1 <!-- Creator : groff version 1.22.3 -->
2 <!-- CreationDate: Fri Apr 29 22:04:24 2016 -->
3 <!DOCTYPE html PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN"
4 "http://www.w3.org/TR/html4/loose.dtd">
7 <meta name="generator" content="groff -Thtml, see www.gnu.org">
8 <meta http-equiv="Content-Type" content="text/html; charset=US-ASCII">
9 <meta name="Content-Style" content="text/css">
10 <style type="text/css">
11 p { margin-top: 0; margin-bottom: 0; vertical-align: top }
12 pre { margin-top: 0; margin-bottom: 0; vertical-align: top }
13 table { margin-top: 0; margin-bottom: 0; vertical-align: top }
14 h1 { text-align: center }
23 <p>LIBARCHIVE_INTERNALS(3) BSD Library Functions Manual
24 LIBARCHIVE_INTERNALS(3)</p>
26 <p style="margin-top: 1em"><b>NAME</b></p>
28 <p style="margin-left:6%;"><b>libarchive_internals</b>
29 — description of libarchive internal interfaces</p>
31 <p style="margin-top: 1em"><b>OVERVIEW</b></p>
33 <p style="margin-left:6%;">The <b>libarchive</b> library
34 provides a flexible interface for reading and writing
35 streaming archive files such as tar and cpio. Internally, it
36 follows a modular layered design that should make it easy to
37 add new archive and compression formats.</p>
39 <p style="margin-top: 1em"><b>GENERAL ARCHITECTURE</b></p>
41 <p style="margin-left:6%;">Externally, libarchive exposes
42 most operations through an opaque, object-style interface.
43 The archive_entry(3) objects store information about a
44 single filesystem object. The rest of the library provides
45 facilities to write archive_entry(3) objects to archive
46 files, read them from archive files, and write them to disk.
47 (There are plans to add a facility to read archive_entry(3)
48 objects from disk as well.)</p>
50 <p style="margin-left:6%; margin-top: 1em">The read and
51 write APIs each have four layers: a public API layer, a
52 format layer that understands the archive file format, a
53 compression layer, and an I/O layer. The I/O layer is
54 completely exposed to clients who can replace it entirely
55 with their own functions.</p>
57 <p style="margin-left:6%; margin-top: 1em">In order to
58 provide as much consistency as possible for clients, some
59 public functions are virtualized. Eventually, it should be
60 possible for clients to open an archive or disk writer, and
61 then use a single set of code to select and write entries,
62 regardless of the target.</p>
64 <p style="margin-top: 1em"><b>READ ARCHITECTURE</b></p>
66 <p style="margin-left:6%;">From the outside, clients use
67 the archive_read(3) API to manipulate an <b>archive</b>
68 object to read entries and bodies from an archive stream.
69 Internally, the <b>archive</b> object is cast to an
70 <b>archive_read</b> object, which holds all read-specific
71 data. The API has four layers: The lowest layer is the I/O
72 layer. This layer can be overridden by clients, but most
73 clients use the packaged I/O callbacks provided, for
74 example, by archive_read_open_memory(3), and
75 archive_read_open_fd(3). The compression layer calls the I/O
76 layer to read bytes and decompresses them for the format
77 layer. The format layer unpacks a stream of uncompressed
78 bytes and creates <b>archive_entry</b> objects from the
79 incoming data. The API layer tracks overall state (for
80 example, it prevents clients from reading data before
81 reading a header) and invokes the format and compression
82 layer operations through registered function pointers. In
83 particular, the API layer drives the format-detection
84 process: When opening the archive, it reads an initial block
85 of data and offers it to each registered compression
86 handler. The one with the highest bid is initialized with
87 the first block. Similarly, the format handlers are polled
88 to see which handler is the best for each archive. (Prior to
89 2.4.0, the format bidders were invoked for each entry, but
90 this design hindered error recovery.)</p>
92 <p style="margin-left:6%; margin-top: 1em"><b>I/O Layer and
93 Client Callbacks</b> <br>
94 The read API goes to some lengths to be nice to clients. As
95 a result, there are few restrictions on the behavior of the
98 <p style="margin-left:6%; margin-top: 1em">The client read
99 callback is expected to provide a block of data on each
100 call. A zero-length return does indicate end of file, but
101 otherwise blocks may be as small as one byte or as large as
102 the entire file. In particular, blocks may be of different
105 <p style="margin-left:6%; margin-top: 1em">The client skip
106 callback returns the number of bytes actually skipped, which
107 may be much smaller than the skip requested. The only
108 requirement is that the skip not be larger. In particular,
109 clients are allowed to return zero for any skip that they
110 don’t want to handle. The skip callback must never be
111 invoked with a negative value.</p>
113 <p style="margin-left:6%; margin-top: 1em">Keep in mind
114 that not all clients are reading from disk: clients reading
115 from networks may provide different-sized blocks on every
116 request and cannot skip at all; advanced clients may use
117 mmap(2) to read the entire file into memory at once and
118 return the entire file to libarchive as a single block;
119 other clients may begin asynchronous I/O operations for the
120 next block on each request.</p>
123 <p style="margin-left:6%; margin-top: 1em"><b>Decompresssion
125 The decompression layer not only handles decompression, it
126 also buffers data so that the format handlers see a much
127 nicer I/O model. The decompression API is a two stage
128 peek/consume model. A read_ahead request specifies a minimum
129 read amount; the decompression layer must provide a pointer
130 to at least that much data. If more data is immediately
131 available, it should return more: the format layer handles
132 bulk data reads by asking for a minimum of one byte and then
133 copying as much data as is available.</p>
135 <p style="margin-left:6%; margin-top: 1em">A subsequent
136 call to the <b>consume</b>() function advances the read
137 pointer. Note that data returned from a <b>read_ahead</b>()
138 call is guaranteed to remain in place until the next call to
139 <b>read_ahead</b>(). Intervening calls to <b>consume</b>()
140 should not cause the data to move.</p>
142 <p style="margin-left:6%; margin-top: 1em">Skip requests
143 must always be handled exactly. Decompression handlers that
144 cannot seek forward should not register a skip handler; the
145 API layer fills in a generic skip handler that reads and
148 <p style="margin-left:6%; margin-top: 1em">A decompression
149 handler has a specific lifecycle:</p>
151 <p>Registration/Configuration</p>
153 <p style="margin-left:17%;">When the client invokes the
154 public support function, the decompression handler invokes
155 the internal <b>__archive_read_register_compression</b>()
156 function to provide bid and initialization functions. This
157 function returns <b>NULL</b> on error or else a pointer to a
158 <b>struct decompressor_t</b>. This structure contains a
159 <i>void * config</i> slot that can be used for storing any
160 customization information.</p>
164 <p style="margin-left:17%; margin-top: 1em">The bid
165 function is invoked with a pointer and size of a block of
166 data. The decompressor can access its config data through
167 the <i>decompressor</i> element of the <b>archive_read</b>
168 object. The bid function is otherwise stateless. In
169 particular, it must not perform any I/O operations.</p>
171 <p style="margin-left:17%; margin-top: 1em">The value
172 returned by the bid function indicates its suitability for
173 handling this data stream. A bid of zero will ensure that
174 this decompressor is never invoked. Return zero if magic
175 number checks fail. Otherwise, your initial implementation
176 should return the number of bits actually checked. For
177 example, if you verify two full bytes and three bits of
178 another byte, bid 19. Note that the initial block may be
179 very short; be careful to only inspect the data you are
180 given. (The current decompressors require two bytes for
181 correct bidding.)</p>
185 <p style="margin-left:17%;">The winning bidder will have
186 its init function called. This function should initialize
187 the remaining slots of the <i>struct decompressor_t</i>
188 object pointed to by the <i>decompressor</i> element of the
189 <i>archive_read</i> object. In particular, it should
190 allocate any working data it needs in the <i>data</i> slot
191 of that structure. The init function is called with the
192 block of data that was used for tasting. At this point, the
193 decompressor is responsible for all I/O requests to the
194 client callbacks. The decompressor is free to read more data
195 as and when necessary.</p>
197 <p>Satisfy I/O requests</p>
199 <p style="margin-left:17%;">The format handler will invoke
200 the <i>read_ahead</i>, <i>consume</i>, and <i>skip</i>
201 functions as needed.</p>
205 <p style="margin-left:17%; margin-top: 1em">The finish
206 method is called only once when the archive is closed. It
207 should release anything stored in the <i>data</i> and
208 <i>config</i> slots of the <i>decompressor</i> object. It
209 should not invoke the client close callback.</p>
211 <p style="margin-left:6%; margin-top: 1em"><b>Format
213 The read formats have a similar lifecycle to the
214 decompression handlers:</p>
218 <p style="margin-left:17%;">Allocate your private data and
219 initialize your pointers.</p>
223 <p style="margin-left:17%; margin-top: 1em">Formats bid by
224 invoking the <b>read_ahead</b>() decompression method but
225 not calling the <b>consume</b>() method. This allows each
226 bidder to look ahead in the input stream. Bidders should not
227 look further ahead than necessary, as long look aheads put
228 pressure on the decompression layer to buffer lots of data.
229 Most formats only require a few hundred bytes of look ahead;
230 look aheads of a few kilobytes are reasonable. (The ISO9660
231 reader sometimes looks ahead by 48k, which should be
232 considered an upper limit.)</p>
236 <p style="margin-left:17%;">The header read is usually the
237 most complex part of any format. There are a few strategies
238 worth mentioning: For formats such as tar or cpio, reading
239 and parsing the header is straightforward since headers
240 alternate with data. For formats that store all header data
241 at the beginning of the file, the first header read request
242 may have to read all headers into memory and store that
243 data, sorted by the location of the file data. Subsequent
244 header read requests will skip forward to the beginning of
245 the file data and return the corresponding header.</p>
249 <p style="margin-left:17%;">The read data interface
250 supports sparse files; this requires that each call return a
251 block of data specifying the file offset and size. This may
252 require you to carefully track the location so that you can
253 return accurate file offsets for each read. Remember that
254 the decompressor will return as much data as it has.
255 Generally, you will want to request one byte, examine the
256 return value to see how much data is available, and possibly
257 trim that to the amount you can use. You should invoke
258 consume for each block just before you return it.</p>
262 <p style="margin-left:17%;">The skip data call should skip
263 over all file data and trailing padding. This is called
264 automatically by the API layer just before each header read.
265 It is also called in response to the client calling the
266 public <b>data_skip</b>() function.</p>
270 <p style="margin-left:17%;">On cleanup, the format should
271 release all of its allocated memory.</p>
273 <p style="margin-left:6%; margin-top: 1em"><b>API Layer</b>
277 <p style="margin-top: 1em"><b>WRITE ARCHITECTURE</b></p>
279 <p style="margin-left:6%;">The write API has a similar set
280 of four layers: an API layer, a format layer, a compression
281 layer, and an I/O layer. The registration here is much
282 simpler because only one format and one compression can be
283 registered at a time.</p>
285 <p style="margin-left:6%; margin-top: 1em"><b>I/O Layer and
286 Client Callbacks</b> <br>
287 XXX To be written XXX</p>
289 <p style="margin-left:6%; margin-top: 1em"><b>Compression
291 XXX To be written XXX</p>
293 <p style="margin-left:6%; margin-top: 1em"><b>Format
295 XXX To be written XXX</p>
297 <p style="margin-left:6%; margin-top: 1em"><b>API Layer</b>
299 XXX To be written XXX</p>
301 <p style="margin-top: 1em"><b>WRITE_DISK
304 <p style="margin-left:6%;">The write_disk API is intended
305 to look just like the write API to clients. Since it does
306 not handle multiple formats or compression, it is not
307 layered internally.</p>
309 <p style="margin-top: 1em"><b>GENERAL SERVICES</b></p>
311 <p style="margin-left:6%;">The <b>archive_read</b>,
312 <b>archive_write</b>, and <b>archive_write_disk</b> objects
313 all contain an initial <b>archive</b> object which provides
314 common support for a set of standard services. (Recall that
315 ANSI/ISO C90 guarantees that you can cast freely between a
316 pointer to a structure and a pointer to the first element of
317 that structure.) The <b>archive</b> object has a magic value
318 that indicates which API this object is associated with,
319 slots for storing error information, and function pointers
320 for virtualized API functions.</p>
322 <p style="margin-top: 1em"><b>MISCELLANEOUS NOTES</b></p>
324 <p style="margin-left:6%;">Connecting existing archiving
325 libraries into libarchive is generally quite difficult. In
326 particular, many existing libraries strongly assume that you
327 are reading from a file; they seek forwards and backwards as
328 necessary to locate various pieces of information. In
329 contrast, libarchive never seeks backwards in its input,
330 which sometimes requires very different approaches.</p>
332 <p style="margin-left:6%; margin-top: 1em">For example,
333 libarchive’s ISO9660 support operates very differently
334 from most ISO9660 readers. The libarchive support utilizes a
335 work-queue design that keeps a list of known entries sorted
336 by their location in the input. Whenever libarchive’s
337 ISO9660 implementation is asked for the next header, checks
338 this list to find the next item on the disk. Directories are
339 parsed when they are encountered and new items are added to
340 the list. This design relies heavily on the ISO9660 image
341 being optimized so that directories always occur earlier on
342 the disk than the files they describe.</p>
344 <p style="margin-left:6%; margin-top: 1em">Depending on the
345 specific format, such approaches may not be possible. The
346 ZIP format specification, for example, allows archivers to
347 store key information only at the end of the file. In
348 theory, it is possible to create ZIP archives that cannot be
349 read without seeking. Fortunately, such archives are very
350 rare, and libarchive can read most ZIP archives, though it
351 cannot always extract as much information as a dedicated ZIP
354 <p style="margin-top: 1em"><b>SEE ALSO</b></p>
356 <p style="margin-left:6%;">archive_entry(3),
357 archive_read(3), archive_write(3), archive_write_disk(3)
360 <p style="margin-top: 1em"><b>HISTORY</b></p>
362 <p style="margin-left:6%;">The <b>libarchive</b> library
363 first appeared in FreeBSD 5.3.</p>
365 <p style="margin-top: 1em"><b>AUTHORS</b></p>
367 <p style="margin-left:6%;">The <b>libarchive</b> library
368 was written by Tim Kientzle <kientzle@acm.org>.</p>
370 <p style="margin-left:6%; margin-top: 1em">BSD
371 January 26, 2011 BSD</p>