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