static errno_t
ipf_add(
const struct ipf_filter* filter,
ipfilter_t *filter_ref,
struct ipfilter_list *head)
{
struct ipfilter *new_filter;
if (filter->name == NULL || (filter->ipf_input == NULL && filter->ipf_output == NULL))
return EINVAL;
MALLOC(new_filter, struct ipfilter*, sizeof(*new_filter), M_IFADDR, M_WAITOK);
if (new_filter == NULL)
return ENOMEM;
lck_mtx_lock(kipf_lock);
new_filter->ipf_filter = *filter;
new_filter->ipf_head = head;
TAILQ_INSERT_HEAD(head, new_filter, ipf_link);
lck_mtx_unlock(kipf_lock);
*filter_ref = (ipfilter_t)new_filter;
/* This will force TCP to re-evaluate its use of TSO */
OSAddAtomic(1, &kipf_count);
routegenid_update();
return 0;
}
int
fuse_filehandle_put(vnode_t vp, vfs_context_t context, fufh_type_t fufh_type)
{
struct fuse_dispatcher fdi;
struct fuse_release_in *fri;
struct fuse_vnode_data *fvdat = VTOFUD(vp);
struct fuse_filehandle *fufh = NULL;
int err = 0;
int isdir = 0;
int op = FUSE_RELEASE;
const bool wait_for_completion = true;
fuse_trace_printf("fuse_filehandle_put(vp=%p, fufh_type=%d)\n",
vp, fufh_type);
fufh = &(fvdat->fufh[fufh_type]);
if (FUFH_IS_VALID(fufh)) {
panic("fuse4x: filehandle_put called on a valid fufh (type=%d)",
fufh_type);
/* NOTREACHED */
}
if (fuse_isdeadfs(vp)) {
goto out;
}
if (vnode_isdir(vp)) {
op = FUSE_RELEASEDIR;
isdir = 1;
}
fdisp_init(&fdi, sizeof(*fri));
fdisp_make_vp(&fdi, op, vp, context);
fri = fdi.indata;
fri->fh = fufh->fh_id;
fri->flags = fufh->open_flags;
if (wait_for_completion) {
if ((err = fdisp_wait_answ(&fdi))) {
goto out;
} else {
fuse_ticket_drop(fdi.tick);
}
} else {
fuse_insert_callback(fdi.tick, NULL);
fuse_insert_message(fdi.tick);
}
out:
OSAddAtomic(-1, (SInt32 *)&fuse_fh_current);
fuse_invalidate_attr(vp);
return err;
}
void
kern_os_free(void * addr)
{
size_t size;
size = kalloc_size(addr);
#if OSALLOCDEBUG
OSAddAtomic(-size, &debug_iomalloc_size);
#endif
kfree_addr(addr);
}
/*
* Allocate kva for pipe circular buffer, the space is pageable
* This routine will 'realloc' the size of a pipe safely, if it fails
* it will retain the old buffer.
* If it fails it will return ENOMEM.
*/
static int
pipespace(struct pipe *cpipe, int size)
{
vm_offset_t buffer;
if (size <= 0)
return(EINVAL);
if ((buffer = (vm_offset_t)kalloc(size)) == 0 )
return(ENOMEM);
/* free old resources if we're resizing */
pipe_free_kmem(cpipe);
cpipe->pipe_buffer.buffer = (caddr_t)buffer;
cpipe->pipe_buffer.size = size;
cpipe->pipe_buffer.in = 0;
cpipe->pipe_buffer.out = 0;
cpipe->pipe_buffer.cnt = 0;
OSAddAtomic(1, &amountpipes);
OSAddAtomic(cpipe->pipe_buffer.size, &amountpipekva);
return (0);
}
errno_t
ipf_remove(
ipfilter_t filter_ref)
{
struct ipfilter *match = (struct ipfilter*)filter_ref;
struct ipfilter_list *head;
if (match == 0 || (match->ipf_head != &ipv4_filters && match->ipf_head != &ipv6_filters))
return EINVAL;
head = match->ipf_head;
lck_mtx_lock(kipf_lock);
TAILQ_FOREACH(match, head, ipf_link) {
if (match == (struct ipfilter*)filter_ref) {
ipf_detach_func ipf_detach = match->ipf_filter.ipf_detach;
void* cookie = match->ipf_filter.cookie;
/*
* Cannot detach when they are filters running
*/
if (kipf_ref) {
kipf_delayed_remove++;
TAILQ_INSERT_TAIL(&tbr_filters, match, ipf_tbr);
match->ipf_filter.ipf_input = 0;
match->ipf_filter.ipf_output = 0;
lck_mtx_unlock(kipf_lock);
} else {
TAILQ_REMOVE(head, match, ipf_link);
lck_mtx_unlock(kipf_lock);
if (ipf_detach)
ipf_detach(cookie);
FREE(match, M_IFADDR);
/* This will force TCP to re-evaluate its use of TSO */
OSAddAtomic(-1, &kipf_count);
if (use_routegenid)
routegenid_update();
}
return 0;
}
}
lck_mtx_unlock(kipf_lock);
return ENOENT;
}
void *
kern_os_malloc(size_t size)
{
void *mem;
if (size == 0) {
return (0);
}
mem = kallocp_tag_bt((vm_size_t *)&size, VM_KERN_MEMORY_LIBKERN);
if (!mem) {
return (0);
}
#if OSALLOCDEBUG
OSAddAtomic(size, &debug_iomalloc_size);
#endif
bzero(mem, size);
return mem;
}
void *
kern_os_realloc(
void * addr,
size_t nsize)
{
void *nmem;
size_t osize;
if (!addr) {
return (kern_os_malloc(nsize));
}
osize = kalloc_size(addr);
if (nsize == osize) {
return (addr);
}
if (nsize == 0) {
kfree_addr(addr);
return (0);
}
nmem = kallocp_tag_bt((vm_size_t *)&nsize, VM_KERN_MEMORY_LIBKERN);
if (!nmem){
kfree_addr(addr);
return (0);
}
#if OSALLOCDEBUG
OSAddAtomic((nsize - osize), &debug_iomalloc_size);
#endif
if (nsize > osize) {
(void)memset((char *)nmem + osize, 0, nsize - osize);
}
(void)memcpy(nmem, addr, (nsize > osize) ? osize : nsize);
kfree_addr(addr);
return (nmem);
}
/*
* Because of the vagaries of how a filehandle can be used, we try not to
* be too smart in here (we try to be smart elsewhere). It is required that
* you come in here only if you really do not have the said filehandle--else
* we panic.
*/
int
fuse_filehandle_get(vnode_t vp,
vfs_context_t context,
fufh_type_t fufh_type,
int mode)
{
struct fuse_dispatcher fdi;
struct fuse_open_in *foi;
struct fuse_open_out *foo;
struct fuse_filehandle *fufh;
struct fuse_vnode_data *fvdat = VTOFUD(vp);
int err = 0;
int isdir = 0;
int oflags = 0;
int op = FUSE_OPEN;
fuse_trace_printf("fuse_filehandle_get(vp=%p, fufh_type=%d, mode=%x)\n",
vp, fufh_type, mode);
fufh = &(fvdat->fufh[fufh_type]);
if (FUFH_IS_VALID(fufh)) {
panic("fuse4x: filehandle_get called despite valid fufh (type=%d)",
fufh_type);
/* NOTREACHED */
}
/*
* Note that this means we are effectively FILTERING OUT open() flags.
*/
(void)mode;
oflags = fuse_filehandle_xlate_to_oflags(fufh_type);
if (vnode_isdir(vp)) {
isdir = 1;
op = FUSE_OPENDIR;
if (fufh_type != FUFH_RDONLY) {
log("fuse4x: non-rdonly fufh requested for directory\n");
fufh_type = FUFH_RDONLY;
}
}
fdisp_init(&fdi, sizeof(*foi));
fdisp_make_vp(&fdi, op, vp, context);
if (vnode_islnk(vp) && (mode & O_SYMLINK)) {
oflags |= O_SYMLINK;
}
foi = fdi.indata;
foi->flags = oflags;
OSAddAtomic(1, (SInt32 *)&fuse_fh_upcall_count);
if ((err = fdisp_wait_answ(&fdi))) {
#if M_FUSE4X_ENABLE_UNSUPPORTED
const char *vname = vnode_getname(vp);
#endif /* M_FUSE4X_ENABLE_UNSUPPORTED */
if (err == ENOENT) {
/*
* See comment in fuse_vnop_reclaim().
*/
cache_purge(vp);
}
#if M_FUSE4X_ENABLE_UNSUPPORTED
log("fuse4x: filehandle_get: failed for %s "
"(type=%d, err=%d, caller=%p)\n",
(vname) ? vname : "?", fufh_type, err,
__builtin_return_address(0));
if (vname) {
vnode_putname(vname);
}
#endif /* M_FUSE4X_ENABLE_UNSUPPORTED */
if (err == ENOENT) {
#if M_FUSE4X_ENABLE_BIGLOCK
struct fuse_data *data = fuse_get_mpdata(vnode_mount(vp));
fuse_biglock_unlock(data->biglock);
#endif
fuse_internal_vnode_disappear(vp, context, REVOKE_SOFT);
#if M_FUSE4X_ENABLE_BIGLOCK
fuse_biglock_lock(data->biglock);
#endif
}
return err;
}
OSAddAtomic(1, (SInt32 *)&fuse_fh_current);
foo = fdi.answ;
fufh->fh_id = foo->fh;
fufh->open_count = 1;
fufh->open_flags = oflags;
fufh->fuse_open_flags = foo->open_flags;
fufh->aux_count = 0;
fuse_ticket_drop(fdi.tick);
return 0;
}
SInt32 OSDecrementAtomic(volatile SInt32 * value)
{
return OSAddAtomic(-1, value);
}
/*
* free:
*
* Free resources
*/
void IOBufferMemoryDescriptor::free()
{
// Cache all of the relevant information on the stack for use
// after we call super::free()!
IOOptionBits flags = _flags;
IOOptionBits internalFlags = _internalFlags;
IOOptionBits options = _options;
vm_size_t size = _capacity;
void * buffer = _buffer;
IOMemoryMap * map = 0;
IOAddressRange * range = _ranges.v64;
vm_offset_t alignment = _alignment;
if (alignment >= page_size)
size = round_page(size);
if (reserved)
{
map = reserved->map;
IODelete( reserved, ExpansionData, 1 );
if (map)
map->release();
}
/* super::free may unwire - deallocate buffer afterwards */
super::free();
if (options & kIOMemoryPageable)
{
#if IOALLOCDEBUG
OSAddAtomicLong(-(round_page(size)), &debug_iomallocpageable_size);
#endif
}
else if (buffer)
{
if (kInternalFlagPageSized & internalFlags) size = round_page(size);
if (kInternalFlagPhysical & internalFlags)
{
IOKernelFreePhysical((mach_vm_address_t) buffer, size);
}
else if (kInternalFlagPageAllocated & internalFlags)
{
uintptr_t page;
page = iopa_free(&gIOBMDPageAllocator, (uintptr_t) buffer, size);
if (page)
{
kmem_free(kernel_map, page, page_size);
}
#if IOALLOCDEBUG
OSAddAtomic(-size, &debug_iomalloc_size);
#endif
IOStatisticsAlloc(kIOStatisticsFreeAligned, size);
}
else if (alignment > 1)
{
IOFreeAligned(buffer, size);
}
else
{
IOFree(buffer, size);
}
}
if (range && (kIOMemoryAsReference & flags))
IODelete(range, IOAddressRange, 1);
}
bool IOBufferMemoryDescriptor::initWithPhysicalMask(
task_t inTask,
IOOptionBits options,
mach_vm_size_t capacity,
mach_vm_address_t alignment,
mach_vm_address_t physicalMask)
{
task_t mapTask = NULL;
vm_map_t vmmap = NULL;
mach_vm_address_t highestMask = 0;
IOOptionBits iomdOptions = kIOMemoryTypeVirtual64 | kIOMemoryAsReference;
IODMAMapSpecification mapSpec;
bool mapped = false;
bool needZero;
if (!capacity) return false;
_options = options;
_capacity = capacity;
_internalFlags = 0;
_internalReserved = 0;
_buffer = 0;
_ranges.v64 = IONew(IOAddressRange, 1);
if (!_ranges.v64)
return (false);
_ranges.v64->address = 0;
_ranges.v64->length = 0;
// make sure super::free doesn't dealloc _ranges before super::init
_flags = kIOMemoryAsReference;
// Grab IOMD bits from the Buffer MD options
iomdOptions |= (options & kIOBufferDescriptorMemoryFlags);
if (!(kIOMemoryMapperNone & options))
{
IOMapper::checkForSystemMapper();
mapped = (0 != IOMapper::gSystem);
}
needZero = (mapped || (0 != (kIOMemorySharingTypeMask & options)));
if (physicalMask && (alignment <= 1))
{
alignment = ((physicalMask ^ (-1ULL)) & (physicalMask - 1));
highestMask = (physicalMask | alignment);
alignment++;
if (alignment < page_size)
alignment = page_size;
}
if ((options & (kIOMemorySharingTypeMask | kIOMapCacheMask | kIOMemoryClearEncrypt)) && (alignment < page_size))
alignment = page_size;
if (alignment >= page_size)
capacity = round_page(capacity);
if (alignment > page_size)
options |= kIOMemoryPhysicallyContiguous;
_alignment = alignment;
if ((capacity + alignment) < _capacity) return (false);
if ((inTask != kernel_task) && !(options & kIOMemoryPageable))
return false;
bzero(&mapSpec, sizeof(mapSpec));
mapSpec.alignment = _alignment;
mapSpec.numAddressBits = 64;
if (highestMask && mapped)
{
if (highestMask <= 0xFFFFFFFF)
mapSpec.numAddressBits = (32 - __builtin_clz((unsigned int) highestMask));
else
mapSpec.numAddressBits = (64 - __builtin_clz((unsigned int) (highestMask >> 32)));
highestMask = 0;
}
// set memory entry cache mode, pageable, purgeable
iomdOptions |= ((options & kIOMapCacheMask) >> kIOMapCacheShift) << kIOMemoryBufferCacheShift;
if (options & kIOMemoryPageable)
{
iomdOptions |= kIOMemoryBufferPageable;
if (options & kIOMemoryPurgeable) iomdOptions |= kIOMemoryBufferPurgeable;
}
else
{
vmmap = kernel_map;
// Buffer shouldn't auto prepare they should be prepared explicitly
// But it never was enforced so what are you going to do?
iomdOptions |= kIOMemoryAutoPrepare;
/* Allocate a wired-down buffer inside kernel space. */
bool contig = (0 != (options & kIOMemoryHostPhysicallyContiguous));
if (!contig && (0 != (options & kIOMemoryPhysicallyContiguous)))
{
contig |= (!mapped);
contig |= (0 != (kIOMemoryMapperNone & options));
#if 0
// treat kIOMemoryPhysicallyContiguous as kIOMemoryHostPhysicallyContiguous for now
contig |= true;
#endif
}
if (contig || highestMask || (alignment > page_size))
{
_internalFlags |= kInternalFlagPhysical;
if (highestMask)
{
_internalFlags |= kInternalFlagPageSized;
capacity = round_page(capacity);
}
_buffer = (void *) IOKernelAllocateWithPhysicalRestrict(
capacity, highestMask, alignment, contig);
}
else if (needZero
&& ((capacity + alignment) <= (page_size - gIOPageAllocChunkBytes)))
{
_internalFlags |= kInternalFlagPageAllocated;
needZero = false;
_buffer = (void *) iopa_alloc(&gIOBMDPageAllocator, &IOBMDPageProc, capacity, alignment);
if (_buffer)
{
IOStatisticsAlloc(kIOStatisticsMallocAligned, capacity);
#if IOALLOCDEBUG
OSAddAtomic(capacity, &debug_iomalloc_size);
#endif
}
}
else if (alignment > 1)
{
_buffer = IOMallocAligned(capacity, alignment);
}
else
{
_buffer = IOMalloc(capacity);
}
if (!_buffer)
{
return false;
}
if (needZero) bzero(_buffer, capacity);
}
if( (options & (kIOMemoryPageable | kIOMapCacheMask))) {
vm_size_t size = round_page(capacity);
// initWithOptions will create memory entry
iomdOptions |= kIOMemoryPersistent;
if( options & kIOMemoryPageable) {
#if IOALLOCDEBUG
OSAddAtomicLong(size, &debug_iomallocpageable_size);
#endif
mapTask = inTask;
if (NULL == inTask)
inTask = kernel_task;
}
else if (options & kIOMapCacheMask)
{
// Prefetch each page to put entries into the pmap
volatile UInt8 * startAddr = (UInt8 *)_buffer;
volatile UInt8 * endAddr = (UInt8 *)_buffer + capacity;
while (startAddr < endAddr)
{
UInt8 dummyVar = *startAddr;
(void) dummyVar;
startAddr += page_size;
}
}
}
_ranges.v64->address = (mach_vm_address_t) _buffer;;
_ranges.v64->length = _capacity;
if (!super::initWithOptions(_ranges.v64, 1, 0,
inTask, iomdOptions, /* System mapper */ 0))
return false;
// give any system mapper the allocation params
if (kIOReturnSuccess != dmaCommandOperation(kIOMDAddDMAMapSpec,
&mapSpec, sizeof(mapSpec)))
return false;
if (mapTask)
{
if (!reserved) {
reserved = IONew( ExpansionData, 1 );
if( !reserved)
return( false );
}
reserved->map = createMappingInTask(mapTask, 0,
kIOMapAnywhere | (options & kIOMapPrefault) | (options & kIOMapCacheMask), 0, 0);
if (!reserved->map)
{
_buffer = 0;
return( false );
}
release(); // map took a retain on this
reserved->map->retain();
removeMapping(reserved->map);
mach_vm_address_t buffer = reserved->map->getAddress();
_buffer = (void *) buffer;
if (kIOMemoryTypeVirtual64 == (kIOMemoryTypeMask & iomdOptions))
_ranges.v64->address = buffer;
}
setLength(_capacity);
return true;
}
bool IOSharedEventQueue::EnqueueTracker(DataArgs * data)
{
uint32_t singleTrackerLen = sizeof(DataArgs);
const UInt32 head = dataQueue->head;
const UInt32 tail = dataQueue->tail;
LOG(LOG_DEBUG, "head=%d", dataQueue->head);
LOG(LOG_DEBUG, "tail=%d", dataQueue->tail);
const UInt32 entrySize = singleTrackerLen+DATA_QUEUE_ENTRY_HEADER_SIZE;
IODataQueueEntry *entry;
if(singleTrackerLen>UINT32_MAX-DATA_QUEUE_ENTRY_HEADER_SIZE)
{
return false;
}
LOG(LOG_DEBUG, "this->getQueueSize()=%d", this->getQueueSize());
if(this->getQueueSize()<tail)
{
return false;
}
if(tail>=head)
{
if(entrySize<=UINT32_MAX-DATA_QUEUE_ENTRY_HEADER_SIZE &&
tail+entrySize<=this->getQueueSize())
{
entry = (IODataQueueEntry*)((uint8_t*)dataQueue->queue+dataQueue->tail);
entry->size=singleTrackerLen;
memcpy(entry->data, data, singleTrackerLen);
OSAddAtomic(entrySize, (SInt32*)&(dataQueue->tail));
}
else if(head>singleTrackerLen)
{
dataQueue->queue->size = singleTrackerLen;
if ( ( getQueueSize() - tail ) >= DATA_QUEUE_ENTRY_HEADER_SIZE )
{
((IODataQueueEntry *)((UInt8 *)dataQueue->queue + tail))->size = entrySize;
}
memcpy(&dataQueue->queue->data, data, singleTrackerLen);
OSCompareAndSwap(dataQueue->tail, entrySize, &dataQueue->tail);
}
else
{
return false;
}
}
else
{
if ( (head - tail) > entrySize )
{
entry = (IODataQueueEntry *)((UInt8 *)dataQueue->queue + tail);
entry->size = singleTrackerLen;
memcpy(&entry->data, data, singleTrackerLen);
OSAddAtomic(entrySize, (SInt32 *)&dataQueue->tail);
}
else
{
return false; // queue is full
}
}
if(head==tail) return true;
//send notification to port if any data is added to queue.
//if ( (this->_status&kSharedEventQueueNotifyWhenAddData) || ( head == tail ) || ( dataQueue->head == tail ))
{
sendDataAvailableNotification();
}
return true;
}
/*
* Look for the request in the cache
* If found then
* return action and optionally reply
* else
* insert it in the cache
*
* The rules are as follows:
* - if in progress, return DROP request
* - if completed within DELAY of the current time, return DROP it
* - if completed a longer time ago return REPLY if the reply was cached or
* return DOIT
* Update/add new request at end of lru list
*/
int
nfsrv_getcache(
struct nfsrv_descript *nd,
struct nfsrv_sock *slp,
mbuf_t *mrepp)
{
struct nfsrvcache *rp;
struct nfsm_chain nmrep;
struct sockaddr *saddr;
int ret, error;
/*
* Don't cache recent requests for reliable transport protocols.
* (Maybe we should for the case of a reconnect, but..)
*/
if (!nd->nd_nam2)
return (RC_DOIT);
lck_mtx_lock(nfsrv_reqcache_mutex);
loop:
for (rp = NFSRCHASH(nd->nd_retxid)->lh_first; rp != 0;
rp = rp->rc_hash.le_next) {
if (nd->nd_retxid == rp->rc_xid && nd->nd_procnum == rp->rc_proc &&
netaddr_match(rp->rc_family, &rp->rc_haddr, nd->nd_nam)) {
if ((rp->rc_flag & RC_LOCKED) != 0) {
rp->rc_flag |= RC_WANTED;
msleep(rp, nfsrv_reqcache_mutex, PZERO-1, "nfsrc", NULL);
goto loop;
}
rp->rc_flag |= RC_LOCKED;
/* If not at end of LRU chain, move it there */
if (rp->rc_lru.tqe_next) {
TAILQ_REMOVE(&nfsrv_reqcache_lruhead, rp, rc_lru);
TAILQ_INSERT_TAIL(&nfsrv_reqcache_lruhead, rp, rc_lru);
}
if (rp->rc_state == RC_UNUSED)
panic("nfsrv cache");
if (rp->rc_state == RC_INPROG) {
OSAddAtomic(1, &nfsstats.srvcache_inproghits);
ret = RC_DROPIT;
} else if (rp->rc_flag & RC_REPSTATUS) {
OSAddAtomic(1, &nfsstats.srvcache_nonidemdonehits);
nd->nd_repstat = rp->rc_status;
error = nfsrv_rephead(nd, slp, &nmrep, 0);
if (error) {
printf("nfsrv cache: reply alloc failed for nonidem request hit\n");
ret = RC_DROPIT;
*mrepp = NULL;
} else {
ret = RC_REPLY;
*mrepp = nmrep.nmc_mhead;
}
} else if (rp->rc_flag & RC_REPMBUF) {
OSAddAtomic(1, &nfsstats.srvcache_nonidemdonehits);
error = mbuf_copym(rp->rc_reply, 0, MBUF_COPYALL, MBUF_WAITOK, mrepp);
if (error) {
printf("nfsrv cache: reply copym failed for nonidem request hit\n");
ret = RC_DROPIT;
} else {
ret = RC_REPLY;
}
} else {
OSAddAtomic(1, &nfsstats.srvcache_idemdonehits);
rp->rc_state = RC_INPROG;
ret = RC_DOIT;
}
rp->rc_flag &= ~RC_LOCKED;
if (rp->rc_flag & RC_WANTED) {
rp->rc_flag &= ~RC_WANTED;
wakeup(rp);
}
lck_mtx_unlock(nfsrv_reqcache_mutex);
return (ret);
}
}
OSAddAtomic(1, &nfsstats.srvcache_misses);
if (nfsrv_reqcache_count < nfsrv_reqcache_size) {
/* try to allocate a new entry */
MALLOC(rp, struct nfsrvcache *, sizeof *rp, M_NFSD, M_WAITOK);
if (rp) {
bzero((char *)rp, sizeof *rp);
nfsrv_reqcache_count++;
rp->rc_flag = RC_LOCKED;
}
} else {
static errno_t
fuse_vfsop_mount(mount_t mp, __unused vnode_t devvp, user_addr_t udata,
vfs_context_t context)
{
int err = 0;
int mntopts = 0;
bool mounted = false;
uint32_t max_read = ~0;
size_t len;
fuse_device_t fdev = NULL;
struct fuse_data *data = NULL;
fuse_mount_args fusefs_args;
struct vfsstatfs *vfsstatfsp = vfs_statfs(mp);
#if M_FUSE4X_ENABLE_BIGLOCK
lck_mtx_t *biglock;
#endif
fuse_trace_printf_vfsop();
if (vfs_isupdate(mp)) {
return ENOTSUP;
}
err = copyin(udata, &fusefs_args, sizeof(fusefs_args));
if (err) {
return EINVAL;
}
/*
* Interesting flags that we can receive from mount or may want to
* otherwise forcibly set include:
*
* MNT_ASYNC
* MNT_AUTOMOUNTED
* MNT_DEFWRITE
* MNT_DONTBROWSE
* MNT_IGNORE_OWNERSHIP
* MNT_JOURNALED
* MNT_NODEV
* MNT_NOEXEC
* MNT_NOSUID
* MNT_NOUSERXATTR
* MNT_RDONLY
* MNT_SYNCHRONOUS
* MNT_UNION
*/
err = ENOTSUP;
#if M_FUSE4X_ENABLE_UNSUPPORTED
vfs_setlocklocal(mp);
#endif /* M_FUSE4X_ENABLE_UNSUPPORTED */
/** Option Processing. **/
if (*fusefs_args.fstypename) {
size_t typenamelen = strlen(fusefs_args.fstypename);
if (typenamelen > FUSE_FSTYPENAME_MAXLEN) {
return EINVAL;
}
snprintf(vfsstatfsp->f_fstypename, MFSTYPENAMELEN, "%s%s",
FUSE_FSTYPENAME_PREFIX, fusefs_args.fstypename);
}
if (!*fusefs_args.fsname)
return EINVAL;
if ((fusefs_args.daemon_timeout > FUSE_MAX_DAEMON_TIMEOUT) ||
(fusefs_args.daemon_timeout < FUSE_MIN_DAEMON_TIMEOUT)) {
return EINVAL;
}
if ((fusefs_args.init_timeout > FUSE_MAX_INIT_TIMEOUT) ||
(fusefs_args.init_timeout < FUSE_MIN_INIT_TIMEOUT)) {
return EINVAL;
}
if (fusefs_args.altflags & FUSE_MOPT_SPARSE) {
mntopts |= FSESS_SPARSE;
}
if (fusefs_args.altflags & FUSE_MOPT_AUTO_CACHE) {
mntopts |= FSESS_AUTO_CACHE;
}
if (fusefs_args.altflags & FUSE_MOPT_AUTO_XATTR) {
if (fusefs_args.altflags & FUSE_MOPT_NATIVE_XATTR) {
return EINVAL;
}
mntopts |= FSESS_AUTO_XATTR;
} else if (fusefs_args.altflags & FUSE_MOPT_NATIVE_XATTR) {
mntopts |= FSESS_NATIVE_XATTR;
}
if (fusefs_args.altflags & FUSE_MOPT_JAIL_SYMLINKS) {
mntopts |= FSESS_JAIL_SYMLINKS;
}
/*
* Note that unlike Linux, which keeps allow_root in user-space and
* passes allow_other in that case to the kernel, we let allow_root
* reach the kernel. The 'if' ordering is important here.
*/
if (fusefs_args.altflags & FUSE_MOPT_ALLOW_ROOT) {
int is_member = 0;
if ((kauth_cred_ismember_gid(kauth_cred_get(), fuse_admin_group, &is_member) != 0) || !is_member) {
log("fuse4x: caller is not a member of fuse4x admin group. "
"Either add user (id=%d) to group (id=%d), "
"or set correct '" SYSCTL_FUSE4X_TUNABLES_ADMIN "' sysctl value.\n",
kauth_cred_getuid(kauth_cred_get()), fuse_admin_group);
return EPERM;
}
mntopts |= FSESS_ALLOW_ROOT;
} else if (fusefs_args.altflags & FUSE_MOPT_ALLOW_OTHER) {
if (!fuse_allow_other && !fuse_vfs_context_issuser(context)) {
int is_member = 0;
if ((kauth_cred_ismember_gid(kauth_cred_get(), fuse_admin_group, &is_member) != 0) || !is_member) {
log("fuse4x: caller is not a member of fuse4x admin group. "
"Either add user (id=%d) to group (id=%d), "
"or set correct '" SYSCTL_FUSE4X_TUNABLES_ADMIN "' sysctl value.\n",
kauth_cred_getuid(kauth_cred_get()), fuse_admin_group);
return EPERM;
}
}
mntopts |= FSESS_ALLOW_OTHER;
}
if (fusefs_args.altflags & FUSE_MOPT_NO_APPLEDOUBLE) {
mntopts |= FSESS_NO_APPLEDOUBLE;
}
if (fusefs_args.altflags & FUSE_MOPT_NO_APPLEXATTR) {
mntopts |= FSESS_NO_APPLEXATTR;
}
if ((fusefs_args.altflags & FUSE_MOPT_FSID) && (fusefs_args.fsid != 0)) {
fsid_t fsid;
mount_t other_mp;
uint32_t target_dev;
target_dev = FUSE_MAKEDEV(FUSE_CUSTOM_FSID_DEVICE_MAJOR,
fusefs_args.fsid);
fsid.val[0] = target_dev;
fsid.val[1] = FUSE_CUSTOM_FSID_VAL1;
other_mp = vfs_getvfs(&fsid);
if (other_mp != NULL) {
return EPERM;
}
vfsstatfsp->f_fsid.val[0] = target_dev;
vfsstatfsp->f_fsid.val[1] = FUSE_CUSTOM_FSID_VAL1;
} else {
vfs_getnewfsid(mp);
}
if (fusefs_args.altflags & FUSE_MOPT_NO_ATTRCACHE) {
mntopts |= FSESS_NO_ATTRCACHE;
}
if (fusefs_args.altflags & FUSE_MOPT_NO_READAHEAD) {
mntopts |= FSESS_NO_READAHEAD;
}
if (fusefs_args.altflags & (FUSE_MOPT_NO_UBC | FUSE_MOPT_DIRECT_IO)) {
mntopts |= FSESS_NO_UBC;
}
if (fusefs_args.altflags & FUSE_MOPT_NO_VNCACHE) {
mntopts |= FSESS_NO_VNCACHE;
}
if (fusefs_args.altflags & FUSE_MOPT_NEGATIVE_VNCACHE) {
if (mntopts & FSESS_NO_VNCACHE) {
return EINVAL;
}
mntopts |= FSESS_NEGATIVE_VNCACHE;
}
if (fusefs_args.altflags & FUSE_MOPT_NO_SYNCWRITES) {
/* Cannot mix 'nosyncwrites' with 'noubc' or 'noreadahead'. */
if (mntopts & (FSESS_NO_READAHEAD | FSESS_NO_UBC)) {
log("fuse4x: cannot mix 'nosyncwrites' with 'noubc' or 'noreadahead'\n");
return EINVAL;
}
mntopts |= FSESS_NO_SYNCWRITES;
vfs_clearflags(mp, MNT_SYNCHRONOUS);
vfs_setflags(mp, MNT_ASYNC);
/* We check for this only if we have nosyncwrites in the first place. */
if (fusefs_args.altflags & FUSE_MOPT_NO_SYNCONCLOSE) {
mntopts |= FSESS_NO_SYNCONCLOSE;
}
} else {
vfs_clearflags(mp, MNT_ASYNC);
vfs_setflags(mp, MNT_SYNCHRONOUS);
}
if (mntopts & FSESS_NO_UBC) {
/* If no buffer cache, disallow exec from file system. */
vfs_setflags(mp, MNT_NOEXEC);
}
vfs_setauthopaque(mp);
vfs_setauthopaqueaccess(mp);
if ((fusefs_args.altflags & FUSE_MOPT_DEFAULT_PERMISSIONS) &&
(fusefs_args.altflags & FUSE_MOPT_DEFER_PERMISSIONS)) {
return EINVAL;
}
if (fusefs_args.altflags & FUSE_MOPT_DEFAULT_PERMISSIONS) {
mntopts |= FSESS_DEFAULT_PERMISSIONS;
vfs_clearauthopaque(mp);
}
if (fusefs_args.altflags & FUSE_MOPT_DEFER_PERMISSIONS) {
mntopts |= FSESS_DEFER_PERMISSIONS;
}
if (fusefs_args.altflags & FUSE_MOPT_EXTENDED_SECURITY) {
mntopts |= FSESS_EXTENDED_SECURITY;
vfs_setextendedsecurity(mp);
}
if (fusefs_args.altflags & FUSE_MOPT_LOCALVOL) {
vfs_setflags(mp, MNT_LOCAL);
}
/* done checking incoming option bits */
err = 0;
vfs_setfsprivate(mp, NULL);
fdev = fuse_device_get(fusefs_args.rdev);
if (!fdev) {
log("fuse4x: invalid device file (number=%d)\n", fusefs_args.rdev);
return EINVAL;
}
fuse_lck_mtx_lock(fdev->mtx);
data = fdev->data;
if (!data) {
fuse_lck_mtx_unlock(fdev->mtx);
return ENXIO;
}
#if M_FUSE4X_ENABLE_BIGLOCK
biglock = data->biglock;
fuse_biglock_lock(biglock);
#endif
if (data->dataflags & FSESS_MOUNTED) {
#if M_FUSE4X_ENABLE_BIGLOCK
fuse_biglock_unlock(biglock);
#endif
fuse_lck_mtx_unlock(fdev->mtx);
return EALREADY;
}
if (!(data->dataflags & FSESS_OPENED)) {
fuse_lck_mtx_unlock(fdev->mtx);
err = ENXIO;
goto out;
}
data->dataflags |= FSESS_MOUNTED;
OSAddAtomic(1, (SInt32 *)&fuse_mount_count);
mounted = true;
if (fdata_dead_get(data)) {
fuse_lck_mtx_unlock(fdev->mtx);
err = ENOTCONN;
goto out;
}
if (!data->daemoncred) {
panic("fuse4x: daemon found but identity unknown");
}
if (fuse_vfs_context_issuser(context) &&
kauth_cred_getuid(vfs_context_ucred(context)) != kauth_cred_getuid(data->daemoncred)) {
fuse_lck_mtx_unlock(fdev->mtx);
err = EPERM;
log("fuse4x: fuse daemon running by user_id=%d does not have privileges to mount on directory %s owned by user_id=%d\n",
kauth_cred_getuid(data->daemoncred), vfsstatfsp->f_mntonname, kauth_cred_getuid(vfs_context_ucred(context)));
goto out;
}
data->mp = mp;
data->fdev = fdev;
data->dataflags |= mntopts;
data->daemon_timeout.tv_sec = fusefs_args.daemon_timeout;
data->daemon_timeout.tv_nsec = 0;
if (data->daemon_timeout.tv_sec) {
data->daemon_timeout_p = &(data->daemon_timeout);
} else {
data->daemon_timeout_p = NULL;
}
data->init_timeout.tv_sec = fusefs_args.init_timeout;
data->init_timeout.tv_nsec = 0;
data->max_read = max_read;
data->fssubtype = fusefs_args.fssubtype;
data->mountaltflags = fusefs_args.altflags;
data->noimplflags = (uint64_t)0;
data->blocksize = fuse_round_size(fusefs_args.blocksize,
FUSE_MIN_BLOCKSIZE, FUSE_MAX_BLOCKSIZE);
data->iosize = fuse_round_size(fusefs_args.iosize,
FUSE_MIN_IOSIZE, FUSE_MAX_IOSIZE);
if (data->iosize < data->blocksize) {
data->iosize = data->blocksize;
}
data->userkernel_bufsize = FUSE_DEFAULT_USERKERNEL_BUFSIZE;
copystr(fusefs_args.fsname, vfsstatfsp->f_mntfromname,
MNAMELEN - 1, &len);
bzero(vfsstatfsp->f_mntfromname + len, MNAMELEN - len);
copystr(fusefs_args.volname, data->volname, MAXPATHLEN - 1, &len);
bzero(data->volname + len, MAXPATHLEN - len);
/* previous location of vfs_setioattr() */
vfs_setfsprivate(mp, data);
fuse_lck_mtx_unlock(fdev->mtx);
/* Send a handshake message to the daemon. */
fuse_send_init(data, context);
struct vfs_attr vfs_attr;
VFSATTR_INIT(&vfs_attr);
// Our vfs_getattr() doesn't look at most *_IS_ACTIVE()'s
err = fuse_vfsop_getattr(mp, &vfs_attr, context);
if (!err) {
vfsstatfsp->f_bsize = vfs_attr.f_bsize;
vfsstatfsp->f_iosize = vfs_attr.f_iosize;
vfsstatfsp->f_blocks = vfs_attr.f_blocks;
vfsstatfsp->f_bfree = vfs_attr.f_bfree;
vfsstatfsp->f_bavail = vfs_attr.f_bavail;
vfsstatfsp->f_bused = vfs_attr.f_bused;
vfsstatfsp->f_files = vfs_attr.f_files;
vfsstatfsp->f_ffree = vfs_attr.f_ffree;
// vfsstatfsp->f_fsid already handled above
vfsstatfsp->f_owner = kauth_cred_getuid(data->daemoncred);
vfsstatfsp->f_flags = vfs_flags(mp);
// vfsstatfsp->f_fstypename already handled above
// vfsstatfsp->f_mntonname handled elsewhere
// vfsstatfsp->f_mnfromname already handled above
vfsstatfsp->f_fssubtype = data->fssubtype;
}
if (fusefs_args.altflags & FUSE_MOPT_BLOCKSIZE) {
vfsstatfsp->f_bsize = data->blocksize;
} else {
//data->blocksize = vfsstatfsp->f_bsize;
}
if (fusefs_args.altflags & FUSE_MOPT_IOSIZE) {
vfsstatfsp->f_iosize = data->iosize;
} else {
//data->iosize = (uint32_t)vfsstatfsp->f_iosize;
vfsstatfsp->f_iosize = data->iosize;
}
out:
if (err) {
vfs_setfsprivate(mp, NULL);
fuse_lck_mtx_lock(fdev->mtx);
data = fdev->data; /* again */
if (mounted) {
OSAddAtomic(-1, (SInt32 *)&fuse_mount_count);
}
if (data) {
data->dataflags &= ~FSESS_MOUNTED;
if (!(data->dataflags & FSESS_OPENED)) {
#if M_FUSE4X_ENABLE_BIGLOCK
assert(biglock == data->biglock);
fuse_biglock_unlock(biglock);
#endif
fuse_device_close_final(fdev);
/* data is gone now */
}
}
fuse_lck_mtx_unlock(fdev->mtx);
} else {
vnode_t fuse_rootvp = NULLVP;
err = fuse_vfsop_root(mp, &fuse_rootvp, context);
if (err) {
goto out; /* go back and follow error path */
}
err = vnode_ref(fuse_rootvp);
(void)vnode_put(fuse_rootvp);
if (err) {
goto out; /* go back and follow error path */
} else {
struct vfsioattr ioattr;
vfs_ioattr(mp, &ioattr);
ioattr.io_devblocksize = data->blocksize;
vfs_setioattr(mp, &ioattr);
}
}
#if M_FUSE4X_ENABLE_BIGLOCK
fuse_lck_mtx_lock(fdev->mtx);
data = fdev->data; /* ...and again */
if(data) {
assert(data->biglock == biglock);
fuse_biglock_unlock(biglock);
}
fuse_lck_mtx_unlock(fdev->mtx);
#endif
return err;
}
Boolean IODataQueue::enqueue(void * data, UInt32 dataSize)
{
const UInt32 head = dataQueue->head; // volatile
const UInt32 tail = dataQueue->tail;
const UInt32 entrySize = dataSize + DATA_QUEUE_ENTRY_HEADER_SIZE;
IODataQueueEntry * entry;
// Check for overflow of entrySize
if (dataSize > UINT32_MAX - DATA_QUEUE_ENTRY_HEADER_SIZE) {
return false;
}
// Check for underflow of (dataQueue->queueSize - tail)
if (dataQueue->queueSize < tail) {
return false;
}
if ( tail >= head )
{
// Is there enough room at the end for the entry?
if ((entrySize <= UINT32_MAX - tail) &&
((tail + entrySize) <= dataQueue->queueSize) )
{
entry = (IODataQueueEntry *)((UInt8 *)dataQueue->queue + tail);
entry->size = dataSize;
memcpy(&entry->data, data, dataSize);
// The tail can be out of bound when the size of the new entry
// exactly matches the available space at the end of the queue.
// The tail can range from 0 to dataQueue->queueSize inclusive.
OSAddAtomic(entrySize, (SInt32 *)&dataQueue->tail);
}
else if ( head > entrySize ) // Is there enough room at the beginning?
{
// Wrap around to the beginning, but do not allow the tail to catch
// up to the head.
dataQueue->queue->size = dataSize;
// We need to make sure that there is enough room to set the size before
// doing this. The user client checks for this and will look for the size
// at the beginning if there isn't room for it at the end.
if ( ( dataQueue->queueSize - tail ) >= DATA_QUEUE_ENTRY_HEADER_SIZE )
{
((IODataQueueEntry *)((UInt8 *)dataQueue->queue + tail))->size = dataSize;
}
memcpy(&dataQueue->queue->data, data, dataSize);
OSCompareAndSwap(dataQueue->tail, entrySize, &dataQueue->tail);
}
else
{
return false; // queue is full
}
}
else
{
// Do not allow the tail to catch up to the head when the queue is full.
// That's why the comparison uses a '>' rather than '>='.
if ( (head - tail) > entrySize )
{
entry = (IODataQueueEntry *)((UInt8 *)dataQueue->queue + tail);
entry->size = dataSize;
memcpy(&entry->data, data, dataSize);
OSAddAtomic(entrySize, (SInt32 *)&dataQueue->tail);
}
else
{
return false; // queue is full
}
}
// Send notification (via mach message) that data is available.
if ( ( head == tail ) /* queue was empty prior to enqueue() */
|| ( dataQueue->head == tail ) ) /* queue was emptied during enqueue() */
{
sendDataAvailableNotification();
}
return true;
}
void
kfree(
void *data,
vm_size_t size)
{
zone_t z;
if (size < MAX_SIZE_ZDLUT)
z = get_zone_dlut(size);
else if (size < kalloc_max_prerounded)
z = get_zone_search(size, k_zindex_start);
else {
/* if size was too large for a zone, then use kmem_free */
vm_map_t alloc_map = kernel_map;
if ((((vm_offset_t) data) >= kalloc_map_min) && (((vm_offset_t) data) <= kalloc_map_max))
alloc_map = kalloc_map;
if (size > kalloc_largest_allocated) {
/*
* work around double FREEs of small MALLOCs
* this used to end up being a nop
* since the pointer being freed from an
* alloc backed by the zalloc world could
* never show up in the kalloc_map... however,
* the kernel_map is a different issue... since it
* was released back into the zalloc pool, a pointer
* would have gotten written over the 'size' that
* the MALLOC was retaining in the first 4 bytes of
* the underlying allocation... that pointer ends up
* looking like a really big size on the 2nd FREE and
* pushes the kfree into the kernel_map... we
* end up removing a ton of virtual space before we panic
* this check causes us to ignore the kfree for a size
* that must be 'bogus'... note that it might not be due
* to the above scenario, but it would still be wrong and
* cause serious damage.
*/
OSAddAtomic(1, &kfree_nop_count);
return;
}
kmem_free(alloc_map, (vm_offset_t)data, size);
kalloc_spin_lock();
kalloc_large_total -= size;
kalloc_large_inuse--;
kalloc_unlock();
KALLOC_ZINFO_SFREE(size);
return;
}
/* free to the appropriate zone */
#ifdef KALLOC_DEBUG
if (size > z->elem_size)
panic("%s: z %p (%s) but requested size %lu", __func__,
z, z->zone_name, (unsigned long)size);
#endif
assert(size <= z->elem_size);
zfree(z, data);
}
static struct kern_coredump_core *
kern_register_coredump_helper_internal(int kern_coredump_config_vers, kern_coredump_callback_config *kc_callbacks,
void *refcon, const char *core_description, boolean_t xnu_callback, boolean_t is64bit,
uint32_t mh_magic, cpu_type_t cpu_type, cpu_subtype_t cpu_subtype)
{
struct kern_coredump_core *core_helper = NULL;
kern_coredump_callback_config *core_callbacks = NULL;
if (kern_coredump_config_vers < KERN_COREDUMP_MIN_CONFIG_VERSION)
return NULL;
if (kc_callbacks == NULL)
return NULL;;
if (core_description == NULL)
return NULL;
if (kc_callbacks->kcc_coredump_get_summary == NULL ||
kc_callbacks->kcc_coredump_save_segment_descriptions == NULL ||
kc_callbacks->kcc_coredump_save_segment_data == NULL ||
kc_callbacks->kcc_coredump_save_thread_state == NULL ||
kc_callbacks->kcc_coredump_save_sw_vers == NULL)
return NULL;
#if !defined(__LP64__)
/* We don't support generating 64-bit cores on 32-bit platforms */
if (is64bit)
return NULL;
#endif
core_helper = kalloc(sizeof(*core_helper));
core_helper->kcc_next = NULL;
core_helper->kcc_refcon = refcon;
if (xnu_callback) {
snprintf((char *)&core_helper->kcc_corename, MACH_CORE_FILEHEADER_NAMELEN, "%s", core_description);
} else {
/* Make sure there's room for the -coproc suffix (16 - NULL char - strlen(-coproc)) */
snprintf((char *)&core_helper->kcc_corename, MACH_CORE_FILEHEADER_NAMELEN, "%.8s-coproc", core_description);
}
core_helper->kcc_is64bit = is64bit;
core_helper->kcc_mh_magic = mh_magic;
core_helper->kcc_cpu_type = cpu_type;
core_helper->kcc_cpu_subtype = cpu_subtype;
core_callbacks = &core_helper->kcc_cb;
core_callbacks->kcc_coredump_init = kc_callbacks->kcc_coredump_init;
core_callbacks->kcc_coredump_get_summary = kc_callbacks->kcc_coredump_get_summary;
core_callbacks->kcc_coredump_save_segment_descriptions = kc_callbacks->kcc_coredump_save_segment_descriptions;
core_callbacks->kcc_coredump_save_segment_data = kc_callbacks->kcc_coredump_save_segment_data;
core_callbacks->kcc_coredump_save_thread_state = kc_callbacks->kcc_coredump_save_thread_state;
core_callbacks->kcc_coredump_save_misc_data = kc_callbacks->kcc_coredump_save_misc_data;
core_callbacks->kcc_coredump_save_sw_vers = kc_callbacks->kcc_coredump_save_sw_vers;
if (xnu_callback) {
assert(kernel_helper == NULL);
kernel_helper = core_helper;
} else {
do {
core_helper->kcc_next = kern_coredump_core_list;
} while (!OSCompareAndSwapPtr(kern_coredump_core_list, core_helper, &kern_coredump_core_list));
}
OSAddAtomic(1, &coredump_registered_count);
kprintf("Registered coredump handler for %s\n", core_description);
return core_helper;
}
static errno_t
fuse_vfsop_unmount(mount_t mp, int mntflags, vfs_context_t context)
{
int err = 0;
int flags = 0;
fuse_device_t fdev;
struct fuse_data *data;
struct fuse_dispatcher fdi;
vnode_t fuse_rootvp = NULLVP;
fuse_trace_printf_vfsop();
if (mntflags & MNT_FORCE) {
flags |= FORCECLOSE;
}
data = fuse_get_mpdata(mp);
if (!data) {
panic("fuse4x: no mount private data in vfs_unmount");
}
#if M_FUSE4X_ENABLE_BIGLOCK
fuse_biglock_lock(data->biglock);
#endif
fdev = data->fdev;
if (fdata_dead_get(data)) {
/*
* If the file system daemon is dead, it's pointless to try to do
* any unmount-time operations that go out to user space. Therefore,
* we pretend that this is a force unmount. However, this isn't of much
* use. That's because if any non-root vnode is in use, the vflush()
* that the kernel does before calling our VFS_UNMOUNT will fail
* if the original unmount wasn't forcible already. That earlier
* vflush is called with SKIPROOT though, so it wouldn't bail out
* on the root vnode being in use.
*
* If we want, we could set FORCECLOSE here so that a non-forced
* unmount will be "upgraded" to a forced unmount if the root vnode
* is busy (you are cd'd to the mount point, for example). It's not
* quite pure to do that though.
*
* flags |= FORCECLOSE;
* log("fuse4x: forcing unmount on a dead file system\n");
*/
} else if (!(data->dataflags & FSESS_INITED)) {
flags |= FORCECLOSE;
log("fuse4x: forcing unmount on not-yet-alive file system\n");
fdata_set_dead(data);
}
fuse_rootvp = data->rootvp;
fuse_trace_printf("%s: Calling vflush(mp, fuse_rootvp, flags=0x%X);\n", __FUNCTION__, flags);
#if M_FUSE4X_ENABLE_BIGLOCK
fuse_biglock_unlock(data->biglock);
#endif
err = vflush(mp, fuse_rootvp, flags);
#if M_FUSE4X_ENABLE_BIGLOCK
fuse_biglock_lock(data->biglock);
#endif
fuse_trace_printf("%s: Done.\n", __FUNCTION__);
if (err) {
#if M_FUSE4X_ENABLE_BIGLOCK
fuse_biglock_unlock(data->biglock);
#endif
return err;
}
if (vnode_isinuse(fuse_rootvp, 1) && !(flags & FORCECLOSE)) {
#if M_FUSE4X_ENABLE_BIGLOCK
fuse_biglock_unlock(data->biglock);
#endif
return EBUSY;
}
if (fdata_dead_get(data)) {
goto alreadydead;
}
fdisp_init(&fdi, 0 /* no data to send along */);
fdisp_make(&fdi, FUSE_DESTROY, mp, FUSE_ROOT_ID, context);
fuse_trace_printf("%s: Waiting for reply from FUSE_DESTROY.\n", __FUNCTION__);
err = fdisp_wait_answ(&fdi);
fuse_trace_printf("%s: Reply received.\n", __FUNCTION__);
if (!err) {
fuse_ticket_drop(fdi.tick);
}
/*
* Note that dounmount() signals a VQ_UNMOUNT VFS event.
*/
fdata_set_dead(data);
alreadydead:
fuse_trace_printf("%s: Calling vnode_rele(fuse_rootp);\n", __FUNCTION__);
#if M_FUSE4X_ENABLE_BIGLOCK
fuse_biglock_unlock(data->biglock);
#endif
vnode_rele(fuse_rootvp); /* We got this reference in fuse_vfsop_mount(). */
#if M_FUSE4X_ENABLE_BIGLOCK
fuse_biglock_lock(data->biglock);
#endif
fuse_trace_printf("%s: Done.\n", __FUNCTION__);
data->rootvp = NULLVP;
fuse_trace_printf("%s: Calling vflush(mp, NULLVP, FORCECLOSE);\n", __FUNCTION__);
#if M_FUSE4X_ENABLE_BIGLOCK
fuse_biglock_unlock(data->biglock);
#endif
(void)vflush(mp, NULLVP, FORCECLOSE);
#if M_FUSE4X_ENABLE_BIGLOCK
fuse_biglock_lock(data->biglock);
#endif
fuse_trace_printf("%s: Done.\n", __FUNCTION__);
fuse_lck_mtx_lock(fdev->mtx);
vfs_setfsprivate(mp, NULL);
data->dataflags &= ~FSESS_MOUNTED;
OSAddAtomic(-1, (SInt32 *)&fuse_mount_count);
#if M_FUSE4X_ENABLE_BIGLOCK
fuse_biglock_unlock(data->biglock);
#endif
if (!(data->dataflags & FSESS_OPENED)) {
/* fdev->data was left for us to clean up */
fuse_device_close_final(fdev);
/* fdev->data is gone now */
}
fuse_lck_mtx_unlock(fdev->mtx);
return 0;
}
static errno_t
fuse_vfsop_mount(mount_t mp, __unused vnode_t devvp, user_addr_t udata,
vfs_context_t context)
{
int err = 0;
int mntopts = 0;
bool mounted = false;
uint32_t drandom = 0;
uint32_t max_read = ~0;
size_t len;
fuse_device_t fdev = NULL;
struct fuse_data *data = NULL;
fuse_mount_args fusefs_args;
struct vfsstatfs *vfsstatfsp = vfs_statfs(mp);
kern_return_t kr;
thread_t init_thread;
#if M_OSXFUSE_ENABLE_BIG_LOCK
fuse_biglock_t *biglock;
#endif
fuse_trace_printf_vfsop();
if (vfs_isupdate(mp)) {
return ENOTSUP;
}
err = copyin(udata, &fusefs_args, sizeof(fusefs_args));
if (err) {
return EINVAL;
}
/*
* Interesting flags that we can receive from mount or may want to
* otherwise forcibly set include:
*
* MNT_ASYNC
* MNT_AUTOMOUNTED
* MNT_DEFWRITE
* MNT_DONTBROWSE
* MNT_IGNORE_OWNERSHIP
* MNT_JOURNALED
* MNT_NODEV
* MNT_NOEXEC
* MNT_NOSUID
* MNT_NOUSERXATTR
* MNT_RDONLY
* MNT_SYNCHRONOUS
* MNT_UNION
*/
#if M_OSXFUSE_ENABLE_UNSUPPORTED
vfs_setlocklocal(mp);
#endif /* M_OSXFUSE_ENABLE_UNSUPPORTED */
/** Option Processing. **/
if (fusefs_args.altflags & FUSE_MOPT_FSTYPENAME) {
size_t typenamelen = strlen(fusefs_args.fstypename);
if ((typenamelen <= 0) || (typenamelen > FUSE_FSTYPENAME_MAXLEN)) {
return EINVAL;
}
snprintf(vfsstatfsp->f_fstypename, MFSTYPENAMELEN, "%s%s",
OSXFUSE_FSTYPENAME_PREFIX, fusefs_args.fstypename);
}
if ((fusefs_args.daemon_timeout > FUSE_MAX_DAEMON_TIMEOUT) ||
(fusefs_args.daemon_timeout < FUSE_MIN_DAEMON_TIMEOUT)) {
return EINVAL;
}
if (fusefs_args.altflags & FUSE_MOPT_SPARSE) {
mntopts |= FSESS_SPARSE;
}
if (fusefs_args.altflags & FUSE_MOPT_SLOW_STATFS) {
mntopts |= FSESS_SLOW_STATFS;
}
if (fusefs_args.altflags & FUSE_MOPT_AUTO_CACHE) {
mntopts |= FSESS_AUTO_CACHE;
}
if (fusefs_args.altflags & FUSE_MOPT_AUTO_XATTR) {
if (fusefs_args.altflags & FUSE_MOPT_NATIVE_XATTR) {
return EINVAL;
}
mntopts |= FSESS_AUTO_XATTR;
} else if (fusefs_args.altflags & FUSE_MOPT_NATIVE_XATTR) {
mntopts |= FSESS_NATIVE_XATTR;
}
if (fusefs_args.altflags & FUSE_MOPT_NO_BROWSE) {
vfs_setflags(mp, MNT_DONTBROWSE);
}
if (fusefs_args.altflags & FUSE_MOPT_JAIL_SYMLINKS) {
mntopts |= FSESS_JAIL_SYMLINKS;
}
/*
* Note that unlike Linux, which keeps allow_root in user-space and
* passes allow_other in that case to the kernel, we let allow_root
* reach the kernel. The 'if' ordering is important here.
*/
if (fusefs_args.altflags & FUSE_MOPT_ALLOW_ROOT) {
int is_member = 0;
if ((kauth_cred_ismember_gid(kauth_cred_get(), fuse_admin_group,
&is_member) == 0) && is_member) {
mntopts |= FSESS_ALLOW_ROOT;
} else {
IOLog("OSXFUSE: caller not a member of OSXFUSE admin group (%d)\n",
fuse_admin_group);
return EPERM;
}
} else if (fusefs_args.altflags & FUSE_MOPT_ALLOW_OTHER) {
if (!fuse_allow_other && !fuse_vfs_context_issuser(context)) {
int is_member = 0;
if ((kauth_cred_ismember_gid(kauth_cred_get(), fuse_admin_group,
&is_member) != 0) || !is_member) {
return EPERM;
}
}
mntopts |= FSESS_ALLOW_OTHER;
}
if (fusefs_args.altflags & FUSE_MOPT_NO_APPLEDOUBLE) {
mntopts |= FSESS_NO_APPLEDOUBLE;
}
if (fusefs_args.altflags & FUSE_MOPT_NO_APPLEXATTR) {
mntopts |= FSESS_NO_APPLEXATTR;
}
if ((fusefs_args.altflags & FUSE_MOPT_FSID) && (fusefs_args.fsid != 0)) {
fsid_t fsid;
mount_t other_mp;
uint32_t target_dev;
target_dev = FUSE_MAKEDEV(FUSE_CUSTOM_FSID_DEVICE_MAJOR,
fusefs_args.fsid);
fsid.val[0] = target_dev;
fsid.val[1] = FUSE_CUSTOM_FSID_VAL1;
other_mp = vfs_getvfs(&fsid);
if (other_mp != NULL) {
return EPERM;
}
vfsstatfsp->f_fsid.val[0] = target_dev;
vfsstatfsp->f_fsid.val[1] = FUSE_CUSTOM_FSID_VAL1;
} else {
vfs_getnewfsid(mp);
}
if (fusefs_args.altflags & FUSE_MOPT_NO_LOCALCACHES) {
mntopts |= FSESS_NO_ATTRCACHE;
mntopts |= FSESS_NO_READAHEAD;
mntopts |= FSESS_NO_UBC;
mntopts |= FSESS_NO_VNCACHE;
}
if (fusefs_args.altflags & FUSE_MOPT_NO_ATTRCACHE) {
mntopts |= FSESS_NO_ATTRCACHE;
}
if (fusefs_args.altflags & FUSE_MOPT_NO_READAHEAD) {
mntopts |= FSESS_NO_READAHEAD;
}
if (fusefs_args.altflags & (FUSE_MOPT_NO_UBC | FUSE_MOPT_DIRECT_IO)) {
mntopts |= FSESS_NO_UBC;
}
if (fusefs_args.altflags & FUSE_MOPT_NO_VNCACHE) {
mntopts |= FSESS_NO_VNCACHE;
}
if (fusefs_args.altflags & FUSE_MOPT_NEGATIVE_VNCACHE) {
if (mntopts & FSESS_NO_VNCACHE) {
return EINVAL;
}
mntopts |= FSESS_NEGATIVE_VNCACHE;
}
if (fusefs_args.altflags & FUSE_MOPT_NO_SYNCWRITES) {
/* Cannot mix 'nosyncwrites' with 'noubc' or 'noreadahead'. */
if (mntopts & (FSESS_NO_READAHEAD | FSESS_NO_UBC)) {
return EINVAL;
}
mntopts |= FSESS_NO_SYNCWRITES;
vfs_clearflags(mp, MNT_SYNCHRONOUS);
vfs_setflags(mp, MNT_ASYNC);
/* We check for this only if we have nosyncwrites in the first place. */
if (fusefs_args.altflags & FUSE_MOPT_NO_SYNCONCLOSE) {
mntopts |= FSESS_NO_SYNCONCLOSE;
}
} else {
vfs_clearflags(mp, MNT_ASYNC);
vfs_setflags(mp, MNT_SYNCHRONOUS);
}
if (mntopts & FSESS_NO_UBC) {
/* If no buffer cache, disallow exec from file system. */
vfs_setflags(mp, MNT_NOEXEC);
}
vfs_setauthopaque(mp);
vfs_setauthopaqueaccess(mp);
if ((fusefs_args.altflags & FUSE_MOPT_DEFAULT_PERMISSIONS) &&
(fusefs_args.altflags & FUSE_MOPT_DEFER_PERMISSIONS)) {
return EINVAL;
}
if (fusefs_args.altflags & FUSE_MOPT_DEFAULT_PERMISSIONS) {
mntopts |= FSESS_DEFAULT_PERMISSIONS;
vfs_clearauthopaque(mp);
}
if (fusefs_args.altflags & FUSE_MOPT_DEFER_PERMISSIONS) {
mntopts |= FSESS_DEFER_PERMISSIONS;
}
if (fusefs_args.altflags & FUSE_MOPT_EXTENDED_SECURITY) {
mntopts |= FSESS_EXTENDED_SECURITY;
vfs_setextendedsecurity(mp);
}
if (fusefs_args.altflags & FUSE_MOPT_LOCALVOL) {
mntopts |= FSESS_LOCALVOL;
vfs_setflags(mp, MNT_LOCAL);
}
/* done checking incoming option bits */
err = 0;
vfs_setfsprivate(mp, NULL);
fdev = fuse_device_get(fusefs_args.rdev);
if (!fdev) {
return EINVAL;
}
fuse_device_lock(fdev);
drandom = fuse_device_get_random(fdev);
if (fusefs_args.random != drandom) {
fuse_device_unlock(fdev);
IOLog("OSXFUSE: failing mount because of mismatched random\n");
return EINVAL;
}
data = fuse_device_get_mpdata(fdev);
if (!data) {
fuse_device_unlock(fdev);
return ENXIO;
}
#if M_OSXFUSE_ENABLE_BIG_LOCK
biglock = data->biglock;
fuse_biglock_lock(biglock);
#endif
if (data->mount_state != FM_NOTMOUNTED) {
#if M_OSXFUSE_ENABLE_BIG_LOCK
fuse_biglock_unlock(biglock);
#endif
fuse_device_unlock(fdev);
return EALREADY;
}
if (!(data->dataflags & FSESS_OPENED)) {
fuse_device_unlock(fdev);
err = ENXIO;
goto out;
}
data->mount_state = FM_MOUNTED;
OSAddAtomic(1, (SInt32 *)&fuse_mount_count);
mounted = true;
if (fdata_dead_get(data)) {
fuse_device_unlock(fdev);
err = ENOTCONN;
goto out;
}
if (!data->daemoncred) {
panic("OSXFUSE: daemon found but identity unknown");
}
if (fuse_vfs_context_issuser(context) &&
kauth_cred_getuid(vfs_context_ucred(context)) != kauth_cred_getuid(data->daemoncred)) {
fuse_device_unlock(fdev);
err = EPERM;
goto out;
}
data->mp = mp;
data->fdev = fdev;
data->dataflags |= mntopts;
data->daemon_timeout.tv_sec = fusefs_args.daemon_timeout;
data->daemon_timeout.tv_nsec = 0;
if (data->daemon_timeout.tv_sec) {
data->daemon_timeout_p = &(data->daemon_timeout);
} else {
data->daemon_timeout_p = (struct timespec *)0;
}
data->max_read = max_read;
data->fssubtype = fusefs_args.fssubtype;
data->mountaltflags = fusefs_args.altflags;
data->noimplflags = (uint64_t)0;
data->blocksize = fuse_round_size(fusefs_args.blocksize,
FUSE_MIN_BLOCKSIZE, FUSE_MAX_BLOCKSIZE);
data->iosize = fuse_round_size(fusefs_args.iosize,
FUSE_MIN_IOSIZE, FUSE_MAX_IOSIZE);
if (data->iosize < data->blocksize) {
data->iosize = data->blocksize;
}
data->userkernel_bufsize = FUSE_DEFAULT_USERKERNEL_BUFSIZE;
copystr(fusefs_args.fsname, vfsstatfsp->f_mntfromname,
MNAMELEN - 1, &len);
bzero(vfsstatfsp->f_mntfromname + len, MNAMELEN - len);
copystr(fusefs_args.volname, data->volname, MAXPATHLEN - 1, &len);
bzero(data->volname + len, MAXPATHLEN - len);
/* previous location of vfs_setioattr() */
vfs_setfsprivate(mp, data);
fuse_device_unlock(fdev);
/* Send a handshake message to the daemon. */
kr = kernel_thread_start(fuse_internal_init, data, &init_thread);
if (kr != KERN_SUCCESS) {
IOLog("OSXFUSE: could not start init thread\n");
err = ENOTCONN;
} else {
thread_deallocate(init_thread);
}
out:
if (err) {
vfs_setfsprivate(mp, NULL);
fuse_device_lock(fdev);
data = fuse_device_get_mpdata(fdev); /* again */
if (mounted) {
OSAddAtomic(-1, (SInt32 *)&fuse_mount_count);
}
if (data) {
data->mount_state = FM_NOTMOUNTED;
if (!(data->dataflags & FSESS_OPENED)) {
#if M_OSXFUSE_ENABLE_BIG_LOCK
assert(biglock == data->biglock);
fuse_biglock_unlock(biglock);
#endif
fuse_device_close_final(fdev);
/* data is gone now */
}
}
fuse_device_unlock(fdev);
} else {
vnode_t fuse_rootvp = NULLVP;
err = fuse_vfsop_root(mp, &fuse_rootvp, context);
if (err) {
goto out; /* go back and follow error path */
}
err = vnode_ref(fuse_rootvp);
#if M_OSXFUSE_ENABLE_BIG_LOCK
/*
* Even though fuse_rootvp will not be reclaimed when calling vnode_put
* because we incremented its usecount by calling vnode_ref release
* biglock just to be safe.
*/
fuse_biglock_unlock(biglock);
#endif /* M_OSXFUSE_ENABLE_BIG_LOCK */
(void)vnode_put(fuse_rootvp);
#if M_OSXFUSE_ENABLE_BIG_LOCK
fuse_biglock_lock(biglock);
#endif
if (err) {
goto out; /* go back and follow error path */
} else {
struct vfsioattr ioattr;
vfs_ioattr(mp, &ioattr);
ioattr.io_maxreadcnt = ioattr.io_maxwritecnt = data->iosize;
ioattr.io_segreadcnt = ioattr.io_segwritecnt = data->iosize / PAGE_SIZE;
ioattr.io_maxsegreadsize = ioattr.io_maxsegwritesize = data->iosize;
ioattr.io_devblocksize = data->blocksize;
vfs_setioattr(mp, &ioattr);
}
}
#if M_OSXFUSE_ENABLE_BIG_LOCK
fuse_device_lock(fdev);
data = fuse_device_get_mpdata(fdev); /* ...and again */
if(data) {
assert(data->biglock == biglock);
fuse_biglock_unlock(biglock);
}
fuse_device_unlock(fdev);
#endif
return err;
}
