#include <stdarg.h>
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/event.h> // for kqueue related stuff
#include <sys/fsevents.h>
#if CONFIG_FSE
#include <sys/namei.h>
#include <sys/filedesc.h>
#include <sys/kernel.h>
#include <sys/file_internal.h>
#include <sys/stat.h>
#include <sys/vnode_internal.h>
#include <sys/mount_internal.h>
#include <sys/proc_internal.h>
#include <sys/kauth.h>
#include <sys/uio.h>
#include <sys/malloc.h>
#include <sys/dirent.h>
#include <sys/attr.h>
#include <sys/sysctl.h>
#include <sys/ubc.h>
#include <machine/cons.h>
#include <miscfs/specfs/specdev.h>
#include <miscfs/devfs/devfs.h>
#include <sys/filio.h>
#include <kern/locks.h>
#include <libkern/OSAtomic.h>
#include <kern/zalloc.h>
#include <mach/mach_time.h>
#include <kern/thread_call.h>
#include <kern/clock.h>
#include <security/audit/audit.h>
#include <bsm/audit_kevents.h>
#include <pexpert/pexpert.h>
#include <libkern/section_keywords.h>
typedef struct kfs_event {
LIST_ENTRY(kfs_event) kevent_list;
int16_t type; u_int16_t flags, len; int32_t refcount; pid_t pid;
uint64_t abstime; ino64_t ino;
dev_t dev;
int32_t mode;
uid_t uid;
gid_t gid;
const char *str;
struct kfs_event *dest; } kfs_event;
#define KFSE_COMBINED_EVENTS 0x0001
#define KFSE_CONTAINS_DROPPED_EVENTS 0x0002
#define KFSE_RECYCLED_EVENT 0x0004
#define KFSE_BEING_CREATED 0x0008
LIST_HEAD(kfse_list, kfs_event) kfse_list_head = LIST_HEAD_INITIALIZER(x);
int num_events_outstanding = 0;
int num_pending_rename = 0;
struct fsevent_handle;
typedef struct fs_event_watcher {
int8_t *event_list; int32_t num_events;
dev_t *devices_not_to_watch; uint32_t num_devices;
int32_t flags;
kfs_event **event_queue;
int32_t eventq_size; int32_t num_readers;
int32_t rd; int32_t wr; int32_t blockers;
int32_t my_id;
uint32_t num_dropped;
uint64_t max_event_id;
struct fsevent_handle *fseh;
pid_t pid;
char proc_name[(2 * MAXCOMLEN) + 1];
} fs_event_watcher;
#define WATCHER_DROPPED_EVENTS 0x0001
#define WATCHER_CLOSING 0x0002
#define WATCHER_WANTS_COMPACT_EVENTS 0x0004
#define WATCHER_WANTS_EXTENDED_INFO 0x0008
#define WATCHER_APPLE_SYSTEM_SERVICE 0x0010 // fseventsd, coreservicesd, mds, revisiond
#define MAX_WATCHERS 8
static fs_event_watcher *watcher_table[MAX_WATCHERS];
#define DEFAULT_MAX_KFS_EVENTS 4096
static int max_kfs_events = DEFAULT_MAX_KFS_EVENTS;
static zone_t event_zone;
static int fs_event_init = 0;
static int16_t fs_event_type_watchers[FSE_MAX_EVENTS];
static dev_t fsevent_unmount_dev = 0;
static int fsevent_unmount_ack_count = 0;
static int watcher_add_event(fs_event_watcher *watcher, kfs_event *kfse);
static void fsevents_wakeup(fs_event_watcher *watcher);
static lck_grp_attr_t * fsevent_group_attr;
static lck_attr_t * fsevent_lock_attr;
static lck_grp_t * fsevent_mutex_group;
static lck_grp_t * fsevent_rw_group;
static lck_rw_t event_handling_lock; static lck_mtx_t watch_table_lock;
static lck_mtx_t event_buf_lock;
static lck_mtx_t event_writer_lock;
__private_extern__ void qsort(
void * array,
size_t nmembers,
size_t member_size,
int (*)(const void *, const void *));
static int
is_ignored_directory(const char *path)
{
if (!path) {
return 0;
}
#define IS_TLD(x) strnstr(__DECONST(char *, path), x, MAXPATHLEN)
if (IS_TLD("/.Spotlight-V100/") ||
IS_TLD("/.MobileBackups/") ||
IS_TLD("/Backups.backupdb/")) {
return 1;
}
#undef IS_TLD
return 0;
}
static void
fsevents_internal_init(void)
{
int i;
if (fs_event_init++ != 0) {
return;
}
for (i = 0; i < FSE_MAX_EVENTS; i++) {
fs_event_type_watchers[i] = 0;
}
memset(watcher_table, 0, sizeof(watcher_table));
fsevent_lock_attr = lck_attr_alloc_init();
fsevent_group_attr = lck_grp_attr_alloc_init();
fsevent_mutex_group = lck_grp_alloc_init("fsevent-mutex", fsevent_group_attr);
fsevent_rw_group = lck_grp_alloc_init("fsevent-rw", fsevent_group_attr);
lck_mtx_init(&watch_table_lock, fsevent_mutex_group, fsevent_lock_attr);
lck_mtx_init(&event_buf_lock, fsevent_mutex_group, fsevent_lock_attr);
lck_mtx_init(&event_writer_lock, fsevent_mutex_group, fsevent_lock_attr);
lck_rw_init(&event_handling_lock, fsevent_rw_group, fsevent_lock_attr);
PE_get_default("kern.maxkfsevents", &max_kfs_events, sizeof(max_kfs_events));
event_zone = zinit(sizeof(kfs_event),
max_kfs_events * sizeof(kfs_event),
max_kfs_events * sizeof(kfs_event),
"fs-event-buf");
if (event_zone == NULL) {
printf("fsevents: failed to initialize the event zone.\n");
}
zone_change(event_zone, Z_EXHAUST, TRUE);
zone_change(event_zone, Z_COLLECT, FALSE);
zone_change(event_zone, Z_CALLERACCT, FALSE);
if (zfill(event_zone, max_kfs_events) < max_kfs_events) {
printf("fsevents: failed to pre-fill the event zone.\n");
}
}
static void
lock_watch_table(void)
{
lck_mtx_lock(&watch_table_lock);
}
static void
unlock_watch_table(void)
{
lck_mtx_unlock(&watch_table_lock);
}
static void
lock_fs_event_list(void)
{
lck_mtx_lock(&event_buf_lock);
}
static void
unlock_fs_event_list(void)
{
lck_mtx_unlock(&event_buf_lock);
}
static void release_event_ref(kfs_event *kfse);
static boolean_t
watcher_cares_about_dev(fs_event_watcher *watcher, dev_t dev)
{
unsigned int i;
if (watcher->devices_not_to_watch == NULL) {
return true;
}
for (i = 0; i < watcher->num_devices; i++) {
if (dev == watcher->devices_not_to_watch[i]) {
return false;
}
}
return true;
}
int
need_fsevent(int type, vnode_t vp)
{
if (type >= 0 && type < FSE_MAX_EVENTS && fs_event_type_watchers[type] == 0) {
return 0;
}
if (vp->v_tag == VT_DEVFS) {
return 0;
}
return 1;
}
#define is_throw_away(x) ((x) == FSE_STAT_CHANGED || (x) == FSE_CONTENT_MODIFIED)
#define KFSE_COMBINED 0x0001
#define KFSE_COLLAPSED 0x0002
#define KFSE_RECYCLED 0x0004
int num_dropped = 0;
int num_parent_switch = 0;
int num_recycled_rename = 0;
static struct timeval last_print;
static int last_event_type = -1;
static void *last_ptr = NULL;
static char last_str[MAXPATHLEN];
static int last_nlen = 0;
static int last_vid = -1;
static uint64_t last_coalesced_time = 0;
static void *last_event_ptr = NULL;
int last_coalesced = 0;
static mach_timebase_info_data_t sTimebaseInfo = { 0, 0 };
int
add_fsevent(int type, vfs_context_t ctx, ...)
{
struct proc *p = vfs_context_proc(ctx);
int i, arg_type, ret;
kfs_event *kfse, *kfse_dest = NULL, *cur;
fs_event_watcher *watcher;
va_list ap;
int error = 0, did_alloc = 0;
dev_t dev = 0;
uint64_t now, elapsed;
char *pathbuff = NULL;
int pathbuff_len;
va_start(ap, ctx);
if (type < 0 || type >= FSE_MAX_EVENTS) {
return EINVAL;
}
if (fs_event_type_watchers[type] == 0) {
va_end(ap);
return 0;
}
now = mach_absolute_time();
lock_fs_event_list();
if (type != FSE_CREATE_FILE && type != FSE_DELETE && type != FSE_RENAME && type != FSE_EXCHANGE && type != FSE_CHOWN && type != FSE_DOCID_CHANGED && type != FSE_DOCID_CREATED && type != FSE_CLONE) {
void *ptr = NULL;
int vid = 0, was_str = 0, nlen = 0;
for (arg_type = va_arg(ap, int32_t); arg_type != FSE_ARG_DONE; arg_type = va_arg(ap, int32_t)) {
switch (arg_type) {
case FSE_ARG_VNODE: {
ptr = va_arg(ap, void *);
vid = vnode_vid((struct vnode *)ptr);
last_str[0] = '\0';
break;
}
case FSE_ARG_STRING: {
nlen = va_arg(ap, int32_t);
ptr = va_arg(ap, void *);
was_str = 1;
break;
}
}
if (ptr != NULL) {
break;
}
}
if (sTimebaseInfo.denom == 0) {
(void) clock_timebase_info(&sTimebaseInfo);
}
elapsed = (now - last_coalesced_time);
if (sTimebaseInfo.denom != sTimebaseInfo.numer) {
if (sTimebaseInfo.denom == 1) {
elapsed *= sTimebaseInfo.numer;
} else {
elapsed = (elapsed * sTimebaseInfo.numer) / (uint64_t)sTimebaseInfo.denom;
}
}
if (type == last_event_type
&& (elapsed < 1000000000)
&&
((vid && vid == last_vid && last_ptr == ptr)
||
(last_str[0] && last_nlen == nlen && ptr && strcmp(last_str, ptr) == 0))
) {
last_coalesced++;
unlock_fs_event_list();
va_end(ap);
return 0;
} else {
last_ptr = ptr;
if (was_str) {
strlcpy(last_str, ptr, sizeof(last_str));
}
last_nlen = nlen;
last_vid = vid;
last_event_type = type;
last_coalesced_time = now;
}
}
va_start(ap, ctx);
kfse = zalloc_noblock(event_zone);
if (kfse && (type == FSE_RENAME || type == FSE_EXCHANGE || type == FSE_CLONE)) {
kfse_dest = zalloc_noblock(event_zone);
if (kfse_dest == NULL) {
did_alloc = 1;
zfree(event_zone, kfse);
kfse = NULL;
}
}
if (kfse == NULL) { unlock_fs_event_list();
lock_watch_table();
for (i = 0; i < MAX_WATCHERS; i++) {
watcher = watcher_table[i];
if (watcher == NULL) {
continue;
}
watcher->flags |= WATCHER_DROPPED_EVENTS;
fsevents_wakeup(watcher);
}
unlock_watch_table();
{
struct timeval current_tv;
num_dropped++;
microuptime(¤t_tv);
if ((current_tv.tv_sec - last_print.tv_sec) > 10) {
int ii;
void *junkptr = zalloc_noblock(event_zone), *listhead = kfse_list_head.lh_first;
printf("add_fsevent: event queue is full! dropping events (num dropped events: %d; num events outstanding: %d).\n", num_dropped, num_events_outstanding);
printf("add_fsevent: kfse_list head %p ; num_pending_rename %d\n", listhead, num_pending_rename);
printf("add_fsevent: zalloc sez: %p\n", junkptr);
printf("add_fsevent: event_zone info: %d 0x%x\n", ((int *)event_zone)[0], ((int *)event_zone)[1]);
lock_watch_table();
for (ii = 0; ii < MAX_WATCHERS; ii++) {
if (watcher_table[ii] == NULL) {
continue;
}
printf("add_fsevent: watcher %s %p: rd %4d wr %4d q_size %4d flags 0x%x\n",
watcher_table[ii]->proc_name,
watcher_table[ii],
watcher_table[ii]->rd, watcher_table[ii]->wr,
watcher_table[ii]->eventq_size, watcher_table[ii]->flags);
}
unlock_watch_table();
last_print = current_tv;
if (junkptr) {
zfree(event_zone, junkptr);
}
}
}
if (pathbuff) {
release_pathbuff(pathbuff);
pathbuff = NULL;
}
return ENOSPC;
}
memset(kfse, 0, sizeof(kfs_event));
kfse->refcount = 1;
OSBitOrAtomic16(KFSE_BEING_CREATED, &kfse->flags);
last_event_ptr = kfse;
kfse->type = type;
kfse->abstime = now;
kfse->pid = p->p_pid;
if (type == FSE_RENAME || type == FSE_EXCHANGE || type == FSE_CLONE) {
memset(kfse_dest, 0, sizeof(kfs_event));
kfse_dest->refcount = 1;
OSBitOrAtomic16(KFSE_BEING_CREATED, &kfse_dest->flags);
kfse_dest->type = type;
kfse_dest->pid = p->p_pid;
kfse_dest->abstime = now;
kfse->dest = kfse_dest;
}
num_events_outstanding++;
if (kfse->type == FSE_RENAME) {
num_pending_rename++;
}
LIST_INSERT_HEAD(&kfse_list_head, kfse, kevent_list);
if (kfse->refcount < 1) {
panic("add_fsevent: line %d: kfse recount %d but should be at least 1\n", __LINE__, kfse->refcount);
}
unlock_fs_event_list();
cur = kfse;
if (type == FSE_DOCID_CREATED || type == FSE_DOCID_CHANGED) {
uint64_t val;
arg_type = va_arg(ap, int32_t);
if (arg_type == FSE_ARG_DEV) {
cur->dev = (dev_t)(va_arg(ap, dev_t));
} else {
cur->dev = (dev_t)0xbadc0de1;
}
arg_type = va_arg(ap, int32_t);
if (arg_type == FSE_ARG_INO) {
cur->ino = (ino64_t)(va_arg(ap, ino64_t));
} else {
cur->ino = 0xbadc0de2;
}
arg_type = va_arg(ap, int32_t);
if (arg_type == FSE_ARG_INO) {
val = (ino64_t)(va_arg(ap, ino64_t));
} else {
val = 0xbadc0de2;
}
__nochk_memcpy(&cur->str, &val, sizeof(ino64_t));
arg_type = va_arg(ap, int32_t);
if (arg_type == FSE_ARG_INT32) {
val = (uint64_t)va_arg(ap, uint32_t);
} else if (arg_type == FSE_ARG_INT64) {
val = (uint64_t)va_arg(ap, uint64_t);
} else {
val = 0xbadc0de3;
}
static_assert(sizeof(cur->uid) + sizeof(cur->gid) == sizeof(val), "gid/uid size mismatch");
static_assert(offsetof(struct kfs_event, gid) - offsetof(struct kfs_event, uid) == sizeof(cur->uid), "unexpected struct kfs_event layout");
memcpy(&cur->uid, &val, sizeof(cur->uid));
memcpy(&cur->gid, (u_int8_t *)&val + sizeof(cur->uid), sizeof(cur->gid));
goto done_with_args;
}
if (type == FSE_UNMOUNT_PENDING) {
arg_type = va_arg(ap, int32_t);
if (arg_type == FSE_ARG_DEV) {
cur->dev = (dev_t)(va_arg(ap, dev_t));
} else {
cur->dev = (dev_t)0xbadc0de1;
}
goto done_with_args;
}
for (arg_type = va_arg(ap, int32_t); arg_type != FSE_ARG_DONE; arg_type = va_arg(ap, int32_t)) {
switch (arg_type) {
case FSE_ARG_VNODE: {
struct vnode *vp;
struct vnode_attr va;
if (kfse->str != NULL) {
cur = kfse_dest;
}
vp = va_arg(ap, struct vnode *);
if (vp == NULL) {
panic("add_fsevent: you can't pass me a NULL vnode ptr (type %d)!\n",
cur->type);
}
VATTR_INIT(&va);
VATTR_WANTED(&va, va_fsid);
VATTR_WANTED(&va, va_fileid);
VATTR_WANTED(&va, va_mode);
VATTR_WANTED(&va, va_uid);
VATTR_WANTED(&va, va_gid);
VATTR_WANTED(&va, va_nlink);
if ((ret = vnode_getattr(vp, &va, vfs_context_kernel())) != 0) {
cur->str = NULL;
error = EINVAL;
goto clean_up;
}
cur->dev = dev = (dev_t)va.va_fsid;
cur->ino = (ino64_t)va.va_fileid;
cur->mode = (int32_t)vnode_vttoif(vnode_vtype(vp)) | va.va_mode;
cur->uid = va.va_uid;
cur->gid = va.va_gid;
if (vp->v_flag & VISHARDLINK) {
cur->mode |= FSE_MODE_HLINK;
if ((vp->v_type == VDIR && va.va_dirlinkcount == 0) || (vp->v_type == VREG && va.va_nlink == 0)) {
cur->mode |= FSE_MODE_LAST_HLINK;
}
}
if (pathbuff == NULL) {
pathbuff = get_pathbuff();
pathbuff_len = MAXPATHLEN;
pathbuff[0] = '\0';
if ((ret = vn_getpath_no_firmlink(vp, pathbuff, &pathbuff_len)) != 0 || pathbuff[0] == '\0') {
cur->flags |= KFSE_CONTAINS_DROPPED_EVENTS;
do {
if (vp->v_parent != NULL) {
vp = vp->v_parent;
} else if (vp->v_mount) {
strlcpy(pathbuff, vp->v_mount->mnt_vfsstat.f_mntonname, MAXPATHLEN);
break;
} else {
vp = NULL;
}
if (vp == NULL) {
break;
}
pathbuff_len = MAXPATHLEN;
ret = vn_getpath_no_firmlink(vp, pathbuff, &pathbuff_len);
} while (ret == ENOSPC);
if (ret != 0 || vp == NULL) {
error = ENOENT;
goto clean_up;
}
}
}
cur->len = pathbuff_len;
cur->str = vfs_addname(pathbuff, pathbuff_len, 0, 0);
if (cur->str == NULL || cur->str[0] == '\0') {
panic("add_fsevent: was not able to add path %s to event %p.\n", pathbuff, cur);
}
release_pathbuff(pathbuff);
pathbuff = NULL;
break;
}
case FSE_ARG_FINFO: {
fse_info *fse;
fse = va_arg(ap, fse_info *);
cur->dev = dev = (dev_t)fse->dev;
cur->ino = (ino64_t)fse->ino;
cur->mode = (int32_t)fse->mode;
cur->uid = (uid_t)fse->uid;
cur->gid = (uid_t)fse->gid;
if ((fse->mode & FSE_MODE_HLINK) && fse->nlink == 0) {
cur->mode |= FSE_MODE_LAST_HLINK;
}
if (cur->mode & FSE_TRUNCATED_PATH) {
cur->flags |= KFSE_CONTAINS_DROPPED_EVENTS;
cur->mode &= ~FSE_TRUNCATED_PATH;
}
break;
}
case FSE_ARG_STRING:
if (kfse->str != NULL) {
cur = kfse_dest;
}
cur->len = (int16_t)(va_arg(ap, int32_t) & 0x7fff);
if (cur->len >= 1) {
cur->str = vfs_addname(va_arg(ap, char *), cur->len, 0, 0);
} else {
printf("add_fsevent: funny looking string length: %d\n", (int)cur->len);
cur->len = 2;
cur->str = vfs_addname("/", cur->len, 0, 0);
}
if (cur->str[0] == 0) {
printf("add_fsevent: bogus looking string (len %d)\n", cur->len);
}
break;
case FSE_ARG_INT32: {
uint32_t ival = (uint32_t)va_arg(ap, int32_t);
kfse->uid = (ino64_t)ival;
break;
}
default:
printf("add_fsevent: unknown type %d\n", arg_type);
(void)va_arg(ap, int32_t);
}
}
done_with_args:
va_end(ap);
OSBitAndAtomic16(~KFSE_BEING_CREATED, &kfse->flags);
if (kfse_dest) {
OSBitAndAtomic16(~KFSE_BEING_CREATED, &kfse_dest->flags);
}
lock_watch_table();
for (i = 0; i < MAX_WATCHERS; i++) {
watcher = watcher_table[i];
if (watcher == NULL) {
continue;
}
if (type < watcher->num_events
&& watcher->event_list[type] == FSE_REPORT
&& watcher_cares_about_dev(watcher, dev)) {
if (watcher_add_event(watcher, kfse) != 0) {
watcher->num_dropped++;
continue;
}
}
}
unlock_watch_table();
clean_up:
if (pathbuff) {
release_pathbuff(pathbuff);
pathbuff = NULL;
}
release_event_ref(kfse);
return error;
}
static void
release_event_ref(kfs_event *kfse)
{
int old_refcount;
kfs_event copy, dest_copy;
old_refcount = OSAddAtomic(-1, &kfse->refcount);
if (old_refcount > 1) {
return;
}
lock_fs_event_list();
if (last_event_ptr == kfse) {
last_event_ptr = NULL;
last_event_type = -1;
last_coalesced_time = 0;
}
if (kfse->refcount < 0) {
panic("release_event_ref: bogus kfse refcount %d\n", kfse->refcount);
}
if (kfse->refcount > 0 || kfse->type == FSE_INVALID) {
unlock_fs_event_list();
return;
}
copy = *kfse;
if (kfse->type != FSE_DOCID_CREATED && kfse->type != FSE_DOCID_CHANGED && kfse->dest && OSAddAtomic(-1, &kfse->dest->refcount) == 1) {
dest_copy = *kfse->dest;
} else {
dest_copy.str = NULL;
dest_copy.len = 0;
dest_copy.type = FSE_INVALID;
}
kfse->pid = kfse->type; kfse->uid = (uid_t)(long)kfse->str; kfse->gid = (gid_t)(long)current_thread();
kfse->str = (char *)0xdeadbeef;
if (dest_copy.type != FSE_INVALID) {
kfse->dest->str = (char *)0xbadc0de; kfse->dest->type = FSE_INVALID;
if (kfse->dest->kevent_list.le_prev != NULL) {
num_events_outstanding--;
LIST_REMOVE(kfse->dest, kevent_list);
memset(&kfse->dest->kevent_list, 0xa5, sizeof(kfse->dest->kevent_list));
}
zfree(event_zone, kfse->dest);
}
{
int otype;
otype = kfse->type;
kfse->type = FSE_INVALID;
if (kfse->kevent_list.le_prev != NULL) {
num_events_outstanding--;
if (otype == FSE_RENAME) {
num_pending_rename--;
}
LIST_REMOVE(kfse, kevent_list);
memset(&kfse->kevent_list, 0, sizeof(kfse->kevent_list));
}
}
zfree(event_zone, kfse);
unlock_fs_event_list();
if (copy.str && copy.type != FSE_DOCID_CREATED && copy.type != FSE_DOCID_CHANGED) {
if (copy.len == 0) { panic("%s:%d: no more fref.vp!\n", __FILE__, __LINE__);
} else { vfs_removename(copy.str);
}
}
if (dest_copy.type != FSE_INVALID && dest_copy.str) {
if (dest_copy.len == 0) {
panic("%s:%d: no more fref.vp!\n", __FILE__, __LINE__);
} else {
vfs_removename(dest_copy.str);
}
}
}
static int
add_watcher(int8_t *event_list, int32_t num_events, int32_t eventq_size, fs_event_watcher **watcher_out, void *fseh)
{
int i;
fs_event_watcher *watcher;
if (eventq_size <= 0 || eventq_size > 100 * max_kfs_events) {
eventq_size = max_kfs_events;
}
MALLOC(watcher,
fs_event_watcher *,
sizeof(fs_event_watcher) + eventq_size * sizeof(kfs_event *),
M_TEMP, M_WAITOK);
if (watcher == NULL) {
return ENOMEM;
}
watcher->event_list = event_list;
watcher->num_events = num_events;
watcher->devices_not_to_watch = NULL;
watcher->num_devices = 0;
watcher->flags = 0;
watcher->event_queue = (kfs_event **)&watcher[1];
watcher->eventq_size = eventq_size;
watcher->rd = 0;
watcher->wr = 0;
watcher->blockers = 0;
watcher->num_readers = 0;
watcher->max_event_id = 0;
watcher->fseh = fseh;
watcher->pid = proc_selfpid();
proc_selfname(watcher->proc_name, sizeof(watcher->proc_name));
watcher->num_dropped = 0;
if (!strncmp(watcher->proc_name, "fseventsd", sizeof(watcher->proc_name)) ||
!strncmp(watcher->proc_name, "coreservicesd", sizeof(watcher->proc_name)) ||
!strncmp(watcher->proc_name, "revisiond", sizeof(watcher->proc_name)) ||
!strncmp(watcher->proc_name, "mds", sizeof(watcher->proc_name))) {
watcher->flags |= WATCHER_APPLE_SYSTEM_SERVICE;
} else {
printf("fsevents: watcher %s (pid: %d) - Using /dev/fsevents directly is unsupported. Migrate to FSEventsFramework\n",
watcher->proc_name, watcher->pid);
}
lock_watch_table();
for (i = 0; i < MAX_WATCHERS; i++) {
if (watcher_table[i] == NULL) {
watcher->my_id = i;
watcher_table[i] = watcher;
break;
}
}
if (i >= MAX_WATCHERS) {
printf("fsevents: too many watchers!\n");
unlock_watch_table();
FREE(watcher, M_TEMP);
return ENOSPC;
}
for (i = 0; i < num_events; i++) {
if (event_list[i] != FSE_IGNORE && i < FSE_MAX_EVENTS) {
fs_event_type_watchers[i]++;
}
}
unlock_watch_table();
*watcher_out = watcher;
return 0;
}
static void
remove_watcher(fs_event_watcher *target)
{
int i, j, counter = 0;
fs_event_watcher *watcher;
kfs_event *kfse;
lock_watch_table();
for (j = 0; j < MAX_WATCHERS; j++) {
watcher = watcher_table[j];
if (watcher != target) {
continue;
}
watcher_table[j] = NULL;
for (i = 0; i < watcher->num_events; i++) {
if (watcher->event_list[i] != FSE_IGNORE && i < FSE_MAX_EVENTS) {
fs_event_type_watchers[i]--;
}
}
if (watcher->flags & WATCHER_CLOSING) {
unlock_watch_table();
return;
}
watcher->flags |= WATCHER_CLOSING;
OSAddAtomic(1, &watcher->num_readers);
unlock_watch_table();
while (watcher->num_readers > 1 && counter++ < 5000) {
lock_watch_table();
fsevents_wakeup(watcher); unlock_watch_table();
tsleep(watcher, PRIBIO, "fsevents-close", 1);
}
if (counter++ >= 5000) {
panic("fsevents: close: still have readers! (%d)\n", watcher->num_readers);
}
lck_rw_lock_exclusive(&event_handling_lock);
while (watcher->rd != watcher->wr) {
kfse = watcher->event_queue[watcher->rd];
watcher->event_queue[watcher->rd] = NULL;
watcher->rd = (watcher->rd + 1) % watcher->eventq_size;
OSSynchronizeIO();
if (kfse != NULL && kfse->type != FSE_INVALID && kfse->refcount >= 1) {
release_event_ref(kfse);
}
}
lck_rw_unlock_exclusive(&event_handling_lock);
if (watcher->event_list) {
FREE(watcher->event_list, M_TEMP);
watcher->event_list = NULL;
}
if (watcher->devices_not_to_watch) {
FREE(watcher->devices_not_to_watch, M_TEMP);
watcher->devices_not_to_watch = NULL;
}
FREE(watcher, M_TEMP);
return;
}
unlock_watch_table();
}
#define EVENT_DELAY_IN_MS 10
static thread_call_t event_delivery_timer = NULL;
static int timer_set = 0;
static void
delayed_event_delivery(__unused void *param0, __unused void *param1)
{
int i;
lock_watch_table();
for (i = 0; i < MAX_WATCHERS; i++) {
if (watcher_table[i] != NULL && watcher_table[i]->rd != watcher_table[i]->wr) {
fsevents_wakeup(watcher_table[i]);
}
}
timer_set = 0;
unlock_watch_table();
}
static void
schedule_event_wakeup(void)
{
uint64_t deadline;
if (event_delivery_timer == NULL) {
event_delivery_timer = thread_call_allocate((thread_call_func_t)delayed_event_delivery, NULL);
}
clock_interval_to_deadline(EVENT_DELAY_IN_MS, 1000 * 1000, &deadline);
thread_call_enter_delayed(event_delivery_timer, deadline);
timer_set = 1;
}
#define MAX_NUM_PENDING 16
static int
watcher_add_event(fs_event_watcher *watcher, kfs_event *kfse)
{
if (kfse->abstime > watcher->max_event_id) {
watcher->max_event_id = kfse->abstime;
}
if (((watcher->wr + 1) % watcher->eventq_size) == watcher->rd) {
watcher->flags |= WATCHER_DROPPED_EVENTS;
fsevents_wakeup(watcher);
return ENOSPC;
}
OSAddAtomic(1, &kfse->refcount);
watcher->event_queue[watcher->wr] = kfse;
OSSynchronizeIO();
watcher->wr = (watcher->wr + 1) % watcher->eventq_size;
int32_t num_pending = 0;
if (watcher->rd < watcher->wr) {
num_pending = watcher->wr - watcher->rd;
}
if (watcher->rd > watcher->wr) {
num_pending = watcher->wr + watcher->eventq_size - watcher->rd;
}
if (num_pending > (watcher->eventq_size * 3 / 4) && !(watcher->flags & WATCHER_APPLE_SYSTEM_SERVICE)) {
lck_rw_lock_exclusive(&event_handling_lock);
while (watcher->rd != watcher->wr) {
kfse = watcher->event_queue[watcher->rd];
watcher->event_queue[watcher->rd] = NULL;
watcher->rd = (watcher->rd + 1) % watcher->eventq_size;
OSSynchronizeIO();
if (kfse != NULL && kfse->type != FSE_INVALID && kfse->refcount >= 1) {
release_event_ref(kfse);
}
}
watcher->flags |= WATCHER_DROPPED_EVENTS;
lck_rw_unlock_exclusive(&event_handling_lock);
printf("fsevents: watcher falling behind: %s (pid: %d) rd: %4d wr: %4d q_size: %4d flags: 0x%x\n",
watcher->proc_name, watcher->pid, watcher->rd, watcher->wr,
watcher->eventq_size, watcher->flags);
fsevents_wakeup(watcher);
} else if (num_pending > MAX_NUM_PENDING) {
fsevents_wakeup(watcher);
} else if (timer_set == 0) {
schedule_event_wakeup();
}
return 0;
}
static int
fill_buff(uint16_t type, int32_t size, const void *data,
char *buff, int32_t *_buff_idx, int32_t buff_sz,
struct uio *uio)
{
int32_t amt, error = 0, buff_idx = *_buff_idx;
uint16_t tmp;
if ((buff_sz - buff_idx) <= (int)(2 * sizeof(uint16_t) + 1)) {
if (buff_idx > uio_resid(uio)) {
error = ENOSPC;
goto get_out;
}
error = uiomove(buff, buff_idx, uio);
if (error) {
goto get_out;
}
buff_idx = 0;
}
memcpy(&buff[buff_idx], &type, sizeof(uint16_t));
buff_idx += sizeof(uint16_t);
tmp = size & 0xffff;
memcpy(&buff[buff_idx], &tmp, sizeof(uint16_t));
buff_idx += sizeof(uint16_t);
while (size > 0) {
amt = (size < (buff_sz - buff_idx)) ? size : (buff_sz - buff_idx);
memcpy(&buff[buff_idx], data, amt);
size -= amt;
buff_idx += amt;
data = (const char *)data + amt;
if (size > (buff_sz - buff_idx)) {
if (buff_idx > uio_resid(uio)) {
error = ENOSPC;
goto get_out;
}
error = uiomove(buff, buff_idx, uio);
if (error) {
goto get_out;
}
buff_idx = 0;
}
if (amt == 0) { break;
}
}
get_out:
*_buff_idx = buff_idx;
return error;
}
static int copy_out_kfse(fs_event_watcher *watcher, kfs_event *kfse, struct uio *uio) __attribute__((noinline));
static int
copy_out_kfse(fs_event_watcher *watcher, kfs_event *kfse, struct uio *uio)
{
int error;
uint16_t tmp16;
int32_t type;
kfs_event *cur;
char evbuff[512];
int evbuff_idx = 0;
if (kfse->type == FSE_INVALID) {
panic("fsevents: copy_out_kfse: asked to copy out an invalid event (kfse %p, refcount %d fref ptr %p)\n", kfse, kfse->refcount, kfse->str);
}
if (kfse->flags & KFSE_BEING_CREATED) {
return 0;
}
if (((kfse->type == FSE_RENAME) || (kfse->type == FSE_CLONE)) && kfse->dest == NULL) {
error = 0;
goto get_out;
}
if (watcher->flags & WATCHER_WANTS_EXTENDED_INFO) {
type = (kfse->type & 0xfff);
if (kfse->flags & KFSE_CONTAINS_DROPPED_EVENTS) {
type |= (FSE_CONTAINS_DROPPED_EVENTS << FSE_FLAG_SHIFT);
} else if (kfse->flags & KFSE_COMBINED_EVENTS) {
type |= (FSE_COMBINED_EVENTS << FSE_FLAG_SHIFT);
}
} else {
type = (int32_t)kfse->type;
}
memcpy(evbuff, &type, sizeof(int32_t));
evbuff_idx += sizeof(int32_t);
memcpy(&evbuff[evbuff_idx], &kfse->pid, sizeof(pid_t));
evbuff_idx += sizeof(pid_t);
cur = kfse;
copy_again:
if (kfse->type == FSE_DOCID_CHANGED || kfse->type == FSE_DOCID_CREATED) {
dev_t dev = cur->dev;
ino64_t ino = cur->ino;
uint64_t ival;
error = fill_buff(FSE_ARG_DEV, sizeof(dev_t), &dev, evbuff, &evbuff_idx, sizeof(evbuff), uio);
if (error != 0) {
goto get_out;
}
error = fill_buff(FSE_ARG_INO, sizeof(ino64_t), &ino, evbuff, &evbuff_idx, sizeof(evbuff), uio);
if (error != 0) {
goto get_out;
}
memcpy(&ino, &cur->str, sizeof(ino64_t));
error = fill_buff(FSE_ARG_INO, sizeof(ino64_t), &ino, evbuff, &evbuff_idx, sizeof(evbuff), uio);
if (error != 0) {
goto get_out;
}
memcpy(&ival, &cur->uid, sizeof(uint64_t)); error = fill_buff(FSE_ARG_INT64, sizeof(uint64_t), &ival, evbuff, &evbuff_idx, sizeof(evbuff), uio);
if (error != 0) {
goto get_out;
}
goto done;
}
if (kfse->type == FSE_UNMOUNT_PENDING) {
dev_t dev = cur->dev;
error = fill_buff(FSE_ARG_DEV, sizeof(dev_t), &dev, evbuff, &evbuff_idx, sizeof(evbuff), uio);
if (error != 0) {
goto get_out;
}
goto done;
}
if (cur->str == NULL || cur->str[0] == '\0') {
printf("copy_out_kfse:2: empty/short path (%s)\n", cur->str);
error = fill_buff(FSE_ARG_STRING, 2, "/", evbuff, &evbuff_idx, sizeof(evbuff), uio);
} else {
error = fill_buff(FSE_ARG_STRING, cur->len, cur->str, evbuff, &evbuff_idx, sizeof(evbuff), uio);
}
if (error != 0) {
goto get_out;
}
if (cur->dev == 0 && cur->ino == 0) {
goto done;
}
if (watcher->flags & WATCHER_WANTS_COMPACT_EVENTS) {
int32_t finfo_size;
finfo_size = sizeof(dev_t) + sizeof(ino64_t) + sizeof(int32_t) + sizeof(uid_t) + sizeof(gid_t);
error = fill_buff(FSE_ARG_FINFO, finfo_size, &cur->ino, evbuff, &evbuff_idx, sizeof(evbuff), uio);
if (error != 0) {
goto get_out;
}
} else {
error = fill_buff(FSE_ARG_DEV, sizeof(dev_t), &cur->dev, evbuff, &evbuff_idx, sizeof(evbuff), uio);
if (error != 0) {
goto get_out;
}
error = fill_buff(FSE_ARG_INO, sizeof(ino64_t), &cur->ino, evbuff, &evbuff_idx, sizeof(evbuff), uio);
if (error != 0) {
goto get_out;
}
error = fill_buff(FSE_ARG_MODE, sizeof(int32_t), &cur->mode, evbuff, &evbuff_idx, sizeof(evbuff), uio);
if (error != 0) {
goto get_out;
}
error = fill_buff(FSE_ARG_UID, sizeof(uid_t), &cur->uid, evbuff, &evbuff_idx, sizeof(evbuff), uio);
if (error != 0) {
goto get_out;
}
error = fill_buff(FSE_ARG_GID, sizeof(gid_t), &cur->gid, evbuff, &evbuff_idx, sizeof(evbuff), uio);
if (error != 0) {
goto get_out;
}
}
if (cur->dest) {
cur = cur->dest;
goto copy_again;
}
done:
error = fill_buff(FSE_ARG_INT64, sizeof(uint64_t), &cur->abstime, evbuff, &evbuff_idx, sizeof(evbuff), uio);
if (error != 0) {
goto get_out;
}
if (sizeof(uint16_t) > sizeof(evbuff) - evbuff_idx) {
if (evbuff_idx > uio_resid(uio)) {
error = ENOSPC;
goto get_out;
}
error = uiomove(evbuff, evbuff_idx, uio);
if (error) {
goto get_out;
}
evbuff_idx = 0;
}
tmp16 = FSE_ARG_DONE;
memcpy(&evbuff[evbuff_idx], &tmp16, sizeof(uint16_t));
evbuff_idx += sizeof(uint16_t);
if (evbuff_idx > uio_resid(uio)) {
error = ENOSPC;
} else {
error = uiomove(evbuff, evbuff_idx, uio);
}
get_out:
return error;
}
static int
fmod_watch(fs_event_watcher *watcher, struct uio *uio)
{
int error = 0;
user_ssize_t last_full_event_resid;
kfs_event *kfse;
uint16_t tmp16;
int skipped;
last_full_event_resid = uio_resid(uio);
if (uio_resid(uio) < 2048 || watcher == NULL) {
return EINVAL;
}
if (watcher->flags & WATCHER_CLOSING) {
return 0;
}
if (OSAddAtomic(1, &watcher->num_readers) != 0) {
OSAddAtomic(-1, &watcher->num_readers);
return EAGAIN;
}
restart_watch:
if (watcher->rd == watcher->wr) {
if (watcher->flags & WATCHER_CLOSING) {
OSAddAtomic(-1, &watcher->num_readers);
return 0;
}
OSAddAtomic(1, &watcher->blockers);
error = tsleep((caddr_t)watcher, PUSER | PCATCH, "fsevents_empty", 0);
OSAddAtomic(-1, &watcher->blockers);
if (error != 0 || (watcher->flags & WATCHER_CLOSING)) {
OSAddAtomic(-1, &watcher->num_readers);
return error;
}
}
if (watcher->flags & WATCHER_DROPPED_EVENTS) {
int32_t val = FSE_EVENTS_DROPPED;
error = uiomove((caddr_t)&val, sizeof(int32_t), uio);
if (error == 0) {
val = 0; error = uiomove((caddr_t)&val, sizeof(int32_t), uio);
tmp16 = FSE_ARG_DONE; error = uiomove((caddr_t)&tmp16, sizeof(int16_t), uio);
last_full_event_resid = uio_resid(uio);
}
if (error) {
OSAddAtomic(-1, &watcher->num_readers);
return error;
}
watcher->flags &= ~WATCHER_DROPPED_EVENTS;
}
skipped = 0;
lck_rw_lock_shared(&event_handling_lock);
while (uio_resid(uio) > 0 && watcher->rd != watcher->wr) {
if (watcher->flags & WATCHER_CLOSING) {
break;
}
kfse = watcher->event_queue[watcher->rd];
if (!kfse || kfse->type == FSE_INVALID || kfse->type >= watcher->num_events || kfse->refcount < 1) {
break;
}
if (watcher->event_list[kfse->type] == FSE_REPORT) {
boolean_t watcher_cares;
if (watcher->devices_not_to_watch == NULL) {
watcher_cares = true;
} else {
lock_watch_table();
watcher_cares = watcher_cares_about_dev(watcher, kfse->dev);
unlock_watch_table();
}
if (watcher_cares) {
if (!(watcher->flags & WATCHER_APPLE_SYSTEM_SERVICE) && kfse->type != FSE_DOCID_CREATED && kfse->type != FSE_DOCID_CHANGED && is_ignored_directory(kfse->str)) {
error = 0;
skipped = 1;
} else {
skipped = 0;
if (last_event_ptr == kfse) {
last_event_ptr = NULL;
last_event_type = -1;
last_coalesced_time = 0;
}
error = copy_out_kfse(watcher, kfse, uio);
if (error != 0) {
uio_setresid(uio, last_full_event_resid);
if (error != ENOENT) {
lck_rw_unlock_shared(&event_handling_lock);
error = 0;
goto get_out;
}
}
last_full_event_resid = uio_resid(uio);
}
}
}
watcher->event_queue[watcher->rd] = NULL;
watcher->rd = (watcher->rd + 1) % watcher->eventq_size;
OSSynchronizeIO();
release_event_ref(kfse);
}
lck_rw_unlock_shared(&event_handling_lock);
if (skipped && error == 0) {
goto restart_watch;
}
get_out:
OSAddAtomic(-1, &watcher->num_readers);
return error;
}
void
fsevent_unmount(__unused struct mount *mp, __unused vfs_context_t ctx)
{
#if CONFIG_EMBEDDED
dev_t dev = mp->mnt_vfsstat.f_fsid.val[0];
int error, waitcount = 0;
struct timespec ts = {.tv_sec = 1, .tv_nsec = 0};
lock_watch_table();
while (fsevent_unmount_dev != 0) {
error = msleep((caddr_t)&fsevent_unmount_dev, &watch_table_lock, PRIBIO, "fsevent_unmount_wait", &ts);
if (error == EWOULDBLOCK) {
error = 0;
}
if (!error && (++waitcount >= 10)) {
error = EWOULDBLOCK;
printf("timeout waiting to signal unmount pending for dev %d (fsevent_unmount_dev %d)\n", dev, fsevent_unmount_dev);
}
if (error) {
unlock_watch_table();
return;
}
}
if (fs_event_type_watchers[FSE_UNMOUNT_PENDING] == 0) {
unlock_watch_table();
return;
}
fsevent_unmount_dev = dev;
fsevent_unmount_ack_count = fs_event_type_watchers[FSE_UNMOUNT_PENDING];
unlock_watch_table();
error = add_fsevent(FSE_UNMOUNT_PENDING, ctx, FSE_ARG_DEV, dev, FSE_ARG_DONE);
lock_watch_table();
waitcount = 0;
while (fsevent_unmount_dev == dev) {
error = msleep((caddr_t)&fsevent_unmount_dev, &watch_table_lock, PRIBIO, "fsevent_unmount_pending", &ts);
if (error == EWOULDBLOCK) {
error = 0;
}
if (!error && (++waitcount >= 10)) {
error = EWOULDBLOCK;
printf("unmount pending ack timeout for dev %d\n", dev);
}
if (error) {
if (fsevent_unmount_dev == dev) {
fsevent_unmount_dev = 0;
fsevent_unmount_ack_count = 0;
}
wakeup((caddr_t)&fsevent_unmount_dev);
break;
}
}
unlock_watch_table();
#endif
}
static int fsevents_installed = 0;
typedef struct fsevent_handle {
UInt32 flags;
SInt32 active;
fs_event_watcher *watcher;
struct klist knotes;
struct selinfo si;
} fsevent_handle;
#define FSEH_CLOSING 0x0001
static int
fseventsf_read(struct fileproc *fp, struct uio *uio,
__unused int flags, __unused vfs_context_t ctx)
{
fsevent_handle *fseh = (struct fsevent_handle *)fp->f_fglob->fg_data;
int error;
error = fmod_watch(fseh->watcher, uio);
return error;
}
#pragma pack(push, 4)
typedef struct fsevent_dev_filter_args32 {
uint32_t num_devices;
user32_addr_t devices;
} fsevent_dev_filter_args32;
typedef struct fsevent_dev_filter_args64 {
uint32_t num_devices;
user64_addr_t devices;
} fsevent_dev_filter_args64;
#pragma pack(pop)
#define FSEVENTS_DEVICE_FILTER_32 _IOW('s', 100, fsevent_dev_filter_args32)
#define FSEVENTS_DEVICE_FILTER_64 _IOW('s', 100, fsevent_dev_filter_args64)
static int
fseventsf_ioctl(struct fileproc *fp, u_long cmd, caddr_t data, vfs_context_t ctx)
{
fsevent_handle *fseh = (struct fsevent_handle *)fp->f_fglob->fg_data;
int ret = 0;
fsevent_dev_filter_args64 *devfilt_args, _devfilt_args;
OSAddAtomic(1, &fseh->active);
if (fseh->flags & FSEH_CLOSING) {
OSAddAtomic(-1, &fseh->active);
return 0;
}
switch (cmd) {
case FIONBIO:
case FIOASYNC:
break;
case FSEVENTS_WANT_COMPACT_EVENTS: {
fseh->watcher->flags |= WATCHER_WANTS_COMPACT_EVENTS;
break;
}
case FSEVENTS_WANT_EXTENDED_INFO: {
fseh->watcher->flags |= WATCHER_WANTS_EXTENDED_INFO;
break;
}
case FSEVENTS_GET_CURRENT_ID: {
*(uint64_t *)data = fseh->watcher->max_event_id;
ret = 0;
break;
}
case FSEVENTS_DEVICE_FILTER_32: {
if (proc_is64bit(vfs_context_proc(ctx))) {
ret = EINVAL;
break;
}
fsevent_dev_filter_args32 *devfilt_args32 = (fsevent_dev_filter_args32 *)data;
devfilt_args = &_devfilt_args;
memset(devfilt_args, 0, sizeof(fsevent_dev_filter_args64));
devfilt_args->num_devices = devfilt_args32->num_devices;
devfilt_args->devices = CAST_USER_ADDR_T(devfilt_args32->devices);
goto handle_dev_filter;
}
case FSEVENTS_DEVICE_FILTER_64:
if (!proc_is64bit(vfs_context_proc(ctx))) {
ret = EINVAL;
break;
}
devfilt_args = (fsevent_dev_filter_args64 *)data;
handle_dev_filter:
{
int new_num_devices;
dev_t *devices_not_to_watch, *tmp = NULL;
if (devfilt_args->num_devices > 256) {
ret = EINVAL;
break;
}
new_num_devices = devfilt_args->num_devices;
if (new_num_devices == 0) {
lock_watch_table();
tmp = fseh->watcher->devices_not_to_watch;
fseh->watcher->devices_not_to_watch = NULL;
fseh->watcher->num_devices = new_num_devices;
unlock_watch_table();
if (tmp) {
FREE(tmp, M_TEMP);
}
break;
}
MALLOC(devices_not_to_watch, dev_t *,
new_num_devices * sizeof(dev_t),
M_TEMP, M_WAITOK);
if (devices_not_to_watch == NULL) {
ret = ENOMEM;
break;
}
ret = copyin(devfilt_args->devices,
(void *)devices_not_to_watch,
new_num_devices * sizeof(dev_t));
if (ret) {
FREE(devices_not_to_watch, M_TEMP);
break;
}
lock_watch_table();
fseh->watcher->num_devices = new_num_devices;
tmp = fseh->watcher->devices_not_to_watch;
fseh->watcher->devices_not_to_watch = devices_not_to_watch;
unlock_watch_table();
if (tmp) {
FREE(tmp, M_TEMP);
}
break;
}
case FSEVENTS_UNMOUNT_PENDING_ACK: {
lock_watch_table();
dev_t dev = *(dev_t *)data;
if (fsevent_unmount_dev == dev) {
if (--fsevent_unmount_ack_count <= 0) {
fsevent_unmount_dev = 0;
wakeup((caddr_t)&fsevent_unmount_dev);
}
} else {
printf("unexpected unmount pending ack %d (%d)\n", dev, fsevent_unmount_dev);
ret = EINVAL;
}
unlock_watch_table();
break;
}
default:
ret = EINVAL;
break;
}
OSAddAtomic(-1, &fseh->active);
return ret;
}
static int
fseventsf_select(struct fileproc *fp, int which, __unused void *wql, vfs_context_t ctx)
{
fsevent_handle *fseh = (struct fsevent_handle *)fp->f_fglob->fg_data;
int ready = 0;
if ((which != FREAD) || (fseh->watcher->flags & WATCHER_CLOSING)) {
return 0;
}
if (fseh->watcher->rd != fseh->watcher->wr) {
ready = 1;
}
if (!ready) {
selrecord(vfs_context_proc(ctx), &fseh->si, wql);
}
return ready;
}
#if NOTUSED
static int
fseventsf_stat(__unused struct fileproc *fp, __unused struct stat *sb, __unused vfs_context_t ctx)
{
return ENOTSUP;
}
#endif
static int
fseventsf_close(struct fileglob *fg, __unused vfs_context_t ctx)
{
fsevent_handle *fseh = (struct fsevent_handle *)fg->fg_data;
fs_event_watcher *watcher;
OSBitOrAtomic(FSEH_CLOSING, &fseh->flags);
while (OSAddAtomic(0, &fseh->active) > 0) {
tsleep((caddr_t)fseh->watcher, PRIBIO, "fsevents-close", 1);
}
watcher = fseh->watcher;
fg->fg_data = NULL;
fseh->watcher = NULL;
remove_watcher(watcher);
FREE(fseh, M_TEMP);
return 0;
}
static void
filt_fsevent_detach(struct knote *kn)
{
fsevent_handle *fseh = (struct fsevent_handle *)kn->kn_hook;
lock_watch_table();
KNOTE_DETACH(&fseh->knotes, kn);
unlock_watch_table();
}
static int
filt_fsevent_common(struct knote *kn, struct kevent_qos_s *kev, long hint)
{
fsevent_handle *fseh = (struct fsevent_handle *)kn->kn_hook;
int activate = 0;
int32_t rd, wr, amt;
int64_t data = 0;
if (NOTE_REVOKE == hint) {
kn->kn_flags |= (EV_EOF | EV_ONESHOT);
activate = 1;
}
rd = fseh->watcher->rd;
wr = fseh->watcher->wr;
if (rd <= wr) {
amt = wr - rd;
} else {
amt = fseh->watcher->eventq_size - (rd - wr);
}
switch (kn->kn_filter) {
case EVFILT_READ:
data = amt;
activate = (data != 0);
break;
case EVFILT_VNODE:
if (kn->kn_sfflags & hint) {
kn->kn_fflags |= hint;
}
if (kn->kn_fflags != 0) {
activate = 1;
}
break;
default:
break;
}
if (activate && kev) {
knote_fill_kevent(kn, kev, data);
}
return activate;
}
static int
filt_fsevent(struct knote *kn, long hint)
{
return filt_fsevent_common(kn, NULL, hint);
}
static int
filt_fsevent_touch(struct knote *kn, struct kevent_qos_s *kev)
{
int res;
lock_watch_table();
kn->kn_sfflags = kev->fflags;
kn->kn_sdata = kev->data;
res = filt_fsevent_common(kn, NULL, 0);
unlock_watch_table();
return res;
}
static int
filt_fsevent_process(struct knote *kn, struct kevent_qos_s *kev)
{
int res;
lock_watch_table();
res = filt_fsevent_common(kn, kev, 0);
unlock_watch_table();
return res;
}
SECURITY_READ_ONLY_EARLY(struct filterops) fsevent_filtops = {
.f_isfd = 1,
.f_attach = NULL,
.f_detach = filt_fsevent_detach,
.f_event = filt_fsevent,
.f_touch = filt_fsevent_touch,
.f_process = filt_fsevent_process,
};
static int
fseventsf_kqfilter(struct fileproc *fp, struct knote *kn,
__unused struct kevent_qos_s *kev)
{
fsevent_handle *fseh = (struct fsevent_handle *)fp->f_fglob->fg_data;
int res;
kn->kn_hook = (void*)fseh;
kn->kn_filtid = EVFILTID_FSEVENT;
lock_watch_table();
KNOTE_ATTACH(&fseh->knotes, kn);
res = filt_fsevent_common(kn, NULL, 0);
unlock_watch_table();
return res;
}
static int
fseventsf_drain(struct fileproc *fp, __unused vfs_context_t ctx)
{
int counter = 0;
fsevent_handle *fseh = (struct fsevent_handle *)fp->f_fglob->fg_data;
while (fseh->watcher->blockers && counter++ < 500) {
lock_watch_table();
fsevents_wakeup(fseh->watcher);
unlock_watch_table();
tsleep((caddr_t)fseh->watcher, PRIBIO, "watcher-close", 1);
}
return 0;
}
static int
fseventsopen(__unused dev_t dev, __unused int flag, __unused int mode, __unused struct proc *p)
{
if (!kauth_cred_issuser(kauth_cred_get())) {
return EPERM;
}
return 0;
}
static int
fseventsclose(__unused dev_t dev, __unused int flag, __unused int mode, __unused struct proc *p)
{
return 0;
}
static int
fseventsread(__unused dev_t dev, __unused struct uio *uio, __unused int ioflag)
{
return EIO;
}
static int
parse_buffer_and_add_events(const char *buffer, int bufsize, vfs_context_t ctx, long *remainder)
{
const fse_info *finfo, *dest_finfo;
const char *path, *ptr, *dest_path, *event_start = buffer;
int path_len, type, dest_path_len, err = 0;
ptr = buffer;
while ((ptr + sizeof(int) + sizeof(fse_info) + 1) < buffer + bufsize) {
type = *(const int *)ptr;
if (type < 0 || type >= FSE_MAX_EVENTS) {
err = EINVAL;
break;
}
ptr += sizeof(int);
finfo = (const fse_info *)ptr;
ptr += sizeof(fse_info);
path = ptr;
while (ptr < buffer + bufsize && *ptr != '\0') {
ptr++;
}
if (ptr >= buffer + bufsize) {
break;
}
ptr++;
path_len = ptr - path;
if (type != FSE_RENAME && type != FSE_EXCHANGE && type != FSE_CLONE) {
event_start = ptr;
err = add_fsevent(type, ctx, FSE_ARG_STRING, path_len, path, FSE_ARG_FINFO, finfo, FSE_ARG_DONE);
if (err) {
break;
}
continue;
}
dest_finfo = (const fse_info *)ptr;
ptr += sizeof(fse_info);
dest_path = ptr;
while (ptr < buffer + bufsize && *ptr != '\0') {
ptr++;
}
if (ptr >= buffer + bufsize) {
break;
}
ptr++; event_start = ptr;
dest_path_len = ptr - dest_path;
if (dest_finfo->ino) {
err = add_fsevent(type, ctx,
FSE_ARG_STRING, path_len, path, FSE_ARG_FINFO, finfo,
FSE_ARG_STRING, dest_path_len, dest_path, FSE_ARG_FINFO, dest_finfo,
FSE_ARG_DONE);
} else {
err = add_fsevent(type, ctx,
FSE_ARG_STRING, path_len, path, FSE_ARG_FINFO, finfo,
FSE_ARG_STRING, dest_path_len, dest_path,
FSE_ARG_DONE);
}
if (err) {
break;
}
}
*remainder = (long)((buffer + bufsize) - event_start);
return err;
}
#define WRITE_BUFFER_SIZE 4096
char *write_buffer = NULL;
static int
fseventswrite(__unused dev_t dev, struct uio *uio, __unused int ioflag)
{
int error = 0, count;
vfs_context_t ctx = vfs_context_current();
long offset = 0, remainder;
lck_mtx_lock(&event_writer_lock);
if (write_buffer == NULL) {
if (kmem_alloc(kernel_map, (vm_offset_t *)&write_buffer, WRITE_BUFFER_SIZE, VM_KERN_MEMORY_FILE)) {
lck_mtx_unlock(&event_writer_lock);
return ENOMEM;
}
}
while (uio_resid(uio)) {
if (uio_resid(uio) > (WRITE_BUFFER_SIZE - offset)) {
count = WRITE_BUFFER_SIZE - offset;
} else {
count = uio_resid(uio);
}
error = uiomove(write_buffer + offset, count, uio);
if (error) {
break;
}
error = parse_buffer_and_add_events(write_buffer, offset + count, ctx, &remainder);
if (error) {
break;
}
if (remainder != 0) {
memmove(write_buffer, (write_buffer + count + offset) - remainder, remainder);
offset = remainder;
} else {
offset = 0;
}
}
lck_mtx_unlock(&event_writer_lock);
return error;
}
static const struct fileops fsevents_fops = {
.fo_type = DTYPE_FSEVENTS,
.fo_read = fseventsf_read,
.fo_write = fo_no_write,
.fo_ioctl = fseventsf_ioctl,
.fo_select = fseventsf_select,
.fo_close = fseventsf_close,
.fo_kqfilter = fseventsf_kqfilter,
.fo_drain = fseventsf_drain,
};
typedef struct fsevent_clone_args32 {
user32_addr_t event_list;
int32_t num_events;
int32_t event_queue_depth;
user32_addr_t fd;
} fsevent_clone_args32;
typedef struct fsevent_clone_args64 {
user64_addr_t event_list;
int32_t num_events;
int32_t event_queue_depth;
user64_addr_t fd;
} fsevent_clone_args64;
#define FSEVENTS_CLONE_32 _IOW('s', 1, fsevent_clone_args32)
#define FSEVENTS_CLONE_64 _IOW('s', 1, fsevent_clone_args64)
static int
fseventsioctl(__unused dev_t dev, u_long cmd, caddr_t data, __unused int flag, struct proc *p)
{
struct fileproc *f;
int fd, error;
fsevent_handle *fseh = NULL;
fsevent_clone_args64 *fse_clone_args, _fse_clone;
int8_t *event_list;
int is64bit = proc_is64bit(p);
switch (cmd) {
case FSEVENTS_CLONE_32: {
if (is64bit) {
return EINVAL;
}
fsevent_clone_args32 *args32 = (fsevent_clone_args32 *)data;
fse_clone_args = &_fse_clone;
memset(fse_clone_args, 0, sizeof(fsevent_clone_args64));
fse_clone_args->event_list = CAST_USER_ADDR_T(args32->event_list);
fse_clone_args->num_events = args32->num_events;
fse_clone_args->event_queue_depth = args32->event_queue_depth;
fse_clone_args->fd = CAST_USER_ADDR_T(args32->fd);
goto handle_clone;
}
case FSEVENTS_CLONE_64:
if (!is64bit) {
return EINVAL;
}
fse_clone_args = (fsevent_clone_args64 *)data;
handle_clone:
if (fse_clone_args->num_events < 0 || fse_clone_args->num_events > 4096) {
return EINVAL;
}
MALLOC(fseh, fsevent_handle *, sizeof(fsevent_handle),
M_TEMP, M_WAITOK);
if (fseh == NULL) {
return ENOMEM;
}
memset(fseh, 0, sizeof(fsevent_handle));
klist_init(&fseh->knotes);
MALLOC(event_list, int8_t *,
fse_clone_args->num_events * sizeof(int8_t),
M_TEMP, M_WAITOK);
if (event_list == NULL) {
FREE(fseh, M_TEMP);
return ENOMEM;
}
error = copyin(fse_clone_args->event_list,
(void *)event_list,
fse_clone_args->num_events * sizeof(int8_t));
if (error) {
FREE(event_list, M_TEMP);
FREE(fseh, M_TEMP);
return error;
}
error = vslock((user_addr_t)fse_clone_args->fd,
sizeof(int32_t));
if (error) {
FREE(event_list, M_TEMP);
FREE(fseh, M_TEMP);
return error;
}
error = add_watcher(event_list,
fse_clone_args->num_events,
fse_clone_args->event_queue_depth,
&fseh->watcher,
fseh);
if (error) {
vsunlock((user_addr_t)fse_clone_args->fd,
sizeof(int32_t), 0);
FREE(event_list, M_TEMP);
FREE(fseh, M_TEMP);
return error;
}
fseh->watcher->fseh = fseh;
error = falloc(p, &f, &fd, vfs_context_current());
if (error) {
remove_watcher(fseh->watcher);
vsunlock((user_addr_t)fse_clone_args->fd,
sizeof(int32_t), 0);
FREE(event_list, M_TEMP);
FREE(fseh, M_TEMP);
return error;
}
proc_fdlock(p);
f->f_fglob->fg_flag = FREAD | FWRITE;
f->f_fglob->fg_ops = &fsevents_fops;
f->f_fglob->fg_data = (caddr_t) fseh;
error = copyout((void *)&fd, fse_clone_args->fd, sizeof(int32_t));
assert(error == 0);
procfdtbl_releasefd(p, fd, NULL);
fp_drop(p, fd, f, 1);
proc_fdunlock(p);
vsunlock((user_addr_t)fse_clone_args->fd,
sizeof(int32_t), 1);
break;
default:
error = EINVAL;
break;
}
return error;
}
static void
fsevents_wakeup(fs_event_watcher *watcher)
{
selwakeup(&watcher->fseh->si);
KNOTE(&watcher->fseh->knotes, NOTE_WRITE | NOTE_NONE);
wakeup((caddr_t)watcher);
}
static struct cdevsw fsevents_cdevsw =
{
fseventsopen,
fseventsclose,
fseventsread,
fseventswrite,
fseventsioctl,
(stop_fcn_t *)&nulldev,
(reset_fcn_t *)&nulldev,
NULL,
eno_select,
eno_mmap,
eno_strat,
eno_getc,
eno_putc,
0
};
void
fsevents_init(void)
{
int ret;
if (fsevents_installed) {
return;
}
fsevents_installed = 1;
ret = cdevsw_add(-1, &fsevents_cdevsw);
if (ret < 0) {
fsevents_installed = 0;
return;
}
devfs_make_node(makedev(ret, 0), DEVFS_CHAR,
UID_ROOT, GID_WHEEL, 0644, "fsevents", 0);
fsevents_internal_init();
}
char *
get_pathbuff(void)
{
char *path;
MALLOC_ZONE(path, char *, MAXPATHLEN, M_NAMEI, M_WAITOK);
return path;
}
void
release_pathbuff(char *path)
{
if (path == NULL) {
return;
}
FREE_ZONE(path, MAXPATHLEN, M_NAMEI);
}
int
get_fse_info(struct vnode *vp, fse_info *fse, __unused vfs_context_t ctx)
{
struct vnode_attr va;
VATTR_INIT(&va);
VATTR_WANTED(&va, va_fsid);
va.va_vaflags |= VA_REALFSID;
VATTR_WANTED(&va, va_fileid);
VATTR_WANTED(&va, va_mode);
VATTR_WANTED(&va, va_uid);
VATTR_WANTED(&va, va_gid);
if (vp->v_flag & VISHARDLINK) {
if (vp->v_type == VDIR) {
VATTR_WANTED(&va, va_dirlinkcount);
} else {
VATTR_WANTED(&va, va_nlink);
}
}
if (vnode_getattr(vp, &va, vfs_context_kernel()) != 0) {
memset(fse, 0, sizeof(fse_info));
return -1;
}
return vnode_get_fse_info_from_vap(vp, fse, &va);
}
int
vnode_get_fse_info_from_vap(vnode_t vp, fse_info *fse, struct vnode_attr *vap)
{
fse->ino = (ino64_t)vap->va_fileid;
fse->dev = (dev_t)vap->va_fsid;
fse->mode = (int32_t)vnode_vttoif(vnode_vtype(vp)) | vap->va_mode;
fse->uid = (uid_t)vap->va_uid;
fse->gid = (gid_t)vap->va_gid;
if (vp->v_flag & VISHARDLINK) {
fse->mode |= FSE_MODE_HLINK;
if (vp->v_type == VDIR) {
fse->nlink = (uint64_t)vap->va_dirlinkcount;
} else {
fse->nlink = (uint64_t)vap->va_nlink;
}
}
return 0;
}
void
create_fsevent_from_kevent(vnode_t vp, uint32_t kevents, struct vnode_attr *vap)
{
int fsevent_type = FSE_CONTENT_MODIFIED, len; char pathbuf[MAXPATHLEN];
fse_info fse;
if (kevents & VNODE_EVENT_DELETE) {
fsevent_type = FSE_DELETE;
} else if (kevents & (VNODE_EVENT_EXTEND | VNODE_EVENT_WRITE)) {
fsevent_type = FSE_CONTENT_MODIFIED;
} else if (kevents & VNODE_EVENT_LINK) {
fsevent_type = FSE_CREATE_FILE;
} else if (kevents & VNODE_EVENT_RENAME) {
fsevent_type = FSE_CREATE_FILE; } else if (kevents & (VNODE_EVENT_FILE_CREATED | VNODE_EVENT_FILE_REMOVED | VNODE_EVENT_DIR_CREATED | VNODE_EVENT_DIR_REMOVED)) {
fsevent_type = FSE_STAT_CHANGED; } else { fsevent_type = FSE_STAT_CHANGED;
}
fse.dev = vap->va_fsid;
fse.ino = vap->va_fileid;
fse.mode = vnode_vttoif(vnode_vtype(vp)) | (uint32_t)vap->va_mode;
if (vp->v_flag & VISHARDLINK) {
fse.mode |= FSE_MODE_HLINK;
if (vp->v_type == VDIR) {
fse.nlink = vap->va_dirlinkcount;
} else {
fse.nlink = vap->va_nlink;
}
}
if (vp->v_type == VDIR) {
fse.mode |= FSE_REMOTE_DIR_EVENT;
}
fse.uid = vap->va_uid;
fse.gid = vap->va_gid;
len = sizeof(pathbuf);
if (vn_getpath_no_firmlink(vp, pathbuf, &len) == 0) {
add_fsevent(fsevent_type, vfs_context_current(), FSE_ARG_STRING, len, pathbuf, FSE_ARG_FINFO, &fse, FSE_ARG_DONE);
}
return;
}
#else
#include <sys/fsevents.h>
char *
get_pathbuff(void)
{
char *path;
MALLOC_ZONE(path, char *, MAXPATHLEN, M_NAMEI, M_WAITOK);
return path;
}
void
release_pathbuff(char *path)
{
FREE_ZONE(path, MAXPATHLEN, M_NAMEI);
}
int
add_fsevent(__unused int type, __unused vfs_context_t ctx, ...)
{
return 0;
}
int
need_fsevent(__unused int type, __unused vnode_t vp)
{
return 0;
}
#endif