syscalls: fsync

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

@@ -65,6 +65,7 @@ extern syscall_sigreturn
 extern syscall_uname
 extern syscall_hostname
 extern syscall_sync
+extern syscall_fsync
 syscalltab:
 dq syscall_print
 dq syscall_mmap
@@ -132,7 +133,8 @@ dq syscall_sigreturn
 dq syscall_uname
 dq syscall_hostname
 dq syscall_sync
-syscallcount equ 66
+dq syscall_fsync
+syscallcount equ 67
 section .text
 global arch_syscall_entry
 ; on entry:

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

@@ -4,7 +4,7 @@
 
 #ifdef SYSCALL_LOGGING
 
-#define SYSCALL_COUNT 66
+#define SYSCALL_COUNT 67
 #define LOGSTR(x) arch_e9_puts(x)
 
 static char *name[] = {
@@ -73,7 +73,8 @@ static char *name[] = {
 	"sigreturn",
 	"uname",
 	"hostname",
-	"sync"
+	"sync",
+	"fsync"
 };
 
 static char *args[] = {
@@ -142,7 +143,8 @@ static char *args[] = {
 	"N/A", // sigreturn
 	"utsname %p", // uname
 	"new %p newsize %lu old %p oldsize %lu", // hostname
-	"N/A" // sync
+	"N/A", // sync
+	"fd %d" // fsync
 };
 
 #endif

kernel-src/fs/ext2.c

@@ -1654,6 +1654,15 @@ static int ext2_rename(vnode_t *sourcedir, char *oldname, vnode_t *targetdir, ch
 	return err;
 }
 
+static int ext2_sync(vnode_t *vnode) {
+	int e = vmmcache_sync(vnode);
+	ext2fs_t *fs = (ext2fs_t *)vnode->vfs;
+	// TODO don't sync the entire disk but rather only the inodes and blocks
+	int e2 = vmmcache_sync(fs->backing);
+	// only the first errors are reported
+	return e ? e : e2;
+}
+
 static int ext2_inactive(vnode_t *vnode) {
 	ext2node_t *node = (ext2node_t *)vnode;
 	ext2fs_t *fs = (ext2fs_t *)vnode->vfs;
@@ -1829,7 +1838,8 @@ static vops_t vnops = {
 	.inactive = ext2_inactive,
 	.rename = ext2_rename,
 	.getpage = ext2_getpage,
-	.putpage = ext2_putpage
+	.putpage = ext2_putpage,
+	.sync = ext2_sync
 };
 
 void ext2_init() {

kernel-src/fs/pipefs.c

@@ -260,7 +260,8 @@ static vops_t vnops = {
 	.resize = pipefs_enodev,
 	.rename = pipefs_enodev,
 	.putpage = pipefs_enodev,
-	.getpage = pipefs_enodev
+	.getpage = pipefs_enodev,
+	.sync = pipefs_enodev
 };
 
 static void ctor(scache_t *cache, void *obj) {

kernel-src/fs/sockfs.c

@@ -186,7 +186,8 @@ static vops_t vnops = {
 	.rename = sockfs_enodev,
 	.ioctl = sockfs_ioctl,
 	.putpage = sockfs_enodev,
-	.getpage = sockfs_enodev
+	.getpage = sockfs_enodev,
+	.sync = sockfs_enodev
 };
 
 static void ctor(scache_t *cache, void *obj) {

kernel-src/fs/tmpfs.c

@@ -493,6 +493,11 @@ static int tmpfs_putpage(vnode_t *node, uintmax_t offset, struct page_t *page) {
 	return 0;
 }
 
+static int tmpfs_sync(vnode_t *node) {
+	// sync is a no-op on tmpfs
+	return 0;
+}
+
 static vfsops_t vfsops = {
 	.mount = tmpfs_mount,
 	.root = tmpfs_root
@@ -520,7 +525,8 @@ static vops_t vnops = {
 	.resize = tmpfs_resize,
 	.rename = tmpfs_rename,
 	.getpage = tmpfs_getpage,
-	.putpage = tmpfs_putpage
+	.putpage = tmpfs_putpage,
+	.sync = tmpfs_sync
 };
 
 static tmpfsnode_t *newnode(vfs_t *vfs, int type) {

kernel-src/include/kernel/vfs.h

@@ -105,6 +105,7 @@ typedef struct vops_t {
 	int (*rename)(vnode_t *source, char *oldname, vnode_t *target, char *newname, int flags);
 	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);
 } vops_t;
 
 #define VFS_INIT(v, o, f) \
@@ -156,6 +157,7 @@ typedef struct vops_t {
 #define VOP_RENAME(s, o, t, n, f) (s)->ops->rename(s, o, t, n, f)
 #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_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) {\

kernel-src/include/kernel/vmmcache.h

@@ -9,7 +9,7 @@ int vmmcache_getpage(vnode_t *vnode, uintmax_t offset, page_t **res);
 int vmmcache_takepage(page_t *page);
 int vmmcache_makedirty(page_t *page);
 int vmmcache_truncate(vnode_t *vnode, uintmax_t offset);
-int vmmcache_syncvnode(vnode_t *vnode);
+int vmmcache_syncvnode(vnode_t *vnode, uintmax_t startoffset, size_t size);
 int vmmcache_sync();
 
 #endif

kernel-src/mm/vmmcache.c

@@ -232,12 +232,73 @@ int vmmcache_truncate(vnode_t *vnode, uintmax_t offset) {
 	return 0;
 }
 
-int vmmcache_syncvnode(vnode_t *vnode) {
-	__assert(!"unimplemented");
+// called with lock held
+// returns with lock released
+static int syncpage(page_t *page) {
+	__assert(page->flags & PAGE_FLAGS_DIRTY);
+	page->flags &= ~PAGE_FLAGS_DIRTY;
+	RELEASE_LOCK();
+	int e = 0;
+	if ((page->flags & PAGE_FLAGS_TRUNCATED) == 0) {
+		e = VOP_PUTPAGE(page->backing, page->offset, (page_t *)page);
+		VOP_RELEASE(page->backing);
+	}
+
+	pmm_release(pmm_getpageaddress((page_t *)page));
+	return e;
 }
 
 static page_t *dirtylist;
 
+int vmmcache_syncvnode(vnode_t *vnode, uintmax_t offset, size_t size) {
+	offset = ROUND_DOWN(offset, PAGE_SIZE);
+	uintmax_t top = offset + size;
+	// overflow check
+	__assert(top > offset);
+	HOLD_LOCK();
+
+	// loop through all vnode pages in memory and check which ones are in the range and are dirty
+	// TODO create a proper vnode dirty list as to not have to loop through the ENTIRE thing in memory
+	page_t *page = vnode->pages;
+	page_t *vnodedirtylist = NULL;
+	for (; page; page = page->vnodenext) {
+		if (page->offset < offset || page->offset >= top)
+			continue;
+
+		// remove from write list and add to an internal list using the write pointers
+		// in a singly linked list way
+		if (page->writenext)
+			page->writenext->writeprev = page->writeprev;
+
+		if (page->writeprev)
+			page->writeprev->writenext = page->writenext;
+		else
+			dirtylist = page->writenext;
+
+		page->writenext = vnodedirtylist;
+		page->writeprev = NULL;
+		vnodedirtylist = page;
+	}
+
+	RELEASE_LOCK();
+
+	int e = 0;
+	while (vnodedirtylist) {
+		// in the case of failure, only the first error to occur will be reported and we will not
+		// retry the write and keep on syncing the pages to disk
+		HOLD_LOCK();
+		page_t *page = vnodedirtylist;
+		vnodedirtylist = vnodedirtylist->writenext;
+		page->writenext = NULL;
+		int error = syncpage(page);
+		if (e == 0)
+			e = error;
+		// syncpage returns with lock released
+	}
+
+	return e;
+}
+
 int vmmcache_sync() {
 	polldesc_t polldesc = {0};
 	int e = poll_initdesc(&polldesc, 1);
@@ -313,22 +374,13 @@ static void writer() {
 			continue;
 		}
 
-		__assert(page->flags & PAGE_FLAGS_DIRTY);
-		page->flags &= ~PAGE_FLAGS_DIRTY;
-
 		dirtylist = page->writenext;
 		if (dirtylist)
 			dirtylist->writeprev = NULL;
 
 		page->writenext = NULL;
-
-		RELEASE_LOCK();
-		if ((page->flags & PAGE_FLAGS_TRUNCATED) == 0) {
-			VOP_PUTPAGE(page->backing, page->offset, (page_t *)page);
-			VOP_RELEASE(page->backing);
-		}
-
-		pmm_release(pmm_getpageaddress((page_t *)page));
+		// TODO notify error on vmmcache_syncvnode
+		syncpage((page_t *)page);
 	}
 }
 

kernel-src/sys/syscalls/sync.c

@@ -6,3 +6,20 @@ syscallret_t syscall_sync(context_t *) {
 	vmmcache_sync();
 	return ret;
 }
+
+syscallret_t syscall_fsync(context_t *context, int fd) {
+	syscallret_t ret = {
+		.ret = -1
+	};
+
+	file_t *file = fd_get(fd);
+	if (file == NULL) {
+		ret.errno = EBADF;
+		return ret;
+	}
+
+	ret.errno = VOP_SYNC(file->vnode);
+
+	fd_release(file);
+	return ret;
+}