Added symlink support to vfs and plumbed it to syscall layer

This commit requires a rebuild of the toolchain.

kernel/fs/vfs.c

@@ -16,6 +16,9 @@
 #include <logging.h>
 #include <hashmap.h>
 
+#define MAX_SYMLINK_DEPTH 8
+#define MAX_SYMLINK_SIZE 4096
+
 tree_t    * fs_tree = NULL; /* File system mountpoint tree */
 fs_node_t * fs_root = NULL; /* Pointer to the root mount fs_node (must be some form of filesystem, even ramdisk) */
 
@@ -365,6 +368,54 @@ fs_node_t *clone_fs(fs_node_t *source) {
 	return source;
 }
 
+int symlink_fs(char * target, char * name) {
+	fs_node_t * parent;
+	char *cwd = (char *)(current_process->wd_name);
+	char *path = canonicalize_path(cwd, name);
+
+	char * parent_path = malloc(strlen(path) + 4);
+	sprintf(parent_path, "%s/..", path);
+
+	char * f_path = path + strlen(path) - 1;
+	while (f_path > path) {
+		if (*f_path == '/') {
+			f_path += 1;
+			break;
+		}
+		f_path--;
+	}
+
+	debug_print(WARNING, "creating symlink %s within %s", f_path, parent_path);
+
+	parent = kopen(parent_path, 0);
+	free(parent_path);
+
+	if (!parent) {
+		free(path);
+		return -1;
+	}
+
+	if (parent->symlink) {
+		parent->symlink(parent, target, f_path);
+	}
+
+	free(path);
+	close_fs(parent);
+
+	return 0;
+}
+
+int readlink_fs(fs_node_t *node, char * buf, uint32_t size) {
+	if (!node) return -1;
+
+	if (node->readlink) {
+		return node->readlink(node, buf, size);
+	} else {
+		return -1;
+	}
+}
+
+
 /**
  * canonicalize_path: Canonicalize a path.
  *
@@ -721,31 +772,14 @@ fs_node_t *get_mount_point(char * path, unsigned int path_depth, char **outpath,
 
 
 
-
-
-/**
- * kopen: Open a file by name.
- *
- * Explore the file system tree to find the appropriate node for
- * for a given path. The path can be relative to the working directory
- * and will be canonicalized by the kernel.
- *
- * @param filename Filename to open
- * @param flags    Flag bits for read/write mode.
- * @returns A file system node element that the caller can free.
- */
-fs_node_t *kopen(char *filename, uint32_t flags) {
+fs_node_t *kopen_recur(char *filename, uint32_t flags, uint32_t symlink_depth, char *relative_to) {
 	/* Simple sanity checks that we actually have a file system */
 	if (!filename) {
 		return NULL;
 	}
 
-	debug_print(INFO, "kopen(%s)", filename);
-
-	/* Reference the current working directory */
-	char *cwd = (char *)(current_process->wd_name);
 	/* Canonicalize the (potentially relative) path... */
-	char *path = canonicalize_path(cwd, filename);
+	char *path = canonicalize_path(relative_to, filename);
 	/* And store the length once to save recalculations */
 	size_t path_len = strlen(path);
 
@@ -791,6 +825,8 @@ fs_node_t *kopen(char *filename, uint32_t flags) {
 	unsigned int depth = 0;
 	/* Find the mountpoint for this file */
 	fs_node_t *node_ptr = get_mount_point(path, path_depth, &path_offset, &depth);
+	debug_print(CRITICAL, "path_offset: %s", path_offset);
+	debug_print(CRITICAL, "depth: %d", depth);
 
 	if (!node_ptr) return NULL;
 
@@ -804,13 +840,80 @@ fs_node_t *kopen(char *filename, uint32_t flags) {
 		/* Search the active directory for the requested directory */
 		debug_print(INFO, "... Searching for %s", path_offset);
 		node_next = finddir_fs(node_ptr, path_offset);
-		free(node_ptr);
+		close_fs(node_ptr);
 		node_ptr = node_next;
 		if (!node_ptr) {
 			/* We failed to find the requested directory */
 			free((void *)path);
 			return NULL;
-		} else if (depth == path_depth - 1) {
+		}
+		/* 
+		 * This test is a little complicated, but we basically always resolve symlinks in the
+		 * of a path (like /home/symlink/file) even if O_NOFOLLOW and O_PATH are set. If we are
+		 * on the leaf of the path then we will look at those flags and act accordingly
+		 */
+		if ((node_ptr->flags & FS_SYMLINK) &&
+				!((flags & O_NOFOLLOW) && (flags & O_PATH) && depth == path_depth - 1)) {
+			/* This ensures we don't return a path when NOFOLLOW is requested but PATH
+			 * isn't passed.
+			 */
+			debug_print(CRITICAL, "resolving symlink at %s", node_ptr->name);
+			if ((flags & O_NOFOLLOW) && depth == path_depth - 1) {
+				/* TODO(gerow): should probably be setting errno from this */
+				debug_print(NOTICE, "Refusing to follow final entry for open with O_NOFOLLOW for %s.", node_ptr->name);
+				free((void *)path);
+				close_fs(node_ptr);
+				return NULL;
+			}
+			if (symlink_depth >= MAX_SYMLINK_DEPTH) {
+				/* TODO(gerow): should probably be setting errno from this */
+				debug_print(WARNING, "Reached max symlink depth on %s.", node_ptr->name);
+				free((void *)path);
+				close_fs(node_ptr);
+				return NULL;
+			}
+			/* 
+			 * This may actually be big enough that we wouldn't want to allocate it on
+			 * the stack, especially considering this function is called recursively
+			 */
+			char symlink_buf[MAX_SYMLINK_SIZE];
+			int len = node_ptr->readlink(node_ptr, symlink_buf, sizeof(symlink_buf));
+			if (len < 0) {
+				/* TODO(gerow): should probably be setting errno from this */
+				debug_print(WARNING, "Got error %d from symlink for %s.", len, node_ptr->name);
+				free((void *)path);
+				close_fs(node_ptr);
+				return NULL;
+			}
+			if (symlink_buf[len - 1] != '\0') {
+				/* TODO(gerow): should probably be setting errno from this */
+				debug_print(WARNING, "readlink for %s doesn't end in a null pointer. That's weird...", len, node_ptr->name);
+				free((void *)path);
+				close_fs(node_ptr);
+				return NULL;
+			}
+			fs_node_t * old_node_ptr = node_ptr;
+			/* Rebuild our path up to this point. This is hella hacky. */
+			char * relpath = malloc(path_len + 1);
+			char * ptr = relpath;
+			memcpy(relpath, path, path_len + 1);
+			for (unsigned int i = 0; i < depth; i++) {
+				while(*ptr != '\0') {
+					ptr++;
+				}
+				*ptr = PATH_SEPARATOR;
+			}
+			node_ptr = kopen_recur(symlink_buf, 0, symlink_depth + 1, relpath);
+			free(relpath);
+			close_fs(old_node_ptr);
+			if (!node_ptr) {
+				/* Dangling symlink? */
+				debug_print(WARNING, "Failed to open symlink path %s. Perhaps it's a dangling symlink?", symlink_buf);
+				free((void *)path);
+				return NULL;
+			}
+		}
+		if (depth == path_depth - 1) {
 			/* We found the file and are done, open the node */
 			open_fs(node_ptr, flags);
 			free((void *)path);
@@ -825,3 +928,20 @@ fs_node_t *kopen(char *filename, uint32_t flags) {
 	return NULL;
 }
 
+/**
+ * kopen: Open a file by name.
+ *
+ * Explore the file system tree to find the appropriate node for
+ * for a given path. The path can be relative to the working directory
+ * and will be canonicalized by the kernel.
+ *
+ * @param filename Filename to open
+ * @param flags    Flag bits for read/write mode.
+ * @returns A file system node element that the caller can free.
+ */
+fs_node_t *kopen(char *filename, uint32_t flags) {
+	debug_print(NOTICE, "kopen(%s)", filename);
+
+	return kopen_recur(filename, flags, 0, (char *)(current_process->wd_name));
+}
+

kernel/include/fs.h

@@ -15,6 +15,8 @@
 #define O_CREAT      0x0200
 #define O_TRUNC      0x0400
 #define O_EXCL       0x0800
+#define O_NOFOLLOW   0x1000
+#define O_PATH       0x2000
 
 #define FS_FILE        0x01
 #define FS_DIRECTORY   0x02
@@ -47,6 +49,8 @@ typedef void (*mkdir_type_t) (struct fs_node *, char *name, uint16_t permission)
 typedef int (*ioctl_type_t) (struct fs_node *, int request, void * argp);
 typedef int (*get_size_type_t) (struct fs_node *);
 typedef int (*chmod_type_t) (struct fs_node *, int mode);
+typedef void (*symlink_type_t) (struct fs_node *, char * name, char * value);
+typedef int (*readlink_type_t) (struct fs_node *, char * buf, size_t size);
 
 typedef struct fs_node {
 	char name[256];         /* The filename. */
@@ -78,6 +82,8 @@ typedef struct fs_node {
 	get_size_type_t get_size;
 	chmod_type_t chmod;
 	unlink_type_t unlink;
+	symlink_type_t symlink;
+	readlink_type_t readlink;
 
 	struct fs_node *ptr;   /* Alias pointer, for symlinks. */
 	uint32_t offset;       /* Offset for read operations XXX move this to new "file descriptor" entry */
@@ -129,6 +135,8 @@ fs_node_t *clone_fs(fs_node_t * source);
 int ioctl_fs(fs_node_t *node, int request, void * argp);
 int chmod_fs(fs_node_t *node, int mode);
 int unlink_fs(char * name);
+int symlink_fs(char * value, char * name);
+int readlink_fs(fs_node_t * node, char * buf, size_t size);
 
 void vfs_install(void);
 void * vfs_mount(char * path, fs_node_t * local_root);

kernel/sys/syscall.c

@@ -663,6 +663,38 @@ static int sys_mount(char * arg, char * mountpoint, char * type, unsigned long f
 	return -EFAULT;
 }
 
+static int sys_symlink(char * target, char * name) {
+	PTR_VALIDATE(target);
+	PTR_VALIDATE(name);
+	return symlink_fs(target, name);
+}
+
+static int sys_readlink(const char * file, char * ptr, int len) {
+	PTR_VALIDATE(file);
+	fs_node_t * node = kopen((char *) file, O_PATH | O_NOFOLLOW);
+	if (!node) {
+		return -ENOENT;
+	}
+	int rv = readlink_fs(node, ptr, len);
+	close_fs(node);
+	return rv;
+}
+
+static int sys_lstat(char * file, uintptr_t st) {
+	int result;
+	PTR_VALIDATE(file);
+	PTR_VALIDATE(st);
+	fs_node_t * fn = kopen(file, O_PATH | O_NOFOLLOW);
+	result = stat_node(fn, st);
+	if (fn) {
+		close_fs(fn);
+	}
+	return result;
+}
+
+/*
+ * System Call Internals
+ */
 static int (*syscalls[])() = {
 	/* System Call Table */
 	[SYS_EXT]          = sys_exit,
@@ -709,6 +741,9 @@ static int (*syscalls[])() = {
 	[SYS_WAITPID]      = sys_waitpid,
 	[SYS_PIPE]         = sys_pipe,
 	[SYS_MOUNT]        = sys_mount,
+	[SYS_SYMLINK]      = sys_symlink,
+	[SYS_READLINK]     = sys_readlink,
+	[SYS_LSTAT]        = sys_lstat,
 };
 
 uint32_t num_syscalls = sizeof(syscalls) / sizeof(*syscalls);

toolchain/patches/newlib/include/syscall_nums.h

@@ -42,3 +42,6 @@
 #define SYS_WAITPID 53
 #define SYS_PIPE 54
 #define SYS_MOUNT 55
+#define SYS_SYMLINK 56
+#define SYS_READLINK 57
+#define SYS_LSTAT 58

toolchain/patches/newlib/toaru/syscalls.c

@@ -89,6 +89,9 @@ DEFN_SYSCALL1(unlink, 52, char *);
 DEFN_SYSCALL3(waitpid, 53, int, int *, int);
 DEFN_SYSCALL1(pipe, 54, int *);
 DEFN_SYSCALL5(mount, SYS_MOUNT, char *, char *, char *, unsigned long, void *);
+DEFN_SYSCALL2(symlink, SYS_SYMLINK, char *, char *);
+DEFN_SYSCALL3(readlink, SYS_READLINK, char *, char *, int);
+DEFN_SYSCALL2(lstat, SYS_LSTAT, char *, void *);
 
 static int toaru_debug_stubs_enabled(void) {
 	static int checked = 0;
@@ -319,8 +322,15 @@ char *getwd(char *buf) {
 	return getcwd(buf, 256);
 }
 
-int lstat(const char *path, struct stat *buf) {
-	return stat(path, buf);
+int lstat(const char *path, struct stat *st) {
+	int ret = syscall_lstat((char *)path, (void *)st);
+	if (ret >= 0) {
+		return ret;
+	} else {
+		errno = -ret;
+		memset(st, 0x00, sizeof(struct stat));
+		return ret;
+	}
 }
 
 int mkdir(const char *pathname, mode_t mode) {
@@ -688,3 +698,24 @@ int mount(char * source, char * target, char * type, unsigned long flags, void *
 	return r;
 }
 
+int symlink(char * target, char * name) {
+	int r = syscall_symlink(target, name);
+
+	if (r < 0) {
+		errno = -r;
+		return -1;
+	}
+
+	return r;
+}
+
+int readlink(char * name, char * buf, size_t len) {
+	int r = syscall_readlink(name, buf, len);
+
+	if (r < 0) {
+		errno = -r;
+		return -1;
+	}
+
+	return r;
+}