X-Git-Url: https://git.saurik.com/apple/xnu.git/blobdiff_plain/b0d623f7f2ae71ed96e60569f61f9a9a27016e80..22ba694c5857e62b5a553b1505dcf2e509177f28:/bsd/kern/kern_fork.c diff --git a/bsd/kern/kern_fork.c b/bsd/kern/kern_fork.c index 666748b52..da7a5395c 100644 --- a/bsd/kern/kern_fork.c +++ b/bsd/kern/kern_fork.c @@ -109,6 +109,7 @@ extern void dtrace_lazy_dofs_duplicate(proc_t, proc_t); #include #include #include +#include #include #include @@ -129,6 +130,9 @@ extern void dtrace_lazy_dofs_duplicate(proc_t, proc_t); #include +#if CONFIG_MEMORYSTATUS +#include +#endif /* XXX routines which should have Mach prototypes, but don't */ void thread_set_parent(thread_t parent, int pid); @@ -137,7 +141,7 @@ void thread_set_child(thread_t child, int pid); void *act_thread_csave(void); -thread_t cloneproc(task_t, proc_t, int); +thread_t cloneproc(task_t, proc_t, int, int); proc_t forkproc(proc_t); void forkproc_free(proc_t); thread_t fork_create_child(task_t parent_task, proc_t child, int inherit_memory, int is64bit); @@ -159,8 +163,8 @@ void proc_vfork_end(proc_t parent_proc); * Notes: Although this function increments a count, a count in * excess of 1 is not currently supported. According to the * POSIX standard, calling anything other than execve() or - * _exit() fillowing a vfork(), including calling vfork() - * itself again, will result in undefned behaviour + * _exit() following a vfork(), including calling vfork() + * itself again, will result in undefined behaviour */ void proc_vfork_begin(proc_t parent_proc) @@ -180,7 +184,7 @@ proc_vfork_begin(proc_t parent_proc) * * Returns: (void) * - * Notes: Decerements the count; currently, reentrancy of vfork() + * Notes: Decrements the count; currently, reentrancy of vfork() * is unsupported on the current process */ void @@ -190,7 +194,6 @@ proc_vfork_end(proc_t parent_proc) parent_proc->p_vforkcnt--; if (parent_proc->p_vforkcnt < 0) panic("vfork cnt is -ve"); - /* resude the vfork count; clear the flag when it goes to 0 */ if (parent_proc->p_vforkcnt == 0) parent_proc->p_lflag &= ~P_LVFORK; proc_unlock(parent_proc); @@ -278,12 +281,8 @@ vfork(proc_t parent_proc, __unused struct vfork_args *uap, int32_t *retval) if ((err = fork1(parent_proc, &child_thread, PROC_CREATE_VFORK)) != 0) { retval[1] = 0; } else { - /* - * kludge: rely on uu_proc being set in the vfork case, - * rather than returning the actual thread. We can remove - * this when we remove the uu_proc/current_proc() kludge. - */ - proc_t child_proc = current_proc(); + uthread_t ut = get_bsdthread_info(current_thread()); + proc_t child_proc = ut->uu_proc; retval[0] = child_proc->p_pid; retval[1] = 1; /* flag child return for user space */ @@ -296,12 +295,12 @@ vfork(proc_t parent_proc, __unused struct vfork_args *uap, int32_t *retval) proc_signalend(child_proc, 0); proc_transend(child_proc, 0); - /* flag the fork has occurred */ proc_knote(parent_proc, NOTE_FORK | child_proc->p_pid); DTRACE_PROC1(create, proc_t, child_proc); + ut->uu_flag &= ~UT_VFORKING; } - return(err); + return (err); } @@ -365,7 +364,7 @@ fork1(proc_t parent_proc, thread_t *child_threadp, int kind) * exceed the limit. The variable nprocs is the current number of * processes, maxproc is the limit. */ - uid = kauth_cred_get()->cr_ruid; + uid = kauth_getruid(); proc_list_lock(); if ((nprocs >= maxproc - 1 && uid != 0) || nprocs >= maxproc) { proc_list_unlock(); @@ -466,7 +465,6 @@ fork1(proc_t parent_proc, thread_t *child_threadp, int kind) AUDIT_ARG(pid, child_proc->p_pid); - AUDIT_SESSION_PROCNEW(child_proc->p_ucred); // XXX END: wants to move to be common code (and safe) /* @@ -482,7 +480,20 @@ fork1(proc_t parent_proc, thread_t *child_threadp, int kind) child_proc->p_vforkact = parent_thread; child_proc->p_stat = SRUN; - parent_uthread->uu_flag |= UT_VFORK; + /* + * Until UT_VFORKING is cleared at the end of the vfork + * syscall, the process identity of this thread is slightly + * murky. + * + * As long as UT_VFORK and it's associated field (uu_proc) + * is set, current_proc() will always return the child process. + * + * However dtrace_proc_selfpid() returns the parent pid to + * ensure that e.g. the proc:::create probe actions accrue + * to the parent. (Otherwise the child magically seems to + * have created itself!) + */ + parent_uthread->uu_flag |= UT_VFORK | UT_VFORKING; parent_uthread->uu_proc = child_proc; parent_uthread->uu_userstate = (void *)act_thread_csave(); parent_uthread->uu_vforkmask = parent_uthread->uu_sigmask; @@ -527,7 +538,7 @@ fork1(proc_t parent_proc, thread_t *child_threadp, int kind) * will, in effect, create a duplicate of it, with only minor * differences. Contrarily, spawned processes do not inherit. */ - if ((child_thread = cloneproc(parent_proc->task, parent_proc, spawn ? FALSE : TRUE)) == NULL) { + if ((child_thread = cloneproc(parent_proc->task, parent_proc, spawn ? FALSE : TRUE, FALSE)) == NULL) { /* Failed to create thread */ err = EAGAIN; goto bad; @@ -570,7 +581,6 @@ fork1(proc_t parent_proc, thread_t *child_threadp, int kind) AUDIT_ARG(pid, child_proc->p_pid); - AUDIT_SESSION_PROCNEW(child_proc->p_ucred); // XXX END: wants to move to be common code (and safe) /* @@ -674,49 +684,53 @@ bad: * this is done by reassociating the parent process structure * with the task, thread, and uthread. * + * Refer to the ASCII art above vfork() to figure out the + * state we're undoing. + * * Parameters: child_proc Child process * retval System call return value array * rval Return value to present to parent * * Returns: void * - * Note: The caller resumes or exits the parent, as appropriate, after - * callling this function. + * Notes: The caller resumes or exits the parent, as appropriate, after + * calling this function. */ void vfork_return(proc_t child_proc, int32_t *retval, int rval) { - proc_t parent_proc = child_proc->p_pptr; - thread_t parent_thread = (thread_t)current_thread(); - uthread_t parent_uthread = (uthread_t)get_bsdthread_info(parent_thread); + task_t parent_task = get_threadtask(child_proc->p_vforkact); + proc_t parent_proc = get_bsdtask_info(parent_task); + thread_t th = current_thread(); + uthread_t uth = get_bsdthread_info(th); + act_thread_catt(uth->uu_userstate); - act_thread_catt(parent_uthread->uu_userstate); - - /* end vfork in parent */ + /* clear vfork state in parent proc structure */ proc_vfork_end(parent_proc); /* REPATRIATE PARENT TASK, THREAD, UTHREAD */ - parent_uthread->uu_userstate = 0; - parent_uthread->uu_flag &= ~UT_VFORK; + uth->uu_userstate = 0; + uth->uu_flag &= ~UT_VFORK; /* restore thread-set-id state */ - if (parent_uthread->uu_flag & UT_WASSETUID) { - parent_uthread->uu_flag |= UT_SETUID; - parent_uthread->uu_flag &= UT_WASSETUID; + if (uth->uu_flag & UT_WASSETUID) { + uth->uu_flag |= UT_SETUID; + uth->uu_flag &= UT_WASSETUID; } - parent_uthread->uu_proc = 0; - parent_uthread->uu_sigmask = parent_uthread->uu_vforkmask; - child_proc->p_lflag &= ~P_LINVFORK; - child_proc->p_vforkact = (void *)0; + uth->uu_proc = 0; + uth->uu_sigmask = uth->uu_vforkmask; + + proc_lock(child_proc); + child_proc->p_lflag &= ~P_LINVFORK; + child_proc->p_vforkact = 0; + proc_unlock(child_proc); - thread_set_parent(parent_thread, rval); + thread_set_parent(th, rval); if (retval) { retval[0] = rval; retval[1] = 0; /* mark parent */ } - - return; } @@ -762,7 +776,8 @@ fork_create_child(task_t parent_task, proc_t child_proc, int inherit_memory, int is64bit, &child_task); if (result != KERN_SUCCESS) { - printf("execve: task_create_internal failed. Code: %d\n", result); + printf("%s: task_create_internal failed. Code: %d\n", + __func__, result); goto bad; } @@ -798,10 +813,17 @@ fork_create_child(task_t parent_task, proc_t child_proc, int inherit_memory, int /* Create a new thread for the child process */ result = thread_create(child_task, &child_thread); if (result != KERN_SUCCESS) { - printf("execve: thread_create failed. Code: %d\n", result); + printf("%s: thread_create failed. Code: %d\n", + __func__, result); task_deallocate(child_task); child_task = NULL; } + + /* + * Tag thread as being the first thread in its task. + */ + thread_set_tag(child_thread, THREAD_TAG_MAINTHREAD); + bad: thread_yield_internal(1); @@ -901,6 +923,8 @@ fork(proc_t parent_proc, __unused struct fork_args *uap, int32_t *retval) * memory from the parent; if this is * non-NULL, then the parent_task must * also be non-NULL + * memstat_internal Whether to track the process in the + * jetsam priority list (if configured) * * Returns: !NULL pointer to new child thread * NULL Failure (unspecified) @@ -922,8 +946,11 @@ fork(proc_t parent_proc, __unused struct fork_args *uap, int32_t *retval) * live with this being somewhat awkward. */ thread_t -cloneproc(task_t parent_task, proc_t parent_proc, int inherit_memory) +cloneproc(task_t parent_task, proc_t parent_proc, int inherit_memory, int memstat_internal) { +#if !CONFIG_MEMORYSTATUS +#pragma unused(memstat_internal) +#endif task_t child_task; proc_t child_proc; thread_t child_thread = NULL; @@ -948,19 +975,19 @@ cloneproc(task_t parent_task, proc_t parent_proc, int inherit_memory) if (parent_proc->p_flag & P_LP64) { task_set_64bit(child_task, TRUE); OSBitOrAtomic(P_LP64, (UInt32 *)&child_proc->p_flag); -#ifdef __ppc__ - /* - * PPC51: ppc64 is limited to 51-bit addresses. - * Memory above that limit is handled specially at - * the pmap level. - */ - pmap_map_sharedpage(child_task, get_map_pmap(get_task_map(child_task))); -#endif /* __ppc__ */ } else { task_set_64bit(child_task, FALSE); OSBitAndAtomic(~((uint32_t)P_LP64), (UInt32 *)&child_proc->p_flag); } +#if CONFIG_MEMORYSTATUS + if (memstat_internal) { + proc_list_lock(); + child_proc->p_memstat_state |= P_MEMSTAT_INTERNAL; + proc_list_unlock(); + } +#endif + /* make child visible */ pinsertchild(parent_proc, child_proc); @@ -1031,6 +1058,9 @@ forkproc_free(proc_t p) /* Stop the profiling clock */ stopprofclock(p); + /* Update the audit session proc count */ + AUDIT_SESSION_PROCEXIT(p); + /* Release the credential reference */ kauth_cred_unref(&p->p_ucred); @@ -1069,6 +1099,7 @@ forkproc(proc_t parent_proc) { proc_t child_proc; /* Our new process */ static int nextpid = 0, pidwrap = 0, nextpidversion = 0; + static uint64_t nextuniqueid = 0; int error = 0; struct session *sessp; uthread_t parent_uthread = (uthread_t)get_bsdthread_info(current_thread()); @@ -1147,6 +1178,8 @@ retry: nprocs++; child_proc->p_pid = nextpid; child_proc->p_idversion = nextpidversion++; + /* kernel process is handcrafted and not from fork, so start from 1 */ + child_proc->p_uniqueid = ++nextuniqueid; #if 1 if (child_proc->p_pid != 0) { if (pfind_locked(child_proc->p_pid) != PROC_NULL) @@ -1180,30 +1213,37 @@ retry: * Increase reference counts on shared objects. * The p_stats and p_sigacts substructs are set in vm_fork. */ - child_proc->p_flag = (parent_proc->p_flag & (P_LP64 | P_TRANSLATED | P_AFFINITY)); + child_proc->p_flag = (parent_proc->p_flag & (P_LP64 | P_TRANSLATED | P_AFFINITY | P_DISABLE_ASLR | P_DELAYIDLESLEEP)); if (parent_proc->p_flag & P_PROFIL) startprofclock(child_proc); + + child_proc->p_vfs_iopolicy = (parent_proc->p_vfs_iopolicy & (P_VFS_IOPOLICY_FORCE_HFS_CASE_SENSITIVITY)); + /* * Note that if the current thread has an assumed identity, this * credential will be granted to the new process. */ child_proc->p_ucred = kauth_cred_get_with_ref(); + /* update cred on proc */ + PROC_UPDATE_CREDS_ONPROC(child_proc); + /* update audit session proc count */ + AUDIT_SESSION_PROCNEW(child_proc); -#ifdef CONFIG_EMBEDDED - lck_mtx_init(&child_proc->p_mlock, proc_lck_grp, proc_lck_attr); - lck_mtx_init(&child_proc->p_fdmlock, proc_lck_grp, proc_lck_attr); -#if CONFIG_DTRACE - lck_mtx_init(&child_proc->p_dtrace_sprlock, proc_lck_grp, proc_lck_attr); -#endif - lck_spin_init(&child_proc->p_slock, proc_lck_grp, proc_lck_attr); -#else /* !CONFIG_EMBEDDED */ +#if CONFIG_FINE_LOCK_GROUPS lck_mtx_init(&child_proc->p_mlock, proc_mlock_grp, proc_lck_attr); lck_mtx_init(&child_proc->p_fdmlock, proc_fdmlock_grp, proc_lck_attr); #if CONFIG_DTRACE lck_mtx_init(&child_proc->p_dtrace_sprlock, proc_lck_grp, proc_lck_attr); #endif lck_spin_init(&child_proc->p_slock, proc_slock_grp, proc_lck_attr); -#endif /* !CONFIG_EMBEDDED */ +#else /* !CONFIG_FINE_LOCK_GROUPS */ + lck_mtx_init(&child_proc->p_mlock, proc_lck_grp, proc_lck_attr); + lck_mtx_init(&child_proc->p_fdmlock, proc_lck_grp, proc_lck_attr); +#if CONFIG_DTRACE + lck_mtx_init(&child_proc->p_dtrace_sprlock, proc_lck_grp, proc_lck_attr); +#endif + lck_spin_init(&child_proc->p_slock, proc_lck_grp, proc_lck_attr); +#endif /* !CONFIG_FINE_LOCK_GROUPS */ klist_init(&child_proc->p_klist); if (child_proc->p_textvp != NULLVP) { @@ -1245,12 +1285,8 @@ retry: /* Intialize new process stats, including start time */ /* non-zeroed portion contains garbage AFAICT */ - bzero(&child_proc->p_stats->pstat_startzero, - (unsigned) ((caddr_t)&child_proc->p_stats->pstat_endzero - - (caddr_t)&child_proc->p_stats->pstat_startzero)); - bzero(&child_proc->p_stats->user_p_prof, sizeof(struct user_uprof)); - microtime(&child_proc->p_start); - child_proc->p_stats->p_start = child_proc->p_start; /* for compat */ + bzero(child_proc->p_stats, sizeof(*child_proc->p_stats)); + microtime_with_abstime(&child_proc->p_start, &child_proc->p_stats->ps_start); if (parent_proc->p_sigacts != NULL) (void)memcpy(child_proc->p_sigacts, @@ -1277,7 +1313,7 @@ retry: TAILQ_INIT(&child_proc->p_aio_doneq); /* Inherit the parent flags for code sign */ - child_proc->p_csflags = parent_proc->p_csflags; + child_proc->p_csflags = (parent_proc->p_csflags & ~CS_KILLED); /* * All processes have work queue locks; cleaned up by @@ -1304,6 +1340,7 @@ retry: child_proc->p_lflag |= P_LREGISTER; } child_proc->p_dispatchqueue_offset = parent_proc->p_dispatchqueue_offset; + child_proc->p_dispatchqueue_serialno_offset = parent_proc->p_dispatchqueue_serialno_offset; #if PSYNCH pth_proc_hashinit(child_proc); #endif /* PSYNCH */ @@ -1321,6 +1358,19 @@ retry: } #endif +#if CONFIG_MEMORYSTATUS + /* Memorystatus + jetsam init */ + child_proc->p_memstat_state = 0; + child_proc->p_memstat_effectivepriority = JETSAM_PRIORITY_DEFAULT; + child_proc->p_memstat_requestedpriority = JETSAM_PRIORITY_DEFAULT; + child_proc->p_memstat_userdata = 0; +#if CONFIG_FREEZE + child_proc->p_memstat_suspendedfootprint = 0; +#endif + child_proc->p_memstat_dirty = 0; + child_proc->p_memstat_idledeadline = 0; +#endif /* CONFIG_MEMORYSTATUS */ + bad: return(child_proc); } @@ -1394,6 +1444,7 @@ uthread_alloc(task_t task, thread_t thread, int noinherit) p = (proc_t) get_bsdtask_info(task); uth = (uthread_t)ut; + uth->uu_thread = thread; /* * Thread inherits credential from the creating thread, if both @@ -1446,6 +1497,9 @@ uthread_alloc(task_t task, thread_t thread, int noinherit) if (p->p_dtrace_ptss_pages != NULL) { uth->t_dtrace_scratch = dtrace_ptss_claim_entry(p); } +#endif +#if CONFIG_MACF + mac_thread_label_init(uth); #endif } @@ -1476,7 +1530,7 @@ uthread_cleanup(task_t task, void *uthread, void * bsd_info) * Calling this routine will clean up any throttle info reference * still inuse by the thread. */ - throttle_lowpri_io(FALSE); + throttle_lowpri_io(0); } /* * Per-thread audit state should never last beyond system @@ -1501,10 +1555,8 @@ uthread_cleanup(task_t task, void *uthread, void * bsd_info) if (uth->uu_allocsize && uth->uu_wqset){ kfree(uth->uu_wqset, uth->uu_allocsize); - sel->count = 0; uth->uu_allocsize = 0; uth->uu_wqset = 0; - sel->wql = 0; } if(uth->pth_name != NULL) @@ -1533,6 +1585,9 @@ uthread_cleanup(task_t task, void *uthread, void * bsd_info) if (tmpptr != NULL) { dtrace_ptss_release_entry(p, tmpptr); } +#endif +#if CONFIG_MACF + mac_thread_label_destroy(uth); #endif } } @@ -1555,6 +1610,13 @@ uthread_cred_free(void *uthread) void uthread_zone_free(void *uthread) { + uthread_t uth = (uthread_t)uthread; + + if (uth->t_tombstone) { + kfree(uth->t_tombstone, sizeof(struct doc_tombstone)); + uth->t_tombstone = NULL; + } + /* and free the uthread itself */ zfree(uthread_zone, uthread); }