use zlib internally to decompress

TODO

-- Internal gunzip (via lib) to speed up stuff, and avoid threads
-	- mmap() if possible for reading
+- mmapped block stream from file
 
 - Index/pkginfo reader: same field multiple times -> memleak
 

src/Makefile

@@ -15,6 +15,8 @@ apk_OBJS = \
 	package.o \
 	archive.o \
 	version.o \
+	io.o \
+	gunzip.o \
 	blob.o \
 	hash.o \
 	md5.o \
@@ -30,7 +32,7 @@ ALL_OBJS = $(apk_OBJS)
 all: $(TARGETS)
 
 apk: $(apk_OBJS)
-	$(CC) $(LDFLAGS) -o $@ $^ $(LIBS)
+	$(CC) $(LDFLAGS) -o $@ $^ $(LIBS) /usr/lib/libz.a
 
 clean::
 	@rm -f $(TARGETS) $(ALL_OBJS)

src/apk_archive.h

@@ -13,8 +13,8 @@
 #define APK_ARCHIVE
 
 #include <sys/types.h>
-#include <pthread.h>
 #include "apk_blob.h"
+#include "apk_io.h"
 
 struct apk_archive_entry {
 	char *name;
@@ -27,16 +27,19 @@ struct apk_archive_entry {
 	mode_t mode;
 	time_t mtime;
 	dev_t device;
-	int read_fd;
 };
 
-typedef int (*apk_archive_entry_parser)(struct apk_archive_entry *entry, void *ctx);
+typedef int (*apk_archive_entry_parser)(void *ctx,
+					const struct apk_archive_entry *ae,
+					struct apk_istream *istream);
 
-pid_t apk_open_gz(int *fd);
-int apk_parse_tar(int fd, apk_archive_entry_parser parser, void *ctx);
-int apk_parse_tar_gz(int fd, apk_archive_entry_parser parser, void *ctx);
-apk_blob_t apk_archive_entry_read(struct apk_archive_entry *ae);
-int apk_archive_entry_extract(struct apk_archive_entry *ae, const char *to);
-pthread_t apk_checksum_and_tee(int *fd, void *ptr);
+struct apk_istream *apk_gunzip_bstream(struct apk_bstream *);
+
+int apk_parse_tar(struct apk_istream *, apk_archive_entry_parser parser, void *ctx);
+int apk_parse_tar_gz(struct apk_bstream *, apk_archive_entry_parser parser, void *ctx);
+
+int apk_archive_entry_extract(const struct apk_archive_entry *ae,
+			      struct apk_istream *is,
+			      const char *to);
 
 #endif

src/apk_io.h

+/* apk_io.h - Alpine Package Keeper (APK)
+ *
+ * Copyright (C) 2008 Timo Teräs <timo.teras@iki.fi>
+ * All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify it 
+ * under the terms of the GNU General Public License version 2 as published
+ * by the Free Software Foundation. See http://www.gnu.org/ for details.
+ */
+
+#ifndef APK_IO
+#define APK_IO
+
+#include "apk_defines.h"
+#include "apk_blob.h"
+
+struct apk_istream {
+	size_t (*read)(void *stream, void *ptr, size_t size);
+	size_t (*splice)(void *stream, int fd, size_t size);
+	void (*close)(void *stream);
+};
+
+struct apk_bstream {
+	size_t (*read)(void *stream, void **ptr);
+	void (*close)(void *stream, csum_t csum);
+};
+
+struct apk_istream *apk_istream_from_fd(int fd);
+struct apk_istream *apk_istream_from_file(const char *file);
+size_t apk_istream_skip(struct apk_istream *istream, size_t size);
+size_t apk_istream_splice(void *stream, int fd, size_t size);
+
+struct apk_bstream *apk_bstream_from_istream(struct apk_istream *istream);
+struct apk_bstream *apk_bstream_from_fd(int fd);
+
+apk_blob_t apk_blob_from_istream(struct apk_istream *istream, size_t size);
+
+#endif

src/apk_package.h

@@ -14,6 +14,7 @@
 
 #include "apk_version.h"
 #include "apk_hash.h"
+#include "apk_io.h"
 
 struct apk_database;
 struct apk_name;
@@ -75,7 +76,7 @@ struct apk_package *apk_pkg_read(struct apk_database *db, const char *name);
 void apk_pkg_free(struct apk_package *pkg);
 
 int apk_pkg_get_state(struct apk_package *pkg);
-int apk_pkg_add_script(struct apk_package *pkg, int fd,
+int apk_pkg_add_script(struct apk_package *pkg, struct apk_istream *is,
 		       unsigned int type, unsigned int size);
 int apk_pkg_run_script(struct apk_package *pkg, int root_fd,
 		       unsigned int type);

src/archive.c

@@ -47,75 +47,62 @@ struct tar_header {
 	char padding[12];   /* 500-512 */
 };
 
-static int get_dev_null(void)
-{
-	static int fd_null = 0;
+#define GET_OCTAL(s) apk_blob_uint(APK_BLOB_PTR_LEN(s, sizeof(s)), 8)
 
-	if (fd_null == 0) {
-		fd_null = open("/dev/null", O_WRONLY);
-		if (fd_null < 0)
-			err(EX_OSFILE, "/dev/null");
-	}
-	return fd_null;
-}
+struct apk_tar_entry_istream {
+	struct apk_istream is;
+	struct apk_istream *tar_is;
+	size_t bytes_left;
+};
 
-pid_t apk_open_gz(int *fd)
+static size_t tar_entry_read(void *stream, void *ptr, size_t size)
 {
-	int pipe_fd[2];
-	pid_t child_pid;
-
-	if (pipe(pipe_fd) < 0)
-		err(EX_OSERR, "pipe");
-
-	child_pid = fork();
-	if (child_pid < 0)
-		err(EX_OSERR, "fork");
-
-	if (child_pid == 0) {
-		close(pipe_fd[0]);
-		dup2(pipe_fd[1], STDOUT_FILENO);
-		dup2(*fd, STDIN_FILENO);
-		dup2(get_dev_null(), STDERR_FILENO);
-		close(pipe_fd[1]);
-		execl(GUNZIP_BINARY, "gunzip", "-c", NULL);
-		err(EX_UNAVAILABLE, GUNZIP_BINARY);
-	}
-
-	close(pipe_fd[1]);
-	*fd = pipe_fd[0];
-
-	return child_pid;
+	struct apk_tar_entry_istream *teis =
+		container_of(stream, struct apk_tar_entry_istream, is);
+
+	if (size > teis->bytes_left)
+		size = teis->bytes_left;
+	size = teis->tar_is->read(teis->tar_is, ptr, size);
+	if (size >= 0)
+		teis->bytes_left -= size;
+	return size;
 }
 
-#define GET_OCTAL(s) apk_blob_uint(APK_BLOB_PTR_LEN(s, sizeof(s)), 8)
-
-static int do_splice(int from_fd, int to_fd, int len)
+static size_t tar_entry_splice(void *stream, int fd, size_t size)
 {
-	int i = 0, r;
-
-	while (i != len) {
-		r = splice(from_fd, NULL, to_fd, NULL, len - i, SPLICE_F_MOVE);
-		if (r == -1)
-			return i;
-		i += r;
-	}
-
-	return i;
+	struct apk_tar_entry_istream *teis =
+		container_of(stream, struct apk_tar_entry_istream, is);
+
+	if (size > teis->bytes_left)
+		size = teis->bytes_left;
+	size = teis->tar_is->splice(teis->tar_is, fd, size);
+	if (size >= 0)
+		teis->bytes_left -= size;
+	return size;
 }
 
-int apk_parse_tar(int fd, apk_archive_entry_parser parser, void *ctx)
+int apk_parse_tar(struct apk_istream *is, apk_archive_entry_parser parser,
+		  void *ctx)
 {
+	struct apk_tar_entry_istream teis = {
+		.is.read = tar_entry_read,
+		.is.splice = tar_entry_splice,
+		.tar_is = is,
+	};
 	struct apk_archive_entry entry;
 	struct tar_header buf;
 	unsigned long offset = 0;
 	int end = 0, r;
+	size_t toskip;
 
 	memset(&entry, 0, sizeof(entry));
-	while (read(fd, &buf, 512) == 512) {
+	while ((r = is->read(is, &buf, 512)) == 512) {
 		offset += 512;
 		if (buf.name[0] == '\0') {
-			if (end)
+			if (end) {
+				r = 0;
 				break;
+			}
 			end++;
 			continue;
 		}
@@ -131,7 +118,6 @@ int apk_parse_tar(int fd, apk_archive_entry_parser parser, void *ctx)
 			.gname = buf.gname,
 			.device = makedev(GET_OCTAL(buf.devmajor),
 					  GET_OCTAL(buf.devminor)),
-			.read_fd = fd,
 		};
 
 		switch (buf.typeflag) {
@@ -139,7 +125,7 @@ int apk_parse_tar(int fd, apk_archive_entry_parser parser, void *ctx)
 			if (entry.name != NULL)
 				free(entry.name);
 			entry.name = malloc(entry.size+1);
-			read(fd, entry.name, entry.size);
+			is->read(is, entry.name, entry.size);
 			offset += entry.size;
 			entry.size = 0;
 			break;
@@ -168,74 +154,48 @@ int apk_parse_tar(int fd, apk_archive_entry_parser parser, void *ctx)
 			break;
 		}
 
+		teis.bytes_left = entry.size;
 		if (entry.mode & S_IFMT) {
 			if (entry.name == NULL)
 				entry.name = strdup(buf.name);
 
 			/* callback parser function */
-			offset += entry.size;
-			r = parser(&entry, ctx);
+			r = parser(ctx, &entry, &teis.is);
 			if (r != 0)
 				return r;
-			offset -= entry.size;
 
 			free(entry.name);
 			entry.name = NULL;
 		}
 
-		if (entry.size)
-			offset += do_splice(fd, get_dev_null(), entry.size);
+		offset += entry.size - teis.bytes_left;
+		toskip = teis.bytes_left;
+		if ((offset + toskip) & 511)
+			toskip += 512 - ((offset + toskip) & 511);
+		offset += toskip;
+		if (toskip != 0)
+			is->read(is, NULL, toskip);
+	}
 
-		/* align to next 512 block */
-		if (offset & 511)
-			offset += do_splice(fd, get_dev_null(),
-					    512 - (offset & 511));
+	if (r != 0) {
+		apk_error("Bad TAR header (r=%d)", r);
+		return -1;
 	}
 
 	return 0;
 }
 
-int apk_parse_tar_gz(int fd, apk_archive_entry_parser parser, void *ctx)
-{
-	pid_t pid;
-	int r, status;
-
-	pid = apk_open_gz(&fd);
-	if (pid < 0)
-		return pid;
-
-	r = apk_parse_tar(fd, parser, ctx);
-	close(fd);
-	waitpid(pid, &status, 0);
-
-	return r;
-}
-
-apk_blob_t apk_archive_entry_read(struct apk_archive_entry *ae)
+int apk_parse_tar_gz(struct apk_bstream *bs, apk_archive_entry_parser parser,
+		     void *ctx)
 {
-	char *str;
-	int pos = 0;
-	ssize_t r;
-
-	str = malloc(ae->size + 1);
-	pos = 0;
-	while (ae->size) {
-		r = read(ae->read_fd, &str[pos], ae->size);
-		if (r < 0) {
-			free(str);
-			return APK_BLOB_NULL;
-		}
-		pos += r;
-		ae->size -= r;
-	}
-	str[pos] = 0;
-
-	return APK_BLOB_PTR_LEN(str, pos+1);
+	return apk_parse_tar(apk_gunzip_bstream(bs), parser, ctx);
 }
 
-int apk_archive_entry_extract(struct apk_archive_entry *ae, const char *fn)
+int apk_archive_entry_extract(const struct apk_archive_entry *ae,
+			      struct apk_istream *is,
+			      const char *fn)
 {
-	int r = -1;
+	int r = -1, fd;
 
 	if (fn == NULL)
 		fn = ae->name;
@@ -251,12 +211,14 @@ int apk_archive_entry_extract(struct apk_archive_entry *ae, const char *fn)
 		break;
 	case S_IFREG:
 		if (ae->link_target == NULL) {
-			r = open(fn, O_WRONLY | O_CREAT, ae->mode & 07777);
-			if (r < 0)
+			fd = open(fn, O_WRONLY | O_CREAT, ae->mode & 07777);
+			if (fd < 0) {
+				r = -1;
 				break;
-			ae->size -= do_splice(ae->read_fd, r, ae->size);
-			close(r);
-			r = ae->size ? -1 : 0;
+			}
+			if (is->splice(is, fd, ae->size) == ae->size)
+				r = 0;
+			close(fd);
 		} else {
 			r = link(ae->link_target, fn);
 		}
@@ -284,74 +246,3 @@ int apk_archive_entry_extract(struct apk_archive_entry *ae, const char *fn)
 	}
 	return r;
 }
-
-struct checksum_and_tee {
-	int in_fd, tee_fd;
-	void *ptr;
-};
-
-static void *__apk_checksum_and_tee(void *arg)
-{
-	struct checksum_and_tee *args = (struct checksum_and_tee *) arg;
-	char buf[2*1024];
-	int r, w, wt;
-	__off64_t offset;
-	csum_ctx_t ctx;
-	int dosplice = 1;
-
-	offset = lseek(args->in_fd, 0, SEEK_CUR);
-	csum_init(&ctx);
-	do {
-		r = read(args->in_fd, buf, sizeof(buf));
-		if (r <= 0)
-			break;
-
-		wt = 0;
-		do {
-			if (dosplice) {
-				w = splice(args->in_fd, &offset, args->tee_fd, NULL,
-					   r - wt, SPLICE_F_MOVE);
-				if (w < 0) {
-					dosplice = 0;
-					continue;
-				}
-			} else {
-				w = write(args->tee_fd, &buf[wt], r - wt);
-				if (w < 0)
-					break;
-			}
-			wt += w;
-		} while (wt != r);
-
-		csum_process(&ctx, buf, r);
-	} while (r == sizeof(buf));
-
-	csum_finish(&ctx, args->ptr);
-	close(args->tee_fd);
-	close(args->in_fd);
-	free(args);
-
-	return NULL;
-}
-
-pthread_t apk_checksum_and_tee(int *fd, void *ptr)
-{
-	struct checksum_and_tee *args;
-	int fds[2];
-	pthread_t tid;
-
-	if (pipe(fds) < 0)
-		return -1;
-
-	fcntl(fds[0], F_SETFD, FD_CLOEXEC);
-	fcntl(fds[1], F_SETFD, FD_CLOEXEC);
-
-	args = malloc(sizeof(*args));
-	*args = (struct checksum_and_tee){ *fd, fds[1], ptr };
-	if (pthread_create(&tid, NULL, __apk_checksum_and_tee, args) < 0)
-		return -1;
-
-	*fd = fds[0];
-	return tid;
-}
-

src/database.c

@@ -376,14 +376,14 @@ static int apk_db_write_scriptdb(struct apk_database *db, int fd)
 	return 0;
 }
 
-static int apk_db_read_scriptdb(struct apk_database *db, int fd)
+static int apk_db_read_scriptdb(struct apk_database *db, struct apk_istream *is)
 {
 	struct apk_package *pkg;
 	struct apk_script_header hdr;
 
-	while (read(fd, &hdr, sizeof(hdr)) == sizeof(hdr)) {
+	while (is->read(is, &hdr, sizeof(hdr)) == sizeof(hdr)) {
 		pkg = apk_db_get_pkg(db, hdr.csum);
-		apk_pkg_add_script(pkg, fd, hdr.type, hdr.size);
+		apk_pkg_add_script(pkg, is, hdr.type, hdr.size);
 	}
 
 	return 0;
@@ -416,6 +416,7 @@ int apk_db_create(const char *root)
 
 static int apk_db_read_config(struct apk_database *db)
 {
+	struct apk_istream *is;
 	struct stat st;
 	char *buf;
 	int fd;
@@ -452,10 +453,10 @@ static int apk_db_read_config(struct apk_database *db)
 		close(fd);
 	}
 
-	fd = open("var/lib/apk/scripts", O_RDONLY);
-	if (fd >= 0) {
-		apk_db_read_scriptdb(db, fd);
-		close(fd);
+	is = apk_istream_from_file("var/lib/apk/scripts");
+	if (is != NULL) {
+		apk_db_read_scriptdb(db, is);
+		is->close(is);
 	}
 
 	return 0;
@@ -664,9 +665,11 @@ int apk_db_recalculate_and_commit(struct apk_database *db)
 	return r;
 }
 
-static int apk_db_install_archive_entry(struct apk_archive_entry *ae,
-					struct install_ctx *ctx)
+static int apk_db_install_archive_entry(void *_ctx,
+					const struct apk_archive_entry *ae,
+					struct apk_istream *is)
 {
+	struct install_ctx *ctx = (struct install_ctx *) _ctx;
 	struct apk_database *db = ctx->db;
 	struct apk_package *pkg = ctx->pkg;
 	apk_blob_t name = APK_BLOB_STR(ae->name);
@@ -698,8 +701,7 @@ static int apk_db_install_archive_entry(struct apk_archive_entry *ae,
 
 	/* Handle script */
 	if (type != APK_SCRIPT_INVALID) {
-		ae->size -= apk_pkg_add_script(pkg, ae->read_fd,
-					       type, ae->size);
+		apk_pkg_add_script(pkg, is, type, ae->size);
 
 		if (type == APK_SCRIPT_GENERIC ||
 		    type == ctx->script) {
@@ -736,7 +738,7 @@ static int apk_db_install_archive_entry(struct apk_archive_entry *ae,
 		if (strncmp(file->filename, ".keep_", 6) == 0)
 			return 0;
 
-		r = apk_archive_entry_extract(ae, NULL);
+		r = apk_archive_entry_extract(ae, is, NULL);
 	} else {
 		if (name.ptr[name.len-1] == '/')
 			name.len--;
@@ -775,10 +777,10 @@ int apk_db_install_pkg(struct apk_database *db,
 		       struct apk_package *oldpkg,
 		       struct apk_package *newpkg)
 {
+	struct apk_bstream *bs;
 	struct install_ctx ctx;
 	csum_t csum;
 	char file[256];
-	pthread_t tid = 0;
 	int fd, r;
 
 	if (fchdir(db->root_fd) < 0)
@@ -813,8 +815,8 @@ int apk_db_install_pkg(struct apk_database *db,
 
 	fcntl(fd, F_SETFD, FD_CLOEXEC);
 
-	tid = apk_checksum_and_tee(&fd, csum);
-	if (tid < 0)
+	bs = apk_bstream_from_fd(fd);
+	if (bs == NULL)
 		goto err_close;
 
 	ctx = (struct install_ctx) {
@@ -823,12 +825,10 @@ int apk_db_install_pkg(struct apk_database *db,
 		.script = (oldpkg == NULL) ?
 			APK_SCRIPT_PRE_INSTALL : APK_SCRIPT_PRE_UPGRADE,
 	};
-	if (apk_parse_tar_gz(fd, (apk_archive_entry_parser)
-			     apk_db_install_archive_entry, &ctx) != 0)
+	if (apk_parse_tar_gz(bs, apk_db_install_archive_entry, &ctx) != 0)
 		goto err_close;
 
-	pthread_join(tid, NULL);
-	close(fd);
+	bs->close(bs, csum);
 
 	db->installed.stats.packages++;
 	hlist_add_head(&newpkg->installed_pkgs_list, &db->installed.packages);
@@ -850,7 +850,5 @@ int apk_db_install_pkg(struct apk_database *db,
 
 err_close:
 	close(fd);
-	if (tid != 0)
-		pthread_join(tid, NULL);
 	return -1;
 }

src/gunzip.c

+/* gunzip.c - Alpine Package Keeper (APK)
+ *
+ * Copyright (C) 2008 Timo Teräs <timo.teras@iki.fi>
+ * All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify it 
+ * under the terms of the GNU General Public License version 2 as published
+ * by the Free Software Foundation. See http://www.gnu.org/ for details.
+ */
+
+#include <stdio.h>
+#include <fcntl.h>
+#include <unistd.h>
+#include <malloc.h>
+#include <zlib.h>
+
+#include "apk_defines.h"
+#include "apk_io.h"
+
+struct apk_gzip_istream {
+	struct apk_istream is;
+	struct apk_bstream *bs;
+	z_stream zs;
+	int z_err;
+};
+
+static size_t gz_read(void *stream, void *ptr, size_t size)
+{
+	struct apk_gzip_istream *gis =
+		container_of(stream, struct apk_gzip_istream, is);
+
+	if (gis->z_err == Z_DATA_ERROR || gis->z_err == Z_ERRNO)
+		return -1;
+	if (gis->z_err == Z_STREAM_END)
+		return 0;
+
+	if (ptr == NULL)
+		return apk_istream_skip(&gis->is, size);
+
+	gis->zs.avail_out = size;
+	gis->zs.next_out  = ptr;
+
+	while (gis->zs.avail_out != 0 && gis->z_err == Z_OK) {
+		if (gis->zs.avail_in == 0) {
+			gis->zs.avail_in = gis->bs->read(gis->bs, (void **) &gis->zs.next_in);
+			if (gis->zs.avail_in < 0) {
+				gis->z_err = Z_DATA_ERROR;
+				return size - gis->zs.avail_out;
+			}
+		}
+
+		gis->z_err = inflate(&gis->zs, Z_NO_FLUSH);
+	}
+
+	if (gis->z_err != Z_OK && gis->z_err != Z_STREAM_END)
+		return -1;
+
+	return size - gis->zs.avail_out;
+}
+
+static void gz_close(void *stream)
+{
+	struct apk_gzip_istream *gis =
+		container_of(stream, struct apk_gzip_istream, is);
+
+	inflateEnd(&gis->zs);
+	free(gis);
+}
+
+struct apk_istream *apk_gunzip_bstream(struct apk_bstream *bs)
+{
+	struct apk_gzip_istream *gis;
+
+	gis = malloc(sizeof(struct apk_gzip_istream));
+	if (gis == NULL)
+		return NULL;
+
+	*gis = (struct apk_gzip_istream) {
+		.is.read = gz_read,
+		.is.splice = apk_istream_splice,
+		.is.close = gz_close,
+		.bs = bs,
+		.z_err = 0,
+	};
+
+	if (inflateInit2(&gis->zs, 15+32) != Z_OK) {
+		free(gis);
+		return NULL;
+	}
+
+	return &gis->is;
+}
+

src/io.c

+/* io.c - Alpine Package Keeper (APK)
+ *
+ * Copyright (C) 2005-2008 Natanael Copa <n@tanael.org>
+ * Copyright (C) 2008 Timo Teräs <timo.teras@iki.fi>
+ * All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify it 
+ * under the terms of the GNU General Public License version 2 as published
+ * by the Free Software Foundation. See http://www.gnu.org/ for details.
+ */
+
+#include <stdio.h>
+#include <fcntl.h>
+#include <unistd.h>
+#include <malloc.h>
+
+#include "apk_defines.h"
+#include "apk_io.h"
+
+struct apk_fd_istream {
+	struct apk_istream is;
+	int fd;
+};
+
+static size_t fd_read(void *stream, void *ptr, size_t size)
+{
+	struct apk_fd_istream *fis =
+		container_of(stream, struct apk_fd_istream, is);
+	size_t i = 0, r;
+
+	if (ptr == NULL) {
+		if (lseek(fis->fd, size, SEEK_CUR) < 0)
+			return -1;
+		return size;
+	}
+
+	while (i < size) {
+		r = read(fis->fd, ptr + i, size - i);
+		if (r < 0)
+			return r;
+		if (r == 0)
+			return i;
+		i += r;
+	}
+
+	return i;
+}
+
+static size_t fd_splice(void *stream