+static int pid_from_task(task_t task)
+{
+ int pid = -1;
+
+ if (task->bsd_info)
+ pid = proc_pid(task->bsd_info);
+
+ return pid;
+}
+
+boolean_t
+kdp_copyin(pmap_t p, uint64_t uaddr, void *dest, size_t size) {
+ size_t rem = size;
+ char *kvaddr = dest;
+
+ while (rem) {
+ ppnum_t upn = pmap_find_phys(p, uaddr);
+ uint64_t phys_src = ptoa_64(upn) | (uaddr & PAGE_MASK);
+ uint64_t phys_dest = kvtophys((vm_offset_t)kvaddr);
+ uint64_t src_rem = PAGE_SIZE - (phys_src & PAGE_MASK);
+ uint64_t dst_rem = PAGE_SIZE - (phys_dest & PAGE_MASK);
+ size_t cur_size = (uint32_t) MIN(src_rem, dst_rem);
+ cur_size = MIN(cur_size, rem);
+
+ if (upn && pmap_valid_page(upn) && phys_dest) {
+ bcopy_phys(phys_src, phys_dest, cur_size);
+ }
+ else
+ break;
+ uaddr += cur_size;
+ kvaddr += cur_size;
+ rem -= cur_size;
+ }
+ return (rem == 0);
+}
+
+
+static void
+kdp_mem_snapshot(struct mem_snapshot *mem_snap)
+{
+ mem_snap->snapshot_magic = STACKSHOT_MEM_SNAPSHOT_MAGIC;
+ mem_snap->free_pages = vm_page_free_count;
+ mem_snap->active_pages = vm_page_active_count;
+ mem_snap->inactive_pages = vm_page_inactive_count;
+ mem_snap->purgeable_pages = vm_page_purgeable_count;
+ mem_snap->wired_pages = vm_page_wire_count;
+ mem_snap->speculative_pages = vm_page_speculative_count;
+ mem_snap->throttled_pages = vm_page_throttled_count;
+}
+
+
+/*
+ * Method for grabbing timer values safely, in the sense that no infinite loop will occur
+ * Certain flavors of the timer_grab function, which would seem to be the thing to use,
+ * can loop infinitely if called while the timer is in the process of being updated.
+ * Unfortunately, it is (rarely) possible to get inconsistent top and bottom halves of
+ * the timer using this method. This seems insoluble, since stackshot runs in a context
+ * where the timer might be half-updated, and has no way of yielding control just long
+ * enough to finish the update.
+ */
+
+static uint64_t safe_grab_timer_value(struct timer *t)
+{
+#if defined(__LP64__)
+ return t->all_bits;
+#else
+ uint64_t time = t->high_bits; /* endian independent grab */
+ time = (time << 32) | t->low_bits;
+ return time;
+#endif
+}
+
+int
+kdp_stackshot(int pid, void *tracebuf, uint32_t tracebuf_size, uint32_t trace_flags, uint32_t dispatch_offset, uint32_t *pbytesTraced)
+{
+ char *tracepos = (char *) tracebuf;
+ char *tracebound = tracepos + tracebuf_size;
+ uint32_t tracebytes = 0;
+ int error = 0;
+
+ task_t task = TASK_NULL;
+ thread_t thread = THREAD_NULL;
+ thread_snapshot_t tsnap = NULL;
+ unsigned framesize = 2 * sizeof(vm_offset_t);
+ struct task ctask;
+ struct thread cthread;
+ struct _vm_map cmap;
+ struct pmap cpmap;
+
+ queue_head_t *task_list = &tasks;
+ boolean_t is_active_list = TRUE;
+
+ boolean_t dispatch_p = ((trace_flags & STACKSHOT_GET_DQ) != 0);
+ boolean_t save_loadinfo_p = ((trace_flags & STACKSHOT_SAVE_LOADINFO) != 0);
+
+ if(trace_flags & STACKSHOT_GET_GLOBAL_MEM_STATS) {
+ if(tracepos + sizeof(struct mem_snapshot) > tracebound) {
+ error = -1;
+ goto error_exit;
+ }
+ kdp_mem_snapshot((struct mem_snapshot *)tracepos);
+ tracepos += sizeof(struct mem_snapshot);
+ }
+
+walk_list:
+ queue_iterate(task_list, task, task_t, tasks) {
+ if ((task == NULL) || (ml_nofault_copy((vm_offset_t) task, (vm_offset_t) &ctask, sizeof(struct task)) != sizeof(struct task)))
+ goto error_exit;
+
+ int task_pid = pid_from_task(task);
+ boolean_t task64 = task_has_64BitAddr(task);
+
+ if (!task->active) {
+ /*
+ * Not interested in terminated tasks without threads, and
+ * at the moment, stackshot can't handle a task without a name.
+ */
+ if (queue_empty(&task->threads) || task_pid == -1) {
+ continue;
+ }
+ }
+
+ /* Trace everything, unless a process was specified */
+ if ((pid == -1) || (pid == task_pid)) {
+ task_snapshot_t task_snap;
+ uint32_t uuid_info_count = 0;
+ mach_vm_address_t uuid_info_addr = 0;
+ boolean_t have_map = (task->map != NULL) &&
+ (ml_nofault_copy((vm_offset_t)(task->map), (vm_offset_t)&cmap, sizeof(struct _vm_map)) == sizeof(struct _vm_map));
+ boolean_t have_pmap = have_map && (cmap.pmap != NULL) &&
+ (ml_nofault_copy((vm_offset_t)(cmap.pmap), (vm_offset_t)&cpmap, sizeof(struct pmap)) == sizeof(struct pmap));
+
+ if (have_pmap && task->active && save_loadinfo_p && task_pid > 0) {
+ // Read the dyld_all_image_infos struct from the task memory to get UUID array count and location
+ if (task64) {
+ struct dyld_all_image_infos64 task_image_infos;
+ if (kdp_copyin(task->map->pmap, task->all_image_info_addr, &task_image_infos, sizeof(struct dyld_all_image_infos64))) {
+ uuid_info_count = (uint32_t)task_image_infos.uuidArrayCount;
+ uuid_info_addr = task_image_infos.uuidArray;
+ }
+ } else {
+ struct dyld_all_image_infos task_image_infos;
+ if (kdp_copyin(task->map->pmap, task->all_image_info_addr, &task_image_infos, sizeof(struct dyld_all_image_infos))) {
+ uuid_info_count = task_image_infos.uuidArrayCount;
+ uuid_info_addr = task_image_infos.uuidArray;
+ }
+ }
+
+ // If we get a NULL uuid_info_addr (which can happen when we catch dyld in the middle of updating
+ // this data structure), we zero the uuid_info_count so that we won't even try to save load info
+ // for this task.
+ if (!uuid_info_addr) {
+ uuid_info_count = 0;
+ }
+ }
+
+ if (tracepos + sizeof(struct task_snapshot) > tracebound) {
+ error = -1;
+ goto error_exit;
+ }
+
+ task_snap = (task_snapshot_t) tracepos;
+ task_snap->snapshot_magic = STACKSHOT_TASK_SNAPSHOT_MAGIC;
+ task_snap->pid = task_pid;
+ task_snap->nloadinfos = uuid_info_count;
+ /* Add the BSD process identifiers */
+ if (task_pid != -1)
+ proc_name_kdp(task, task_snap->p_comm, sizeof(task_snap->p_comm));
+ else
+ task_snap->p_comm[0] = '\0';
+ task_snap->ss_flags = 0;
+ if (task64)
+ task_snap->ss_flags |= kUser64_p;
+ if (!task->active)
+ task_snap->ss_flags |= kTerminatedSnapshot;
+
+ task_snap->suspend_count = task->suspend_count;
+ task_snap->task_size = have_pmap ? pmap_resident_count(task->map->pmap) : 0;
+ task_snap->faults = task->faults;
+ task_snap->pageins = task->pageins;
+ task_snap->cow_faults = task->cow_faults;
+
+ task_snap->user_time_in_terminated_threads = task->total_user_time;
+ task_snap->system_time_in_terminated_threads = task->total_system_time;
+ tracepos += sizeof(struct task_snapshot);
+
+ if (task_pid > 0 && uuid_info_count > 0) {
+ uint32_t uuid_info_size = (uint32_t)(task64 ? sizeof(struct dyld_uuid_info64) : sizeof(struct dyld_uuid_info));
+ uint32_t uuid_info_array_size = uuid_info_count * uuid_info_size;
+
+ if (tracepos + uuid_info_array_size > tracebound) {
+ error = -1;
+ goto error_exit;
+ }
+
+ // Copy in the UUID info array
+ // It may be nonresident, in which case just fix up nloadinfos to 0 in the task_snap
+ if (have_pmap && !kdp_copyin(task->map->pmap, uuid_info_addr, tracepos, uuid_info_array_size))
+ task_snap->nloadinfos = 0;
+ else
+ tracepos += uuid_info_array_size;
+ }
+
+ queue_iterate(&task->threads, thread, thread_t, task_threads){
+ if ((thread == NULL) || (ml_nofault_copy((vm_offset_t) thread, (vm_offset_t) &cthread, sizeof(struct thread)) != sizeof(struct thread)))
+ goto error_exit;
+
+ if (((tracepos + 4 * sizeof(struct thread_snapshot)) > tracebound)) {
+ error = -1;
+ goto error_exit;
+ }
+ /* Populate the thread snapshot header */
+ tsnap = (thread_snapshot_t) tracepos;
+ tsnap->thread_id = thread_tid(thread);
+ tsnap->state = thread->state;
+ tsnap->wait_event = thread->wait_event;
+ tsnap->continuation = (uint64_t) (uintptr_t) thread->continuation;
+ tsnap->user_time = safe_grab_timer_value(&thread->user_timer);
+ tsnap->system_time = safe_grab_timer_value(&thread->system_timer);
+ tsnap->snapshot_magic = STACKSHOT_THREAD_SNAPSHOT_MAGIC;
+ tracepos += sizeof(struct thread_snapshot);
+ tsnap->ss_flags = 0;
+
+ if (dispatch_p && (task != kernel_task) && (task->active) && have_pmap) {
+ uint64_t dqkeyaddr = thread_dispatchqaddr(thread);
+ if (dqkeyaddr != 0) {
+ uint64_t dqaddr = 0;
+ if (kdp_copyin(task->map->pmap, dqkeyaddr, &dqaddr, (task64 ? 8 : 4)) && (dqaddr != 0)) {
+ uint64_t dqserialnumaddr = dqaddr + dispatch_offset;
+ uint64_t dqserialnum = 0;
+ if (kdp_copyin(task->map->pmap, dqserialnumaddr, &dqserialnum, (task64 ? 8 : 4))) {
+ tsnap->ss_flags |= kHasDispatchSerial;
+ *(uint64_t *)tracepos = dqserialnum;
+ tracepos += 8;
+ }
+ }
+ }
+ }
+/* Call through to the machine specific trace routines
+ * Frames are added past the snapshot header.
+ */
+ tracebytes = 0;
+ if (thread->kernel_stack != 0) {
+#if defined(__LP64__)
+ tracebytes = machine_trace_thread64(thread, tracepos, tracebound, MAX_FRAMES, FALSE);
+ tsnap->ss_flags |= kKernel64_p;
+ framesize = 16;
+#else
+ tracebytes = machine_trace_thread(thread, tracepos, tracebound, MAX_FRAMES, FALSE);
+ framesize = 8;
+#endif
+ }
+ tsnap->nkern_frames = tracebytes/framesize;
+ tracepos += tracebytes;
+ tracebytes = 0;
+ /* Trace user stack, if any */
+ if (task->active && thread->task->map != kernel_map) {
+ /* 64-bit task? */
+ if (task_has_64BitAddr(thread->task)) {
+ tracebytes = machine_trace_thread64(thread, tracepos, tracebound, MAX_FRAMES, TRUE);
+ tsnap->ss_flags |= kUser64_p;
+ framesize = 16;
+ }
+ else {
+ tracebytes = machine_trace_thread(thread, tracepos, tracebound, MAX_FRAMES, TRUE);
+ framesize = 8;
+ }
+ }
+ tsnap->nuser_frames = tracebytes/framesize;
+ tracepos += tracebytes;
+ tracebytes = 0;
+ }
+ }
+ }
+
+ if (is_active_list) {
+ is_active_list = FALSE;
+ task_list = &terminated_tasks;
+ goto walk_list;
+ }
+
+error_exit:
+ /* Release stack snapshot wait indicator */
+ kdp_snapshot_postflight();
+
+ *pbytesTraced = (uint32_t)(tracepos - (char *) tracebuf);
+
+ return error;