X-Git-Url: https://git.saurik.com/apple/xnu.git/blobdiff_plain/ebb1b9f42b62218f29061826217bb0f71cd375a6..fe8ab488e9161c46dd9885d58fc52996dc0249ff:/bsd/kern/kern_exec.c diff --git a/bsd/kern/kern_exec.c b/bsd/kern/kern_exec.c index 4c17cd232..3d2710538 100644 --- a/bsd/kern/kern_exec.c +++ b/bsd/kern/kern_exec.c @@ -33,8 +33,6 @@ * the terms and conditions for use and redistribution. */ -#include - /*- * Copyright (c) 1982, 1986, 1991, 1993 * The Regents of the University of California. All rights reserved. @@ -81,6 +79,7 @@ * Version 2.0. */ #include +#include #include #include @@ -109,9 +108,12 @@ #include /* ubc_map() */ #include #include +#include #include #include +#include + #include #include @@ -128,6 +130,8 @@ #include /* thread_wakeup() */ #include #include +#include +#include #if CONFIG_MACF #include @@ -138,23 +142,40 @@ #include #include #include +#include +#include + +#include #include +#include + +#if CONFIG_MEMORYSTATUS +#include +#endif + #if CONFIG_DTRACE /* Do not include dtrace.h, it redefines kmem_[alloc/free] */ extern void (*dtrace_fasttrap_exec_ptr)(proc_t); +extern void (*dtrace_proc_waitfor_exec_ptr)(proc_t); extern void (*dtrace_helpers_cleanup)(proc_t); extern void dtrace_lazy_dofs_destroy(proc_t); +/* + * Since dtrace_proc_waitfor_exec_ptr can be added/removed in dtrace_subr.c, + * we will store its value before actually calling it. + */ +static void (*dtrace_proc_waitfor_hook)(proc_t) = NULL; + #include #endif /* support for child creation in exec after vfork */ -thread_t fork_create_child(task_t parent_task, proc_t child_proc, int inherit_memory, int is64bit); +thread_t fork_create_child(task_t parent_task, coalition_t parent_coalition, proc_t child_proc, int inherit_memory, int is64bit); void vfork_exit(proc_t p, int rv); int setsigvec(proc_t, thread_t, int, struct __kern_sigaction *, boolean_t in_sigstart); -extern void proc_apply_task_networkbg_internal(proc_t); +extern void proc_apply_task_networkbg_internal(proc_t, thread_t); /* * Mach things for which prototypes are unavailable from Mach headers @@ -170,8 +191,13 @@ kern_return_t ipc_object_copyin( ipc_object_t *objectp); void ipc_port_release_send(ipc_port_t); +#if DEVELOPMENT || DEBUG +void task_importance_update_owner_info(task_t); +#endif + extern struct savearea *get_user_regs(thread_t); +__attribute__((noinline)) int __EXEC_WAITING_ON_TASKGATED_CODE_SIGNATURE_UPCALL__(mach_port_t task_access_port, int32_t new_pid); #include #include @@ -200,7 +226,7 @@ extern struct savearea *get_user_regs(thread_t); #define IS_EOL(ch) ((ch == '#') || (ch == '\n')) extern vm_map_t bsd_pageable_map; -extern struct fileops vnops; +extern const struct fileops vnops; #define ROUND_PTR(type, addr) \ (type *)( ( (uintptr_t)(addr) + 16 - 1) \ @@ -218,15 +244,14 @@ static int exec_add_apple_strings(struct image_params *imgp); static int exec_handle_sugid(struct image_params *imgp); static int sugid_scripts = 0; SYSCTL_INT (_kern, OID_AUTO, sugid_scripts, CTLFLAG_RW | CTLFLAG_LOCKED, &sugid_scripts, 0, ""); -static kern_return_t create_unix_stack(vm_map_t map, user_addr_t user_stack, - int customstack, proc_t p); +static kern_return_t create_unix_stack(vm_map_t map, load_result_t* load_result, proc_t p); static int copyoutptr(user_addr_t ua, user_addr_t ptr, int ptr_size); static void exec_resettextvp(proc_t, struct image_params *); static int check_for_signature(proc_t, struct image_params *); - -/* We don't want this one exported */ -__private_extern__ -int open1(vfs_context_t, struct nameidata *, int, struct vnode_attr *, int32_t *); +static void exec_prefault_data(proc_t, struct image_params *, load_result_t *); +static errno_t exec_handle_port_actions(struct image_params *imgp, short psa_flags, boolean_t * portwatch_present, ipc_port_t * portwatch_ports); +static errno_t exec_handle_spawnattr_policy(proc_t p, int psa_apptype, uint64_t psa_qos_clamp, + ipc_port_t * portwatch_ports, int portwatch_count); /* * exec_add_user_string @@ -374,96 +399,6 @@ exec_reset_save_path(struct image_params *imgp) return (0); } -#ifdef IMGPF_POWERPC -/* - * exec_powerpc32_imgact - * - * Implicitly invoke the PowerPC handler for a byte-swapped image magic - * number. This may happen either as a result of an attempt to invoke a - * PowerPC image directly, or indirectly as the interpreter used in an - * interpreter script. - * - * Parameters; struct image_params * image parameter block - * - * Returns: -1 not an PowerPC image (keep looking) - * -3 Success: exec_archhandler_ppc: relookup - * >0 Failure: exec_archhandler_ppc: error number - * - * Note: This image activator does not handle the case of a direct - * invocation of the exec_archhandler_ppc, since in that case, the - * exec_archhandler_ppc itself is not a PowerPC binary; instead, - * binary image activators must recognize the exec_archhandler_ppc; - * This is managed in exec_check_permissions(). - * - * Note: This image activator is limited to 32 bit powerpc images; - * if support for 64 bit powerpc images is desired, it would - * be more in line with this design to write a separate 64 bit - * image activator. - */ -static int -exec_powerpc32_imgact(struct image_params *imgp) -{ - struct mach_header *mach_header = (struct mach_header *)imgp->ip_vdata; - int error; - size_t len = 0; - - /* - * Make sure it's a PowerPC binary. If we've already redirected - * from an interpreted file once, don't do it again. - */ - if (mach_header->magic != MH_CIGAM) { - /* - * If it's a cross-architecture 64 bit binary, then claim - * it, but refuse to run it. - */ - if (mach_header->magic == MH_CIGAM_64) - return (EBADARCH); - return (-1); - } - - /* If there is no exec_archhandler_ppc, we can't run it */ - if (exec_archhandler_ppc.path[0] == 0) - return (EBADARCH); - - /* Remember the type of the original file for later grading */ - if (!imgp->ip_origcputype) { - imgp->ip_origcputype = - OSSwapBigToHostInt32(mach_header->cputype); - imgp->ip_origcpusubtype = - OSSwapBigToHostInt32(mach_header->cpusubtype); - } - - /* - * The PowerPC flag will be set by the exec_check_permissions() - * call anyway; however, we set this flag here so that the relookup - * in execve() does not follow symbolic links, as a side effect. - */ - imgp->ip_flags |= IMGPF_POWERPC; - - /* impute an interpreter */ - error = copystr(exec_archhandler_ppc.path, imgp->ip_interp_buffer, - IMG_SHSIZE, &len); - if (error) - return (error); - - exec_reset_save_path(imgp); - exec_save_path(imgp, CAST_USER_ADDR_T(imgp->ip_interp_buffer), - UIO_SYSSPACE); - - /* - * provide a replacement string for p->p_comm; we have to use an - * alternate buffer for this, rather than replacing it directly, - * since the exec may fail and return to the parent. In that case, - * we would have erroneously changed the parent p->p_comm instead. - */ - strlcpy(imgp->ip_p_comm, imgp->ip_ndp->ni_cnd.cn_nameptr, MAXCOMLEN+1); - /* +1 to allow MAXCOMLEN characters to be copied */ - - return (-3); -} -#endif /* IMGPF_POWERPC */ - - /* * exec_shell_imgact * @@ -511,11 +446,6 @@ exec_shell_imgact(struct image_params *imgp) return (-1); } -#ifdef IMGPF_POWERPC - if ((imgp->ip_flags & IMGPF_POWERPC) != 0) - return (EBADARCH); -#endif /* IMGPF_POWERPC */ - imgp->ip_flags |= IMGPF_INTERPRET; imgp->ip_interp_sugid_fd = -1; imgp->ip_interp_buffer[0] = '\0'; @@ -606,7 +536,6 @@ exec_shell_imgact(struct image_params *imgp) return(error); fp->f_fglob->fg_flag = FREAD; - fp->f_fglob->fg_type = DTYPE_VNODE; fp->f_fglob->fg_ops = &vnops; fp->f_fglob->fg_data = (caddr_t)imgp->ip_vp; @@ -668,6 +597,44 @@ exec_fat_imgact(struct image_params *imgp) goto bad; } +#if DEVELOPMENT || DEBUG + if (cpu_type() == CPU_TYPE_ARM64) { + uint32_t fat_nfat_arch = OSSwapBigToHostInt32(fat_header->nfat_arch); + struct fat_arch *archs; + int vfexec = (imgp->ip_flags & IMGPF_VFORK_EXEC); + int spawn = (imgp->ip_flags & IMGPF_SPAWN); + + archs = (struct fat_arch *)(imgp->ip_vdata + sizeof(struct fat_header)); + + /* ip_vdata always has PAGE_SIZE of data */ + if (PAGE_SIZE >= (sizeof(struct fat_header) + (fat_nfat_arch + 1) * sizeof(struct fat_arch))) { + if (fat_nfat_arch > 0 + && OSSwapBigToHostInt32(archs[fat_nfat_arch].cputype) == CPU_TYPE_ARM64) { + + /* rdar://problem/15001727 */ + printf("Attempt to execute malformed binary %s\n", imgp->ip_strings); + + proc_lock(p); + p->p_csflags |= CS_KILLED; + proc_unlock(p); + + /* + * We can't stop the system call, so make sure the child never executes + * For vfork exec, the current implementation has not set up the thread in the + * child process, so we cannot signal it. Return an error code in that case. + */ + if (!vfexec && !spawn) { + psignal(p, SIGKILL); + error = 0; + } else { + error = EBADEXEC; + } + goto bad; + } + } + } +#endif + /* If posix_spawn binprefs exist, respect those prefs. */ psa = (struct _posix_spawnattr *) imgp->ip_px_sa; if (psa != NULL && psa->psa_binprefs[0] != 0) { @@ -776,7 +743,6 @@ exec_mach_imgact(struct image_params *imgp) struct mach_header *mach_header = (struct mach_header *)imgp->ip_vdata; proc_t p = vfs_context_proc(imgp->ip_vfs_context); int error = 0; - int vfexec = 0; task_t task; task_t new_task = NULL; /* protected by vfexec */ thread_t thread; @@ -786,14 +752,21 @@ exec_mach_imgact(struct image_params *imgp) load_return_t lret; load_result_t load_result; struct _posix_spawnattr *psa = NULL; - int spawn = (imgp->ip_flags & IMGPF_SPAWN); - int apptype = 0; + int spawn = (imgp->ip_flags & IMGPF_SPAWN); + int vfexec = (imgp->ip_flags & IMGPF_VFORK_EXEC); /* * make sure it's a Mach-O 1.0 or Mach-O 2.0 binary; the difference * is a reserved field on the end, so for the most part, we can - * treat them as if they were identical. - */ + * treat them as if they were identical. Reverse-endian Mach-O + * binaries are recognized but not compatible. + */ + if ((mach_header->magic == MH_CIGAM) || + (mach_header->magic == MH_CIGAM_64)) { + error = EBADARCH; + goto bad; + } + if ((mach_header->magic != MH_MAGIC) && (mach_header->magic != MH_MAGIC_64)) { error = -1; @@ -816,15 +789,6 @@ exec_mach_imgact(struct image_params *imgp) thread = current_thread(); uthread = get_bsdthread_info(thread); - /* - * Save off the vfexec state up front; we have to do this, because - * we need to know if we were in this state initially subsequent to - * creating the backing task, thread, and uthread for the child - * process (from the vfs_context_t from in img_parms). - */ - if (uthread->uu_flag & UT_VFORK) - vfexec = 1; /* Mark in exec */ - if ((mach_header->cputype & CPU_ARCH_ABI64) == CPU_ARCH_ABI64) imgp->ip_flags |= IMGPF_IS_64BIT; @@ -854,8 +818,7 @@ exec_mach_imgact(struct image_params *imgp) goto bad; } grade: - if (!grade_binary(imgp->ip_origcputype & ~CPU_SUBTYPE_LIB64, - imgp->ip_origcpusubtype & ~CPU_SUBTYPE_MASK)) { + if (!grade_binary(imgp->ip_origcputype, imgp->ip_origcpusubtype & ~CPU_SUBTYPE_MASK)) { error = EBADARCH; goto bad; } @@ -874,21 +837,6 @@ grade: AUDIT_ARG(envv, imgp->ip_endargv, imgp->ip_envc, imgp->ip_endenvv - imgp->ip_endargv); -#ifdef IMGPF_POWERPC - /* - * XXX - * - * Should be factored out; this is here because we might be getting - * invoked this way as the result of a shell script, and the check - * in exec_check_permissions() is not interior to the jump back up - * to the "encapsulated_binary:" label in exec_activate_image(). - */ - if (imgp->ip_vattr->va_fsid == exec_archhandler_ppc.fsid && - imgp->ip_vattr->va_fileid == exec_archhandler_ppc.fileid) { - imgp->ip_flags |= IMGPF_POWERPC; - } -#endif /* IMGPF_POWERPC */ - /* * We are being called to activate an image subsequent to a vfork() * operation; in this case, we know that our task, thread, and @@ -898,7 +846,7 @@ grade: */ if (vfexec || spawn) { if (vfexec) { - imgp->ip_new_thread = fork_create_child(task, p, FALSE, (imgp->ip_flags & IMGPF_IS_64BIT)); + imgp->ip_new_thread = fork_create_child(task, COALITION_NULL, p, FALSE, (imgp->ip_flags & IMGPF_IS_64BIT)); if (imgp->ip_new_thread == NULL) { error = ENOMEM; goto bad; @@ -933,6 +881,10 @@ grade: * NOTE: An error after this point indicates we have potentially * destroyed or overwritten some process state while attempting an * execve() following a vfork(), which is an unrecoverable condition. + * We send the new process an immediate SIGKILL to avoid it executing + * any instructions in the mutated address space. For true spawns, + * this is not the case, and "too late" is still not too late to + * return an error code to the parent process. */ /* @@ -945,6 +897,11 @@ grade: goto badtoolate; } + proc_lock(p); + p->p_cputype = imgp->ip_origcputype; + p->p_cpusubtype = imgp->ip_origcpusubtype; + proc_unlock(p); + vm_map_set_user_wire_limit(get_task_map(task), p->p_rlimit[RLIMIT_MEMLOCK].rlim_cur); /* @@ -952,9 +909,10 @@ grade: * requested them on exec. */ if (load_result.csflags & CS_VALID) { - imgp->ip_csflags |= load_result.csflags & + imgp->ip_csflags |= load_result.csflags & (CS_VALID| - CS_HARD|CS_KILL|CS_EXEC_SET_HARD|CS_EXEC_SET_KILL); + CS_HARD|CS_KILL|CS_ENFORCEMENT|CS_REQUIRE_LV|CS_DYLD_PLATFORM| + CS_EXEC_SET_HARD|CS_EXEC_SET_KILL|CS_EXEC_SET_ENFORCEMENT); } else { imgp->ip_csflags &= ~CS_VALID; } @@ -963,6 +921,10 @@ grade: imgp->ip_csflags |= CS_HARD; if (p->p_csflags & CS_EXEC_SET_KILL) imgp->ip_csflags |= CS_KILL; + if (p->p_csflags & CS_EXEC_SET_ENFORCEMENT) + imgp->ip_csflags |= CS_ENFORCEMENT; + if (p->p_csflags & CS_EXEC_SET_INSTALLER) + imgp->ip_csflags |= CS_INSTALLER; /* @@ -971,10 +933,6 @@ grade: vm_map_exec(get_task_map(task), task, (void *) p->p_fd->fd_rdir, -#ifdef IMGPF_POWERPC - imgp->ip_flags & IMGPF_POWERPC ? - CPU_TYPE_POWERPC : -#endif cpu_type()); /* @@ -986,19 +944,23 @@ grade: * deal with set[ug]id. */ error = exec_handle_sugid(imgp); + if (error) { + goto badtoolate; + } + + /* + * deal with voucher on exec-calling thread. + */ + if (imgp->ip_new_thread == NULL) + thread_set_mach_voucher(current_thread(), IPC_VOUCHER_NULL); /* Make sure we won't interrupt ourself signalling a partial process */ if (!vfexec && !spawn && (p->p_lflag & P_LTRACED)) psignal(p, SIGTRAP); - if (error) { - goto badtoolate; - } - if (load_result.unixproc && create_unix_stack(get_task_map(task), - load_result.user_stack, - load_result.customstack, + &load_result, p) != KERN_SUCCESS) { error = load_return_to_errno(LOAD_NOSPACE); goto badtoolate; @@ -1035,14 +997,17 @@ grade: error = copyoutptr(load_result.mach_header, ap, new_ptr_size); if (error) { - if (vfexec || spawn) - vm_map_switch(old_map); + if (vfexec || spawn) + vm_map_switch(old_map); goto badtoolate; } task_set_dyld_info(task, load_result.all_image_info_addr, load_result.all_image_info_size); } + /* Avoid immediate VM faults back into kernel */ + exec_prefault_data(p, imgp, &load_result); + if (vfexec || spawn) { vm_map_switch(old_map); } @@ -1094,6 +1059,13 @@ grade: pal_dbg_set_task_name( p->task ); +#if DEVELOPMENT || DEBUG + /* + * Update the pid an proc name for importance base if any + */ + task_importance_update_owner_info(p->task); +#endif + memcpy(&p->p_uuid[0], &load_result.uuid[0], sizeof(p->p_uuid)); // dtrace code cleanup needed @@ -1128,8 +1100,8 @@ grade: */ proc_lock(p); if (p->p_dtrace_probes && dtrace_fasttrap_exec_ptr) { - (*dtrace_fasttrap_exec_ptr)(p); - } + (*dtrace_fasttrap_exec_ptr)(p); + } proc_unlock(p); #endif @@ -1154,18 +1126,11 @@ grade: } } -#ifdef IMGPF_POWERPC /* - * Mark the process as powerpc or not. If powerpc, set the affinity - * flag, which will be used for grading binaries in future exec's - * from the process. + * Ensure the 'translated' and 'affinity' flags are cleared, since we + * no longer run PowerPC binaries. */ - if (((imgp->ip_flags & IMGPF_POWERPC) != 0)) - OSBitOrAtomic(P_TRANSLATED, &p->p_flag); - else -#endif /* IMGPF_POWERPC */ - OSBitAndAtomic(~((uint32_t)P_TRANSLATED), &p->p_flag); - OSBitAndAtomic(~((uint32_t)P_AFFINITY), &p->p_flag); + OSBitAndAtomic(~((uint32_t)(P_TRANSLATED | P_AFFINITY)), &p->p_flag); /* * If posix_spawned with the START_SUSPENDED flag, stop the @@ -1179,22 +1144,6 @@ grade: proc_unlock(p); (void) task_suspend(p->task); } - if ((psa->psa_flags & POSIX_SPAWN_OSX_TALAPP_START) || (psa->psa_flags & POSIX_SPAWN_OSX_DBCLIENT_START) || (psa->psa_flags & POSIX_SPAWN_IOS_APP_START)) { - if ((psa->psa_flags & POSIX_SPAWN_OSX_TALAPP_START)) - apptype = PROC_POLICY_OSX_APPTYPE_TAL; - else if (psa->psa_flags & POSIX_SPAWN_OSX_DBCLIENT_START) - apptype = PROC_POLICY_OSX_APPTYPE_DBCLIENT; - else if (psa->psa_flags & POSIX_SPAWN_IOS_APP_START) - apptype = PROC_POLICY_IOS_APPTYPE; - else - apptype = 0; - proc_set_task_apptype(p->task, apptype); - if ((apptype == PROC_POLICY_OSX_APPTYPE_TAL) || - (apptype == PROC_POLICY_OSX_APPTYPE_DBCLIENT)) { - - proc_apply_task_networkbg_internal(p); - } - } } /* @@ -1218,15 +1167,32 @@ grade: psignal_vfork(p, new_task, thread, SIGTRAP); } + goto done; + badtoolate: -if (!spawn) - proc_knote(p, NOTE_EXEC); + /* Don't allow child process to execute any instructions */ + if (!spawn) { + if (vfexec) { + psignal_vfork(p, new_task, thread, SIGKILL); + } else { + psignal(p, SIGKILL); + } - if (vfexec || spawn) { + /* We can't stop this system call at this point, so just pretend we succeeded */ + error = 0; + } + +done: + if (!spawn) { + /* notify only if it has not failed due to FP Key error */ + if ((p->p_lflag & P_LTERM_DECRYPTFAIL) == 0) + proc_knote(p, NOTE_EXEC); + } + + /* Drop extra references for cases where we don't expect the caller to clean up */ + if (vfexec || (spawn && error == 0)) { task_deallocate(new_task); thread_deallocate(thread); - if (error) - error = 0; } bad: @@ -1249,9 +1215,6 @@ struct execsw { } execsw[] = { { exec_mach_imgact, "Mach-o Binary" }, { exec_fat_imgact, "Fat Binary" }, -#ifdef IMGPF_POWERPC - { exec_powerpc32_imgact, "PowerPC binary" }, -#endif /* IMGPF_POWERPC */ { exec_shell_imgact, "Interpreter Script" }, { NULL, NULL} }; @@ -1285,7 +1248,7 @@ struct execsw { static int exec_activate_image(struct image_params *imgp) { - struct nameidata nd; + struct nameidata *ndp = NULL; int error; int resid; int once = 1; /* save SGUID-ness for interpreted files */ @@ -1295,7 +1258,7 @@ exec_activate_image(struct image_params *imgp) error = execargs_alloc(imgp); if (error) - goto bad; + goto bad_notrans; error = exec_save_path(imgp, imgp->ip_user_fname, imgp->ip_seg); if (error) { @@ -1305,15 +1268,21 @@ exec_activate_image(struct image_params *imgp) /* Use imgp->ip_strings, which contains the copyin-ed exec path */ DTRACE_PROC1(exec, uintptr_t, imgp->ip_strings); - NDINIT(&nd, LOOKUP, OP_LOOKUP, FOLLOW | LOCKLEAF | AUDITVNPATH1, + MALLOC(ndp, struct nameidata *, sizeof(*ndp), M_TEMP, M_WAITOK | M_ZERO); + if (ndp == NULL) { + error = ENOMEM; + goto bad_notrans; + } + + NDINIT(ndp, LOOKUP, OP_LOOKUP, FOLLOW | LOCKLEAF | AUDITVNPATH1, UIO_SYSSPACE, CAST_USER_ADDR_T(imgp->ip_strings), imgp->ip_vfs_context); again: - error = namei(&nd); + error = namei(ndp); if (error) goto bad_notrans; - imgp->ip_ndp = &nd; /* successful namei(); call nameidone() later */ - imgp->ip_vp = nd.ni_vp; /* if set, need to vnode_put() at some point */ + imgp->ip_ndp = ndp; /* successful namei(); call nameidone() later */ + imgp->ip_vp = ndp->ni_vp; /* if set, need to vnode_put() at some point */ /* * Before we start the transition from binary A to binary B, make @@ -1327,7 +1296,7 @@ again: proc_unlock(p); goto bad_notrans; } - error = proc_transstart(p, 1); + error = proc_transstart(p, 1, 0); proc_unlock(p); if (error) goto bad_notrans; @@ -1381,28 +1350,26 @@ encapsulated_binary: } mac_vnode_label_copy(imgp->ip_vp->v_label, imgp->ip_scriptlabelp); + + /* + * Take a ref of the script vnode for later use. + */ + if (imgp->ip_scriptvp) + vnode_put(imgp->ip_scriptvp); + if (vnode_getwithref(imgp->ip_vp) == 0) + imgp->ip_scriptvp = imgp->ip_vp; #endif - nameidone(&nd); + nameidone(ndp); vnode_put(imgp->ip_vp); imgp->ip_vp = NULL; /* already put */ imgp->ip_ndp = NULL; /* already nameidone */ /* Use imgp->ip_strings, which exec_shell_imgact reset to the interpreter */ - NDINIT(&nd, LOOKUP, OP_LOOKUP, FOLLOW | LOCKLEAF, + NDINIT(ndp, LOOKUP, OP_LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, CAST_USER_ADDR_T(imgp->ip_strings), imgp->ip_vfs_context); -#ifdef IMGPF_POWERPC - /* - * PowerPC does not follow symlinks because the - * code which sets exec_archhandler_ppc.fsid and - * exec_archhandler_ppc.fileid doesn't follow them. - */ - if (imgp->ip_flags & IMGPF_POWERPC) - nd.ni_cnd.cn_flags &= ~FOLLOW; -#endif /* IMGPF_POWERPC */ - proc_transend(p, 0); goto again; @@ -1418,7 +1385,7 @@ encapsulated_binary: if (error == 0 && kauth_authorize_fileop_has_listeners()) { kauth_authorize_fileop(vfs_context_ucred(imgp->ip_vfs_context), KAUTH_FILEOP_EXEC, - (uintptr_t)nd.ni_vp, 0); + (uintptr_t)ndp->ni_vp, 0); } bad: @@ -1429,10 +1396,86 @@ bad_notrans: execargs_free(imgp); if (imgp->ip_ndp) nameidone(imgp->ip_ndp); + if (ndp) + FREE(ndp, M_TEMP); return (error); } + +/* + * exec_handle_spawnattr_policy + * + * Description: Decode and apply the posix_spawn apptype, qos clamp, and watchport ports to the task. + * + * Parameters: proc_t p process to apply attributes to + * int psa_apptype posix spawn attribute apptype + * + * Returns: 0 Success + */ +static errno_t +exec_handle_spawnattr_policy(proc_t p, int psa_apptype, uint64_t psa_qos_clamp, + ipc_port_t * portwatch_ports, int portwatch_count) +{ + int apptype = TASK_APPTYPE_NONE; + int qos_clamp = THREAD_QOS_UNSPECIFIED; + + if ((psa_apptype & POSIX_SPAWN_PROC_TYPE_MASK) != 0) { + int proctype = psa_apptype & POSIX_SPAWN_PROC_TYPE_MASK; + + switch(proctype) { + case POSIX_SPAWN_PROC_TYPE_DAEMON_INTERACTIVE: + apptype = TASK_APPTYPE_DAEMON_INTERACTIVE; + break; + case POSIX_SPAWN_PROC_TYPE_DAEMON_STANDARD: + apptype = TASK_APPTYPE_DAEMON_STANDARD; + break; + case POSIX_SPAWN_PROC_TYPE_DAEMON_ADAPTIVE: + apptype = TASK_APPTYPE_DAEMON_ADAPTIVE; + break; + case POSIX_SPAWN_PROC_TYPE_DAEMON_BACKGROUND: + apptype = TASK_APPTYPE_DAEMON_BACKGROUND; + break; + case POSIX_SPAWN_PROC_TYPE_APP_DEFAULT: + apptype = TASK_APPTYPE_APP_DEFAULT; + break; + case POSIX_SPAWN_PROC_TYPE_APP_TAL: + apptype = TASK_APPTYPE_APP_TAL; + break; + default: + apptype = TASK_APPTYPE_NONE; + /* TODO: Should an invalid value here fail the spawn? */ + break; + } + } + + if (psa_qos_clamp != POSIX_SPAWN_PROC_CLAMP_NONE) { + switch (psa_qos_clamp) { + case POSIX_SPAWN_PROC_CLAMP_UTILITY: + qos_clamp = THREAD_QOS_UTILITY; + break; + case POSIX_SPAWN_PROC_CLAMP_BACKGROUND: + qos_clamp = THREAD_QOS_BACKGROUND; + break; + case POSIX_SPAWN_PROC_CLAMP_MAINTENANCE: + qos_clamp = THREAD_QOS_MAINTENANCE; + break; + default: + qos_clamp = THREAD_QOS_UNSPECIFIED; + /* TODO: Should an invalid value here fail the spawn? */ + break; + } + } + + if (psa_apptype != TASK_APPTYPE_NONE || qos_clamp != THREAD_QOS_UNSPECIFIED) { + proc_set_task_spawnpolicy(p->task, apptype, qos_clamp, + portwatch_ports, portwatch_count); + } + + return (0); +} + + /* * exec_handle_port_actions * @@ -1448,64 +1491,78 @@ bad_notrans: * ENOTSUP Illegal posix_spawn attr flag was set */ static errno_t -exec_handle_port_actions(struct image_params *imgp, short psa_flags) +exec_handle_port_actions(struct image_params *imgp, short psa_flags, boolean_t * portwatch_present, ipc_port_t * portwatch_ports) { _posix_spawn_port_actions_t pacts = imgp->ip_px_spa; proc_t p = vfs_context_proc(imgp->ip_vfs_context); _ps_port_action_t *act = NULL; task_t task = p->task; ipc_port_t port = NULL; - errno_t ret = KERN_SUCCESS; + errno_t ret = 0; int i; + *portwatch_present = FALSE; + for (i = 0; i < pacts->pspa_count; i++) { act = &pacts->pspa_actions[i]; if (ipc_object_copyin(get_task_ipcspace(current_task()), - CAST_MACH_PORT_TO_NAME(act->new_port), - MACH_MSG_TYPE_COPY_SEND, - (ipc_object_t *) &port) != KERN_SUCCESS) - return EINVAL; - - if (ret) - return ret; + act->new_port, MACH_MSG_TYPE_COPY_SEND, + (ipc_object_t *) &port) != KERN_SUCCESS) { + ret = EINVAL; + goto done; + } switch (act->port_type) { - case PSPA_SPECIAL: - /* Only allowed when not under vfork */ - if (!(psa_flags & POSIX_SPAWN_SETEXEC)) - return ENOTSUP; - ret = (task_set_special_port(task, - act->which, - port) == KERN_SUCCESS) ? 0 : EINVAL; - break; - case PSPA_EXCEPTION: - /* Only allowed when not under vfork */ - if (!(psa_flags & POSIX_SPAWN_SETEXEC)) - return ENOTSUP; - ret = (task_set_exception_ports(task, - act->mask, - port, - act->behavior, - act->flavor) == KERN_SUCCESS) ? 0 : EINVAL; - break; + case PSPA_SPECIAL: + /* Only allowed when not under vfork */ + if (!(psa_flags & POSIX_SPAWN_SETEXEC)) + ret = ENOTSUP; + else if (task_set_special_port(task, + act->which, port) != KERN_SUCCESS) + ret = EINVAL; + break; + + case PSPA_EXCEPTION: + /* Only allowed when not under vfork */ + if (!(psa_flags & POSIX_SPAWN_SETEXEC)) + ret = ENOTSUP; + else if (task_set_exception_ports(task, + act->mask, port, act->behavior, + act->flavor) != KERN_SUCCESS) + ret = EINVAL; + break; #if CONFIG_AUDIT - case PSPA_AU_SESSION: - ret = audit_session_spawnjoin(p, - port); - break; + case PSPA_AU_SESSION: + ret = audit_session_spawnjoin(p, port); + break; #endif - default: - ret = EINVAL; + case PSPA_IMP_WATCHPORTS: + if (portwatch_ports != NULL) { + *portwatch_present = TRUE; + /* hold on to this till end of spawn */ + portwatch_ports[i] = port; + ret = 0; + } else + ipc_port_release_send(port); + break; + default: + ret = EINVAL; + break; } + /* action failed, so release port resources */ + if (ret) { ipc_port_release_send(port); - return ret; + break; } } - return ret; +done: + if (0 != ret) + DTRACE_PROC1(spawn__port__failure, mach_port_name_t, act->new_port); + return (ret); } /* @@ -1548,28 +1605,41 @@ exec_handle_file_actions(struct image_params *imgp, short psa_flags) * context of UIO_SYSSPACE, and casts the address * argument to a user_addr_t. */ - struct vnode_attr va; - struct nameidata nd; + char *bufp = NULL; + struct vnode_attr *vap; + struct nameidata *ndp; int mode = psfa->psfaa_openargs.psfao_mode; struct dup2_args dup2a; struct close_nocancel_args ca; int origfd; - VATTR_INIT(&va); + MALLOC(bufp, char *, sizeof(*vap) + sizeof(*ndp), M_TEMP, M_WAITOK | M_ZERO); + if (bufp == NULL) { + error = ENOMEM; + break; + } + + vap = (struct vnode_attr *) bufp; + ndp = (struct nameidata *) (bufp + sizeof(*vap)); + + VATTR_INIT(vap); /* Mask off all but regular access permissions */ mode = ((mode &~ p->p_fd->fd_cmask) & ALLPERMS) & ~S_ISTXT; - VATTR_SET(&va, va_mode, mode & ACCESSPERMS); + VATTR_SET(vap, va_mode, mode & ACCESSPERMS); - NDINIT(&nd, LOOKUP, OP_OPEN, FOLLOW | AUDITVNPATH1, UIO_SYSSPACE, + NDINIT(ndp, LOOKUP, OP_OPEN, FOLLOW | AUDITVNPATH1, UIO_SYSSPACE, CAST_USER_ADDR_T(psfa->psfaa_openargs.psfao_path), imgp->ip_vfs_context); error = open1(imgp->ip_vfs_context, - &nd, + ndp, psfa->psfaa_openargs.psfao_oflag, - &va, + vap, + fileproc_alloc_init, NULL, ival); + FREE(bufp, M_TEMP); + /* * If there's an error, or we get the right fd by * accident, then drop out here. This is easier than @@ -1633,20 +1703,27 @@ exec_handle_file_actions(struct image_params *imgp, short psa_flags) break; case PSFA_INHERIT: { - struct fileproc *fp; - int fd = psfa->psfaa_filedes; + struct fcntl_nocancel_args fcntla; /* * Check to see if the descriptor exists, and * ensure it's -not- marked as close-on-exec. - * [Less code than the equivalent F_GETFD/F_SETFD.] + * + * Attempting to "inherit" a guarded fd will + * result in a error. */ - proc_fdlock(p); - if ((error = fp_lookup(p, fd, &fp, 1)) == 0) { - *fdflags(p, fd) &= ~UF_EXCLOSE; - (void) fp_drop(p, fd, fp, 1); + fcntla.fd = psfa->psfaa_filedes; + fcntla.cmd = F_GETFD; + if ((error = fcntl_nocancel(p, &fcntla, ival)) != 0) + break; + + if ((ival[0] & FD_CLOEXEC) == FD_CLOEXEC) { + fcntla.fd = psfa->psfaa_filedes; + fcntla.cmd = F_SETFD; + fcntla.arg = ival[0] & ~FD_CLOEXEC; + error = fcntl_nocancel(p, &fcntla, ival); } - proc_fdunlock(p); + } break; @@ -1657,8 +1734,15 @@ exec_handle_file_actions(struct image_params *imgp, short psa_flags) /* All file actions failures are considered fatal, per POSIX */ - if (error) + if (error) { + if (PSFA_OPEN == psfa->psfaa_type) { + DTRACE_PROC1(spawn__open__failure, uintptr_t, + psfa->psfaa_openargs.psfao_path); + } else { + DTRACE_PROC1(spawn__fd__failure, int, psfa->psfaa_filedes); + } break; + } } if (error != 0 || (psa_flags & POSIX_SPAWN_CLOEXEC_DEFAULT) == 0) @@ -1696,6 +1780,102 @@ exec_handle_file_actions(struct image_params *imgp, short psa_flags) return (0); } +#if CONFIG_MACF +/* + * exec_spawnattr_getmacpolicyinfo + */ +void * +exec_spawnattr_getmacpolicyinfo(const void *macextensions, const char *policyname, size_t *lenp) +{ + const struct _posix_spawn_mac_policy_extensions *psmx = macextensions; + int i; + + if (psmx == NULL) + return NULL; + + for (i = 0; i < psmx->psmx_count; i++) { + const _ps_mac_policy_extension_t *extension = &psmx->psmx_extensions[i]; + if (strncmp(extension->policyname, policyname, sizeof(extension->policyname)) == 0) { + if (lenp != NULL) + *lenp = extension->datalen; + return extension->datap; + } + } + + if (lenp != NULL) + *lenp = 0; + return NULL; +} + +static int +spawn_copyin_macpolicyinfo(const struct user__posix_spawn_args_desc *px_args, _posix_spawn_mac_policy_extensions_t *psmxp) +{ + _posix_spawn_mac_policy_extensions_t psmx = NULL; + int error = 0; + int copycnt = 0; + int i = 0; + + *psmxp = NULL; + + if (px_args->mac_extensions_size < PS_MAC_EXTENSIONS_SIZE(1) || + px_args->mac_extensions_size > PAGE_SIZE) { + error = EINVAL; + goto bad; + } + + MALLOC(psmx, _posix_spawn_mac_policy_extensions_t, px_args->mac_extensions_size, M_TEMP, M_WAITOK); + if ((error = copyin(px_args->mac_extensions, psmx, px_args->mac_extensions_size)) != 0) + goto bad; + + if (PS_MAC_EXTENSIONS_SIZE(psmx->psmx_count) > px_args->mac_extensions_size) { + error = EINVAL; + goto bad; + } + + for (i = 0; i < psmx->psmx_count; i++) { + _ps_mac_policy_extension_t *extension = &psmx->psmx_extensions[i]; + if (extension->datalen == 0 || extension->datalen > PAGE_SIZE) { + error = EINVAL; + goto bad; + } + } + + for (copycnt = 0; copycnt < psmx->psmx_count; copycnt++) { + _ps_mac_policy_extension_t *extension = &psmx->psmx_extensions[copycnt]; + void *data = NULL; + + MALLOC(data, void *, extension->datalen, M_TEMP, M_WAITOK); + if ((error = copyin(extension->data, data, extension->datalen)) != 0) { + FREE(data, M_TEMP); + goto bad; + } + extension->datap = data; + } + + *psmxp = psmx; + return 0; + +bad: + if (psmx != NULL) { + for (i = 0; i < copycnt; i++) + FREE(psmx->psmx_extensions[i].datap, M_TEMP); + FREE(psmx, M_TEMP); + } + return error; +} + +static void +spawn_free_macpolicyinfo(_posix_spawn_mac_policy_extensions_t psmx) +{ + int i; + + if (psmx == NULL) + return; + for (i = 0; i < psmx->psmx_count; i++) + FREE(psmx->psmx_extensions[i].datap, M_TEMP); + FREE(psmx, M_TEMP); +} +#endif /* CONFIG_MACF */ /* * posix_spawn @@ -1748,6 +1928,9 @@ posix_spawn(proc_t ap, struct posix_spawn_args *uap, int32_t *retval) boolean_t spawn_no_exec = FALSE; boolean_t proc_transit_set = TRUE; boolean_t exec_done = FALSE; + int portwatch_count = 0; + ipc_port_t * portwatch_ports = NULL; + vm_size_t px_sa_offset = offsetof(struct _posix_spawnattr, psa_ports); /* * Allocate a big chunk for locals instead of using stack since these @@ -1772,6 +1955,7 @@ posix_spawn(proc_t ap, struct posix_spawn_args *uap, int32_t *retval) imgp->ip_flags = (is_64 ? IMGPF_WAS_64BIT : IMGPF_NONE); imgp->ip_p_comm = alt_p_comm; /* for PowerPC */ imgp->ip_seg = (is_64 ? UIO_USERSPACE64 : UIO_USERSPACE32); + imgp->ip_mac_return = 0; if (uap->adesc != USER_ADDR_NULL) { if(is_64) { @@ -1791,17 +1975,23 @@ posix_spawn(proc_t ap, struct posix_spawn_args *uap, int32_t *retval) px_args.file_actions = CAST_USER_ADDR_T(px_args32.file_actions); px_args.port_actions_size = px_args32.port_actions_size; px_args.port_actions = CAST_USER_ADDR_T(px_args32.port_actions); + px_args.mac_extensions_size = px_args32.mac_extensions_size; + px_args.mac_extensions = CAST_USER_ADDR_T(px_args32.mac_extensions); } if (error) goto bad; if (px_args.attr_size != 0) { /* - * This could lose some of the port_actions pointer, - * but we already have it from px_args. + * We are not copying the port_actions pointer, + * because we already have it from px_args. + * This is a bit fragile: */ - if ((error = copyin(px_args.attrp, &px_sa, sizeof(px_sa))) != 0) + + if ((error = copyin(px_args.attrp, &px_sa, px_sa_offset) != 0)) goto bad; + + bzero( (void *)( (unsigned long) &px_sa + px_sa_offset), sizeof(px_sa) - px_sa_offset ); imgp->ip_px_sa = &px_sa; } @@ -1823,6 +2013,12 @@ posix_spawn(proc_t ap, struct posix_spawn_args *uap, int32_t *retval) if ((error = copyin(px_args.file_actions, px_sfap, px_args.file_actions_size)) != 0) goto bad; + + /* Verify that the action count matches the struct size */ + if (PSF_ACTIONS_SIZE(px_sfap->psfa_act_count) != px_args.file_actions_size) { + error = EINVAL; + goto bad; + } } if (px_args.port_actions_size != 0) { /* Limit port_actions to one page of data */ @@ -1843,7 +2039,19 @@ posix_spawn(proc_t ap, struct posix_spawn_args *uap, int32_t *retval) if ((error = copyin(px_args.port_actions, px_spap, px_args.port_actions_size)) != 0) goto bad; + + /* Verify that the action count matches the struct size */ + if (PS_PORT_ACTIONS_SIZE(px_spap->pspa_count) != px_args.port_actions_size) { + error = EINVAL; + goto bad; + } + } +#if CONFIG_MACF + if (px_args.mac_extensions_size != 0) { + if ((error = spawn_copyin_macpolicyinfo(&px_args, (_posix_spawn_mac_policy_extensions_t *)&imgp->ip_px_smpx)) != 0) + goto bad; } +#endif /* CONFIG_MACF */ } /* set uthread to parent */ @@ -1867,14 +2075,63 @@ posix_spawn(proc_t ap, struct posix_spawn_args *uap, int32_t *retval) * and execve(). */ if (imgp->ip_px_sa == NULL || !(px_sa.psa_flags & POSIX_SPAWN_SETEXEC)){ - if ((error = fork1(p, &imgp->ip_new_thread, PROC_CREATE_SPAWN)) != 0) + + /* + * Set the new task's coalition, if it is requested. + * TODO: privilege check - 15365900 + */ + coalition_t coal = COALITION_NULL; +#if CONFIG_COALITIONS + if (imgp->ip_px_sa) { + uint64_t cid = px_sa.psa_coalitionid; + if (cid != 0) { +#if COALITION_DEBUG + printf("%s: searching for coalition ID %llu\n", __func__, cid); +#endif + coal = coalition_find_and_activate_by_id(cid); + if (coal == COALITION_NULL) { +#if COALITION_DEBUG + printf("%s: could not find coalition ID %llu (perhaps it has been terminated or reaped)\n", __func__, cid); +#endif + error = ESRCH; + goto bad; + } + } + } +#endif /* CONFIG_COALITIONS */ + + error = fork1(p, &imgp->ip_new_thread, PROC_CREATE_SPAWN, coal); + + if (error != 0) { + if (coal != COALITION_NULL) { +#if CONFIG_COALITIONS + coalition_remove_active(coal); + coalition_release(coal); +#endif /* CONFIG_COALITIONS */ + } goto bad; + } imgp->ip_flags |= IMGPF_SPAWN; /* spawn w/o exec */ spawn_no_exec = TRUE; /* used in later tests */ + + if (coal != COALITION_NULL) { +#if CONFIG_COALITIONS + coalition_remove_active(coal); + coalition_release(coal); +#endif /* CONFIG_COALITIONS */ + } } - if (spawn_no_exec) + if (spawn_no_exec) { p = (proc_t)get_bsdthreadtask_info(imgp->ip_new_thread); + + /* + * We had to wait until this point before firing the + * proc:::create probe, otherwise p would not point to the + * child process. + */ + DTRACE_PROC1(create, proc_t, p); + } assert(p != NULL); /* By default, the thread everyone plays with is the parent */ @@ -1907,14 +2164,35 @@ posix_spawn(proc_t ap, struct posix_spawn_args *uap, int32_t *retval) } /* Has spawn port actions? */ - if (imgp->ip_px_spa != NULL) { - /* - * The check for the POSIX_SPAWN_SETEXEC flag is done in - * exec_handle_port_actions(). + if (imgp->ip_px_spa != NULL) { + boolean_t is_adaptive = FALSE; + boolean_t portwatch_present = FALSE; + + /* Will this process become adaptive? The apptype isn't ready yet, so we can't look there. */ + if (imgp->ip_px_sa != NULL && px_sa.psa_apptype == POSIX_SPAWN_PROC_TYPE_DAEMON_ADAPTIVE) + is_adaptive = TRUE; + + /* + * portwatch only: + * Allocate a place to store the ports we want to bind to the new task + * We can't bind them until after the apptype is set. */ + if (px_spap->pspa_count != 0 && is_adaptive) { + portwatch_count = px_spap->pspa_count; + MALLOC(portwatch_ports, ipc_port_t *, (sizeof(ipc_port_t) * portwatch_count), M_TEMP, M_WAITOK | M_ZERO); + } else { + portwatch_ports = NULL; + } + if ((error = exec_handle_port_actions(imgp, - imgp->ip_px_sa != NULL ? px_sa.psa_flags : 0)) != 0) + imgp->ip_px_sa != NULL ? px_sa.psa_flags : 0, &portwatch_present, portwatch_ports)) != 0) goto bad; + + if (portwatch_present == FALSE && portwatch_ports != NULL) { + FREE(portwatch_ports, M_TEMP); + portwatch_ports = NULL; + portwatch_count = 0; + } } /* Has spawn attr? */ @@ -1961,6 +2239,11 @@ posix_spawn(proc_t ap, struct posix_spawn_args *uap, int32_t *retval) /* * Disable ASLR for the spawned process. */ + /* + * But only do so if we are not embedded; embedded allows for a + * boot-arg (-disable_aslr) to deal with this (which itself is + * only honored on DEVELOPMENT or DEBUG builds of xnu). + */ if (px_sa.psa_flags & _POSIX_SPAWN_DISABLE_ASLR) OSBitOrAtomic(P_DISABLE_ASLR, &p->p_flag); @@ -2003,7 +2286,7 @@ posix_spawn(proc_t ap, struct posix_spawn_args *uap, int32_t *retval) * Activate the image */ error = exec_activate_image(imgp); - + if (error == 0) { /* process completed the exec */ exec_done = TRUE; @@ -2059,10 +2342,38 @@ posix_spawn(proc_t ap, struct posix_spawn_args *uap, int32_t *retval) error = setsigvec(p, child_thread, sig + 1, &vec, spawn_no_exec); } } + + /* + * Activate the CPU usage monitor, if requested. This is done via a task-wide, per-thread CPU + * usage limit, which will generate a resource exceeded exception if any one thread exceeds the + * limit. + * + * Userland gives us interval in seconds, and the kernel SPI expects nanoseconds. + */ + if (px_sa.psa_cpumonitor_percent != 0) { + /* + * Always treat a CPU monitor activation coming from spawn as entitled. Requiring + * an entitlement to configure the monitor a certain way seems silly, since + * whomever is turning it on could just as easily choose not to do so. + * + * XXX - Ignore the parameters that we get from userland. The spawnattr method of + * activating the monitor always gets the system default parameters. Once we have + * an explicit spawn SPI for configuring the defaults, we can revert this to + * respect the params passed in from userland. + */ + error = proc_set_task_ruse_cpu(p->task, + TASK_POLICY_RESOURCE_ATTRIBUTE_NOTIFY_EXC, + PROC_POLICY_CPUMON_DEFAULTS, 0, + 0, TRUE); + } } bad: + if (error == 0) { + /* reset delay idle sleep status if set */ + if ((p->p_flag & P_DELAYIDLESLEEP) == P_DELAYIDLESLEEP) + OSBitAndAtomic(~((uint32_t)P_DELAYIDLESLEEP), &p->p_flag); /* upon successful spawn, re/set the proc control state */ if (imgp->ip_px_sa != NULL) { switch (px_sa.psa_pcontrol) { @@ -2082,6 +2393,15 @@ bad: }; } exec_resettextvp(p, imgp); + +#if CONFIG_MEMORYSTATUS && CONFIG_JETSAM + /* Has jetsam attributes? */ + if (imgp->ip_px_sa != NULL && (px_sa.psa_jetsam_flags & POSIX_SPAWN_JETSAM_SET)) { + memorystatus_update(p, px_sa.psa_priority, 0, (px_sa.psa_jetsam_flags & POSIX_SPAWN_JETSAM_USE_EFFECTIVE_PRIORITY), + TRUE, px_sa.psa_high_water_mark, (px_sa.psa_jetsam_flags & POSIX_SPAWN_JETSAM_HIWATER_BACKGROUND), + (px_sa.psa_jetsam_flags & POSIX_SPAWN_JETSAM_MEMLIMIT_FATAL)); + } +#endif } /* @@ -2106,8 +2426,54 @@ bad: /* flag the 'fork' has occurred */ proc_knote(p->p_pptr, NOTE_FORK | p->p_pid); /* then flag exec has occurred */ - proc_knote(p, NOTE_EXEC); - DTRACE_PROC1(create, proc_t, p); + /* notify only if it has not failed due to FP Key error */ + if ((p->p_lflag & P_LTERM_DECRYPTFAIL) == 0) + proc_knote(p, NOTE_EXEC); + } else { + /* reset the importance attribute from our previous life */ + task_importance_reset(p->task); + + /* reset atm context from task */ + task_atm_reset(p->task); + } + + /* + * Apply the spawnattr policy, apptype (which primes the task for importance donation), + * and bind any portwatch ports to the new task. + * This must be done after the exec so that the child's thread is ready, + * and after the in transit state has been released, because priority is + * dropped here so we need to be prepared for a potentially long preemption interval + * + * TODO: Consider splitting this up into separate phases + */ + if (error == 0 && imgp->ip_px_sa != NULL) { + struct _posix_spawnattr *psa = (struct _posix_spawnattr *) imgp->ip_px_sa; + + exec_handle_spawnattr_policy(p, psa->psa_apptype, psa->psa_qos_clamp, + portwatch_ports, portwatch_count); + } + + /* Apply the main thread qos */ + if (error == 0) { + thread_t main_thread = (imgp->ip_new_thread != NULL) ? imgp->ip_new_thread : current_thread(); + + task_set_main_thread_qos(p->task, main_thread); + } + + /* + * Release any ports we kept around for binding to the new task + * We need to release the rights even if the posix_spawn has failed. + */ + if (portwatch_ports != NULL) { + for (int i = 0; i < portwatch_count; i++) { + ipc_port_t port = NULL; + if ((port = portwatch_ports[i]) != NULL) { + ipc_port_release_send(port); + } + } + FREE(portwatch_ports, M_TEMP); + portwatch_ports = NULL; + portwatch_count = 0; } /* @@ -2132,6 +2498,8 @@ bad: if (imgp != NULL) { if (imgp->ip_vp) vnode_put(imgp->ip_vp); + if (imgp->ip_scriptvp) + vnode_put(imgp->ip_scriptvp); if (imgp->ip_strings) execargs_free(imgp); if (imgp->ip_px_sfa != NULL) @@ -2140,6 +2508,8 @@ bad: FREE(imgp->ip_px_spa, M_TEMP); #if CONFIG_MACF + if (imgp->ip_px_smpx != NULL) + spawn_free_macpolicyinfo(imgp->ip_px_smpx); if (imgp->ip_execlabelp) mac_cred_label_free(imgp->ip_execlabelp); if (imgp->ip_scriptlabelp) @@ -2147,34 +2517,48 @@ bad: #endif } - if (error) { - DTRACE_PROC1(exec__failure, int, error); +#if CONFIG_DTRACE + if (spawn_no_exec) { + /* + * In the original DTrace reference implementation, + * posix_spawn() was a libc routine that just + * did vfork(2) then exec(2). Thus the proc::: probes + * are very fork/exec oriented. The details of this + * in-kernel implementation of posix_spawn() is different + * (while producing the same process-observable effects) + * particularly w.r.t. errors, and which thread/process + * is constructing what on behalf of whom. + */ + if (error) { + DTRACE_PROC1(spawn__failure, int, error); + } else { + DTRACE_PROC(spawn__success); + /* + * Some DTrace scripts, e.g. newproc.d in + * /usr/bin, rely on the the 'exec-success' + * probe being fired in the child after the + * new process image has been constructed + * in order to determine the associated pid. + * + * So, even though the parent built the image + * here, for compatibility, mark the new thread + * so 'exec-success' fires on it as it leaves + * the kernel. + */ + dtrace_thread_didexec(imgp->ip_new_thread); + } } else { - /* - * temporary - so dtrace call to current_proc() - * returns the child process instead of the parent. - */ - if (imgp != NULL && imgp->ip_flags & IMGPF_SPAWN) { - p->p_lflag |= P_LINVFORK; - p->p_vforkact = current_thread(); - uthread->uu_proc = p; - uthread->uu_flag |= UT_VFORK; - } - - DTRACE_PROC(exec__success); - - /* - * temporary - so dtrace call to current_proc() - * returns the child process instead of the parent. - */ - if (imgp != NULL && imgp->ip_flags & IMGPF_SPAWN) { - p->p_lflag &= ~P_LINVFORK; - p->p_vforkact = NULL; - uthread->uu_proc = PROC_NULL; - uthread->uu_flag &= ~UT_VFORK; - } + if (error) { + DTRACE_PROC1(exec__failure, int, error); + } else { + DTRACE_PROC(exec__success); + } } + if ((dtrace_proc_waitfor_hook = dtrace_proc_waitfor_exec_ptr) != NULL) + (*dtrace_proc_waitfor_hook)(p); +#endif + /* Return to both the parent and the child? */ if (imgp != NULL && spawn_no_exec) { /* @@ -2209,7 +2593,7 @@ bad: } else { /* - * Return" to the child + * Return to the child * * Note: the image activator earlier dropped the * task/thread references to the newly spawned @@ -2256,6 +2640,8 @@ execve(proc_t p, struct execve_args *uap, int32_t *retval) struct __mac_execve_args muap; int err; + memoryshot(VM_EXECVE, DBG_FUNC_NONE); + muap.fname = uap->fname; muap.argp = uap->argp; muap.envp = uap->envp; @@ -2302,6 +2688,7 @@ __mac_execve(proc_t p, struct __mac_execve_args *uap, int32_t *retval) char alt_p_comm[sizeof(p->p_comm)] = {0}; /* for PowerPC */ int is_64 = IS_64BIT_PROCESS(p); struct vfs_context context; + struct uthread *uthread; context.vc_thread = current_thread(); context.vc_ucred = kauth_cred_proc_ref(p); /* XXX must NOT be kauth_cred_get() */ @@ -2328,6 +2715,12 @@ __mac_execve(proc_t p, struct __mac_execve_args *uap, int32_t *retval) imgp->ip_flags = (is_64 ? IMGPF_WAS_64BIT : IMGPF_NONE) | ((p->p_flag & P_DISABLE_ASLR) ? IMGPF_DISABLE_ASLR : IMGPF_NONE); imgp->ip_p_comm = alt_p_comm; /* for PowerPC */ imgp->ip_seg = (is_64 ? UIO_USERSPACE64 : UIO_USERSPACE32); + imgp->ip_mac_return = 0; + + uthread = get_bsdthread_info(current_thread()); + if (uthread->uu_flag & UT_VFORK) { + imgp->ip_flags |= IMGPF_VFORK_EXEC; + } #if CONFIG_MACF if (uap->mac_p != USER_ADDR_NULL) { @@ -2353,6 +2746,8 @@ __mac_execve(proc_t p, struct __mac_execve_args *uap, int32_t *retval) } if (imgp->ip_vp != NULLVP) vnode_put(imgp->ip_vp); + if (imgp->ip_scriptvp != NULLVP) + vnode_put(imgp->ip_scriptvp); if (imgp->ip_strings) execargs_free(imgp); #if CONFIG_MACF @@ -2362,14 +2757,27 @@ __mac_execve(proc_t p, struct __mac_execve_args *uap, int32_t *retval) mac_vnode_label_free(imgp->ip_scriptlabelp); #endif if (!error) { - struct uthread *uthread; - /* Sever any extant thread affinity */ thread_affinity_exec(current_thread()); + thread_t main_thread = (imgp->ip_new_thread != NULL) ? imgp->ip_new_thread : current_thread(); + + task_set_main_thread_qos(p->task, main_thread); + + /* reset task importance */ + task_importance_reset(p->task); + + /* reset atm context from task */ + task_atm_reset(p->task); + DTRACE_PROC(exec__success); - uthread = get_bsdthread_info(current_thread()); - if (uthread->uu_flag & UT_VFORK) { + +#if CONFIG_DTRACE + if ((dtrace_proc_waitfor_hook = dtrace_proc_waitfor_exec_ptr) != NULL) + (*dtrace_proc_waitfor_hook)(p); +#endif + + if (imgp->ip_flags & IMGPF_VFORK_EXEC) { vfork_return(p, retval, p->p_pid); (void)thread_resume(imgp->ip_new_thread); } @@ -2922,7 +3330,7 @@ bad: } static char * -random_hex_str(char *str, int len) +random_hex_str(char *str, int len, boolean_t embedNUL) { uint64_t low, high, value; int idx; @@ -2940,6 +3348,14 @@ random_hex_str(char *str, int len) high = random(); value = high << 32 | low; + if (embedNUL) { + /* + * Zero a byte to protect against C string vulnerabilities + * e.g. for userland __stack_chk_guard. + */ + value &= ~(0xffull << 8); + } + str[0] = '0'; str[1] = 'x'; for (idx = 2; idx < len - 1; idx++) { @@ -2969,6 +3385,14 @@ random_hex_str(char *str, int len) #define ENTROPY_VALUES 2 #define ENTROPY_KEY "malloc_entropy=" +/* + * System malloc engages nanozone for UIAPP. + */ +#define NANO_ENGAGE_KEY "MallocNanoZone=1" + +#define PFZ_KEY "pfz=" +extern user32_addr_t commpage_text32_location; +extern user64_addr_t commpage_text64_location; /* * Build up the contents of the apple[] string vector */ @@ -2976,23 +3400,56 @@ static int exec_add_apple_strings(struct image_params *imgp) { int i, error; - int new_ptr_size = (imgp->ip_flags & IMGPF_IS_64BIT) ? 8 : 4; + int new_ptr_size=4; char guard[19]; char guard_vec[strlen(GUARD_KEY) + 19 * GUARD_VALUES + 1]; char entropy[19]; char entropy_vec[strlen(ENTROPY_KEY) + 19 * ENTROPY_VALUES + 1]; + char pfz_string[strlen(PFZ_KEY) + 16 + 4 +1]; + + if( imgp->ip_flags & IMGPF_IS_64BIT) { + new_ptr_size = 8; + snprintf(pfz_string, sizeof(pfz_string),PFZ_KEY "0x%llx",commpage_text64_location); + } else { + snprintf(pfz_string, sizeof(pfz_string),PFZ_KEY "0x%x",commpage_text32_location); + } + /* exec_save_path stored the first string */ imgp->ip_applec = 1; + /* adding the pfz string */ + error = exec_add_user_string(imgp, CAST_USER_ADDR_T(pfz_string),UIO_SYSSPACE,FALSE); + if(error) + goto bad; + imgp->ip_applec++; + + /* adding the NANO_ENGAGE_KEY key */ + if (imgp->ip_px_sa) { + int proc_flags = (((struct _posix_spawnattr *) imgp->ip_px_sa)->psa_flags); + + if ((proc_flags & _POSIX_SPAWN_NANO_ALLOCATOR) == _POSIX_SPAWN_NANO_ALLOCATOR) { + char uiapp_string[strlen(NANO_ENGAGE_KEY) + 1]; + + snprintf(uiapp_string, sizeof(uiapp_string), NANO_ENGAGE_KEY); + error = exec_add_user_string(imgp, CAST_USER_ADDR_T(uiapp_string),UIO_SYSSPACE,FALSE); + if (error) + goto bad; + imgp->ip_applec++; + } + } + /* * Supply libc with a collection of random values to use when * implementing -fstack-protector. + * + * (The first random string always contains an embedded NUL so that + * __stack_chk_guard also protects against C string vulnerabilities) */ (void)strlcpy(guard_vec, GUARD_KEY, sizeof (guard_vec)); for (i = 0; i < GUARD_VALUES; i++) { - random_hex_str(guard, sizeof (guard)); + random_hex_str(guard, sizeof (guard), i == 0); if (i) (void)strlcat(guard_vec, ",", sizeof (guard_vec)); (void)strlcat(guard_vec, guard, sizeof (guard_vec)); @@ -3008,7 +3465,7 @@ exec_add_apple_strings(struct image_params *imgp) */ (void)strlcpy(entropy_vec, ENTROPY_KEY, sizeof(entropy_vec)); for (i = 0; i < ENTROPY_VALUES; i++) { - random_hex_str(entropy, sizeof (entropy)); + random_hex_str(entropy, sizeof (entropy), FALSE); if (i) (void)strlcat(entropy_vec, ",", sizeof (entropy_vec)); (void)strlcat(entropy_vec, entropy, sizeof (entropy_vec)); @@ -3090,6 +3547,13 @@ exec_check_permissions(struct image_params *imgp) if ((vp->v_mount->mnt_flag & MNT_NOSUID) || (p->p_lflag & P_LTRACED)) vap->va_mode &= ~(VSUID | VSGID); + /* + * Disable _POSIX_SPAWN_ALLOW_DATA_EXEC and _POSIX_SPAWN_DISABLE_ASLR + * flags for setuid/setgid binaries. + */ + if (vap->va_mode & (VSUID | VSGID)) + imgp->ip_flags &= ~(IMGPF_ALLOW_DATA_EXEC | IMGPF_DISABLE_ASLR); + #if CONFIG_MACF error = mac_vnode_check_exec(imgp->ip_vfs_context, vp, imgp); if (error) @@ -3116,18 +3580,6 @@ exec_check_permissions(struct image_params *imgp) #endif -#ifdef IMGPF_POWERPC - /* - * If the file we are about to attempt to load is the exec_handler_ppc, - * which is determined by matching the vattr fields against previously - * cached values, then we set the PowerPC environment flag. - */ - if (vap->va_fsid == exec_archhandler_ppc.fsid && - vap->va_fileid == exec_archhandler_ppc.fileid) { - imgp->ip_flags |= IMGPF_POWERPC; - } -#endif /* IMGPF_POWERPC */ - /* XXX May want to indicate to underlying FS that vnode is open */ return (error); @@ -3167,9 +3619,11 @@ exec_handle_sugid(struct image_params *imgp) proc_t p = vfs_context_proc(imgp->ip_vfs_context); int i; int leave_sugid_clear = 0; + int mac_reset_ipc = 0; int error = 0; #if CONFIG_MACF - int mac_transition; + int mac_transition, disjoint_cred = 0; + int label_update_return = 0; /* * Determine whether a call to update the MAC label will result in the @@ -3183,8 +3637,12 @@ exec_handle_sugid(struct image_params *imgp) mac_transition = mac_cred_check_label_update_execve( imgp->ip_vfs_context, imgp->ip_vp, + imgp->ip_arch_offset, + imgp->ip_scriptvp, imgp->ip_scriptlabelp, - imgp->ip_execlabelp, p); + imgp->ip_execlabelp, + p, + imgp->ip_px_smpx); #endif OSBitAndAtomic(~((uint32_t)P_SUGID), &p->p_flag); @@ -3249,12 +3707,19 @@ handle_mac_transition: * modifying any others sharing it. */ if (mac_transition) { - kauth_cred_t my_cred; - if (kauth_proc_label_update_execve(p, + kauth_proc_label_update_execve(p, imgp->ip_vfs_context, imgp->ip_vp, + imgp->ip_arch_offset, + imgp->ip_scriptvp, imgp->ip_scriptlabelp, - imgp->ip_execlabelp)) { + imgp->ip_execlabelp, + &imgp->ip_csflags, + imgp->ip_px_smpx, + &disjoint_cred, /* will be non zero if disjoint */ + &label_update_return); + + if (disjoint_cred) { /* * If updating the MAC label resulted in a * disjoint credential, flag that we need to @@ -3266,22 +3731,13 @@ handle_mac_transition: */ leave_sugid_clear = 0; } - - my_cred = kauth_cred_proc_ref(p); - mac_task_label_update_cred(my_cred, p->task); - kauth_cred_unref(&my_cred); + + imgp->ip_mac_return = label_update_return; } -#endif /* CONFIG_MACF */ + + mac_reset_ipc = mac_proc_check_inherit_ipc_ports(p, p->p_textvp, p->p_textoff, imgp->ip_vp, imgp->ip_arch_offset, imgp->ip_scriptvp); - /* - * Have mach reset the task and thread ports. - * We don't want anyone who had the ports before - * a setuid exec to be able to access/control the - * task/thread after. - */ - ipc_task_reset(p->task); - ipc_thread_reset((imgp->ip_new_thread != NULL) ? - imgp->ip_new_thread : current_thread()); +#endif /* CONFIG_MACF */ /* * If 'leave_sugid_clear' is non-zero, then we passed the @@ -3289,62 +3745,93 @@ handle_mac_transition: * the previous cred was a member of the VSGID group, but * that it was not the default at the time of the execve, * and that the post-labelling credential was not disjoint. - * So we don't set the P_SUGID on the basis of simply - * running this code. - */ - if (!leave_sugid_clear) - OSBitOrAtomic(P_SUGID, &p->p_flag); - - /* - * Radar 2261856; setuid security hole fix - * XXX For setuid processes, attempt to ensure that - * stdin, stdout, and stderr are already allocated. - * We do not want userland to accidentally allocate - * descriptors in this range which has implied meaning - * to libc. + * So we don't set the P_SUGID or reset mach ports and fds + * on the basis of simply running this code. */ - for (i = 0; i < 3; i++) { + if (mac_reset_ipc || !leave_sugid_clear) { + /* + * Have mach reset the task and thread ports. + * We don't want anyone who had the ports before + * a setuid exec to be able to access/control the + * task/thread after. + */ + ipc_task_reset(p->task); + ipc_thread_reset((imgp->ip_new_thread != NULL) ? + imgp->ip_new_thread : current_thread()); + } - if (p->p_fd->fd_ofiles[i] != NULL) - continue; + if (!leave_sugid_clear) { + /* + * Flag the process as setuid. + */ + OSBitOrAtomic(P_SUGID, &p->p_flag); /* - * Do the kernel equivalent of - * - * (void) open("/dev/null", O_RDONLY); + * Radar 2261856; setuid security hole fix + * XXX For setuid processes, attempt to ensure that + * stdin, stdout, and stderr are already allocated. + * We do not want userland to accidentally allocate + * descriptors in this range which has implied meaning + * to libc. */ + for (i = 0; i < 3; i++) { - struct fileproc *fp; - int indx; + if (p->p_fd->fd_ofiles[i] != NULL) + continue; - if ((error = falloc(p, - &fp, &indx, imgp->ip_vfs_context)) != 0) - continue; + /* + * Do the kernel equivalent of + * + * if i == 0 + * (void) open("/dev/null", O_RDONLY); + * else + * (void) open("/dev/null", O_WRONLY); + */ - struct nameidata nd1; + struct fileproc *fp; + int indx; + int flag; + struct nameidata *ndp = NULL; - NDINIT(&nd1, LOOKUP, OP_OPEN, FOLLOW, UIO_SYSSPACE, - CAST_USER_ADDR_T("/dev/null"), - imgp->ip_vfs_context); + if (i == 0) + flag = FREAD; + else + flag = FWRITE; - if ((error = vn_open(&nd1, FREAD, 0)) != 0) { - fp_free(p, indx, fp); - break; - } + if ((error = falloc(p, + &fp, &indx, imgp->ip_vfs_context)) != 0) + continue; + + MALLOC(ndp, struct nameidata *, sizeof(*ndp), M_TEMP, M_WAITOK | M_ZERO); + if (ndp == NULL) { + error = ENOMEM; + break; + } + + NDINIT(ndp, LOOKUP, OP_OPEN, FOLLOW, UIO_SYSSPACE, + CAST_USER_ADDR_T("/dev/null"), + imgp->ip_vfs_context); - struct fileglob *fg = fp->f_fglob; + if ((error = vn_open(ndp, flag, 0)) != 0) { + fp_free(p, indx, fp); + break; + } + + struct fileglob *fg = fp->f_fglob; + + fg->fg_flag = flag; + fg->fg_ops = &vnops; + fg->fg_data = ndp->ni_vp; - fg->fg_flag = FREAD; - fg->fg_type = DTYPE_VNODE; - fg->fg_ops = &vnops; - fg->fg_data = nd1.ni_vp; + vnode_put(ndp->ni_vp); - vnode_put(nd1.ni_vp); + proc_fdlock(p); + procfdtbl_releasefd(p, indx, NULL); + fp_drop(p, indx, fp, 1); + proc_fdunlock(p); - proc_fdlock(p); - procfdtbl_releasefd(p, indx, NULL); - fp_drop(p, indx, fp, 1); - proc_fdunlock(p); + FREE(ndp, M_TEMP); + } } } #if CONFIG_MACF @@ -3361,6 +3848,7 @@ handle_mac_transition: goto handle_mac_transition; } } + #endif /* CONFIG_MACF */ /* @@ -3390,72 +3878,90 @@ handle_mac_transition: * limits on stack growth, if they end up being needed. * * Parameters: p Process to set stack on - * user_stack Address to set stack for process to - * customstack FALSE if no custom stack in binary - * map Address map in which to allocate the - * new stack, if 'customstack' is FALSE + * load_result Information from mach-o load commands + * map Address map in which to allocate the new stack * * Returns: KERN_SUCCESS Stack successfully created * !KERN_SUCCESS Mach failure code */ static kern_return_t -create_unix_stack(vm_map_t map, user_addr_t user_stack, int customstack, +create_unix_stack(vm_map_t map, load_result_t* load_result, proc_t p) { mach_vm_size_t size, prot_size; mach_vm_offset_t addr, prot_addr; kern_return_t kr; + mach_vm_address_t user_stack = load_result->user_stack; + proc_lock(p); p->user_stack = user_stack; proc_unlock(p); - if (!customstack) { + if (!load_result->prog_allocated_stack) { /* * Allocate enough space for the maximum stack size we * will ever authorize and an extra page to act as - * a guard page for stack overflows. + * a guard page for stack overflows. For default stacks, + * vm_initial_limit_stack takes care of the extra guard page. + * Otherwise we must allocate it ourselves. */ - size = mach_vm_round_page(MAXSSIZ); -#if STACK_GROWTH_UP - addr = mach_vm_trunc_page(user_stack); -#else /* STACK_GROWTH_UP */ - addr = mach_vm_trunc_page(user_stack - size); -#endif /* STACK_GROWTH_UP */ + + size = mach_vm_round_page(load_result->user_stack_size); + if (load_result->prog_stack_size) + size += PAGE_SIZE; + addr = mach_vm_trunc_page(load_result->user_stack - size); kr = mach_vm_allocate(map, &addr, size, VM_MAKE_TAG(VM_MEMORY_STACK) | - VM_FLAGS_FIXED); + VM_FLAGS_FIXED); if (kr != KERN_SUCCESS) { - return kr; + /* If can't allocate at default location, try anywhere */ + addr = 0; + kr = mach_vm_allocate(map, &addr, size, + VM_MAKE_TAG(VM_MEMORY_STACK) | + VM_FLAGS_ANYWHERE); + if (kr != KERN_SUCCESS) + return kr; + + user_stack = addr + size; + load_result->user_stack = user_stack; + + proc_lock(p); + p->user_stack = user_stack; + proc_unlock(p); } + /* * And prevent access to what's above the current stack * size limit for this process. */ prot_addr = addr; -#if STACK_GROWTH_UP - prot_addr += unix_stack_size(p); -#endif /* STACK_GROWTH_UP */ - prot_addr = mach_vm_round_page(prot_addr); - prot_size = mach_vm_trunc_page(size - unix_stack_size(p)); + if (load_result->prog_stack_size) + prot_size = PAGE_SIZE; + else + prot_size = mach_vm_trunc_page(size - unix_stack_size(p)); kr = mach_vm_protect(map, - prot_addr, - prot_size, - FALSE, - VM_PROT_NONE); + prot_addr, + prot_size, + FALSE, + VM_PROT_NONE); if (kr != KERN_SUCCESS) { (void) mach_vm_deallocate(map, addr, size); return kr; } } + return KERN_SUCCESS; } #include -static char init_program_name[128] = "/sbin/launchd"; - -struct execve_args init_exec_args; +static const char * init_programs[] = { +#if DEVELOPMENT || DEBUG + "/usr/local/sbin/launchd.development", +#endif + "/sbin/launchd", +}; /* * load_init_program @@ -3475,75 +3981,77 @@ struct execve_args init_exec_args; void load_init_program(proc_t p) { - vm_offset_t init_addr; - int argc = 0; + vm_offset_t init_addr, addr; + int argc; uint32_t argv[3]; + unsigned int i; int error; int retval[2]; - - /* - * Copy out program name. - */ + const char *init_program_name; + struct execve_args init_exec_args; init_addr = VM_MIN_ADDRESS; - (void) vm_allocate(current_map(), &init_addr, PAGE_SIZE, - VM_FLAGS_ANYWHERE); + (void) vm_allocate(current_map(), &init_addr, PAGE_SIZE, VM_FLAGS_ANYWHERE); if (init_addr == 0) init_addr++; + + for (i = 0; i < sizeof(init_programs)/sizeof(init_programs[0]); i++) { + + init_program_name = init_programs[i]; + addr = init_addr; + argc = 0; - (void) copyout((caddr_t) init_program_name, CAST_USER_ADDR_T(init_addr), - (unsigned) sizeof(init_program_name)+1); - - argv[argc++] = (uint32_t)init_addr; - init_addr += sizeof(init_program_name); - init_addr = (vm_offset_t)ROUND_PTR(char, init_addr); + /* + * Copy out program name. + */ + (void) copyout(init_program_name, CAST_USER_ADDR_T(addr), strlen(init_program_name)+1); - /* - * Put out first (and only) argument, similarly. - * Assumes everything fits in a page as allocated - * above. - */ - if (boothowto & RB_SINGLE) { - const char *init_args = "-s"; + argv[argc++] = (uint32_t)addr; + addr += strlen(init_program_name)+1; + addr = (vm_offset_t)ROUND_PTR(char, addr); - copyout(init_args, CAST_USER_ADDR_T(init_addr), - strlen(init_args)); + /* + * Put out first (and only) argument, similarly. + * Assumes everything fits in a page as allocated above. + */ + if (boothowto & RB_SINGLE) { + const char *init_args = "-s"; - argv[argc++] = (uint32_t)init_addr; - init_addr += strlen(init_args); - init_addr = (vm_offset_t)ROUND_PTR(char, init_addr); + copyout(init_args, CAST_USER_ADDR_T(addr), strlen(init_args)+1); - } + argv[argc++] = (uint32_t)addr; + addr += strlen(init_args)+1; + addr = (vm_offset_t)ROUND_PTR(char, addr); + } - /* - * Null-end the argument list - */ - argv[argc] = 0; - - /* - * Copy out the argument list. - */ + /* + * Null-end the argument list + */ + argv[argc] = 0; - (void) copyout((caddr_t) argv, CAST_USER_ADDR_T(init_addr), - (unsigned) sizeof(argv)); - - /* - * Set up argument block for fake call to execve. - */ + /* + * Copy out the argument list. + */ + (void) copyout(argv, CAST_USER_ADDR_T(addr), sizeof(argv)); - init_exec_args.fname = CAST_USER_ADDR_T(argv[0]); - init_exec_args.argp = CAST_USER_ADDR_T((char **)init_addr); - init_exec_args.envp = CAST_USER_ADDR_T(0); + /* + * Set up argument block for fake call to execve. + */ + init_exec_args.fname = CAST_USER_ADDR_T(argv[0]); + init_exec_args.argp = CAST_USER_ADDR_T((char **)addr); + init_exec_args.envp = CAST_USER_ADDR_T(0); - /* - * So that mach_init task is set with uid,gid 0 token - */ - set_security_token(p); + /* + * So that init task is set with uid,gid 0 token + */ + set_security_token(p); - error = execve(p,&init_exec_args,retval); - if (error) - panic("Process 1 exec of %s failed, errno %d", - init_program_name, error); + error = execve(p, &init_exec_args, retval); + if (!error) + return; + } + + panic("Process 1 exec of %s failed, errno %d", init_program_name, error); } /* @@ -3587,6 +4095,7 @@ load_return_to_errno(load_return_t lrtn) case LOAD_IOERROR: return EIO; case LOAD_FAILURE: + case LOAD_DECRYPTFAIL: default: return EBADEXEC; } @@ -3646,9 +4155,9 @@ execargs_lock_unlock(void) { lck_mtx_unlock(execargs_cache_lock); } -static void +static wait_result_t execargs_lock_sleep(void) { - lck_mtx_sleep(execargs_cache_lock, LCK_SLEEP_DEFAULT, &execargs_free_count, THREAD_UNINT); + return(lck_mtx_sleep(execargs_cache_lock, LCK_SLEEP_DEFAULT, &execargs_free_count, THREAD_INTERRUPTIBLE)); } static kern_return_t @@ -3687,14 +4196,19 @@ static int execargs_alloc(struct image_params *imgp) { kern_return_t kret; + wait_result_t res; int i, cache_index = -1; execargs_lock_lock(); while (execargs_free_count == 0) { execargs_waiters++; - execargs_lock_sleep(); + res = execargs_lock_sleep(); execargs_waiters--; + if (res != THREAD_AWAKENED) { + execargs_lock_unlock(); + return (EINTR); + } } execargs_free_count--; @@ -3814,15 +4328,73 @@ exec_resettextvp(proc_t p, struct image_params *imgp) } -static int +/* + * If the process is not signed or if it contains entitlements, we + * need to communicate through the task_access_port to taskgated. + * + * taskgated will provide a detached code signature if present, and + * will enforce any restrictions on entitlements. + */ + +static boolean_t +taskgated_required(proc_t p, boolean_t *require_success) +{ + size_t length; + void *blob; + int error; + + if ((p->p_csflags & CS_VALID) == 0) { + *require_success = FALSE; + return TRUE; + } + + error = cs_entitlements_blob_get(p, &blob, &length); + if (error == 0 && blob != NULL) { + /* + * fatal on the desktop when entitlements are present, + * unless we started in single-user mode + */ + if ((boothowto & RB_SINGLE) == 0) + *require_success = TRUE; + /* + * Allow initproc to run without causing taskgated to launch + */ + if (p == initproc) { + *require_success = FALSE; + return FALSE; + } + + return TRUE; + } + + *require_success = FALSE; + return 0; +} + +/* + * __EXEC_WAITING_ON_TASKGATED_CODE_SIGNATURE_UPCALL__ + * + * Description: Waits for the userspace daemon to respond to the request + * we made. Function declared non inline to be visible in + * stackshots and spindumps as well as debugging. + */ +__attribute__((noinline)) int +__EXEC_WAITING_ON_TASKGATED_CODE_SIGNATURE_UPCALL__(mach_port_t task_access_port, int32_t new_pid) +{ + return find_code_signature(task_access_port, new_pid); +} + +static int check_for_signature(proc_t p, struct image_params *imgp) { - void *blob = NULL; - size_t length = 0; mach_port_t port = NULL; kern_return_t kr = KERN_FAILURE; int error = EACCES; + boolean_t unexpected_failure = FALSE; unsigned char hash[SHA1_RESULTLEN]; + boolean_t require_success = FALSE; + int spawn = (imgp->ip_flags & IMGPF_SPAWN); + int vfexec = (imgp->ip_flags & IMGPF_VFORK_EXEC); /* * Override inherited code signing flags with the @@ -3837,19 +4409,20 @@ check_for_signature(proc_t p, struct image_params *imgp) if(p->p_csflags & (CS_HARD|CS_KILL)) { vm_map_switch_protect(get_task_map(p->task), TRUE); } - - /* If the process is not signed or if it contains - * entitlements, we need to communicate through the - * task_access_port to taskgated. taskgated will provide a - * detached code signature if present, and will enforce any - * restrictions on entitlements. taskgated returns - * KERN_SUCCESS if it has completed its work and the exec - * should continue, or KERN_FAILURE if the exec should fail. + + /* + * image activation may be failed due to policy + * which is unexpected but security framework does not + * approve of exec, kill and return immediately. */ - error = cs_entitlements_blob_get(p, &blob, &length); + if (imgp->ip_mac_return != 0) { + error = imgp->ip_mac_return; + unexpected_failure = TRUE; + goto done; + } - /* if signed and no entitlements, then we're done here */ - if ((p->p_csflags & CS_VALID) && NULL == blob) { + /* check if callout to taskgated is needed */ + if (!taskgated_required(p, &require_success)) { error = 0; goto done; } @@ -3857,17 +4430,30 @@ check_for_signature(proc_t p, struct image_params *imgp) kr = task_get_task_access_port(p->task, &port); if (KERN_SUCCESS != kr || !IPC_PORT_VALID(port)) { error = 0; -#if !CONFIG_EMBEDDED - /* fatal on the desktop when entitlements are present */ - if (NULL != blob) + if (require_success) error = EACCES; -#endif goto done; } - kr = find_code_signature(port, p->p_pid); - if (KERN_SUCCESS != kr) { + /* + * taskgated returns KERN_SUCCESS if it has completed its work + * and the exec should continue, KERN_FAILURE if the exec should + * fail, or it may error out with different error code in an + * event of mig failure (e.g. process was signalled during the + * rpc call, taskgated died, mig server died etc.). + */ + + kr = __EXEC_WAITING_ON_TASKGATED_CODE_SIGNATURE_UPCALL__(port, p->p_pid); + switch (kr) { + case KERN_SUCCESS: + error = 0; + break; + case KERN_FAILURE: + error = EACCES; + goto done; + default: error = EACCES; + unexpected_failure = TRUE; goto done; } @@ -3885,9 +4471,152 @@ check_for_signature(proc_t p, struct image_params *imgp) } done: - if (0 != error) + if (0 != error) { + if (!unexpected_failure) + p->p_csflags |= CS_KILLED; /* make very sure execution fails */ - psignal(p, SIGKILL); + if (vfexec || spawn) { + psignal_vfork(p, p->task, imgp->ip_new_thread, SIGKILL); + error = 0; + } else { + psignal(p, SIGKILL); + } + } return error; } +/* + * Typically as soon as we start executing this process, the + * first instruction will trigger a VM fault to bring the text + * pages (as executable) into the address space, followed soon + * thereafter by dyld data structures (for dynamic executable). + * To optimize this, as well as improve support for hardware + * debuggers that can only access resident pages present + * in the process' page tables, we prefault some pages if + * possible. Errors are non-fatal. + */ +static void exec_prefault_data(proc_t p __unused, struct image_params *imgp, load_result_t *load_result) +{ + int ret; + size_t expected_all_image_infos_size; + + /* + * Prefault executable or dyld entry point. + */ + vm_fault(current_map(), + vm_map_trunc_page(load_result->entry_point, + vm_map_page_mask(current_map())), + VM_PROT_READ | VM_PROT_EXECUTE, + FALSE, + THREAD_UNINT, NULL, 0); + + if (imgp->ip_flags & IMGPF_IS_64BIT) { + expected_all_image_infos_size = sizeof(struct user64_dyld_all_image_infos); + } else { + expected_all_image_infos_size = sizeof(struct user32_dyld_all_image_infos); + } + + /* Decode dyld anchor structure from */ + if (load_result->dynlinker && + load_result->all_image_info_addr && + load_result->all_image_info_size >= expected_all_image_infos_size) { + union { + struct user64_dyld_all_image_infos infos64; + struct user32_dyld_all_image_infos infos32; + } all_image_infos; + + /* + * Pre-fault to avoid copyin() going through the trap handler + * and recovery path. + */ + vm_fault(current_map(), + vm_map_trunc_page(load_result->all_image_info_addr, + vm_map_page_mask(current_map())), + VM_PROT_READ | VM_PROT_WRITE, + FALSE, + THREAD_UNINT, NULL, 0); + if ((load_result->all_image_info_addr & PAGE_MASK) + expected_all_image_infos_size > PAGE_SIZE) { + /* all_image_infos straddles a page */ + vm_fault(current_map(), + vm_map_trunc_page(load_result->all_image_info_addr + expected_all_image_infos_size - 1, + vm_map_page_mask(current_map())), + VM_PROT_READ | VM_PROT_WRITE, + FALSE, + THREAD_UNINT, NULL, 0); + } + + ret = copyin(load_result->all_image_info_addr, + &all_image_infos, + expected_all_image_infos_size); + if (ret == 0 && all_image_infos.infos32.version >= 9) { + + user_addr_t notification_address; + user_addr_t dyld_image_address; + user_addr_t dyld_version_address; + user_addr_t dyld_all_image_infos_address; + user_addr_t dyld_slide_amount; + + if (imgp->ip_flags & IMGPF_IS_64BIT) { + notification_address = all_image_infos.infos64.notification; + dyld_image_address = all_image_infos.infos64.dyldImageLoadAddress; + dyld_version_address = all_image_infos.infos64.dyldVersion; + dyld_all_image_infos_address = all_image_infos.infos64.dyldAllImageInfosAddress; + } else { + notification_address = all_image_infos.infos32.notification; + dyld_image_address = all_image_infos.infos32.dyldImageLoadAddress; + dyld_version_address = all_image_infos.infos32.dyldVersion; + dyld_all_image_infos_address = all_image_infos.infos32.dyldAllImageInfosAddress; + } + + /* + * dyld statically sets up the all_image_infos in its Mach-O + * binary at static link time, with pointers relative to its default + * load address. Since ASLR might slide dyld before its first + * instruction is executed, "dyld_slide_amount" tells us how far + * dyld was loaded compared to its default expected load address. + * All other pointers into dyld's image should be adjusted by this + * amount. At some point later, dyld will fix up pointers to take + * into account the slide, at which point the all_image_infos_address + * field in the structure will match the runtime load address, and + * "dyld_slide_amount" will be 0, if we were to consult it again. + */ + + dyld_slide_amount = load_result->all_image_info_addr - dyld_all_image_infos_address; + +#if 0 + kprintf("exec_prefault: 0x%016llx 0x%08x 0x%016llx 0x%016llx 0x%016llx 0x%016llx\n", + (uint64_t)load_result->all_image_info_addr, + all_image_infos.infos32.version, + (uint64_t)notification_address, + (uint64_t)dyld_image_address, + (uint64_t)dyld_version_address, + (uint64_t)dyld_all_image_infos_address); +#endif + + vm_fault(current_map(), + vm_map_trunc_page(notification_address + dyld_slide_amount, + vm_map_page_mask(current_map())), + VM_PROT_READ | VM_PROT_EXECUTE, + FALSE, + THREAD_UNINT, NULL, 0); + vm_fault(current_map(), + vm_map_trunc_page(dyld_image_address + dyld_slide_amount, + vm_map_page_mask(current_map())), + VM_PROT_READ | VM_PROT_EXECUTE, + FALSE, + THREAD_UNINT, NULL, 0); + vm_fault(current_map(), + vm_map_trunc_page(dyld_version_address + dyld_slide_amount, + vm_map_page_mask(current_map())), + VM_PROT_READ, + FALSE, + THREAD_UNINT, NULL, 0); + vm_fault(current_map(), + vm_map_trunc_page(dyld_all_image_infos_address + dyld_slide_amount, + vm_map_page_mask(current_map())), + VM_PROT_READ | VM_PROT_WRITE, + FALSE, + THREAD_UNINT, NULL, 0); + } + } +}