summaryrefslogtreecommitdiffstats
path: root/lib/libarchive/archive_private.h
diff options
context:
space:
mode:
authorkientzle <kientzle@FreeBSD.org>2004-02-09 23:22:54 +0000
committerkientzle <kientzle@FreeBSD.org>2004-02-09 23:22:54 +0000
commitaf9413b539628b0758b4af66ffbb879fba49cc6d (patch)
tree20e1d80fd0a1d288a08af1696ee258bcd08f41d0 /lib/libarchive/archive_private.h
parent6b6e533f7e91602d4ccb3353351eb9f816c2daf2 (diff)
downloadFreeBSD-src-af9413b539628b0758b4af66ffbb879fba49cc6d.zip
FreeBSD-src-af9413b539628b0758b4af66ffbb879fba49cc6d.tar.gz
Initial import of libarchive.
What it is: A library for reading and writing various streaming archive formats, especially tar and cpio. Being a library, it should be easy to incorporate into pkg_* tools, sysinstall, and any other place that needs to read or write such archives. Features: * Full automatic detection of both compression and archive format. * Extensible internal architecture to make it easy to add new formats. * Support for "pax interchange format," a new POSIX-standard tar format that eliminates essentially all of the restrictions of historic formats. * BSD license Thanks to: jkh for pushing me to start this work, gordon for encouraging me to commit it, bde for answering endless style questions, and many others for feedback and encouragement. Status: Pretty good overall, though there are still a few rough edges and the library could always use more testing. Feedback eagerly solicited.
Diffstat (limited to 'lib/libarchive/archive_private.h')
-rw-r--r--lib/libarchive/archive_private.h267
1 files changed, 267 insertions, 0 deletions
diff --git a/lib/libarchive/archive_private.h b/lib/libarchive/archive_private.h
new file mode 100644
index 0000000..58c093b
--- /dev/null
+++ b/lib/libarchive/archive_private.h
@@ -0,0 +1,267 @@
+/*-
+ * Copyright (c) 2003-2004 Tim Kientzle
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ * notice, this list of conditions and the following disclaimer
+ * in this position and unchanged.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ * notice, this list of conditions and the following disclaimer in the
+ * documentation and/or other materials provided with the distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE AUTHOR(S) ``AS IS'' AND ANY EXPRESS OR
+ * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
+ * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
+ * IN NO EVENT SHALL THE AUTHOR(S) BE LIABLE FOR ANY DIRECT, INDIRECT,
+ * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
+ * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
+ * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ *
+ * $FreeBSD$
+ */
+
+#ifndef ARCHIVE_PRIVATE_H_INCLUDED
+#define ARCHIVE_PRIVATE_H_INCLUDED
+
+#include <stdint.h>
+
+#include "archive.h"
+#include "archive_string.h"
+
+#define ARCHIVE_WRITE_MAGIC (0xb0c5c0deU)
+#define ARCHIVE_READ_MAGIC (0xdeb0c5U)
+
+/*
+ * This is used by archive_extract to keep track of non-writable
+ * directories so that they can be initially restored writable, then
+ * fixed up at end. This also handles mtime/atime fixups.
+ */
+struct archive_extract_dir_entry {
+ struct archive_extract_dir_entry *next;
+ mode_t mode;
+ int64_t mtime;
+ int64_t atime;
+ unsigned long mtime_nanos;
+ unsigned long atime_nanos;
+ /* Note: ctime cannot be restored, so don't bother */
+ char *name;
+};
+
+struct archive {
+ /*
+ * The magic/state values are used to sanity-check the
+ * client's usage. If an API function is called at a
+ * rediculous time, or the client passes us an invalid
+ * pointer, these values allow me to catch that.
+ */
+ unsigned magic;
+ unsigned state;
+
+ struct archive_entry *entry;
+
+ /*
+ * Space to store per-entry strings. Most header strings are
+ * copied here from the format-specific header, in order to
+ * gaurantee null-termination. Maybe these should go into
+ * per-format storage?
+ */
+ struct archive_string entry_name;
+ struct archive_string entry_linkname;
+ struct archive_string entry_uname;
+ struct archive_string entry_gname;
+
+ /* Utility: Pointer to a block of nulls. */
+ const char *nulls;
+ size_t null_length;
+
+ /*
+ * Used to limit reads of entry data. Eventually, each reader
+ * will be able to register it's own read_data routine and these
+ * will move into the per-format data for the formats that use them.
+ */
+ uint64_t entry_bytes_remaining;
+ uint64_t entry_padding; /* Skip this much after entry data. */
+
+ uid_t user_uid; /* UID of current user. */
+
+ /* Callbacks to open/read/write/close archive stream. */
+ archive_open_callback *client_opener;
+ archive_read_callback *client_reader;
+ archive_write_callback *client_writer;
+ archive_close_callback *client_closer;
+ void *client_data;
+
+ /*
+ * Blocking information. Note that bytes_in_last_block is
+ * misleadingly named; I should find a better name. These
+ * control the final output from all compressors, including
+ * compression_none.
+ */
+ int bytes_per_block;
+ int bytes_in_last_block;
+
+ /*
+ * These control whether data within a gzip/bzip2 compressed
+ * stream gets padded or not. If pad_uncompressed is set,
+ * the data will be padded to a full block before being
+ * compressed. The pad_uncompressed_byte determines the value
+ * that will be used for padding. Note that these have no
+ * effect on compression "none."
+ */
+ int pad_uncompressed;
+ int pad_uncompressed_byte; /* TODO: Support this. */
+
+ /*
+ * PAX extended header data. When reading,
+ * name/linkname/uname/gname fields may point into here. This
+ * should be moved into per-format data storage.
+ */
+ struct archive_string pax_header;
+
+ /*
+ * GNU header fields. These should be moved into format-specific
+ * storage.
+ */
+ struct archive_string gnu_name;
+ struct archive_string gnu_linkname;
+ int gnu_header_recursion_depth;
+
+ /* Position in UNCOMPRESSED data stream. */
+ intmax_t file_position;
+ /* File offset of beginning of most recently-read header. */
+ intmax_t header_position;
+
+ /*
+ * Detection functions for decompression: bid functions are
+ * given a block of data from the beginning of the stream and
+ * can bid on whether or not they support the data stream.
+ * General guideline: bid the number of bits that you actually
+ * test, e.g., 16 if you test a 2-byte magic value. The
+ * highest bidder will have their init function invoked, which
+ * can set up pointers to specific handlers.
+ *
+ * On write, the client just invokes an archive_write_set function
+ * which sets up the data here directly.
+ */
+ int compression_code; /* Currently active compression. */
+ const char *compression_name;
+ struct {
+ int (*bid)(const void *buff, size_t);
+ int (*init)(struct archive *, const void *buff, size_t);
+ } decompressors[4];
+ /* Read/write data stream (with compression). */
+ void *compression_data; /* Data for (de)compressor. */
+ int (*compression_init)(struct archive *); /* Initialize. */
+ int (*compression_finish)(struct archive *);
+ ssize_t (*compression_write)(struct archive *, const void *, size_t);
+ /*
+ * Read uses a peek/consume I/O model: the decompression code
+ * returns a pointer to the requested block and advances the
+ * file position only when requested by a consume call. This
+ * reduces copying and also simplifies look-ahead for format
+ * detection.
+ */
+ ssize_t (*compression_read_ahead)(struct archive *,
+ const void **, size_t request);
+ ssize_t (*compression_read_consume)(struct archive *, size_t);
+
+ /*
+ * Format detection is mostly the same as compression
+ * detection, with two significant differences: The bidders
+ * use the read_ahead calls above to examine the stream rather
+ * than having the supervisor hand them a block of data to
+ * examine, and the auction is repeated for every header.
+ * Winning bidders should set the archive_format and
+ * archive_format_name appropriately. Bid routines should
+ * check archive_format and decline to bid if the format of
+ * the last header was incompatible.
+ *
+ * Again, write support is considerably simpler because there's
+ * no need for an auction.
+ */
+ int archive_format;
+ const char *archive_format_name;
+
+ struct archive_format_descriptor {
+ int (*bid)(struct archive *);
+ int (*read_header)(struct archive *, struct archive_entry *);
+ int (*cleanup)(struct archive *);
+ void *format_data; /* Format-specific data for readers. */
+ } formats[4];
+ struct archive_format_descriptor *format; /* Active format. */
+
+ /*
+ * Storage for format-specific data. Note that there can be
+ * multiple format readers active at one time, so we need to
+ * allow for multiple format readers to have their data
+ * available. The pformat_data slot here is the solution: on
+ * read, it's set up in the bid phase and is gauranteed to
+ * always point to a void* variable that the format can use.
+ */
+ void **pformat_data; /* Pointer to current format_data. */
+ void *format_data; /* Used by writers. */
+
+ /*
+ * Pointers to format-specific functions. On read, these are
+ * initialized in the bid process. On write, they're initialized by
+ * archive_write_set_format_XXX() calls.
+ */
+ int (*format_init)(struct archive *); /* Only used on write. */
+ int (*format_finish)(struct archive *);
+ int (*format_finish_entry)(struct archive *);
+ int (*format_write_header)(struct archive *,
+ struct archive_entry *);
+ int (*format_write_data)(struct archive *,
+ const void *buff, size_t);
+
+ /*
+ * Various information needed by archive_extract.
+ */
+ struct archive_string extract_mkdirpath;
+ struct archive_extract_dir_entry *archive_extract_dir_list;
+ void (*cleanup_archive_extract)(struct archive *);
+
+ int archive_error_number;
+ const char *error;
+ struct archive_string error_string;
+};
+
+
+/* Utility function to format a USTAR header into a buffer. */
+int
+__archive_write_format_header_ustar(struct archive *, char buff[512],
+ struct archive_entry *);
+
+#define ARCHIVE_STATE_ANY 0xFFFFU
+#define ARCHIVE_STATE_NEW 1U
+#define ARCHIVE_STATE_HEADER 2U
+#define ARCHIVE_STATE_DATA 4U
+#define ARCHIVE_STATE_EOF 8U
+#define ARCHIVE_STATE_CLOSED 0x10U
+#define ARCHIVE_STATE_FATAL 0x8000U
+
+/* Check magic value and state; exit if it isn't valid. */
+void
+__archive_check_magic(struct archive *, unsigned magic,
+ unsigned state, const char *func);
+
+#define archive_check_magic(a,m,s) \
+ __archive_check_magic((a), (m), (s), __func__)
+
+int __archive_read_register_format(struct archive *a,
+ void *format_data,
+ int (*bid)(struct archive *),
+ int (*read_header)(struct archive *, struct archive_entry *),
+ int (*cleanup)(struct archive *));
+
+int __archive_read_register_compression(struct archive *a,
+ int (*bid)(const void *, size_t),
+ int (*init)(struct archive *, const void *, size_t));
+
+#endif
OpenPOWER on IntegriCloud