X-Git-Url: https://git.saurik.com/apple/xnu.git/blobdiff_plain/4a3eedf9ecc9bbe3f3a5c6ce5e53ad199d639d32..fe8ab488e9161c46dd9885d58fc52996dc0249ff:/bsd/kern/kern_exec.c diff --git a/bsd/kern/kern_exec.c b/bsd/kern/kern_exec.c index 6b2702d7b..3d2710538 100644 --- a/bsd/kern/kern_exec.c +++ b/bsd/kern/kern_exec.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2000-2007 Apple Inc. All rights reserved. + * Copyright (c) 2000-2011 Apple Inc. All rights reserved. * * @APPLE_OSREFERENCE_LICENSE_HEADER_START@ * @@ -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 @@ -109,19 +108,31 @@ #include /* ubc_map() */ #include #include +#include #include +#include + +#include -#include +#include #include #include +#include #include +#include #include #include #include #include +#include /* thread_wakeup() */ +#include +#include +#include +#include + #if CONFIG_MACF #include #include @@ -131,20 +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, int, struct __user_sigaction *); +int setsigvec(proc_t, thread_t, int, struct __kern_sigaction *, boolean_t in_sigstart); +extern void proc_apply_task_networkbg_internal(proc_t, thread_t); /* * Mach things for which prototypes are unavailable from Mach headers @@ -160,13 +191,19 @@ 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 #include #include +#include #include #include #include @@ -175,16 +212,6 @@ extern struct savearea *get_user_regs(thread_t); #include -/* - * SIZE_MAXPTR The maximum size of a user space pointer, in bytes - * SIZE_IMG_STRSPACE The available string space, minus two pointers; we - * define it interms of the maximum, since we don't - * know the pointer size going in, until after we've - * parsed the executable image. - */ -#define SIZE_MAXPTR 8 /* 64 bits */ -#define SIZE_IMG_STRSPACE (NCARGS - 2 * SIZE_MAXPTR) - /* * EAI_ITERLIMIT The maximum number of times to iterate an image * activator in exec_activate_image() before treating @@ -192,11 +219,17 @@ extern struct savearea *get_user_regs(thread_t); */ #define EAI_ITERLIMIT 10 +/* + * For #! interpreter parsing + */ +#define IS_WHITESPACE(ch) ((ch == ' ') || (ch == '\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 *)( ( (unsigned)(addr) + 16 - 1) \ + (type *)( ( (uintptr_t)(addr) + 16 - 1) \ & ~(16 - 1) ) struct image_params; /* Forward */ @@ -207,25 +240,28 @@ static int execargs_alloc(struct image_params *imgp); static int execargs_free(struct image_params *imgp); static int exec_check_permissions(struct image_params *imgp); static int exec_extract_strings(struct image_params *imgp); +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, &sugid_scripts, 0, ""); -static kern_return_t create_unix_stack(vm_map_t map, user_addr_t user_stack, - int customstack, proc_t p); +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, 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 *); - -/* We don't want this one exported */ -__private_extern__ -int open1(vfs_context_t, struct nameidata *, int, struct vnode_attr *, register_t *); +static int check_for_signature(proc_t, struct image_params *); +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_string + * exec_add_user_string * * Add the requested string to the string space area. * * Parameters; struct image_params * image parameter block * user_addr_t string to add to strings area + * int segment from which string comes + * boolean_t TRUE if string contributes to NCARGS * * Returns: 0 Success * !0 Failure errno from copyinstr() @@ -233,29 +269,41 @@ int open1(vfs_context_t, struct nameidata *, int, struct vnode_attr *, register * Implicit returns: * (imgp->ip_strendp) updated location of next add, if any * (imgp->ip_strspace) updated byte count of space remaining + * (imgp->ip_argspace) updated byte count of space in NCARGS */ static int -exec_add_string(struct image_params *imgp, user_addr_t str) +exec_add_user_string(struct image_params *imgp, user_addr_t str, int seg, boolean_t is_ncargs) { - int error = 0; - - do { - size_t len = 0; - if (imgp->ip_strspace <= 0) { + int error = 0; + + do { + size_t len = 0; + int space; + + if (is_ncargs) + space = imgp->ip_argspace; /* by definition smaller than ip_strspace */ + else + space = imgp->ip_strspace; + + if (space <= 0) { error = E2BIG; break; } - if (IS_UIO_SYS_SPACE(imgp->ip_seg)) { + + if (!UIO_SEG_IS_USER_SPACE(seg)) { char *kstr = CAST_DOWN(char *,str); /* SAFE */ - error = copystr(kstr, imgp->ip_strendp, imgp->ip_strspace, &len); + error = copystr(kstr, imgp->ip_strendp, space, &len); } else { - error = copyinstr(str, imgp->ip_strendp, imgp->ip_strspace, - &len); + error = copyinstr(str, imgp->ip_strendp, space, &len); } + imgp->ip_strendp += len; imgp->ip_strspace -= len; + if (is_ncargs) + imgp->ip_argspace -= len; + } while (error == ENAMETOOLONG); - + return error; } @@ -265,11 +313,10 @@ exec_add_string(struct image_params *imgp, user_addr_t str) * To support new app package launching for Mac OS X, the dyld needs the * first argument to execve() stored on the user stack. * - * Save the executable path name at the top of the strings area and set + * Save the executable path name at the bottom of the strings area and set * the argument vector pointer to the location following that to indicate * the start of the argument and environment tuples, setting the remaining - * string space count to the size of the string area minus the path length - * and a reserve for two pointers. + * string space count to the size of the string area minus the path length. * * Parameters; struct image_params * image parameter block * char * path used to invoke program @@ -283,8 +330,9 @@ exec_add_string(struct image_params *imgp, user_addr_t str) * Implicit returns: * (imgp->ip_strings) saved path * (imgp->ip_strspace) space remaining in ip_strings - * (imgp->ip_argv) beginning of argument list * (imgp->ip_strendp) start of remaining copy area + * (imgp->ip_argspace) space remaining of NCARGS + * (imgp->ip_applec) Initial applev[0] * * Note: We have to do this before the initial namei() since in the * path contains symbolic links, namei() will overwrite the @@ -298,10 +346,7 @@ exec_save_path(struct image_params *imgp, user_addr_t path, int seg) { int error; size_t len; - char *kpath = CAST_DOWN(char *,path); /* SAFE */ - - imgp->ip_strendp = imgp->ip_strings; - imgp->ip_strspace = SIZE_IMG_STRSPACE; + char *kpath; len = MIN(MAXPATHLEN, imgp->ip_strspace); @@ -310,7 +355,8 @@ exec_save_path(struct image_params *imgp, user_addr_t path, int seg) case UIO_USERSPACE64: /* Same for copyin()... */ error = copyinstr(path, imgp->ip_strings, len, &len); break; - case UIO_SYSSPACE32: + case UIO_SYSSPACE: + kpath = CAST_DOWN(char *,path); /* SAFE */ error = copystr(kpath, imgp->ip_strings, len, &len); break; default: @@ -321,96 +367,37 @@ exec_save_path(struct image_params *imgp, user_addr_t path, int seg) if (!error) { imgp->ip_strendp += len; imgp->ip_strspace -= len; - imgp->ip_argv = imgp->ip_strendp; } return(error); } -#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. + * exec_reset_save_path * - * Parameters; struct image_params * image parameter block + * If we detect a shell script, we need to reset the string area + * state so that the interpreter can be saved onto the stack. + + * 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 + * Returns: int 0 Success * - * 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(). + * Implicit returns: + * (imgp->ip_strings) saved path + * (imgp->ip_strspace) space remaining in ip_strings + * (imgp->ip_strendp) start of remaining copy area + * (imgp->ip_argspace) space remaining of NCARGS * - * 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) +exec_reset_save_path(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_name, - IMG_SHSIZE, &len); - if (error) - return (error); - - /* - * provide a replacement string for p->p_comm; we have to use an - * 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); + imgp->ip_strendp = imgp->ip_strings; + imgp->ip_argspace = NCARGS; + imgp->ip_strspace = ( NCARGS + PAGE_SIZE ); - return (-3); + return (0); } -#endif /* IMGPF_POWERPC */ - /* * exec_shell_imgact @@ -438,14 +425,12 @@ exec_shell_imgact(struct image_params *imgp) { char *vdata = imgp->ip_vdata; char *ihp; - char *line_endp; + char *line_startp, *line_endp; char *interp; - char temp[16]; proc_t p; struct fileproc *fp; int fd; int error; - size_t len; /* * Make sure it's a shell script. If we've already redirected @@ -461,71 +446,83 @@ 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'; - /* Check to see if SUGID scripts are permitted. If they aren't then + /* Check to see if SUGID scripts are permitted. If they aren't then * clear the SUGID bits. * imgp->ip_vattr is known to be valid. - */ - if (sugid_scripts == 0) { - imgp->ip_origvattr->va_mode &= ~(VSUID | VSGID); + */ + if (sugid_scripts == 0) { + imgp->ip_origvattr->va_mode &= ~(VSUID | VSGID); } - /* Find the nominal end of the interpreter line */ - for( ihp = &vdata[2]; *ihp != '\n' && *ihp != '#'; ihp++) { - if (ihp >= &vdata[IMG_SHSIZE]) + /* Try to find the first non-whitespace character */ + for( ihp = &vdata[2]; ihp < &vdata[IMG_SHSIZE]; ihp++ ) { + if (IS_EOL(*ihp)) { + /* Did not find interpreter, "#!\n" */ return (ENOEXEC); + } else if (IS_WHITESPACE(*ihp)) { + /* Whitespace, like "#! /bin/sh\n", keep going. */ + } else { + /* Found start of interpreter */ + break; + } + } + + if (ihp == &vdata[IMG_SHSIZE]) { + /* All whitespace, like "#! " */ + return (ENOEXEC); } - line_endp = ihp; - ihp = &vdata[2]; - /* Skip over leading spaces - until the interpreter name */ - while ( ihp < line_endp && ((*ihp == ' ') || (*ihp == '\t'))) - ihp++; + line_startp = ihp; - /* - * Find the last non-whitespace character before the end of line or - * the beginning of a comment; this is our new end of line. - */ - for (;line_endp > ihp && ((*line_endp == ' ') || (*line_endp == '\t')); line_endp--) - continue; + /* Try to find the end of the interpreter+args string */ + for ( ; ihp < &vdata[IMG_SHSIZE]; ihp++ ) { + if (IS_EOL(*ihp)) { + /* Got it */ + break; + } else { + /* Still part of interpreter or args */ + } + } - /* Empty? */ - if (line_endp == ihp) + if (ihp == &vdata[IMG_SHSIZE]) { + /* A long line, like "#! blah blah blah" without end */ return (ENOEXEC); + } - /* copy the interpreter name */ - interp = imgp->ip_interp_name; - while ((ihp < line_endp) && (*ihp != ' ') && (*ihp != '\t')) - *interp++ = *ihp++; - *interp = '\0'; + /* Backtrack until we find the last non-whitespace */ + while (IS_EOL(*ihp) || IS_WHITESPACE(*ihp)) { + ihp--; + } - exec_save_path(imgp, CAST_USER_ADDR_T(imgp->ip_interp_name), - UIO_SYSSPACE32); + /* The character after the last non-whitespace is our logical end of line */ + line_endp = ihp + 1; - ihp = &vdata[2]; - while (ihp < line_endp) { - /* Skip leading whitespace before each argument */ - while ((*ihp == ' ') || (*ihp == '\t')) - ihp++; + /* + * Now we have pointers to the usable part of: + * + * "#! /usr/bin/int first second third \n" + * ^ line_startp ^ line_endp + */ - if (ihp >= line_endp) - break; + /* copy the interpreter name */ + interp = imgp->ip_interp_buffer; + for ( ihp = line_startp; (ihp < line_endp) && !IS_WHITESPACE(*ihp); ihp++) + *interp++ = *ihp; + *interp = '\0'; - /* We have an argument; copy it */ - while ((ihp < line_endp) && (*ihp != ' ') && (*ihp != '\t')) { - *imgp->ip_strendp++ = *ihp++; - imgp->ip_strspace--; - } - *imgp->ip_strendp++ = 0; - imgp->ip_strspace--; - imgp->ip_argc++; - } + exec_reset_save_path(imgp); + exec_save_path(imgp, CAST_USER_ADDR_T(imgp->ip_interp_buffer), + UIO_SYSSPACE); + + /* Copy the entire interpreter + args for later processing into argv[] */ + interp = imgp->ip_interp_buffer; + for ( ihp = line_startp; (ihp < line_endp); ihp++) + *interp++ = *ihp; + *interp = '\0'; /* * If we have a SUID oder SGID script, create a file descriptor @@ -539,7 +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; @@ -549,10 +545,7 @@ exec_shell_imgact(struct image_params *imgp) proc_fdunlock(p); vnode_ref(imgp->ip_vp); - snprintf(temp, sizeof(temp), "/dev/fd/%d", fd); - error = copyoutstr(temp, imgp->ip_user_fname, sizeof(temp), &len); - if (error) - return(error); + imgp->ip_interp_sugid_fd = fd; } return (-3); @@ -604,6 +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) { @@ -661,7 +692,7 @@ use_arch: /* Read the Mach-O header out of fat_arch */ error = vn_rdwr(UIO_READ, imgp->ip_vp, imgp->ip_vdata, PAGE_SIZE, fat_arch.offset, - UIO_SYSSPACE32, (IO_UNIT|IO_NODELOCKED), + UIO_SYSSPACE, (IO_UNIT|IO_NODELOCKED), cred, &resid, p); if (error) { goto bad; @@ -712,7 +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; @@ -722,12 +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 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; @@ -750,9 +789,6 @@ exec_mach_imgact(struct image_params *imgp) thread = current_thread(); uthread = get_bsdthread_info(thread); - 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; @@ -782,8 +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; } @@ -793,39 +828,33 @@ grade: if (error) goto bad; - /* - * Hack for binary compatability; put three NULs on the end of the - * string area, and round it up to the next word boundary. This - * ensures padding with NULs to the boundary. - */ - imgp->ip_strendp[0] = 0; - imgp->ip_strendp[1] = 0; - imgp->ip_strendp[2] = 0; - imgp->ip_strendp += (((imgp->ip_strendp - imgp->ip_strings) + NBPW-1) & ~(NBPW-1)); + error = exec_add_apple_strings(imgp); + if (error) + goto bad; + + AUDIT_ARG(argv, imgp->ip_startargv, imgp->ip_argc, + imgp->ip_endargv - imgp->ip_startargv); + 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(). + * We are being called to activate an image subsequent to a vfork() + * operation; in this case, we know that our task, thread, and + * uthread are actually those of our parent, and our proc, which we + * obtained indirectly from the image_params vfs_context_t, is the + * new child process. */ - if (imgp->ip_vattr->va_fsid == exec_archhandler_ppc.fsid && - imgp->ip_vattr->va_fileid == (uint64_t)((u_long)exec_archhandler_ppc.fileid)) { - imgp->ip_flags |= IMGPF_POWERPC; - } -#endif /* IMGPF_POWERPC */ - - if (vfexec) { - imgp->ip_vfork_thread = fork_create_child(task, p, FALSE, (imgp->ip_flags & IMGPF_IS_64BIT)); - if (imgp->ip_vfork_thread == NULL) { - error = ENOMEM; - goto bad; + if (vfexec || spawn) { + if (vfexec) { + 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; + } } + /* reset local idea of thread, uthread, task */ - thread = imgp->ip_vfork_thread; + thread = imgp->ip_new_thread; uthread = get_bsdthread_info(thread); task = new_task = get_threadtask(thread); map = get_task_map(task); @@ -837,32 +866,27 @@ grade: * We set these flags here; this is OK, since if we fail after * this point, we have already destroyed the parent process anyway. */ + task_set_dyld_info(task, MACH_VM_MIN_ADDRESS, 0); if (imgp->ip_flags & IMGPF_IS_64BIT) { task_set_64bit(task, TRUE); - OSBitOrAtomic(P_LP64, (UInt32 *)&p->p_flag); + OSBitOrAtomic(P_LP64, &p->p_flag); } else { task_set_64bit(task, FALSE); - OSBitAndAtomic(~((uint32_t)P_LP64), (UInt32 *)&p->p_flag); + OSBitAndAtomic(~((uint32_t)P_LP64), &p->p_flag); } /* * Load the Mach-O file. - */ - - /* + * * NOTE: An error after this point indicates we have potentially - * destroyed or overwrote some process state while attempting an + * 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. */ - /* - * We reset the task to 64-bit (or not) here. It may have picked up - * a new map, and we need that to reflect its true 64-bit nature. - */ - - task_set_64bit(task, - ((imgp->ip_flags & IMGPF_IS_64BIT) == IMGPF_IS_64BIT)); - /* * Actually load the image file we previously decided to load. */ @@ -873,6 +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); /* @@ -880,9 +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; } @@ -891,53 +921,52 @@ 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; - /* load_machfile() maps the vnode */ - (void)ubc_map(imgp->ip_vp, PROT_READ | PROT_EXEC); - /* * Set up the system reserved areas in the new address space. */ 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()); /* - * Close file descriptors - * which specify close-on-exec. + * Close file descriptors which specify close-on-exec. */ - fdexec(p); + fdexec(p, psa != NULL ? psa->psa_flags : 0); /* * deal with set[ug]id. */ error = exec_handle_sugid(imgp); + if (error) { + goto badtoolate; + } - proc_knote(p, NOTE_EXEC); + /* + * deal with voucher on exec-calling thread. + */ + if (imgp->ip_new_thread == NULL) + thread_set_mach_voucher(current_thread(), IPC_VOUCHER_NULL); - if (!vfexec && (p->p_lflag & P_LTRACED)) + /* 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; } - if (vfexec) { + if (vfexec || spawn) { old_map = vm_map_switch(get_task_map(task)); } @@ -951,7 +980,7 @@ grade: ap = p->user_stack; error = exec_copyout_strings(imgp, &ap); if (error) { - if (vfexec) + if (vfexec || spawn) vm_map_switch(old_map); goto badtoolate; } @@ -961,23 +990,25 @@ grade: if (load_result.dynlinker) { uint64_t ap; + int new_ptr_size = (imgp->ip_flags & IMGPF_IS_64BIT) ? 8 : 4; /* Adjust the stack */ - if (imgp->ip_flags & IMGPF_IS_64BIT) { - ap = thread_adjuserstack(thread, -8); - error = copyoutptr(load_result.mach_header, ap, 8); - } else { - ap = thread_adjuserstack(thread, -4); - error = suword(ap, load_result.mach_header); - } + ap = thread_adjuserstack(thread, -new_ptr_size); + error = copyoutptr(load_result.mach_header, ap, new_ptr_size); + if (error) { - if (vfexec) - 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); } - if (vfexec) { + /* Avoid immediate VM faults back into kernel */ + exec_prefault_data(p, imgp, &load_result); + + if (vfexec || spawn) { vm_map_switch(old_map); } /* Set the entry point */ @@ -1026,6 +1057,18 @@ grade: p->p_comm[imgp->ip_ndp->ni_cnd.cn_namelen] = '\0'; } + 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 #if CONFIG_DTRACE /* * Invalidate any predicate evaluation already cached for this thread by DTrace. @@ -1057,8 +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 @@ -1070,11 +1113,11 @@ grade: */ kdbg_trace_string(p, &dbg_arg1, &dbg_arg2, &dbg_arg3, &dbg_arg4); - if (vfexec) { + if (vfexec || spawn) { KERNEL_DEBUG_CONSTANT1((TRACEDBG_CODE(DBG_TRACE_DATA, 2)) | DBG_FUNC_NONE, - p->p_pid ,0,0,0, (unsigned int)thread); + p->p_pid ,0,0,0, (uintptr_t)thread_tid(thread)); KERNEL_DEBUG_CONSTANT1((TRACEDBG_CODE(DBG_TRACE_STRING, 2)) | DBG_FUNC_NONE, - dbg_arg1, dbg_arg2, dbg_arg3, dbg_arg4, (unsigned int)thread); + dbg_arg1, dbg_arg2, dbg_arg3, dbg_arg4, (uintptr_t)thread_tid(thread)); } else { KERNEL_DEBUG_CONSTANT((TRACEDBG_CODE(DBG_TRACE_DATA, 2)) | DBG_FUNC_NONE, p->p_pid ,0,0,0,0); @@ -1083,18 +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, (UInt32 *)&p->p_flag); - else -#endif /* IMGPF_POWERPC */ - OSBitAndAtomic(~((uint32_t)P_TRANSLATED), (UInt32 *)&p->p_flag); - OSBitAndAtomic(~((uint32_t)P_AFFINITY), (UInt32 *)&p->p_flag); + OSBitAndAtomic(~((uint32_t)(P_TRANSLATED | P_AFFINITY)), &p->p_flag); /* * If posix_spawned with the START_SUSPENDED flag, stop the @@ -1112,9 +1148,10 @@ grade: /* * mark as execed, wakeup the process that vforked (if any) and tell - * it that it now has it's own resources back + * it that it now has its own resources back */ - OSBitOrAtomic(P_EXEC, (UInt32 *)&p->p_flag); + OSBitOrAtomic(P_EXEC, &p->p_flag); + proc_resetregister(p); if (p->p_pptr && (p->p_lflag & P_LPPWAIT)) { proc_lock(p); p->p_lflag &= ~P_LPPWAIT; @@ -1122,16 +1159,40 @@ grade: wakeup((caddr_t)p->p_pptr); } + /* + * Pay for our earlier safety; deliver the delayed signals from + * the incomplete vfexec process now that it's complete. + */ if (vfexec && (p->p_lflag & P_LTRACED)) { psignal_vfork(p, new_task, thread, SIGTRAP); } + goto done; + badtoolate: - if (vfexec) { + /* Don't allow child process to execute any instructions */ + if (!spawn) { + if (vfexec) { + psignal_vfork(p, new_task, thread, SIGKILL); + } else { + psignal(p, SIGKILL); + } + + /* 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: @@ -1154,9 +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} }; @@ -1190,39 +1248,58 @@ 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 */ int i; int iterlimit = EAI_ITERLIMIT; + proc_t p = vfs_context_proc(imgp->ip_vfs_context); error = execargs_alloc(imgp); if (error) - goto bad; + goto bad_notrans; - /* - * XXXAUDIT: Note: the double copyin introduces an audit - * race. To correct this race, we must use a single - * copyin(), e.g. by passing a flag to namei to indicate an - * external path buffer is being used. - */ error = exec_save_path(imgp, imgp->ip_user_fname, imgp->ip_seg); if (error) { - goto bad; + goto bad_notrans; } + /* Use imgp->ip_strings, which contains the copyin-ed exec path */ DTRACE_PROC1(exec, uintptr_t, imgp->ip_strings); - NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNPATH1, - imgp->ip_seg, imgp->ip_user_fname, imgp->ip_vfs_context); + 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; - imgp->ip_ndp = &nd; /* successful namei(); call nameidone() later */ - imgp->ip_vp = nd.ni_vp; /* if set, need to vnode_put() at some point */ + goto bad_notrans; + 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 + * sure another thread hasn't started exiting the process. We grab + * the proc lock to check p_lflag initially, and the transition + * mechanism ensures that the value doesn't change after we release + * the lock. + */ + proc_lock(p); + if (p->p_lflag & P_LEXIT) { + proc_unlock(p); + goto bad_notrans; + } + error = proc_transstart(p, 1, 0); + proc_unlock(p); + if (error) + goto bad_notrans; error = exec_check_permissions(imgp); if (error) @@ -1235,7 +1312,7 @@ again: } error = vn_rdwr(UIO_READ, imgp->ip_vp, imgp->ip_vdata, PAGE_SIZE, 0, - UIO_SYSSPACE32, IO_NODELOCKED, + UIO_SYSSPACE, IO_NODELOCKED, vfs_context_ucred(imgp->ip_vfs_context), &resid, vfs_context_proc(imgp->ip_vfs_context)); if (error) @@ -1272,26 +1349,28 @@ encapsulated_binary: break; } mac_vnode_label_copy(imgp->ip_vp->v_label, - imgp->ip_scriptlabelp); + 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(ndp); + vnode_put(imgp->ip_vp); imgp->ip_vp = NULL; /* already put */ - nd.ni_cnd.cn_nameiop = LOOKUP; - nd.ni_cnd.cn_flags = (nd.ni_cnd.cn_flags & HASBUF) | - (FOLLOW | LOCKLEAF); + imgp->ip_ndp = NULL; /* already nameidone */ -#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 */ + /* Use imgp->ip_strings, which exec_shell_imgact reset to the interpreter */ + NDINIT(ndp, LOOKUP, OP_LOOKUP, FOLLOW | LOCKLEAF, + UIO_SYSSPACE, CAST_USER_ADDR_T(imgp->ip_strings), imgp->ip_vfs_context); - nd.ni_segflg = UIO_SYSSPACE32; - nd.ni_dirp = CAST_USER_ADDR_T(imgp->ip_interp_name); + proc_transend(p, 0); goto again; default: @@ -1306,104 +1385,212 @@ 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: + proc_transend(p, 0); + +bad_notrans: if (imgp->ip_strings) execargs_free(imgp); if (imgp->ip_ndp) nameidone(imgp->ip_ndp); + if (ndp) + FREE(ndp, M_TEMP); return (error); } + /* - * exec_handle_port_actions + * exec_handle_spawnattr_policy * - * Description: Go through the _posix_port_actions_t contents, - * calling task_set_special_port and task_set_exception_ports - * for the current task. + * Description: Decode and apply the posix_spawn apptype, qos clamp, and watchport ports to the task. * - * Parameters: struct image_params * Image parameter block + * Parameters: proc_t p process to apply attributes to + * int psa_apptype posix spawn attribute apptype * - * Returns: 0 Success - * KERN_FAILURE Failure + * Returns: 0 Success */ -static int -exec_handle_port_actions(struct image_params *imgp) +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) { - _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; - kern_return_t ret = KERN_SUCCESS; - int i; - - for (i = 0; i < pacts->pspa_count; i++) { - act = &pacts->pspa_actions[i]; + int apptype = TASK_APPTYPE_NONE; + int qos_clamp = THREAD_QOS_UNSPECIFIED; - ret = ipc_object_copyin(get_task_ipcspace(current_task()), - (mach_port_name_t) act->new_port, - MACH_MSG_TYPE_COPY_SEND, - (ipc_object_t *) &port); + if ((psa_apptype & POSIX_SPAWN_PROC_TYPE_MASK) != 0) { + int proctype = psa_apptype & POSIX_SPAWN_PROC_TYPE_MASK; - if (ret) - return ret; - - switch (act->port_type) { - case PSPA_SPECIAL: - ret = task_set_special_port(task, - act->which, - port); + 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 PSPA_EXCEPTION: - ret = task_set_exception_ports(task, - act->mask, - port, - act->behavior, - act->flavor); + case POSIX_SPAWN_PROC_TYPE_APP_TAL: + apptype = TASK_APPTYPE_APP_TAL; break; default: - ret = KERN_FAILURE; + apptype = TASK_APPTYPE_NONE; + /* TODO: Should an invalid value here fail the spawn? */ + break; } - /* action failed, so release port resources */ - if (ret) { - ipc_port_release_send(port); - return ret; + } + + 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; } } - return ret; + 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_file_actions + * exec_handle_port_actions * - * Description: Go through the _posix_file_actions_t contents applying the - * open, close, and dup2 operations to the open file table for - * the current process. + * Description: Go through the _posix_port_actions_t contents, + * calling task_set_special_port, task_set_exception_ports + * and/or audit_session_spawnjoin for the current task. * * Parameters: struct image_params * Image parameter block + * short psa_flags posix spawn attribute flags * * Returns: 0 Success - * ??? - * - * Note: Actions are applied in the order specified, with the credential + * EINVAL Failure + * ENOTSUP Illegal posix_spawn attr flag was set + */ +static errno_t +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 = 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()), + 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)) + 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; +#endif + 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); + break; + } + } + +done: + if (0 != ret) + DTRACE_PROC1(spawn__port__failure, mach_port_name_t, act->new_port); + return (ret); +} + +/* + * exec_handle_file_actions + * + * Description: Go through the _posix_file_actions_t contents applying the + * open, close, and dup2 operations to the open file table for + * the current process. + * + * Parameters: struct image_params * Image parameter block + * + * Returns: 0 Success + * ??? + * + * Note: Actions are applied in the order specified, with the credential * of the parent process. This is done to permit the parent * process to utilize POSIX_SPAWN_RESETIDS to drop privilege in * the child following operations the child may in fact not be * normally permitted to perform. */ static int -exec_handle_file_actions(struct image_params *imgp) +exec_handle_file_actions(struct image_params *imgp, short psa_flags) { int error = 0; int action; proc_t p = vfs_context_proc(imgp->ip_vfs_context); _posix_spawn_file_actions_t px_sfap = imgp->ip_px_sfa; - register_t ival[2]; /* dummy retval for system calls) */ + int ival[2]; /* dummy retval for system calls) */ for (action = 0; action < px_sfap->psfa_act_count; action++) { _psfa_action_t *psfa = &px_sfap->psfa_act_acts[ action]; @@ -1415,35 +1602,48 @@ exec_handle_file_actions(struct image_params *imgp) * a path argument, which is normally copied in from * user space; because of this, we have to support an * open from kernel space that passes an address space - * context oof UIO_SYSSPACE, and casts the address + * 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, 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 that - * rearchitecting all the open code to preallocate fd + * accident, then drop out here. This is easier than + * reworking all the open code to preallocate fd * slots, and internally taking one as an argument. */ if (error || ival[0] == psfa->psfaa_filedes) @@ -1502,18 +1702,180 @@ exec_handle_file_actions(struct image_params *imgp) } break; + case PSFA_INHERIT: { + struct fcntl_nocancel_args fcntla; + + /* + * Check to see if the descriptor exists, and + * ensure it's -not- marked as close-on-exec. + * + * Attempting to "inherit" a guarded fd will + * result in a error. + */ + 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); + } + + } + break; + default: error = EINVAL; break; } + /* 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; + } } - return (error); + if (error != 0 || (psa_flags & POSIX_SPAWN_CLOEXEC_DEFAULT) == 0) + return (error); + + /* + * If POSIX_SPAWN_CLOEXEC_DEFAULT is set, behave (during + * this spawn only) as if "close on exec" is the default + * disposition of all pre-existing file descriptors. In this case, + * the list of file descriptors mentioned in the file actions + * are the only ones that can be inherited, so mark them now. + * + * The actual closing part comes later, in fdexec(). + */ + proc_fdlock(p); + for (action = 0; action < px_sfap->psfa_act_count; action++) { + _psfa_action_t *psfa = &px_sfap->psfa_act_acts[action]; + int fd = psfa->psfaa_filedes; + + switch (psfa->psfaa_type) { + case PSFA_DUP2: + fd = psfa->psfaa_openargs.psfao_oflag; + /*FALLTHROUGH*/ + case PSFA_OPEN: + case PSFA_INHERIT: + *fdflags(p, fd) |= UF_INHERIT; + break; + + case PSFA_CLOSE: + break; + } + } + proc_fdunlock(p); + + 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 @@ -1539,53 +1901,67 @@ exec_handle_file_actions(struct image_params *imgp) * exec_activate_image:??? * mac_execve_enter:??? * - * TODO: More gracefully handle failures after vfork - * Expect to need __mac_posix_spawn() at some point... + * TODO: Expect to need __mac_posix_spawn() at some point... * Handle posix_spawnattr_t * Handle posix_spawn_file_actions_t */ int -posix_spawn(proc_t ap, struct posix_spawn_args *uap, register_t *retval) +posix_spawn(proc_t ap, struct posix_spawn_args *uap, int32_t *retval) { proc_t p = ap; /* quiet bogus GCC vfork() warning */ user_addr_t pid = uap->pid; - register_t ival[2]; /* dummy retval for vfork() */ - struct image_params image_params, *imgp; - struct vnode_attr va; - struct vnode_attr origva; + int ival[2]; /* dummy retval for setpgid() */ + char *bufp = NULL; + struct image_params *imgp; + struct vnode_attr *vap; + struct vnode_attr *origvap; struct uthread *uthread = 0; /* compiler complains if not set to 0*/ int error, sig; - task_t task; - int numthreads; char alt_p_comm[sizeof(p->p_comm)] = {0}; /* for PowerPC */ int is_64 = IS_64BIT_PROCESS(p); - int undo_vfork = 0; struct vfs_context context; struct user__posix_spawn_args_desc px_args; struct _posix_spawnattr px_sa; _posix_spawn_file_actions_t px_sfap = NULL; _posix_spawn_port_actions_t px_spap = NULL; - struct __user_sigaction vec; + struct __kern_sigaction vec; + 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); - imgp = &image_params; + /* + * Allocate a big chunk for locals instead of using stack since these + * structures are pretty big. + */ + MALLOC(bufp, char *, (sizeof(*imgp) + sizeof(*vap) + sizeof(*origvap)), M_TEMP, M_WAITOK | M_ZERO); + imgp = (struct image_params *) bufp; + if (bufp == NULL) { + error = ENOMEM; + goto bad; + } + vap = (struct vnode_attr *) (bufp + sizeof(*imgp)); + origvap = (struct vnode_attr *) (bufp + sizeof(*imgp) + sizeof(*vap)); /* Initialize the common data in the image_params structure */ - bzero(imgp, sizeof(*imgp)); imgp->ip_user_fname = uap->path; imgp->ip_user_argv = uap->argv; imgp->ip_user_envv = uap->envp; - imgp->ip_vattr = &va; - imgp->ip_origvattr = &origva; + imgp->ip_vattr = vap; + imgp->ip_origvattr = origvap; imgp->ip_vfs_context = &context; 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) { error = copyin(uap->adesc, &px_args, sizeof(px_args)); } else { - struct _posix_spawn_args_desc px_args32; + struct user32__posix_spawn_args_desc px_args32; error = copyin(uap->adesc, &px_args32, sizeof(px_args32)); @@ -1599,17 +1975,23 @@ posix_spawn(proc_t ap, struct posix_spawn_args *uap, register_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; } @@ -1631,6 +2013,12 @@ posix_spawn(proc_t ap, struct posix_spawn_args *uap, register_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 */ @@ -1651,49 +2039,168 @@ posix_spawn(proc_t ap, struct posix_spawn_args *uap, register_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 */ + uthread = get_bsdthread_info(current_thread()); + + /* + * ; this does not result in a behaviour change + * relative to Leopard, so there should not be any existing code + * which depends on it. + */ + if (uthread->uu_flag & UT_VFORK) { + error = EINVAL; + goto bad; } + /* + * If we don't have the extension flag that turns "posix_spawn()" + * into "execve() with options", then we will be creating a new + * process which does not inherit memory from the parent process, + * which is one of the most expensive things about using fork() + * and execve(). + */ if (imgp->ip_px_sa == NULL || !(px_sa.psa_flags & POSIX_SPAWN_SETEXEC)){ - if ((error = vfork(p, NULL, ival)) != 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; - undo_vfork = 1; + } + 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 */ + } } - /* "reenter the kernel" on a new vfork()'ed process */ - uthread = get_bsdthread_info(current_thread()); - if (undo_vfork) - p = uthread->uu_proc; + 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 */ context.vc_thread = current_thread(); context.vc_ucred = p->p_ucred; /* XXX must NOT be kauth_cred_get() */ + /* + * However, if we're not in the setexec case, redirect the context + * to the newly created process instead + */ + if (spawn_no_exec) + context.vc_thread = imgp->ip_new_thread; + /* * Post fdcopy(), pre exec_handle_sugid() - this is where we want * to handle the file_actions. Since vfork() also ends up setting * us into the parent process group, and saved off the signal flags, * this is also where we want to handle the spawn flags. */ + /* Has spawn file actions? */ - if (imgp->ip_px_sfa != NULL && - (error = exec_handle_file_actions(imgp)) != 0) { - goto bad; + if (imgp->ip_px_sfa != NULL) { + /* + * The POSIX_SPAWN_CLOEXEC_DEFAULT flag + * is handled in exec_handle_file_actions(). + */ + if ((error = exec_handle_file_actions(imgp, + imgp->ip_px_sa != NULL ? px_sa.psa_flags : 0)) != 0) + goto bad; } /* Has spawn port actions? */ - if (imgp->ip_px_spa != NULL) { - /* Only allowed when not under vfork */ - if (!(px_sa.psa_flags & POSIX_SPAWN_SETEXEC)) { - error = ENOTSUP; - goto bad; + 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)) != 0) + + if ((error = exec_handle_port_actions(imgp, + 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? */ if (imgp->ip_px_sa != NULL) { - /* Set the process group ID of the child process */ + /* + * Set the process group ID of the child process; this has + * to happen before the image activation. + */ if (px_sa.psa_flags & POSIX_SPAWN_SETPGROUP) { struct setpgid_args spga; spga.pid = p->p_pid; @@ -1705,11 +2212,15 @@ posix_spawn(proc_t ap, struct posix_spawn_args *uap, register_t *retval) if((error = setpgid(p, &spga, ival)) != 0) goto bad; } + /* * Reset UID/GID to parent's RUID/RGID; This works only * because the operation occurs *after* the vfork() and * before the call to exec_handle_sugid() by the image - * activator called from exec_activate_image(). + * activator called from exec_activate_image(). POSIX + * requires that any setuid/setgid bits on the process + * image will take precedence over the spawn attributes + * (re)setting them. * * The use of p_ucred is safe, since we are acting on the * new process, and it has no threads other than the one @@ -1717,20 +2228,109 @@ posix_spawn(proc_t ap, struct posix_spawn_args *uap, register_t *retval) */ if (px_sa.psa_flags & POSIX_SPAWN_RESETIDS) { kauth_cred_t my_cred = p->p_ucred; - kauth_cred_t my_new_cred = kauth_cred_setuidgid(my_cred, my_cred->cr_ruid, my_cred->cr_rgid); - if (my_new_cred != my_cred) + kauth_cred_t my_new_cred = kauth_cred_setuidgid(my_cred, kauth_cred_getruid(my_cred), kauth_cred_getrgid(my_cred)); + if (my_new_cred != my_cred) { p->p_ucred = my_new_cred; + /* update cred on proc */ + PROC_UPDATE_CREDS_ONPROC(p); + } + } + + /* + * 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); + + /* + * Forcibly disallow execution from data pages for the spawned process + * even if it would otherwise be permitted by the architecture default. + */ + if (px_sa.psa_flags & _POSIX_SPAWN_ALLOW_DATA_EXEC) + imgp->ip_flags |= IMGPF_ALLOW_DATA_EXEC; + } + + /* + * Disable ASLR during image activation. This occurs either if the + * _POSIX_SPAWN_DISABLE_ASLR attribute was found above or if + * P_DISABLE_ASLR was inherited from the parent process. + */ + if (p->p_flag & P_DISABLE_ASLR) + imgp->ip_flags |= IMGPF_DISABLE_ASLR; + + /* + * Clear transition flag so we won't hang if exec_activate_image() causes + * an automount (and launchd does a proc sysctl to service it). + * + * , . + */ + if (spawn_no_exec) { + proc_transend(p, 0); + proc_transit_set = 0; + } + +#if MAC_SPAWN /* XXX */ + if (uap->mac_p != USER_ADDR_NULL) { + error = mac_execve_enter(uap->mac_p, imgp); + if (error) + goto bad; + } +#endif + + /* + * Activate the image + */ + error = exec_activate_image(imgp); + + if (error == 0) { + /* process completed the exec */ + exec_done = TRUE; + } else if (error == -1) { + /* Image not claimed by any activator? */ + error = ENOEXEC; + } + + /* + * If we have a spawn attr, and it contains signal related flags, + * the we need to process them in the "context" of the new child + * process, so we have to process it following image activation, + * prior to making the thread runnable in user space. This is + * necessitated by some signal information being per-thread rather + * than per-process, and we don't have the new allocation in hand + * until after the image is activated. + */ + if (!error && imgp->ip_px_sa != NULL) { + thread_t child_thread = current_thread(); + uthread_t child_uthread = uthread; + + /* + * If we created a new child thread, then the thread and + * uthread are different than the current ones; otherwise, + * we leave them, since we are in the exec case instead. + */ + if (spawn_no_exec) { + child_thread = imgp->ip_new_thread; + child_uthread = get_bsdthread_info(child_thread); } + /* * Mask a list of signals, instead of them being unmasked, if * they were unmasked in the parent; note that some signals * are not maskable. */ if (px_sa.psa_flags & POSIX_SPAWN_SETSIGMASK) - uthread->uu_sigmask = (px_sa.psa_sigmask & ~sigcantmask); + child_uthread->uu_sigmask = (px_sa.psa_sigmask & ~sigcantmask); /* * Default a list of signals instead of ignoring them, if - * they were ignored in the parent. + * they were ignored in the parent. Note that we pass + * spawn_no_exec to setsigvec() to indicate that we called + * fork1() and therefore do not need to call proc_signalstart() + * internally. */ if (px_sa.psa_flags & POSIX_SPAWN_SETSIGDEF) { vec.sa_handler = SIG_DFL; @@ -1738,107 +2338,276 @@ posix_spawn(proc_t ap, struct posix_spawn_args *uap, register_t *retval) vec.sa_mask = 0; vec.sa_flags = 0; for (sig = 0; sig < NSIG; sig++) - if (px_sa.psa_sigdefault && 1 << sig) { - error = setsigvec(p, sig, &vec); + if (px_sa.psa_sigdefault & (1 << sig)) { + 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) { + case POSIX_SPAWN_PCONTROL_THROTTLE: + p->p_pcaction = P_PCTHROTTLE; + break; + case POSIX_SPAWN_PCONTROL_SUSPEND: + p->p_pcaction = P_PCSUSP; + break; + case POSIX_SPAWN_PCONTROL_KILL: + p->p_pcaction = P_PCKILL; + break; + case POSIX_SPAWN_PCONTROL_NONE: + default: + p->p_pcaction = 0; + break; + }; + } + 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 } /* - * XXXAUDIT: Currently, we only audit the pathname of the binary. - * There may also be poor interaction with dyld. - */ + * If we successfully called fork1(), we always need to do this; + * we identify this case by noting the IMGPF_SPAWN flag. This is + * because we come back from that call with signals blocked in the + * child, and we have to unblock them, but we want to wait until + * after we've performed any spawn actions. This has to happen + * before check_for_signature(), which uses psignal. + */ + if (spawn_no_exec) { + if (proc_transit_set) + proc_transend(p, 0); - task = current_task(); + /* + * Drop the signal lock on the child which was taken on our + * behalf by forkproc()/cloneproc() to prevent signals being + * received by the child in a partially constructed state. + */ + proc_signalend(p, 0); + + /* flag the 'fork' has occurred */ + proc_knote(p->p_pptr, NOTE_FORK | p->p_pid); + /* then flag exec has occurred */ + /* 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); - /* If we're not in vfork, don't permit a mutithreaded task to exec */ - if (!(uthread->uu_flag & UT_VFORK)) { - if (task != kernel_task) { - numthreads = get_task_numacts(task); - if (numthreads <= 0 ) { - error = EINVAL; - goto bad; - } - if (numthreads > 1) { - error = ENOTSUP; - goto bad; + /* 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; } -#if MAC_SPAWN /* XXX */ - if (uap->mac_p != USER_ADDR_NULL) { - error = mac_execve_enter(uap->mac_p, imgp); - if (error) - goto bad; + /* + * We have to delay operations which might throw a signal until after + * the signals have been unblocked; however, we want that to happen + * after exec_resettextvp() so that the textvp is correct when they + * fire. + */ + if (error == 0) { + error = check_for_signature(p, imgp); + + /* + * Pay for our earlier safety; deliver the delayed signals from + * the incomplete spawn process now that it's complete. + */ + if (imgp != NULL && spawn_no_exec && (p->p_lflag & P_LTRACED)) { + psignal_vfork(p, p->task, imgp->ip_new_thread, SIGTRAP); + } } -#endif - if ((error = exec_activate_image(imgp)) != 0) - goto bad; -bad: - /* Image not claimed by any activator? */ - if (error == -1) - error = ENOEXEC; - if (error == 0) { - exec_resettextvp(p, imgp); - } - if (imgp->ip_vp) - vnode_put(imgp->ip_vp); - if (imgp->ip_strings) - execargs_free(imgp); - if (imgp->ip_px_sfa != NULL) - FREE(imgp->ip_px_sfa, M_TEMP); - if (imgp->ip_px_spa != NULL) - FREE(imgp->ip_px_spa, M_TEMP); + 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) + FREE(imgp->ip_px_sfa, M_TEMP); + if (imgp->ip_px_spa != NULL) + FREE(imgp->ip_px_spa, M_TEMP); + #if CONFIG_MACF - if (imgp->ip_execlabelp) - mac_cred_label_free(imgp->ip_execlabelp); - if (imgp->ip_scriptlabelp) - mac_vnode_label_free(imgp->ip_scriptlabelp); + 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) + mac_vnode_label_free(imgp->ip_scriptlabelp); #endif - if (undo_vfork) { + } + +#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 { if (error) { DTRACE_PROC1(exec__failure, int, error); - vfork_exit(p, W_EXITCODE(-1, 0)); } 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) { /* - * Returning to the parent process... - * * If the parent wants the pid, copy it out */ if (pid != USER_ADDR_NULL) (void)suword(pid, p->p_pid); retval[0] = error; - /* - * Override inherited code signing flags with the - * ones for the process that is being successfully - * loaded - */ - proc_lock(p); - p->p_csflags = imgp->ip_csflags; - proc_unlock(p); - vfork_return(p, NULL, error); - (void)thread_resume(imgp->ip_vfork_thread); - } - if (!error) { /* - * Override inherited code signing flags with the - * ones for the process that is being successfully - * loaded + * If we had an error, perform an internal reap ; this is + * entirely safe, as we have a real process backing us. */ - proc_lock(p); - p->p_csflags = imgp->ip_csflags; - proc_unlock(p); - DTRACE_PROC(exec__success); - } else { - DTRACE_PROC1(exec__failure, int, error); - } + if (error) { + proc_list_lock(); + p->p_listflag |= P_LIST_DEADPARENT; + proc_list_unlock(); + proc_lock(p); + /* make sure no one else has killed it off... */ + if (p->p_stat != SZOMB && p->exit_thread == NULL) { + p->exit_thread = current_thread(); + proc_unlock(p); + exit1(p, 1, (int *)NULL); + if (exec_done == FALSE) { + task_deallocate(get_threadtask(imgp->ip_new_thread)); + thread_deallocate(imgp->ip_new_thread); + } + } else { + /* someone is doing it for us; just skip it */ + proc_unlock(p); + } + } else { + /* + * Return to the child + * + * Note: the image activator earlier dropped the + * task/thread references to the newly spawned + * process; this is OK, since we still have suspended + * queue references on them, so we should be fine + * with the delayed resume of the thread here. + */ + (void)thread_resume(imgp->ip_new_thread); + } + } + if (bufp != NULL) { + FREE(bufp, M_TEMP); + } + return(error); } @@ -1866,11 +2635,13 @@ bad: */ /* ARGSUSED */ int -execve(proc_t p, struct execve_args *uap, register_t *retval) +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; @@ -1907,61 +2678,48 @@ execve(proc_t p, struct execve_args *uap, register_t *retval) * TODO: Dynamic linker header address on stack is copied via suword() */ int -__mac_execve(proc_t p, struct __mac_execve_args *uap, register_t *retval) +__mac_execve(proc_t p, struct __mac_execve_args *uap, int32_t *retval) { - struct image_params image_params, *imgp; - struct vnode_attr va; - struct vnode_attr origva; - struct uthread *uthread; + char *bufp = NULL; + struct image_params *imgp; + struct vnode_attr *vap; + struct vnode_attr *origvap; int error; - task_t task; - int numthreads; 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() */ - imgp = &image_params; - + /* Allocate a big chunk for locals instead of using stack since these + * structures a pretty big. + */ + MALLOC(bufp, char *, (sizeof(*imgp) + sizeof(*vap) + sizeof(*origvap)), M_TEMP, M_WAITOK | M_ZERO); + imgp = (struct image_params *) bufp; + if (bufp == NULL) { + error = ENOMEM; + goto exit_with_error; + } + vap = (struct vnode_attr *) (bufp + sizeof(*imgp)); + origvap = (struct vnode_attr *) (bufp + sizeof(*imgp) + sizeof(*vap)); + /* Initialize the common data in the image_params structure */ - bzero(imgp, sizeof(*imgp)); imgp->ip_user_fname = uap->fname; imgp->ip_user_argv = uap->argp; imgp->ip_user_envv = uap->envp; - imgp->ip_vattr = &va; - imgp->ip_origvattr = &origva; + imgp->ip_vattr = vap; + imgp->ip_origvattr = origvap; imgp->ip_vfs_context = &context; - imgp->ip_flags = (is_64 ? IMGPF_WAS_64BIT : IMGPF_NONE); + 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; - /* - * XXXAUDIT: Currently, we only audit the pathname of the binary. - * There may also be poor interaction with dyld. - */ - - task = current_task(); uthread = get_bsdthread_info(current_thread()); - - /* If we're not in vfork, don't permit a mutithreaded task to exec */ - if (!(uthread->uu_flag & UT_VFORK)) { - if (task != kernel_task) { - proc_lock(p); - numthreads = get_task_numacts(task); - if (numthreads <= 0 ) { - proc_unlock(p); - kauth_cred_unref(&context.vc_ucred); - return(EINVAL); - } - if (numthreads > 1) { - proc_unlock(p); - kauth_cred_unref(&context.vc_ucred); - return(ENOTSUP); - } - proc_unlock(p); - } + if (uthread->uu_flag & UT_VFORK) { + imgp->ip_flags |= IMGPF_VFORK_EXEC; } #if CONFIG_MACF @@ -1969,14 +2727,12 @@ __mac_execve(proc_t p, struct __mac_execve_args *uap, register_t *retval) error = mac_execve_enter(uap->mac_p, imgp); if (error) { kauth_cred_unref(&context.vc_ucred); - return (error); + goto exit_with_error; } } #endif - proc_transstart(p, 0); error = exec_activate_image(imgp); - proc_transend(p, 0); kauth_cred_unref(&context.vc_ucred); @@ -1986,9 +2742,12 @@ __mac_execve(proc_t p, struct __mac_execve_args *uap, register_t *retval) if (error == 0) { exec_resettextvp(p, imgp); + error = check_for_signature(p, imgp); } 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 @@ -1998,23 +2757,38 @@ __mac_execve(proc_t p, struct __mac_execve_args *uap, register_t *retval) mac_vnode_label_free(imgp->ip_scriptlabelp); #endif if (!error) { - /* - * Override inherited code signing flags with the - * ones for the process that is being successfully - * loaded - */ - proc_lock(p); - p->p_csflags = imgp->ip_csflags; - proc_unlock(p); + /* 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); - 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_vfork_thread); + (void)thread_resume(imgp->ip_new_thread); } } else { DTRACE_PROC1(exec__failure, int, error); } + +exit_with_error: + if (bufp != NULL) { + FREE(bufp, M_TEMP); + } return(error); } @@ -2067,8 +2841,6 @@ copyinptr(user_addr_t froma, user_addr_t *toptr, int ptr_size) * Returns: 0 Success * EFAULT Bad 'ua' * - * Implicit returns: - * *ptr_size Modified */ static int copyoutptr(user_addr_t ua, user_addr_t ptr, int ptr_size) @@ -2077,7 +2849,7 @@ copyoutptr(user_addr_t ua, user_addr_t ptr, int ptr_size) if (ptr_size == 4) { /* 64 bit value containing 32 bit address */ - unsigned int i = CAST_DOWN(unsigned int,ua); /* SAFE */ + unsigned int i = CAST_DOWN_EXPLICIT(unsigned int,ua); /* SAFE */ error = copyout(&i, ptr, 4); } else { @@ -2105,85 +2877,156 @@ copyoutptr(user_addr_t ua, user_addr_t ptr, int ptr_size) * Note: The strings segment layout is backward, from the beginning * of the top of the stack to consume the minimal amount of * space possible; the returned stack pointer points to the - * end of the area consumed (stacks grow upward). + * end of the area consumed (stacks grow downward). * * argc is an int; arg[i] are pointers; env[i] are pointers; - * exec_path is a pointer; the 0's are (void *)NULL's + * the 0's are (void *)NULL's * * The stack frame layout is: * - * +-------------+ - * sp-> | argc | - * +-------------+ - * | arg[0] | - * +-------------+ - * : - * : - * +-------------+ - * | arg[argc-1] | - * +-------------+ - * | 0 | - * +-------------+ - * | env[0] | - * +-------------+ - * : - * : - * +-------------+ - * | env[n] | - * +-------------+ - * | 0 | - * +-------------+ - * | exec_path | In MacOS X PR2 Beaker2E the path passed to exec() is - * +-------------+ passed on the stack just after the trailing 0 of the - * | 0 | the envp[] array as a pointer to a string. - * +-------------+ - * | PATH AREA | - * +-------------+ - * | STRING AREA | - * : - * : - * | | <- p->user_stack - * +-------------+ + * +-------------+ <- p->user_stack + * | 16b | + * +-------------+ + * | STRING AREA | + * | : | + * | : | + * | : | + * +- -- -- -- --+ + * | PATH AREA | + * +-------------+ + * | 0 | + * +-------------+ + * | applev[n] | + * +-------------+ + * : + * : + * +-------------+ + * | applev[1] | + * +-------------+ + * | exec_path / | + * | applev[0] | + * +-------------+ + * | 0 | + * +-------------+ + * | env[n] | + * +-------------+ + * : + * : + * +-------------+ + * | env[0] | + * +-------------+ + * | 0 | + * +-------------+ + * | arg[argc-1] | + * +-------------+ + * : + * : + * +-------------+ + * | arg[0] | + * +-------------+ + * | argc | + * sp-> +-------------+ * * Although technically a part of the STRING AREA, we treat the PATH AREA as * a separate entity. This allows us to align the beginning of the PATH AREA * to a pointer boundary so that the exec_path, env[i], and argv[i] pointers * which preceed it on the stack are properly aligned. - * - * TODO: argc copied with suword(), which takes a 64 bit address */ + static int exec_copyout_strings(struct image_params *imgp, user_addr_t *stackp) { proc_t p = vfs_context_proc(imgp->ip_vfs_context); int ptr_size = (imgp->ip_flags & IMGPF_IS_64BIT) ? 8 : 4; - char *argv = imgp->ip_argv; /* modifiable copy of argv */ + int ptr_area_size; + void *ptr_buffer_start, *ptr_buffer; + int string_size; + user_addr_t string_area; /* *argv[], *env[] */ - user_addr_t path_area; /* package launch path */ - user_addr_t ptr_area; /* argv[], env[], exec_path */ + user_addr_t ptr_area; /* argv[], env[], applev[] */ + user_addr_t argc_area; /* argc */ user_addr_t stack; - int stringc = imgp->ip_argc + imgp->ip_envc; - int len; int error; - int strspace; + + unsigned i; + struct copyout_desc { + char *start_string; + int count; +#if CONFIG_DTRACE + user_addr_t *dtrace_cookie; +#endif + boolean_t null_term; + } descriptors[] = { + { + .start_string = imgp->ip_startargv, + .count = imgp->ip_argc, +#if CONFIG_DTRACE + .dtrace_cookie = &p->p_dtrace_argv, +#endif + .null_term = TRUE + }, + { + .start_string = imgp->ip_endargv, + .count = imgp->ip_envc, +#if CONFIG_DTRACE + .dtrace_cookie = &p->p_dtrace_envp, +#endif + .null_term = TRUE + }, + { + .start_string = imgp->ip_strings, + .count = 1, +#if CONFIG_DTRACE + .dtrace_cookie = NULL, +#endif + .null_term = FALSE + }, + { + .start_string = imgp->ip_endenvv, + .count = imgp->ip_applec - 1, /* exec_path handled above */ +#if CONFIG_DTRACE + .dtrace_cookie = NULL, +#endif + .null_term = TRUE + } + }; stack = *stackp; - unsigned patharea_len = imgp->ip_argv - imgp->ip_strings; - int envc_add = 0; - /* - * Set up pointers to the beginning of the string area, the beginning - * of the path area, and the beginning of the pointer area (actually, - * the location of argc, an int, which may be smaller than a pointer, - * but we use ptr_size worth of space for it, for alignment). + * All previous contributors to the string area + * should have aligned their sub-area */ - string_area = stack - (((imgp->ip_strendp - imgp->ip_strings) + ptr_size-1) & ~(ptr_size-1)) - ptr_size; - path_area = string_area - ((patharea_len + ptr_size-1) & ~(ptr_size-1)); - ptr_area = path_area - ((imgp->ip_argc + imgp->ip_envc + 4 + envc_add) * ptr_size) - ptr_size /*argc*/; + if (imgp->ip_strspace % ptr_size != 0) { + error = EINVAL; + goto bad; + } - /* Return the initial stack address: the location of argc */ - *stackp = ptr_area; + /* Grow the stack down for the strings we've been building up */ + string_size = imgp->ip_strendp - imgp->ip_strings; + stack -= string_size; + string_area = stack; + + /* + * Need room for one pointer for each string, plus + * one for the NULLs terminating the argv, envv, and apple areas. + */ + ptr_area_size = (imgp->ip_argc + imgp->ip_envc + imgp->ip_applec + 3) * + ptr_size; + stack -= ptr_area_size; + ptr_area = stack; + + /* We'll construct all the pointer arrays in our string buffer, + * which we already know is aligned properly, and ip_argspace + * was used to verify we have enough space. + */ + ptr_buffer_start = ptr_buffer = (void *)imgp->ip_strendp; + + /* + * Need room for pointer-aligned argc slot. + */ + stack -= ptr_size; + argc_area = stack; /* * Record the size of the arguments area so that sysctl_procargs() @@ -2191,92 +3034,73 @@ exec_copyout_strings(struct image_params *imgp, user_addr_t *stackp) */ proc_lock(p); p->p_argc = imgp->ip_argc; - p->p_argslen = (int)(stack - path_area); + p->p_argslen = (int)(*stackp - string_area); proc_unlock(p); + /* Return the initial stack address: the location of argc */ + *stackp = stack; /* - * Support for new app package launching for Mac OS X allocates - * the "path" at the begining of the imgp->ip_strings buffer. - * copy it just before the string area. + * Copy out the entire strings area. */ - len = 0; - error = copyoutstr(imgp->ip_strings, path_area, - patharea_len, - (size_t *)&len); + error = copyout(imgp->ip_strings, string_area, + string_size); if (error) goto bad; + for (i = 0; i < sizeof(descriptors)/sizeof(descriptors[0]); i++) { + char *cur_string = descriptors[i].start_string; + int j; - /* Save a NULL pointer below it */ - (void)copyoutptr(0LL, path_area - ptr_size, ptr_size); - - /* Save the pointer to "path" just below it */ - (void)copyoutptr(path_area, path_area - 2*ptr_size, ptr_size); +#if CONFIG_DTRACE + if (descriptors[i].dtrace_cookie) { + proc_lock(p); + *descriptors[i].dtrace_cookie = ptr_area + ((uintptr_t)ptr_buffer - (uintptr_t)ptr_buffer_start); /* dtrace convenience */ + proc_unlock(p); + } +#endif /* CONFIG_DTRACE */ - /* - * ptr_size for 2 NULL one each ofter arg[argc -1] and env[n] - * ptr_size for argc - * skip over saved path, ptr_size for pointer to path, - * and ptr_size for the NULL after pointer to path. - */ + /* + * For each segment (argv, envv, applev), copy as many pointers as requested + * to our pointer buffer. + */ + for (j = 0; j < descriptors[i].count; j++) { + user_addr_t cur_address = string_area + (cur_string - imgp->ip_strings); + + /* Copy out the pointer to the current string. Alignment has been verified */ + if (ptr_size == 8) { + *(uint64_t *)ptr_buffer = (uint64_t)cur_address; + } else { + *(uint32_t *)ptr_buffer = (uint32_t)cur_address; + } + + ptr_buffer = (void *)((uintptr_t)ptr_buffer + ptr_size); + cur_string += strlen(cur_string) + 1; /* Only a NUL between strings in the same area */ + } - /* argc (int32, stored in a ptr_size area) */ - (void)suword(ptr_area, imgp->ip_argc); - ptr_area += sizeof(int); - /* pad to ptr_size, if 64 bit image, to ensure user stack alignment */ - if (imgp->ip_flags & IMGPF_IS_64BIT) { - (void)suword(ptr_area, 0); /* int, not long: ignored */ - ptr_area += sizeof(int); + if (descriptors[i].null_term) { + if (ptr_size == 8) { + *(uint64_t *)ptr_buffer = 0ULL; + } else { + *(uint32_t *)ptr_buffer = 0; + } + + ptr_buffer = (void *)((uintptr_t)ptr_buffer + ptr_size); + } } -#if CONFIG_DTRACE - p->p_dtrace_argv = ptr_area; /* user_addr_t &argv[0] for dtrace convenience */ -#endif /* CONFIG_DTRACE */ - /* - * We use (string_area - path_area) here rather than the more - * intuitive (imgp->ip_argv - imgp->ip_strings) because we are - * interested in the length of the PATH_AREA in user space, - * rather than the actual length of the execution path, since - * it includes alignment padding of the PATH_AREA + STRING_AREA - * to a ptr_size boundary. + * Copy out all our pointer arrays in bulk. */ - strspace = SIZE_IMG_STRSPACE - (string_area - path_area); - for (;;) { - if (stringc == imgp->ip_envc) { - /* argv[n] = NULL */ - (void)copyoutptr(0LL, ptr_area, ptr_size); - ptr_area += ptr_size; -#if CONFIG_DTRACE - p->p_dtrace_envp = ptr_area; /* user_addr_t &env[0] for dtrace convenience */ -#endif /* CONFIG_DTRACE */ - } - if (--stringc < 0) - break; - - /* pointer: argv[n]/env[n] */ - (void)copyoutptr(string_area, ptr_area, ptr_size); + error = copyout(ptr_buffer_start, ptr_area, + ptr_area_size); + if (error) + goto bad; - /* string : argv[n][]/env[n][] */ - do { - if (strspace <= 0) { - error = E2BIG; - break; - } - error = copyoutstr(argv, string_area, - (unsigned)strspace, - (size_t *)&len); - string_area += len; - argv += len; - strspace -= len; - } while (error == ENAMETOOLONG); - if (error == EFAULT || error == E2BIG) - break; /* bad stack - user's problem */ - ptr_area += ptr_size; - } - /* env[n] = NULL */ - (void)copyoutptr(0LL, ptr_area, ptr_size); + /* argc (int32, stored in a ptr_size area) */ + error = copyoutptr((user_addr_t)imgp->ip_argc, argc_area, ptr_size); + if (error) + goto bad; bad: return(error); @@ -2289,6 +3113,11 @@ bad: * Copy arguments and environment from user space into work area; we may * have already copied some early arguments into the work area, and if * so, any arguments opied in are appended to those already there. + * This function is the primary manipulator of ip_argspace, since + * these are the arguments the client of execve(2) knows about. After + * each argv[]/envv[] string is copied, we charge the string length + * and argv[]/envv[] pointer slot to ip_argspace, so that we can + * full preflight the arg list size. * * Parameters: struct image_params * the image parameter block * @@ -2298,6 +3127,8 @@ bad: * Implicit returns; * (imgp->ip_argc) Count of arguments, updated * (imgp->ip_envc) Count of environment strings, updated + * (imgp->ip_argspace) Count of remaining of NCARGS + * (imgp->ip_interp_buffer) Interpreter and args (mutated in place) * * * Note: The argument and environment vectors are user space pointers @@ -2308,37 +3139,100 @@ exec_extract_strings(struct image_params *imgp) { int error = 0; int ptr_size = (imgp->ip_flags & IMGPF_WAS_64BIT) ? 8 : 4; + int new_ptr_size = (imgp->ip_flags & IMGPF_IS_64BIT) ? 8 : 4; user_addr_t argv = imgp->ip_user_argv; user_addr_t envv = imgp->ip_user_envv; /* - * If the argument vector is NULL, this is the system startup - * bootstrap from load_init_program(), and there's nothing to do + * Adjust space reserved for the path name by however much padding it + * needs. Doing this here since we didn't know if this would be a 32- + * or 64-bit process back in exec_save_path. */ - if (imgp->ip_user_argv == 0LL) - goto bad; - - /* Now, get rest of arguments */ + while (imgp->ip_strspace % new_ptr_size != 0) { + *imgp->ip_strendp++ = '\0'; + imgp->ip_strspace--; + /* imgp->ip_argspace--; not counted towards exec args total */ + } /* - * If we are running an interpreter, replace the av[0] that was - * passed to execve() with the fully qualified path name that was - * passed to execve() for interpreters which do not use the PATH - * to locate their script arguments. + * From now on, we start attributing string space to ip_argspace */ - if((imgp->ip_flags & IMGPF_INTERPRET) != 0 && argv != 0LL) { + imgp->ip_startargv = imgp->ip_strendp; + imgp->ip_argc = 0; + + if((imgp->ip_flags & IMGPF_INTERPRET) != 0) { user_addr_t arg; + char *argstart, *ch; + + /* First, the arguments in the "#!" string are tokenized and extracted. */ + argstart = imgp->ip_interp_buffer; + while (argstart) { + ch = argstart; + while (*ch && !IS_WHITESPACE(*ch)) { + ch++; + } - error = copyinptr(argv, &arg, ptr_size); - if (error) - goto bad; - if (arg != 0LL && arg != (user_addr_t)-1) { - argv += ptr_size; - error = exec_add_string(imgp, imgp->ip_user_fname); + if (*ch == '\0') { + /* last argument, no need to NUL-terminate */ + error = exec_add_user_string(imgp, CAST_USER_ADDR_T(argstart), UIO_SYSSPACE, TRUE); + argstart = NULL; + } else { + /* NUL-terminate */ + *ch = '\0'; + error = exec_add_user_string(imgp, CAST_USER_ADDR_T(argstart), UIO_SYSSPACE, TRUE); + + /* + * Find the next string. We know spaces at the end of the string have already + * been stripped. + */ + argstart = ch + 1; + while (IS_WHITESPACE(*argstart)) { + argstart++; + } + } + + /* Error-check, regardless of whether this is the last interpreter arg or not */ if (error) goto bad; + if (imgp->ip_argspace < new_ptr_size) { + error = E2BIG; + goto bad; + } + imgp->ip_argspace -= new_ptr_size; /* to hold argv[] entry */ imgp->ip_argc++; } + + if (argv != 0LL) { + /* + * If we are running an interpreter, replace the av[0] that was + * passed to execve() with the path name that was + * passed to execve() for interpreters which do not use the PATH + * to locate their script arguments. + */ + error = copyinptr(argv, &arg, ptr_size); + if (error) + goto bad; + if (arg != 0LL) { + argv += ptr_size; /* consume without using */ + } + } + + if (imgp->ip_interp_sugid_fd != -1) { + char temp[19]; /* "/dev/fd/" + 10 digits + NUL */ + snprintf(temp, sizeof(temp), "/dev/fd/%d", imgp->ip_interp_sugid_fd); + error = exec_add_user_string(imgp, CAST_USER_ADDR_T(temp), UIO_SYSSPACE, TRUE); + } else { + error = exec_add_user_string(imgp, imgp->ip_user_fname, imgp->ip_seg, TRUE); + } + + if (error) + goto bad; + if (imgp->ip_argspace < new_ptr_size) { + error = E2BIG; + goto bad; + } + imgp->ip_argspace -= new_ptr_size; /* to hold argv[] entry */ + imgp->ip_argc++; } while (argv != 0LL) { @@ -2348,23 +3242,37 @@ exec_extract_strings(struct image_params *imgp) if (error) goto bad; - argv += ptr_size; if (arg == 0LL) { break; - } else if (arg == (user_addr_t)-1) { - /* Um... why would it be -1? */ - error = EFAULT; - goto bad; } + + argv += ptr_size; + /* * av[n...] = arg[n] */ - error = exec_add_string(imgp, arg); + error = exec_add_user_string(imgp, arg, imgp->ip_seg, TRUE); if (error) goto bad; + if (imgp->ip_argspace < new_ptr_size) { + error = E2BIG; + goto bad; + } + imgp->ip_argspace -= new_ptr_size; /* to hold argv[] entry */ imgp->ip_argc++; } + /* Save space for argv[] NULL terminator */ + if (imgp->ip_argspace < new_ptr_size) { + error = E2BIG; + goto bad; + } + imgp->ip_argspace -= new_ptr_size; + + /* Note where the args ends and env begins. */ + imgp->ip_endargv = imgp->ip_strendp; + imgp->ip_envc = 0; + /* Now, get the environment */ while (envv != 0LL) { user_addr_t env; @@ -2376,29 +3284,214 @@ exec_extract_strings(struct image_params *imgp) envv += ptr_size; if (env == 0LL) { break; - } else if (env == (user_addr_t)-1) { - error = EFAULT; - goto bad; } /* * av[n...] = env[n] */ - error = exec_add_string(imgp, env); + error = exec_add_user_string(imgp, env, imgp->ip_seg, TRUE); if (error) goto bad; + if (imgp->ip_argspace < new_ptr_size) { + error = E2BIG; + goto bad; + } + imgp->ip_argspace -= new_ptr_size; /* to hold envv[] entry */ imgp->ip_envc++; } + + /* Save space for envv[] NULL terminator */ + if (imgp->ip_argspace < new_ptr_size) { + error = E2BIG; + goto bad; + } + imgp->ip_argspace -= new_ptr_size; + + /* Align the tail of the combined argv+envv area */ + while (imgp->ip_strspace % new_ptr_size != 0) { + if (imgp->ip_argspace < 1) { + error = E2BIG; + goto bad; + } + *imgp->ip_strendp++ = '\0'; + imgp->ip_strspace--; + imgp->ip_argspace--; + } + + /* Note where the envv ends and applev begins. */ + imgp->ip_endenvv = imgp->ip_strendp; + + /* + * From now on, we are no longer charging argument + * space to ip_argspace. + */ + bad: return error; } +static char * +random_hex_str(char *str, int len, boolean_t embedNUL) +{ + uint64_t low, high, value; + int idx; + char digit; + + /* A 64-bit value will only take 16 characters, plus '0x' and NULL. */ + if (len > 19) + len = 19; + + /* We need enough room for at least 1 digit */ + if (len < 4) + return (NULL); + + low = random(); + 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++) { + digit = value & 0xf; + value = value >> 4; + if (digit < 10) + str[idx] = '0' + digit; + else + str[idx] = 'a' + (digit - 10); + } + str[idx] = '\0'; + return (str); +} + +/* + * Libc has an 8-element array set up for stack guard values. It only fills + * in one of those entries, and both gcc and llvm seem to use only a single + * 8-byte guard. Until somebody needs more than an 8-byte guard value, don't + * do the work to construct them. + */ +#define GUARD_VALUES 1 +#define GUARD_KEY "stack_guard=" + +/* + * System malloc needs some entropy when it is initialized. + */ +#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 + */ +static int +exec_add_apple_strings(struct image_params *imgp) +{ + int i, error; + 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), i == 0); + if (i) + (void)strlcat(guard_vec, ",", sizeof (guard_vec)); + (void)strlcat(guard_vec, guard, sizeof (guard_vec)); + } + + error = exec_add_user_string(imgp, CAST_USER_ADDR_T(guard_vec), UIO_SYSSPACE, FALSE); + if (error) + goto bad; + imgp->ip_applec++; + + /* + * Supply libc with entropy for system malloc. + */ + (void)strlcpy(entropy_vec, ENTROPY_KEY, sizeof(entropy_vec)); + for (i = 0; i < ENTROPY_VALUES; i++) { + random_hex_str(entropy, sizeof (entropy), FALSE); + if (i) + (void)strlcat(entropy_vec, ",", sizeof (entropy_vec)); + (void)strlcat(entropy_vec, entropy, sizeof (entropy_vec)); + } + + error = exec_add_user_string(imgp, CAST_USER_ADDR_T(entropy_vec), UIO_SYSSPACE, FALSE); + if (error) + goto bad; + imgp->ip_applec++; + + /* Align the tail of the combined applev area */ + while (imgp->ip_strspace % new_ptr_size != 0) { + *imgp->ip_strendp++ = '\0'; + imgp->ip_strspace--; + } + +bad: + return error; +} #define unix_stack_size(p) (p->p_rlimit[RLIMIT_STACK].rlim_cur) /* * exec_check_permissions * - * Decription: Verify that the file that is being attempted to be executed + * Description: Verify that the file that is being attempted to be executed * is in fact allowed to be executed based on it POSIX file * permissions and other access control criteria * @@ -2440,7 +3533,7 @@ exec_check_permissions(struct image_params *imgp) * will always succeed, and we don't want to happen unless the * file really is executable. */ - if ((vap->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0) + if (!vfs_authopaque(vnode_mount(vp)) && ((vap->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0)) return (EACCES); /* Disallow zero length files */ @@ -2450,13 +3543,16 @@ exec_check_permissions(struct image_params *imgp) imgp->ip_arch_offset = (user_size_t)0; imgp->ip_arch_size = vap->va_data_size; - /* Disable setuid-ness for traced programs or if MNT_NOSUID */ - if ((vp->v_mount->mnt_flag & MNT_NOSUID) || (p->p_lflag & P_LTRACED)) { - vap->va_mode &= ~(VSUID | VSGID); -#if CONFIG_MACF - imgp->ip_no_trans = 1; -#endif - } + /* Disable setuid-ness for traced programs or if MNT_NOSUID */ + 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); @@ -2484,18 +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 == (uint64_t)((u_long)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); @@ -2534,24 +3618,38 @@ exec_handle_sugid(struct image_params *imgp) kauth_cred_t cred = vfs_context_ucred(imgp->ip_vfs_context); proc_t p = vfs_context_proc(imgp->ip_vfs_context); int i; - int is_member = 0; + int leave_sugid_clear = 0; + int mac_reset_ipc = 0; int error = 0; - struct vnode *dev_null = NULLVP; #if CONFIG_MACF - kauth_cred_t my_cred; -#endif + int mac_transition, disjoint_cred = 0; + int label_update_return = 0; -#if CONFIG_MACF - int mac_transition; - mac_transition = mac_cred_check_label_update_execve(imgp->ip_vfs_context, imgp->ip_vp, - imgp->ip_scriptlabelp, imgp->ip_execlabelp, p); + /* + * Determine whether a call to update the MAC label will result in the + * credential changing. + * + * Note: MAC policies which do not actually end up modifying + * the label subsequently are strongly encouraged to + * return 0 for this check, since a non-zero answer will + * slow down the exec fast path for normal binaries. + */ + 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_px_smpx); #endif - OSBitAndAtomic(~((uint32_t)P_SUGID), (UInt32 *)&p->p_flag); + OSBitAndAtomic(~((uint32_t)P_SUGID), &p->p_flag); /* * Order of the following is important; group checks must go last, - * as we use the success of the 'is_member' check combined with the + * as we use the success of the 'ismember' check combined with the * failure of the explicit match to indicate that we will be setting * the egid of the process even though the new process did not * require VSUID/VSGID bits in order for it to set the new group as @@ -2565,12 +3663,14 @@ exec_handle_sugid(struct image_params *imgp) */ if (((imgp->ip_origvattr->va_mode & VSUID) != 0 && kauth_cred_getuid(cred) != imgp->ip_origvattr->va_uid) || + ((imgp->ip_origvattr->va_mode & VSGID) != 0 && + ((kauth_cred_ismember_gid(cred, imgp->ip_origvattr->va_gid, &leave_sugid_clear) || !leave_sugid_clear) || + (kauth_cred_getgid(cred) != imgp->ip_origvattr->va_gid)))) { + #if CONFIG_MACF - mac_transition || /* A policy wants to transition */ +/* label for MAC transition and neither VSUID nor VSGID */ +handle_mac_transition: #endif - ((imgp->ip_origvattr->va_mode & VSGID) != 0 && - ((kauth_cred_ismember_gid(cred, imgp->ip_origvattr->va_gid, &is_member) || !is_member) || - (cred->cr_gid != imgp->ip_origvattr->va_gid)))) { /* * Replace the credential with a copy of itself if euid or @@ -2590,128 +3690,177 @@ exec_handle_sugid(struct image_params *imgp) */ if (imgp->ip_origvattr->va_mode & VSUID) { p->p_ucred = kauth_cred_setresuid(p->p_ucred, KAUTH_UID_NONE, imgp->ip_origvattr->va_uid, imgp->ip_origvattr->va_uid, KAUTH_UID_NONE); + /* update cred on proc */ + PROC_UPDATE_CREDS_ONPROC(p); } if (imgp->ip_origvattr->va_mode & VSGID) { p->p_ucred = kauth_cred_setresgid(p->p_ucred, KAUTH_GID_NONE, imgp->ip_origvattr->va_gid, imgp->ip_origvattr->va_gid); + /* update cred on proc */ + PROC_UPDATE_CREDS_ONPROC(p); } #if CONFIG_MACF /* - * XXXMAC: In FreeBSD, we set P_SUGID on a MAC transition - * to protect against debuggers being attached by an - * insufficiently privileged process onto the result of - * a transition to a more privileged credential. This is - * too conservative on FreeBSD, but we need to do - * something similar here, or risk vulnerability. - * - * Before we make the call into the MAC policies, get a new + * If a policy has indicated that it will transition the label, + * before making the call into the MAC policies, get a new * duplicate credential, so they can modify it without * modifying any others sharing it. */ - if (mac_transition && !imgp->ip_no_trans) { + if (mac_transition) { kauth_proc_label_update_execve(p, - imgp->ip_vfs_context, - imgp->ip_vp, - imgp->ip_scriptlabelp, imgp->ip_execlabelp); - - my_cred = kauth_cred_proc_ref(p); - mac_task_label_update_cred(my_cred, p->task); - kauth_cred_unref(&my_cred); + imgp->ip_vfs_context, + imgp->ip_vp, + imgp->ip_arch_offset, + imgp->ip_scriptvp, + imgp->ip_scriptlabelp, + 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 + * set the P_SUGID bit. This protects + * against debuggers being attached by an + * insufficiently privileged process onto the + * result of a transition to a more privileged + * credential. + */ + leave_sugid_clear = 0; + } + + imgp->ip_mac_return = label_update_return; } -#endif + + 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); + +#endif /* CONFIG_MACF */ + /* - * 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. + * If 'leave_sugid_clear' is non-zero, then we passed the + * VSUID and MACF checks, and successfully determined that + * 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 or reset mach ports and fds + * on the basis of simply running this code. */ - if (current_task() == p->task) { + 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(current_thread()); + ipc_thread_reset((imgp->ip_new_thread != NULL) ? + imgp->ip_new_thread : current_thread()); } - /* - * If 'is_member' is non-zero, then we passed the VSUID and - * MACF checks, and successfully determined that the previous - * cred was a member of the VSGID group, but that it was not - * the default at the time of the execve. So we don't set the - * P_SUGID on the basis of simply running this code. - */ - if (!is_member) - OSBitOrAtomic(P_SUGID, (UInt32 *)&p->p_flag); + if (!leave_sugid_clear) { + /* + * Flag the process as setuid. + */ + OSBitOrAtomic(P_SUGID, &p->p_flag); - /* Cache the vnode for /dev/null the first time around */ - if (dev_null == NULLVP) { - struct nameidata nd1; + /* + * 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++) { - NDINIT(&nd1, LOOKUP, FOLLOW, UIO_SYSSPACE32, - CAST_USER_ADDR_T("/dev/null"), - imgp->ip_vfs_context); + if (p->p_fd->fd_ofiles[i] != NULL) + continue; - if ((error = vn_open(&nd1, FREAD, 0)) == 0) { - dev_null = nd1.ni_vp; /* - * vn_open returns with both a use_count - * and an io_count on the found vnode - * drop the io_count, but keep the use_count + * Do the kernel equivalent of + * + * if i == 0 + * (void) open("/dev/null", O_RDONLY); + * else + * (void) open("/dev/null", O_WRONLY); */ - vnode_put(nd1.ni_vp); - } - } - /* Radar 2261856; setuid security hole fix */ - /* Patch from OpenBSD: A. Ramesh */ - /* - * 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. - */ - if (dev_null != NULLVP) { - for (i = 0; i < 3; i++) { struct fileproc *fp; int indx; + int flag; + struct nameidata *ndp = NULL; - if (p->p_fd->fd_ofiles[i] != NULL) - continue; + if (i == 0) + flag = FREAD; + else + flag = FWRITE; - if ((error = falloc(p, &fp, &indx, imgp->ip_vfs_context)) != 0) + if ((error = falloc(p, + &fp, &indx, imgp->ip_vfs_context)) != 0) continue; - if ((error = vnode_ref_ext(dev_null, FREAD)) != 0) { + 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); + + if ((error = vn_open(ndp, flag, 0)) != 0) { fp_free(p, indx, fp); break; } - 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)dev_null; - + struct fileglob *fg = fp->f_fglob; + + fg->fg_flag = flag; + fg->fg_ops = &vnops; + fg->fg_data = ndp->ni_vp; + + vnode_put(ndp->ni_vp); + proc_fdlock(p); procfdtbl_releasefd(p, indx, NULL); fp_drop(p, indx, fp, 1); proc_fdunlock(p); + + FREE(ndp, M_TEMP); } - /* - * for now we need to drop the reference immediately - * since we don't have any mechanism in place to - * release it before starting to unmount "/dev" - * during a reboot/shutdown - */ - vnode_rele(dev_null); - dev_null = NULLVP; } } +#if CONFIG_MACF + else { + /* + * We are here because we were told that the MAC label will + * be transitioned, and the binary is not VSUID or VSGID; to + * deal with this case, we could either duplicate a lot of + * code, or we can indicate we want to default the P_SUGID + * bit clear and jump back up. + */ + if (mac_transition) { + leave_sugid_clear = 1; + goto handle_mac_transition; + } + } + +#endif /* CONFIG_MACF */ /* * Implement the semantic where the effective user and group become * the saved user and group in exec'ed programs. */ - p->p_ucred = kauth_cred_setsvuidgid(p->p_ucred, kauth_cred_getuid(p->p_ucred), p->p_ucred->cr_gid); + p->p_ucred = kauth_cred_setsvuidgid(p->p_ucred, kauth_cred_getuid(p->p_ucred), kauth_cred_getgid(p->p_ucred)); + /* update cred on proc */ + PROC_UPDATE_CREDS_ONPROC(p); - /* XXX Obsolete; security token should not be separate from cred */ + /* Update the process' identity version and set the security token */ + p->p_idversion++; set_security_token(p); return(error); @@ -2729,72 +3878,90 @@ exec_handle_sugid(struct image_params *imgp) * 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 @@ -2814,75 +3981,77 @@ struct execve_args init_exec_args; void load_init_program(proc_t p) { - vm_offset_t init_addr; - int argc = 0; - char *argv[3]; + vm_offset_t init_addr, addr; + int argc; + uint32_t argv[3]; + unsigned int i; int error; - register_t retval[2]; - - /* - * Copy out program name. - */ + int retval[2]; + 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++] = (char *) 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++] = (char *)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] = NULL; - - /* - * 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\n", - 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); } /* @@ -2926,6 +4095,7 @@ load_return_to_errno(load_return_t lrtn) case LOAD_IOERROR: return EIO; case LOAD_FAILURE: + case LOAD_DECRYPTFAIL: default: return EBADEXEC; } @@ -2938,8 +4108,6 @@ load_return_to_errno(load_return_t lrtn) #include #include -extern semaphore_t execve_semaphore; - /* * execargs_alloc * @@ -2972,30 +4140,110 @@ extern semaphore_t execve_semaphore; * not modified its environment, we can't really know that it's * really a block there as well. */ + + +static int execargs_waiters = 0; +lck_mtx_t *execargs_cache_lock; + +static void +execargs_lock_lock(void) { + lck_mtx_lock_spin(execargs_cache_lock); +} + +static void +execargs_lock_unlock(void) { + lck_mtx_unlock(execargs_cache_lock); +} + +static wait_result_t +execargs_lock_sleep(void) { + return(lck_mtx_sleep(execargs_cache_lock, LCK_SLEEP_DEFAULT, &execargs_free_count, THREAD_INTERRUPTIBLE)); +} + +static kern_return_t +execargs_purgeable_allocate(char **execarg_address) { + kern_return_t kr = vm_allocate(bsd_pageable_map, (vm_offset_t *)execarg_address, BSD_PAGEABLE_SIZE_PER_EXEC, VM_FLAGS_ANYWHERE | VM_FLAGS_PURGABLE); + assert(kr == KERN_SUCCESS); + return kr; +} + +static kern_return_t +execargs_purgeable_reference(void *execarg_address) { + int state = VM_PURGABLE_NONVOLATILE; + kern_return_t kr = vm_purgable_control(bsd_pageable_map, (vm_offset_t) execarg_address, VM_PURGABLE_SET_STATE, &state); + + assert(kr == KERN_SUCCESS); + return kr; +} + +static kern_return_t +execargs_purgeable_volatilize(void *execarg_address) { + int state = VM_PURGABLE_VOLATILE | VM_PURGABLE_ORDERING_OBSOLETE; + kern_return_t kr; + kr = vm_purgable_control(bsd_pageable_map, (vm_offset_t) execarg_address, VM_PURGABLE_SET_STATE, &state); + + assert(kr == KERN_SUCCESS); + + return kr; +} + +static void +execargs_wakeup_waiters(void) { + thread_wakeup(&execargs_free_count); +} + static int execargs_alloc(struct image_params *imgp) { kern_return_t kret; + wait_result_t res; + int i, cache_index = -1; - kret = semaphore_wait(execve_semaphore); - if (kret != KERN_SUCCESS) - switch (kret) { - default: - return (EINVAL); - case KERN_INVALID_ADDRESS: - case KERN_PROTECTION_FAILURE: - return (EACCES); - case KERN_ABORTED: - case KERN_OPERATION_TIMED_OUT: + execargs_lock_lock(); + + while (execargs_free_count == 0) { + execargs_waiters++; + res = execargs_lock_sleep(); + execargs_waiters--; + if (res != THREAD_AWAKENED) { + execargs_lock_unlock(); return (EINTR); } + } + + execargs_free_count--; + + for (i = 0; i < execargs_cache_size; i++) { + vm_offset_t element = execargs_cache[i]; + if (element) { + cache_index = i; + imgp->ip_strings = (char *)(execargs_cache[i]); + execargs_cache[i] = 0; + break; + } + } + + assert(execargs_free_count >= 0); + + execargs_lock_unlock(); + + if (cache_index == -1) { + kret = execargs_purgeable_allocate(&imgp->ip_strings); + } + else + kret = execargs_purgeable_reference(imgp->ip_strings); - kret = kmem_alloc_pageable(bsd_pageable_map, (vm_offset_t *)&imgp->ip_strings, NCARGS + PAGE_SIZE); - imgp->ip_vdata = imgp->ip_strings + NCARGS; + assert(kret == KERN_SUCCESS); if (kret != KERN_SUCCESS) { - semaphore_signal(execve_semaphore); return (ENOMEM); } + + /* last page used to read in file headers */ + imgp->ip_vdata = imgp->ip_strings + ( NCARGS + PAGE_SIZE ); + imgp->ip_strendp = imgp->ip_strings; + imgp->ip_argspace = NCARGS; + imgp->ip_strspace = ( NCARGS + PAGE_SIZE ); + return (0); } @@ -3016,23 +4264,34 @@ static int execargs_free(struct image_params *imgp) { kern_return_t kret; + int i; + boolean_t needs_wakeup = FALSE; + + kret = execargs_purgeable_volatilize(imgp->ip_strings); - kmem_free(bsd_pageable_map, (vm_offset_t)imgp->ip_strings, NCARGS + PAGE_SIZE); - imgp->ip_strings = NULL; - - kret = semaphore_signal(execve_semaphore); - switch (kret) { - case KERN_INVALID_ADDRESS: - case KERN_PROTECTION_FAILURE: - return (EINVAL); - case KERN_ABORTED: - case KERN_OPERATION_TIMED_OUT: - return (EINTR); - case KERN_SUCCESS: - return(0); - default: - return (EINVAL); + execargs_lock_lock(); + execargs_free_count++; + + for (i = 0; i < execargs_cache_size; i++) { + vm_offset_t element = execargs_cache[i]; + if (element == 0) { + execargs_cache[i] = (vm_offset_t) imgp->ip_strings; + imgp->ip_strings = NULL; + break; + } } + + assert(imgp->ip_strings == NULL); + + if (execargs_waiters > 0) + needs_wakeup = TRUE; + + execargs_lock_unlock(); + + if (needs_wakeup == TRUE) + execargs_wakeup_waiters(); + + return ((kret == KERN_SUCCESS ? 0 : EINVAL)); } static void @@ -3069,3 +4328,295 @@ exec_resettextvp(proc_t p, struct image_params *imgp) } +/* + * 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) +{ + 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 + * ones for the process that is being successfully + * loaded + */ + proc_lock(p); + p->p_csflags = imgp->ip_csflags; + proc_unlock(p); + + /* Set the switch_protect flag on the map */ + if(p->p_csflags & (CS_HARD|CS_KILL)) { + vm_map_switch_protect(get_task_map(p->task), TRUE); + } + + /* + * image activation may be failed due to policy + * which is unexpected but security framework does not + * approve of exec, kill and return immediately. + */ + if (imgp->ip_mac_return != 0) { + error = imgp->ip_mac_return; + unexpected_failure = TRUE; + goto done; + } + + /* check if callout to taskgated is needed */ + if (!taskgated_required(p, &require_success)) { + error = 0; + goto done; + } + + kr = task_get_task_access_port(p->task, &port); + if (KERN_SUCCESS != kr || !IPC_PORT_VALID(port)) { + error = 0; + if (require_success) + error = EACCES; + goto done; + } + + /* + * 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; + } + + /* Only do this if exec_resettextvp() did not fail */ + if (p->p_textvp != NULLVP) { + /* + * If there's a new code directory, mark this process + * as signed. + */ + if (0 == ubc_cs_getcdhash(p->p_textvp, p->p_textoff, hash)) { + proc_lock(p); + p->p_csflags |= CS_VALID; + proc_unlock(p); + } + } + +done: + if (0 != error) { + if (!unexpected_failure) + p->p_csflags |= CS_KILLED; + /* make very sure execution fails */ + 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); + } + } +}