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