X-Git-Url: https://git.saurik.com/apple/xnu.git/blobdiff_plain/2d21ac55c334faf3a56e5634905ed6987fc787d4..5ba3f43ea354af8ad55bea84372a2bc834d8757c:/bsd/kern/kern_exit.c diff --git a/bsd/kern/kern_exit.c b/bsd/kern/kern_exit.c index 7bc8b1d74..b8f5def79 100644 --- a/bsd/kern/kern_exit.c +++ b/bsd/kern/kern_exit.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2000-2007 Apple Inc. All rights reserved. + * Copyright (c) 2000-2016 Apple Inc. All rights reserved. * * @APPLE_OSREFERENCE_LICENSE_HEADER_START@ * @@ -74,6 +74,7 @@ #include #include +#include #include "compat_43.h" @@ -94,77 +95,105 @@ #include #include #include +#include +#include +#include #include #include #include #include -#include /* fdfree */ -#if SYSV_SHM -#include /* shmexit */ -#endif -#include /* acct_process */ +#include +#include /* fdfree */ +#include /* acct_process */ +#include +#include /* kevent_proc_copy_uptrs */ +#include -#include +#include #include #include +#include +#include +#include #include #include #include +#include #include #include #include #include +#include +#include + +#include + +#include + +#if SYSV_SHM +#include /* shmexit */ +#endif /* SYSV_SHM */ +#if CONFIG_PERSONAS +#include +#endif /* CONFIG_PERSONAS */ +#if CONFIG_MEMORYSTATUS +#include +#endif /* CONFIG_MEMORYSTATUS */ #if CONFIG_DTRACE /* Do not include dtrace.h, it redefines kmem_[alloc/free] */ -extern void (*dtrace_fasttrap_exit_ptr)(proc_t); -extern void (*dtrace_helpers_cleanup)(proc_t); -extern void dtrace_lazy_dofs_destroy(proc_t); - +void dtrace_proc_exit(proc_t p); #include -#endif - +#endif /* CONFIG_DTRACE */ #if CONFIG_MACF -#include +#include +#include #include -#endif - -#include -#include -#include -#include /* init_process */ - -#include +#endif /* CONFIG_MACF */ -extern char init_task_failure_data[]; -void proc_prepareexit(proc_t p, int rv); +void proc_prepareexit(proc_t p, int rv, boolean_t perf_notify); +void gather_populate_corpse_crashinfo(proc_t p, task_t corpse_task, + mach_exception_data_type_t code, mach_exception_data_type_t subcode, + uint64_t *udata_buffer, int num_udata, void *reason); +mach_exception_data_type_t proc_encode_exit_exception_code(proc_t p); void vfork_exit(proc_t p, int rv); -void vproc_exit(proc_t p); -__private_extern__ void munge_rusage(struct rusage *a_rusage_p, struct user_rusage *a_user_rusage_p); -static int reap_child_locked(proc_t parent, proc_t child, int deadparent, int locked, int droplock); +__private_extern__ void munge_user64_rusage(struct rusage *a_rusage_p, struct user64_rusage *a_user_rusage_p); +__private_extern__ void munge_user32_rusage(struct rusage *a_rusage_p, struct user32_rusage *a_user_rusage_p); +static int reap_child_locked(proc_t parent, proc_t child, int deadparent, int reparentedtoinit, int locked, int droplock); +static void populate_corpse_crashinfo(proc_t p, task_t corpse_task, + struct rusage_superset *rup, mach_exception_data_type_t code, + mach_exception_data_type_t subcode, uint64_t *udata_buffer, + int num_udata, os_reason_t reason); +static void proc_update_corpse_exception_codes(proc_t p, mach_exception_data_type_t *code, mach_exception_data_type_t *subcode); +extern int proc_pidpathinfo_internal(proc_t p, uint64_t arg, char *buffer, uint32_t buffersize, int32_t *retval); +static __attribute__((noinline)) void launchd_crashed_panic(proc_t p, int rv); +extern void proc_piduniqidentifierinfo(proc_t p, struct proc_uniqidentifierinfo *p_uniqidinfo); +extern void task_coalition_ids(task_t task, uint64_t ids[COALITION_NUM_TYPES]); +extern uint64_t get_task_phys_footprint_limit(task_t); +int proc_list_uptrs(void *p, uint64_t *udata_buffer, int size); +extern uint64_t task_corpse_get_crashed_thread_id(task_t corpse_task); + /* * Things which should have prototypes in headers, but don't */ -void *get_bsduthreadarg(thread_t); void proc_exit(proc_t p); int wait1continue(int result); int waitidcontinue(int result); -int *get_bsduthreadrval(thread_t); kern_return_t sys_perf_notify(thread_t thread, int pid); -kern_return_t abnormal_exit_notify(mach_exception_data_type_t code, - mach_exception_data_type_t subcode); -int in_shutdown(void); -void workqueue_exit(struct proc *); +kern_return_t task_exception_notify(exception_type_t exception, + mach_exception_data_type_t code, mach_exception_data_type_t subcode); +kern_return_t task_violated_guard(mach_exception_code_t, mach_exception_subcode_t, void *); void delay(int); - +void gather_rusage_info(proc_t p, rusage_info_current *ru, int flavor); + /* * NOTE: Source and target may *NOT* overlap! * XXX Should share code with bsd/dev/ppc/unix_signal.c */ -static void -siginfo_64to32(user_siginfo_t *in, siginfo_t *out) +void +siginfo_user_to_user32(user_siginfo_t *in, user32_siginfo_t *out) { out->si_signo = in->si_signo; out->si_errno = in->si_errno; @@ -172,23 +201,486 @@ siginfo_64to32(user_siginfo_t *in, siginfo_t *out) out->si_pid = in->si_pid; out->si_uid = in->si_uid; out->si_status = in->si_status; - out->si_addr = CAST_DOWN(void *,in->si_addr); + out->si_addr = CAST_DOWN_EXPLICIT(user32_addr_t,in->si_addr); /* following cast works for sival_int because of padding */ - out->si_value.sival_ptr = CAST_DOWN(void *,in->si_value.sival_ptr); + out->si_value.sival_ptr = CAST_DOWN_EXPLICIT(user32_addr_t,in->si_value.sival_ptr); out->si_band = in->si_band; /* range reduction */ - out->__pad[0] = in->pad[0]; /* mcontext.ss.r1 */ } +void +siginfo_user_to_user64(user_siginfo_t *in, user64_siginfo_t *out) +{ + out->si_signo = in->si_signo; + out->si_errno = in->si_errno; + out->si_code = in->si_code; + out->si_pid = in->si_pid; + out->si_uid = in->si_uid; + out->si_status = in->si_status; + out->si_addr = in->si_addr; + /* following cast works for sival_int because of padding */ + out->si_value.sival_ptr = in->si_value.sival_ptr; + out->si_band = in->si_band; /* range reduction */ +} + +static int +copyoutsiginfo(user_siginfo_t *native, boolean_t is64, user_addr_t uaddr) +{ + if (is64) { + user64_siginfo_t sinfo64; + + bzero(&sinfo64, sizeof (sinfo64)); + siginfo_user_to_user64(native, &sinfo64); + return (copyout(&sinfo64, uaddr, sizeof (sinfo64))); + } else { + user32_siginfo_t sinfo32; + + bzero(&sinfo32, sizeof (sinfo32)); + siginfo_user_to_user32(native, &sinfo32); + return (copyout(&sinfo32, uaddr, sizeof (sinfo32))); + } +} + +void gather_populate_corpse_crashinfo(proc_t p, task_t corpse_task, + mach_exception_data_type_t code, mach_exception_data_type_t subcode, + uint64_t *udata_buffer, int num_udata, void *reason) +{ + struct rusage_superset rup; + + gather_rusage_info(p, &rup.ri, RUSAGE_INFO_CURRENT); + rup.ri.ri_phys_footprint = 0; + populate_corpse_crashinfo(p, corpse_task, &rup, code, subcode, + udata_buffer, num_udata, reason); +} + +static void proc_update_corpse_exception_codes(proc_t p, mach_exception_data_type_t *code, mach_exception_data_type_t *subcode) +{ + mach_exception_data_type_t code_update = *code; + mach_exception_data_type_t subcode_update = *subcode; + if (p->p_exit_reason == OS_REASON_NULL) { + return; + } + + switch (p->p_exit_reason->osr_namespace) { + case OS_REASON_JETSAM: + if (p->p_exit_reason->osr_code == JETSAM_REASON_MEMORY_PERPROCESSLIMIT) { + /* Update the code with EXC_RESOURCE code for high memory watermark */ + EXC_RESOURCE_ENCODE_TYPE(code_update, RESOURCE_TYPE_MEMORY); + EXC_RESOURCE_ENCODE_FLAVOR(code_update, FLAVOR_HIGH_WATERMARK); + EXC_RESOURCE_HWM_ENCODE_LIMIT(code_update, ((get_task_phys_footprint_limit(p->task)) >> 20)); + subcode_update = 0; + break; + } + + break; + default: + break; + } + + *code = code_update; + *subcode = subcode_update; + return; +} + +mach_exception_data_type_t proc_encode_exit_exception_code(proc_t p) +{ + uint64_t subcode = 0; + + if (p->p_exit_reason == OS_REASON_NULL) { + return 0; + } + + /* Embed first 32 bits of osr_namespace and osr_code in exception code */ + ENCODE_OSR_NAMESPACE_TO_MACH_EXCEPTION_CODE(subcode, p->p_exit_reason->osr_namespace); + ENCODE_OSR_CODE_TO_MACH_EXCEPTION_CODE(subcode, p->p_exit_reason->osr_code); + return (mach_exception_data_type_t)subcode; +} + +static void +populate_corpse_crashinfo(proc_t p, task_t corpse_task, struct rusage_superset *rup, + mach_exception_data_type_t code, mach_exception_data_type_t subcode, + uint64_t *udata_buffer, int num_udata, os_reason_t reason) +{ + mach_vm_address_t uaddr = 0; + mach_exception_data_type_t exc_codes[EXCEPTION_CODE_MAX]; + exc_codes[0] = code; + exc_codes[1] = subcode; + cpu_type_t cputype; + struct proc_uniqidentifierinfo p_uniqidinfo; + struct proc_workqueueinfo pwqinfo; + int retval = 0; + uint64_t crashed_threadid = task_corpse_get_crashed_thread_id(corpse_task); + unsigned int pflags = 0; + uint64_t max_footprint_mb; + uint64_t max_footprint; + void *crash_info_ptr = task_get_corpseinfo(corpse_task); + +#if CONFIG_MEMORYSTATUS + int memstat_dirty_flags = 0; +#endif + + if (KERN_SUCCESS == kcdata_get_memory_addr(crash_info_ptr, TASK_CRASHINFO_EXCEPTION_CODES, sizeof(exc_codes), &uaddr)) { + kcdata_memcpy(crash_info_ptr, uaddr, exc_codes, sizeof(exc_codes)); + } + + if (KERN_SUCCESS == kcdata_get_memory_addr(crash_info_ptr, TASK_CRASHINFO_PID, sizeof(p->p_pid), &uaddr)) { + kcdata_memcpy(crash_info_ptr, uaddr, &p->p_pid, sizeof(p->p_pid)); + } + + if (KERN_SUCCESS == kcdata_get_memory_addr(crash_info_ptr, TASK_CRASHINFO_PPID, sizeof(p->p_ppid), &uaddr)) { + kcdata_memcpy(crash_info_ptr, uaddr, &p->p_ppid, sizeof(p->p_ppid)); + } + + /* Don't include the crashed thread ID if there's an exit reason that indicates it's irrelevant */ + if ((p->p_exit_reason == OS_REASON_NULL) || !(p->p_exit_reason->osr_flags & OS_REASON_FLAG_NO_CRASHED_TID)) { + if (KERN_SUCCESS == kcdata_get_memory_addr(crash_info_ptr, TASK_CRASHINFO_CRASHED_THREADID, sizeof(uint64_t), &uaddr)) { + kcdata_memcpy(crash_info_ptr, uaddr, &crashed_threadid, sizeof(uint64_t)); + } + } + + if (KERN_SUCCESS == + kcdata_get_memory_addr(crash_info_ptr, TASK_CRASHINFO_BSDINFOWITHUNIQID, sizeof(struct proc_uniqidentifierinfo), &uaddr)) { + proc_piduniqidentifierinfo(p, &p_uniqidinfo); + kcdata_memcpy(crash_info_ptr, uaddr, &p_uniqidinfo, sizeof(struct proc_uniqidentifierinfo)); + } + + if (KERN_SUCCESS == kcdata_get_memory_addr(crash_info_ptr, TASK_CRASHINFO_RUSAGE_INFO, sizeof(rusage_info_current), &uaddr)) { + kcdata_memcpy(crash_info_ptr, uaddr, &rup->ri, sizeof(rusage_info_current)); + } + + if (KERN_SUCCESS == kcdata_get_memory_addr(crash_info_ptr, TASK_CRASHINFO_PROC_CSFLAGS, sizeof(p->p_csflags), &uaddr)) { + kcdata_memcpy(crash_info_ptr, uaddr, &p->p_csflags, sizeof(p->p_csflags)); + } + + if (KERN_SUCCESS == kcdata_get_memory_addr(crash_info_ptr, TASK_CRASHINFO_PROC_NAME, sizeof(p->p_comm), &uaddr)) { + kcdata_memcpy(crash_info_ptr, uaddr, &p->p_comm, sizeof(p->p_comm)); + } + + if (KERN_SUCCESS == kcdata_get_memory_addr(crash_info_ptr, TASK_CRASHINFO_PROC_STARTTIME, sizeof(p->p_start), &uaddr)) { + struct timeval64 t64; + t64.tv_sec = (int64_t)p->p_start.tv_sec; + t64.tv_usec = (int64_t)p->p_start.tv_usec; + kcdata_memcpy(crash_info_ptr, uaddr, &t64, sizeof(t64)); + } + + if (KERN_SUCCESS == kcdata_get_memory_addr(crash_info_ptr, TASK_CRASHINFO_USERSTACK, sizeof(p->user_stack), &uaddr)) { + kcdata_memcpy(crash_info_ptr, uaddr, &p->user_stack, sizeof(p->user_stack)); + } + + if (KERN_SUCCESS == kcdata_get_memory_addr(crash_info_ptr, TASK_CRASHINFO_ARGSLEN, sizeof(p->p_argslen), &uaddr)) { + kcdata_memcpy(crash_info_ptr, uaddr, &p->p_argslen, sizeof(p->p_argslen)); + } + + if (KERN_SUCCESS == kcdata_get_memory_addr(crash_info_ptr, TASK_CRASHINFO_PROC_ARGC, sizeof(p->p_argc), &uaddr)) { + kcdata_memcpy(crash_info_ptr, uaddr, &p->p_argc, sizeof(p->p_argc)); + } + + if (KERN_SUCCESS == kcdata_get_memory_addr(crash_info_ptr, TASK_CRASHINFO_PROC_PATH, MAXPATHLEN, &uaddr)) { + char *buf = (char *) kalloc(MAXPATHLEN); + if (buf != NULL) { + bzero(buf, MAXPATHLEN); + proc_pidpathinfo_internal(p, 0, buf, MAXPATHLEN, &retval); + kcdata_memcpy(crash_info_ptr, uaddr, buf, MAXPATHLEN); + kfree(buf, MAXPATHLEN); + } + } + + pflags = p->p_flag & (P_LP64 | P_SUGID); + if (KERN_SUCCESS == kcdata_get_memory_addr(crash_info_ptr, TASK_CRASHINFO_PROC_FLAGS, sizeof(pflags), &uaddr)) { + kcdata_memcpy(crash_info_ptr, uaddr, &pflags, sizeof(pflags)); + } + + if (KERN_SUCCESS == kcdata_get_memory_addr(crash_info_ptr, TASK_CRASHINFO_UID, sizeof(p->p_uid), &uaddr)) { + kcdata_memcpy(crash_info_ptr, uaddr, &p->p_uid, sizeof(p->p_uid)); + } + + if (KERN_SUCCESS == kcdata_get_memory_addr(crash_info_ptr, TASK_CRASHINFO_GID, sizeof(p->p_gid), &uaddr)) { + kcdata_memcpy(crash_info_ptr, uaddr, &p->p_gid, sizeof(p->p_gid)); + } + + cputype = cpu_type() & ~CPU_ARCH_MASK; + if (IS_64BIT_PROCESS(p)) + cputype |= CPU_ARCH_ABI64; + + if (KERN_SUCCESS == kcdata_get_memory_addr(crash_info_ptr, TASK_CRASHINFO_CPUTYPE, sizeof(cpu_type_t), &uaddr)) { + kcdata_memcpy(crash_info_ptr, uaddr, &cputype, sizeof(cpu_type_t)); + } + + if (KERN_SUCCESS == kcdata_get_memory_addr(crash_info_ptr, TASK_CRASHINFO_MEMORY_LIMIT, sizeof(max_footprint_mb), &uaddr)) { + max_footprint = get_task_phys_footprint_limit(p->task); + max_footprint_mb = max_footprint >> 20; + kcdata_memcpy(crash_info_ptr, uaddr, &max_footprint_mb, sizeof(max_footprint_mb)); + } + + bzero(&pwqinfo, sizeof(struct proc_workqueueinfo)); + retval = fill_procworkqueue(p, &pwqinfo); + if (retval == 0) { + if (KERN_SUCCESS == kcdata_get_memory_addr(crash_info_ptr, TASK_CRASHINFO_WORKQUEUEINFO, sizeof(struct proc_workqueueinfo), &uaddr)) { + kcdata_memcpy(crash_info_ptr, uaddr, &pwqinfo, sizeof(struct proc_workqueueinfo)); + } + } + + if (KERN_SUCCESS == kcdata_get_memory_addr(crash_info_ptr, TASK_CRASHINFO_RESPONSIBLE_PID, sizeof(p->p_responsible_pid), &uaddr)) { + kcdata_memcpy(crash_info_ptr, uaddr, &p->p_responsible_pid, sizeof(p->p_responsible_pid)); + } + +#if CONFIG_COALITIONS + if (KERN_SUCCESS == kcdata_get_memory_addr_for_array(crash_info_ptr, TASK_CRASHINFO_COALITION_ID, sizeof(uint64_t), COALITION_NUM_TYPES, &uaddr)) { + uint64_t coalition_ids[COALITION_NUM_TYPES]; + task_coalition_ids(p->task, coalition_ids); + kcdata_memcpy(crash_info_ptr, uaddr, coalition_ids, sizeof(coalition_ids)); + } +#endif /* CONFIG_COALITIONS */ + +#if CONFIG_MEMORYSTATUS + memstat_dirty_flags = memorystatus_dirty_get(p); + if (KERN_SUCCESS == kcdata_get_memory_addr(crash_info_ptr, TASK_CRASHINFO_DIRTY_FLAGS, sizeof(memstat_dirty_flags), &uaddr)) { + kcdata_memcpy(crash_info_ptr, uaddr, &memstat_dirty_flags, sizeof(memstat_dirty_flags)); + } +#endif + + if (p->p_exit_reason != OS_REASON_NULL && reason == OS_REASON_NULL) { + reason = p->p_exit_reason; + } + if (reason != OS_REASON_NULL) { + if (KERN_SUCCESS == kcdata_get_memory_addr(crash_info_ptr, EXIT_REASON_SNAPSHOT, sizeof(struct exit_reason_snapshot), &uaddr)) { + struct exit_reason_snapshot ers = { + .ers_namespace = reason->osr_namespace, + .ers_code = reason->osr_code, + .ers_flags = reason->osr_flags + }; + + kcdata_memcpy(crash_info_ptr, uaddr, &ers, sizeof(ers)); + } + + if (reason->osr_kcd_buf != 0) { + uint32_t reason_buf_size = kcdata_memory_get_used_bytes(&reason->osr_kcd_descriptor); + assert(reason_buf_size != 0); + + if (KERN_SUCCESS == kcdata_get_memory_addr(crash_info_ptr, KCDATA_TYPE_NESTED_KCDATA, reason_buf_size, &uaddr)) { + kcdata_memcpy(crash_info_ptr, uaddr, reason->osr_kcd_buf, reason_buf_size); + } + } + } + + if (num_udata > 0) { + if (KERN_SUCCESS == kcdata_get_memory_addr_for_array(crash_info_ptr, TASK_CRASHINFO_UDATA_PTRS, + sizeof(uint64_t), num_udata, &uaddr)) { + kcdata_memcpy(crash_info_ptr, uaddr, udata_buffer, sizeof(uint64_t) * num_udata); + } + } +} + +/* + * We only parse exit reason kcdata blobs for launchd when it dies + * and we're going to panic. + * + * Meant to be called immediately before panicking. + */ +char * +launchd_exit_reason_get_string_desc(os_reason_t exit_reason) +{ + kcdata_iter_t iter; + + if (exit_reason == OS_REASON_NULL || exit_reason->osr_kcd_buf == NULL || + exit_reason->osr_bufsize == 0) { + return NULL; + } + + iter = kcdata_iter(exit_reason->osr_kcd_buf, exit_reason->osr_bufsize); + if (!kcdata_iter_valid(iter)) { +#if DEBUG || DEVELOPMENT + printf("launchd exit reason has invalid exit reason buffer\n"); +#endif + return NULL; + } + + if (kcdata_iter_type(iter) != KCDATA_BUFFER_BEGIN_OS_REASON) { +#if DEBUG || DEVELOPMENT + printf("launchd exit reason buffer type mismatch, expected %d got %d\n", + KCDATA_BUFFER_BEGIN_OS_REASON, kcdata_iter_type(iter)); +#endif + return NULL; + } + + iter = kcdata_iter_find_type(iter, EXIT_REASON_USER_DESC); + if (!kcdata_iter_valid(iter)) { + return NULL; + } + + return (char *)kcdata_iter_payload(iter); +} + +static __attribute__((noinline)) void +launchd_crashed_panic(proc_t p, int rv) +{ + char *launchd_exit_reason_desc = launchd_exit_reason_get_string_desc(p->p_exit_reason); + + if (p->p_exit_reason == OS_REASON_NULL) { + printf("pid 1 exited -- no exit reason available -- (signal %d, exit %d)\n", + WTERMSIG(rv), WEXITSTATUS(rv)); + } else { + printf("pid 1 exited -- exit reason namespace %d subcode 0x%llx, description %s\n", + p->p_exit_reason->osr_namespace, p->p_exit_reason->osr_code, launchd_exit_reason_desc ? + launchd_exit_reason_desc : "none"); + } + + const char *launchd_crashed_prefix_str; + + if (strnstr(p->p_name, "preinit", sizeof(p->p_name))) { + launchd_crashed_prefix_str = "LTE preinit process exited"; + } else { + launchd_crashed_prefix_str = "initproc exited"; + } + +#if (DEVELOPMENT || DEBUG) && CONFIG_COREDUMP + /* + * For debugging purposes, generate a core file of initproc before + * panicking. Leave at least 300 MB free on the root volume, and ignore + * the process's corefile ulimit. fsync() the file to ensure it lands on disk + * before the panic hits. + */ + + int err; + uint64_t coredump_start = mach_absolute_time(); + uint64_t coredump_end; + clock_sec_t tv_sec; + clock_usec_t tv_usec; + uint32_t tv_msec; + + + err = coredump(p, 300, COREDUMP_IGNORE_ULIMIT | COREDUMP_FULLFSYNC); + + coredump_end = mach_absolute_time(); + + absolutetime_to_microtime(coredump_end - coredump_start, &tv_sec, &tv_usec); + + tv_msec = tv_usec / 1000; + + if (err != 0) { + printf("Failed to generate initproc core file: error %d, took %d.%03d seconds\n", + err, (uint32_t)tv_sec, tv_msec); + } else { + printf("Generated initproc core file in %d.%03d seconds\n", + (uint32_t)tv_sec, tv_msec); + } +#endif /* (DEVELOPMENT || DEBUG) && CONFIG_COREDUMP */ + + sync(p, (void *)NULL, (int *)NULL); + + if (p->p_exit_reason == OS_REASON_NULL) { + panic_with_options(0, NULL, DEBUGGER_OPTION_INITPROC_PANIC, "%s -- no exit reason available -- (signal %d, exit status %d %s)", + launchd_crashed_prefix_str, WTERMSIG(rv), WEXITSTATUS(rv), ((p->p_csflags & CS_KILLED) ? "CS_KILLED" : "")); + } else { + panic_with_options(0, NULL, DEBUGGER_OPTION_INITPROC_PANIC, "%s %s -- exit reason namespace %d subcode 0x%llx description: %." LAUNCHD_PANIC_REASON_STRING_MAXLEN "s", + ((p->p_csflags & CS_KILLED) ? "CS_KILLED" : ""), + launchd_crashed_prefix_str, p->p_exit_reason->osr_namespace, p->p_exit_reason->osr_code, + launchd_exit_reason_desc ? launchd_exit_reason_desc : "none"); + } +} + +#define OS_REASON_IFLAG_USER_FAULT 0x1 + +#define OS_REASON_TOTAL_USER_FAULTS_PER_PROC 5 + +static int +abort_with_payload_internal(proc_t p, + uint32_t reason_namespace, uint64_t reason_code, + user_addr_t payload, uint32_t payload_size, + user_addr_t reason_string, uint64_t reason_flags, + uint32_t internal_flags) +{ + os_reason_t exit_reason = OS_REASON_NULL; + kern_return_t kr = KERN_SUCCESS; + + if (internal_flags & OS_REASON_IFLAG_USER_FAULT) { + uint32_t old_value = atomic_load_explicit(&p->p_user_faults, + memory_order_relaxed); + for (;;) { + if (old_value >= OS_REASON_TOTAL_USER_FAULTS_PER_PROC) { + return EQFULL; + } + // this reloads the value in old_value + if (atomic_compare_exchange_strong_explicit(&p->p_user_faults, + &old_value, old_value + 1, memory_order_relaxed, + memory_order_relaxed)) { + break; + } + } + } + + KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_PROC, BSD_PROC_EXITREASON_CREATE) | DBG_FUNC_NONE, + p->p_pid, reason_namespace, + reason_code, 0, 0); + + exit_reason = build_userspace_exit_reason(reason_namespace, reason_code, + payload, payload_size, reason_string, reason_flags); + + if (internal_flags & OS_REASON_IFLAG_USER_FAULT) { + mach_exception_code_t code = 0; + + EXC_GUARD_ENCODE_TYPE(code, GUARD_TYPE_USER); /* simulated EXC_GUARD */ + EXC_GUARD_ENCODE_FLAVOR(code, 0); + EXC_GUARD_ENCODE_TARGET(code, reason_namespace); + + if (exit_reason == OS_REASON_NULL) { + kr = KERN_RESOURCE_SHORTAGE; + } else { + kr = task_violated_guard(code, reason_code, exit_reason); + } + os_reason_free(exit_reason); + } else { + /* + * We use SIGABRT (rather than calling exit directly from here) so that + * the debugger can catch abort_with_{reason,payload} calls. + */ + psignal_try_thread_with_reason(p, current_thread(), SIGABRT, exit_reason); + } + + switch (kr) { + case KERN_SUCCESS: + return 0; + case KERN_NOT_SUPPORTED: + return ENOTSUP; + case KERN_INVALID_ARGUMENT: + return EINVAL; + case KERN_RESOURCE_SHORTAGE: + default: + return EBUSY; + } +} + +int +abort_with_payload(struct proc *cur_proc, struct abort_with_payload_args *args, + __unused void *retval) +{ + abort_with_payload_internal(cur_proc, args->reason_namespace, + args->reason_code, args->payload, args->payload_size, + args->reason_string, args->reason_flags, 0); + + return 0; +} + +int +os_fault_with_payload(struct proc *cur_proc, + struct os_fault_with_payload_args *args, __unused int *retval) +{ + return abort_with_payload_internal(cur_proc, args->reason_namespace, + args->reason_code, args->payload, args->payload_size, + args->reason_string, args->reason_flags, OS_REASON_IFLAG_USER_FAULT); +} + + /* * exit -- * Death of process. */ +__attribute__((noreturn)) void exit(proc_t p, struct exit_args *uap, int *retval) { + p->p_xhighbits = ((uint32_t)(uap->rval) & 0xFF000000) >> 24; exit1(p, W_EXITCODE(uap->rval, 0), retval); - /* drop funnel before we return */ thread_exception_return(); /* NOTREACHED */ while (TRUE) @@ -203,10 +695,28 @@ exit(proc_t p, struct exit_args *uap, int *retval) */ int exit1(proc_t p, int rv, int *retval) +{ + return exit1_internal(p, rv, retval, TRUE, TRUE, 0); +} + +int +exit1_internal(proc_t p, int rv, int *retval, boolean_t thread_can_terminate, boolean_t perf_notify, + int jetsam_flags) +{ + return exit_with_reason(p, rv, retval, thread_can_terminate, perf_notify, jetsam_flags, OS_REASON_NULL); +} + +/* + * NOTE: exit_with_reason drops a reference on the passed exit_reason + */ +int +exit_with_reason(proc_t p, int rv, int *retval, boolean_t thread_can_terminate, boolean_t perf_notify, + int jetsam_flags, struct os_reason *exit_reason) { thread_t self = current_thread(); struct task *task = p->task; struct uthread *ut; + int error = 0; /* * If a thread in this task has already @@ -215,11 +725,17 @@ exit1(proc_t p, int rv, int *retval) */ ut = get_bsdthread_info(self); - if (ut->uu_flag & UT_VFORK) { - vfork_exit(p, rv); - vfork_return(p , retval, p->p_pid); - unix_syscall_return(0); - /* NOT REACHED */ + if ((p == current_proc()) && + (ut->uu_flag & UT_VFORK)) { + os_reason_free(exit_reason); + if (!thread_can_terminate) { + return EINVAL; + } + + vfork_exit(p, rv); + vfork_return(p , retval, p->p_pid); + unix_syscall_return(0); + /* NOT REACHED */ } /* @@ -228,60 +744,128 @@ exit1(proc_t p, int rv, int *retval) * which is currently required by mac_audit_postselect(). */ + /* + * The BSM token contains two components: an exit status as passed + * to exit(), and a return value to indicate what sort of exit it + * was. The exit status is WEXITSTATUS(rv), but it's not clear + * what the return value is. + */ + AUDIT_ARG(exit, WEXITSTATUS(rv), 0); + /* + * TODO: what to audit here when jetsam calls exit and the uthread, + * 'ut' does not belong to the proc, 'p'. + */ AUDIT_SYSCALL_EXIT(SYS_exit, p, ut, 0); /* Exit is always successfull */ DTRACE_PROC1(exit, int, CLD_EXITED); + /* mark process is going to exit and pull out of DBG/disk throttle */ + /* TODO: This should be done after becoming exit thread */ + proc_set_task_policy(p->task, TASK_POLICY_ATTRIBUTE, + TASK_POLICY_TERMINATED, TASK_POLICY_ENABLE); + proc_lock(p); + error = proc_transstart(p, 1, (jetsam_flags ? 1 : 0)); + if (error == EDEADLK) { + /* + * If proc_transstart() returns EDEADLK, then another thread + * is either exec'ing or exiting. Return an error and allow + * the other thread to continue. + */ + proc_unlock(p); + os_reason_free(exit_reason); + if (current_proc() == p){ + if (p->exit_thread == self) { + printf("exit_thread failed to exit, leaving process %s[%d] in unkillable limbo\n", + p->p_comm, p->p_pid); + } + + if (thread_can_terminate) { + thread_exception_return(); + } + } + + return error; + } + while (p->exit_thread != self) { if (sig_try_locked(p) <= 0) { + proc_transend(p, 1); + os_reason_free(exit_reason); + if (get_threadtask(self) != task) { proc_unlock(p); return(0); } proc_unlock(p); + thread_terminate(self); + if (!thread_can_terminate) { + return 0; + } + thread_exception_return(); /* NOTREACHED */ } sig_lock_to_exit(p); } -#if !CONFIG_EMBEDDED /* BER_XXX */ - if (p->p_pid == 1) { - proc_unlock(p); - printf("pid 1 exited (signal %d, exit %d)", - WTERMSIG(rv), WEXITSTATUS(rv)); - panic("%s died\nState at Last Exception:\n\n%s", - (p->p_comm[0] != '\0' ? - p->p_comm : - "launchd"), - init_task_failure_data); + + if (exit_reason != OS_REASON_NULL) { + KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_PROC, BSD_PROC_EXITREASON_COMMIT) | DBG_FUNC_NONE, + p->p_pid, exit_reason->osr_namespace, + exit_reason->osr_code, 0, 0); } -#endif + + assert(p->p_exit_reason == OS_REASON_NULL); + p->p_exit_reason = exit_reason; p->p_lflag |= P_LEXIT; p->p_xstat = rv; + p->p_lflag |= jetsam_flags; + proc_transend(p, 1); proc_unlock(p); - proc_prepareexit(p, rv); + proc_prepareexit(p, rv, perf_notify); - /* task terminate will call proc_terminate and that cleans it up */ + /* Last thread to terminate will call proc_exit() */ task_terminate_internal(task); return(0); } void -proc_prepareexit(proc_t p, int rv) +proc_prepareexit(proc_t p, int rv, boolean_t perf_notify) { - mach_exception_data_type_t code, subcode; + mach_exception_data_type_t code = 0, subcode = 0; + struct uthread *ut; thread_t self = current_thread(); ut = get_bsdthread_info(self); + struct rusage_superset *rup; + int kr = 0; + int create_corpse = FALSE; + + if (p == initproc) { + launchd_crashed_panic(p, rv); + /* NOTREACHED */ + } - /* If a core should be generated, notify crash reporter */ - if (!in_shutdown() && hassigprop(WTERMSIG(rv), SA_CORE)) { + /* + * Generate a corefile/crashlog if: + * The process doesn't have an exit reason that indicates no crash report should be created + * AND any of the following are true: + * - The process was terminated due to a fatal signal that generates a core + * - The process was killed due to a code signing violation + * - The process has an exit reason that indicates we should generate a crash report + * + * The first condition is necessary because abort_with_reason()/payload() use SIGABRT + * (which normally triggers a core) but may indicate that no crash report should be created. + */ + if (!(PROC_HAS_EXITREASON(p) && (PROC_EXITREASON_FLAGS(p) & OS_REASON_FLAG_NO_CRASH_REPORT)) && + (hassigprop(WTERMSIG(rv), SA_CORE) || ((p->p_csflags & CS_KILLED) != 0) || + (PROC_HAS_EXITREASON(p) && (PROC_EXITREASON_FLAGS(p) & + OS_REASON_FLAG_GENERATE_CRASH_REPORT)))) { /* * Workaround for processes checking up on PT_DENY_ATTACH: * should be backed out post-Leopard (details in 5431025). @@ -300,13 +884,81 @@ proc_prepareexit(proc_t p, int rv) ((ut->uu_exception & 0x0f) << 20) | ((int)ut->uu_code & 0xfffff); subcode = ut->uu_subcode; - (void) abnormal_exit_notify(code, subcode); + + kr = task_exception_notify(EXC_CRASH, code, subcode); + + /* Nobody handled EXC_CRASH?? remember to make corpse */ + if (kr != 0) { + create_corpse = TRUE; + } } skipcheck: - /* Notify the perf server */ - (void)sys_perf_notify(self, p->p_pid); + /* Notify the perf server? */ + if (perf_notify) { + (void)sys_perf_notify(self, p->p_pid); + } + + + /* stash the usage into corpse data if making_corpse == true */ + if (create_corpse == TRUE) { + kr = task_mark_corpse(p->task); + if (kr != KERN_SUCCESS) { + if (kr == KERN_NO_SPACE) { + printf("Process[%d] has no vm space for corpse info.\n", p->p_pid); + } else if (kr == KERN_NOT_SUPPORTED) { + printf("Process[%d] was destined to be corpse. But corpse is disabled by config.\n", p->p_pid); + } else { + printf("Process[%d] crashed: %s. Too many corpses being created.\n", p->p_pid, p->p_comm); + } + create_corpse = FALSE; + } + } + /* + * Before this process becomes a zombie, stash resource usage + * stats in the proc for external observers to query + * via proc_pid_rusage(). + * + * If the zombie allocation fails, just punt the stats. + */ + MALLOC_ZONE(rup, struct rusage_superset *, + sizeof (*rup), M_ZOMBIE, M_WAITOK); + if (rup != NULL) { + gather_rusage_info(p, &rup->ri, RUSAGE_INFO_CURRENT); + rup->ri.ri_phys_footprint = 0; + rup->ri.ri_proc_exit_abstime = mach_absolute_time(); + + /* + * Make the rusage_info visible to external observers + * only after it has been completely filled in. + */ + p->p_ru = rup; + } + if (create_corpse) { + int est_knotes = 0, num_knotes = 0; + uint64_t *buffer = NULL; + int buf_size = 0; + + /* Get all the udata pointers from kqueue */ + est_knotes = kevent_proc_copy_uptrs(p, NULL, 0); + if (est_knotes > 0) { + buf_size = (est_knotes + 32) * sizeof(uint64_t); + buffer = (uint64_t *) kalloc(buf_size); + num_knotes = kevent_proc_copy_uptrs(p, buffer, buf_size); + if (num_knotes > est_knotes + 32) { + num_knotes = est_knotes + 32; + } + } + + /* Update the code, subcode based on exit reason */ + proc_update_corpse_exception_codes(p, &code, &subcode); + populate_corpse_crashinfo(p, p->task, rup, + code, subcode, buffer, num_knotes, NULL); + if (buffer != NULL) { + kfree(buffer, buf_size); + } + } /* * Remove proc from allproc queue and from pidhash chain. * Need to do this before we do anything that can block. @@ -316,6 +968,10 @@ skipcheck: proc_list_lock(); +#if CONFIG_MEMORYSTATUS + memorystatus_remove(p, TRUE); +#endif + LIST_REMOVE(p, p_list); LIST_INSERT_HEAD(&zombproc, p, p_list); /* Place onto zombproc. */ /* will not be visible via proc_find */ @@ -344,64 +1000,59 @@ proc_exit(proc_t p) proc_t q; proc_t pp; struct task *task = p->task; - boolean_t fstate; vnode_t tvp = NULLVP; struct pgrp * pg; struct session *sessp; struct uthread * uth; + pid_t pid; + int exitval; + int knote_hint; - /* This can happen if thread_terminate of the single thread - * process - */ - - uth = (struct uthread *)get_bsdthread_info(current_thread()); + uth = current_uthread(); proc_lock(p); + proc_transstart(p, 1, 0); if( !(p->p_lflag & P_LEXIT)) { + /* + * This can happen if a thread_terminate() occurs + * in a single-threaded process. + */ p->p_lflag |= P_LEXIT; + proc_transend(p, 1); proc_unlock(p); - proc_prepareexit(p, 0); + proc_prepareexit(p, 0, TRUE); + (void) task_terminate_internal(task); proc_lock(p); + } else { + proc_transend(p, 1); } p->p_lflag |= P_LPEXIT; - proc_unlock(p); - -#if CONFIG_DTRACE - /* - * Free any outstanding lazy dof entries. It is imperative we - * always call dtrace_lazy_dofs_destroy, rather than null check - * and call if !NULL. If we NULL test, during lazy dof faulting - * we can race with the faulting code and proceed from here to - * beyond the helpers cleanup. The lazy dof faulting will then - * install new helpers which will never be cleaned up, and leak. - */ - dtrace_lazy_dofs_destroy(p); /* - * Clean up any DTrace helper actions or probes for the process. + * Other kernel threads may be in the middle of signalling this process. + * Wait for those threads to wrap it up before making the process + * disappear on them. */ - if (p->p_dtrace_helpers != NULL) { - (*dtrace_helpers_cleanup)(p); + if ((p->p_lflag & P_LINSIGNAL) || (p->p_sigwaitcnt > 0)) { + p->p_sigwaitcnt++; + while ((p->p_lflag & P_LINSIGNAL) || (p->p_sigwaitcnt > 1)) + msleep(&p->p_sigmask, &p->p_mlock, PWAIT, "proc_sigdrain", NULL); + p->p_sigwaitcnt--; } - /* - * Clean up any DTrace probes associated with this process. - */ - /* - * APPLE NOTE: We release ptss pages/entries in dtrace_fasttrap_exit_ptr(), - * call this after dtrace_helpers_cleanup() - */ - proc_lock(p); - if (p->p_dtrace_probes && dtrace_fasttrap_exit_ptr) { - (*dtrace_fasttrap_exit_ptr)(p); - } proc_unlock(p); + pid = p->p_pid; + exitval = p->p_xstat; + KERNEL_DEBUG_CONSTANT_IST(KDEBUG_COMMON, + BSDDBG_CODE(DBG_BSD_PROC, BSD_PROC_EXIT) | DBG_FUNC_START, + pid, exitval, 0, 0, 0); + +#if CONFIG_DTRACE + dtrace_proc_exit(p); #endif - /* XXX Zombie allocation may fail, in which case stats get lost */ - MALLOC_ZONE(p->p_ru, struct rusage *, - sizeof (*p->p_ru), M_ZOMBIE, M_WAITOK); + nspace_proc_exit(p); /* * need to cancel async IO requests that can be cancelled and wait for those @@ -410,7 +1061,10 @@ proc_exit(proc_t p) proc_refdrain(p); - workqueue_exit(p); + /* if any pending cpu limits action, clear it */ + task_clear_cpuusage(p->task, TRUE); + + workqueue_mark_exiting(p); _aio_exit( p ); @@ -420,6 +1074,23 @@ proc_exit(proc_t p) */ fdfree(p); + /* + * Once all the knotes, kqueues & workloops are destroyed, get rid of the + * workqueue. + */ + workqueue_exit(p); + + if (uth->uu_lowpri_window) { + /* + * task is marked as a low priority I/O type + * and the I/O we issued while in flushing files on close + * collided with normal I/O operations... + * no need to throttle this thread since its going away + * but we do need to update our bookeeping w/r to throttled threads + */ + throttle_lowpri_io(0); + } + #if SYSV_SHM /* Close ref SYSV Shared memory*/ if (p->vm_shm) @@ -430,18 +1101,19 @@ proc_exit(proc_t p) semexit(p); #endif +#if PSYNCH + pth_proc_hashdelete(p); +#endif /* PSYNCH */ + sessp = proc_session(p); if (SESS_LEADER(p, sessp)) { - /* Protected by funnel for tty accesses */ - fstate = thread_funnel_set(kernel_flock, TRUE); - if (sessp->s_ttyvp != NULLVP) { struct vnode *ttyvp; int ttyvid; + int cttyflag = 0; struct vfs_context context; - struct tty * tp; - + struct tty *tp; /* * Controlling process. @@ -449,51 +1121,72 @@ proc_exit(proc_t p) * drain controlling terminal * and revoke access to controlling terminal. */ - tp = sessp->s_ttyp; - + session_lock(sessp); + tp = SESSION_TP(sessp); if ((tp != TTY_NULL) && (tp->t_session == sessp)) { - tty_pgsignal(tp, SIGHUP, 1); - (void) ttywait(tp); + session_unlock(sessp); + /* - * The tty could have been revoked - * if we blocked. + * We're going to SIGHUP the foreground process + * group. It can't change from this point on + * until the revoke is complete. + * The process group changes under both the tty + * lock and proc_list_lock but we need only one */ + tty_lock(tp); + ttysetpgrphup(tp); + tty_unlock(tp); - session_lock(sessp); - ttyvp = sessp->s_ttyvp; - ttyvid = sessp->s_ttyvid; - sessp->s_ttyvp = NULL; - sessp->s_ttyvid = 0; - sessp->s_ttyp = NULL; - sessp->s_ttypgrpid = NO_PID; - session_unlock(sessp); + tty_pgsignal(tp, SIGHUP, 1); - if ((ttyvp != NULLVP) && (vnode_getwithvid(ttyvp, ttyvid) == 0)) { - context.vc_thread = proc_thread(p); /* XXX */ - context.vc_ucred = kauth_cred_proc_ref(p); - VNOP_REVOKE(ttyvp, REVOKEALL, &context); - vnode_put(ttyvp); - kauth_cred_unref(&context.vc_ucred); - } - } else { session_lock(sessp); - ttyvp = sessp->s_ttyvp; - sessp->s_ttyvp = NULL; - sessp->s_ttyvid = 0; - sessp->s_ttyp = NULL; - sessp->s_ttypgrpid = NO_PID; - session_unlock(sessp); + tp = SESSION_TP(sessp); + } + cttyflag = sessp->s_flags & S_CTTYREF; + sessp->s_flags &= ~S_CTTYREF; + ttyvp = sessp->s_ttyvp; + ttyvid = sessp->s_ttyvid; + sessp->s_ttyvp = NULLVP; + sessp->s_ttyvid = 0; + sessp->s_ttyp = TTY_NULL; + sessp->s_ttypgrpid = NO_PID; + session_unlock(sessp); + + if ((ttyvp != NULLVP) && (vnode_getwithvid(ttyvp, ttyvid) == 0)) { + if (tp != TTY_NULL) { + tty_lock(tp); + (void) ttywait(tp); + tty_unlock(tp); + } + context.vc_thread = proc_thread(p); /* XXX */ + context.vc_ucred = kauth_cred_proc_ref(p); + VNOP_REVOKE(ttyvp, REVOKEALL, &context); + if (cttyflag) { + /* + * Release the extra usecount taken in cttyopen. + * usecount should be released after VNOP_REVOKE is called. + * This usecount was taken to ensure that + * the VNOP_REVOKE results in a close to + * the tty since cttyclose is a no-op. + */ + vnode_rele(ttyvp); + } + vnode_put(ttyvp); + kauth_cred_unref(&context.vc_ucred); + ttyvp = NULLVP; + } + if (tp) { + /* + * This is cleared even if not set. This is also done in + * spec_close to ensure that the flag is cleared. + */ + tty_lock(tp); + ttyclrpgrphup(tp); + tty_unlock(tp); + + ttyfree(tp); } - if (ttyvp) - vnode_rele(ttyvp); - /* - * s_ttyp is not zero'd; we use this to indicate - * that the session once had a controlling terminal. - * (for logging and informational purposes) - */ } - - (void) thread_funnel_set(kernel_flock, fstate); session_lock(sessp); sessp->s_leader = NULL; session_unlock(sessp); @@ -508,10 +1201,18 @@ proc_exit(proc_t p) (void)acct_process(p); proc_list_lock(); + + if ((p->p_listflag & P_LIST_EXITCOUNT) == P_LIST_EXITCOUNT) { + p->p_listflag &= ~P_LIST_EXITCOUNT; + proc_shutdown_exitcount--; + if (proc_shutdown_exitcount == 0) + wakeup(&proc_shutdown_exitcount); + } + /* wait till parentrefs are dropped and grant no more */ proc_childdrainstart(p); while ((q = p->p_children.lh_first) != NULL) { - q->p_listflag |= P_LIST_DEADPARENT; + int reparentedtoinit = (q->p_listflag & P_LIST_DEADPARENT) ? 1 : 0; if (q->p_stat == SZOMB) { if (p != q->p_pptr) panic("parent child linkage broken"); @@ -525,22 +1226,52 @@ proc_exit(proc_t p) * if the reap is already in progress. So we get * the reference here exclusively and their can be * no waiters. So there is no need for a wakeup - * after we are done. AlsO the reap frees the structure + * after we are done. Also the reap frees the structure * and the proc struct cannot be used for wakeups as well. * It is safe to use q here as this is system reap */ - (void)reap_child_locked(p, q, 1, 1, 0); + (void)reap_child_locked(p, q, 1, reparentedtoinit, 1, 0); } else { - proc_reparentlocked(q, initproc, 0, 1); /* * Traced processes are killed * since their existence means someone is messing up. */ if (q->p_lflag & P_LTRACED) { + struct proc *opp; + + /* + * Take a reference on the child process to + * ensure it doesn't exit and disappear between + * the time we drop the list_lock and attempt + * to acquire its proc_lock. + */ + if (proc_ref_locked(q) != q) + continue; + proc_list_unlock(); + + opp = proc_find(q->p_oppid); + if (opp != PROC_NULL) { + proc_list_lock(); + q->p_oppid = 0; + proc_list_unlock(); + proc_reparentlocked(q, opp, 0, 0); + proc_rele(opp); + } else { + /* original parent exited while traced */ + proc_list_lock(); + q->p_listflag |= P_LIST_DEADPARENT; + q->p_oppid = 0; + proc_list_unlock(); + proc_reparentlocked(q, initproc, 0, 0); + } + proc_lock(q); q->p_lflag &= ~P_LTRACED; + if (q->sigwait_thread) { + thread_t thread = q->sigwait_thread; + proc_unlock(q); /* * The sigwait_thread could be stopped at a @@ -549,13 +1280,19 @@ proc_exit(proc_t p) * the first thread in the task. So any attempts to kill * the process would result into a deadlock on q->sigwait. */ - thread_resume((thread_t)q->sigwait_thread); - clear_wait(q->sigwait_thread, THREAD_INTERRUPTED); - threadsignal((thread_t)q->sigwait_thread, SIGKILL, 0); - } else + thread_resume(thread); + clear_wait(thread, THREAD_INTERRUPTED); + threadsignal(thread, SIGKILL, 0, TRUE); + } else { proc_unlock(q); + } + psignal(q, SIGKILL); proc_list_lock(); + proc_rele_locked(q); + } else { + q->p_listflag |= P_LIST_DEADPARENT; + proc_reparentlocked(q, initproc, 0, 1); } } } @@ -563,6 +1300,15 @@ proc_exit(proc_t p) proc_childdrainend(p); proc_list_unlock(); +#if CONFIG_MACF + /* + * Notify MAC policies that proc is dead. + * This should be replaced with proper label management + * (rdar://problem/32126399). + */ + mac_proc_notify_exit(p); +#endif + /* * Release reference to text vnode */ @@ -570,63 +1316,18 @@ proc_exit(proc_t p) p->p_textvp = NULL; if (tvp != NULLVP) { vnode_rele(tvp); - } - - /* - * Save exit status and final rusage info, adding in child rusage - * info and self times. If we were unable to allocate a zombie - * structure, this information is lost. - */ - /* No need for locking here as no one than this thread can access this */ - if (p->p_ru != NULL) { - *p->p_ru = p->p_stats->p_ru; - - timerclear(&p->p_ru->ru_utime); - timerclear(&p->p_ru->ru_stime); - - if (task) { - task_basic_info_32_data_t tinfo; - task_thread_times_info_data_t ttimesinfo; - task_events_info_data_t teventsinfo; - mach_msg_type_number_t task_info_stuff, task_ttimes_stuff; - mach_msg_type_number_t task_events_stuff; - struct timeval ut,st; - - task_info_stuff = TASK_BASIC_INFO_32_COUNT; - task_info(task, TASK_BASIC2_INFO_32, - (task_info_t)&tinfo, &task_info_stuff); - p->p_ru->ru_utime.tv_sec = tinfo.user_time.seconds; - p->p_ru->ru_utime.tv_usec = tinfo.user_time.microseconds; - p->p_ru->ru_stime.tv_sec = tinfo.system_time.seconds; - p->p_ru->ru_stime.tv_usec = tinfo.system_time.microseconds; - - p->p_ru->ru_maxrss = tinfo.resident_size; - - task_ttimes_stuff = TASK_THREAD_TIMES_INFO_COUNT; - task_info(task, TASK_THREAD_TIMES_INFO, - (task_info_t)&ttimesinfo, &task_ttimes_stuff); + } - ut.tv_sec = ttimesinfo.user_time.seconds; - ut.tv_usec = ttimesinfo.user_time.microseconds; - st.tv_sec = ttimesinfo.system_time.seconds; - st.tv_usec = ttimesinfo.system_time.microseconds; - timeradd(&ut,&p->p_ru->ru_utime,&p->p_ru->ru_utime); - timeradd(&st,&p->p_ru->ru_stime,&p->p_ru->ru_stime); - - task_events_stuff = TASK_EVENTS_INFO_COUNT; - task_info(task, TASK_EVENTS_INFO, - (task_info_t)&teventsinfo, &task_events_stuff); - - p->p_ru->ru_minflt = (teventsinfo.faults - - teventsinfo.pageins); - p->p_ru->ru_majflt = teventsinfo.pageins; - p->p_ru->ru_nivcsw = (teventsinfo.csw - - p->p_ru->ru_nvcsw); - if (p->p_ru->ru_nivcsw < 0) - p->p_ru->ru_nivcsw = 0; - } + /* + * Save exit status and final rusage info, adding in child rusage + * info and self times. If we were unable to allocate a zombie + * structure, this information is lost. + */ + if (p->p_ru != NULL) { + calcru(p, &p->p_stats->p_ru.ru_utime, &p->p_stats->p_ru.ru_stime, NULL); + p->p_ru->ru = p->p_stats->p_ru; - ruadd(p->p_ru, &p->p_stats->p_cru); + ruadd(&(p->p_ru->ru), &p->p_stats->p_cru); } /* @@ -645,21 +1346,7 @@ proc_exit(proc_t p) } } - proc_spinlock(p); - if (thread_call_cancel(p->p_rcall)) - p->p_ractive--; - - while (p->p_ractive > 0) { - proc_spinunlock(p); - - delay(1); - - proc_spinlock(p); - } - proc_spinunlock(p); - - thread_call_free(p->p_rcall); - p->p_rcall = NULL; + proc_free_realitimer(p); /* * Other substructures are freed from wait(). @@ -673,6 +1360,7 @@ proc_exit(proc_t p) proc_limitdrop(p, 1); p->p_limit = NULL; + vm_purgeable_disown(p->task); /* * Finish up by terminating the task @@ -682,7 +1370,8 @@ proc_exit(proc_t p) p->task = TASK_NULL; set_bsdtask_info(task, NULL); - proc_knote(p, NOTE_EXIT); + knote_hint = NOTE_EXIT | (p->p_xstat & 0xffff); + proc_knote(p, knote_hint); /* mark the thread as the one that is doing proc_exit * no need to hold proc lock in uthread_free @@ -694,6 +1383,8 @@ proc_exit(proc_t p) pp = proc_parent(p); if (pp->p_flag & P_NOCLDWAIT) { + if (p->p_ru != NULL) { + proc_lock(pp); #if 3839178 /* * If the parent is ignoring SIGCHLD, then POSIX requires @@ -708,29 +1399,30 @@ proc_exit(proc_t p) * zombie to init. If we were unable to allocate a * zombie structure, this information is lost. */ - if (p->p_ru != NULL) { - proc_lock(pp); - ruadd(&pp->p_stats->p_cru, p->p_ru); + ruadd(&pp->p_stats->p_cru, &p->p_ru->ru); +#endif /* !3839178 */ + update_rusage_info_child(&pp->p_stats->ri_child, &p->p_ru->ri); proc_unlock(pp); } -#endif /* !3839178 */ - + /* kernel can reap this one, no need to move it to launchd */ proc_list_lock(); p->p_listflag |= P_LIST_DEADPARENT; proc_list_unlock(); } - if ((p->p_listflag & P_LIST_DEADPARENT) == 0) { + if ((p->p_listflag & P_LIST_DEADPARENT) == 0 || p->p_oppid) { if (pp != initproc) { proc_lock(pp); pp->si_pid = p->p_pid; + pp->p_xhighbits = p->p_xhighbits; + p->p_xhighbits = 0; pp->si_status = p->p_xstat; pp->si_code = CLD_EXITED; /* * p_ucred usage is safe as it is an exiting process * and reference is dropped in reap */ - pp->si_uid = p->p_ucred->cr_ruid; + pp->si_uid = kauth_cred_getruid(p->p_ucred); proc_unlock(pp); } /* mark as a zombie */ @@ -739,6 +1431,9 @@ proc_exit(proc_t p) * The write is to an int and is coherent. Also parent is * keyed off of list lock for reaping */ + KERNEL_DEBUG_CONSTANT_IST(KDEBUG_COMMON, + BSDDBG_CODE(DBG_BSD_PROC, BSD_PROC_EXIT) | DBG_FUNC_END, + pid, exitval, 0, 0, 0); p->p_stat = SZOMB; /* * The current process can be reaped so, no one @@ -760,13 +1455,17 @@ proc_exit(proc_t p) * keyed off of list lock for reaping */ proc_list_lock(); - p->p_stat = SZOMB; + KERNEL_DEBUG_CONSTANT_IST(KDEBUG_COMMON, + BSDDBG_CODE(DBG_BSD_PROC, BSD_PROC_EXIT) | DBG_FUNC_END, + pid, exitval, 0, 0, 0); /* check for sysctl zomb lookup */ while ((p->p_listflag & P_LIST_WAITING) == P_LIST_WAITING) { msleep(&p->p_stat, proc_list_mlock, PWAIT, "waitcoll", 0); } /* safe to use p as this is a system reap */ + p->p_stat = SZOMB; p->p_listflag |= P_LIST_WAITING; + /* * This is a named reference and it is not granted * if the reap is already in progress. So we get @@ -776,9 +1475,18 @@ proc_exit(proc_t p) * and the proc struct cannot be used for wakeups as well. * It is safe to use p here as this is system reap */ - (void)reap_child_locked(pp, p, 1, 1, 1); + (void)reap_child_locked(pp, p, 1, 0, 1, 1); /* list lock dropped by reap_child_locked */ } + if (uth->uu_lowpri_window) { + /* + * task is marked as a low priority I/O type and we've + * somehow picked up another throttle during exit processing... + * no need to throttle this thread since its going away + * but we do need to update our bookeeping w/r to throttled threads + */ + throttle_lowpri_io(0); + } proc_rele(pp); @@ -802,47 +1510,85 @@ proc_exit(proc_t p) * 1 Process was reaped */ static int -reap_child_locked(proc_t parent, proc_t child, int deadparent, int locked, int droplock) +reap_child_locked(proc_t parent, proc_t child, int deadparent, int reparentedtoinit, int locked, int droplock) { - proc_t trace_parent; /* Traced parent process, if tracing */ + proc_t trace_parent = PROC_NULL; /* Traced parent process, if tracing */ + if (locked == 1) + proc_list_unlock(); + /* * If we got the child via a ptrace 'attach', * we need to give it back to the old parent. + * + * Exception: someone who has been reparented to launchd before being + * ptraced can simply be reaped, refer to radar 5677288 + * p_oppid -> ptraced + * trace_parent == initproc -> away from launchd + * reparentedtoinit -> came to launchd by reparenting */ - if (locked == 1) - proc_list_unlock(); - if (child->p_oppid && (trace_parent = proc_find(child->p_oppid))) { + if (child->p_oppid) { + int knote_hint; + pid_t oppid; + proc_lock(child); + oppid = child->p_oppid; child->p_oppid = 0; + knote_hint = NOTE_EXIT | (child->p_xstat & 0xffff); proc_unlock(child); - if (trace_parent != initproc) { - /* - * proc internal fileds and p_ucred usage safe - * here as child is dead and is not reaped or - * reparented yet - */ - proc_lock(trace_parent); - trace_parent->si_pid = child->p_pid; - trace_parent->si_status = child->p_xstat; - trace_parent->si_code = CLD_CONTINUED; - trace_parent->si_uid = child->p_ucred->cr_ruid; - proc_unlock(trace_parent); - } - proc_reparentlocked(child, trace_parent, 1, 0); - psignal(trace_parent, SIGCHLD); - proc_list_lock(); - wakeup((caddr_t)trace_parent); - child->p_listflag &= ~P_LIST_WAITING; - wakeup(&child->p_stat); - proc_list_unlock(); - proc_rele(trace_parent); - if ((locked == 1) && (droplock == 0)) + + if ((trace_parent = proc_find(oppid)) + && !((trace_parent == initproc) && reparentedtoinit)) { + + if (trace_parent != initproc) { + /* + * proc internal fileds and p_ucred usage safe + * here as child is dead and is not reaped or + * reparented yet + */ + proc_lock(trace_parent); + trace_parent->si_pid = child->p_pid; + trace_parent->si_status = child->p_xstat; + trace_parent->si_code = CLD_CONTINUED; + trace_parent->si_uid = kauth_cred_getruid(child->p_ucred); + proc_unlock(trace_parent); + } + proc_reparentlocked(child, trace_parent, 1, 0); + + /* resend knote to original parent (and others) after reparenting */ + proc_knote(child, knote_hint); + + psignal(trace_parent, SIGCHLD); proc_list_lock(); - return (0); - } + wakeup((caddr_t)trace_parent); + child->p_listflag &= ~P_LIST_WAITING; + wakeup(&child->p_stat); + proc_list_unlock(); + proc_rele(trace_parent); + if ((locked == 1) && (droplock == 0)) + proc_list_lock(); + return (0); + } + /* + * If we can't reparent (e.g. the original parent exited while child was being debugged, or + * original parent is the same as the debugger currently exiting), we still need to satisfy + * the knote lifecycle for other observers on the system. While the debugger was attached, + * the NOTE_EXIT would not have been broadcast during initial child termination. + */ + proc_knote(child, knote_hint); + + if (trace_parent != PROC_NULL) { + proc_rele(trace_parent); + } + } + +#pragma clang diagnostic push +#pragma clang diagnostic ignored "-Wdeprecated-declarations" proc_knote(child, NOTE_REAP); +#pragma clang diagnostic pop + + proc_knote_drain(child); child->p_xstat = 0; if (child->p_ru) { @@ -857,7 +1603,8 @@ reap_child_locked(proc_t parent, proc_t child, int deadparent, int locked, int d */ if (!(parent->p_flag & P_NOCLDWAIT)) #endif /* 3839178 */ - ruadd(&parent->p_stats->p_cru, child->p_ru); + ruadd(&parent->p_stats->p_cru, &child->p_ru->ru); + update_rusage_info_child(&parent->p_stats->ri_child, &child->p_ru->ri); proc_unlock(parent); FREE_ZONE(child->p_ru, sizeof *child->p_ru, M_ZOMBIE); child->p_ru = NULL; @@ -865,19 +1612,24 @@ reap_child_locked(proc_t parent, proc_t child, int deadparent, int locked, int d printf("Warning : lost p_ru for %s\n", child->p_comm); } + AUDIT_SESSION_PROCEXIT(child); + /* * Decrement the count of procs running with this uid. * p_ucred usage is safe here as it is an exited process. * and refernce is dropped after these calls down below * (locking protection is provided by list lock held in chgproccnt) */ - (void)chgproccnt(child->p_ucred->cr_ruid, -1); - -#if CONFIG_LCTX - ALLLCTX_LOCK; - leavelctx(child); - ALLLCTX_UNLOCK; +#if CONFIG_PERSONAS + /* + * persona_proc_drop calls chgproccnt(-1) on the persona uid, + * and (+1) on the child->p_ucred uid + */ + persona_proc_drop(child); #endif + (void)chgproccnt(kauth_cred_getruid(child->p_ucred), -1); + + os_reason_free(child->p_exit_reason); /* * Free up credentials. @@ -910,15 +1662,33 @@ reap_child_locked(proc_t parent, proc_t child, int deadparent, int locked, int d proc_checkdeadrefs(child); nprocs--; + if (deadparent) { + /* + * If a child zombie is being reaped because its parent + * is exiting, make sure we update the list flag + */ + child->p_listflag |= P_LIST_DEADPARENT; + } + proc_list_unlock(); +#if CONFIG_FINE_LOCK_GROUPS + lck_mtx_destroy(&child->p_mlock, proc_mlock_grp); + lck_mtx_destroy(&child->p_ucred_mlock, proc_ucred_mlock_grp); + lck_mtx_destroy(&child->p_fdmlock, proc_fdmlock_grp); +#if CONFIG_DTRACE + lck_mtx_destroy(&child->p_dtrace_sprlock, proc_lck_grp); +#endif + lck_spin_destroy(&child->p_slock, proc_slock_grp); +#else /* CONFIG_FINE_LOCK_GROUPS */ lck_mtx_destroy(&child->p_mlock, proc_lck_grp); + lck_mtx_destroy(&child->p_ucred_mlock, proc_lck_grp); lck_mtx_destroy(&child->p_fdmlock, proc_lck_grp); #if CONFIG_DTRACE lck_mtx_destroy(&child->p_dtrace_sprlock, proc_lck_grp); #endif lck_spin_destroy(&child->p_slock, proc_lck_grp); - workqueue_destroy_lock(child); +#endif /* CONFIG_FINE_LOCK_GROUPS */ FREE_ZONE(child, sizeof *child, M_PROC); if ((locked == 1) && (droplock == 0)) @@ -931,34 +1701,44 @@ reap_child_locked(proc_t parent, proc_t child, int deadparent, int locked, int d int wait1continue(int result) { - void *vt; + proc_t p; thread_t thread; + uthread_t uth; + struct _wait4_data *wait4_data; + struct wait4_nocancel_args *uap; int *retval; - proc_t p; if (result) return(result); p = current_proc(); thread = current_thread(); - vt = get_bsduthreadarg(thread); - retval = get_bsduthreadrval(thread); - return(wait4(p, (struct wait4_args *)vt, retval)); + uth = (struct uthread *)get_bsdthread_info(thread); + + wait4_data = &uth->uu_kevent.uu_wait4_data; + uap = wait4_data->args; + retval = wait4_data->retval; + return(wait4_nocancel(p, uap, retval)); } int -wait4(proc_t q, struct wait4_args *uap, register_t *retval) +wait4(proc_t q, struct wait4_args *uap, int32_t *retval) { __pthread_testcancel(1); return(wait4_nocancel(q, (struct wait4_nocancel_args *)uap, retval)); } int -wait4_nocancel(proc_t q, struct wait4_nocancel_args *uap, register_t *retval) +wait4_nocancel(proc_t q, struct wait4_nocancel_args *uap, int32_t *retval) { int nfound; + int sibling_count; proc_t p; int status, error; + uthread_t uth; + struct _wait4_data *wait4_data; + + AUDIT_ARG(pid, uap->pid); if (uap->pid == 0) uap->pid = -q->p_pgrpid; @@ -967,7 +1747,11 @@ loop: proc_list_lock(); loop1: nfound = 0; - for (p = q->p_children.lh_first; p != 0; p = p->p_sibling.le_next) { + sibling_count = 0; + + PCHILDREN_FOREACH(q, p) { + if ( p->p_sibling.le_next != 0 ) + sibling_count++; if (uap->pid != WAIT_ANY && p->p_pid != uap->pid && p->p_pgrpid != -(uap->pid)) @@ -985,6 +1769,8 @@ loop1: if (p->p_stat == SZOMB) { + int reparentedtoinit = (p->p_listflag & P_LIST_DEADPARENT) ? 1 : 0; + proc_list_unlock(); #if CONFIG_MACF if ((error = mac_proc_check_wait(q, p)) != 0) @@ -1005,16 +1791,18 @@ loop1: error = ENOMEM; } else { if (IS_64BIT_PROCESS(q)) { - struct user_rusage my_rusage; - munge_rusage(p->p_ru, &my_rusage); + struct user64_rusage my_rusage; + munge_user64_rusage(&p->p_ru->ru, &my_rusage); error = copyout((caddr_t)&my_rusage, uap->rusage, sizeof (my_rusage)); } else { - error = copyout((caddr_t)p->p_ru, + struct user32_rusage my_rusage; + munge_user32_rusage(&p->p_ru->ru, &my_rusage); + error = copyout((caddr_t)&my_rusage, uap->rusage, - sizeof (struct rusage)); + sizeof (my_rusage)); } } /* information unavailable? */ @@ -1022,13 +1810,28 @@ loop1: goto out; } - /* Clean up */ - if (!reap_child_locked(q, p, 0, 0, 0)) { - proc_list_lock(); - p->p_listflag &= ~P_LIST_WAITING; - wakeup(&p->p_stat); - proc_list_unlock(); + /* Conformance change for 6577252. + * When SIGCHLD is blocked and wait() returns because the status + * of a child process is available and there are no other + * children processes, then any pending SIGCHLD signal is cleared. + */ + if ( sibling_count == 0 ) { + int mask = sigmask(SIGCHLD); + uth = current_uthread(); + + if ( (uth->uu_sigmask & mask) != 0 ) { + /* we are blocking SIGCHLD signals. clear any pending SIGCHLD. + * This locking looks funny but it is protecting access to the + * thread via p_uthlist. + */ + proc_lock(q); + uth->uu_siglist &= ~mask; /* clear pending signal */ + proc_unlock(q); + } } + + /* Clean up */ + (void)reap_child_locked(q, p, 0, reparentedtoinit, 0, 0); return (0); } @@ -1065,7 +1868,7 @@ loop1: #endif /* Prevent other process for waiting for this event */ - OSBitAndAtomic(~((uint32_t)P_CONTINUED), (UInt32 *)&p->p_flag); + OSBitAndAtomic(~((uint32_t)P_CONTINUED), &p->p_flag); retval[0] = p->p_pid; if (uap->status) { status = W_STOPCODE(SIGCONT); @@ -1091,6 +1894,12 @@ loop1: return (0); } + /* Save arguments for continuation. Backing storage is in uthread->uu_arg, and will not be deallocated */ + uth = current_uthread(); + wait4_data = &uth->uu_kevent.uu_wait4_data; + wait4_data->args = uap; + wait4_data->retval = retval; + if ((error = msleep0((caddr_t)q, proc_list_mlock, PWAIT | PCATCH | PDROP, "wait", 0, wait1continue))) return (error); @@ -1103,21 +1912,34 @@ out: return (error); } +#if DEBUG +#define ASSERT_LCK_MTX_OWNED(lock) \ + lck_mtx_assert(lock, LCK_MTX_ASSERT_OWNED) +#else +#define ASSERT_LCK_MTX_OWNED(lock) /* nothing */ +#endif int waitidcontinue(int result) { - void *vt; + proc_t p; thread_t thread; + uthread_t uth; + struct _waitid_data *waitid_data; + struct waitid_nocancel_args *uap; int *retval; if (result) - return(result); + return (result); + p = current_proc(); thread = current_thread(); - vt = get_bsduthreadarg(thread); - retval = get_bsduthreadrval(thread); - return(waitid(current_proc(), (struct waitid_args *)vt, retval)); + uth = (struct uthread *)get_bsdthread_info(thread); + + waitid_data = &uth->uu_kevent.uu_waitid_data; + uap = waitid_data->args; + retval = waitid_data->retval; + return(waitid_nocancel(p, uap, retval)); } /* @@ -1126,7 +1948,7 @@ waitidcontinue(int result) * * Parameters: uap->idtype one of P_PID, P_PGID, P_ALL * uap->id pid_t or gid_t or ignored - * uap->infop Address of signinfo_t struct in + * uap->infop Address of siginfo_t struct in * user space into which to return status * uap->options flag values * @@ -1134,36 +1956,29 @@ waitidcontinue(int result) * !0 Error returning status to user space */ int -waitid(proc_t q, struct waitid_args *uap, register_t *retval) +waitid(proc_t q, struct waitid_args *uap, int32_t *retval) { __pthread_testcancel(1); - return(waitid_nocancel(q, (struct waitid_nocancel_args *)uap, retval)); + return (waitid_nocancel(q, (struct waitid_nocancel_args *)uap, retval)); } int -waitid_nocancel(proc_t q, struct waitid_nocancel_args *uap, __unused register_t *retval) +waitid_nocancel(proc_t q, struct waitid_nocancel_args *uap, + __unused int32_t *retval) { - user_siginfo_t collect64; /* siginfo data to return to caller */ - + user_siginfo_t siginfo; /* siginfo data to return to caller */ + boolean_t caller64 = IS_64BIT_PROCESS(q); int nfound; proc_t p; int error; + uthread_t uth; + struct _waitid_data *waitid_data; - /* - * Forced validation of options for T.waitpid 21; should be a TSD! - * This will pass the test, but note that we have more bits than the - * standard specifies that we will allow in, in this case. The test - * passes because they light all the bits, not just the ones we allow, - * and so the following check returns EINVAL like the test wants. - */ - if (((uap->options & (WNOHANG|WNOWAIT|WCONTINUED|WUNTRACED|WSTOPPED|WEXITED)) != uap->options) || - (uap->options == 0)) + if (uap->options == 0 || + (uap->options & ~(WNOHANG|WNOWAIT|WCONTINUED|WSTOPPED|WEXITED))) return (EINVAL); /* bits set that aren't recognized */ - /* - * Overly critical options checking, per POSIX - */ - switch(uap->idtype) { + switch (uap->idtype) { case P_PID: /* child with process ID equal to... */ case P_PGID: /* child with process group ID equal to... */ if (((int)uap->id) < 0) @@ -1177,8 +1992,9 @@ loop: proc_list_lock(); loop1: nfound = 0; - for (p = q->p_children.lh_first; p != 0; p = p->p_sibling.le_next) { - switch(uap->idtype) { + + PCHILDREN_FOREACH(q, p) { + switch (uap->idtype) { case P_PID: /* child with process ID equal to... */ if (p->p_pid != (pid_t)uap->id) continue; @@ -1198,67 +2014,45 @@ loop1: * the single return for waited process guarantee. */ if (p->p_listflag & P_LIST_WAITING) { - (void)msleep(&p->p_stat, proc_list_mlock, PWAIT, "waitidcoll", 0); + (void) msleep(&p->p_stat, proc_list_mlock, + PWAIT, "waitidcoll", 0); goto loop1; } p->p_listflag |= P_LIST_WAITING; /* mark busy */ nfound++; - /* - * Types of processes we are interested in - * - * XXX Don't know what to do for WCONTINUED?!? - */ - switch(p->p_stat) { + bzero(&siginfo, sizeof (siginfo)); + + switch (p->p_stat) { case SZOMB: /* Exited */ if (!(uap->options & WEXITED)) break; - - /* drop the lock and the thread is going to return */ proc_list_unlock(); +#if CONFIG_MACF + if ((error = mac_proc_check_wait(q, p)) != 0) + goto out; +#endif + siginfo.si_signo = SIGCHLD; + siginfo.si_pid = p->p_pid; + siginfo.si_status = (WEXITSTATUS(p->p_xstat) & 0x00FFFFFF) | (((uint32_t)(p->p_xhighbits) << 24) & 0xFF000000); + p->p_xhighbits = 0; + if (WIFSIGNALED(p->p_xstat)) { + siginfo.si_code = WCOREDUMP(p->p_xstat) ? + CLD_DUMPED : CLD_KILLED; + } else + siginfo.si_code = CLD_EXITED; - /* Collect "siginfo" information for caller */ - collect64.si_signo = SIGCHLD; - collect64.si_code = 0; - collect64.si_errno = 0; - collect64.si_pid = 0; - collect64.si_uid = 0; - collect64.si_addr = 0; - collect64.si_status = WEXITSTATUS(p->p_xstat); - collect64.si_band = 0; - - if (IS_64BIT_PROCESS(p)) { - error = copyout((caddr_t)&collect64, - uap->infop, - sizeof(collect64)); - } else { - siginfo_t collect; - siginfo_64to32(&collect64,&collect); - error = copyout((caddr_t)&collect, - uap->infop, - sizeof(collect)); - } - /* information unavailable? */ - if (error) + if ((error = copyoutsiginfo(&siginfo, + caller64, uap->infop)) != 0) goto out; /* Prevent other process for waiting for this event? */ if (!(uap->options & WNOWAIT)) { - /* Clean up */ - if (!reap_child_locked(q, p, 0, 0, 0)) { - proc_list_lock(); - p->p_listflag &= ~P_LIST_WAITING; - wakeup(&p->p_stat); - proc_list_unlock(); - } - } else { - proc_list_lock(); - p->p_listflag &= ~P_LIST_WAITING; - proc_list_unlock(); + (void) reap_child_locked(q, p, 0, 0, 0, 0); + return (0); } - - return (0); + goto out; case SSTOP: /* Stopped */ /* @@ -1274,35 +2068,18 @@ loop1: */ if ((p->p_lflag & P_LWAITED) != 0) break; - - /* drop the lock and the thread is going to return */ proc_list_unlock(); +#if CONFIG_MACF + if ((error = mac_proc_check_wait(q, p)) != 0) + goto out; +#endif + siginfo.si_signo = SIGCHLD; + siginfo.si_pid = p->p_pid; + siginfo.si_status = p->p_xstat; /* signal number */ + siginfo.si_code = CLD_STOPPED; - /* Collect "siginfo" information for caller */ - collect64.si_signo = SIGCHLD; - collect64.si_code = 0; - collect64.si_errno = 0; - collect64.si_pid = 0; - collect64.si_uid = 0; - collect64.si_addr = 0; - proc_lock(p); - collect64.si_status = p->p_xstat; - proc_unlock(p); - collect64.si_band = 0; - - if (IS_64BIT_PROCESS(p)) { - error = copyout((caddr_t)&collect64, - uap->infop, - sizeof(collect64)); - } else { - siginfo_t collect; - siginfo_64to32(&collect64,&collect); - error = copyout((caddr_t)&collect, - uap->infop, - sizeof(collect)); - } - /* information unavailable? */ - if (error) + if ((error = copyoutsiginfo(&siginfo, + caller64, uap->infop)) != 0) goto out; /* Prevent other process for waiting for this event? */ @@ -1311,12 +2088,9 @@ loop1: p->p_lflag |= P_LWAITED; proc_unlock(p); } - - error = 0; goto out; - default: /* All others */ - /* ...meaning Continued */ + default: /* All other states => Continued */ if (!(uap->options & WCONTINUED)) break; @@ -1327,54 +2101,40 @@ loop1: */ if ((p->p_flag & P_CONTINUED) == 0) break; - - /* drop the lock and the thread is going to return */ proc_list_unlock(); - - /* Collect "siginfo" information for caller */ +#if CONFIG_MACF + if ((error = mac_proc_check_wait(q, p)) != 0) + goto out; +#endif + siginfo.si_signo = SIGCHLD; + siginfo.si_code = CLD_CONTINUED; proc_lock(p); - collect64.si_signo = SIGCHLD; - collect64.si_code = CLD_CONTINUED; - collect64.si_errno = 0; - collect64.si_pid = p->p_contproc; - collect64.si_uid = 0; - collect64.si_addr = 0; - collect64.si_status = p->p_xstat; - collect64.si_band = 0; + siginfo.si_pid = p->p_contproc; + siginfo.si_status = p->p_xstat; proc_unlock(p); - if (IS_64BIT_PROCESS(p)) { - error = copyout((caddr_t)&collect64, - uap->infop, - sizeof(collect64)); - } else { - siginfo_t collect; - siginfo_64to32(&collect64,&collect); - error = copyout((caddr_t)&collect, - uap->infop, - sizeof(collect)); - } - /* information unavailable? */ - if (error) + if ((error = copyoutsiginfo(&siginfo, + caller64, uap->infop)) != 0) goto out; /* Prevent other process for waiting for this event? */ if (!(uap->options & WNOWAIT)) { - OSBitAndAtomic(~((uint32_t)P_CONTINUED), (UInt32 *)&p->p_flag); + OSBitAndAtomic(~((uint32_t)P_CONTINUED), + &p->p_flag); } - - error = 0; goto out; } - /* LIST LOCK IS HELD HERE */ + ASSERT_LCK_MTX_OWNED(proc_list_mlock); + /* Not a process we are interested in; go on to next child */ - + p->p_listflag &= ~P_LIST_WAITING; wakeup(&p->p_stat); } + ASSERT_LCK_MTX_OWNED(proc_list_mlock); - /* list lock is always held */ /* No child processes that could possibly satisfy the request? */ + if (nfound == 0) { proc_list_unlock(); return (ECHILD); @@ -1382,10 +2142,30 @@ loop1: if (uap->options & WNOHANG) { proc_list_unlock(); +#if CONFIG_MACF + if ((error = mac_proc_check_wait(q, p)) != 0) + return (error); +#endif + /* + * The state of the siginfo structure in this case + * is undefined. Some implementations bzero it, some + * (like here) leave it untouched for efficiency. + * + * Thus the most portable check for "no matching pid with + * WNOHANG" is to store a zero into si_pid before + * invocation, then check for a non-zero value afterwards. + */ return (0); } - if ((error = msleep0((caddr_t)q, proc_list_mlock, PWAIT | PCATCH | PDROP, "waitid", 0, waitidcontinue))) + /* Save arguments for continuation. Backing storage is in uthread->uu_arg, and will not be deallocated */ + uth = current_uthread(); + waitid_data = &uth->uu_kevent.uu_waitid_data; + waitid_data->args = uap; + waitid_data->retval = retval; + + if ((error = msleep0(q, proc_list_mlock, + PWAIT | PCATCH | PDROP, "waitid", 0, waitidcontinue)) != 0) return (error); goto loop; @@ -1414,7 +2194,7 @@ proc_reparentlocked(proc_t child, proc_t parent, int cansignal, int locked) oldparent = child->p_pptr; #if __PROC_INTERNAL_DEBUG if (oldparent == PROC_NULL) - panic("proc_reparent: process %x does not have a parent\n", (unsigned int)child); + panic("proc_reparent: process %p does not have a parent\n", child); #endif LIST_REMOVE(child, p_sibling); @@ -1440,44 +2220,6 @@ proc_reparentlocked(proc_t child, proc_t parent, int cansignal, int locked) proc_list_lock(); } -/* - * Make the current process an "init" process, meaning - * that it doesn't have a parent, and that it won't be - * gunned down by kill(-1, 0). - */ -kern_return_t -init_process(__unused struct init_process_args *args) -{ - proc_t p = current_proc(); - - AUDIT_MACH_SYSCALL_ENTER(AUE_INITPROCESS); - if (suser(kauth_cred_get(), &p->p_acflag)) { - AUDIT_MACH_SYSCALL_EXIT(KERN_NO_ACCESS); - return(KERN_NO_ACCESS); - } - - if (p->p_pid != 1 && p->p_pgrpid != p->p_pid) - enterpgrp(p, p->p_pid, 0); - OSBitOrAtomic(P_SYSTEM, (UInt32 *)&p->p_flag); - - /* - * Take us out of the sibling chain, and - * out of our parent's child chain. - */ - proc_list_lock(); - LIST_REMOVE(p, p_sibling); - p->p_sibling.le_prev = NULL; - p->p_sibling.le_next = NULL; - p->p_pptr = kernproc; - p->p_ppid = 0; - proc_list_unlock(); - - - AUDIT_MACH_SYSCALL_EXIT(KERN_SUCCESS); - return(KERN_SUCCESS); -} - - /* * Exit: deallocate address space and other resources, change proc state * to zombie, and unlink proc from allproc and parent's lists. Save exit @@ -1541,6 +2283,10 @@ vfork_exit_internal(proc_t p, int rv, int forceexit) proc_list_lock(); +#if CONFIG_MEMORYSTATUS + memorystatus_remove(p, TRUE); +#endif + LIST_REMOVE(p, p_list); LIST_INSERT_HEAD(&zombproc, p, p_list); /* Place onto zombproc. */ /* will not be visible via proc_find */ @@ -1554,45 +2300,22 @@ vfork_exit_internal(proc_t p, int rv, int forceexit) p->p_sigignore = ~0; proc_unlock(p); - proc_spinlock(p); - if (thread_call_cancel(p->p_rcall)) - p->p_ractive--; - - while (p->p_ractive > 0) { - proc_spinunlock(p); - - delay(1); - - proc_spinlock(p); - } - proc_spinunlock(p); - - thread_call_free(p->p_rcall); - p->p_rcall = NULL; - ut->uu_siglist = 0; - vproc_exit(p); -} + /* begin vproc_exit */ -void -vproc_exit(proc_t p) -{ proc_t q; proc_t pp; - + vnode_t tvp; -#ifdef FIXME - struct task *task = p->task; -#endif + struct pgrp * pg; struct session *sessp; - boolean_t fstate; + struct rusage_superset *rup; /* XXX Zombie allocation may fail, in which case stats get lost */ - MALLOC_ZONE(p->p_ru, struct rusage *, - sizeof (*p->p_ru), M_ZOMBIE, M_WAITOK); - + MALLOC_ZONE(rup, struct rusage_superset *, + sizeof (*rup), M_ZOMBIE, M_WAITOK); proc_refdrain(p); @@ -1605,14 +2328,12 @@ vproc_exit(proc_t p) sessp = proc_session(p); if (SESS_LEADER(p, sessp)) { - /* Protected by funnel for tty accesses */ - fstate = thread_funnel_set(kernel_flock, TRUE); - if (sessp->s_ttyvp != NULLVP) { struct vnode *ttyvp; int ttyvid; + int cttyflag = 0; struct vfs_context context; - struct tty * tp; + struct tty *tp; /* * Controlling process. @@ -1620,51 +2341,72 @@ vproc_exit(proc_t p) * drain controlling terminal * and revoke access to controlling terminal. */ - tp = sessp->s_ttyp; - + session_lock(sessp); + tp = SESSION_TP(sessp); if ((tp != TTY_NULL) && (tp->t_session == sessp)) { - tty_pgsignal(tp, SIGHUP, 1); - (void) ttywait(tp); + session_unlock(sessp); + /* - * The tty could have been revoked - * if we blocked. + * We're going to SIGHUP the foreground process + * group. It can't change from this point on + * until the revoke is complete. + * The process group changes under both the tty + * lock and proc_list_lock but we need only one */ + tty_lock(tp); + ttysetpgrphup(tp); + tty_unlock(tp); - session_lock(sessp); - ttyvp = sessp->s_ttyvp; - ttyvid = sessp->s_ttyvid; - sessp->s_ttyvp = NULL; - sessp->s_ttyvid = 0; - sessp->s_ttyp = NULL; - sessp->s_ttypgrpid = NO_PID; - session_unlock(sessp); + tty_pgsignal(tp, SIGHUP, 1); - if ((ttyvp != NULLVP) && (vnode_getwithvid(ttyvp, ttyvid) == 0)) { - context.vc_thread = proc_thread(p); /* XXX */ - context.vc_ucred = kauth_cred_proc_ref(p); - VNOP_REVOKE(ttyvp, REVOKEALL, &context); - vnode_put(ttyvp); - kauth_cred_unref(&context.vc_ucred); - } - } else { session_lock(sessp); - ttyvp = sessp->s_ttyvp; - sessp->s_ttyvp = NULL; - sessp->s_ttyvid = 0; - sessp->s_ttyp = NULL; - sessp->s_ttypgrpid = NO_PID; - session_unlock(sessp); + tp = SESSION_TP(sessp); } - if (ttyvp) - vnode_rele(ttyvp); - /* - * s_ttyp is not zero'd; we use this to indicate - * that the session once had a controlling terminal. - * (for logging and informational purposes) - */ - } - (void) thread_funnel_set(kernel_flock, fstate); + cttyflag = sessp->s_flags & S_CTTYREF; + sessp->s_flags &= ~S_CTTYREF; + ttyvp = sessp->s_ttyvp; + ttyvid = sessp->s_ttyvid; + sessp->s_ttyvp = NULL; + sessp->s_ttyvid = 0; + sessp->s_ttyp = TTY_NULL; + sessp->s_ttypgrpid = NO_PID; + session_unlock(sessp); + + if ((ttyvp != NULLVP) && (vnode_getwithvid(ttyvp, ttyvid) == 0)) { + if (tp != TTY_NULL) { + tty_lock(tp); + (void) ttywait(tp); + tty_unlock(tp); + } + context.vc_thread = proc_thread(p); /* XXX */ + context.vc_ucred = kauth_cred_proc_ref(p); + VNOP_REVOKE(ttyvp, REVOKEALL, &context); + if (cttyflag) { + /* + * Release the extra usecount taken in cttyopen. + * usecount should be released after VNOP_REVOKE is called. + * This usecount was taken to ensure that + * the VNOP_REVOKE results in a close to + * the tty since cttyclose is a no-op. + */ + vnode_rele(ttyvp); + } + vnode_put(ttyvp); + kauth_cred_unref(&context.vc_ucred); + ttyvp = NULLVP; + } + if (tp) { + /* + * This is cleared even if not set. This is also done in + * spec_close to ensure that the flag is cleared. + */ + tty_lock(tp); + ttyclrpgrphup(tp); + tty_unlock(tp); + ttyfree(tp); + } + } session_lock(sessp); sessp->s_leader = NULL; session_unlock(sessp); @@ -1680,7 +2422,6 @@ vproc_exit(proc_t p) proc_list_lock(); proc_childdrainstart(p); while ((q = p->p_children.lh_first) != NULL) { - q->p_listflag |= P_LIST_DEADPARENT; if (q->p_stat == SZOMB) { if (p != q->p_pptr) panic("parent child linkage broken"); @@ -1698,18 +2439,39 @@ vproc_exit(proc_t p) * and the proc struct cannot be used for wakeups as well. * It is safe to use q here as this is system reap */ - (void)reap_child_locked(p, q, 1, 1, 0); + (void)reap_child_locked(p, q, 1, 0, 1, 0); } else { - proc_reparentlocked(q, initproc, 0, 1); /* * Traced processes are killed * since their existence means someone is messing up. */ if (q->p_lflag & P_LTRACED) { + struct proc *opp; + proc_list_unlock(); + + opp = proc_find(q->p_oppid); + if (opp != PROC_NULL) { + proc_list_lock(); + q->p_oppid = 0; + proc_list_unlock(); + proc_reparentlocked(q, opp, 0, 0); + proc_rele(opp); + } else { + /* original parent exited while traced */ + proc_list_lock(); + q->p_listflag |= P_LIST_DEADPARENT; + q->p_oppid = 0; + proc_list_unlock(); + proc_reparentlocked(q, initproc, 0, 0); + } + proc_lock(q); q->p_lflag &= ~P_LTRACED; + if (q->sigwait_thread) { + thread_t thread = q->sigwait_thread; + proc_unlock(q); /* * The sigwait_thread could be stopped at a @@ -1718,14 +2480,18 @@ vproc_exit(proc_t p) * the first thread in the task. So any attempts to kill * the process would result into a deadlock on q->sigwait. */ - thread_resume((thread_t)q->sigwait_thread); - clear_wait(q->sigwait_thread, THREAD_INTERRUPTED); - threadsignal((thread_t)q->sigwait_thread, SIGKILL, 0); - } else + thread_resume(thread); + clear_wait(thread, THREAD_INTERRUPTED); + threadsignal(thread, SIGKILL, 0, TRUE); + } else { proc_unlock(q); - + } + psignal(q, SIGKILL); proc_list_lock(); + } else { + q->p_listflag |= P_LIST_DEADPARENT; + proc_reparentlocked(q, initproc, 0, 1); } } } @@ -1747,26 +2513,25 @@ vproc_exit(proc_t p) * info and self times. If we were unable to allocate a zombie * structure, this information is lost. */ - /* No need for locking here as no one than this thread can access this */ - if (p->p_ru != NULL) { - *p->p_ru = p->p_stats->p_ru; - timerclear(&p->p_ru->ru_utime); - timerclear(&p->p_ru->ru_stime); + if (rup != NULL) { + rup->ru = p->p_stats->p_ru; + timerclear(&rup->ru.ru_utime); + timerclear(&rup->ru.ru_stime); #ifdef FIXME if (task) { - task_basic_info_data_t tinfo; + mach_task_basic_info_data_t tinfo; task_thread_times_info_data_t ttimesinfo; int task_info_stuff, task_ttimes_stuff; struct timeval ut,st; - task_info_stuff = TASK_BASIC_INFO_COUNT; - task_info(task, TASK_BASIC_INFO, + task_info_stuff = MACH_TASK_BASIC_INFO_COUNT; + task_info(task, MACH_TASK_BASIC_INFO, &tinfo, &task_info_stuff); - p->p_ru->ru_utime.tv_sec = tinfo.user_time.seconds; - p->p_ru->ru_utime.tv_usec = tinfo.user_time.microseconds; - p->p_ru->ru_stime.tv_sec = tinfo.system_time.seconds; - p->p_ru->ru_stime.tv_usec = tinfo.system_time.microseconds; + p->p_ru->ru.ru_utime.tv_sec = tinfo.user_time.seconds; + p->p_ru->ru.ru_utime.tv_usec = tinfo.user_time.microseconds; + p->p_ru->ru.ru_stime.tv_sec = tinfo.system_time.seconds; + p->p_ru->ru.ru_stime.tv_usec = tinfo.system_time.microseconds; task_ttimes_stuff = TASK_THREAD_TIMES_INFO_COUNT; task_info(task, TASK_THREAD_TIMES_INFO, @@ -1776,12 +2541,22 @@ vproc_exit(proc_t p) ut.tv_usec = ttimesinfo.user_time.microseconds; st.tv_sec = ttimesinfo.system_time.seconds; st.tv_usec = ttimesinfo.system_time.microseconds; - timeradd(&ut,&p->p_ru->ru_utime,&p->p_ru->ru_utime); - timeradd(&st,&p->p_ru->ru_stime,&p->p_ru->ru_stime); + timeradd(&ut,&p->p_ru->ru.ru_utime,&p->p_ru->ru.ru_utime); + timeradd(&st,&p->p_ru->ru.ru_stime,&p->p_ru->ru.ru_stime); } #endif /* FIXME */ - ruadd(p->p_ru, &p->p_stats->p_cru); + ruadd(&rup->ru, &p->p_stats->p_cru); + + gather_rusage_info(p, &rup->ri, RUSAGE_INFO_CURRENT); + rup->ri.ri_phys_footprint = 0; + rup->ri.ri_proc_exit_abstime = mach_absolute_time(); + + /* + * Now that we have filled in the rusage info, make it + * visible to an external observer via proc_pid_rusage(). + */ + p->p_ru = rup; } /* @@ -1800,6 +2575,12 @@ vproc_exit(proc_t p) } } +#if PSYNCH + pth_proc_hashdelete(p); +#endif /* PSYNCH */ + + proc_free_realitimer(p); + /* * Other substructures are freed from wait(). */ @@ -1827,13 +2608,15 @@ vproc_exit(proc_t p) if (pp != initproc) { proc_lock(pp); pp->si_pid = p->p_pid; + pp->p_xhighbits = p->p_xhighbits; + p->p_xhighbits = 0; pp->si_status = p->p_xstat; pp->si_code = CLD_EXITED; /* * p_ucred usage is safe as it is an exiting process * and reference is dropped in reap */ - pp->si_uid = p->p_ucred->cr_ruid; + pp->si_uid = kauth_cred_getruid(p->p_ucred); proc_unlock(pp); } /* mark as a zombie */ @@ -1853,12 +2636,13 @@ vproc_exit(proc_t p) proc_list_unlock(); } else { proc_list_lock(); - p->p_stat = SZOMB; /* check for lookups by zomb sysctl */ while ((p->p_listflag & P_LIST_WAITING) == P_LIST_WAITING) { msleep(&p->p_stat, proc_list_mlock, PWAIT, "waitcoll", 0); } + p->p_stat = SZOMB; p->p_listflag |= P_LIST_WAITING; + /* * This is a named reference and it is not granted * if the reap is already in progress. So we get @@ -1868,7 +2652,7 @@ vproc_exit(proc_t p) * and the proc struct cannot be used for wakeups as well. * It is safe to use p here as this is system reap */ - (void)reap_child_locked(pp, p, 0, 1, 1); + (void)reap_child_locked(pp, p, 0, 0, 1, 1); /* list lock dropped by reap_child_locked */ } proc_rele(pp); @@ -1878,11 +2662,11 @@ vproc_exit(proc_t p) /* * munge_rusage * LP64 support - long is 64 bits if we are dealing with a 64 bit user - * process. We munge the kernel (32 bit) version of rusage into the + * process. We munge the kernel version of rusage into the * 64 bit version. */ __private_extern__ void -munge_rusage(struct rusage *a_rusage_p, struct user_rusage *a_user_rusage_p) +munge_user64_rusage(struct rusage *a_rusage_p, struct user64_rusage *a_user_rusage_p) { /* timeval changes size, so utime and stime need special handling */ a_user_rusage_p->ru_utime.tv_sec = a_rusage_p->ru_utime.tv_sec; @@ -1908,3 +2692,32 @@ munge_rusage(struct rusage *a_rusage_p, struct user_rusage *a_user_rusage_p) a_user_rusage_p->ru_nvcsw = a_rusage_p->ru_nvcsw; a_user_rusage_p->ru_nivcsw = a_rusage_p->ru_nivcsw; } + +/* For a 64-bit kernel and 32-bit userspace, munging may be needed */ +__private_extern__ void +munge_user32_rusage(struct rusage *a_rusage_p, struct user32_rusage *a_user_rusage_p) +{ + /* timeval changes size, so utime and stime need special handling */ + a_user_rusage_p->ru_utime.tv_sec = a_rusage_p->ru_utime.tv_sec; + a_user_rusage_p->ru_utime.tv_usec = a_rusage_p->ru_utime.tv_usec; + a_user_rusage_p->ru_stime.tv_sec = a_rusage_p->ru_stime.tv_sec; + a_user_rusage_p->ru_stime.tv_usec = a_rusage_p->ru_stime.tv_usec; + /* + * everything else can be a direct assign. We currently ignore + * the loss of precision + */ + a_user_rusage_p->ru_maxrss = a_rusage_p->ru_maxrss; + a_user_rusage_p->ru_ixrss = a_rusage_p->ru_ixrss; + a_user_rusage_p->ru_idrss = a_rusage_p->ru_idrss; + a_user_rusage_p->ru_isrss = a_rusage_p->ru_isrss; + a_user_rusage_p->ru_minflt = a_rusage_p->ru_minflt; + a_user_rusage_p->ru_majflt = a_rusage_p->ru_majflt; + a_user_rusage_p->ru_nswap = a_rusage_p->ru_nswap; + a_user_rusage_p->ru_inblock = a_rusage_p->ru_inblock; + a_user_rusage_p->ru_oublock = a_rusage_p->ru_oublock; + a_user_rusage_p->ru_msgsnd = a_rusage_p->ru_msgsnd; + a_user_rusage_p->ru_msgrcv = a_rusage_p->ru_msgrcv; + a_user_rusage_p->ru_nsignals = a_rusage_p->ru_nsignals; + a_user_rusage_p->ru_nvcsw = a_rusage_p->ru_nvcsw; + a_user_rusage_p->ru_nivcsw = a_rusage_p->ru_nivcsw; +}