source: rtems/cpukit/zlib/examples/gzlog.c @ d42bd2f

4.104.115
Last change on this file since d42bd2f was d42bd2f, checked in by Ralf Corsepius <ralf.corsepius@…>, on Apr 25, 2008 at 11:14:33 PM

Import from zlib-1.2.4

  • Property mode set to 100644
File size: 40.5 KB
Line 
1/*
2 * gzlog.c
3 * Copyright (C) 2004, 2008 Mark Adler, all rights reserved
4 * For conditions of distribution and use, see copyright notice in gzlog.h
5 * version 2.0, 25 Apr 2008
6 */
7
8/*
9   gzlog provides a mechanism for frequently appending short strings to a gzip
10   file that is efficient both in execution time and compression ratio.  The
11   strategy is to write the short strings in an uncompressed form to the end of
12   the gzip file, only compressing when the amount of uncompressed data has
13   reached a given threshold.
14
15   gzlog also provides protection against interruptions in the process due to
16   system crashes.  The status of the operation is recorded in an extra field
17   in the gzip file, and is only updated once the gzip file is brought to a
18   valid state.  The last data to be appended or compressed is saved in an
19   auxiliary file, so that if the operation is interrupted, it can be completed
20   the next time an append operation is attempted.
21
22   gzlog maintains another auxiliary file with the last 32K of data from the
23   compressed portion, which is preloaded for the compression of the subsequent
24   data.  This minimizes the impact to the compression ratio of appending.
25 */
26
27/*
28   Operations Concept:
29
30   Files (log name "foo"):
31   foo.gz -- gzip file with the complete log
32   foo.add -- last message to append or last data to compress
33   foo.dict -- dictionary of the last 32K of data for next compression
34   foo.temp -- temporary dictionary file for compression after this one
35   foo.lock -- lock file for reading and writing the other files
36   foo.repairs -- log file for log file recovery operations (not compressed)
37
38   gzip file structure:
39   - fixed-length (no file name) header with extra field (see below)
40   - compressed data ending initially with empty stored block
41   - uncompressed data filling out originally empty stored block and
42     subsequent stored blocks as needed (16K max each)
43   - gzip trailer
44   - no junk at end (no other gzip streams)
45
46   When appending data, the information in the first three items above plus the
47   foo.add file are sufficient to recover an interrupted append operation.  The
48   extra field has the necessary information to restore the start of the last
49   stored block and determine where to append the data in the foo.add file, as
50   well as the crc and length of the gzip data before the append operation.
51
52   The foo.add file is created before the gzip file is marked for append, and
53   deleted after the gzip file is marked as complete.  So if the append
54   operation is interrupted, the data to add will still be there.  If due to
55   some external force, the foo.add file gets deleted between when the append
56   operation was interrupted and when recovery is attempted, the gzip file will
57   still be restored, but without the appended data.
58
59   When compressing data, the information in the first two items above plus the
60   foo.add file are sufficient to recover an interrupted compress operation.
61   The extra field has the necessary information to find the end of the
62   compressed data, and contains both the crc and length of just the compressed
63   data and of the complete set of data including the contents of the foo.add
64   file.
65
66   Again, the foo.add file is maintained during the compress operation in case
67   of an interruption.  If in the unlikely event the foo.add file with the data
68   to be compressed is missing due to some external force, a gzip file with
69   just the previous compressed data will be reconstructed.  In this case, all
70   of the data that was to be compressed is lost (approximately one megabyte).
71   This will not occur if all that happened was an interruption of the compress
72   operation.
73
74   The third state that is marked is the replacement of the old dictionary with
75   the new dictionary after a compress operation.  Once compression is
76   complete, the gzip file is marked as being in the replace state.  This
77   completes the gzip file, so an interrupt after being so marked does not
78   result in recompression.  Then the dictionary file is replaced, and the gzip
79   file is marked as completed.  This state prevents the possibility of
80   restarting compression with the wrong dictionary file.
81
82   All three operations are wrapped by a lock/unlock procedure.  In order to
83   gain exclusive access to the log files, first a foo.lock file must be
84   exclusively created.  When all operations are complete, the lock is
85   released by deleting the foo.lock file.  If when attempting to create the
86   lock file, it already exists and the modify time of the lock file is more
87   than five minutes old (set by the PATIENCE define below), then the old
88   lock file is considered stale and deleted, and the exclusive creation of
89   the lock file is retried.  To assure that there are no false assessments
90   of the staleness of the lock file, the operations periodically touch the
91   lock file to update the modified date.
92
93   Following is the definition of the extra field with all of the information
94   required to enable the above append and compress operations and their
95   recovery if interrupted.  Multi-byte values are stored little endian
96   (consistent with the gzip format).  File pointers are eight bytes long.
97   The crc's and lengths for the gzip trailer are four bytes long.  (Note that
98   the length at the end of a gzip file is used for error checking only, and
99   for large files is actually the length modulo 2^32.)  The stored block
100   length is two bytes long.  The gzip extra field two-byte identification is
101   "ap" for append.  It is assumed that writing the extra field to the file is
102   an "atomic" operation.  That is, either all of the extra field is written
103   to the file, or none of it is, if the operation is interrupted right at the
104   point of updating the extra field.  This is a reasonable assumption, since
105   the extra field is within the first 52 bytes of the file, which is smaller
106   than any expected block size for a mass storage device (usually 512 bytes or
107   larger).
108
109   Extra field (35 bytes):
110   - Pointer to first stored block length -- this points to the two-byte length
111     of the first stored block, which is followed by the two-byte, one's
112     complement of that length.  The stored block length is preceded by the
113     three-bit header of the stored block, which is the actual start of the
114     stored block in the deflate format.  See the bit offset field below.
115   - Pointer to the last stored block length.  This is the same as above, but
116     for the last stored block of the uncompressed data in the gzip file.
117     Initially this is the same as the first stored block length pointer.
118     When the stored block gets to 16K (see the MAX_STORE define), then a new
119     stored block as added, at which point the last stored block length pointer
120     is different from the first stored block length pointer.  When they are
121     different, the first bit of the last stored block header is eight bits, or
122     one byte back from the block length.
123   - Compressed data crc and length.  This is the crc and length of the data
124     that is in the compressed portion of the deflate stream.  These are used
125     only in the event that the foo.add file containing the data to compress is
126     lost after a compress operation is interrupted.
127   - Total data crc and length.  This is the crc and length of all of the data
128     stored in the gzip file, compressed and uncompressed.  It is used to
129     reconstruct the gzip trailer when compressing, as well as when recovering
130     interrupted operations.
131   - Final stored block length.  This is used to quickly find where to append,
132     and allows the restoration of the original final stored block state when
133     an append operation is interrupted.
134   - First stored block start as the number of bits back from the final stored
135     block first length byte.  This value is in the range of 3..10, and is
136     stored as the low three bits of the final byte of the extra field after
137     subtracting three (0..7).  This allows the last-block bit of the stored
138     block header to be updated when a new stored block is added, for the case
139     when the first stored block and the last stored block are the same.  (When
140     they are different, the numbers of bits back is known to be eight.)  This
141     also allows for new compressed data to be appended to the old compressed
142     data in the compress operation, overwriting the previous first stored
143     block, or for the compressed data to be terminated and a valid gzip file
144     reconstructed on the off chance that a compression operation was
145     interrupted and the data to compress in the foo.add file was deleted.
146   - The operation in process.  This is the next two bits in the last byte (the
147     bits under the mask 0x18).  The are interpreted as 0: nothing in process,
148     1: append in process, 2: compress in process, 3: replace in process.
149   - The top three bits of the last byte in the extra field are reserved and
150     are currently set to zero.
151
152   Main procedure:
153   - Exclusively create the foo.lock file using the O_CREAT and O_EXCL modes of
154     the system open() call.  If the modify time of an existing lock file is
155     more than PATIENCE seconds old, then the lock file is deleted and the
156     exclusive create is retried.
157   - Load the extra field from the foo.gz file, and see if an operation was in
158     progress but not completed.  If so, apply the recovery procedure below.
159   - Perform the append procedure with the provided data.
160   - If the uncompressed data in the foo.gz file is 1MB or more, apply the
161     compress procedure.
162   - Delete the foo.lock file.
163
164   Append procedure:
165   - Put what to append in the foo.add file so that the operation can be
166     restarted if this procedure is interrupted.
167   - Mark the foo.gz extra field with the append operation in progress.
168   + Restore the original last-block bit and stored block length of the last
169     stored block from the information in the extra field, in case a previous
170     append operation was interrupted.
171   - Append the provided data to the last stored block, creating new stored
172     blocks as needed and updating the stored blocks last-block bits and
173     lengths.
174   - Update the crc and length with the new data, and write the gzip trailer.
175   - Write over the extra field (with a single write operation) with the new
176     pointers, lengths, and crc's, and mark the gzip file as not in process.
177     Though there is still a foo.add file, it will be ignored since nothing
178     is in process.  If a foo.add file is leftover from a previously
179     completed operation, it is truncated when writing new data to it.
180   - Delete the foo.add file.
181
182   Compress and replace procedures:
183   - Read all of the uncompressed data in the stored blocks in foo.gz and write
184     it to foo.add.  Also write foo.temp with the last 32K of that data to
185     provide a dictionary for the next invocation of this procedure.
186   - Rewrite the extra field marking foo.gz with a compression in process.
187   * If there is no data provided to compress (due to a missing foo.add file
188     when recovering), reconstruct and truncate the foo.gz file to contain
189     only the previous compressed data and proceed to the step after the next
190     one.  Otherwise ...
191   - Compress the data with the dictionary in foo.dict, and write to the
192     foo.gz file starting at the bit immediately following the last previously
193     compressed block.  If there is no foo.dict, proceed anyway with the
194     compression at slightly reduced efficiency.  (For the foo.dict file to be
195     missing requires some external failure beyond simply the interruption of
196     a compress operation.)  During this process, the foo.lock file is
197     periodically touched to assure that that file is not considered stale by
198     another process before we're done.  The deflation is terminated with a
199     non-last empty static block (10 bits long), that is then located and
200     written over by a last-bit-set empty stored block.
201   - Append the crc and length of the data in the gzip file (previously
202     calculated during the append operations).
203   - Write over the extra field with the updated stored block offsets, bits
204     back, crc's, and lengths, and mark foo.gz as in process for a replacement
205     of the dictionary.
206   @ Delete the foo.add file.
207   - Replace foo.dict with foo.temp.
208   - Write over the extra field, marking foo.gz as complete.
209
210   Recovery procedure:
211   - If not a replace recovery, read in the foo.add file, and provide that data
212     to the appropriate recovery below.  If there is no foo.add file, provide
213     a zero data length to the recovery.  In that case, the append recovery
214     restores the foo.gz to the previous compressed + uncompressed data state.
215     For the the compress recovery, a missing foo.add file results in foo.gz
216     being restored to the previous compressed-only data state.
217   - Append recovery:
218     - Pick up append at + step above
219   - Compress recovery:
220     - Pick up compress at * step above
221   - Replace recovery:
222     - Pick up compress at @ step above
223   - Log the repair with a date stamp in foo.repairs
224 */
225
226#include <sys/types.h>
227#include <stdio.h>      /* rename, fopen, fprintf, fclose */
228#include <stdlib.h>     /* malloc, free */
229#include <string.h>     /* strlen, strrchr, strcpy, strncpy, strcmp */
230#include <fcntl.h>      /* open */
231#include <unistd.h>     /* lseek, read, write, close, unlink, sleep, */
232                        /* ftruncate, fsync */
233#include <errno.h>      /* errno */
234#include <time.h>       /* time, ctime */
235#include <sys/stat.h>   /* stat */
236#include <sys/time.h>   /* utimes */
237#include "zlib.h"       /* crc32 */
238
239#include "gzlog.h"      /* header for external access */
240
241#define local static
242typedef unsigned int uint;
243typedef unsigned long ulong;
244
245/* Macro for debugging to deterministically force recovery operations */
246#ifdef DEBUG
247    #include <setjmp.h>         /* longjmp */
248    jmp_buf gzlog_jump;         /* where to go back to */
249    int gzlog_bail = 0;         /* which point to bail at (1..8) */
250    int gzlog_count = -1;       /* number of times through to wait */
251#   define BAIL(n) do { if (n == gzlog_bail && gzlog_count-- == 0) \
252                            longjmp(gzlog_jump, gzlog_bail); } while (0)
253#else
254#   define BAIL(n)
255#endif
256
257/* how old the lock file can be in seconds before considering it stale */
258#define PATIENCE 300
259
260/* maximum stored block size in Kbytes -- must be in 1..63 */
261#define MAX_STORE 16
262
263/* number of stored Kbytes to trigger compression (must be >= 32 to allow
264   dictionary construction, and <= 204 * MAX_STORE, in order for >> 10 to
265   discard the stored block headers contribution of five bytes each) */
266#define TRIGGER 1024
267
268/* size of a deflate dictionary (this cannot be changed) */
269#define DICT 32768U
270
271/* values for the operation (2 bits) */
272#define NO_OP 0
273#define APPEND_OP 1
274#define COMPRESS_OP 2
275#define REPLACE_OP 3
276
277/* macros to extract little-endian integers from an unsigned byte buffer */
278#define PULL2(p) ((p)[0]+((uint)((p)[1])<<8))
279#define PULL4(p) (PULL2(p)+((ulong)PULL2(p+2)<<16))
280#define PULL8(p) (PULL4(p)+((off_t)PULL4(p+4)<<32))
281
282/* macros to store integers into a byte buffer in little-endian order */
283#define PUT2(p,a) do {(p)[0]=a;(p)[1]=(a)>>8;} while(0)
284#define PUT4(p,a) do {PUT2(p,a);PUT2(p+2,a>>16);} while(0)
285#define PUT8(p,a) do {PUT4(p,a);PUT4(p+4,a>>32);} while(0)
286
287/* internal structure for log information */
288#define LOGID "\106\035\172"    /* should be three non-zero characters */
289struct log {
290    char id[4];     /* contains LOGID to detect inadvertent overwrites */
291    int fd;         /* file descriptor for .gz file, opened read/write */
292    char *path;     /* allocated path, e.g. "/var/log/foo" or "foo" */
293    char *end;      /* end of path, for appending suffices such as ".gz" */
294    off_t first;    /* offset of first stored block first length byte */
295    int back;       /* location of first block id in bits back from first */
296    uint stored;    /* bytes currently in last stored block */
297    off_t last;     /* offset of last stored block first length byte */
298    ulong ccrc;     /* crc of compressed data */
299    ulong clen;     /* length (modulo 2^32) of compressed data */
300    ulong tcrc;     /* crc of total data */
301    ulong tlen;     /* length (modulo 2^32) of total data */
302    time_t lock;    /* last modify time of our lock file */
303};
304
305/* gzip header for gzlog */
306local unsigned char log_gzhead[] = {
307    0x1f, 0x8b,                 /* magic gzip id */
308    8,                          /* compression method is deflate */
309    4,                          /* there is an extra field (no file name) */
310    0, 0, 0, 0,                 /* no modification time provided */
311    0, 0xff,                    /* no extra flags, no OS specified */
312    39, 0, 'a', 'p', 35, 0      /* extra field with "ap" subfield */
313                                /* 35 is EXTRA, 39 is EXTRA + 4 */
314};
315
316#define HEAD sizeof(log_gzhead)     /* should be 16 */
317
318/* initial gzip extra field content (52 == HEAD + EXTRA + 1) */
319local unsigned char log_gzext[] = {
320    52, 0, 0, 0, 0, 0, 0, 0,    /* offset of first stored block length */
321    52, 0, 0, 0, 0, 0, 0, 0,    /* offset of last stored block length */
322    0, 0, 0, 0, 0, 0, 0, 0,     /* compressed data crc and length */
323    0, 0, 0, 0, 0, 0, 0, 0,     /* total data crc and length */
324    0, 0,                       /* final stored block data length */
325    5                           /* op is NO_OP, last bit 8 bits back */
326};
327
328#define EXTRA sizeof(log_gzext)     /* should be 35 */
329
330/* initial gzip data and trailer */
331local unsigned char log_gzbody[] = {
332    1, 0, 0, 0xff, 0xff,        /* empty stored block (last) */
333    0, 0, 0, 0,                 /* crc */
334    0, 0, 0, 0                  /* uncompressed length */
335};
336
337#define BODY sizeof(log_gzbody)
338
339/* Exclusively create foo.lock in order to negotiate exclusive access to the
340   foo.* files.  If the modify time of an existing lock file is greater than
341   PATIENCE seconds in the past, then consider the lock file to have been
342   abandoned, delete it, and try the exclusive create again.  Save the lock
343   file modify time for verification of ownership.  Return 0 on success, or -1
344   on failure, usually due to an access restriction or invalid path.  Note that
345   if stat() or unlink() fails, it may be due to another process noticing the
346   abandoned lock file a smidge sooner and deleting it, so those are not
347   flagged as an error. */
348local int log_lock(struct log *log)
349{
350    int fd;
351    struct stat st;
352
353    strcpy(log->end, ".lock");
354    while ((fd = open(log->path, O_CREAT | O_EXCL, 0644)) < 0) {
355        if (errno != EEXIST)
356            return -1;
357        if (stat(log->path, &st) == 0 && time(NULL) - st.st_mtime > PATIENCE) {
358            unlink(log->path);
359            continue;
360        }
361        sleep(2);       /* relinquish the CPU for two seconds while waiting */
362    }
363    close(fd);
364    if (stat(log->path, &st) == 0)
365        log->lock = st.st_mtime;
366    return 0;
367}
368
369/* Update the modify time of the lock file to now, in order to prevent another
370   task from thinking that the lock is stale.  Save the lock file modify time
371   for verification of ownership. */
372local void log_touch(struct log *log)
373{
374    struct stat st;
375
376    strcpy(log->end, ".lock");
377    utimes(log->path, NULL);
378    if (stat(log->path, &st) == 0)
379        log->lock = st.st_mtime;
380}
381
382/* Check the log file modify time against what is expected.  Return true if
383   this is not our lock.  If it is our lock, touch it to keep it. */
384local int log_check(struct log *log)
385{
386    struct stat st;
387
388    strcpy(log->end, ".lock");
389    if (stat(log->path, &st) || st.st_mtime != log->lock)
390        return 1;
391    log_touch(log);
392    return 0;
393}
394
395/* Unlock a previously acquired lock, but only if it's ours. */
396local void log_unlock(struct log *log)
397{
398    if (log_check(log))
399        return;
400    strcpy(log->end, ".lock");
401    unlink(log->path);
402    log->lock = 0;
403}
404
405/* Check the gzip header and read in the extra field, filling in the values in
406   the log structure.  Return op on success or -1 if the gzip header was not as
407   expected.  op is the current operation in progress last written to the extra
408   field.  This assumes that the gzip file has already been opened, with the
409   file descriptor log->fd. */
410local int log_head(struct log *log)
411{
412    int op;
413    unsigned char buf[HEAD + EXTRA];
414
415    if (lseek(log->fd, 0, SEEK_SET) < 0 ||
416        read(log->fd, buf, HEAD + EXTRA) != HEAD + EXTRA ||
417        memcmp(buf, log_gzhead, HEAD)) {
418        return -1;
419    }
420    log->first = PULL8(buf + HEAD);
421    log->last = PULL8(buf + HEAD + 8);
422    log->ccrc = PULL4(buf + HEAD + 16);
423    log->clen = PULL4(buf + HEAD + 20);
424    log->tcrc = PULL4(buf + HEAD + 24);
425    log->tlen = PULL4(buf + HEAD + 28);
426    log->stored = PULL2(buf + HEAD + 32);
427    log->back = 3 + (buf[HEAD + 34] & 7);
428    op = (buf[HEAD + 34] >> 3) & 3;
429    return op;
430}
431
432/* Write over the extra field contents, marking the operation as op.  Use fsync
433   to assure that the device is written to, and in the requested order.  This
434   operation, and only this operation, is assumed to be atomic in order to
435   assure that the log is recoverable in the event of an interruption at any
436   point in the process.  Return -1 if the write to foo.gz failed. */
437local int log_mark(struct log *log, int op)
438{
439    int ret;
440    unsigned char ext[EXTRA];
441
442    PUT8(ext, log->first);
443    PUT8(ext + 8, log->last);
444    PUT4(ext + 16, log->ccrc);
445    PUT4(ext + 20, log->clen);
446    PUT4(ext + 24, log->tcrc);
447    PUT4(ext + 28, log->tlen);
448    PUT2(ext + 32, log->stored);
449    ext[34] = log->back - 3 + (op << 3);
450    fsync(log->fd);
451    ret = lseek(log->fd, HEAD, SEEK_SET) < 0 ||
452          write(log->fd, ext, EXTRA) != EXTRA ? -1 : 0;
453    fsync(log->fd);
454    return ret;
455}
456
457/* Rewrite the last block header bits and subsequent zero bits to get to a byte
458   boundary, setting the last block bit if last is true, and then write the
459   remainder of the stored block header (length and one's complement).  Leave
460   the file pointer after the end of the last stored block data.  Return -1 if
461   there is a read or write failure on the foo.gz file */
462local int log_last(struct log *log, int last)
463{
464    int back, len, mask;
465    unsigned char buf[6];
466
467    /* determine the locations of the bytes and bits to modify */
468    back = log->last == log->first ? log->back : 8;
469    len = back > 8 ? 2 : 1;                 /* bytes back from log->last */
470    mask = 0x80 >> ((back - 1) & 7);        /* mask for block last-bit */
471
472    /* get the byte to modify (one or two back) into buf[0] -- don't need to
473       read the byte if the last-bit is eight bits back, since in that case
474       the entire byte will be modified */
475    buf[0] = 0;
476    if (back != 8 && (lseek(log->fd, log->last - len, SEEK_SET) < 0 ||
477                      read(log->fd, buf, 1) != 1))
478        return -1;
479
480    /* change the last-bit of the last stored block as requested -- note
481       that all bits above the last-bit are set to zero, per the type bits
482       of a stored block being 00 and per the convention that the bits to
483       bring the stream to a byte boundary are also zeros */
484    buf[1] = 0;
485    buf[2 - len] = (*buf & (mask - 1)) + (last ? mask : 0);
486
487    /* write the modified stored block header and lengths, move the file
488       pointer to after the last stored block data */
489    PUT2(buf + 2, log->stored);
490    PUT2(buf + 4, log->stored ^ 0xffff);
491    return lseek(log->fd, log->last - len, SEEK_SET) < 0 ||
492           write(log->fd, buf + 2 - len, len + 4) != len + 4 ||
493           lseek(log->fd, log->stored, SEEK_CUR) < 0 ? -1 : 0;
494}
495
496/* Append len bytes from data to the locked and open log file.  len may be zero
497   if recovering and no .add file was found.  In that case, the previous state
498   of the foo.gz file is restored.  The data is appended uncompressed in
499   deflate stored blocks.  Return -1 if there was an error reading or writing
500   the foo.gz file. */
501local int log_append(struct log *log, unsigned char *data, size_t len)
502{
503    uint put;
504    off_t end;
505    unsigned char buf[8];
506
507    /* set the last block last-bit and length, in case recovering an
508       interrupted append, then position the file pointer to append to the
509       block */
510    if (log_last(log, 1))
511        return -1;
512
513    /* append, adding stored blocks and updating the offset of the last stored
514       block as needed, and update the total crc and length */
515    while (len) {
516        /* append as much as we can to the last block */
517        put = (MAX_STORE << 10) - log->stored;
518        if (put > len)
519            put = (uint)len;
520        if (put) {
521            if (write(log->fd, data, put) != put)
522                return -1;
523            BAIL(1);
524            log->tcrc = crc32(log->tcrc, data, put);
525            log->tlen += put;
526            log->stored += put;
527            data += put;
528            len -= put;
529        }
530
531        /* if we need to, add a new empty stored block */
532        if (len) {
533            /* mark current block as not last */
534            if (log_last(log, 0))
535                return -1;
536
537            /* point to new, empty stored block */
538            log->last += 4 + log->stored + 1;
539            log->stored = 0;
540        }
541
542        /* mark last block as last, update its length */
543        if (log_last(log, 1))
544            return -1;
545        BAIL(2);
546    }
547
548    /* write the new crc and length trailer, and truncate just in case (could
549       be recovering from partial append with a missing foo.add file) */
550    PUT4(buf, log->tcrc);
551    PUT4(buf + 4, log->tlen);
552    if (write(log->fd, buf, 8) != 8 ||
553        (end = lseek(log->fd, 0, SEEK_CUR)) < 0 || ftruncate(log->fd, end))
554        return -1;
555
556    /* write the extra field, marking the log file as done, delete .add file */
557    if (log_mark(log, NO_OP))
558        return -1;
559    strcpy(log->end, ".add");
560    unlink(log->path);          /* ignore error, since may not exist */
561    return 0;
562}
563
564/* Replace the foo.dict file with the foo.temp file.  Also delete the foo.add
565   file, since the compress operation may have been interrupted before that was
566   done.  Returns 1 if memory could not be allocated, or -1 if reading or
567   writing foo.gz fails, or if the rename fails for some reason other than
568   foo.temp not existing.  foo.temp not existing is a permitted error, since
569   the replace operation may have been interrupted after the rename is done,
570   but before foo.gz is marked as complete. */
571local int log_replace(struct log *log)
572{
573    int ret;
574    char *dest;
575
576    /* delete foo.add file */
577    strcpy(log->end, ".add");
578    unlink(log->path);         /* ignore error, since may not exist */
579    BAIL(3);
580
581    /* rename foo.name to foo.dict, replacing foo.dict if it exists */
582    strcpy(log->end, ".dict");
583    dest = malloc(strlen(log->path) + 1);
584    if (dest == NULL)
585        return -2;
586    strcpy(dest, log->path);
587    strcpy(log->end, ".temp");
588    ret = rename(log->path, dest);
589    free(dest);
590    if (ret && errno != ENOENT)
591        return -1;
592    BAIL(4);
593
594    /* mark the foo.gz file as done */
595    return log_mark(log, NO_OP);
596}
597
598/* Compress the len bytes at data and append the compressed data to the
599   foo.gz deflate data immediately after the previous compressed data.  This
600   overwrites the previous uncompressed data, which was stored in foo.add
601   and is the data provided in data[0..len-1].  If this operation is
602   interrupted, it picks up at the start of this routine, with the foo.add
603   file read in again.  If there is no data to compress (len == 0), then we
604   simply terminate the foo.gz file after the previously compressed data,
605   appending a final empty stored block and the gzip trailer.  Return -1 if
606   reading or writing the log.gz file failed, or -2 if there was a memory
607   allocation failure. */
608local int log_compress(struct log *log, unsigned char *data, size_t len)
609{
610    int fd;
611    uint got, max;
612    ssize_t dict;
613    off_t end;
614    z_stream strm;
615    unsigned char buf[DICT];
616
617    /* compress and append compressed data */
618    if (len) {
619        /* set up for deflate, allocating memory */
620        strm.zalloc = Z_NULL;
621        strm.zfree = Z_NULL;
622        strm.opaque = Z_NULL;
623        if (deflateInit2(&strm, Z_DEFAULT_COMPRESSION, Z_DEFLATED, -15, 8,
624                         Z_DEFAULT_STRATEGY) != Z_OK)
625            return -2;
626
627        /* read in dictionary (last 32K of data that was compressed) */
628        strcpy(log->end, ".dict");
629        fd = open(log->path, O_RDONLY, 0);
630        if (fd >= 0) {
631            dict = read(fd, buf, DICT);
632            close(fd);
633            if (dict < 0) {
634                deflateEnd(&strm);
635                return -1;
636            }
637            if (dict)
638                deflateSetDictionary(&strm, buf, (uint)dict);
639        }
640        log_touch(log);
641
642        /* prime deflate with last bits of previous block, position write
643           pointer to write those bits and overwrite what follows */
644        if (lseek(log->fd, log->first - (log->back > 8 ? 2 : 1),
645                SEEK_SET) < 0 ||
646            read(log->fd, buf, 1) != 1 || lseek(log->fd, -1, SEEK_CUR) < 0) {
647            deflateEnd(&strm);
648            return -1;
649        }
650        deflatePrime(&strm, (8 - log->back) & 7, *buf);
651
652        /* compress, finishing with a partial non-last empty static block */
653        strm.next_in = data;
654        max = (((uint)0 - 1) >> 1) + 1; /* in case int smaller than size_t */
655        do {
656            strm.avail_in = len > max ? max : (uint)len;
657            len -= strm.avail_in;
658            do {
659                strm.avail_out = DICT;
660                strm.next_out = buf;
661                deflate(&strm, len ? Z_NO_FLUSH : Z_PARTIAL_FLUSH);
662                got = DICT - strm.avail_out;
663                if (got && write(log->fd, buf, got) != got) {
664                    deflateEnd(&strm);
665                    return -1;
666                }
667                log_touch(log);
668            } while (strm.avail_out == 0);
669        } while (len);
670        deflateEnd(&strm);
671        BAIL(5);
672
673        /* find start of empty static block -- scanning backwards the first one
674           bit is the second bit of the block, if the last byte is zero, then
675           we know the byte before that has a one in the top bit, since an
676           empty static block is ten bits long */
677        if ((log->first = lseek(log->fd, -1, SEEK_CUR)) < 0 ||
678            read(log->fd, buf, 1) != 1)
679            return -1;
680        log->first++;
681        if (*buf) {
682            log->back = 1;
683            while ((*buf & ((uint)1 << (8 - log->back++))) == 0)
684                ;       /* guaranteed to terminate, since *buf != 0 */
685        }
686        else
687            log->back = 10;
688
689        /* update compressed crc and length */
690        log->ccrc = log->tcrc;
691        log->clen = log->tlen;
692    }
693    else {
694        /* no data to compress -- fix up existing gzip stream */
695        log->tcrc = log->ccrc;
696        log->tlen = log->clen;
697    }
698
699    /* complete and truncate gzip stream */
700    log->last = log->first;
701    log->stored = 0;
702    PUT4(buf, log->tcrc);
703    PUT4(buf + 4, log->tlen);
704    if (log_last(log, 1) || write(log->fd, buf, 8) != 8 ||
705        (end = lseek(log->fd, 0, SEEK_CUR)) < 0 || ftruncate(log->fd, end))
706        return -1;
707    BAIL(6);
708
709    /* mark as being in the replace operation */
710    if (log_mark(log, REPLACE_OP))
711        return -1;
712
713    /* execute the replace operation and mark the file as done */
714    return log_replace(log);
715}
716
717/* log a repair record to the .repairs file */
718local void log_log(struct log *log, int op, char *record)
719{
720    time_t now;
721    FILE *rec;
722
723    now = time(NULL);
724    strcpy(log->end, ".repairs");
725    rec = fopen(log->path, "a");
726    if (rec == NULL)
727        return;
728    fprintf(rec, "%.24s %s recovery: %s\n", ctime(&now), op == APPEND_OP ?
729            "append" : (op == COMPRESS_OP ? "compress" : "replace"), record);
730    fclose(rec);
731    return;
732}
733
734/* Recover the interrupted operation op.  First read foo.add for recovering an
735   append or compress operation.  Return -1 if there was an error reading or
736   writing foo.gz or reading an existing foo.add, or -2 if there was a memory
737   allocation failure. */
738local int log_recover(struct log *log, int op)
739{
740    int fd, ret = 0;
741    unsigned char *data = NULL;
742    size_t len = 0;
743    struct stat st;
744
745    /* log recovery */
746    log_log(log, op, "start");
747
748    /* load foo.add file if expected and present */
749    if (op == APPEND_OP || op == COMPRESS_OP) {
750        strcpy(log->end, ".add");
751        if (stat(log->path, &st) == 0 && st.st_size) {
752            len = (size_t)(st.st_size);
753            if (len != st.st_size || (data = malloc(st.st_size)) == NULL) {
754                log_log(log, op, "allocation failure");
755                return -2;
756            }
757            if ((fd = open(log->path, O_RDONLY, 0)) < 0) {
758                log_log(log, op, ".add file read failure");
759                return -1;
760            }
761            ret = read(fd, data, len) != len;
762            close(fd);
763            if (ret) {
764                log_log(log, op, ".add file read failure");
765                return -1;
766            }
767            log_log(log, op, "loaded .add file");
768        }
769        else
770            log_log(log, op, "missing .add file!");
771    }
772
773    /* recover the interrupted operation */
774    switch (op) {
775    case APPEND_OP:
776        ret = log_append(log, data, len);
777        break;
778    case COMPRESS_OP:
779        ret = log_compress(log, data, len);
780        break;
781    case REPLACE_OP:
782        ret = log_replace(log);
783    }
784
785    /* log status */
786    log_log(log, op, ret ? "failure" : "complete");
787
788    /* clean up */
789    if (data != NULL)
790        free(data);
791    return ret;
792}
793
794/* Close the foo.gz file (if open) and release the lock. */
795local void log_close(struct log *log)
796{
797    if (log->fd >= 0)
798        close(log->fd);
799    log->fd = -1;
800    log_unlock(log);
801}
802
803/* Open foo.gz, verify the header, and load the extra field contents, after
804   first creating the foo.lock file to gain exclusive access to the foo.*
805   files.  If foo.gz does not exist or is empty, then write the initial header,
806   extra, and body content of an empty foo.gz log file.  If there is an error
807   creating the lock file due to access restrictions, or an error reading or
808   writing the foo.gz file, or if the foo.gz file is not a proper log file for
809   this object (e.g. not a gzip file or does not contain the expected extra
810   field), then return true.  If there is an error, the lock is released.
811   Otherwise, the lock is left in place. */
812local int log_open(struct log *log)
813{
814    int op;
815
816    /* release open file resource if left over -- can occur if lock lost
817       between gzlog_open() and gzlog_write() */
818    if (log->fd >= 0)
819        close(log->fd);
820    log->fd = -1;
821
822    /* negotiate exclusive access */
823    if (log_lock(log) < 0)
824        return -1;
825
826    /* open the log file, foo.gz */
827    strcpy(log->end, ".gz");
828    log->fd = open(log->path, O_RDWR | O_CREAT, 0644);
829    if (log->fd < 0) {
830        log_close(log);
831        return -1;
832    }
833
834    /* if new, initialize foo.gz with an empty log, delete old dictionary */
835    if (lseek(log->fd, 0, SEEK_END) == 0) {
836        if (write(log->fd, log_gzhead, HEAD) != HEAD ||
837            write(log->fd, log_gzext, EXTRA) != EXTRA ||
838            write(log->fd, log_gzbody, BODY) != BODY) {
839            log_close(log);
840            return -1;
841        }
842        strcpy(log->end, ".dict");
843        unlink(log->path);
844    }
845
846    /* verify log file and load extra field information */
847    if ((op = log_head(log)) < 0) {
848        log_close(log);
849        return -1;
850    }
851
852    /* check for interrupted process and if so, recover */
853    if (op != NO_OP && log_recover(log, op)) {
854        log_close(log);
855        return -1;
856    }
857
858    /* touch the lock file to prevent another process from grabbing it */
859    log_touch(log);
860    return 0;
861}
862
863/* See gzlog.h for the description of the external methods below */
864gzlog *gzlog_open(char *path)
865{
866    size_t n;
867    struct log *log;
868
869    /* check arguments */
870    if (path == NULL || *path == 0)
871        return NULL;
872
873    /* allocate and initialize log structure */
874    log = malloc(sizeof(struct log));
875    if (log == NULL)
876        return NULL;
877    strcpy(log->id, LOGID);
878    log->fd = -1;
879
880    /* save path and end of path for name construction */
881    n = strlen(path);
882    log->path = malloc(n + 9);              /* allow for ".repairs" */
883    if (log->path == NULL) {
884        free(log);
885        return NULL;
886    }
887    strcpy(log->path, path);
888    log->end = log->path + n;
889
890    /* gain exclusive access and verify log file -- may perform a
891       recovery operation if needed */
892    if (log_open(log)) {
893        free(log->path);
894        free(log);
895        return NULL;
896    }
897
898    /* return pointer to log structure */
899    return log;
900}
901
902/* gzlog_compress() return values:
903    0: all good
904   -1: file i/o error (usually access issue)
905   -2: memory allocation failure
906   -3: invalid log pointer argument */
907int gzlog_compress(gzlog *logd)
908{
909    int fd, ret;
910    uint block;
911    size_t len, next;
912    unsigned char *data, buf[5];
913    struct log *log = logd;
914
915    /* check arguments */
916    if (log == NULL || strcmp(log->id, LOGID) || len < 0)
917        return -3;
918
919    /* see if we lost the lock -- if so get it again and reload the extra
920       field information (it probably changed), recover last operation if
921       necessary */
922    if (log_check(log) && log_open(log))
923        return -1;
924
925    /* create space for uncompressed data */
926    len = ((size_t)(log->last - log->first) & ~(((size_t)1 << 10) - 1)) +
927          log->stored;
928    if ((data = malloc(len)) == NULL)
929        return -2;
930
931    /* do statement here is just a cheap trick for error handling */
932    do {
933        /* read in the uncompressed data */
934        if (lseek(log->fd, log->first - 1, SEEK_SET) < 0)
935            break;
936        next = 0;
937        while (next < len) {
938            if (read(log->fd, buf, 5) != 5)
939                break;
940            block = PULL2(buf + 1);
941            if (next + block > len ||
942                read(log->fd, (char *)data + next, block) != block)
943                break;
944            next += block;
945        }
946        if (lseek(log->fd, 0, SEEK_CUR) != log->last + 4 + log->stored)
947            break;
948        log_touch(log);
949
950        /* write the uncompressed data to the .add file */
951        strcpy(log->end, ".add");
952        fd = open(log->path, O_WRONLY | O_CREAT | O_TRUNC, 0644);
953        if (fd < 0)
954            break;
955        ret = write(fd, data, len) != len;
956        if (ret | close(fd))
957            break;
958        log_touch(log);
959
960        /* write the dictionary for the next compress to the .temp file */
961        strcpy(log->end, ".temp");
962        fd = open(log->path, O_WRONLY | O_CREAT | O_TRUNC, 0644);
963        if (fd < 0)
964            break;
965        next = DICT > len ? len : DICT;
966        ret = write(fd, (char *)data + len - next, next) != next;
967        if (ret | close(fd))
968            break;
969        log_touch(log);
970
971        /* roll back to compressed data, mark the compress in progress */
972        log->last = log->first;
973        log->stored = 0;
974        if (log_mark(log, COMPRESS_OP))
975            break;
976        BAIL(7);
977
978        /* compress and append the data (clears mark) */
979        ret = log_compress(log, data, len);
980        free(data);
981        return ret;
982    } while (0);
983
984    /* broke out of do above on i/o error */
985    free(data);
986    return -1;
987}
988
989/* gzlog_write() return values:
990    0: all good
991   -1: file i/o error (usually access issue)
992   -2: memory allocation failure
993   -3: invalid log pointer argument */
994int gzlog_write(gzlog *logd, void *data, size_t len)
995{
996    int fd, ret;
997    struct log *log = logd;
998
999    /* check arguments */
1000    if (log == NULL || strcmp(log->id, LOGID) || len < 0)
1001        return -3;
1002    if (data == NULL || len == 0)
1003        return 0;
1004
1005    /* see if we lost the lock -- if so get it again and reload the extra
1006       field information (it probably changed), recover last operation if
1007       necessary */
1008    if (log_check(log) && log_open(log))
1009        return -1;
1010
1011    /* create and write .add file */
1012    strcpy(log->end, ".add");
1013    fd = open(log->path, O_WRONLY | O_CREAT | O_TRUNC, 0644);
1014    if (fd < 0)
1015        return -1;
1016    ret = write(fd, data, len) != len;
1017    if (ret | close(fd))
1018        return -1;
1019    log_touch(log);
1020
1021    /* mark log file with append in progress */
1022    if (log_mark(log, APPEND_OP))
1023        return -1;
1024    BAIL(8);
1025
1026    /* append data (clears mark) */
1027    if (log_append(log, data, len))
1028        return -1;
1029
1030    /* check to see if it's time to compress -- if not, then done */
1031    if (((log->last - log->first) >> 10) + (log->stored >> 10) < TRIGGER)
1032        return 0;
1033
1034    /* time to compress */
1035    return gzlog_compress(log);
1036}
1037
1038/* gzlog_close() return values:
1039    0: ok
1040   -3: invalid log pointer argument */
1041int gzlog_close(gzlog *logd)
1042{
1043    struct log *log = logd;
1044
1045    /* check arguments */
1046    if (log == NULL || strcmp(log->id, LOGID))
1047        return -3;
1048
1049    /* close the log file and release the lock */
1050    log_close(log);
1051
1052    /* free structure and return */
1053    if (log->path != NULL)
1054        free(log->path);
1055    strcpy(log->id, "bad");
1056    free(log);
1057    return 0;
1058}
Note: See TracBrowser for help on using the repository browser.