2 Copyright (C) 1998, 1999, 2000, 2001, 2002, 2003, 2004, 2005, 2006,
3 2007, 2008, 2009, 2010 Free Software Foundation, Inc.
5 This file is part of GNU Wget.
7 GNU Wget is free software; you can redistribute it and/or modify
8 it under the terms of the GNU General Public License as published by
9 the Free Software Foundation; either version 3 of the License, or
10 (at your option) any later version.
12 GNU Wget is distributed in the hope that it will be useful,
13 but WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 GNU General Public License for more details.
17 You should have received a copy of the GNU General Public License
18 along with Wget. If not, see <http://www.gnu.org/licenses/>.
20 Additional permission under GNU GPL version 3 section 7
22 If you modify this program, or any covered work, by linking or
23 combining it with the OpenSSL project's OpenSSL library (or a
24 modified version of that library), containing parts covered by the
25 terms of the OpenSSL or SSLeay licenses, the Free Software Foundation
26 grants you additional permission to convey the resulting work.
27 Corresponding Source for a non-source form of such a combination
28 shall include the source code for the parts of OpenSSL used as well
29 as that of the covered work. */
45 VMS log files are often VFC record format, not stream, so fputs() can
46 produce multiple records, even when there's no newline terminator in
47 the buffer. The result is unsightly output with spurious newlines.
48 Using fprintf() instead of fputs(), along with inhibiting some
49 fflush() activity below, seems to solve the problem.
52 # define FPUTS( s, f) fprintf( (f), "%s", (s))
54 # define FPUTS( s, f) fputs( (s), (f))
55 #endif /* def __VMS [else] */
57 /* This file implements support for "logging". Logging means printing
58 output, plus several additional features:
60 - Cataloguing output by importance. You can specify that a log
61 message is "verbose" or "debug", and it will not be printed unless
62 in verbose or debug mode, respectively.
64 - Redirecting the log to the file. When Wget's output goes to the
65 terminal, and Wget receives SIGHUP, all further output is
66 redirected to a log file. When this is the case, Wget can also
67 print the last several lines of "context" to the log file so that
68 it does not begin in the middle of a line. For this to work, the
69 logging code stores the last several lines of context. Callers may
70 request for certain output not to be stored.
72 - Inhibiting output. When Wget receives SIGHUP, but redirecting
73 the output fails, logging is inhibited. */
76 /* The file descriptor used for logging. This is NULL before log_init
77 is called; logging functions log to stderr then. log_init sets it
78 either to stderr or to a file pointer obtained from fopen(). If
79 logging is inhibited, logfp is set back to NULL. */
82 /* If true, it means logging is inhibited, i.e. nothing is printed or
84 static bool inhibit_logging;
86 /* Whether the last output lines are stored for use as context. */
87 static bool save_context_p;
89 /* Whether the log is flushed after each command. */
90 static bool flush_log_p = true;
92 /* Whether any output has been received while flush_log_p was 0. */
93 static bool needs_flushing;
95 /* In the event of a hang-up, and if its output was on a TTY, Wget
96 redirects its output to `wget-log'.
98 For the convenience of reading this newly-created log, we store the
99 last several lines ("screenful", hence the choice of 24) of Wget
100 output, and dump them as context when the time comes. */
101 #define SAVED_LOG_LINES 24
103 /* log_lines is a circular buffer that stores SAVED_LOG_LINES lines of
104 output. log_line_current always points to the position in the
105 buffer that will be written to next. When log_line_current reaches
106 SAVED_LOG_LINES, it is reset to zero.
108 The problem here is that we'd have to either (re)allocate and free
109 strings all the time, or limit the lines to an arbitrary number of
110 characters. Instead of settling for either of these, we do both:
111 if the line is smaller than a certain "usual" line length (128
112 chars by default), a preallocated memory is used. The rare lines
113 that are longer than 128 characters are malloc'ed and freed
114 separately. This gives good performance with minimum memory
115 consumption and fragmentation. */
117 #define STATIC_LENGTH 128
119 static struct log_ln {
120 char static_line[STATIC_LENGTH + 1]; /* statically allocated
122 char *malloced_line; /* malloc'ed line, for lines of output
123 larger than 80 characters. */
124 char *content; /* this points either to malloced_line
125 or to the appropriate static_line.
126 If this is NULL, it means the line
127 has not yet been used. */
128 } log_lines[SAVED_LOG_LINES];
130 /* The current position in the ring. */
131 static int log_line_current = -1;
133 /* Whether the most recently written line was "trailing", i.e. did not
134 finish with \n. This is an important piece of information because
135 the code is always careful to append data to trailing lines, rather
136 than create new ones. */
137 static bool trailing_line;
139 static void check_redirect_output (void);
141 #define ROT_ADVANCE(num) do { \
142 if (++num >= SAVED_LOG_LINES) \
146 /* Free the log line index with NUM. This calls free on
147 ln->malloced_line if it's non-NULL, and it also resets
148 ln->malloced_line and ln->content to NULL. */
151 free_log_line (int num)
153 struct log_ln *ln = log_lines + num;
154 if (ln->malloced_line)
156 xfree (ln->malloced_line);
157 ln->malloced_line = NULL;
162 /* Append bytes in the range [start, end) to one line in the log. The
163 region is not supposed to contain newlines, except for the last
164 character (at end[-1]). */
167 saved_append_1 (const char *start, const char *end)
169 int len = end - start;
173 /* First, check whether we need to append to an existing line or to
177 /* Create a new line. */
180 if (log_line_current == -1)
181 log_line_current = 0;
183 free_log_line (log_line_current);
184 ln = log_lines + log_line_current;
185 if (len > STATIC_LENGTH)
187 ln->malloced_line = strdupdelim (start, end);
188 ln->content = ln->malloced_line;
192 memcpy (ln->static_line, start, len);
193 ln->static_line[len] = '\0';
194 ln->content = ln->static_line;
199 /* Append to the last line. If the line is malloc'ed, we just
200 call realloc and append the new string. If the line is
201 static, we have to check whether appending the new string
202 would make it exceed STATIC_LENGTH characters, and if so,
203 convert it to malloc(). */
204 struct log_ln *ln = log_lines + log_line_current;
205 if (ln->malloced_line)
207 /* Resize malloc'ed line and append. */
208 int old_len = strlen (ln->malloced_line);
209 ln->malloced_line = xrealloc (ln->malloced_line, old_len + len + 1);
210 memcpy (ln->malloced_line + old_len, start, len);
211 ln->malloced_line[old_len + len] = '\0';
212 /* might have changed due to realloc */
213 ln->content = ln->malloced_line;
217 int old_len = strlen (ln->static_line);
218 if (old_len + len > STATIC_LENGTH)
220 /* Allocate memory and concatenate the old and the new
222 ln->malloced_line = xmalloc (old_len + len + 1);
223 memcpy (ln->malloced_line, ln->static_line,
225 memcpy (ln->malloced_line + old_len, start, len);
226 ln->malloced_line[old_len + len] = '\0';
227 ln->content = ln->malloced_line;
231 /* Just append to the old, statically allocated
233 memcpy (ln->static_line + old_len, start, len);
234 ln->static_line[old_len + len] = '\0';
235 ln->content = ln->static_line;
239 trailing_line = !(end[-1] == '\n');
241 ROT_ADVANCE (log_line_current);
244 /* Log the contents of S, as explained above. If S consists of
245 multiple lines, they are logged separately. If S does not end with
246 a newline, it will form a "trailing" line, to which things will get
247 appended the next time this function is called. */
250 saved_append (const char *s)
254 const char *end = strchr (s, '\n');
256 end = s + strlen (s);
259 saved_append_1 (s, end);
264 /* Check X against opt.verbose and opt.quiet. The semantics is as
267 * LOG_ALWAYS - print the message unconditionally;
269 * LOG_NOTQUIET - print the message if opt.quiet is non-zero;
271 * LOG_NONVERBOSE - print the message if opt.verbose is zero;
273 * LOG_VERBOSE - print the message if opt.verbose is non-zero. */
274 #define CHECK_VERBOSE(x) \
283 case LOG_NONVERBOSE: \
284 if (opt.verbose || opt.quiet) \
292 /* Returns the file descriptor for logging. This is LOGFP, except if
293 called before log_init, in which case it returns stderr. This is
294 useful in case someone calls a logging function before log_init.
296 If logging is inhibited, return NULL. */
308 /* Log a literal string S. The string is logged as-is, without a
312 logputs (enum log_options o, const char *s)
316 check_redirect_output ();
317 if ((fp = get_log_fp ()) == NULL)
327 needs_flushing = true;
330 struct logvprintf_state {
336 /* Print a message to the log. A copy of message will be saved to
337 saved_log, for later reusal by log_dump_context().
339 Normally we'd want this function to loop around vsnprintf until
340 sufficient room is allocated, as the Linux man page recommends.
341 However each call to vsnprintf() must be preceded by va_start and
342 followed by va_end. Since calling va_start/va_end is possible only
343 in the function that contains the `...' declaration, we cannot call
344 vsnprintf more than once. Therefore this function saves its state
345 to logvprintf_state and signals the parent to call it again.
347 (An alternative approach would be to use va_copy, but that's not
351 log_vprintf_internal (struct logvprintf_state *state, const char *fmt,
355 char *write_ptr = smallmsg;
356 int available_size = sizeof (smallmsg);
358 FILE *fp = get_log_fp ();
362 /* In the simple case just call vfprintf(), to avoid needless
363 allocation and games with vsnprintf(). */
364 vfprintf (fp, fmt, args);
368 if (state->allocated != 0)
370 write_ptr = state->bigmsg;
371 available_size = state->allocated;
374 /* The GNU coding standards advise not to rely on the return value
375 of sprintf(). However, vsnprintf() is a relatively new function
376 missing from legacy systems. Therefore I consider it safe to
377 assume that its return value is meaningful. On the systems where
378 vsnprintf() is not available, we use the implementation from
379 snprintf.c which does return the correct value. */
380 numwritten = vsnprintf (write_ptr, available_size, fmt, args);
382 /* vsnprintf() will not step over the limit given by available_size.
383 If it fails, it returns either -1 (older implementations) or the
384 number of characters (not counting the terminating \0) that
385 *would have* been written if there had been enough room (C99).
386 In the former case, we double available_size and malloc to get a
387 larger buffer, and try again. In the latter case, we use the
388 returned information to build a buffer of the correct size. */
390 if (numwritten == -1)
392 /* Writing failed, and we don't know the needed size. Try
393 again with doubled size. */
394 int newsize = available_size << 1;
395 state->bigmsg = xrealloc (state->bigmsg, newsize);
396 state->allocated = newsize;
399 else if (numwritten >= available_size)
401 /* Writing failed, but we know exactly how much space we
403 int newsize = numwritten + 1;
404 state->bigmsg = xrealloc (state->bigmsg, newsize);
405 state->allocated = newsize;
409 /* Writing succeeded. */
410 saved_append (write_ptr);
411 FPUTS (write_ptr, fp);
413 xfree (state->bigmsg);
419 needs_flushing = true;
424 /* Flush LOGFP. Useful while flushing is disabled. */
428 FILE *fp = get_log_fp ();
432 On VMS, flush only for a terminal. See note at FPUTS macro, above.
435 if (isatty( fileno( fp)))
439 #else /* def __VMS */
441 #endif /* def __VMS [else] */
443 needs_flushing = false;
446 /* Enable or disable log flushing. */
448 log_set_flush (bool flush)
450 if (flush == flush_log_p)
455 /* Disable flushing by setting flush_log_p to 0. */
460 /* Reenable flushing. If anything was printed in no-flush mode,
461 flush the log now. */
468 /* (Temporarily) disable storing log to memory. Returns the old
469 status of storing, with which this function can be called again to
470 reestablish storing. */
473 log_set_save_context (bool savep)
475 bool old = save_context_p;
476 save_context_p = savep;
480 /* Print a message to the screen or to the log. The first argument
481 defines the verbosity of the message, and the rest are as in
485 logprintf (enum log_options o, const char *fmt, ...)
488 struct logvprintf_state lpstate;
491 check_redirect_output ();
499 va_start (args, fmt);
500 done = log_vprintf_internal (&lpstate, fmt, args);
507 /* The same as logprintf(), but does anything only if opt.debug is
510 debug_logprintf (const char *fmt, ...)
515 struct logvprintf_state lpstate;
518 check_redirect_output ();
525 va_start (args, fmt);
526 done = log_vprintf_internal (&lpstate, fmt, args);
532 #endif /* ENABLE_DEBUG */
534 /* Open FILE and set up a logging stream. If FILE cannot be opened,
535 exit with status of 1. */
537 log_init (const char *file, bool appendp)
541 logfp = fopen (file, appendp ? "a" : "w");
544 fprintf (stderr, "%s: %s: %s\n", exec_name, file, strerror (errno));
550 /* The log goes to stderr to avoid collisions with the output if
551 the user specifies `-O -'. #### Francois Pinard suggests
552 that it's a better idea to print to stdout by default, and to
553 stderr only if the user actually specifies `-O -'. He says
554 this inconsistency is harder to document, but is overall
555 easier on the user. */
560 && isatty (fileno (logfp))
564 /* If the output is a TTY, enable save context, i.e. store
565 the most recent several messages ("context") and dump
566 them to a log file in case SIGHUP or SIGUSR1 is received
567 (or Ctrl+Break is pressed under Windows). */
568 save_context_p = true;
573 /* Close LOGFP, inhibit further logging and free the memory associated
583 inhibit_logging = true;
584 save_context_p = false;
586 for (i = 0; i < SAVED_LOG_LINES; i++)
588 log_line_current = -1;
589 trailing_line = false;
592 /* Dump saved lines to logfp. */
594 log_dump_context (void)
596 int num = log_line_current;
597 FILE *fp = get_log_fp ();
607 struct log_ln *ln = log_lines + num;
609 FPUTS (ln->content, fp);
612 while (num != log_line_current);
614 if (log_lines[log_line_current].content)
615 FPUTS (log_lines[log_line_current].content, fp);
619 /* String escape functions. */
621 /* Return the number of non-printable characters in SOURCE.
622 Non-printable characters are determined as per c-ctype.c. */
625 count_nonprint (const char *source)
629 for (p = source, cnt = 0; *p; p++)
635 /* Copy SOURCE to DEST, escaping non-printable characters.
637 Non-printable refers to anything outside the non-control ASCII
638 range (32-126) which means that, for example, CR, LF, and TAB are
639 considered non-printable along with ESC, BS, and other control
640 chars. This is by design: it makes sure that messages from remote
641 servers cannot be easily used to deceive the users by mimicking
642 Wget's output. Disallowing non-ASCII characters is another
643 necessary security measure, which makes sure that remote servers
644 cannot garble the screen or guess the local charset and perform
647 Of course, the above mandates that escnonprint only be used in
648 contexts expected to be ASCII, such as when printing host names,
649 URL components, HTTP headers, FTP server messages, and the like.
651 ESCAPE is the leading character of the escape sequence. BASE
652 should be the base of the escape sequence, and must be either 8 for
655 DEST must point to a location with sufficient room to store an
656 encoded version of SOURCE. */
659 copy_and_escape (const char *source, char *dest, char escape, int base)
661 const char *from = source;
665 /* Copy chars from SOURCE to DEST, escaping non-printable ones. */
669 while ((c = *from++) != '\0')
675 *to++ = '0' + (c >> 6);
676 *to++ = '0' + ((c >> 3) & 7);
677 *to++ = '0' + (c & 7);
681 while ((c = *from++) != '\0')
687 *to++ = XNUM_TO_DIGIT (c >> 4);
688 *to++ = XNUM_TO_DIGIT (c & 0xf);
702 static struct ringel ring[RING_SIZE]; /* ring data */
705 escnonprint_internal (const char *str, char escape, int base)
707 static int ringpos; /* current ring position */
710 assert (base == 8 || base == 16);
712 nprcnt = count_nonprint (str);
714 /* If there are no non-printable chars in STR, don't bother
715 copying anything, just return STR. */
719 /* Set up a pointer to the current ring position, so we can write
720 simply r->X instead of ring[ringpos].X. */
721 struct ringel *r = ring + ringpos;
723 /* Every non-printable character is replaced with the escape char
724 and three (or two, depending on BASE) *additional* chars. Size
725 must also include the length of the original string and one
726 additional char for the terminating \0. */
727 int needed_size = strlen (str) + 1 + (base == 8 ? 3 * nprcnt : 2 * nprcnt);
729 /* If the current buffer is uninitialized or too small,
731 if (r->buffer == NULL || r->size < needed_size)
733 r->buffer = xrealloc (r->buffer, needed_size);
734 r->size = needed_size;
737 copy_and_escape (str, r->buffer, escape, base);
738 ringpos = (ringpos + 1) % RING_SIZE;
743 /* Return a pointer to a static copy of STR with the non-printable
744 characters escaped as \ooo. If there are no non-printable
745 characters in STR, STR is returned. See copy_and_escape for more
746 information on which characters are considered non-printable.
748 DON'T call this function on translated strings because escaping
749 will break them. Don't call it on literal strings from the source,
750 which are by definition trusted. If newlines are allowed in the
751 string, escape and print it line by line because escaping the whole
752 string will convert newlines to \012. (This is so that expectedly
753 single-line messages cannot use embedded newlines to mimic Wget's
754 output and deceive the user.)
756 escnonprint doesn't quote its escape character because it is notf
757 meant as a general and reversible quoting mechanism, but as a quick
758 way to defang binary junk sent by malicious or buggy servers.
760 NOTE: since this function can return a pointer to static data, be
761 careful to copy its result before calling it again. However, to be
762 more useful with printf, it maintains an internal ring of static
763 buffers to return. Currently the ring size is 3, which means you
764 can print up to three values in the same printf; if more is needed,
768 escnonprint (const char *str)
770 return escnonprint_internal (str, '\\', 8);
773 /* Return a pointer to a static copy of STR with the non-printable
774 characters escaped as %XX. If there are no non-printable
775 characters in STR, STR is returned.
777 See escnonprint for usage details. */
780 escnonprint_uri (const char *str)
782 return escnonprint_internal (str, '%', 16);
789 for (i = 0; i < countof (ring); i++)
790 xfree_null (ring[i].buffer);
793 /* When SIGHUP or SIGUSR1 are received, the output is redirected
794 elsewhere. Such redirection is only allowed once. */
795 static enum { RR_NONE, RR_REQUESTED, RR_DONE } redirect_request = RR_NONE;
796 static const char *redirect_request_signal_name;
798 /* Redirect output to `wget-log'. */
801 redirect_output (void)
804 logfp = unique_create (DEFAULT_LOGFILE, false, &logfile);
807 fprintf (stderr, _("\n%s received, redirecting output to %s.\n"),
808 redirect_request_signal_name, quote (logfile));
810 /* Dump the context output to the newly opened log. */
815 /* Eek! Opening the alternate log file has failed. Nothing we
816 can do but disable printing completely. */
817 fprintf (stderr, _("\n%s received.\n"), redirect_request_signal_name);
818 fprintf (stderr, _("%s: %s; disabling logging.\n"),
819 logfile, strerror (errno));
820 inhibit_logging = true;
822 save_context_p = false;
825 /* Check whether a signal handler requested the output to be
829 check_redirect_output (void)
831 if (redirect_request == RR_REQUESTED)
833 redirect_request = RR_DONE;
838 /* Request redirection at a convenient time. This may be called from
842 log_request_redirect_output (const char *signal_name)
844 if (redirect_request == RR_NONE && save_context_p)
845 /* Request output redirection. The request will be processed by
846 check_redirect_output(), which is called from entry point log
848 redirect_request = RR_REQUESTED;
849 redirect_request_signal_name = signal_name;