X-Git-Url: https://git.saurik.com/apple/xnu.git/blobdiff_plain/d1ecb069dfe24481e4a83f44cb5217a2b06746d7..99c3a10404e5d1ef94397ab4df5a8b74711fc4d3:/bsd/kern/bsd_init.c diff --git a/bsd/kern/bsd_init.c b/bsd/kern/bsd_init.c index e2868a40c..2a04fff66 100644 --- a/bsd/kern/bsd_init.c +++ b/bsd/kern/bsd_init.c @@ -106,6 +106,7 @@ #include #include #include +#include #include @@ -132,10 +133,12 @@ #include /* for mcache_init() */ #include /* for mbinit() */ #include /* for knote_init() */ +#include /* for memorystatus_init() */ #include /* for aio_init() */ #include /* for psem_cache_init() */ #include /* for dlil_init() */ #include /* for proto_kpi_init() */ +#include /* for iptap_init() */ #include /* for pipeinit() */ #include /* for socketinit() */ #include /* for domaininit() */ @@ -151,6 +154,9 @@ #include /* for tty_init() */ #include /* for utun_register_control() */ #include /* for net_str_id_init() */ +#include /* for netsrc_init() */ +#include /* for nstat_init() */ +#include /* for assert() */ #include @@ -162,6 +168,10 @@ #include +#if NFSCLIENT +#include +#endif + #if CONFIG_IMAGEBOOT #include #endif @@ -171,12 +181,13 @@ #endif #include +#include +#include void * get_user_regs(thread_t); /* XXX kludge for */ void IOKitInitializeTime(void); /* XXX */ void IOSleep(unsigned int); /* XXX */ void loopattach(void); /* XXX */ -void vc_progress_set(boolean_t, uint32_t); /* XXX */ const char copyright[] = "Copyright (c) 1982, 1986, 1989, 1991, 1993\n\t" @@ -214,14 +225,6 @@ char hostname[MAXHOSTNAMELEN]; int hostnamelen; char domainname[MAXDOMNAMELEN]; int domainnamelen; -#if defined(__i386__) || defined(__x86_64__) -struct exec_archhandler exec_archhandler_ppc = { - .path = "/usr/libexec/oah/translate", -}; -const char * const kRosettaStandIn_str = "/usr/libexec/oah/RosettaNonGrata"; -#else /* __i386__ */ -struct exec_archhandler exec_archhandler_ppc; -#endif /* __i386__ */ char rootdevice[16]; /* hfs device names have at least 9 chars */ @@ -242,26 +245,30 @@ extern void klogwakeup(void); extern void file_lock_init(void); extern void kmeminit(void); extern void bsd_bufferinit(void); +extern void throttle_init(void); -extern int srv; +extern int serverperfmode; extern int ncl; vm_map_t bsd_pageable_map; vm_map_t mb_map; -static int bsd_simul_execs = BSD_SIMUL_EXECS; -static int bsd_pageable_map_size = BSD_PAGABLE_MAP_SIZE; -__private_extern__ int execargs_cache_size = BSD_SIMUL_EXECS; -__private_extern__ int execargs_free_count = BSD_SIMUL_EXECS; +static int bsd_simul_execs; +static int bsd_pageable_map_size; +__private_extern__ int execargs_cache_size = 0; +__private_extern__ int execargs_free_count = 0; __private_extern__ vm_offset_t * execargs_cache = NULL; -void bsd_exec_setup(int); +void bsd_exec_setup(int) __attribute__((aligned(4096))); + +__private_extern__ int bootarg_vnode_cache_defeat = 0; /* - * Set to disable grading 64 bit Mach-o binaries as executable, for testing; - * Intel only. + * Prevent kernel-based ASLR from being used, for testing. */ -__private_extern__ int bootarg_no64exec = 0; +#if DEVELOPMENT || DEBUG +__private_extern__ int bootarg_disable_aslr = 0; +#endif int cmask = CMASK; extern int customnbuf; @@ -274,6 +281,7 @@ static void parse_bsd_args(void); extern task_t bsd_init_task; extern char init_task_failure_data[]; extern void time_zone_slock_init(void); +extern void select_wait_queue_init(void); static void process_name(const char *, proc_t); static void setconf(void); @@ -289,17 +297,21 @@ extern void sysv_sem_lock_init(void); #if SYSV_MSG extern void sysv_msg_lock_init(void); #endif -extern void pthread_init(void); +#if !defined(SECURE_KERNEL) /* kmem access not enabled by default; can be changed with boot-args */ +/* We don't need to keep this symbol around in RELEASE kernel */ int setup_kmem = 0; +#endif -/* size of kernel trace buffer, disabled by default */ -unsigned int new_nkdbufs = 0; +#if CONFIG_MACF +#if defined (__i386__) || defined (__x86_64__) +/* MACF policy_check configuration flags; see policy_check.c for details */ +int policy_check_flags = 0; -/* mach leak logging */ -int log_leaks = 0; -int turn_on_log_leaks = 0; +extern int check_policy_init(int); +#endif +#endif /* CONFIG_MACF */ extern void stackshot_lock_init(void); @@ -309,7 +321,6 @@ extern void stackshot_lock_init(void); extern void dtrace_postinit(void); #endif - /* * Initialization code. * Called from cold start routine as @@ -343,8 +354,6 @@ struct rlimit vm_initial_limit_core = { DFLCSIZ, MAXCSIZ }; extern thread_t cloneproc(task_t, proc_t, int); extern int (*mountroot)(void); -extern int netboot_mountroot(void); /* netboot.c */ -extern int netboot_setup(void); lck_grp_t * proc_lck_grp; lck_grp_t * proc_slock_grp; @@ -375,20 +384,27 @@ void (*unmountroot_pre_hook)(void); * of the uu_context.vc_ucred field so that the uthread structure can be * used like any other. */ +extern void run_bringup_tests(void); + +extern void IOServicePublishResource(const char *, boolean_t); + void bsd_init(void) { struct uthread *ut; unsigned int i; -#if __i386__ || __x86_64__ - int error; -#endif struct vfs_context context; kern_return_t ret; struct ucred temp_cred; + struct posix_cred temp_pcred; +#if NFSCLIENT || CONFIG_IMAGEBOOT + boolean_t netboot = FALSE; +#endif #define bsd_init_kprintf(x...) /* kprintf("bsd_init: " x) */ + throttle_init(); + kernel_flock = funnel_alloc(KERNEL_FUNNEL); if (kernel_flock == (funnel_t *)0 ) { panic("bsd_init: Failed to allocate kernel funnel"); @@ -427,7 +443,7 @@ bsd_init(void) proc_lck_grp_attr= lck_grp_attr_alloc_init(); proc_lck_grp = lck_grp_alloc_init("proc", proc_lck_grp_attr); -#ifndef CONFIG_EMBEDDED +#if CONFIG_FINE_LOCK_GROUPS proc_slock_grp = lck_grp_alloc_init("proc-slock", proc_lck_grp_attr); proc_fdmlock_grp = lck_grp_alloc_init("proc-fdmlock", proc_lck_grp_attr); proc_mlock_grp = lck_grp_alloc_init("proc-mlock", proc_lck_grp_attr); @@ -440,20 +456,21 @@ bsd_init(void) #endif #endif -#ifdef CONFIG_EMBEDDED - proc_list_mlock = lck_mtx_alloc_init(proc_lck_grp, proc_lck_attr); - proc_klist_mlock = lck_mtx_alloc_init(proc_lck_grp, proc_lck_attr); - lck_mtx_init(&kernproc->p_mlock, proc_lck_grp, proc_lck_attr); - lck_mtx_init(&kernproc->p_fdmlock, proc_lck_grp, proc_lck_attr); - lck_spin_init(&kernproc->p_slock, proc_lck_grp, proc_lck_attr); -#else +#if CONFIG_FINE_LOCK_GROUPS proc_list_mlock = lck_mtx_alloc_init(proc_mlock_grp, proc_lck_attr); proc_klist_mlock = lck_mtx_alloc_init(proc_mlock_grp, proc_lck_attr); lck_mtx_init(&kernproc->p_mlock, proc_mlock_grp, proc_lck_attr); lck_mtx_init(&kernproc->p_fdmlock, proc_fdmlock_grp, proc_lck_attr); lck_spin_init(&kernproc->p_slock, proc_slock_grp, proc_lck_attr); +#else + proc_list_mlock = lck_mtx_alloc_init(proc_lck_grp, proc_lck_attr); + proc_klist_mlock = lck_mtx_alloc_init(proc_lck_grp, proc_lck_attr); + lck_mtx_init(&kernproc->p_mlock, proc_lck_grp, proc_lck_attr); + lck_mtx_init(&kernproc->p_fdmlock, proc_lck_grp, proc_lck_attr); + lck_spin_init(&kernproc->p_slock, proc_lck_grp, proc_lck_attr); #endif + assert(bsd_simul_execs != 0); execargs_cache_lock = lck_mtx_alloc_init(proc_lck_grp, proc_lck_attr); execargs_cache_size = bsd_simul_execs; execargs_free_count = bsd_simul_execs; @@ -473,6 +490,14 @@ bsd_init(void) */ mac_policy_initbsd(); kernproc->p_mac_enforce = 0; + +#if defined (__i386__) || defined (__x86_64__) + /* + * We currently only support this on i386/x86_64, as that is the + * only lock code we have instrumented so far. + */ + check_policy_init(policy_check_flags); +#endif #endif /* MAC */ /* @@ -483,15 +508,16 @@ bsd_init(void) kernproc->p_pgrp = &pgrp0; LIST_INSERT_HEAD(PGRPHASH(0), &pgrp0, pg_hash); LIST_INIT(&pgrp0.pg_members); -#ifdef CONFIG_EMBEDDED - lck_mtx_init(&pgrp0.pg_mlock, proc_lck_grp, proc_lck_attr); -#else +#ifdef CONFIG_FINE_LOCK_GROUPS lck_mtx_init(&pgrp0.pg_mlock, proc_mlock_grp, proc_lck_attr); +#else + lck_mtx_init(&pgrp0.pg_mlock, proc_lck_grp, proc_lck_attr); #endif /* There is no other bsd thread this point and is safe without pgrp lock */ LIST_INSERT_HEAD(&pgrp0.pg_members, kernproc, p_pglist); kernproc->p_listflag |= P_LIST_INPGRP; kernproc->p_pgrpid = 0; + kernproc->p_uniqueid = 0; pgrp0.pg_session = &session0; pgrp0.pg_membercnt = 1; @@ -499,10 +525,10 @@ bsd_init(void) session0.s_count = 1; session0.s_leader = kernproc; session0.s_listflags = 0; -#ifdef CONFIG_EMBEDDED - lck_mtx_init(&session0.s_mlock, proc_lck_grp, proc_lck_attr); -#else +#ifdef CONFIG_FINE_LOCK_GROUPS lck_mtx_init(&session0.s_mlock, proc_mlock_grp, proc_lck_attr); +#else + lck_mtx_init(&session0.s_mlock, proc_lck_grp, proc_lck_attr); #endif LIST_INSERT_HEAD(SESSHASH(0), &session0, s_hash); proc_list_unlock(); @@ -515,6 +541,14 @@ bsd_init(void) kernproc->p_stat = SRUN; kernproc->p_flag = P_SYSTEM; + kernproc->p_lflag = 0; + kernproc->p_ladvflag = 0; + +#if DEVELOPMENT || DEBUG + if (bootarg_disable_aslr) + kernproc->p_flag |= P_DISABLE_ASLR; +#endif + kernproc->p_nice = NZERO; kernproc->p_pptr = kernproc; @@ -531,15 +565,22 @@ bsd_init(void) */ bsd_init_kprintf("calling bzero\n"); bzero(&temp_cred, sizeof(temp_cred)); - temp_cred.cr_ngroups = 1; + bzero(&temp_pcred, sizeof(temp_pcred)); + temp_pcred.cr_ngroups = 1; - temp_cred.cr_audit.as_aia_p = &audit_default_aia; - /* XXX the following will go away with cr_au */ - temp_cred.cr_au.ai_auid = AU_DEFAUDITID; + temp_cred.cr_audit.as_aia_p = audit_default_aia_p; bsd_init_kprintf("calling kauth_cred_create\n"); + /* + * We have to label the temp cred before we create from it to + * properly set cr_ngroups, or the create will fail. + */ + posix_cred_label(&temp_cred, &temp_pcred); kernproc->p_ucred = kauth_cred_create(&temp_cred); + /* update cred on proc */ + PROC_UPDATE_CREDS_ONPROC(kernproc); + /* give the (already exisiting) initial thread a reference on it */ bsd_init_kprintf("calling kauth_cred_ref\n"); kauth_cred_ref(kernproc->p_ucred); @@ -598,6 +639,7 @@ bsd_init(void) vm_offset_t minimum; bsd_init_kprintf("calling kmem_suballoc\n"); + assert(bsd_pageable_map_size != 0); ret = kmem_suballoc(kernel_map, &minimum, (vm_size_t)bsd_pageable_map_size, @@ -630,15 +672,15 @@ bsd_init(void) bsd_init_kprintf("calling IOKitInitializeTime\n"); IOKitInitializeTime(); - if (turn_on_log_leaks && !new_nkdbufs) - new_nkdbufs = 200000; - start_kern_tracing(new_nkdbufs); - if (turn_on_log_leaks) - log_leaks = 1; - bsd_init_kprintf("calling ubc_init\n"); ubc_init(); + /* + * Initialize device-switches. + */ + bsd_init_kprintf("calling devsw_init() \n"); + devsw_init(); + /* Initialize the file systems. */ bsd_init_kprintf("calling vfsinit\n"); vfsinit(); @@ -702,6 +744,8 @@ bsd_init(void) psem_cache_init(); bsd_init_kprintf("calling time_zone_slock_init\n"); time_zone_slock_init(); + bsd_init_kprintf("calling select_wait_queue_init\n"); + select_wait_queue_init(); /* Stack snapshot facility lock */ stackshot_lock_init(); @@ -724,16 +768,26 @@ bsd_init(void) socketinit(); bsd_init_kprintf("calling domaininit\n"); domaininit(); + iptap_init(); #endif /* SOCKETS */ kernproc->p_fd->fd_cdir = NULL; kernproc->p_fd->fd_rdir = NULL; -#if CONFIG_EMBEDDED - /* Initialize kernel memory status notifications */ - bsd_init_kprintf("calling kern_memorystatus_init\n"); - kern_memorystatus_init(); +#if CONFIG_FREEZE +#ifndef CONFIG_MEMORYSTATUS + #error "CONFIG_FREEZE defined without matching CONFIG_MEMORYSTATUS" #endif + /* Initialise background freezing */ + bsd_init_kprintf("calling memorystatus_freeze_init\n"); + memorystatus_freeze_init(); +#endif + +#if CONFIG_MEMORYSTATUS + /* Initialize kernel memory status notifications */ + bsd_init_kprintf("calling memorystatus_init\n"); + memorystatus_init(); +#endif /* CONFIG_MEMORYSTATUS */ #ifdef GPROF /* Initialize kernel profiling. */ @@ -780,6 +834,8 @@ bsd_init(void) /* register user tunnel kernel control handler */ utun_register_control(); + netsrc_init(); + nstat_init(); #endif /* NETWORKING */ bsd_init_kprintf("calling vnode_pager_bootstrap\n"); @@ -794,61 +850,22 @@ bsd_init(void) bsd_init_kprintf("calling inittodr\n"); inittodr(0); -#if CONFIG_EMBEDDED - { - /* print out early VM statistics */ - kern_return_t kr1; - vm_statistics_data_t stat; - mach_msg_type_number_t count; - - count = HOST_VM_INFO_COUNT; - kr1 = host_statistics(host_self(), - HOST_VM_INFO, - (host_info_t)&stat, - &count); - kprintf("Mach Virtual Memory Statistics (page size of 4096) bytes\n" - "Pages free:\t\t\t%u.\n" - "Pages active:\t\t\t%u.\n" - "Pages inactive:\t\t\t%u.\n" - "Pages wired down:\t\t%u.\n" - "\"Translation faults\":\t\t%u.\n" - "Pages copy-on-write:\t\t%u.\n" - "Pages zero filled:\t\t%u.\n" - "Pages reactivated:\t\t%u.\n" - "Pageins:\t\t\t%u.\n" - "Pageouts:\t\t\t%u.\n" - "Object cache: %u hits of %u lookups (%d%% hit rate)\n", - - stat.free_count, - stat.active_count, - stat.inactive_count, - stat.wire_count, - stat.faults, - stat.cow_faults, - stat.zero_fill_count, - stat.reactivations, - stat.pageins, - stat.pageouts, - stat.hits, - stat.lookups, - (stat.hits == 0) ? 100 : - ((stat.lookups * 100) / stat.hits)); - } -#endif /* CONFIG_EMBEDDED */ - /* Mount the root file system. */ while( TRUE) { int err; bsd_init_kprintf("calling setconf\n"); setconf(); +#if NFSCLIENT + netboot = (mountroot == netboot_mountroot); +#endif bsd_init_kprintf("vfs_mountroot\n"); if (0 == (err = vfs_mountroot())) break; rootdevice[0] = '\0'; #if NFSCLIENT - if (mountroot == netboot_mountroot) { + if (netboot) { PE_display_icon( 0, "noroot"); /* XXX a netboot-specific icon would be nicer */ vc_progress_set(FALSE, 0); for (i=1; 1; i*=2) { @@ -880,8 +897,10 @@ bsd_init(void) filedesc0.fd_cdir = rootvnode; #if NFSCLIENT - if (mountroot == netboot_mountroot) { + if (netboot) { int err; + + netboot = TRUE; /* post mount setup */ if ((err = netboot_setup()) != 0) { PE_display_icon( 0, "noroot"); /* XXX a netboot-specific icon would be nicer */ @@ -903,19 +922,12 @@ bsd_init(void) * See if a system disk image is present. If so, mount it and * switch the root vnode to point to it */ - - if(imageboot_needed()) { - int err; - - /* An image was found */ - if((err = imageboot_setup())) { - /* - * this is not fatal. Keep trying to root - * off the original media - */ - printf("%s: imageboot could not find root, %d\n", - __FUNCTION__, err); - } + if (netboot == FALSE && imageboot_needed()) { + /* + * An image was found. No turning back: we're booted + * with a kernel from the disk image. + */ + imageboot_setup(); } #endif /* CONFIG_IMAGEBOOT */ @@ -943,18 +955,8 @@ bsd_init(void) kernproc->p_flag |= P_LP64; printf("Kernel is LP64\n"); #endif -#if __i386__ || __x86_64__ - /* this should be done after the root filesystem is mounted */ - error = set_archhandler(kernproc, CPU_TYPE_POWERPC); - // 10/30/08 - gab: - // if default 'translate' can't be found, see if the understudy is available - if (ENOENT == error) { - strlcpy(exec_archhandler_ppc.path, kRosettaStandIn_str, MAXPATHLEN); - error = set_archhandler(kernproc, CPU_TYPE_POWERPC); - } - if (error) /* XXX make more generic */ - exec_archhandler_ppc.path[0] = 0; -#endif + + pal_kernel_announce(); bsd_init_kprintf("calling mountroot_post_hook\n"); @@ -1117,12 +1119,14 @@ parse_bsd_args(void) if (PE_parse_boot_argn("-x", namep, sizeof (namep))) /* safe boot */ boothowto |= RB_SAFEBOOT; - if (PE_parse_boot_argn("-l", namep, sizeof (namep))) /* leaks logging */ - turn_on_log_leaks = 1; + /* disable vnode_cache_is_authorized() by setting vnode_cache_defeat */ + if (PE_parse_boot_argn("-vnode_cache_defeat", namep, sizeof (namep))) + bootarg_vnode_cache_defeat = 1; - /* disable 64 bit grading */ - if (PE_parse_boot_argn("-no64exec", namep, sizeof (namep))) - bootarg_no64exec = 1; +#if DEVELOPMENT || DEBUG + if (PE_parse_boot_argn("-disable_aslr", namep, sizeof (namep))) + bootarg_disable_aslr = 1; +#endif PE_parse_boot_argn("ncl", &ncl, sizeof (ncl)); if (PE_parse_boot_argn("nbuf", &max_nbuf_headers, @@ -1132,11 +1136,20 @@ parse_bsd_args(void) #if !defined(SECURE_KERNEL) PE_parse_boot_argn("kmem", &setup_kmem, sizeof (setup_kmem)); #endif - PE_parse_boot_argn("trace", &new_nkdbufs, sizeof (new_nkdbufs)); + +#if CONFIG_MACF +#if defined (__i386__) || defined (__x86_64__) + PE_parse_boot_argn("policy_check", &policy_check_flags, sizeof (policy_check_flags)); +#endif +#endif /* CONFIG_MACF */ if (PE_parse_boot_argn("msgbuf", &msgbuf, sizeof (msgbuf))) { log_setsize(msgbuf); } + + if (PE_parse_boot_argn("-novfscache", namep, sizeof(namep))) { + nc_disabled = 1; + } } void @@ -1165,10 +1178,13 @@ bsd_exec_setup(int scale) break; } - bsd_pageable_map_size = (bsd_simul_execs * (NCARGS + PAGE_SIZE)); + bsd_pageable_map_size = (bsd_simul_execs * BSD_PAGEABLE_SIZE_PER_EXEC); } #if !NFSCLIENT +int +netboot_root(void); + int netboot_root(void) {