h>
42 #include <sys/systm.h>
43 #include <sys/utsname.h>
44 #include <sys/sysconfig.h>
45 #include <sys/systeminfo.h>
46 #include <sys/zone.h>
47 #include <sys/stat.h>
48 #include <sys/mntent.h>
49 #include <sys/ctfs.h>
50 #include <sys/priv.h>
51 #include <sys/acctctl.h>
52 #include <libgen.h>
53 #include <bsm/audit.h>
54 #include <sys/crypto/ioctl.h>
55 #include <sys/fs/zfs.h>
56 #include <sys/zfs_ioctl.h>
57 #include <sys/ucontext.h>
58 #include <sys/mntio.h>
59 #include <sys/mnttab.h>
60 #include <sys/attr.h>
61 #include <sys/lofi.h>
62 #include <atomic.h>
63 #include <sys/acl.h>
64 #include <sys/socket.h>
65
66 #include <s10_brand.h>
67 #include <brand_misc.h>
68 #include <s10_misc.h>
69 #include <s10_signal.h>
70
71 /*
72 * See usr/src/lib/brand/shared/brand/common/brand_util.c for general
73 * emulation notes.
74 */
75
76 static zoneid_t zoneid;
77 static boolean_t emul_global_zone = B_FALSE;
78 static s10_emul_bitmap_t emul_bitmap;
79 pid_t zone_init_pid;
80
81 /*
82 * S10_FEATURE_IS_PRESENT is a macro that helps facilitate conditional
83 * emulation. For each constant N defined in the s10_emulated_features
84 * enumeration in usr/src/uts/common/brand/solaris10/s10_brand.h,
85 * S10_FEATURE_IS_PRESENT(N) is true iff the feature/backport represented by N
86 * is present in the Solaris 10 image hosted within the zone. In other words,
87 * S10_FEATURE_IS_PRESENT(N) is true iff the file /usr/lib/brand/solaris10/M,
88 * where M is the enum value of N, was present in the zone when the zone booted.
89 *
90 *
91 * *** Sample Usage
92 *
93 * Suppose that you need to backport a fix to Solaris 10 and there is
94 * emulation in place for the fix. Suppose further that the emulation won't be
95 * needed if the fix is backported (i.e., if the fix is present in the hosted
96 * Solaris 10 environment, then the brand won't need the emulation). Then if
97 * you add a constant named "S10_FEATURE_X" to the end of the
98 * s10_emulated_features enumeration that represents the backported fix and
99 * S10_FEATURE_X evaluates to four, then you should create a file named
100 * /usr/lib/brand/solaris10/4 as part of your backport. Additionally, you
101 * should retain the aforementioned emulation but modify it so that it's
102 * performed only when S10_FEATURE_IS_PRESENT(S10_FEATURE_X) is false. Thus the
103 * emulation function should look something like the following:
104 *
105 * static int
106 * my_emul_function(sysret_t *rv, ...)
107 * {
108 * if (S10_FEATURE_IS_PRESENT(S10_FEATURE_X)) {
109 * // Don't emulate
110 * return (__systemcall(rv, ...));
111 * } else {
112 * // Emulate whatever needs to be emulated when the
113 * // backport isn't present in the Solaris 10 image.
114 * }
115 * }
116 */
117 #define S10_FEATURE_IS_PRESENT(s10_emulated_features_constant) \
118 ((emul_bitmap[(s10_emulated_features_constant) >> 3] & \
119 (1 << ((s10_emulated_features_constant) & 0x7))) != 0)
120
121 brand_sysent_table_t brand_sysent_table[];
122
123 #define S10_UTS_RELEASE "5.10"
124 #define S10_UTS_VERSION "Generic_Virtual"
125
126 /*
127 * If the ioctl fd's major doesn't match "major", then pass through the
128 * ioctl, since it is not the expected device. major should be a
129 * pointer to a static dev_t initialized to -1, and devname should be
130 * the path of the device.
131 *
132 * Returns 1 if the ioctl was handled (in which case *err contains the
133 * error code), or 0 if it still needs handling.
134 */
135 static int
passthru_otherdev_ioctl(dev_t * majordev,const char * devname,int * err,sysret_t * rval,int fdes,int cmd,intptr_t arg)136 passthru_otherdev_ioctl(dev_t *majordev, const char *devname, int *err,
137 sysret_t *rval, int fdes, int cmd, intptr_t arg)
138 {
139 struct stat sbuf;
140
141 if (*majordev == (dev_t)-1) {
142 if ((*err = __systemcall(rval, SYS_fstatat + 1024,
143 AT_FDCWD, devname, &sbuf, 0) != 0) != 0)
144 goto doioctl;
145
146 *majordev = major(sbuf.st_rdev);
147 }
148
149 if ((*err = __systemcall(rval, SYS_fstatat + 1024, fdes,
150 NULL, &sbuf, 0)) != 0)
151 goto doioctl;
152
153 if (major(sbuf.st_rdev) == *majordev)
154 return (0);
155
156 doioctl:
157 *err = (__systemcall(rval, SYS_ioctl + 1024, fdes, cmd, arg));
158 return (1);
159 }
160
161 /*
162 * Figures out the PID of init for the zone. Also returns a boolean
163 * indicating whether this process currently has that pid: if so,
164 * then at this moment, we are init.
165 */
166 static boolean_t
get_initpid_info(void)167 get_initpid_info(void)
168 {
169 pid_t pid;
170 sysret_t rval;
171 int err;
172
173 /*
174 * Determine the current process PID and the PID of the zone's init.
175 * We use care not to call getpid() here, because we're not supposed
176 * to call getpid() until after the program is fully linked-- the
177 * first call to getpid() is a signal from the linker to debuggers
178 * that linking has been completed.
179 */
180 if ((err = __systemcall(&rval, SYS_brand,
181 B_S10_PIDINFO, &pid, &zone_init_pid)) != 0) {
182 brand_abort(err, "Failed to get init's pid");
183 }
184
185 /*
186 * Note that we need to be cautious with the pid we get back--
187 * it should not be stashed and used in place of getpid(), since
188 * we might fork(2). So we keep zone_init_pid and toss the pid
189 * we otherwise got.
190 */
191 if (pid == zone_init_pid)
192 return (B_TRUE);
193
194 return (B_FALSE);
195 }
196
197 /* Free the thread-local storage provided by mntfs_get_mntentbuf(). */
198 static void
mntfs_free_mntentbuf(void * arg)199 mntfs_free_mntentbuf(void *arg)
200 {
201 struct mntentbuf *embufp = arg;
202
203 if (embufp == NULL)
204 return;
205 if (embufp->mbuf_emp)
206 free(embufp->mbuf_emp);
207 if (embufp->mbuf_buf)
208 free(embufp->mbuf_buf);
209 bzero(embufp, sizeof (struct mntentbuf));
210 free(embufp);
211 }
212
213 /* Provide the thread-local storage required by mntfs_ioctl(). */
214 static struct mntentbuf *
mntfs_get_mntentbuf(size_t size)215 mntfs_get_mntentbuf(size_t size)
216 {
217 static mutex_t keylock;
218 static thread_key_t key;
219 static int once_per_keyname = 0;
220 void *tsd = NULL;
221 struct mntentbuf *embufp;
222
223 /* Create the key. */
224 if (!once_per_keyname) {
225 (void) mutex_lock(&keylock);
226 if (!once_per_keyname) {
227 if (thr_keycreate(&key, mntfs_free_mntentbuf)) {
228 (void) mutex_unlock(&keylock);
229 return (NULL);
230 } else {
231 once_per_keyname++;
232 }
233 }
234 (void) mutex_unlock(&keylock);
235 }
236
237 /*
238 * The thread-specific datum for this key is the address of a struct
239 * mntentbuf. If this is the first time here then we allocate the struct
240 * and its contents, and associate its address with the thread; if there
241 * are any problems then we abort.
242 */
243 if (thr_getspecific(key, &tsd))
244 return (NULL);
245 if (tsd == NULL) {
246 if (!(embufp = calloc(1, sizeof (struct mntentbuf))) ||
247 !(embufp->mbuf_emp = malloc(sizeof (struct extmnttab))) ||
248 thr_setspecific(key, embufp)) {
249 mntfs_free_mntentbuf(embufp);
250 return (NULL);
251 }
252 } else {
253 embufp = tsd;
254 }
255
256 /* Return the buffer, resizing it if necessary. */
257 if (size > embufp->mbuf_bufsize) {
258 if (embufp->mbuf_buf)
259 free(embufp->mbuf_buf);
260 if ((embufp->mbuf_buf = malloc(size)) == NULL) {
261 embufp->mbuf_bufsize = 0;
262 return (NULL);
263 } else {
264 embufp->mbuf_bufsize = size;
265 }
266 }
267 return (embufp);
268 }
269
270 /*
271 * The MNTIOC_GETMNTENT command in this release differs from that in early
272 * versions of Solaris 10.
273 *
274 * Previously, the command would copy a pointer to a struct extmnttab to an
275 * address provided as an argument. The pointer would be somewhere within a
276 * mapping already present within the user's address space. In addition, the
277 * text to which the struct's members pointed would also be within a
278 * pre-existing mapping. Now, the user is required to allocate memory for both
279 * the struct and the text buffer, and to pass the address of each within a
280 * struct mntentbuf. In order to conceal these details from a Solaris 10 client
281 * we allocate some thread-local storage in which to create the necessary data
282 * structures; this is static, thread-safe memory that will be cleaned up
283 * without the caller's intervention.
284 *
285 * MNTIOC_GETEXTMNTENT and MNTIOC_GETMNTANY are new in this release; they should
286 * not work for older clients.
287 */
288 int
mntfs_ioctl(sysret_t * rval,int fdes,int cmd,intptr_t arg)289 mntfs_ioctl(sysret_t *rval, int fdes, int cmd, intptr_t arg)
290 {
291 int err;
292 struct stat statbuf;
293 struct mntentbuf *embufp;
294 static size_t bufsize = MNT_LINE_MAX;
295
296 /* Do not emulate mntfs commands from up-to-date clients. */
297 if (S10_FEATURE_IS_PRESENT(S10_FEATURE_ALTERED_MNTFS_IOCTL))
298 return (__systemcall(rval, SYS_ioctl + 1024, fdes, cmd, arg));
299
300 /* Do not emulate mntfs commands directed at other file systems. */
301 if ((err = __systemcall(rval, SYS_fstatat + 1024,
302 fdes, NULL, &statbuf, 0)) != 0)
303 return (err);
304 if (strcmp(statbuf.st_fstype, MNTTYPE_MNTFS) != 0)
305 return (__systemcall(rval, SYS_ioctl + 1024, fdes, cmd, arg));
306
307 if (cmd == MNTIOC_GETEXTMNTENT || cmd == MNTIOC_GETMNTANY)
308 return (EINVAL);
309
310 if ((embufp = mntfs_get_mntentbuf(bufsize)) == NULL)
311 return (ENOMEM);
312
313 /*
314 * MNTIOC_GETEXTMNTENT advances the file pointer once it has
315 * successfully copied out the result to the address provided. We
316 * therefore need to check the user-supplied address now since the
317 * one we'll be providing is guaranteed to work.
318 */
319 if (brand_uucopy(&embufp->mbuf_emp, (void *)arg, sizeof (void *)) != 0)
320 return (EFAULT);
321
322 /*
323 * Keep retrying for as long as we fail for want of a large enough
324 * buffer.
325 */
326 for (;;) {
327 if ((err = __systemcall(rval, SYS_ioctl + 1024, fdes,
328 MNTIOC_GETEXTMNTENT, embufp)) != 0)
329 return (err);
330
331 if (rval->sys_rval1 == MNTFS_TOOLONG) {
332 /* The buffer wasn't large enough. */
333 (void) atomic_swap_ulong((unsigned long *)&bufsize,
334 2 * embufp->mbuf_bufsize);
335 if ((embufp = mntfs_get_mntentbuf(bufsize)) == NULL)
336 return (ENOMEM);
337 } else {
338 break;
339 }
340 }
341
342 if (brand_uucopy(&embufp->mbuf_emp, (void *)arg, sizeof (void *)) != 0)
343 return (EFAULT);
344
345 return (0);
346 }
347
348 /*
349 * Assign the structure member value from the s (source) structure to the
350 * d (dest) structure.
351 */
352 #define struct_assign(d, s, val) (((d).val) = ((s).val))
353
354 /*
355 * The CRYPTO_GET_FUNCTION_LIST parameter structure crypto_function_list_t
356 * changed between S10 and Nevada, so we have to emulate the old S10
357 * crypto_function_list_t structure when interposing on the ioctl syscall.
358 */
359 typedef struct s10_crypto_function_list {
360 boolean_t fl_digest_init;
361 boolean_t fl_digest;
362 boolean_t fl_digest_update;
363 boolean_t fl_digest_key;
364 boolean_t fl_digest_final;
365
366 boolean_t fl_encrypt_init;
367 boolean_t fl_encrypt;
368 boolean_t fl_encrypt_update;
369 boolean_t fl_encrypt_final;
370
371 boolean_t fl_decrypt_init;
372 boolean_t fl_decrypt;
373 boolean_t fl_decrypt_update;
374 boolean_t fl_decrypt_final;
375
376 boolean_t fl_mac_init;
377 boolean_t fl_mac;
378 boolean_t fl_mac_update;
379 boolean_t fl_mac_final;
380
381 boolean_t fl_sign_init;
382 boolean_t fl_sign;
383 boolean_t fl_sign_update;
384 boolean_t fl_sign_final;
385 boolean_t fl_sign_recover_init;
386 boolean_t fl_sign_recover;
387
388 boolean_t fl_verify_init;
389 boolean_t fl_verify;
390 boolean_t fl_verify_update;
391 boolean_t fl_verify_final;
392 boolean_t fl_verify_recover_init;
393 boolean_t fl_verify_recover;
394
395 boolean_t fl_digest_encrypt_update;
396 boolean_t fl_decrypt_digest_update;
397 boolean_t fl_sign_encrypt_update;
398 boolean_t fl_decrypt_verify_update;
399
400 boolean_t fl_seed_random;
401 boolean_t fl_generate_random;
402
403 boolean_t fl_session_open;
404 boolean_t fl_session_close;
405 boolean_t fl_session_login;
406 boolean_t fl_session_logout;
407
408 boolean_t fl_object_create;
409 boolean_t fl_object_copy;
410 boolean_t fl_object_destroy;
411 boolean_t fl_object_get_size;
412 boolean_t fl_object_get_attribute_value;
413 boolean_t fl_object_set_attribute_value;
414 boolean_t fl_object_find_init;
415 boolean_t fl_object_find;
416 boolean_t fl_object_find_final;
417
418 boolean_t fl_key_generate;
419 boolean_t fl_key_generate_pair;
420 boolean_t fl_key_wrap;
421 boolean_t fl_key_unwrap;
422 boolean_t fl_key_derive;
423
424 boolean_t fl_init_token;
425 boolean_t fl_init_pin;
426 boolean_t fl_set_pin;
427
428 boolean_t prov_is_hash_limited;
429 uint32_t prov_hash_threshold;
430 uint32_t prov_hash_limit;
431 } s10_crypto_function_list_t;
432
433 typedef struct s10_crypto_get_function_list {
434 uint_t fl_return_value;
435 crypto_provider_id_t fl_provider_id;
436 s10_crypto_function_list_t fl_list;
437 } s10_crypto_get_function_list_t;
438
439 /*
440 * The structure returned by the CRYPTO_GET_FUNCTION_LIST ioctl on /dev/crypto
441 * increased in size due to:
442 * 6482533 Threshold for HW offload via PKCS11 interface
443 * between S10 and Nevada. This is a relatively simple process of filling
444 * in the S10 structure fields with the Nevada data.
445 *
446 * We stat the device to make sure that the ioctl is meant for /dev/crypto.
447 *
448 */
449 static int
crypto_ioctl(sysret_t * rval,int fdes,int cmd,intptr_t arg)450 crypto_ioctl(sysret_t *rval, int fdes, int cmd, intptr_t arg)
451 {
452 int err;
453 s10_crypto_get_function_list_t s10_param;
454 crypto_get_function_list_t native_param;
455 static dev_t crypto_dev = (dev_t)-1;
456
457 if (passthru_otherdev_ioctl(&crypto_dev, "/dev/crypto", &err,
458 rval, fdes, cmd, arg) == 1)
459 return (err);
460
461 if (brand_uucopy((const void *)arg, &s10_param, sizeof (s10_param))
462 != 0)
463 return (EFAULT);
464 struct_assign(native_param, s10_param, fl_provider_id);
465 if ((err = __systemcall(rval, SYS_ioctl + 1024, fdes, cmd,
466 &native_param)) != 0)
467 return (err);
468
469 struct_assign(s10_param, native_param, fl_return_value);
470 struct_assign(s10_param, native_param, fl_provider_id);
471
472 struct_assign(s10_param, native_param, fl_list.fl_digest_init);
473 struct_assign(s10_param, native_param, fl_list.fl_digest);
474 struct_assign(s10_param, native_param, fl_list.fl_digest_update);
475 struct_assign(s10_param, native_param, fl_list.fl_digest_key);
476 struct_assign(s10_param, native_param, fl_list.fl_digest_final);
477
478 struct_assign(s10_param, native_param, fl_list.fl_encrypt_init);
479 struct_assign(s10_param, native_param, fl_list.fl_encrypt);
480 struct_assign(s10_param, native_param, fl_list.fl_encrypt_update);
481 struct_assign(s10_param, native_param, fl_list.fl_encrypt_final);
482
483 struct_assign(s10_param, native_param, fl_list.fl_decrypt_init);
484 struct_assign(s10_param, native_param, fl_list.fl_decrypt);
485 struct_assign(s10_param, native_param, fl_list.fl_decrypt_update);
486 struct_assign(s10_param, native_param, fl_list.fl_decrypt_final);
487
488 struct_assign(s10_param, native_param, fl_list.fl_mac_init);
489 struct_assign(s10_param, native_param, fl_list.fl_mac);
490 struct_assign(s10_param, native_param, fl_list.fl_mac_update);
491 struct_assign(s10_param, native_param, fl_list.fl_mac_final);
492
493 struct_assign(s10_param, native_param, fl_list.fl_sign_init);
494 struct_assign(s10_param, native_param, fl_list.fl_sign);
495 struct_assign(s10_param, native_param, fl_list.fl_sign_update);
496 struct_assign(s10_param, native_param, fl_list.fl_sign_final);
497 struct_assign(s10_param, native_param, fl_list.fl_sign_recover_init);
498 struct_assign(s10_param, native_param, fl_list.fl_sign_recover);
499
500 struct_assign(s10_param, native_param, fl_list.fl_verify_init);
501 struct_assign(s10_param, native_param, fl_list.fl_verify);
502 struct_assign(s10_param, native_param, fl_list.fl_verify_update);
503 struct_assign(s10_param, native_param, fl_list.fl_verify_final);
504 struct_assign(s10_param, native_param, fl_list.fl_verify_recover_init);
505 struct_assign(s10_param, native_param, fl_list.fl_verify_recover);
506
507 struct_assign(s10_param, native_param,
508 fl_list.fl_digest_encrypt_update);
509 struct_assign(s10_param, native_param,
510 fl_list.fl_decrypt_digest_update);
511 struct_assign(s10_param, native_param, fl_list.fl_sign_encrypt_update);
512 struct_assign(s10_param, native_param,
513 fl_list.fl_decrypt_verify_update);
514
515 struct_assign(s10_param, native_param, fl_list.fl_seed_random);
516 struct_assign(s10_param, native_param, fl_list.fl_generate_random);
517
518 struct_assign(s10_param, native_param, fl_list.fl_session_open);
519 struct_assign(s10_param, native_param, fl_list.fl_session_close);
520 struct_assign(s10_param, native_param, fl_list.fl_session_login);
521 struct_assign(s10_param, native_param, fl_list.fl_session_logout);
522
523 struct_assign(s10_param, native_param, fl_list.fl_object_create);
524 struct_assign(s10_param, native_param, fl_list.fl_object_copy);
525 struct_assign(s10_param, native_param, fl_list.fl_object_destroy);
526 struct_assign(s10_param, native_param, fl_list.fl_object_get_size);
527 struct_assign(s10_param, native_param,
528 fl_list.fl_object_get_attribute_value);
529 struct_assign(s10_param, native_param,
530 fl_list.fl_object_set_attribute_value);
531 struct_assign(s10_param, native_param, fl_list.fl_object_find_init);
532 struct_assign(s10_param, native_param, fl_list.fl_object_find);
533 struct_assign(s10_param, native_param, fl_list.fl_object_find_final);
534
535 struct_assign(s10_param, native_param, fl_list.fl_key_generate);
536 struct_assign(s10_param, native_param, fl_list.fl_key_generate_pair);
537 struct_assign(s10_param, native_param, fl_list.fl_key_wrap);
538 struct_assign(s10_param, native_param, fl_list.fl_key_unwrap);
539 struct_assign(s10_param, native_param, fl_list.fl_key_derive);
540
541 struct_assign(s10_param, native_param, fl_list.fl_init_token);
542 struct_assign(s10_param, native_param, fl_list.fl_init_pin);
543 struct_assign(s10_param, native_param, fl_list.fl_set_pin);
544
545 struct_assign(s10_param, native_param, fl_list.prov_is_hash_limited);
546 struct_assign(s10_param, native_param, fl_list.prov_hash_threshold);
547 struct_assign(s10_param, native_param, fl_list.prov_hash_limit);
548
549 return (brand_uucopy(&s10_param, (void *)arg, sizeof (s10_param)));
550 }
551
552 /*
553 * The process contract CT_TGET and CT_TSET parameter structure ct_param_t
554 * changed between S10 and Nevada, so we have to emulate the old S10
555 * ct_param_t structure when interposing on the ioctl syscall.
556 */
557 typedef struct s10_ct_param {
558 uint32_t ctpm_id;
559 uint32_t ctpm_pad;
560 uint64_t ctpm_value;
561 } s10_ct_param_t;
562
563 /*
564 * We have to emulate process contract ioctls for init(8) because the
565 * ioctl parameter structure changed between S10 and Nevada. This is
566 * a relatively simple process of filling Nevada structure fields,
567 * shuffling values, and initiating a native system call.
568 *
569 * For now, we'll assume that all consumers of CT_TGET and CT_TSET will
570 * need emulation. We'll issue a stat to make sure that the ioctl
571 * is meant for the contract file system.
572 *
573 */
574 static int
ctfs_ioctl(sysret_t * rval,int fdes,int cmd,intptr_t arg)575 ctfs_ioctl(sysret_t *rval, int fdes, int cmd, intptr_t arg)
576 {
577 int err;
578 s10_ct_param_t s10param;
579 ct_param_t param;
580 struct stat statbuf;
581
582 if ((err = __systemcall(rval, SYS_fstatat + 1024,
583 fdes, NULL, &statbuf, 0)) != 0)
584 return (err);
585 if (strcmp(statbuf.st_fstype, MNTTYPE_CTFS) != 0)
586 return (__systemcall(rval, SYS_ioctl + 1024, fdes, cmd, arg));
587
588 if (brand_uucopy((const void *)arg, &s10param, sizeof (s10param)) != 0)
589 return (EFAULT);
590 param.ctpm_id = s10param.ctpm_id;
591 param.ctpm_size = sizeof (uint64_t);
592 param.ctpm_value = &s10param.ctpm_value;
593 if ((err = __systemcall(rval, SYS_ioctl + 1024, fdes, cmd, ¶m))
594 != 0)
595 return (err);
596
597 if (cmd == CT_TGET)
598 return (brand_uucopy(&s10param, (void *)arg,
599 sizeof (s10param)));
600
601 return (0);
602 }
603
604 /*
605 * ZFS ioctls have changed in each Solaris 10 (S10) release as well as in
606 * Solaris Next. The brand wraps ZFS commands so that the native commands
607 * are used, but we want to be sure no command sneaks in that uses ZFS
608 * without our knowledge. We'll abort the process if we see a ZFS ioctl.
609 */
610 static int
zfs_ioctl(sysret_t * rval,int fdes,int cmd,intptr_t arg)611 zfs_ioctl(sysret_t *rval, int fdes, int cmd, intptr_t arg)
612 {
613 static dev_t zfs_dev = (dev_t)-1;
614 int err;
615
616 if (passthru_otherdev_ioctl(&zfs_dev, ZFS_DEV, &err,
617 rval, fdes, cmd, arg) == 1)
618 return (err);
619
620 brand_abort(0, "ZFS ioctl!");
621 /*NOTREACHED*/
622 return (0);
623 }
624
625 struct s10_lofi_ioctl {
626 uint32_t li_id;
627 boolean_t li_force;
628 char li_filename[MAXPATHLEN + 1];
629 };
630
631 static int
lofi_ioctl(sysret_t * rval,int fdes,int cmd,intptr_t arg)632 lofi_ioctl(sysret_t *rval, int fdes, int cmd, intptr_t arg)
633 {
634 static dev_t lofi_dev = (dev_t)-1;
635 struct s10_lofi_ioctl s10_param;
636 struct lofi_ioctl native_param;
637 int err;
638
639 if (passthru_otherdev_ioctl(&lofi_dev, "/dev/lofictl", &err,
640 rval, fdes, cmd, arg) == 1)
641 return (err);
642
643 if (brand_uucopy((const void *)arg, &s10_param,
644 sizeof (s10_param)) != 0)
645 return (EFAULT);
646
647 /*
648 * Somewhat weirdly, EIO is what the S10 lofi driver would
649 * return for unrecognised cmds.
650 */
651 if (cmd >= LOFI_CHECK_COMPRESSED)
652 return (EIO);
653
654 bzero(&native_param, sizeof (native_param));
655
656 struct_assign(native_param, s10_param, li_id);
657 struct_assign(native_param, s10_param, li_force);
658
659 /*
660 * Careful here, this has changed from [MAXPATHLEN + 1] to
661 * [MAXPATHLEN].
662 */
663 bcopy(s10_param.li_filename, native_param.li_filename,
664 sizeof (native_param.li_filename));
665 native_param.li_filename[MAXPATHLEN - 1] = '\0';
666
667 err = __systemcall(rval, SYS_ioctl + 1024, fdes, cmd, &native_param);
668
669 struct_assign(s10_param, native_param, li_id);
670 /* li_force is input-only */
671
672 bcopy(native_param.li_filename, s10_param.li_filename,
673 sizeof (native_param.li_filename));
674
675 (void) brand_uucopy(&s10_param, (void *)arg, sizeof (s10_param));
676 return (err);
677 }
678
679 int
s10_ioctl(sysret_t * rval,int fdes,int cmd,intptr_t arg)680 s10_ioctl(sysret_t *rval, int fdes, int cmd, intptr_t arg)
681 {
682 switch (cmd) {
683 case CRYPTO_GET_FUNCTION_LIST:
684 return (crypto_ioctl(rval, fdes, cmd, arg));
685 case CT_TGET:
686 /*FALLTHRU*/
687 case CT_TSET:
688 return (ctfs_ioctl(rval, fdes, cmd, arg));
689 case MNTIOC_GETMNTENT:
690 /*FALLTHRU*/
691 case MNTIOC_GETEXTMNTENT:
692 /*FALLTHRU*/
693 case MNTIOC_GETMNTANY:
694 return (mntfs_ioctl(rval, fdes, cmd, arg));
695 }
696
697 switch (cmd & ~0xff) {
698 case ZFS_IOC:
699 return (zfs_ioctl(rval, fdes, cmd, arg));
700
701 case LOFI_IOC_BASE:
702 return (lofi_ioctl(rval, fdes, cmd, arg));
703
704 default:
705 break;
706 }
707
708 return (__systemcall(rval, SYS_ioctl + 1024, fdes, cmd, arg));
709 }
710
711 /*
712 * Unfortunately, pwrite()'s behavior differs between S10 and Nevada when
713 * applied to files opened with O_APPEND. The offset argument is ignored and
714 * the buffer is appended to the target file in S10, whereas the current file
715 * position is ignored in Nevada (i.e., pwrite() acts as though the target file
716 * wasn't opened with O_APPEND). This is a result of the fix for CR 6655660
717 * (pwrite() must ignore the O_APPEND/FAPPEND flag).
718 *
719 * We emulate the old S10 pwrite() behavior by checking whether the target file
720 * was opened with O_APPEND. If it was, then invoke the write() system call
721 * instead of pwrite(); otherwise, invoke the pwrite() system call as usual.
722 */
723 static int
s10_pwrite(sysret_t * rval,int fd,const void * bufferp,size_t num_bytes,off_t offset)724 s10_pwrite(sysret_t *rval, int fd, const void *bufferp, size_t num_bytes,
725 off_t offset)
726 {
727 int err;
728
729 if ((err = __systemcall(rval, SYS_fcntl + 1024, fd, F_GETFL)) != 0)
730 return (err);
731 if (rval->sys_rval1 & O_APPEND)
732 return (__systemcall(rval, SYS_write + 1024, fd, bufferp,
733 num_bytes));
734 return (__systemcall(rval, SYS_pwrite + 1024, fd, bufferp, num_bytes,
735 offset));
736 }
737
738 #if !defined(_LP64)
739 /*
740 * This is the large file version of the pwrite() system call for 32-bit
741 * processes. This exists for the same reason that s10_pwrite() exists; see
742 * the comment above s10_pwrite().
743 */
744 static int
s10_pwrite64(sysret_t * rval,int fd,const void * bufferp,size32_t num_bytes,uint32_t offset_1,uint32_t offset_2)745 s10_pwrite64(sysret_t *rval, int fd, const void *bufferp, size32_t num_bytes,
746 uint32_t offset_1, uint32_t offset_2)
747 {
748 int err;
749
750 if ((err = __systemcall(rval, SYS_fcntl + 1024, fd, F_GETFL)) != 0)
751 return (err);
752 if (rval->sys_rval1 & O_APPEND)
753 return (__systemcall(rval, SYS_write + 1024, fd, bufferp,
754 num_bytes));
755 return (__systemcall(rval, SYS_pwrite64 + 1024, fd, bufferp,
756 num_bytes, offset_1, offset_2));
757 }
758 #endif /* !_LP64 */
759
760 /*
761 * These are convenience macros that s10_getdents_common() uses. Both treat
762 * their arguments, which should be character pointers, as dirent pointers or
763 * dirent64 pointers and yield their d_name and d_reclen fields. These
764 * macros shouldn't be used outside of s10_getdents_common().
765 */
766 #define dirent_name(charptr) ((charptr) + name_offset)
767 #define dirent_reclen(charptr) \
768 (*(unsigned short *)(uintptr_t)((charptr) + reclen_offset))
769
770 /*
771 * This function contains code that is common to both s10_getdents() and
772 * s10_getdents64(). See the comment above s10_getdents() for details.
773 *
774 * rval, fd, buf, and nbyte should be passed unmodified from s10_getdents()
775 * and s10_getdents64(). getdents_syscall_id should be either SYS_getdents
776 * or SYS_getdents64. name_offset should be the the byte offset of
777 * the d_name field in the dirent structures passed to the kernel via the
778 * syscall represented by getdents_syscall_id. reclen_offset should be
779 * the byte offset of the d_reclen field in the aforementioned dirent
780 * structures.
781 */
782 static int
s10_getdents_common(sysret_t * rval,int fd,char * buf,size_t nbyte,int getdents_syscall_id,size_t name_offset,size_t reclen_offset)783 s10_getdents_common(sysret_t *rval, int fd, char *buf, size_t nbyte,
784 int getdents_syscall_id, size_t name_offset, size_t reclen_offset)
785 {
786 int err;
787 size_t buf_size;
788 char *local_buf;
789 char *buf_current;
790
791 /*
792 * Use a special brand operation, B_S10_ISFDXATTRDIR, to determine
793 * whether the specified file descriptor refers to an extended file
794 * attribute directory. If it doesn't, then SYS_getdents won't
795 * reveal extended file attributes, in which case we can simply
796 * hand the syscall to the native kernel.
797 */
798 if ((err = __systemcall(rval, SYS_brand + 1024, B_S10_ISFDXATTRDIR,
799 fd)) != 0)
800 return (err);
801 if (rval->sys_rval1 == 0)
802 return (__systemcall(rval, getdents_syscall_id + 1024, fd, buf,
803 nbyte));
804
805 /*
806 * The file descriptor refers to an extended file attributes directory.
807 * We need to create a dirent buffer that's as large as buf into which
808 * the native SYS_getdents will store the special extended file
809 * attribute directory's entries. We can't dereference buf because
810 * it might be an invalid pointer!
811 */
812 if (nbyte > MAXGETDENTS_SIZE)
813 nbyte = MAXGETDENTS_SIZE;
814 local_buf = (char *)malloc(nbyte);
815 if (local_buf == NULL) {
816 /*
817 * getdents(2) doesn't return an error code indicating a memory
818 * allocation error and it doesn't make sense to return any of
819 * its documented error codes for a malloc(3C) failure. We'll
820 * use ENOMEM even though getdents(2) doesn't use it because it
821 * best describes the failure.
822 */
823 (void) B_TRUSS_POINT_3(rval, getdents_syscall_id, ENOMEM, fd,
824 buf, nbyte);
825 rval->sys_rval1 = -1;
826 rval->sys_rval2 = 0;
827 return (EIO);
828 }
829
830 /*
831 * Issue a native SYS_getdents syscall but use our local dirent buffer
832 * instead of buf. This will allow us to examine the returned dirent
833 * structures immediately and copy them to buf later. That way the
834 * calling process won't be able to see the dirent structures until
835 * we finish examining them.
836 */
837 if ((err = __systemcall(rval, getdents_syscall_id + 1024, fd, local_buf,
838 nbyte)) != 0) {
839 free(local_buf);
840 return (err);
841 }
842 buf_size = rval->sys_rval1;
843 if (buf_size == 0) {
844 free(local_buf);
845 return (0);
846 }
847
848 /*
849 * Look for SUNWattr_ro (VIEW_READONLY) and SUNWattr_rw
850 * (VIEW_READWRITE) in the directory entries and remove them
851 * from the dirent buffer.
852 */
853 for (buf_current = local_buf;
854 (size_t)(buf_current - local_buf) < buf_size; /* cstyle */) {
855 if (strcmp(dirent_name(buf_current), VIEW_READONLY) != 0 &&
856 strcmp(dirent_name(buf_current), VIEW_READWRITE) != 0) {
857 /*
858 * The dirent refers to an attribute that should
859 * be visible to Solaris 10 processes. Keep it
860 * and examine the next entry in the buffer.
861 */
862 buf_current += dirent_reclen(buf_current);
863 } else {
864 /*
865 * We found either SUNWattr_ro (VIEW_READONLY)
866 * or SUNWattr_rw (VIEW_READWRITE). Remove it
867 * from the dirent buffer by decrementing
868 * buf_size by the size of the entry and
869 * overwriting the entry with the remaining
870 * entries.
871 */
872 buf_size -= dirent_reclen(buf_current);
873 (void) memmove(buf_current, buf_current +
874 dirent_reclen(buf_current), buf_size -
875 (size_t)(buf_current - local_buf));
876 }
877 }
878
879 /*
880 * Copy local_buf into buf so that the calling process can see
881 * the results.
882 */
883 if ((err = brand_uucopy(local_buf, buf, buf_size)) != 0) {
884 free(local_buf);
885 rval->sys_rval1 = -1;
886 rval->sys_rval2 = 0;
887 return (err);
888 }
889 rval->sys_rval1 = buf_size;
890 free(local_buf);
891 return (0);
892 }
893
894 /*
895 * Solaris Next added two special extended file attributes, SUNWattr_ro and
896 * SUNWattr_rw, which are called "extended system attributes". They have
897 * special semantics (e.g., a process cannot unlink SUNWattr_ro) and should
898 * not appear in solaris10-branded zones because no Solaris 10 applications,
899 * including system commands such as tar(1), are coded to correctly handle these
900 * special attributes.
901 *
902 * This emulation function solves the aforementioned problem by emulating
903 * the getdents(2) syscall and filtering both system attributes out of resulting
904 * directory entry lists. The emulation function only filters results when
905 * the given file descriptor refers to an extended file attribute directory.
906 * Filtering getdents(2) results is expensive because it requires dynamic
907 * memory allocation; however, the performance cost is tolerable because
908 * we don't expect Solaris 10 processes to frequently examine extended file
909 * attribute directories.
910 *
911 * The brand's emulation library needs two getdents(2) emulation functions
912 * because getdents(2) comes in two flavors: non-largefile-aware getdents(2)
913 * and largefile-aware getdents64(2). s10_getdents() handles the non-largefile-
914 * aware case for 32-bit processes and all getdents(2) syscalls for 64-bit
915 * processes (64-bit processes use largefile-aware interfaces by default).
916 * See s10_getdents64() below for the largefile-aware getdents64(2) emulation
917 * function for 32-bit processes.
918 */
919 static int
s10_getdents(sysret_t * rval,int fd,struct dirent * buf,size_t nbyte)920 s10_getdents(sysret_t *rval, int fd, struct dirent *buf, size_t nbyte)
921 {
922 return (s10_getdents_common(rval, fd, (char *)buf, nbyte, SYS_getdents,
923 offsetof(struct dirent, d_name),
924 offsetof(struct dirent, d_reclen)));
925 }
926
927 #ifndef _LP64
928 /*
929 * This is the largefile-aware version of getdents(2) for 32-bit processes.
930 * This exists for the same reason that s10_getdents() exists. See the comment
931 * above s10_getdents().
932 */
933 static int
s10_getdents64(sysret_t * rval,int fd,struct dirent64 * buf,size_t nbyte)934 s10_getdents64(sysret_t *rval, int fd, struct dirent64 *buf, size_t nbyte)
935 {
936 return (s10_getdents_common(rval, fd, (char *)buf, nbyte,
937 SYS_getdents64, offsetof(struct dirent64, d_name),
938 offsetof(struct dirent64, d_reclen)));
939 }
940 #endif /* !_LP64 */
941
942 #define S10_TRIVIAL_ACL_CNT 6
943 #define NATIVE_TRIVIAL_ACL_CNT 3
944
945 /*
946 * Check if the ACL qualifies as a trivial ACL based on the native
947 * interpretation.
948 */
949 static boolean_t
has_trivial_native_acl(int cmd,int cnt,const char * fname,int fd)950 has_trivial_native_acl(int cmd, int cnt, const char *fname, int fd)
951 {
952 int i, err;
953 sysret_t rval;
954 ace_t buf[NATIVE_TRIVIAL_ACL_CNT];
955
956 if (fname != NULL)
957 err = __systemcall(&rval, SYS_pathconf + 1024, fname,
958 _PC_ACL_ENABLED);
959 else
960 err = __systemcall(&rval, SYS_fpathconf + 1024, fd,
961 _PC_ACL_ENABLED);
962 if (err != 0 || rval.sys_rval1 != _ACL_ACE_ENABLED)
963 return (B_FALSE);
964
965 /*
966 * If we just got the ACL cnt, we don't need to get it again, its
967 * passed in as the cnt arg.
968 */
969 if (cmd != ACE_GETACLCNT) {
970 if (fname != NULL) {
971 if (__systemcall(&rval, SYS_acl + 1024, fname,
972 ACE_GETACLCNT, 0, NULL) != 0)
973 return (B_FALSE);
974 } else {
975 if (__systemcall(&rval, SYS_facl + 1024, fd,
976 ACE_GETACLCNT, 0, NULL) != 0)
977 return (B_FALSE);
978 }
979 cnt = rval.sys_rval1;
980 }
981
982 if (cnt != NATIVE_TRIVIAL_ACL_CNT)
983 return (B_FALSE);
984
985 if (fname != NULL) {
986 if (__systemcall(&rval, SYS_acl + 1024, fname, ACE_GETACL, cnt,
987 buf) != 0)
988 return (B_FALSE);
989 } else {
990 if (__systemcall(&rval, SYS_facl + 1024, fd, ACE_GETACL, cnt,
991 buf) != 0)
992 return (B_FALSE);
993 }
994
995 /*
996 * The following is based on the logic from the native OS
997 * ace_trivial_common() to determine if the native ACL is trivial.
998 */
999 for (i = 0; i < cnt; i++) {
1000 switch (buf[i].a_flags & ACE_TYPE_FLAGS) {
1001 case ACE_OWNER:
1002 case ACE_GROUP|ACE_IDENTIFIER_GROUP:
1003 case ACE_EVERYONE:
1004 break;
1005 default:
1006 return (B_FALSE);
1007 }
1008
1009 if (buf[i].a_flags & (ACE_FILE_INHERIT_ACE|
1010 ACE_DIRECTORY_INHERIT_ACE|ACE_NO_PROPAGATE_INHERIT_ACE|
1011 ACE_INHERIT_ONLY_ACE))
1012 return (B_FALSE);
1013
1014 /*
1015 * Special check for some special bits
1016 *
1017 * Don't allow anybody to deny reading basic
1018 * attributes or a files ACL.
1019 */
1020 if (buf[i].a_access_mask & (ACE_READ_ACL|ACE_READ_ATTRIBUTES) &&
1021 buf[i].a_type == ACE_ACCESS_DENIED_ACE_TYPE)
1022 return (B_FALSE);
1023
1024 /*
1025 * Delete permissions are never set by default
1026 */
1027 if (buf[i].a_access_mask & (ACE_DELETE|ACE_DELETE_CHILD))
1028 return (B_FALSE);
1029 /*
1030 * only allow owner@ to have
1031 * write_acl/write_owner/write_attributes/write_xattr/
1032 */
1033 if (buf[i].a_type == ACE_ACCESS_ALLOWED_ACE_TYPE &&
1034 (!(buf[i].a_flags & ACE_OWNER) && (buf[i].a_access_mask &
1035 (ACE_WRITE_OWNER|ACE_WRITE_ACL| ACE_WRITE_ATTRIBUTES|
1036 ACE_WRITE_NAMED_ATTRS))))
1037 return (B_FALSE);
1038
1039 }
1040
1041 return (B_TRUE);
1042 }
1043
1044 /*
1045 * The following logic is based on the S10 adjust_ace_pair_common() code.
1046 */
1047 static void
s10_adjust_ace_mask(void * pair,size_t access_off,size_t pairsize,mode_t mode)1048 s10_adjust_ace_mask(void *pair, size_t access_off, size_t pairsize, mode_t mode)
1049 {
1050 char *datap = (char *)pair;
1051 uint32_t *amask0 = (uint32_t *)(uintptr_t)(datap + access_off);
1052 uint32_t *amask1 = (uint32_t *)(uintptr_t)(datap + pairsize +
1053 access_off);
1054
1055 if (mode & S_IROTH)
1056 *amask1 |= ACE_READ_DATA;
1057 else
1058 *amask0 |= ACE_READ_DATA;
1059 if (mode & S_IWOTH)
1060 *amask1 |= ACE_WRITE_DATA|ACE_APPEND_DATA;
1061 else
1062 *amask0 |= ACE_WRITE_DATA|ACE_APPEND_DATA;
1063 if (mode & S_IXOTH)
1064 *amask1 |= ACE_EXECUTE;
1065 else
1066 *amask0 |= ACE_EXECUTE;
1067 }
1068
1069 /*
1070 * Construct a trivial S10 style ACL.
1071 */
1072 static int
make_trivial_s10_acl(const char * fname,int fd,ace_t * bp)1073 make_trivial_s10_acl(const char *fname, int fd, ace_t *bp)
1074 {
1075 int err;
1076 sysret_t rval;
1077 struct stat64 buf;
1078 ace_t trivial_s10_acl[] = {
1079 {(uint_t)-1, 0, ACE_OWNER, ACE_ACCESS_DENIED_ACE_TYPE},
1080 {(uint_t)-1, ACE_WRITE_ACL|ACE_WRITE_OWNER|ACE_WRITE_ATTRIBUTES|
1081 ACE_WRITE_NAMED_ATTRS, ACE_OWNER,
1082 ACE_ACCESS_ALLOWED_ACE_TYPE},
1083 {(uint_t)-1, 0, ACE_GROUP|ACE_IDENTIFIER_GROUP,
1084 ACE_ACCESS_DENIED_ACE_TYPE},
1085 {(uint_t)-1, 0, ACE_GROUP|ACE_IDENTIFIER_GROUP,
1086 ACE_ACCESS_ALLOWED_ACE_TYPE},
1087 {(uint_t)-1, ACE_WRITE_ACL|ACE_WRITE_OWNER|ACE_WRITE_ATTRIBUTES|
1088 ACE_WRITE_NAMED_ATTRS, ACE_EVERYONE,
1089 ACE_ACCESS_DENIED_ACE_TYPE},
1090 {(uint_t)-1, ACE_READ_ACL|ACE_READ_ATTRIBUTES|
1091 ACE_READ_NAMED_ATTRS|ACE_SYNCHRONIZE, ACE_EVERYONE,
1092 ACE_ACCESS_ALLOWED_ACE_TYPE}
1093 };
1094
1095 if (fname != NULL) {
1096 if ((err = __systemcall(&rval, SYS_fstatat64 + 1024, AT_FDCWD,
1097 fname, &buf, 0)) != 0)
1098 return (err);
1099 } else {
1100 if ((err = __systemcall(&rval, SYS_fstatat64 + 1024, fd,
1101 NULL, &buf, 0)) != 0)
1102 return (err);
1103 }
1104
1105 s10_adjust_ace_mask(&trivial_s10_acl[0], offsetof(ace_t, a_access_mask),
1106 sizeof (ace_t), (buf.st_mode & 0700) >> 6);
1107 s10_adjust_ace_mask(&trivial_s10_acl[2], offsetof(ace_t, a_access_mask),
1108 sizeof (ace_t), (buf.st_mode & 0070) >> 3);
1109 s10_adjust_ace_mask(&trivial_s10_acl[4], offsetof(ace_t, a_access_mask),
1110 sizeof (ace_t), buf.st_mode & 0007);
1111
1112 if (brand_uucopy(&trivial_s10_acl, bp, sizeof (trivial_s10_acl)) != 0)
1113 return (EFAULT);
1114
1115 return (0);
1116 }
1117
1118 /*
1119 * The definition of a trivial ace-style ACL (used by ZFS and NFSv4) has been
1120 * simplified since S10. Instead of 6 entries on a trivial S10 ACE ACL we now
1121 * have 3 streamlined entries. The new, simpler trivial style confuses S10
1122 * commands such as 'ls -v' or 'cp -p' which don't see the expected S10 trivial
1123 * ACL entries and thus assume that there is a complex ACL on the file.
1124 *
1125 * See: PSARC/2010/029 Improved ACL interoperability
1126 *
1127 * Note that the trival ACL detection code is implemented in acl_trival() in
1128 * lib/libsec/common/aclutils.c. It always uses the acl() syscall (not the
1129 * facl syscall) to determine if an ACL is trivial. However, we emulate both
1130 * acl() and facl() so that the two provide consistent results.
1131 *
1132 * We don't currently try to emulate setting of ACLs since the primary
1133 * consumer of this feature is SMB or NFSv4 servers, neither of which are
1134 * supported in solaris10-branded zones. If ACLs are used they must be set on
1135 * files using the native OS interpretation.
1136 */
1137 int
s10_acl(sysret_t * rval,const char * fname,int cmd,int nentries,void * aclbufp)1138 s10_acl(sysret_t *rval, const char *fname, int cmd, int nentries, void *aclbufp)
1139 {
1140 int res;
1141
1142 res = __systemcall(rval, SYS_acl + 1024, fname, cmd, nentries, aclbufp);
1143
1144 switch (cmd) {
1145 case ACE_GETACLCNT:
1146 if (res == 0 && has_trivial_native_acl(ACE_GETACLCNT,
1147 rval->sys_rval1, fname, 0)) {
1148 rval->sys_rval1 = S10_TRIVIAL_ACL_CNT;
1149 }
1150 break;
1151 case ACE_GETACL:
1152 if (res == 0 &&
1153 has_trivial_native_acl(ACE_GETACL, 0, fname, 0) &&
1154 nentries >= S10_TRIVIAL_ACL_CNT) {
1155 res = make_trivial_s10_acl(fname, 0, aclbufp);
1156 rval->sys_rval1 = S10_TRIVIAL_ACL_CNT;
1157 }
1158 break;
1159 }
1160
1161 return (res);
1162 }
1163
1164 int
s10_facl(sysret_t * rval,int fdes,int cmd,int nentries,void * aclbufp)1165 s10_facl(sysret_t *rval, int fdes, int cmd, int nentries, void *aclbufp)
1166 {
1167 int res;
1168
1169 res = __systemcall(rval, SYS_facl + 1024, fdes, cmd, nentries, aclbufp);
1170
1171 switch (cmd) {
1172 case ACE_GETACLCNT:
1173 if (res == 0 && has_trivial_native_acl(ACE_GETACLCNT,
1174 rval->sys_rval1, NULL, fdes)) {
1175 rval->sys_rval1 = S10_TRIVIAL_ACL_CNT;
1176 }
1177 break;
1178 case ACE_GETACL:
1179 if (res == 0 &&
1180 has_trivial_native_acl(ACE_GETACL, 0, NULL, fdes) &&
1181 nentries >= S10_TRIVIAL_ACL_CNT) {
1182 res = make_trivial_s10_acl(NULL, fdes, aclbufp);
1183 rval->sys_rval1 = S10_TRIVIAL_ACL_CNT;
1184 }
1185 break;
1186 }
1187
1188 return (res);
1189 }
1190
1191 #define S10_AC_PROC (0x1 << 28)
1192 #define S10_AC_TASK (0x2 << 28)
1193 #define S10_AC_FLOW (0x4 << 28)
1194 #define S10_AC_MODE(x) ((x) & 0xf0000000)
1195 #define S10_AC_OPTION(x) ((x) & 0x0fffffff)
1196
1197 /*
1198 * The mode shift, mode mask and option mask for acctctl have changed. The
1199 * mode is currently the top full byte and the option is the lower 3 full bytes.
1200 */
1201 int
s10_acctctl(sysret_t * rval,int cmd,void * buf,size_t bufsz)1202 s10_acctctl(sysret_t *rval, int cmd, void *buf, size_t bufsz)
1203 {
1204 int mode = S10_AC_MODE(cmd);
1205 int option = S10_AC_OPTION(cmd);
1206
1207 switch (mode) {
1208 case S10_AC_PROC:
1209 mode = AC_PROC;
1210 break;
1211 case S10_AC_TASK:
1212 mode = AC_TASK;
1213 break;
1214 case S10_AC_FLOW:
1215 mode = AC_FLOW;
1216 break;
1217 default:
1218 return (B_TRUSS_POINT_3(rval, SYS_acctctl, EINVAL, cmd, buf,
1219 bufsz));
1220 }
1221
1222 return (__systemcall(rval, SYS_acctctl + 1024, mode | option, buf,
1223 bufsz));
1224 }
1225
1226 /*
1227 * The Audit Policy parameters have changed due to:
1228 * 6466722 audituser and AUDIT_USER are defined, unused, undocumented and
1229 * should be removed.
1230 *
1231 * In S10 we had the following flag:
1232 * #define AUDIT_USER 0x0040
1233 * which doesn't exist in Solaris Next where the subsequent flags are shifted
1234 * down. For example, in S10 we had:
1235 * #define AUDIT_GROUP 0x0080
1236 * but on Solaris Next we have:
1237 * #define AUDIT_GROUP 0x0040
1238 * AUDIT_GROUP has the value AUDIT_USER had in S10 and all of the subsequent
1239 * bits are also shifted one place.
1240 *
1241 * When we're getting or setting the Audit Policy parameters we need to
1242 * shift the outgoing or incoming bits into their proper positions. Since
1243 * S10_AUDIT_USER was always unused, we always clear that bit on A_GETPOLICY.
1244 *
1245 * The command we care about, BSM_AUDITCTL, passes the most parameters (3),
1246 * so declare this function to take up to 4 args and just pass them on.
1247 * The number of parameters for s10_auditsys needs to be equal to the BSM_*
1248 * subcommand that has the most parameters, since we want to pass all
1249 * parameters through, regardless of which subcommands we interpose on.
1250 *
1251 * Note that the auditsys system call uses the SYSENT_AP macro wrapper instead
1252 * of the more common SYSENT_CI macro. This means the return value is a
1253 * SE_64RVAL so the syscall table uses RV_64RVAL.
1254 */
1255
1256 #define S10_AUDIT_HMASK 0xffffffc0
1257 #define S10_AUDIT_LMASK 0x3f
1258 #define S10_AUC_NOSPACE 0x3
1259
1260 int
s10_auditsys(sysret_t * rval,int bsmcmd,intptr_t a0,intptr_t a1,intptr_t a2)1261 s10_auditsys(sysret_t *rval, int bsmcmd, intptr_t a0, intptr_t a1, intptr_t a2)
1262 {
1263 int err;
1264 uint32_t m;
1265
1266 if (bsmcmd != BSM_AUDITCTL)
1267 return (__systemcall(rval, SYS_auditsys + 1024, bsmcmd, a0, a1,
1268 a2));
1269
1270 if ((int)a0 == A_GETPOLICY) {
1271 if ((err = __systemcall(rval, SYS_auditsys + 1024, bsmcmd, a0,
1272 &m, a2)) != 0)
1273 return (err);
1274 m = ((m & S10_AUDIT_HMASK) << 1) | (m & S10_AUDIT_LMASK);
1275 if (brand_uucopy(&m, (void *)a1, sizeof (m)) != 0)
1276 return (EFAULT);
1277 return (0);
1278
1279 } else if ((int)a0 == A_SETPOLICY) {
1280 if (brand_uucopy((const void *)a1, &m, sizeof (m)) != 0)
1281 return (EFAULT);
1282 m = ((m >> 1) & S10_AUDIT_HMASK) | (m & S10_AUDIT_LMASK);
1283 return (__systemcall(rval, SYS_auditsys + 1024, bsmcmd, a0, &m,
1284 a2));
1285 } else if ((int)a0 == A_GETCOND) {
1286 if ((err = __systemcall(rval, SYS_auditsys + 1024, bsmcmd, a0,
1287 &m, a2)) != 0)
1288 return (err);
1289 if (m == AUC_NOSPACE)
1290 m = S10_AUC_NOSPACE;
1291 if (brand_uucopy(&m, (void *)a1, sizeof (m)) != 0)
1292 return (EFAULT);
1293 return (0);
1294 } else if ((int)a0 == A_SETCOND) {
1295 if (brand_uucopy((const void *)a1, &m, sizeof (m)) != 0)
1296 return (EFAULT);
1297 if (m == S10_AUC_NOSPACE)
1298 m = AUC_NOSPACE;
1299 return (__systemcall(rval, SYS_auditsys + 1024, bsmcmd, a0, &m,
1300 a2));
1301 }
1302
1303 return (__systemcall(rval, SYS_auditsys + 1024, bsmcmd, a0, a1, a2));
1304 }
1305
1306 /*
1307 * Determine whether the executable passed to SYS_exec or SYS_execve is a
1308 * native executable. The s10_npreload.so invokes the B_S10_NATIVE brand
1309 * operation which patches up the processes exec info to eliminate any trace
1310 * of the wrapper. That will make pgrep and other commands that examine
1311 * process' executable names and command-line parameters work properly.
1312 */
1313 static int
s10_exec_native(sysret_t * rval,const char * fname,const char ** argp,const char ** envp)1314 s10_exec_native(sysret_t *rval, const char *fname, const char **argp,
1315 const char **envp)
1316 {
1317 const char *filename = fname;
1318 char path[64];
1319 int err;
1320
1321 /* Get a copy of the executable we're trying to run */
1322 path[0] = '\0';
1323 (void) brand_uucopystr(filename, path, sizeof (path));
1324
1325 /* Check if we're trying to run a native binary */
1326 if (strncmp(path, "/.SUNWnative/usr/lib/brand/solaris10/s10_native",
1327 sizeof (path)) != 0)
1328 return (0);
1329
1330 /* Skip the first element in the argv array */
1331 argp++;
1332
1333 /*
1334 * The the path of the dynamic linker is the second parameter
1335 * of s10_native_exec().
1336 */
1337 if (brand_uucopy(argp, &filename, sizeof (char *)) != 0)
1338 return (EFAULT);
1339
1340 /* If an exec call succeeds, it never returns */
1341 err = __systemcall(rval, SYS_brand + 1024, B_EXEC_NATIVE, filename,
1342 argp, envp, NULL, NULL, NULL);
1343 brand_assert(err != 0);
1344 return (err);
1345 }
1346
1347 /*
1348 * Interpose on the SYS_exec syscall to detect native wrappers.
1349 */
1350 int
s10_exec(sysret_t * rval,const char * fname,const char ** argp)1351 s10_exec(sysret_t *rval, const char *fname, const char **argp)
1352 {
1353 int err;
1354
1355 if ((err = s10_exec_native(rval, fname, argp, NULL)) != 0)
1356 return (err);
1357
1358 /* If an exec call succeeds, it never returns */
1359 err = __systemcall(rval, SYS_execve + 1024, fname, argp, NULL, 0);
1360 brand_assert(err != 0);
1361 return (err);
1362 }
1363
1364 /*
1365 * Interpose on the SYS_execve syscall to detect native wrappers.
1366 */
1367 int
s10_execve(sysret_t * rval,const char * fname,const char ** argp,const char ** envp)1368 s10_execve(sysret_t *rval, const char *fname, const char **argp,
1369 const char **envp)
1370 {
1371 int err;
1372
1373 if ((err = s10_exec_native(rval, fname, argp, envp)) != 0)
1374 return (err);
1375
1376 /* If an exec call succeeds, it never returns */
1377 err = __systemcall(rval, SYS_execve + 1024, fname, argp, envp, 0);
1378 brand_assert(err != 0);
1379 return (err);
1380 }
1381
1382 /*
1383 * S10's issetugid() syscall is now a subcode to privsys().
1384 */
1385 static int
s10_issetugid(sysret_t * rval)1386 s10_issetugid(sysret_t *rval)
1387 {
1388 return (__systemcall(rval, SYS_privsys + 1024, PRIVSYS_ISSETUGID,
1389 0, 0, 0, 0, 0));
1390 }
1391
1392 /*
1393 * S10's socket() syscall does not split type and flags
1394 */
1395 static int
s10_so_socket(sysret_t * rval,int domain,int type,int protocol,char * devpath,int version)1396 s10_so_socket(sysret_t *rval, int domain, int type, int protocol,
1397 char *devpath, int version)
1398 {
1399 if ((type & ~SOCK_TYPE_MASK) != 0) {
1400 errno = EINVAL;
1401 return (-1);
1402 }
1403 return (__systemcall(rval, SYS_so_socket + 1024, domain, type,
1404 protocol, devpath, version));
1405 }
1406
1407 /*
1408 * S10's pipe() syscall has a different calling convention
1409 */
1410 static int
s10_pipe(sysret_t * rval)1411 s10_pipe(sysret_t *rval)
1412 {
1413 int fds[2], err;
1414 if ((err = __systemcall(rval, SYS_pipe + 1024, fds, 0)) != 0)
1415 return (err);
1416
1417 rval->sys_rval1 = fds[0];
1418 rval->sys_rval2 = fds[1];
1419 return (0);
1420 }
1421
1422 /*
1423 * S10's accept() syscall takes three arguments
1424 */
1425 static int
s10_accept(sysret_t * rval,int sock,struct sockaddr * addr,uint_t * addrlen,int version)1426 s10_accept(sysret_t *rval, int sock, struct sockaddr *addr, uint_t *addrlen,
1427 int version)
1428 {
1429 return (__systemcall(rval, SYS_accept + 1024, sock, addr, addrlen,
1430 version, 0));
1431 }
1432
1433 static long
s10_uname(sysret_t * rv,uintptr_t p1)1434 s10_uname(sysret_t *rv, uintptr_t p1)
1435 {
1436 struct utsname un, *unp = (struct utsname *)p1;
1437 int rev, err;
1438
1439 if ((err = __systemcall(rv, SYS_uname + 1024, &un)) != 0)
1440 return (err);
1441
1442 rev = atoi(&un.release[2]);
1443 brand_assert(rev >= 11);
1444 bzero(un.release, _SYS_NMLN);
1445 (void) strlcpy(un.release, S10_UTS_RELEASE, _SYS_NMLN);
1446 bzero(un.version, _SYS_NMLN);
1447 (void) strlcpy(un.version, S10_UTS_VERSION, _SYS_NMLN);
1448
1449 /* copy out the modified uname info */
1450 return (brand_uucopy(&un, unp, sizeof (un)));
1451 }
1452
1453 int
s10_sysconfig(sysret_t * rv,int which)1454 s10_sysconfig(sysret_t *rv, int which)
1455 {
1456 long value;
1457
1458 /*
1459 * We must interpose on the sysconfig(2) requests
1460 * that deal with the realtime signal number range.
1461 * All others get passed to the native sysconfig(2).
1462 */
1463 switch (which) {
1464 case _CONFIG_RTSIG_MAX:
1465 value = S10_SIGRTMAX - S10_SIGRTMIN + 1;
1466 break;
1467 case _CONFIG_SIGRT_MIN:
1468 value = S10_SIGRTMIN;
1469 break;
1470 case _CONFIG_SIGRT_MAX:
1471 value = S10_SIGRTMAX;
1472 break;
1473 default:
1474 return (__systemcall(rv, SYS_sysconfig + 1024, which));
1475 }
1476
1477 (void) B_TRUSS_POINT_1(rv, SYS_sysconfig, 0, which);
1478 rv->sys_rval1 = value;
1479 rv->sys_rval2 = 0;
1480
1481 return (0);
1482 }
1483
1484 int
s10_sysinfo(sysret_t * rv,int command,char * buf,long count)1485 s10_sysinfo(sysret_t *rv, int command, char *buf, long count)
1486 {
1487 char *value;
1488 int len;
1489
1490 /*
1491 * We must interpose on the sysinfo(2) commands SI_RELEASE and
1492 * SI_VERSION; all others get passed to the native sysinfo(2)
1493 * command.
1494 */
1495 switch (command) {
1496 case SI_RELEASE:
1497 value = S10_UTS_RELEASE;
1498 break;
1499
1500 case SI_VERSION:
1501 value = S10_UTS_VERSION;
1502 break;
1503
1504 default:
1505 /*
1506 * The default action is to pass the command to the
1507 * native sysinfo(2) syscall.
1508 */
1509 return (__systemcall(rv, SYS_systeminfo + 1024,
1510 command, buf, count));
1511 }
1512
1513 len = strlen(value) + 1;
1514 if (count > 0) {
1515 if (brand_uucopystr(value, buf, count) != 0)
1516 return (EFAULT);
1517
1518 /*
1519 * Assure NULL termination of buf as brand_uucopystr() doesn't.
1520 */
1521 if (len > count && brand_uucopy("\0", buf + (count - 1), 1)
1522 != 0)
1523 return (EFAULT);
1524 }
1525
1526 /*
1527 * On success, sysinfo(2) returns the size of buffer required to hold
1528 * the complete value plus its terminating NULL byte.
1529 */
1530 (void) B_TRUSS_POINT_3(rv, SYS_systeminfo, 0, command, buf, count);
1531 rv->sys_rval1 = len;
1532 rv->sys_rval2 = 0;
1533 return (0);
1534 }
1535
1536 #if defined(__x86)
1537 #if defined(__amd64)
1538 /*
1539 * 64-bit x86 LWPs created by SYS_lwp_create start here if they need to set
1540 * their %fs registers to the legacy Solaris 10 selector value.
1541 *
1542 * This function does three things:
1543 *
1544 * 1. Trap to the kernel so that it can set %fs to the legacy Solaris 10
1545 * selector value.
1546 * 2. Read the LWP's true entry point (the entry point supplied by libc
1547 * when SYS_lwp_create was invoked) from %r14.
1548 * 3. Eliminate this function's stack frame and pass control to the LWP's
1549 * true entry point.
1550 *
1551 * See the comment above s10_lwp_create_correct_fs() (see below) for the reason
1552 * why this function exists.
1553 */
1554 /*ARGSUSED*/
1555 static void
s10_lwp_create_entry_point(void * ulwp_structp)1556 s10_lwp_create_entry_point(void *ulwp_structp)
1557 {
1558 sysret_t rval;
1559
1560 /*
1561 * The new LWP's %fs register is initially zero, but libc won't
1562 * function correctly when %fs is zero. Change the LWP's %fs register
1563 * via SYS_brand.
1564 */
1565 (void) __systemcall(&rval, SYS_brand + 1024, B_S10_FSREGCORRECTION);
1566
1567 /*
1568 * Jump to the true entry point, which is stored in %r14.
1569 * Remove our stack frame before jumping so that
1570 * s10_lwp_create_entry_point() won't be seen in stack traces.
1571 *
1572 * NOTE: s10_lwp_create_entry_point() pushes %r12 onto its stack frame
1573 * so that it can use it as a temporary register. We don't restore %r12
1574 * in this assembly block because we don't care about its value (and
1575 * neither does _lwp_start()). Besides, the System V ABI AMD64
1576 * Actirecture Processor Supplement doesn't specify that %r12 should
1577 * have a special value when LWPs start, so we can ignore its value when
1578 * we jump to the true entry point. Furthermore, %r12 is a callee-saved
1579 * register, so the true entry point should push %r12 onto its stack
1580 * before using the register. We ignore %r14 after we read it for
1581 * similar reasons.
1582 *
1583 * NOTE: The compiler will generate a function epilogue for this
1584 * function despite the fact that the LWP will never execute it.
1585 * We could hand-code this entire function in assembly to eliminate
1586 * the epilogue, but the epilogue is only three or four instructions,
1587 * so we wouldn't save much space. Besides, why would we want
1588 * to create yet another ugly, hard-to-maintain assembly function when
1589 * we could write most of it in C?
1590 */
1591 __asm__ __volatile__(
1592 "movq %0, %%rdi\n\t" /* pass ulwp_structp as arg1 */
1593 "movq %%rbp, %%rsp\n\t" /* eliminate the stack frame */
1594 "popq %%rbp\n\t"
1595 "jmp *%%r14\n\t" /* jump to the true entry point */
1596 : : "r" (ulwp_structp));
1597 /*NOTREACHED*/
1598 }
1599
1600 /*
1601 * The S10 libc expects that %fs will be nonzero for new 64-bit x86 LWPs but the
1602 * Nevada kernel clears %fs for such LWPs. Unforunately, new LWPs do not issue
1603 * SYS_lwp_private (see s10_lwp_private() below) after they are created, so
1604 * we must ensure that new LWPs invoke a brand operation that sets %fs to a
1605 * nonzero value immediately after their creation.
1606 *
1607 * The easiest way to do this is to make new LWPs start at a special function,
1608 * s10_lwp_create_entry_point() (see its definition above), that invokes the
1609 * brand operation that corrects %fs. We'll store the entry points of new LWPs
1610 * in their %r14 registers so that s10_lwp_create_entry_point() can find and
1611 * call them after invoking the special brand operation. %r14 is a callee-saved
1612 * register; therefore, any functions invoked by s10_lwp_create_entry_point()
1613 * and all functions dealing with signals (e.g., sigacthandler()) will preserve
1614 * %r14 for s10_lwp_create_entry_point().
1615 *
1616 * The Nevada kernel can safely work with nonzero %fs values because the kernel
1617 * configures per-thread %fs segment descriptors so that the legacy %fs selector
1618 * value will still work. See the comment in lwp_load() regarding %fs and
1619 * %fsbase in 64-bit x86 processes.
1620 *
1621 * This emulation exists thanks to CRs 6467491 and 6501650.
1622 */
1623 static int
s10_lwp_create_correct_fs(sysret_t * rval,ucontext_t * ucp,int flags,id_t * new_lwp)1624 s10_lwp_create_correct_fs(sysret_t *rval, ucontext_t *ucp, int flags,
1625 id_t *new_lwp)
1626 {
1627 ucontext_t s10_uc;
1628
1629 /*
1630 * Copy the supplied ucontext_t structure to the local stack
1631 * frame and store the new LWP's entry point (the value of %rip
1632 * stored in the ucontext_t) in the new LWP's %r14 register.
1633 * Then make s10_lwp_create_entry_point() the new LWP's entry
1634 * point.
1635 */
1636 if (brand_uucopy(ucp, &s10_uc, sizeof (s10_uc)) != 0)
1637 return (EFAULT);
1638
1639 s10_uc.uc_mcontext.gregs[REG_R14] = s10_uc.uc_mcontext.gregs[REG_RIP];
1640 s10_uc.uc_mcontext.gregs[REG_RIP] = (greg_t)s10_lwp_create_entry_point;
1641
1642 /* fix up the signal mask */
1643 if (s10_uc.uc_flags & UC_SIGMASK)
1644 (void) s10sigset_to_native(&s10_uc.uc_sigmask,
1645 &s10_uc.uc_sigmask);
1646
1647 /*
1648 * Issue SYS_lwp_create to create the new LWP. We pass the
1649 * modified ucontext_t to make sure that the new LWP starts at
1650 * s10_lwp_create_entry_point().
1651 */
1652 return (__systemcall(rval, SYS_lwp_create + 1024, &s10_uc,
1653 flags, new_lwp));
1654 }
1655 #endif /* __amd64 */
1656
1657 /*
1658 * SYS_lwp_private is issued by libc_init() to set %fsbase in 64-bit x86
1659 * processes. The Nevada kernel sets %fs to zero but the S10 libc expects
1660 * %fs to be nonzero. We'll pass the issued system call to the kernel untouched
1661 * and invoke a brand operation to set %fs to the legacy S10 selector value.
1662 *
1663 * This emulation exists thanks to CRs 6467491 and 6501650.
1664 */
1665 static int
s10_lwp_private(sysret_t * rval,int cmd,int which,uintptr_t base)1666 s10_lwp_private(sysret_t *rval, int cmd, int which, uintptr_t base)
1667 {
1668 #if defined(__amd64)
1669 int err;
1670
1671 /*
1672 * The current LWP's %fs register should be zero. Determine whether the
1673 * Solaris 10 libc with which we're working functions correctly when %fs
1674 * is zero by calling thr_main() after issuing the SYS_lwp_private
1675 * syscall. If thr_main() barfs (returns -1), then change the LWP's %fs
1676 * register via SYS_brand and patch brand_sysent_table so that issuing
1677 * SYS_lwp_create executes s10_lwp_create_correct_fs() rather than the
1678 * default s10_lwp_create(). s10_lwp_create_correct_fs() will
1679 * guarantee that new LWPs will have correct %fs values.
1680 */
1681 if ((err = __systemcall(rval, SYS_lwp_private + 1024, cmd, which,
1682 base)) != 0)
1683 return (err);
1684 if (thr_main() == -1) {
1685 /*
1686 * SYS_lwp_private is only issued by libc_init(), which is
1687 * executed when libc is first loaded by ld.so.1. Thus we
1688 * are guaranteed to be single-threaded at this point. Even
1689 * if we were multithreaded at this point, writing a 64-bit
1690 * value to the st_callc field of a brand_sysent_table
1691 * entry is guaranteed to be atomic on 64-bit x86 chips
1692 * as long as the field is not split across cache lines
1693 * (It shouldn't be.). See chapter 8, section 1.1 of
1694 * "The Intel 64 and IA32 Architectures Software Developer's
1695 * Manual," Volume 3A for more details.
1696 */
1697 brand_sysent_table[SYS_lwp_create].st_callc =
1698 (sysent_cb_t)(uintptr_t)s10_lwp_create_correct_fs;
1699 return (__systemcall(rval, SYS_brand + 1024,
1700 B_S10_FSREGCORRECTION));
1701 }
1702 return (0);
1703 #else /* !__amd64 */
1704 return (__systemcall(rval, SYS_lwp_private + 1024, cmd, which, base));
1705 #endif /* !__amd64 */
1706 }
1707 #endif /* __x86 */
1708
1709 /*
1710 * The Opensolaris versions of lwp_mutex_timedlock() and lwp_mutex_trylock()
1711 * add an extra argument to the interfaces, a uintptr_t value for the mutex's
1712 * mutex_owner field. The Solaris 10 libc assigns the mutex_owner field at
1713 * user-level, so we just make the extra argument be zero in both syscalls.
1714 */
1715
1716 static int
s10_lwp_mutex_timedlock(sysret_t * rval,lwp_mutex_t * lp,timespec_t * tsp)1717 s10_lwp_mutex_timedlock(sysret_t *rval, lwp_mutex_t *lp, timespec_t *tsp)
1718 {
1719 return (__systemcall(rval, SYS_lwp_mutex_timedlock + 1024, lp, tsp, 0));
1720 }
1721
1722 static int
s10_lwp_mutex_trylock(sysret_t * rval,lwp_mutex_t * lp)1723 s10_lwp_mutex_trylock(sysret_t *rval, lwp_mutex_t *lp)
1724 {
1725 return (__systemcall(rval, SYS_lwp_mutex_trylock + 1024, lp, 0));
1726 }
1727
1728 /*
1729 * If the emul_global_zone flag is set then emulate some aspects of the
1730 * zone system call. In particular, emulate the global zone ID on the
1731 * ZONE_LOOKUP subcommand and emulate some of the global zone attributes
1732 * on the ZONE_GETATTR subcommand. If the flag is not set or we're performing
1733 * some other operation, simply pass the calls through.
1734 */
1735 int
s10_zone(sysret_t * rval,int cmd,void * arg1,void * arg2,void * arg3,void * arg4)1736 s10_zone(sysret_t *rval, int cmd, void *arg1, void *arg2, void *arg3,
1737 void *arg4)
1738 {
1739 char *aval;
1740 int len;
1741 zoneid_t zid;
1742 int attr;
1743 char *buf;
1744 size_t bufsize;
1745
1746 /*
1747 * We only emulate the zone syscall for a subset of specific commands,
1748 * otherwise we just pass the call through.
1749 */
1750 if (!emul_global_zone)
1751 return (__systemcall(rval, SYS_zone + 1024, cmd, arg1, arg2,
1752 arg3, arg4));
1753
1754 switch (cmd) {
1755 case ZONE_LOOKUP:
1756 (void) B_TRUSS_POINT_1(rval, SYS_zone, 0, cmd);
1757 rval->sys_rval1 = GLOBAL_ZONEID;
1758 rval->sys_rval2 = 0;
1759 return (0);
1760
1761 case ZONE_GETATTR:
1762 zid = (zoneid_t)(uintptr_t)arg1;
1763 attr = (int)(uintptr_t)arg2;
1764 buf = (char *)arg3;
1765 bufsize = (size_t)arg4;
1766
1767 /*
1768 * If the request is for the global zone then we're emulating
1769 * that, otherwise pass this thru.
1770 */
1771 if (zid != GLOBAL_ZONEID)
1772 goto passthru;
1773
1774 switch (attr) {
1775 case ZONE_ATTR_NAME:
1776 aval = GLOBAL_ZONENAME;
1777 break;
1778
1779 case ZONE_ATTR_BRAND:
1780 aval = NATIVE_BRAND_NAME;
1781 break;
1782 default:
1783 /*
1784 * We only emulate a subset of the attrs, use the
1785 * real zone id to pass thru the rest.
1786 */
1787 arg1 = (void *)(uintptr_t)zoneid;
1788 goto passthru;
1789 }
1790
1791 (void) B_TRUSS_POINT_5(rval, SYS_zone, 0, cmd, zid, attr,
1792 buf, bufsize);
1793
1794 len = strlen(aval) + 1;
1795 if (len > bufsize)
1796 return (ENAMETOOLONG);
1797
1798 if (buf != NULL) {
1799 if (len == 1) {
1800 if (brand_uucopy("\0", buf, 1) != 0)
1801 return (EFAULT);
1802 } else {
1803 if (brand_uucopystr(aval, buf, len) != 0)
1804 return (EFAULT);
1805
1806 /*
1807 * Assure NULL termination of "buf" as
1808 * brand_uucopystr() does NOT.
1809 */
1810 if (brand_uucopy("\0", buf + (len - 1), 1) != 0)
1811 return (EFAULT);
1812 }
1813 }
1814
1815 rval->sys_rval1 = len;
1816 rval->sys_rval2 = 0;
1817 return (0);
1818
1819 default:
1820 break;
1821 }
1822
1823 passthru:
1824 return (__systemcall(rval, SYS_zone + 1024, cmd, arg1, arg2, arg3,
1825 arg4));
1826 }
1827
1828 /*ARGSUSED*/
1829 int
brand_init(int argc,char * argv[],char * envp[])1830 brand_init(int argc, char *argv[], char *envp[])
1831 {
1832 sysret_t rval;
1833 ulong_t ldentry;
1834 int err;
1835 char *bname;
1836
1837 brand_pre_init();
1838
1839 /*
1840 * Cache the pid of the zone's init process and determine if
1841 * we're init(8) for the zone. Remember: we might be init
1842 * now, but as soon as we fork(2) we won't be.
1843 */
1844 (void) get_initpid_info();
1845
1846 /* get the current zoneid */
1847 err = __systemcall(&rval, SYS_zone, ZONE_LOOKUP, NULL);
1848 brand_assert(err == 0);
1849 zoneid = (zoneid_t)rval.sys_rval1;
1850
1851 /* Get the zone's emulation bitmap. */
1852 if ((err = __systemcall(&rval, SYS_zone, ZONE_GETATTR, zoneid,
1853 S10_EMUL_BITMAP, emul_bitmap, sizeof (emul_bitmap))) != 0) {
1854 brand_abort(err, "The zone's patch level is unsupported");
1855 /*NOTREACHED*/
1856 }
1857
1858 bname = basename(argv[0]);
1859
1860 /*
1861 * In general we want the S10 commands that are zone-aware to continue
1862 * to behave as they normally do within a zone. Since these commands
1863 * are zone-aware, they should continue to "do the right thing".
1864 * However, some zone-aware commands aren't going to work the way
1865 * we expect them to inside the branded zone. In particular, the pkg
1866 * and patch commands will not properly manage all pkgs/patches
1867 * unless the commands think they are running in the global zone. For
1868 * these commands we want to emulate the global zone.
1869 *
1870 * We don't do any emulation for pkgcond since it is typically used
1871 * in pkg/patch postinstall scripts and we want those scripts to do
1872 * the right thing inside a zone.
1873 *
1874 * One issue is the handling of hollow pkgs. Since the pkgs are
1875 * hollow, they won't use pkgcond in their postinstall scripts. These
1876 * pkgs typically are installing drivers so we handle that by
1877 * replacing add_drv and rem_drv in the s10_boot script.
1878 */
1879 if (strcmp("pkgadd", bname) == 0 || strcmp("pkgrm", bname) == 0 ||
1880 strcmp("patchadd", bname) == 0 || strcmp("patchrm", bname) == 0)
1881 emul_global_zone = B_TRUE;
1882
1883 ldentry = brand_post_init(S10_VERSION, argc, argv, envp);
1884
1885 brand_runexe(argv, ldentry);
1886 /*NOTREACHED*/
1887 brand_abort(0, "brand_runexe() returned");
1888 return (-1);
1889 }
1890
1891 /*
1892 * This table must have at least NSYSCALL entries in it.
1893 *
1894 * The second parameter of each entry in the brand_sysent_table
1895 * contains the number of parameters and flags that describe the
1896 * syscall return value encoding. See the block comments at the
1897 * top of this file for more information about the syscall return
1898 * value flags and when they should be used.
1899 */
1900 brand_sysent_table_t brand_sysent_table[] = {
1901 #if defined(__sparc) && !defined(__sparcv9)
1902 EMULATE(brand_indir, 9 | RV_64RVAL), /* 0 */
1903 #else
1904 NOSYS, /* 0 */
1905 #endif
1906 NOSYS, /* 1 */
1907 EMULATE(s10_forkall, 0 | RV_32RVAL2), /* 2 */
1908 NOSYS, /* 3 */
1909 NOSYS, /* 4 */
1910 EMULATE(s10_open, 3 | RV_DEFAULT), /* 5 */
1911 NOSYS, /* 6 */
1912 EMULATE(s10_wait, 0 | RV_32RVAL2), /* 7 */
1913 EMULATE(s10_creat, 2 | RV_DEFAULT), /* 8 */
1914 EMULATE(s10_link, 2 | RV_DEFAULT), /* 9 */
1915 EMULATE(s10_unlink, 1 | RV_DEFAULT), /* 10 */
1916 EMULATE(s10_exec, 2 | RV_DEFAULT), /* 11 */
1917 NOSYS, /* 12 */
1918 NOSYS, /* 13 */
1919 EMULATE(s10_mknod, 3 | RV_DEFAULT), /* 14 */
1920 EMULATE(s10_chmod, 2 | RV_DEFAULT), /* 15 */
1921 EMULATE(s10_chown, 3 | RV_DEFAULT), /* 16 */
1922 NOSYS, /* 17 */
1923 EMULATE(s10_stat, 2 | RV_DEFAULT), /* 18 */
1924 NOSYS, /* 19 */
1925 NOSYS, /* 20 */
1926 NOSYS, /* 21 */
1927 EMULATE(s10_umount, 1 | RV_DEFAULT), /* 22 */
1928 NOSYS, /* 23 */
1929 NOSYS, /* 24 */
1930 NOSYS, /* 25 */
1931 NOSYS, /* 26 */
1932 NOSYS, /* 27 */
1933 EMULATE(s10_fstat, 2 | RV_DEFAULT), /* 28 */
1934 NOSYS, /* 29 */
1935 EMULATE(s10_utime, 2 | RV_DEFAULT), /* 30 */
1936 NOSYS, /* 31 */
1937 NOSYS, /* 32 */
1938 EMULATE(s10_access, 2 | RV_DEFAULT), /* 33 */
1939 NOSYS, /* 34 */
1940 NOSYS, /* 35 */
1941 NOSYS, /* 36 */
1942 EMULATE(s10_kill, 2 | RV_DEFAULT), /* 37 */
1943 NOSYS, /* 38 */
1944 NOSYS, /* 39 */
1945 NOSYS, /* 40 */
1946 EMULATE(s10_dup, 1 | RV_DEFAULT), /* 41 */
1947 EMULATE(s10_pipe, 0 | RV_32RVAL2), /* 42 */
1948 NOSYS, /* 43 */
1949 NOSYS, /* 44 */
1950 NOSYS, /* 45 */
1951 NOSYS, /* 46 */
1952 NOSYS, /* 47 */
1953 NOSYS, /* 48 */
1954 NOSYS, /* 49 */
1955 NOSYS, /* 50 */
1956 NOSYS, /* 51 */
1957 NOSYS, /* 52 */
1958 NOSYS, /* 53 */
1959 EMULATE(s10_ioctl, 3 | RV_DEFAULT), /* 54 */
1960 NOSYS, /* 55 */
1961 NOSYS, /* 56 */
1962 NOSYS, /* 57 */
1963 NOSYS, /* 58 */
1964 EMULATE(s10_execve, 3 | RV_DEFAULT), /* 59 */
1965 NOSYS, /* 60 */
1966 NOSYS, /* 61 */
1967 NOSYS, /* 62 */
1968 NOSYS, /* 63 */
1969 NOSYS, /* 64 */
1970 NOSYS, /* 65 */
1971 NOSYS, /* 66 */
1972 NOSYS, /* 67 */
1973 NOSYS, /* 68 */
1974 NOSYS, /* 69 */
1975 NOSYS, /* 70 */
1976 EMULATE(s10_acctctl, 3 | RV_DEFAULT), /* 71 */
1977 NOSYS, /* 72 */
1978 NOSYS, /* 73 */
1979 NOSYS, /* 74 */
1980 EMULATE(s10_issetugid, 0 | RV_DEFAULT), /* 75 */
1981 EMULATE(s10_fsat, 6 | RV_DEFAULT), /* 76 */
1982 NOSYS, /* 77 */
1983 NOSYS, /* 78 */
1984 EMULATE(s10_rmdir, 1 | RV_DEFAULT), /* 79 */
1985 EMULATE(s10_mkdir, 2 | RV_DEFAULT), /* 80 */
1986 EMULATE(s10_getdents, 3 | RV_DEFAULT), /* 81 */
1987 NOSYS, /* 82 */
1988 NOSYS, /* 83 */
1989 NOSYS, /* 84 */
1990 NOSYS, /* 85 */
1991 NOSYS, /* 86 */
1992 EMULATE(s10_poll, 3 | RV_DEFAULT), /* 87 */
1993 EMULATE(s10_lstat, 2 | RV_DEFAULT), /* 88 */
1994 EMULATE(s10_symlink, 2 | RV_DEFAULT), /* 89 */
1995 EMULATE(s10_readlink, 3 | RV_DEFAULT), /* 90 */
1996 NOSYS, /* 91 */
1997 NOSYS, /* 92 */
1998 EMULATE(s10_fchmod, 2 | RV_DEFAULT), /* 93 */
1999 EMULATE(s10_fchown, 3 | RV_DEFAULT), /* 94 */
2000 EMULATE(s10_sigprocmask, 3 | RV_DEFAULT), /* 95 */
2001 EMULATE(s10_sigsuspend, 1 | RV_DEFAULT), /* 96 */
2002 NOSYS, /* 97 */
2003 EMULATE(s10_sigaction, 3 | RV_DEFAULT), /* 98 */
2004 EMULATE(s10_sigpending, 2 | RV_DEFAULT), /* 99 */
2005 NOSYS, /* 100 */
2006 NOSYS, /* 101 */
2007 NOSYS, /* 102 */
2008 NOSYS, /* 103 */
2009 NOSYS, /* 104 */
2010 NOSYS, /* 105 */
2011 NOSYS, /* 106 */
2012 EMULATE(s10_waitid, 4 | RV_DEFAULT), /* 107 */
2013 EMULATE(s10_sigsendsys, 2 | RV_DEFAULT), /* 108 */
2014 NOSYS, /* 109 */
2015 NOSYS, /* 110 */
2016 NOSYS, /* 111 */
2017 NOSYS, /* 112 */
2018 NOSYS, /* 113 */
2019 NOSYS, /* 114 */
2020 NOSYS, /* 115 */
2021 NOSYS, /* 116 */
2022 NOSYS, /* 117 */
2023 NOSYS, /* 118 */
2024 NOSYS, /* 119 */
2025 NOSYS, /* 120 */
2026 NOSYS, /* 121 */
2027 NOSYS, /* 122 */
2028 #if defined(__x86)
2029 EMULATE(s10_xstat, 3 | RV_DEFAULT), /* 123 */
2030 EMULATE(s10_lxstat, 3 | RV_DEFAULT), /* 124 */
2031 EMULATE(s10_fxstat, 3 | RV_DEFAULT), /* 125 */
2032 EMULATE(s10_xmknod, 4 | RV_DEFAULT), /* 126 */
2033 #else
2034 NOSYS, /* 123 */
2035 NOSYS, /* 124 */
2036 NOSYS, /* 125 */
2037 NOSYS, /* 126 */
2038 #endif
2039 NOSYS, /* 127 */
2040 NOSYS, /* 128 */
2041 NOSYS, /* 129 */
2042 EMULATE(s10_lchown, 3 | RV_DEFAULT), /* 130 */
2043 NOSYS, /* 131 */
2044 NOSYS, /* 132 */
2045 NOSYS, /* 133 */
2046 EMULATE(s10_rename, 2 | RV_DEFAULT), /* 134 */
2047 EMULATE(s10_uname, 1 | RV_DEFAULT), /* 135 */
2048 NOSYS, /* 136 */
2049 EMULATE(s10_sysconfig, 1 | RV_DEFAULT), /* 137 */
2050 NOSYS, /* 138 */
2051 EMULATE(s10_sysinfo, 3 | RV_DEFAULT), /* 139 */
2052 NOSYS, /* 140 */
2053 NOSYS, /* 141 */
2054 NOSYS, /* 142 */
2055 EMULATE(s10_fork1, 0 | RV_32RVAL2), /* 143 */
2056 EMULATE(s10_sigtimedwait, 3 | RV_DEFAULT), /* 144 */
2057 NOSYS, /* 145 */
2058 NOSYS, /* 146 */
2059 EMULATE(s10_lwp_sema_wait, 1 | RV_DEFAULT), /* 147 */
2060 NOSYS, /* 148 */
2061 NOSYS, /* 149 */
2062 NOSYS, /* 150 */
2063 NOSYS, /* 151 */
2064 NOSYS, /* 152 */
2065 NOSYS, /* 153 */
2066 EMULATE(s10_utimes, 2 | RV_DEFAULT), /* 154 */
2067 NOSYS, /* 155 */
2068 NOSYS, /* 156 */
2069 NOSYS, /* 157 */
2070 NOSYS, /* 158 */
2071 EMULATE(s10_lwp_create, 3 | RV_DEFAULT), /* 159 */
2072 NOSYS, /* 160 */
2073 NOSYS, /* 161 */
2074 NOSYS, /* 162 */
2075 EMULATE(s10_lwp_kill, 2 | RV_DEFAULT), /* 163 */
2076 NOSYS, /* 164 */
2077 EMULATE(s10_lwp_sigmask, 3 | RV_32RVAL2), /* 165 */
2078 #if defined(__x86)
2079 EMULATE(s10_lwp_private, 3 | RV_DEFAULT), /* 166 */
2080 #else
2081 NOSYS, /* 166 */
2082 #endif
2083 NOSYS, /* 167 */
2084 NOSYS, /* 168 */
2085 EMULATE(s10_lwp_mutex_lock, 1 | RV_DEFAULT), /* 169 */
2086 NOSYS, /* 170 */
2087 NOSYS, /* 171 */
2088 NOSYS, /* 172 */
2089 NOSYS, /* 173 */
2090 EMULATE(s10_pwrite, 4 | RV_DEFAULT), /* 174 */
2091 NOSYS, /* 175 */
2092 NOSYS, /* 176 */
2093 NOSYS, /* 177 */
2094 NOSYS, /* 178 */
2095 NOSYS, /* 179 */
2096 NOSYS, /* 180 */
2097 NOSYS, /* 181 */
2098 NOSYS, /* 182 */
2099 NOSYS, /* 183 */
2100 NOSYS, /* 184 */
2101 EMULATE(s10_acl, 4 | RV_DEFAULT), /* 185 */
2102 EMULATE(s10_auditsys, 4 | RV_64RVAL), /* 186 */
2103 NOSYS, /* 187 */
2104 NOSYS, /* 188 */
2105 NOSYS, /* 189 */
2106 EMULATE(s10_sigqueue, 4 | RV_DEFAULT), /* 190 */
2107 NOSYS, /* 191 */
2108 NOSYS, /* 192 */
2109 NOSYS, /* 193 */
2110 NOSYS, /* 194 */
2111 NOSYS, /* 195 */
2112 NOSYS, /* 196 */
2113 NOSYS, /* 197 */
2114 NOSYS, /* 198 */
2115 NOSYS, /* 199 */
2116 EMULATE(s10_facl, 4 | RV_DEFAULT), /* 200 */
2117 NOSYS, /* 201 */
2118 NOSYS, /* 202 */
2119 NOSYS, /* 203 */
2120 NOSYS, /* 204 */
2121 EMULATE(s10_signotify, 3 | RV_DEFAULT), /* 205 */
2122 NOSYS, /* 206 */
2123 NOSYS, /* 207 */
2124 NOSYS, /* 208 */
2125 NOSYS, /* 209 */
2126 EMULATE(s10_lwp_mutex_timedlock, 2 | RV_DEFAULT), /* 210 */
2127 NOSYS, /* 211 */
2128 NOSYS, /* 212 */
2129 #if defined(_LP64)
2130 NOSYS, /* 213 */
2131 #else
2132 EMULATE(s10_getdents64, 3 | RV_DEFAULT), /* 213 */
2133 #endif
2134 NOSYS, /* 214 */
2135 #if defined(_LP64)
2136 NOSYS, /* 215 */
2137 NOSYS, /* 216 */
2138 NOSYS, /* 217 */
2139 #else
2140 EMULATE(s10_stat64, 2 | RV_DEFAULT), /* 215 */
2141 EMULATE(s10_lstat64, 2 | RV_DEFAULT), /* 216 */
2142 EMULATE(s10_fstat64, 2 | RV_DEFAULT), /* 217 */
2143 #endif
2144 NOSYS, /* 218 */
2145 NOSYS, /* 219 */
2146 NOSYS, /* 220 */
2147 NOSYS, /* 221 */
2148 NOSYS, /* 222 */
2149 #if defined(_LP64)
2150 NOSYS, /* 223 */
2151 NOSYS, /* 224 */
2152 NOSYS, /* 225 */
2153 #else
2154 EMULATE(s10_pwrite64, 5 | RV_DEFAULT), /* 223 */
2155 EMULATE(s10_creat64, 2 | RV_DEFAULT), /* 224 */
2156 EMULATE(s10_open64, 3 | RV_DEFAULT), /* 225 */
2157 #endif
2158 NOSYS, /* 226 */
2159 EMULATE(s10_zone, 5 | RV_DEFAULT), /* 227 */
2160 NOSYS, /* 228 */
2161 NOSYS, /* 229 */
2162 EMULATE(s10_so_socket, 5 | RV_DEFAULT), /* 230 */
2163 NOSYS, /* 231 */
2164 NOSYS, /* 232 */
2165 NOSYS, /* 233 */
2166 EMULATE(s10_accept, 4 | RV_DEFAULT), /* 234 */
2167 NOSYS, /* 235 */
2168 NOSYS, /* 236 */
2169 NOSYS, /* 237 */
2170 NOSYS, /* 238 */
2171 NOSYS, /* 239 */
2172 NOSYS, /* 240 */
2173 NOSYS, /* 241 */
2174 NOSYS, /* 242 */
2175 NOSYS, /* 243 */
2176 NOSYS, /* 244 */
2177 NOSYS, /* 245 */
2178 NOSYS, /* 246 */
2179 NOSYS, /* 247 */
2180 NOSYS, /* 248 */
2181 NOSYS, /* 249 */
2182 NOSYS, /* 250 */
2183 EMULATE(s10_lwp_mutex_trylock, 1 | RV_DEFAULT), /* 251 */
2184 NOSYS, /* 252 */
2185 NOSYS, /* 253 */
2186 NOSYS, /* 254 */
2187 NOSYS /* 255 */
2188 };
2189