X-Git-Url: https://git.saurik.com/apple/xnu.git/blobdiff_plain/ccc36f2f2d89f9115c479db4439aa5c88de5b44a..5ba3f43ea354af8ad55bea84372a2bc834d8757c:/bsd/kern/kern_exit.c?ds=sidebyside diff --git a/bsd/kern/kern_exit.c b/bsd/kern/kern_exit.c index 157ec5f05..b8f5def79 100644 --- a/bsd/kern/kern_exit.c +++ b/bsd/kern/kern_exit.c @@ -1,23 +1,29 @@ /* - * Copyright (c) 2000-2004 Apple Computer, Inc. All rights reserved. + * Copyright (c) 2000-2016 Apple Inc. All rights reserved. * - * @APPLE_LICENSE_HEADER_START@ + * @APPLE_OSREFERENCE_LICENSE_HEADER_START@ * - * The contents of this file constitute Original Code as defined in and - * are subject to the Apple Public Source License Version 1.1 (the - * "License"). You may not use this file except in compliance with the - * License. Please obtain a copy of the License at - * http://www.apple.com/publicsource and read it before using this file. + * This file contains Original Code and/or Modifications of Original Code + * as defined in and that are subject to the Apple Public Source License + * Version 2.0 (the 'License'). You may not use this file except in + * compliance with the License. The rights granted to you under the License + * may not be used to create, or enable the creation or redistribution of, + * unlawful or unlicensed copies of an Apple operating system, or to + * circumvent, violate, or enable the circumvention or violation of, any + * terms of an Apple operating system software license agreement. * - * This Original Code and all software distributed under the License are - * distributed on an "AS IS" basis, WITHOUT WARRANTY OF ANY KIND, EITHER + * Please obtain a copy of the License at + * http://www.opensource.apple.com/apsl/ and read it before using this file. + * + * The Original Code and all software distributed under the License are + * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES, * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY, - * FITNESS FOR A PARTICULAR PURPOSE OR NON-INFRINGEMENT. Please see the - * License for the specific language governing rights and limitations - * under the License. + * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT. + * Please see the License for the specific language governing rights and + * limitations under the License. * - * @APPLE_LICENSE_HEADER_END@ + * @APPLE_OSREFERENCE_LICENSE_HEADER_END@ */ /* Copyright (c) 1995, 1997 Apple Computer, Inc. All Rights Reserved */ /* @@ -59,67 +65,622 @@ * * @(#)kern_exit.c 8.7 (Berkeley) 2/12/94 */ +/* + * NOTICE: This file was modified by SPARTA, Inc. in 2005 to introduce + * support for mandatory and extensible security protections. This notice + * is included in support of clause 2.2 (b) of the Apple Public License, + * Version 2.0. + */ #include #include +#include #include "compat_43.h" #include #include #include +#include #include +#include #include #include #include #include -#include #include -#include -#include +#include +#include #include #include #include #include +#include +#include +#include #include #include - -#include +#include +#include +#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 -#if KTRACE -#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] */ +void dtrace_proc_exit(proc_t p); +#include +#endif /* CONFIG_DTRACE */ +#if CONFIG_MACF +#include +#include +#include +#endif /* CONFIG_MACF */ + +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); +__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 proc_exit(proc_t p); +int wait1continue(int result); +int waitidcontinue(int result); +kern_return_t sys_perf_notify(thread_t thread, int pid); +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 + */ +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; + 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 = 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_EXPLICIT(user32_addr_t,in->si_value.sival_ptr); + out->si_band = in->si_band; /* range reduction */ +} + +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 -extern char init_task_failure_data[]; -int exit1 __P((struct proc *, int, int *)); -void proc_prepareexit(struct proc *p); -int vfork_exit(struct proc *p, int rv); -void vproc_exit(struct proc *p); + 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. */ -struct exit_args { - int rval; -}; +__attribute__((noreturn)) void -exit(p, uap, retval) - struct proc *p; - struct exit_args *uap; - int *retval; +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_funnel_set(kernel_flock, FALSE); thread_exception_return(); /* NOTREACHED */ while (TRUE) @@ -133,16 +694,29 @@ exit(p, uap, retval) * status and rusage for wait(). Check for child processes and orphan them. */ int -exit1(p, rv, retval) - register struct proc *p; - int rv; - int * retval; +exit1(proc_t p, int rv, int *retval) { - register struct proc *q, *nq; - thread_act_t self = current_act(); + 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; - register int i,s; struct uthread *ut; + int error = 0; /* * If a thread in this task has already @@ -151,114 +725,347 @@ exit1(p, rv, retval) */ ut = get_bsdthread_info(self); - if (ut->uu_flag & P_VFORK) { - if (!vfork_exit(p, rv)) { - vfork_return(self, p->p_pptr, p , retval); - unix_syscall_return(0); - /* NOT REACHED */ - } - return(EINVAL); + 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 */ } - AUDIT_SYSCALL_EXIT(0, p, ut); /* Exit is always successfull */ - signal_lock(p); + + /* + * The parameter list of audit_syscall_exit() was augmented to + * take the Darwin syscall number as the first parameter, + * 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) { - signal_unlock(p); + proc_unlock(p); return(0); } - signal_unlock(p); + proc_unlock(p); + thread_terminate(self); - thread_funnel_set(kernel_flock, FALSE); + if (!thread_can_terminate) { + return 0; + } + thread_exception_return(); /* NOTREACHED */ } sig_lock_to_exit(p); } - signal_unlock(p); - if (p->p_pid == 1) { - printf("pid 1 exited (signal %d, exit %d)", - WTERMSIG(rv), WEXITSTATUS(rv)); - panic("init died\nState at Last Exception:\n\n%s", - 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); } - s = splsched(); - p->p_flag |= P_WEXIT; - splx(s); - proc_prepareexit(p); + 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; - /* task terminate will call proc_terminate and that cleans it up */ + proc_transend(p, 1); + proc_unlock(p); + + proc_prepareexit(p, rv, perf_notify); + + /* Last thread to terminate will call proc_exit() */ task_terminate_internal(task); return(0); } void -proc_prepareexit(struct proc *p) +proc_prepareexit(proc_t p, int rv, boolean_t perf_notify) { - int s; + mach_exception_data_type_t code = 0, subcode = 0; + struct uthread *ut; - exception_data_t code[EXCEPTION_CODE_MAX]; - thread_act_t self = current_act(); + 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 */ + } - code[0] = 0xFF000001; /* Set terminate code */ - code[1] = p->p_pid; /* Pass out the pid */ - (void)sys_perf_notify(p->task, &code, 2); /* Notify the perf server */ + /* + * 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). + */ + if ((SIGSEGV == WTERMSIG(rv)) && + (p->p_pptr->p_lflag & P_LNOATTACH)) { + goto skipcheck; + } + /* + * Crash Reporter looks for the signal value, original exception + * type, and low 20 bits of the original code in code[0] + * (8, 4, and 20 bits respectively). code[1] is unmodified. + */ + code = ((WTERMSIG(rv) & 0xff) << 24) | + ((ut->uu_exception & 0x0f) << 20) | + ((int)ut->uu_code & 0xfffff); + subcode = ut->uu_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? */ + 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. * Not doing causes things like mount() find this on allproc * in partially cleaned state. */ + + 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. */ - LIST_REMOVE(p, p_hash); + /* will not be visible via proc_find */ + p->p_listflag |= P_LIST_EXITED; + + proc_list_unlock(); + #ifdef PGINPROF vmsizmon(); #endif /* * If parent is waiting for us to exit or exec, - * P_PPWAIT is set; we will wakeup the parent below. + * P_LPPWAIT is set; we will wakeup the parent below. */ - p->p_flag &= ~(P_TRACED | P_PPWAIT); - p->p_sigignore = ~0; - p->p_siglist = 0; - ut = get_bsdthread_info(self); + proc_lock(p); + p->p_lflag &= ~(P_LTRACED | P_LPPWAIT); + p->p_sigignore = ~(sigcantmask); ut->uu_siglist = 0; - untimeout(realitexpire, (caddr_t)p->p_pid); + proc_unlock(p); } void -proc_exit(struct proc *p) +proc_exit(proc_t p) { - register struct proc *q, *nq, *pp; + proc_t q; + proc_t pp; struct task *task = p->task; - register int i,s; - boolean_t funnel_state; + vnode_t tvp = NULLVP; + struct pgrp * pg; + struct session *sessp; + struct uthread * uth; + pid_t pid; + int exitval; + int knote_hint; + + 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, TRUE); + (void) task_terminate_internal(task); + proc_lock(p); + } else { + proc_transend(p, 1); + } - /* This can happen if thread_terminate of the single thread - * process - */ + p->p_lflag |= P_LPEXIT; - funnel_state = thread_funnel_set(kernel_flock, TRUE); - if( !(p->p_flag & P_WEXIT)) { - s = splsched(); - p->p_flag |= P_WEXIT; - splx(s); - proc_prepareexit(p); + /* + * 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_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--; } - MALLOC_ZONE(p->p_ru, struct rusage *, - sizeof (*p->p_ru), M_ZOMBIE, M_WAITOK); + 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 + + nspace_proc_exit(p); /* * need to cancel async IO requests that can be cancelled and wait for those * already active. MAY BLOCK! */ + + proc_refdrain(p); + + /* if any pending cpu limits action, clear it */ + task_clear_cpuusage(p->task, TRUE); + + workqueue_mark_exiting(p); + _aio_exit( p ); /* @@ -267,17 +1074,46 @@ proc_exit(struct proc *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) shmexit(p); +#endif +#if SYSV_SEM /* Release SYSV semaphores */ semexit(p); +#endif - if (SESS_LEADER(p)) { - register struct session *sp = p->p_session; +#if PSYNCH + pth_proc_hashdelete(p); +#endif /* PSYNCH */ + + sessp = proc_session(p); + if (SESS_LEADER(p, sessp)) { - if (sp->s_ttyvp) { + if (sessp->s_ttyvp != NULLVP) { struct vnode *ttyvp; + int ttyvid; + int cttyflag = 0; + struct vfs_context context; + struct tty *tp; /* * Controlling process. @@ -285,114 +1121,215 @@ proc_exit(struct proc *p) * drain controlling terminal * and revoke access to controlling terminal. */ - if (sp->s_ttyp->t_session == sp) { - if (sp->s_ttyp->t_pgrp) - pgsignal(sp->s_ttyp->t_pgrp, SIGHUP, 1); - (void) ttywait(sp->s_ttyp); + session_lock(sessp); + tp = SESSION_TP(sessp); + if ((tp != TTY_NULL) && (tp->t_session == sessp)) { + 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 */ - if (sp->s_ttyvp) - VOP_REVOKE(sp->s_ttyvp, REVOKEALL); + tty_lock(tp); + ttysetpgrphup(tp); + tty_unlock(tp); + + tty_pgsignal(tp, SIGHUP, 1); + + session_lock(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); } - ttyvp = sp->s_ttyvp; - sp->s_ttyvp = NULL; - if (ttyvp) - vrele(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) - */ } - sp->s_leader = NULL; + session_lock(sessp); + sessp->s_leader = NULL; + session_unlock(sessp); } + session_rele(sessp); + + pg = proc_pgrp(p); + fixjobc(p, pg, 0); + pg_rele(pg); - fixjobc(p, p->p_pgrp, 0); p->p_rlimit[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY; (void)acct_process(p); -#if KTRACE - /* - * release trace file - */ - p->p_traceflag = 0; /* don't trace the vrele() */ - if (p->p_tracep) { - struct vnode *tvp = p->p_tracep; - p->p_tracep = NULL; + proc_list_lock(); - if (UBCINFOEXISTS(tvp)) - ubc_rele(tvp); - vrele(tvp); + 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); } -#endif - q = p->p_children.lh_first; - if (q) /* only need this if any child is S_ZOMB */ - wakeup((caddr_t) initproc); - for (; q != 0; q = nq) { - nq = q->p_sibling.le_next; - proc_reparent(q, initproc); - /* - * Traced processes are killed - * since their existence means someone is messing up. - */ - if (q->p_flag & P_TRACED) { - q->p_flag &= ~P_TRACED; - if (q->sigwait_thread) { + /* wait till parentrefs are dropped and grant no more */ + proc_childdrainstart(p); + while ((q = p->p_children.lh_first) != NULL) { + 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"); + /* check for sysctl zomb lookup */ + while ((q->p_listflag & P_LIST_WAITING) == P_LIST_WAITING) { + msleep(&q->p_stat, proc_list_mlock, PWAIT, "waitcoll", 0); + } + q->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 + * 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 + * 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, reparentedtoinit, 1, 0); + } else { + /* + * Traced processes are killed + * since their existence means someone is messing up. + */ + if (q->p_lflag & P_LTRACED) { + struct proc *opp; + /* - * The sigwait_thread could be stopped at a - * breakpoint. Wake it up to kill. - * Need to do this as it could be a thread which is not - * the first thread in the task. So any attempts to kill - * the process would result into a deadlock on q->sigwait. + * 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. */ - thread_resume((thread_act_t)q->sigwait_thread); - clear_wait(q->sigwait_thread, THREAD_INTERRUPTED); - threadsignal((thread_act_t)q->sigwait_thread, SIGKILL, 0); + 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 + * breakpoint. Wake it up to kill. + * Need to do this as it could be a thread which is not + * the first thread in the task. So any attempts to kill + * the process would result into a deadlock on q->sigwait. + */ + 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); } - psignal(q, SIGKILL); } } + proc_childdrainend(p); + proc_list_unlock(); + +#if CONFIG_MACF /* - * Save exit status and final rusage info, adding in child rusage - * info and self times. + * Notify MAC policies that proc is dead. + * This should be replaced with proper label management + * (rdar://problem/32126399). */ - *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_data_t tinfo; - task_thread_times_info_data_t ttimesinfo; - int task_info_stuff, task_ttimes_stuff; - struct timeval ut,st; + mac_proc_notify_exit(p); +#endif - task_info_stuff = TASK_BASIC_INFO_COUNT; - task_info(task, 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; + /* + * Release reference to text vnode + */ + tvp = p->p_textvp; + p->p_textvp = NULL; + if (tvp != NULLVP) { + vnode_rele(tvp); + } - task_ttimes_stuff = TASK_THREAD_TIMES_INFO_COUNT; - task_info(task, TASK_THREAD_TIMES_INFO, - &ttimesinfo, &task_ttimes_stuff); + /* + * 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; - 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); + ruadd(&(p->p_ru->ru), &p->p_stats->p_cru); } - ruadd(p->p_ru, &p->p_stats->p_cru); - /* * Free up profiling buffers. */ @@ -405,25 +1342,25 @@ proc_exit(struct proc *p) for (; p1 != NULL; p1 = pn) { pn = p1->pr_next; - kfree((vm_offset_t)p1, sizeof *p1); + kfree(p1, sizeof *p1); } } + proc_free_realitimer(p); + /* * Other substructures are freed from wait(). */ - FREE_ZONE(p->p_stats, sizeof *p->p_stats, M_SUBPROC); + FREE_ZONE(p->p_stats, sizeof *p->p_stats, M_PSTATS); p->p_stats = NULL; - FREE_ZONE(p->p_sigacts, sizeof *p->p_sigacts, M_SUBPROC); + FREE_ZONE(p->p_sigacts, sizeof *p->p_sigacts, M_SIGACTS); p->p_sigacts = NULL; - if (--p->p_limit->p_refcnt == 0) - FREE_ZONE(p->p_limit, sizeof *p->p_limit, M_SUBPROC); + proc_limitdrop(p, 1); p->p_limit = NULL; - /* Free the auditing info */ - audit_proc_free(p); + vm_purgeable_disown(p->task); /* * Finish up by terminating the task @@ -431,353 +1368,856 @@ proc_exit(struct proc *p) * member of the task exiting). */ p->task = TASK_NULL; - //task->proc = NULL; set_bsdtask_info(task, NULL); - KNOTE(&p->p_klist, 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 + */ + uth->uu_flag |= UT_PROCEXIT; /* * Notify parent that we're gone. */ - if (p->p_pptr->p_flag & P_NOCLDWAIT) { - struct proc * pp = p->p_pptr; + 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 + * us to not add the resource usage to the parent process - + * we are only going to hand it off to init to get reaped. + * We should contest the standard in this case on the basis + * of RLIMIT_CPU. + */ +#else /* !3839178 */ /* * Add child resource usage to parent before giving - * zombie to init + * zombie to init. If we were unable to allocate a + * zombie structure, this information is lost. + */ + 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); + } + + /* 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 || 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 = kauth_cred_getruid(p->p_ucred); + proc_unlock(pp); + } + /* mark as a zombie */ + /* No need to take proc lock as all refs are drained and + * no one except parent (reaping ) can look at this. + * 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 + * can depend on this */ - ruadd(&p->p_pptr->p_stats->p_cru, p->p_ru); - proc_reparent(p, initproc); - /* If there are no more children wakeup parent */ - if (LIST_EMPTY(&pp->p_children)) - wakeup((caddr_t)pp); + psignal(pp, SIGCHLD); + + /* and now wakeup the parent */ + proc_list_lock(); + wakeup((caddr_t)pp); + proc_list_unlock(); + } else { + /* should be fine as parent proc would be initproc */ + /* mark as a zombie */ + /* No need to take proc lock as all refs are drained and + * no one except parent (reaping ) can look at this. + * The write is to an int and is coherent. Also parent is + * keyed off of list lock for reaping + */ + proc_list_lock(); + 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 + * 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 + * 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, 0, 1, 1); + /* list lock dropped by reap_child_locked */ } - /* should be fine as parent proc would be initproc */ - pp = p->p_pptr; - if (pp != initproc) { - pp->si_pid = p->p_pid; - pp->si_status = p->p_xstat; - pp->si_code = CLD_EXITED; - pp->si_uid = p->p_cred->p_ruid; + 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); } - psignal(pp, SIGCHLD); + proc_rele(pp); - /* mark as a zombie */ - p->p_stat = SZOMB; +} - /* and now wakeup the parent */ - wakeup((caddr_t)p->p_pptr); - (void) thread_funnel_set(kernel_flock, funnel_state); -} +/* + * reap_child_locked + * + * Description: Given a process from which all status information needed + * has already been extracted, if the process is a ptrace + * attach process, detach it and give it back to its real + * parent, else recover all resources remaining associated + * with it. + * + * Parameters: proc_t parent Parent of process being reaped + * proc_t child Process to reap + * + * Returns: 0 Process was not reaped because it + * came from an attach + * 1 Process was reaped + */ +static int +reap_child_locked(proc_t parent, proc_t child, int deadparent, int reparentedtoinit, int locked, int droplock) +{ + 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 (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 = 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(); + 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); + } -struct wait4_args { - int pid; - int *status; - int options; - struct rusage *rusage; -}; + /* + * 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 COMPAT_43 -int -owait(p, uap, retval) - struct proc *p; - void *uap; - int *retval; -{ - struct wait4_args *a; + 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 - a = (struct wait4_args *)get_bsduthreadarg(current_act()); + proc_knote_drain(child); - a->options = 0; - a->rusage = NULL; - a->pid = WAIT_ANY; - a->status = NULL; - return (wait1(p, a, retval, 1)); -} + child->p_xstat = 0; + if (child->p_ru) { + proc_lock(parent); +#if 3839178 + /* + * If the parent is ignoring SIGCHLD, then POSIX requires + * us to not add the resource usage to the parent process - + * we are only going to hand it off to init to get reaped. + * We should contest the standard in this case on the basis + * of RLIMIT_CPU. + */ + if (!(parent->p_flag & P_NOCLDWAIT)) +#endif /* 3839178 */ + 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; + } else { + printf("Warning : lost p_ru for %s\n", child->p_comm); + } -int -wait4(p, uap, retval) - struct proc *p; - struct wait4_args *uap; - int *retval; -{ - return (wait1(p, uap, retval, 0)); -} + 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) + */ +#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. + */ + if (IS_VALID_CRED(child->p_ucred)) { + kauth_cred_unref(&child->p_ucred); + } + + /* XXXX Note NOT SAFE TO USE p_ucred from this point onwards */ + + /* + * Finally finished with old proc entry. + * Unlink it from its process group and free it. + */ + leavepgrp(child); -struct owait3_args { - int *status; - int options; - struct rusage *rusage; -}; + proc_list_lock(); + LIST_REMOVE(child, p_list); /* off zombproc */ + parent->p_childrencnt--; + LIST_REMOVE(child, p_sibling); + /* If there are no more children wakeup parent */ + if ((deadparent != 0) && (LIST_EMPTY(&parent->p_children))) + wakeup((caddr_t)parent); /* with list lock held */ + child->p_listflag &= ~P_LIST_WAITING; + wakeup(&child->p_stat); + + /* Take it out of process hash */ + LIST_REMOVE(child, p_hash); + child->p_listflag &= ~P_LIST_INHASH; + 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; + } -int -owait3(p, uap, retval) - struct proc *p; - struct owait3_args *uap; - int *retval; -{ - struct wait4_args *a; + proc_list_unlock(); - a = (struct wait4_args *)get_bsduthreadarg(current_act()); +#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); +#endif /* CONFIG_FINE_LOCK_GROUPS */ - a->rusage = uap->rusage; - a->options = uap->options; - a->status = uap->status; - a->pid = WAIT_ANY; + FREE_ZONE(child, sizeof *child, M_PROC); + if ((locked == 1) && (droplock == 0)) + proc_list_lock(); - return (wait1(p, a, retval, 1)); + return (1); } -#else -#define wait1 wait4 -#endif int -wait1continue(result) +wait1continue(int result) { - void *vt; - thread_act_t thread; + proc_t p; + thread_t thread; + uthread_t uth; + struct _wait4_data *wait4_data; + struct wait4_nocancel_args *uap; int *retval; - struct proc *p; if (result) return(result); p = current_proc(); - thread = current_act(); - vt = (void *)get_bsduthreadarg(thread); - retval = (int *)get_bsduthreadrval(thread); - return(wait1((struct proc *)p, (struct wait4_args *)vt, retval, 0)); + thread = current_thread(); + 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 -wait1(q, uap, retval, compat) - register struct proc *q; - register struct wait4_args *uap; - register_t *retval; -#if COMPAT_43 - int compat; -#endif +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, int32_t *retval) { - register int nfound; - register struct proc *p, *t; + int nfound; + int sibling_count; + proc_t p; int status, error; - struct vnode *tvp; + uthread_t uth; + struct _wait4_data *wait4_data; + + AUDIT_ARG(pid, uap->pid); -retry: if (uap->pid == 0) - uap->pid = -q->p_pgid; + uap->pid = -q->p_pgrpid; 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_pgid != -(uap->pid)) + p->p_pgrpid != -(uap->pid)) continue; + nfound++; - if (p->p_flag & P_WAITING) { - (void)tsleep(&p->p_stat, PWAIT, "waitcoll", 0); - goto loop; + + /* XXX This is racy because we don't get the lock!!!! */ + + if (p->p_listflag & P_LIST_WAITING) { + (void)msleep(&p->p_stat, proc_list_mlock, PWAIT, "waitcoll", 0); + goto loop1; } - p->p_flag |= P_WAITING; /* only allow single thread to wait() */ + p->p_listflag |= P_LIST_WAITING; /* only allow single thread to wait() */ + if (p->p_stat == SZOMB) { - retval[0] = p->p_pid; -#if COMPAT_43 - if (compat) - retval[1] = p->p_xstat; - else + 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) + goto out; #endif + retval[0] = p->p_pid; if (uap->status) { - status = p->p_xstat; /* convert to int */ - if (error = copyout((caddr_t)&status, - (caddr_t)uap->status, - sizeof(status))) { - p->p_flag &= ~P_WAITING; - wakeup(&p->p_stat); - return (error); - } - } - if (uap->rusage && - (error = copyout((caddr_t)p->p_ru, - (caddr_t)uap->rusage, - sizeof (struct rusage)))) { - p->p_flag &= ~P_WAITING; - wakeup(&p->p_stat); - return (error); + /* Legacy apps expect only 8 bits of status */ + status = 0xffff & p->p_xstat; /* convert to int */ + error = copyout((caddr_t)&status, + uap->status, + sizeof(status)); + if (error) + goto out; } - /* - * If we got the child via a ptrace 'attach', - * we need to give it back to the old parent. - */ - if (p->p_oppid && (t = pfind(p->p_oppid))) { - p->p_oppid = 0; - proc_reparent(p, t); - if (t != initproc) { - t->si_pid = p->p_pid; - t->si_status = p->p_xstat; - t->si_code = CLD_CONTINUED; - t->si_uid = p->p_cred->p_ruid; + if (uap->rusage) { + if (p->p_ru == NULL) { + error = ENOMEM; + } else { + if (IS_64BIT_PROCESS(q)) { + 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 { + struct user32_rusage my_rusage; + munge_user32_rusage(&p->p_ru->ru, &my_rusage); + error = copyout((caddr_t)&my_rusage, + uap->rusage, + sizeof (my_rusage)); + } } - psignal(t, SIGCHLD); - wakeup((caddr_t)t); - p->p_flag &= ~P_WAITING; - wakeup(&p->p_stat); - return (0); - } - p->p_xstat = 0; - if (p->p_ru) { - ruadd(&q->p_stats->p_cru, p->p_ru); - FREE_ZONE(p->p_ru, sizeof *p->p_ru, M_ZOMBIE); - p->p_ru = NULL; - } else { - printf("Warning : lost p_ru for %s\n", p->p_comm); + /* information unavailable? */ + if (error) + goto out; } - /* - * Decrement the count of procs running with this uid. - */ - (void)chgproccnt(p->p_cred->p_ruid, -1); - - /* - * Free up credentials. + /* 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 (--p->p_cred->p_refcnt == 0) { - struct ucred *ucr = p->p_ucred; - struct pcred *pcr; - - if (ucr != NOCRED) { - p->p_ucred = NOCRED; - crfree(ucr); + 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); } - pcr = p->p_cred; - p->p_cred = NULL; - FREE_ZONE(pcr, sizeof *pcr, M_SUBPROC); } + + /* Clean up */ + (void)reap_child_locked(q, p, 0, reparentedtoinit, 0, 0); - /* - * Release reference to text vnode - */ - tvp = p->p_textvp; - p->p_textvp = NULL; - if (tvp) - vrele(tvp); - - /* - * Finally finished with old proc entry. - * Unlink it from its process group and free it. - */ - leavepgrp(p); - LIST_REMOVE(p, p_list); /* off zombproc */ - LIST_REMOVE(p, p_sibling); - p->p_flag &= ~P_WAITING; - FREE_ZONE(p, sizeof *p, M_PROC); - nprocs--; - wakeup(&p->p_stat); return (0); } - if (p->p_stat == SSTOP && (p->p_flag & P_WAITED) == 0 && - (p->p_flag & P_TRACED || uap->options & WUNTRACED)) { - p->p_flag |= P_WAITED; + if (p->p_stat == SSTOP && (p->p_lflag & P_LWAITED) == 0 && + (p->p_lflag & P_LTRACED || uap->options & WUNTRACED)) { + proc_list_unlock(); +#if CONFIG_MACF + if ((error = mac_proc_check_wait(q, p)) != 0) + goto out; +#endif + proc_lock(p); + p->p_lflag |= P_LWAITED; + proc_unlock(p); retval[0] = p->p_pid; -#if COMPAT_43 - if (compat) { - retval[1] = W_STOPCODE(p->p_xstat); - error = 0; + if (uap->status) { + status = W_STOPCODE(p->p_xstat); + error = copyout((caddr_t)&status, + uap->status, + sizeof(status)); } else + error = 0; + goto out; + } + /* + * If we are waiting for continued processses, and this + * process was continued + */ + if ((uap->options & WCONTINUED) && + (p->p_flag & P_CONTINUED)) { + proc_list_unlock(); +#if CONFIG_MACF + if ((error = mac_proc_check_wait(q, p)) != 0) + goto out; #endif + + /* Prevent other process for waiting for this event */ + OSBitAndAtomic(~((uint32_t)P_CONTINUED), &p->p_flag); + retval[0] = p->p_pid; if (uap->status) { - status = W_STOPCODE(p->p_xstat); + status = W_STOPCODE(SIGCONT); error = copyout((caddr_t)&status, - (caddr_t)uap->status, + uap->status, sizeof(status)); } else error = 0; - p->p_flag &= ~P_WAITING; - wakeup(&p->p_stat); - return (error); + goto out; } - p->p_flag &= ~P_WAITING; + p->p_listflag &= ~P_LIST_WAITING; wakeup(&p->p_stat); } - if (nfound == 0) + /* list lock is held when we get here any which way */ + if (nfound == 0) { + proc_list_unlock(); return (ECHILD); + } if (uap->options & WNOHANG) { retval[0] = 0; + proc_list_unlock(); return (0); } - if (error = tsleep0((caddr_t)q, PWAIT | PCATCH, "wait", 0, wait1continue)) + /* 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); goto loop; +out: + proc_list_lock(); + p->p_listflag &= ~P_LIST_WAITING; + wakeup(&p->p_stat); + proc_list_unlock(); + return (error); } -/* - * make process 'parent' the new parent of process 'child'. - */ -void -proc_reparent(child, parent) - register struct proc *child; - register struct proc *parent; +#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) { + proc_t p; + thread_t thread; + uthread_t uth; + struct _waitid_data *waitid_data; + struct waitid_nocancel_args *uap; + int *retval; - if (child->p_pptr == parent) - return; + if (result) + return (result); - LIST_REMOVE(child, p_sibling); - LIST_INSERT_HEAD(&parent->p_children, child, p_sibling); - child->p_pptr = parent; + p = current_proc(); + thread = current_thread(); + 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)); } /* - * 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). + * Description: Suspend the calling thread until one child of the process + * containing the calling thread changes state. + * + * Parameters: uap->idtype one of P_PID, P_PGID, P_ALL + * uap->id pid_t or gid_t or ignored + * uap->infop Address of siginfo_t struct in + * user space into which to return status + * uap->options flag values + * + * Returns: 0 Success + * !0 Error returning status to user space */ -kern_return_t -init_process(void) +int +waitid(proc_t q, struct waitid_args *uap, int32_t *retval) { - register struct proc *p = current_proc(); + __pthread_testcancel(1); + return (waitid_nocancel(q, (struct waitid_nocancel_args *)uap, retval)); +} - AUDIT_MACH_SYSCALL_ENTER(AUE_INITPROCESS); - if (suser(p->p_ucred, &p->p_acflag)) { - AUDIT_MACH_SYSCALL_EXIT(KERN_NO_ACCESS); - return(KERN_NO_ACCESS); +int +waitid_nocancel(proc_t q, struct waitid_nocancel_args *uap, + __unused int32_t *retval) +{ + 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; + + if (uap->options == 0 || + (uap->options & ~(WNOHANG|WNOWAIT|WCONTINUED|WSTOPPED|WEXITED))) + return (EINVAL); /* bits set that aren't recognized */ + + 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) + return (EINVAL); + break; + case P_ALL: /* any child */ + break; } - if (p->p_pid != 1 && p->p_pgid != p->p_pid) - enterpgrp(p, p->p_pid, 0); - p->p_flag |= P_SYSTEM; +loop: + proc_list_lock(); +loop1: + nfound = 0; + + 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; + break; + case P_PGID: /* child with process group ID equal to... */ + if (p->p_pgrpid != (pid_t)uap->id) + continue; + break; + case P_ALL: /* any child */ + break; + } - /* - * Take us out of the sibling chain, and - * out of our parent's child chain. - */ - LIST_REMOVE(p, p_sibling); - p->p_sibling.le_prev = NULL; - p->p_sibling.le_next = NULL; - p->p_pptr = kernproc; + /* XXX This is racy because we don't get the lock!!!! */ + + /* + * Wait collision; go to sleep and restart; used to maintain + * 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); + goto loop1; + } + p->p_listflag |= P_LIST_WAITING; /* mark busy */ + + nfound++; + + bzero(&siginfo, sizeof (siginfo)); + + switch (p->p_stat) { + case SZOMB: /* Exited */ + if (!(uap->options & WEXITED)) + break; + 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; + + if ((error = copyoutsiginfo(&siginfo, + caller64, uap->infop)) != 0) + goto out; + + /* Prevent other process for waiting for this event? */ + if (!(uap->options & WNOWAIT)) { + (void) reap_child_locked(q, p, 0, 0, 0, 0); + return (0); + } + goto out; + + case SSTOP: /* Stopped */ + /* + * If we are not interested in stopped processes, then + * ignore this one. + */ + if (!(uap->options & WSTOPPED)) + break; + + /* + * If someone has already waited it, we lost a race + * to be the one to return status. + */ + if ((p->p_lflag & P_LWAITED) != 0) + break; + 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; + + if ((error = copyoutsiginfo(&siginfo, + caller64, uap->infop)) != 0) + goto out; + + /* Prevent other process for waiting for this event? */ + if (!(uap->options & WNOWAIT)) { + proc_lock(p); + p->p_lflag |= P_LWAITED; + proc_unlock(p); + } + goto out; + + default: /* All other states => Continued */ + if (!(uap->options & WCONTINUED)) + break; + + /* + * If the flag isn't set, then this process has not + * been stopped and continued, or the status has + * already been reaped by another caller of waitid(). + */ + if ((p->p_flag & P_CONTINUED) == 0) + break; + proc_list_unlock(); +#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); + siginfo.si_pid = p->p_contproc; + siginfo.si_status = p->p_xstat; + proc_unlock(p); + + 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), + &p->p_flag); + } + goto out; + } + 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); + + /* No child processes that could possibly satisfy the request? */ + + if (nfound == 0) { + proc_list_unlock(); + return (ECHILD); + } + + 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); + } + + /* 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); - AUDIT_MACH_SYSCALL_EXIT(KERN_SUCCESS); - return(KERN_SUCCESS); + goto loop; +out: + proc_list_lock(); + p->p_listflag &= ~P_LIST_WAITING; + wakeup(&p->p_stat); + proc_list_unlock(); + return (error); } +/* + * make process 'parent' the new parent of process 'child'. + */ void -process_terminate_self(void) +proc_reparentlocked(proc_t child, proc_t parent, int cansignal, int locked) { - struct proc *p = current_proc(); + proc_t oldparent = PROC_NULL; - if (p != NULL) { - exit1(p, W_EXITCODE(0, SIGKILL), (int *)NULL); - /*NOTREACHED*/ - } + if (child->p_pptr == parent) + return; + + if (locked == 0) + proc_list_lock(); + + oldparent = child->p_pptr; +#if __PROC_INTERNAL_DEBUG + if (oldparent == PROC_NULL) + panic("proc_reparent: process %p does not have a parent\n", child); +#endif + + LIST_REMOVE(child, p_sibling); +#if __PROC_INTERNAL_DEBUG + if (oldparent->p_childrencnt == 0) + panic("process children count already 0\n"); +#endif + oldparent->p_childrencnt--; +#if __PROC_INTERNAL_DEBUG1 + if (oldparent->p_childrencnt < 0) + panic("process children count -ve\n"); +#endif + LIST_INSERT_HEAD(&parent->p_children, child, p_sibling); + parent->p_childrencnt++; + child->p_pptr = parent; + child->p_ppid = parent->p_pid; + + proc_list_unlock(); + + if ((cansignal != 0) && (initproc == parent) && (child->p_stat == SZOMB)) + psignal(initproc, SIGCHLD); + if (locked == 1) + proc_list_lock(); } /* @@ -786,31 +2226,53 @@ process_terminate_self(void) * status and rusage for wait(). Check for child processes and orphan them. */ -int -vfork_exit(p, rv) - struct proc *p; - int rv; +void +vfork_exit(proc_t p, int rv) +{ + vfork_exit_internal(p, rv, 0); +} + +void +vfork_exit_internal(proc_t p, int rv, int forceexit) { - register struct proc *q, *nq; - thread_act_t self = current_act(); + thread_t self = current_thread(); +#ifdef FIXME struct task *task = p->task; - register int i,s; +#endif struct uthread *ut; - exception_data_t code[EXCEPTION_CODE_MAX]; - ut = get_bsdthread_info(self); - if (p->exit_thread) { - return(1); - } - p->exit_thread = self; - - s = splsched(); - p->p_flag |= P_WEXIT; - splx(s); + /* + * If a thread in this task has already + * called exit(), then halt any others + * right here. + */ - code[0] = 0xFF000001; /* Set terminate code */ - code[1] = p->p_pid; /* Pass out the pid */ - (void)sys_perf_notify(p->task, &code, 2); /* Notify the perf server */ + ut = get_bsdthread_info(self); + + + proc_lock(p); + if ((p->p_lflag & P_LPEXIT) == P_LPEXIT) { + /* + * This happens when a parent exits/killed and vfork is in progress + * other threads. But shutdown code for ex has already called exit1() + */ + proc_unlock(p); + return; + } + p->p_lflag |= (P_LEXIT | P_LPEXIT); + proc_unlock(p); + + if (forceexit == 0) { + /* + * parent of a vfork child has already called exit() and the + * thread that has vfork in proress terminates. So there is no + * separate address space here and it has already been marked for + * termination. This was never covered before and could cause problems + * if we block here for outside code. + */ + /* Notify the perf server */ + (void)sys_perf_notify(self, p->p_pid); + } /* * Remove proc from allproc queue and from pidhash chain. @@ -818,36 +2280,44 @@ vfork_exit(p, rv) * Not doing causes things like mount() find this on allproc * in partially cleaned state. */ + + 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. */ - LIST_REMOVE(p, p_hash); - /* - * If parent is waiting for us to exit or exec, - * P_PPWAIT is set; we will wakeup the parent below. - */ - p->p_flag &= ~(P_TRACED | P_PPWAIT); + /* will not be visible via proc_find */ + p->p_listflag |= P_LIST_EXITED; + + proc_list_unlock(); + + proc_lock(p); + p->p_xstat = rv; + p->p_lflag &= ~(P_LTRACED | P_LPPWAIT); p->p_sigignore = ~0; - p->p_siglist = 0; + proc_unlock(p); ut->uu_siglist = 0; - untimeout(realitexpire, (caddr_t)p->p_pid); - p->p_xstat = rv; + /* begin vproc_exit */ - vproc_exit(p); - return(0); -} + proc_t q; + proc_t pp; -void -vproc_exit(struct proc *p) -{ - register struct proc *q, *nq, *pp; - struct task *task = p->task; - register int i,s; - boolean_t funnel_state; + vnode_t tvp; + + struct pgrp * pg; + struct session *sessp; + struct rusage_superset *rup; - MALLOC_ZONE(p->p_ru, struct rusage *, - sizeof (*p->p_ru), M_ZOMBIE, M_WAITOK); + /* XXX Zombie allocation may fail, in which case stats get lost */ + MALLOC_ZONE(rup, struct rusage_superset *, + sizeof (*rup), M_ZOMBIE, M_WAITOK); + + proc_refdrain(p); /* * Close open files and release open-file table. @@ -855,11 +2325,15 @@ vproc_exit(struct proc *p) */ fdfree(p); - if (SESS_LEADER(p)) { - register struct session *sp = p->p_session; - - if (sp->s_ttyvp) { + sessp = proc_session(p); + if (SESS_LEADER(p, sessp)) { + + if (sessp->s_ttyvp != NULLVP) { struct vnode *ttyvp; + int ttyvid; + int cttyflag = 0; + struct vfs_context context; + struct tty *tp; /* * Controlling process. @@ -867,99 +2341,197 @@ vproc_exit(struct proc *p) * drain controlling terminal * and revoke access to controlling terminal. */ - if (sp->s_ttyp->t_session == sp) { - if (sp->s_ttyp->t_pgrp) - pgsignal(sp->s_ttyp->t_pgrp, SIGHUP, 1); - (void) ttywait(sp->s_ttyp); + session_lock(sessp); + tp = SESSION_TP(sessp); + if ((tp != TTY_NULL) && (tp->t_session == sessp)) { + 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 */ - if (sp->s_ttyvp) - VOP_REVOKE(sp->s_ttyvp, REVOKEALL); + tty_lock(tp); + ttysetpgrphup(tp); + tty_unlock(tp); + + tty_pgsignal(tp, SIGHUP, 1); + + session_lock(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 = 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); } - ttyvp = sp->s_ttyvp; - sp->s_ttyvp = NULL; - if (ttyvp) - vrele(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) - */ } - sp->s_leader = NULL; + session_lock(sessp); + sessp->s_leader = NULL; + session_unlock(sessp); } + session_rele(sessp); + + pg = proc_pgrp(p); + fixjobc(p, pg, 0); + pg_rele(pg); - fixjobc(p, p->p_pgrp, 0); p->p_rlimit[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY; -#if KTRACE - /* - * release trace file - */ - p->p_traceflag = 0; /* don't trace the vrele() */ - if (p->p_tracep) { - struct vnode *tvp = p->p_tracep; - p->p_tracep = NULL; + proc_list_lock(); + proc_childdrainstart(p); + while ((q = p->p_children.lh_first) != NULL) { + if (q->p_stat == SZOMB) { + if (p != q->p_pptr) + panic("parent child linkage broken"); + /* check for lookups by zomb sysctl */ + while ((q->p_listflag & P_LIST_WAITING) == P_LIST_WAITING) { + msleep(&q->p_stat, proc_list_mlock, PWAIT, "waitcoll", 0); + } + q->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 + * 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 + * 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, 0, 1, 0); + } else { + /* + * 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); + } - if (UBCINFOEXISTS(tvp)) - ubc_rele(tvp); - vrele(tvp); - } -#endif + 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 + * breakpoint. Wake it up to kill. + * Need to do this as it could be a thread which is not + * the first thread in the task. So any attempts to kill + * the process would result into a deadlock on q->sigwait. + */ + thread_resume(thread); + clear_wait(thread, THREAD_INTERRUPTED); + threadsignal(thread, SIGKILL, 0, TRUE); + } else { + proc_unlock(q); + } - q = p->p_children.lh_first; - if (q) /* only need this if any child is S_ZOMB */ - wakeup((caddr_t) initproc); - for (; q != 0; q = nq) { - nq = q->p_sibling.le_next; - proc_reparent(q, initproc); - /* - * Traced processes are killed - * since their existence means someone is messing up. - */ - if (q->p_flag & P_TRACED) { - q->p_flag &= ~P_TRACED; - if (q->sigwait_thread) { - /* - * The sigwait_thread could be stopped at a - * breakpoint. Wake it up to kill. - * Need to do this as it could be a thread which is not - * the first thread in the task. So any attempts to kill - * the process would result into a deadlock on q->sigwait. - */ - thread_resume((thread_act_t)q->sigwait_thread); - clear_wait(q->sigwait_thread, THREAD_INTERRUPTED); - threadsignal((thread_act_t)q->sigwait_thread, SIGKILL, 0); + psignal(q, SIGKILL); + proc_list_lock(); + } else { + q->p_listflag |= P_LIST_DEADPARENT; + proc_reparentlocked(q, initproc, 0, 1); } - psignal(q, SIGKILL); } } + proc_childdrainend(p); + proc_list_unlock(); + /* - * Save exit status and final rusage info, adding in child rusage - * info and self times. + * Release reference to text vnode */ - *p->p_ru = p->p_stats->p_ru; + tvp = p->p_textvp; + p->p_textvp = NULL; + if (tvp != NULLVP) { + vnode_rele(tvp); + } - timerclear(&p->p_ru->ru_utime); - timerclear(&p->p_ru->ru_stime); + /* + * 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 (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; + if (task) { + 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, @@ -969,12 +2541,23 @@ vproc_exit(struct proc *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; + } /* * Free up profiling buffers. @@ -988,21 +2571,26 @@ vproc_exit(struct proc *p) for (; p1 != NULL; p1 = pn) { pn = p1->pr_next; - kfree((vm_offset_t)p1, sizeof *p1); + kfree(p1, sizeof *p1); } } +#if PSYNCH + pth_proc_hashdelete(p); +#endif /* PSYNCH */ + + proc_free_realitimer(p); + /* * Other substructures are freed from wait(). */ - FREE_ZONE(p->p_stats, sizeof *p->p_stats, M_SUBPROC); + FREE_ZONE(p->p_stats, sizeof *p->p_stats, M_PSTATS); p->p_stats = NULL; - FREE_ZONE(p->p_sigacts, sizeof *p->p_sigacts, M_SUBPROC); + FREE_ZONE(p->p_sigacts, sizeof *p->p_sigacts, M_SIGACTS); p->p_sigacts = NULL; - if (--p->p_limit->p_refcnt == 0) - FREE_ZONE(p->p_limit, sizeof *p->p_limit, M_SUBPROC); + proc_limitdrop(p, 1); p->p_limit = NULL; /* @@ -1015,18 +2603,121 @@ vproc_exit(struct proc *p) /* * Notify parent that we're gone. */ - pp = p->p_pptr; - if (pp != initproc) { - pp->si_pid = p->p_pid; - pp->si_status = p->p_xstat; - pp->si_code = CLD_EXITED; - pp->si_uid = p->p_cred->p_ruid; + pp = proc_parent(p); + if ((p->p_listflag & P_LIST_DEADPARENT) == 0) { + 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 = kauth_cred_getruid(p->p_ucred); + proc_unlock(pp); + } + /* mark as a zombie */ + /* mark as a zombie */ + /* No need to take proc lock as all refs are drained and + * no one except parent (reaping ) can look at this. + * The write is to an int and is coherent. Also parent is + * keyed off of list lock for reaping + */ + p->p_stat = SZOMB; + + psignal(pp, SIGCHLD); + + /* and now wakeup the parent */ + proc_list_lock(); + wakeup((caddr_t)pp); + proc_list_unlock(); + } else { + proc_list_lock(); + /* 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 + * 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 + * 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, 0, 1, 1); + /* list lock dropped by reap_child_locked */ } - psignal(p->p_pptr, SIGCHLD); + proc_rele(pp); +} - /* mark as a zombie */ - p->p_stat = SZOMB; + +/* + * munge_rusage + * LP64 support - long is 64 bits if we are dealing with a 64 bit user + * process. We munge the kernel version of rusage into the + * 64 bit version. + */ +__private_extern__ void +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; + 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, since there is no loss + * of precision implied boing 32->64. + */ + 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; +} - /* and now wakeup the parent */ - wakeup((caddr_t)p->p_pptr); +/* 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; }