vfs: advisory locking + flock syscall

kernel-src/arch/x86-64/syscall.asm

@@ -93,6 +93,7 @@ extern syscall_sigpending
 extern syscall_killthread
 extern syscall_shutdown
 extern syscall_nice
+extern syscall_flock
 syscalltab:
 dq syscall_print
 dq syscall_mmap
@@ -188,7 +189,8 @@ dq syscall_sigpending
 dq syscall_killthread
 dq syscall_shutdown
 dq syscall_nice
-syscallcount equ 94
+dq syscall_flock
+syscallcount equ 95
 section .text
 global arch_syscall_entry
 ; on entry:

kernel-src/arch/x86-64/syscalllog.c

@@ -5,7 +5,7 @@
 
 #ifdef SYSCALL_LOGGING
 
-#define SYSCALL_COUNT 94
+#define SYSCALL_COUNT 95
 #define LOGSTR(x) arch_e9_puts(x)
 
 static char *name[] = {
@@ -102,7 +102,8 @@ static char *name[] = {
 	"sigpending",
 	"killthread",
 	"shutdown",
-	"nice"
+	"nice",
+	"flock"
 };
 
 static char *args[] = {
@@ -199,7 +200,8 @@ static char *args[] = {
 	"sigset %p", // sigpending
 	"pid %d tid %d signal %d", // killthread
 	"how %d", // shutdown
-	"off %d" // nice
+	"off %d", // nice
+	"fd %d op %d" // flock
 };
 
 #endif

kernel-src/fs/devfs.c

@@ -442,6 +442,14 @@ static int devfs_lock(vnode_t *vnode) {
 static int devfs_unlock(vnode_t *vnode) {
 	return 0;
 }
+	
+static int devfs_advlock(vnode_t *node, int op, advlock_t *advlock) {
+	devnode_t *devnode = (devnode_t *)node;
+	if (devnode->master)
+		devnode = devnode->master;
+
+	return vfs_advlock((vnode_t *)devnode, op, advlock);
+}
 
 static int devfs_enodev() {
 	return ENODEV;
@@ -479,6 +487,7 @@ static vops_t vnops = {
 	.putpage = devfs_putpage,
 	.getpage = devfs_getpage,
 	.sync = devfs_sync,
+	.advlock = devfs_advlock,
 	.lock = devfs_lock,
 	.unlock = devfs_unlock
 };

kernel-src/fs/ext2.c

@@ -1823,6 +1823,10 @@ static int ext2_unlock(vnode_t *vnode) {
 	return 0;
 }
 
+static int ext2_advlock(vnode_t *node, int op, advlock_t *advlock) {
+	return vfs_advlock(node, op, advlock);
+}
+
 static int ext2_inactive(vnode_t *vnode) {
 	ext2node_t *node = (ext2node_t *)vnode;
 	ext2fs_t *fs = (ext2fs_t *)vnode->vfs;
@@ -2010,6 +2014,7 @@ static vops_t vnops = {
 	.getpage = ext2_getpage,
 	.putpage = ext2_putpage,
 	.sync = ext2_sync,
+	.advlock = ext2_advlock,
 	.lock = ext2_lock,
 	.unlock = ext2_unlock
 };

kernel-src/fs/file.c

@@ -20,6 +20,7 @@ static bool ctor(scache_t *cache, void *obj) {
 	file->offset = 0;
 	file->flags = 0;
 	file->mode = 0;
+	file->advlock = NULL;
 	file->refcount = 1;
 	return true;
 }
@@ -39,12 +40,18 @@ static void cleanfile(file_t *file) {
 		VOP_RELEASE(file->vnode);
 	}
 
-	file->vnode = NULL;
 	file->refcount = 1;
 	file->offset = 0;
 	file->flags = 0;
 	file->mode = 0;
 	MUTEX_INIT(&file->mutex);
+	if (file->advlock) {
+		// try unlocking it if applicable
+		VOP_ADVLOCK(file->vnode, ADVLOCK_UNLOCK, file->advlock);
+		ADVLOCK_UNREF(file->advlock);
+		file->advlock = NULL;
+	}
+	file->vnode = NULL;
 	slab_free(filecache, file);
 }
 

kernel-src/fs/pipefs.c

@@ -364,6 +364,10 @@ static int pipefs_unlock(vnode_t *vnode) {
 	return 0;
 }
 
+static int pipefs_advlock(vnode_t *node, int op, advlock_t *advlock) {
+	return vfs_advlock(node, op, advlock);
+}
+
 static int pipefs_enodev() {
 	return ENODEV;
 }
@@ -392,6 +396,7 @@ static vops_t vnops = {
 	.putpage = pipefs_enodev,
 	.getpage = pipefs_enodev,
 	.sync = pipefs_enodev,
+	.advlock = pipefs_advlock,
 	.lock = pipefs_lock,
 	.unlock = pipefs_unlock
 };

kernel-src/fs/sockfs.c

@@ -229,6 +229,10 @@ static int sockfs_unlock(vnode_t *) {
 	return 0;
 }
 
+static int sockfs_advlock(vnode_t *node, int op, advlock_t *advlock) {
+	return vfs_advlock(node, op, advlock);
+}
+
 static int sockfs_enodev() {
 	return ENODEV;
 }
@@ -258,6 +262,7 @@ static vops_t vnops = {
 	.putpage = sockfs_enodev,
 	.getpage = sockfs_enodev,
 	.sync = sockfs_enodev,
+	.advlock = sockfs_advlock,
 	.lock = sockfs_lock,
 	.unlock = sockfs_unlock
 };

kernel-src/fs/tmpfs.c

@@ -503,6 +503,10 @@ static int tmpfs_sync(vnode_t *node) {
 	return 0;
 }
 
+static int tmpfs_advlock(vnode_t *node, int op, advlock_t *advlock) {
+	return vfs_advlock(node, op, advlock);
+}
+
 static int tmpfs_lock(vnode_t *node) {
 	MUTEX_ACQUIRE(&node->lock);
 	return 0;
@@ -542,6 +546,7 @@ static vops_t vnops = {
 	.getpage = tmpfs_getpage,
 	.putpage = tmpfs_putpage,
 	.sync = tmpfs_sync,
+	.advlock = tmpfs_advlock,
 	.lock = tmpfs_lock,
 	.unlock = tmpfs_unlock
 };

kernel-src/fs/vfs.c

@@ -12,6 +12,7 @@
 #include <kernel/pipefs.h>
 #include <kernel/auth.h>
 #include <kernel/init.h>
+#include <kernel/slab.h>
 
 #define PATHNAME_MAX 512
 #define MAXLINKDEPTH 64
@@ -21,6 +22,7 @@ vnode_t *vfsroot;
 
 static mutex_t listlock;
 static vfs_t *vfslist;
+static scache_t *advlock_cache;
 
 static cred_t *getcred() {
 	if (current_thread() == NULL || current_thread()->proc == NULL)
@@ -50,6 +52,15 @@ static vops_t vnops = {
 	.unlock = noop
 };
 
+static bool advlock_ctor(scache_t *cache, void *_obj) {
+	advlock_t *obj = _obj;
+	obj->type = ADVLOCK_UNLOCK;
+	obj->shared_count = 0;
+	obj->refcount = 1;
+	EVENT_INITHEADER(&obj->unlock_event);
+	return true;
+}
+
 void vfs_init() {
 	__assert(hashtable_init(&fstable, 20) == 0);
 	vfsroot = alloc(sizeof(vnode_t));
@@ -58,10 +69,21 @@ void vfs_init() {
 	vfsroot->type = V_TYPE_DIR;
 	vfsroot->refcount = 1;
 	vfsroot->ops = &vnops;
+	advlock_cache = slab_newcache(sizeof(advlock_t), 0, advlock_ctor, NULL);
+	__assert(advlock_cache);
 }
 
 INIT_ROUTINE_DEFINE(vfs, INIT_ROUTINE_FLAGS_NONE, vfs_init, bsp_early);
 
+advlock_t *advlock_allocate(void) {
+	return slab_allocate(advlock_cache);
+}
+
+void advlock_free(advlock_t *advlock) {
+	advlock->refcount = 1;
+	slab_free(advlock_cache, advlock);
+}
+
 int vfs_register(vfsops_t *ops, char *name) {
 	return hashtable_set(&fstable, ops, name, strlen(name), true);
 }
@@ -166,6 +188,140 @@ int vfs_close(vnode_t *node, int flags) {
 	return err;
 }
 
+static void adv_unlock(vnode_t *vnode, int op, advlock_t *lock) {
+	MUTEX_ACQUIRE(&vnode->adv_mutex);
+
+	switch (lock->type) {
+		// is this lock not attached anywhere?
+		case ADVLOCK_UNLOCK:
+			goto leave;
+		case ADVLOCK_EXCLUSIVE:
+			__assert(vnode->advlock == lock);
+
+			EVENT_SIGNAL(&lock->unlock_event);
+			ADVLOCK_UNREF(lock);
+			vnode->advlock = NULL;
+
+			lock->type = ADVLOCK_UNLOCK;
+
+			break;
+		case ADVLOCK_SHARED:
+			__assert(vnode->advlock && vnode->advlock->type == ADVLOCK_SHARED);
+			lock->type = ADVLOCK_UNLOCK;
+
+			if (--vnode->advlock->shared_count == 0) {
+				// we do not change the type as this will be destroyed
+				// (shared locks allocate an extra advlock)
+				EVENT_SIGNAL(&vnode->advlock->unlock_event);
+				ADVLOCK_UNREF(vnode->advlock);
+				vnode->advlock = NULL;
+			}
+	}
+
+	leave:
+	MUTEX_RELEASE(&vnode->adv_mutex);
+}
+
+static int adv_lock(vnode_t *vnode, int op, advlock_t *lock, bool non_blocking) {
+	retry:
+	MUTEX_ACQUIRE(&vnode->adv_mutex);
+	int error = 0;
+
+	// do we have to convert the lock?
+	if ((lock->type == ADVLOCK_EXCLUSIVE && op == ADVLOCK_SHARED) ||
+		(lock->type == ADVLOCK_SHARED && op == ADVLOCK_EXCLUSIVE)) {
+		error = EDEADLK;
+		goto leave;
+	}
+
+	// have we already taken it?
+	if (lock->type == op) {
+		goto leave;
+	}
+
+	if ((op == ADVLOCK_EXCLUSIVE && vnode->advlock && vnode->advlock != lock) ||
+	    (op == ADVLOCK_SHARED    && vnode->advlock && vnode->advlock->type == ADVLOCK_EXCLUSIVE)
+	) {
+		// can't take lock, sleep
+		if (non_blocking) {
+			error = EWOULDBLOCK;
+			goto leave;
+		}
+
+		advlock_t *sleeping_on = vnode->advlock;
+		ADVLOCK_REF(sleeping_on);
+
+		eventlistener_t listener;
+		EVENT_INITLISTENER(&listener);
+		EVENT_ATTACH(&listener, &sleeping_on->unlock_event);
+
+		MUTEX_RELEASE(&vnode->adv_mutex);
+
+		error = EVENT_WAIT(&listener, 0);
+		EVENT_DETACHALL(&listener);
+
+		ADVLOCK_UNREF(sleeping_on);
+
+		// interrupted by a signal?
+		if (error)
+			return error;
+
+		goto retry;
+	}
+
+	// can take the lock!
+	lock->type = op;
+
+	if (vnode->advlock == NULL) {
+		// unlocked advlock
+		if (op == ADVLOCK_SHARED) {
+			// shared locks allocate a dummy one
+			vnode->advlock = advlock_allocate();
+			if (vnode->advlock == NULL) {
+				error = ENOMEM;
+				goto leave;
+			}
+		} else {
+			// exclusive lock
+			vnode->advlock = lock;
+			ADVLOCK_REF(vnode->advlock);
+		}
+	} else {
+		// shared lock on shared advlock
+		++vnode->advlock->shared_count;
+	}
+
+	leave:
+	MUTEX_RELEASE(&vnode->adv_mutex);
+	return error;
+}
+
+// expected to be called from inside filesystem specific code
+// adv_mutex is expected to be unlocked
+// returns 0 if successful lock, EWOULDBLOCK if unsucessful and non-blocking
+// a ADVLOCK_UNLOCK operation decreases the shared count or releases an exclusive lock if applicable, waking all waiters in case of a successful unlock
+// a ADVLOCK_SHARED operation takes an unlocked lock or increments the shared count of a shared one
+// a ADVLOCK_EXCLUSIVE operation either places the lock or sleeps waiting to do so
+int vfs_advlock(vnode_t *node, int op, advlock_t *lock) {
+	int error = 0;
+	int non_blocking = op & ADVLOCK_NON_BLOCKING;
+	op &= ~ADVLOCK_NON_BLOCKING;
+
+	if (op == ADVLOCK_UNLOCK) {
+		adv_unlock(node, op, lock);
+	} else {
+		retry:
+		error = adv_lock(node, op, lock, non_blocking);
+		if (error == EDEADLK) {
+			// EDEADLK is a way of adv_lock signaling that the lock should be converted, and thus must be unlocked and then relocked non-atomically
+			adv_unlock(node, ADVLOCK_UNLOCK, lock);
+			goto retry;
+		}
+	}
+
+	return error;
+}
+
 // if node is not NULL, then a reference is kept and the newnode is returned in node
 // ref is expected to be unlocked.
 // the new node is returned locked

kernel-src/include/kernel/file.h

@@ -4,6 +4,7 @@
 #include <kernel/vfs.h>
 #include <mutex.h>
 #include <kernel/scheduler.h>
+#include <kernel/event.h>
 
 typedef struct file_t {
 	vnode_t *vnode;
@@ -12,6 +13,7 @@ typedef struct file_t {
 	mode_t mode;
 	uintmax_t offset;
 	int flags;
+	advlock_t *advlock;
 } file_t;
 
 typedef struct fd_t {

kernel-src/include/kernel/poll.h

@@ -31,7 +31,8 @@ typedef struct polldata {
 } polldata_t;
 
 #define POLL_INITHEADER(x) \
-	SPINLOCK_INIT((x)->lock);
+	SPINLOCK_INIT((x)->lock); \
+	(x)->data = NULL;
 
 int poll_initdesc(polldesc_t *, size_t size);
 void poll_add(pollheader_t *, polldata_t *, int events);

kernel-src/include/kernel/vfs.h

@@ -9,6 +9,7 @@
 #include <time.h>
 #include <errno.h>
 #include <kernel/cred.h>
+#include <kernel/event.h>
 
 #define V_ATTR_MODE	1
 #define V_ATTR_UID	2
@@ -71,6 +72,21 @@ typedef struct vfs_t {
 	int flags;
 } vfs_t;
 
+#define ADVLOCK_SHARED 1
+#define ADVLOCK_EXCLUSIVE 2
+#define ADVLOCK_NON_BLOCKING 4
+#define ADVLOCK_UNLOCK 8
+
+typedef struct advlock_t {
+	int type;
+	int refcount;
+	int shared_count;
+	eventheader_t unlock_event;
+} advlock_t;
+
+#define ADVLOCK_REF(x) __atomic_fetch_add(&(x)->refcount, 1, __ATOMIC_SEQ_CST);
+#define ADVLOCK_UNREF(x) if (__atomic_sub_fetch(&(x)->refcount, 1, __ATOMIC_SEQ_CST) == 0) advlock_free(x);
+
 #define V_FLAGS_ROOT 1
 
 #define V_TYPE_REGULAR	0
@@ -105,6 +121,9 @@ typedef struct vnode_t {
 
 	mutex_t pages_mutex;
 	struct page_t *pages;
+
+	mutex_t adv_mutex;
+	advlock_t *advlock;
 } vnode_t;
 
 typedef struct vfsops_t {
@@ -143,6 +162,7 @@ typedef struct vops_t {
 	int (*getpage)(vnode_t *node, uintmax_t offset, struct page_t *page);
 	int (*putpage)(vnode_t *node, uintmax_t offset, struct page_t *page);
 	int (*sync)(vnode_t *node);
+	int (*advlock)(vnode_t *node, int op, advlock_t *advlock);
 	int (*lock)(vnode_t *node);
 	int (*unlock)(vnode_t *node);
 } vops_t;
@@ -164,6 +184,8 @@ typedef struct vops_t {
 	MUTEX_INIT(&(vn)->lock); \
 	MUTEX_INIT(&(vn)->size_lock); \
 	MUTEX_INIT(&(vn)->pages_mutex); \
+	MUTEX_INIT(&(vn)->adv_mutex); \
+	(vn)->advlock = NULL; \
 	(vn)->refcount = 1; \
 	(vn)->flags = f; \
 	(vn)->type = t; \
@@ -201,6 +223,7 @@ typedef struct vops_t {
 #define VOP_GETPAGE(v, o, p) (v)->ops->getpage(v, o, p)
 #define VOP_PUTPAGE(v, o, p) (v)->ops->putpage(v, o, p)
 #define VOP_SYNC(v) (v)->ops->sync(v)
+#define VOP_ADVLOCK(v, o, a) (v)->ops->advlock(v, o, a)
 #define VOP_HOLD(v) __atomic_add_fetch(&(v)->refcount, 1, __ATOMIC_SEQ_CST)
 #define VOP_RELEASE(v) {\
 		if (__atomic_sub_fetch(&(v)->refcount, 1, __ATOMIC_SEQ_CST) == 0) {\
@@ -211,6 +234,9 @@ typedef struct vops_t {
 
 extern vnode_t *vfsroot;
 
+advlock_t *advlock_allocate(void);
+void advlock_free(advlock_t *advlock);
+
 void vfs_init();
 int vfs_mount(vnode_t *backing, vnode_t *pathref, char *path, char *name, void *data);
 int vfs_register(vfsops_t *vfsops, char *name);
@@ -226,6 +252,7 @@ int vfs_link(vnode_t *destref, char *destpath, vnode_t *linkref, char *linkpath,
 int vfs_rename(vnode_t *srcref, char *srcpath, vnode_t *dstref, char *dstpath, int flags);
 int vfs_unlink(vnode_t *ref, char *path);
 int vfs_pollstub(vnode_t *node, struct polldata *, int events);
+int vfs_advlock(vnode_t *node, int op, advlock_t *lock);
 void vfs_inactive(vnode_t *node);
 
 #define VFS_LOOKUP_PARENT 	0x20000000

kernel-src/sys/syscalls/flock.c

@@ -0,0 +1,74 @@
+#include <kernel/syscalls.h>
+#include <kernel/file.h>
+#include <kernel/vfs.h>
+
+syscallret_t syscall_flock(context_t *, int fd, int op_flag) {
+	syscallret_t ret = {
+		.ret = -1
+	};
+
+	int op = op_flag &= ~ADVLOCK_NON_BLOCKING;
+	switch (op) {
+		case ADVLOCK_UNLOCK:
+		case ADVLOCK_SHARED:
+		case ADVLOCK_EXCLUSIVE:
+			break;
+		default:
+			ret.errno = EINVAL;
+			return ret;
+	}
+
+	file_t *file = fd_get(fd);
+	if (file == NULL) {
+		ret.errno = EBADF;
+		return ret;
+	}
+
+	MUTEX_ACQUIRE(&file->mutex);
+
+	advlock_t *advlock = file->advlock;
+	if (advlock)
+		ADVLOCK_REF(advlock);
+
+	if (op == ADVLOCK_UNLOCK && advlock == NULL) {
+		// trying to unlock a lock which does not exist, just leave
+		ret.errno = 0;
+		ret.ret = 0;
+		goto leave;
+	}
+
+	if (op != ADVLOCK_UNLOCK && advlock == NULL) {
+		// locking and the lock does not exist yet, allocate it
+		advlock = advlock_allocate();
+		if (advlock == NULL) {
+			ret.errno = ENOMEM;
+			goto leave;
+		}
+
+		file->advlock = advlock;
+		advlock->refcount = 2;
+	}
+
+	vnode_t *vn = file->vnode;
+	VOP_HOLD(vn);
+
+	// release the file so that it can be closed etc in case we block
+	MUTEX_RELEASE(&file->mutex);
+	fd_release(file);
+
+	ret.errno = VOP_ADVLOCK(vn, op_flag, advlock);
+	ret.ret = ret.errno ? -1 : 0;
+
+	VOP_RELEASE(vn);
+	ADVLOCK_UNREF(advlock);
+
+	return ret;
+
+	leave:
+	if (advlock)
+		ADVLOCK_UNREF(advlock);
+	MUTEX_RELEASE(&file->mutex);
+	fd_release(file);
+
+	return ret;
+}

patches/mlibc/jinx-working-patch.patch

@@ -84,7 +84,7 @@ index d5ba7b2..a86353f 100644
  	long error = syscall(SYSCALL_ARCHCTL, &ret, func, (uint64_t)arg);
  	if(error)
 diff --git mlibc-clean/sysdeps/astral/generic/generic.cpp mlibc-workdir/sysdeps/astral/generic/generic.cpp
-index bb85dba..a541845 100644
+index bb85dba..5e9e21a 100644
 --- mlibc-clean/sysdeps/astral/generic/generic.cpp
 +++ mlibc-workdir/sysdeps/astral/generic/generic.cpp
 @@ -31,6 +31,12 @@ namespace mlibc {
@@ -100,10 +100,15 @@ index bb85dba..a541845 100644
  	int sys_getsockopt(int fd, int layer, int number,
  		void *__restrict buffer, socklen_t *__restrict size) {
  		(void)size;
-@@ -40,6 +46,18 @@ namespace mlibc {
+@@ -40,6 +46,23 @@ namespace mlibc {
  		return 0;
  	}
  
++	int sys_flock(int fd, int options) {
++		long r;
++		return syscall(SYSCALL_FLOCK, &r, fd, options);
++	}
++
 +	int sys_nice(int nice, int *ret) {
 +		long r = -1;
 +		long e = syscall(SYSCALL_NICE, &r, nice);
@@ -119,7 +124,7 @@ index bb85dba..a541845 100644
  	int sys_tgkill(int pid, int tid, int sig) {
  		long ret;
  		return syscall(SYSCALL_KILLTHREAD, &ret, pid, tid, sig);
-@@ -161,7 +179,18 @@ namespace mlibc {
+@@ -161,7 +184,18 @@ namespace mlibc {
  		*bytes_written = writec;
  		return error;
  	}
@@ -139,7 +144,7 @@ index bb85dba..a541845 100644
  	#ifndef MLIBC_BUILDING_RTLD
  
  	typedef struct {
-@@ -376,6 +405,12 @@ namespace mlibc {
+@@ -376,6 +410,12 @@ namespace mlibc {
  		return ret;
  	}
  
@@ -152,7 +157,7 @@ index bb85dba..a541845 100644
  	#ifndef MLIBC_BUILDING_RTLD
  
  	[[noreturn]] void sys_thread_exit() {
-@@ -771,7 +806,7 @@ namespace mlibc {
+@@ -771,7 +811,7 @@ namespace mlibc {
  
  	int sys_tcb_set(void *pointer) {
  		long r;
@@ -179,15 +184,16 @@ index d738e02..7e4f2a1 100644
  #ifdef __cplusplus
  extern "C" {
 diff --git mlibc-clean/sysdeps/astral/include/astral/syscall.h mlibc-workdir/sysdeps/astral/include/astral/syscall.h
-index 5f42d62..09a5d5f 100644
+index 5f42d62..6fbb253 100644
 --- mlibc-clean/sysdeps/astral/include/astral/syscall.h
 +++ mlibc-workdir/sysdeps/astral/include/astral/syscall.h
-@@ -96,6 +96,8 @@
+@@ -96,6 +96,9 @@
  #define SYSCALL_SIGTIMEDWAIT 89
  #define SYSCALL_SIGPENDING 90
  #define SYSCALL_KILLTHREAD 91
 +#define SYSCALL_SHUTDOWN 92
 +#define SYSCALL_NICE 93
++#define SYSCALL_FLOCK 94
  
  #ifndef __MLIBC_ABI_ONLY