X-Git-Url: https://git.dogcows.com/gitweb?a=blobdiff_plain;f=src%2Fcreate.c;h=30fe90bd9ae52728af5f3b763711bfe8b317deaa;hb=8111f9d34df33cb06072275b2ec7dd10ef119d08;hp=5bc51ae6f1f6a1171771b99a2f6ee7acadb4861a;hpb=f5116dd1dbd4b1c510005934be3563102eb7b3be;p=chaz%2Ftar diff --git a/src/create.c b/src/create.c index 5bc51ae..30fe90b 100644 --- a/src/create.c +++ b/src/create.c @@ -1,5 +1,5 @@ /* Create a tar archive. - Copyright 1985, 92, 93, 94, 96, 97, 1999 Free Software Foundation, Inc. + Copyright 1985, 92, 93, 94, 96, 97, 99, 2000 Free Software Foundation, Inc. Written by John Gilmore, on 1985-08-25. This program is free software; you can redistribute it and/or modify it @@ -33,39 +33,23 @@ struct utimbuf }; #endif +#include + #include "common.h" +#include #ifndef MSDOS extern dev_t ar_dev; extern ino_t ar_ino; #endif -extern struct name *gnu_list_name; - -/* This module is the only one that cares about `struct link's. */ - struct link { - struct link *next; dev_t dev; ino_t ino; - short linkcount; char name[1]; }; - -struct link *linklist = NULL; /* points to first link in list */ -/* Base 64 digits; see Internet RFC 2045 Table 1. */ -char const base_64_digits[64] = -{ - 'A', 'B', 'C', 'D', 'E', 'F', 'G', 'H', 'I', 'J', 'K', 'L', 'M', - 'N', 'O', 'P', 'Q', 'R', 'S', 'T', 'U', 'V', 'W', 'X', 'Y', 'Z', - 'a', 'b', 'c', 'd', 'e', 'f', 'g', 'h', 'i', 'j', 'k', 'l', 'm', - 'n', 'o', 'p', 'q', 'r', 's', 't', 'u', 'v', 'w', 'x', 'y', 'z', - '0', '1', '2', '3', '4', '5', '6', '7', '8', '9', '+', '/' -}; -#define base_8_digits (base_64_digits + 26 * 2) - /* The maximum uintmax_t value that can be represented with DIGITS digits, assuming that each digit is BITS_PER_DIGIT wide. */ #define MAX_VAL_WITH_DIGITS(digits, bits_per_digit) \ @@ -73,98 +57,154 @@ char const base_64_digits[64] = ? ((uintmax_t) 1 << ((digits) * (bits_per_digit))) - 1 \ : (uintmax_t) -1) -/* Convert VALUE to a representation suitable for tar headers, - using base 1 << BITS_PER_DIGIT. - Use the digits in DIGIT_CHAR[0] ... DIGIT_CHAR[base - 1]. +/* Convert VALUE to an octal representation suitable for tar headers. Output to buffer WHERE with size SIZE. The result is undefined if SIZE is 0 or if VALUE is too large to fit. */ static void -to_base (uintmax_t value, int bits_per_digit, char const *digit_char, - char *where, size_t size) +to_octal (uintmax_t value, char *where, size_t size) { uintmax_t v = value; size_t i = size; - unsigned digit_mask = (1 << bits_per_digit) - 1; do { - where[--i] = digit_char[v & digit_mask]; - v >>= bits_per_digit; + where[--i] = '0' + (v & ((1 << LG_8) - 1)); + v >>= LG_8; } while (i); } -/* NEGATIVE is nonzero if VALUE was negative before being cast to - uintmax_t; its original bitpattern can be deduced from VALSIZE, its - original size before casting. Convert VALUE to external form, - using SUBSTITUTE (...) if VALUE won't fit. Output to buffer WHERE - with size SIZE. TYPE is the kind of value being output (useful for - diagnostics). Prefer the POSIX format of SIZE - 1 octal digits - (with leading zero digits), followed by '\0'. If this won't work, - and if GNU format is allowed, use '+' or '-' followed by SIZE - 1 - base-64 digits. If neither format works, use SUBSTITUTE (...) - instead. Pass to SUBSTITUTE the address of an 0-or-1 flag - recording whether the substitute value is negative. */ +/* Convert NEGATIVE VALUE to a base-256 representation suitable for + tar headers. NEGATIVE is 1 if VALUE was negative before being cast + to uintmax_t, 0 otherwise. Output to buffer WHERE with size SIZE. + The result is undefined if SIZE is 0 or if VALUE is too large to + fit. */ + +static void +to_base256 (int negative, uintmax_t value, char *where, size_t size) +{ + uintmax_t v = value; + uintmax_t propagated_sign_bits = + ((uintmax_t) - negative << (CHAR_BIT * sizeof v - LG_256)); + size_t i = size; + + do + { + where[--i] = v & ((1 << LG_256) - 1); + v = propagated_sign_bits | (v >> LG_256); + } + while (i); +} + +/* Convert NEGATIVE VALUE (which was originally of size VALSIZE) to + external form, using SUBSTITUTE (...) if VALUE won't fit. Output + to buffer WHERE with size SIZE. NEGATIVE is 1 iff VALUE was + negative before being cast to uintmax_t; its original bitpattern + can be deduced from VALSIZE, its original size before casting. + TYPE is the kind of value being output (useful for diagnostics). + Prefer the POSIX format of SIZE - 1 octal digits (with leading zero + digits), followed by '\0'. If this won't work, and if GNU or + OLDGNU format is allowed, use '\200' followed by base-256, or (if + NEGATIVE is nonzero) '\377' followed by two's complement base-256. + If neither format works, use SUBSTITUTE (...) instead. Pass to + SUBSTITUTE the address of an 0-or-1 flag recording whether the + substitute value is negative. */ static void to_chars (int negative, uintmax_t value, size_t valsize, uintmax_t (*substitute) PARAMS ((int *)), char *where, size_t size, const char *type) { - uintmax_t v = negative ? -value : value; + int base256_allowed = (archive_format == GNU_FORMAT + || archive_format == OLDGNU_FORMAT); - if (! negative && v <= MAX_VAL_WITH_DIGITS (size - 1, LG_8)) + /* Generate the POSIX octal representation if the number fits. */ + if (! negative && value <= MAX_VAL_WITH_DIGITS (size - 1, LG_8)) { where[size - 1] = '\0'; - to_base (v, LG_8, base_8_digits, where, size - 1); + to_octal (value, where, size - 1); } - else if (v <= MAX_VAL_WITH_DIGITS (size - 1, LG_64) - && archive_format == GNU_FORMAT) + + /* Otherwise, generate the base-256 representation if we are + generating an old or new GNU format and if the number fits. */ + else if (((negative ? -1 - value : value) + <= MAX_VAL_WITH_DIGITS (size - 1, LG_256)) + && base256_allowed) { - where[0] = negative ? '-' : '+'; - to_base (v, LG_64, base_64_digits, where + 1, size - 1); + where[0] = negative ? -1 : 1 << (LG_256 - 1); + to_base256 (negative, value, where + 1, size - 1); } - else if (negative - && archive_format != GNU_FORMAT - && valsize * CHAR_BIT <= (size - 1) * LG_8) + + /* Otherwise, if the number is negative, and if it would not cause + ambiguity on this host by confusing positive with negative + values, then generate the POSIX octal representation of the value + modulo 2**(field bits). The resulting tar file is + machine-dependent, since it depends on the host word size. Yuck! + But this is the traditional behavior. */ + else if (negative && valsize * CHAR_BIT <= (size - 1) * LG_8) { + static int warned_once; + if (! warned_once) + { + warned_once = 1; + WARN ((0, 0, _("Generating negative octal headers"))); + } where[size - 1] = '\0'; - to_base (value & MAX_VAL_WITH_DIGITS (valsize * CHAR_BIT, 1), - LG_8, base_8_digits, where, size - 1); + to_octal (value & MAX_VAL_WITH_DIGITS (valsize * CHAR_BIT, 1), + where, size - 1); } + + /* Otherwise, output a substitute value if possible (with a + warning), and an error message if not. */ else { - uintmax_t maxval = (archive_format == GNU_FORMAT - ? MAX_VAL_WITH_DIGITS (size - 1, LG_64) + uintmax_t maxval = (base256_allowed + ? MAX_VAL_WITH_DIGITS (size - 1, LG_256) : MAX_VAL_WITH_DIGITS (size - 1, LG_8)); - char buf1[UINTMAX_STRSIZE_BOUND + 1]; - char buf2[UINTMAX_STRSIZE_BOUND + 1]; - char buf3[UINTMAX_STRSIZE_BOUND + 1]; - char *value_string = STRINGIFY_BIGINT (v, buf1 + 1); - char *maxval_string = STRINGIFY_BIGINT (maxval, buf2 + 1); - char const *minval_string = - (archive_format == GNU_FORMAT - ? "0" - : (maxval_string[-1] = '-', maxval_string - 1)); + char valbuf[UINTMAX_STRSIZE_BOUND + 1]; + char maxbuf[UINTMAX_STRSIZE_BOUND]; + char minbuf[UINTMAX_STRSIZE_BOUND + 1]; + char const *minval_string; + char const *maxval_string = STRINGIFY_BIGINT (maxval, maxbuf); + char const *value_string; + + if (base256_allowed) + { + uintmax_t m = maxval + 1 ? maxval + 1 : maxval / 2 + 1; + char *p = STRINGIFY_BIGINT (m, minbuf + 1); + *--p = '-'; + minval_string = p; + } + else + minval_string = "0"; + if (negative) - *--value_string = '-'; + { + char *p = STRINGIFY_BIGINT (- value, valbuf + 1); + *--p = '-'; + value_string = p; + } + else + value_string = STRINGIFY_BIGINT (value, valbuf); + if (substitute) { int negsub; uintmax_t sub = substitute (&negsub) & maxval; - uintmax_t s = (negsub &= archive_format == GNU_FORMAT) ? -sub : sub; - char *sub_string = STRINGIFY_BIGINT (s, buf3 + 1); + uintmax_t s = (negsub &= archive_format == GNU_FORMAT) ? - sub : sub; + char subbuf[UINTMAX_STRSIZE_BOUND + 1]; + char *sub_string = STRINGIFY_BIGINT (s, subbuf + 1); if (negsub) *--sub_string = '-'; - WARN ((0, 0, _("%s value %s out of range %s..%s; substituting %s"), - type, value_string, minval_string, maxval_string, + WARN ((0, 0, _("value %s out of %s range %s..%s; substituting %s"), + value_string, type, minval_string, maxval_string, sub_string)); - to_chars (negsub, s, valsize, NULL, where, size, type); + to_chars (negsub, s, valsize, 0, where, size, type); } else - ERROR ((0, 0, _("%s value %s out of range %s..%s"), - type, value_string, minval_string, maxval_string)); + ERROR ((0, 0, _("value %s out of %s range %s..%s"), + value_string, type, minval_string, maxval_string)); } } @@ -193,19 +233,20 @@ gid_to_chars (gid_t v, char *p, size_t s) void major_to_chars (major_t v, char *p, size_t s) { - to_chars (v < 0, (uintmax_t) v, sizeof v, NULL, p, s, "major_t"); + to_chars (v < 0, (uintmax_t) v, sizeof v, 0, p, s, "major_t"); } void minor_to_chars (minor_t v, char *p, size_t s) { - to_chars (v < 0, (uintmax_t) v, sizeof v, NULL, p, s, "minor_t"); + to_chars (v < 0, (uintmax_t) v, sizeof v, 0, p, s, "minor_t"); } void mode_to_chars (mode_t v, char *p, size_t s) { /* In the common case where the internal and external mode bits are the same, + and we are not using POSIX or GNU format, propagate all unknown bits to the external mode. This matches historical practice. Otherwise, just copy the bits we know about. */ @@ -214,7 +255,9 @@ mode_to_chars (mode_t v, char *p, size_t s) if (S_ISUID == TSUID && S_ISGID == TSGID && S_ISVTX == TSVTX && S_IRUSR == TUREAD && S_IWUSR == TUWRITE && S_IXUSR == TUEXEC && S_IRGRP == TGREAD && S_IWGRP == TGWRITE && S_IXGRP == TGEXEC - && S_IROTH == TOREAD && S_IWOTH == TOWRITE && S_IXOTH == TOEXEC) + && S_IROTH == TOREAD && S_IWOTH == TOWRITE && S_IXOTH == TOEXEC + && archive_format != POSIX_FORMAT + && archive_format != GNU_FORMAT) { negative = v < 0; u = v; @@ -235,25 +278,25 @@ mode_to_chars (mode_t v, char *p, size_t s) | (v & S_IWOTH ? TOWRITE : 0) | (v & S_IXOTH ? TOEXEC : 0)); } - to_chars (negative, u, sizeof v, NULL, p, s, "mode_t"); + to_chars (negative, u, sizeof v, 0, p, s, "mode_t"); } void off_to_chars (off_t v, char *p, size_t s) { - to_chars (v < 0, (uintmax_t) v, sizeof v, NULL, p, s, "off_t"); + to_chars (v < 0, (uintmax_t) v, sizeof v, 0, p, s, "off_t"); } void size_to_chars (size_t v, char *p, size_t s) { - to_chars (0, (uintmax_t) v, sizeof v, NULL, p, s, "size_t"); + to_chars (0, (uintmax_t) v, sizeof v, 0, p, s, "size_t"); } void time_to_chars (time_t v, char *p, size_t s) { - to_chars (v < 0, (uintmax_t) v, sizeof v, NULL, p, s, "time_t"); + to_chars (v < 0, (uintmax_t) v, sizeof v, 0, p, s, "time_t"); } static uintmax_t @@ -281,28 +324,22 @@ uid_to_chars (uid_t v, char *p, size_t s) void uintmax_to_chars (uintmax_t v, char *p, size_t s) { - to_chars (0, v, sizeof v, NULL, p, s, "uintmax_t"); + to_chars (0, v, sizeof v, 0, p, s, "uintmax_t"); } /* Writing routines. */ -/*-----------------------------------------------------------------------. -| Just zeroes out the buffer so we don't confuse ourselves with leftover | -| data. | -`-----------------------------------------------------------------------*/ - +/* Zero out the buffer so we don't confuse ourselves with leftover + data. */ static void clear_buffer (char *buffer) { memset (buffer, 0, BLOCKSIZE); } -/*-------------------------------------------------------------------------. -| Write the EOT block(s). We zero at least two blocks, through | -| the end of the record. Old tar, as previous versions of GNU tar, writes | -| garbage after two zeroed blocks. | -`-------------------------------------------------------------------------*/ - +/* Write the EOT block(s). Zero at least two blocks, through the end + of the record. Old tar, as previous versions of GNU tar, writes + garbage after two zeroed blocks. */ void write_eot (void) { @@ -314,9 +351,7 @@ write_eot (void) set_next_block_after (pointer); } -/*-----------------------------------------------------. -| Write a GNUTYPE_LONGLINK or GNUTYPE_LONGNAME block. | -`-----------------------------------------------------*/ +/* Write a GNUTYPE_LONGLINK or GNUTYPE_LONGNAME block. */ /* FIXME: Cross recursion between start_header and write_long! */ @@ -357,10 +392,8 @@ write_long (const char *p, char type) /* Header handling. */ -/*---------------------------------------------------------------------. -| Make a header block for the file name whose stat info is st. Return | -| header pointer for success, NULL if the name is too long. | -`---------------------------------------------------------------------*/ +/* Make a header block for the file whose stat info is st, + and return its address. */ static union block * start_header (const char *name, struct stat *st) @@ -377,7 +410,7 @@ start_header (const char *name, struct stat *st) if (!warned_once) { warned_once = 1; - WARN ((0, 0, _("Removing `%.*s' prefix from archive names"), + WARN ((0, 0, _("Removing leading `%.*s' from member names"), (int) prefix_len, name)); } name += prefix_len; @@ -389,15 +422,22 @@ start_header (const char *name, struct stat *st) if (!warned_once) { warned_once = 1; - WARN ((0, 0, _("Removing leading `/' from archive names"))); + WARN ((0, 0, _("Removing leading `/' from member names"))); } name++; } - } - /* Check the file name and put it in the block. */ + { + static int warned_once; + if (! warned_once && contains_dot_dot (name)) + { + warned_once = 1; + WARN ((0, 0, _("Member names contain `..'"))); + } + } + } - if (strlen (name) >= (size_t) NAME_FIELD_SIZE) + if (sizeof header->header.name <= strlen (name)) write_long (name, GNUTYPE_LONGNAME); header = find_next_block (); memset (header->buffer, 0, sizeof (union block)); @@ -492,11 +532,8 @@ start_header (const char *name, struct stat *st) return header; } -/*-------------------------------------------------------------------------. -| Finish off a filled-in header block and write it out. We also print the | -| file name and/or full info if verbose is on. | -`-------------------------------------------------------------------------*/ - +/* Finish off a filled-in header block and write it out. We also + print the file name and/or full info if verbose is on. */ void finish_header (union block *header) { @@ -504,11 +541,11 @@ finish_header (union block *header) int sum; char *p; - memcpy (header->header.chksum, CHKBLANKS, sizeof (header->header.chksum)); + memcpy (header->header.chksum, CHKBLANKS, sizeof header->header.chksum); sum = 0; p = header->buffer; - for (i = sizeof (*header); i-- != 0; ) + for (i = sizeof *header; i-- != 0; ) /* We can't use unsigned char here because of old compilers, e.g. V7. */ sum += 0xFF & *p++; @@ -524,8 +561,6 @@ finish_header (union block *header) uintmax_to_chars ((uintmax_t) sum, header->header.chksum, 7); - set_next_block_after (header); - if (verbose_option && header->header.typeflag != GNUTYPE_LONGLINK && header->header.typeflag != GNUTYPE_LONGNAME) @@ -537,16 +572,15 @@ finish_header (union block *header) current_format = archive_format; print_header (); } + + set_next_block_after (header); } /* Sparse file processing. */ -/*-------------------------------------------------------------------------. -| Takes a blockful of data and basically cruises through it to see if it's | -| made *entirely* of zeros, returning a 0 the instant it finds something | -| that is a nonzero, i.e., useful data. | -`-------------------------------------------------------------------------*/ - +/* Takes a blockful of data and basically cruises through it to see if + it's made *entirely* of zeros, returning a 0 the instant it finds + something that is a nonzero, i.e., useful data. */ static int zero_block_p (char *buffer) { @@ -558,10 +592,6 @@ zero_block_p (char *buffer) return 1; } -/*---. -| ? | -`---*/ - static void init_sparsearray (void) { @@ -571,8 +601,7 @@ init_sparsearray (void) /* Make room for our scratch space -- initially is 10 elts long. */ - sparsearray = (struct sp_array *) - xmalloc (sp_array_size * sizeof (struct sp_array)); + sparsearray = xmalloc (sp_array_size * sizeof (struct sp_array)); for (counter = 0; counter < sp_array_size; counter++) { sparsearray[counter].offset = 0; @@ -580,28 +609,21 @@ init_sparsearray (void) } } -/*---. -| ? | -`---*/ - -static void -find_new_file_size (off_t *filesize, int highest_index) +static off_t +find_new_file_size (int sparses) { - int counter; - - *filesize = 0; - for (counter = 0; - sparsearray[counter].numbytes && counter <= highest_index; - counter++) - *filesize += sparsearray[counter].numbytes; + int i; + off_t s = 0; + for (i = 0; i < sparses; i++) + s += sparsearray[i].numbytes; + return s; } -/*-----------------------------------------------------------------------. -| Make one pass over the file NAME, studying where any non-zero data is, | -| that is, how far into the file each instance of data is, and how many | -| bytes are there. Save this information in the sparsearray, which will | -| later be translated into header information. | -`-----------------------------------------------------------------------*/ +/* Make one pass over the file NAME, studying where any non-zero data + is, that is, how far into the file each instance of data is, and + how many bytes are there. Save this information in the + sparsearray, which will later be translated into header + information. */ /* There is little point in trimming small amounts of null data at the head and tail of blocks, only avoid dumping full null blocks. */ @@ -615,7 +637,7 @@ deal_with_sparse (char *name, union block *header) size_t numbytes = 0; off_t offset = 0; int file; - int sparse_index = 0; + int sparses = 0; ssize_t count; char buffer[BLOCKSIZE]; @@ -629,15 +651,14 @@ deal_with_sparse (char *name, union block *header) init_sparsearray (); clear_buffer (buffer); - while (count = safe_read (file, buffer, sizeof buffer), count != 0) + while (0 < (count = safe_read (file, buffer, sizeof buffer))) { /* Realloc the scratch area as necessary. FIXME: should reallocate only at beginning of a new instance of non-zero data. */ - if (sparse_index > sp_array_size - 1) + if (sp_array_size <= sparses) { - - sparsearray = (struct sp_array *) + sparsearray = xrealloc (sparsearray, 2 * sp_array_size * sizeof (struct sp_array)); sp_array_size *= 2; @@ -651,14 +672,14 @@ deal_with_sparse (char *name, union block *header) { if (numbytes) { - sparsearray[sparse_index++].numbytes = numbytes; + sparsearray[sparses++].numbytes = numbytes; numbytes = 0; } } else { if (!numbytes) - sparsearray[sparse_index].offset = offset; + sparsearray[sparses].offset = offset; numbytes += count; } @@ -669,7 +690,7 @@ deal_with_sparse (char *name, union block *header) if (!zero_block_p (buffer)) { if (!numbytes) - sparsearray[sparse_index].offset = offset; + sparsearray[sparses].offset = offset; numbytes += count; } else @@ -690,53 +711,36 @@ deal_with_sparse (char *name, union block *header) } if (numbytes) - sparsearray[sparse_index++].numbytes = numbytes; + sparsearray[sparses++].numbytes = numbytes; else { - sparsearray[sparse_index].offset = offset - 1; - sparsearray[sparse_index++].numbytes = 1; + sparsearray[sparses].offset = offset - 1; + sparsearray[sparses++].numbytes = 1; } - close (file); - return sparse_index - 1; + return close (file) == 0 && 0 <= count ? sparses : 0; } -/*---. -| ? | -`---*/ - static int finish_sparse_file (int file, off_t *sizeleft, off_t fullsize, char *name) { union block *start; size_t bufsize; - int sparse_index = 0; + int sparses = 0; ssize_t count; while (*sizeleft > 0) { start = find_next_block (); memset (start->buffer, 0, BLOCKSIZE); - bufsize = sparsearray[sparse_index].numbytes; - if (!bufsize) - { - /* We blew it, maybe. */ - char buf1[UINTMAX_STRSIZE_BOUND]; - char buf2[UINTMAX_STRSIZE_BOUND]; - - ERROR ((0, 0, _("Wrote %s of %s bytes to file %s"), - STRINGIFY_BIGINT (fullsize - *sizeleft, buf1), - STRINGIFY_BIGINT (fullsize, buf2), - name)); - break; - } + bufsize = sparsearray[sparses].numbytes; + if (! bufsize) + abort (); - if (lseek (file, sparsearray[sparse_index++].offset, SEEK_SET) < 0) + if (lseek (file, sparsearray[sparses++].offset, SEEK_SET) < 0) { - char buf[UINTMAX_STRSIZE_BOUND]; - ERROR ((0, errno, _("lseek error at byte %s in file %s"), - STRINGIFY_BIGINT (sparsearray[sparse_index - 1].offset, buf), - name)); + (ignore_failed_read_option ? seek_warn_details : seek_error_details) + (name, sparsearray[sparses - 1].offset); break; } @@ -745,28 +749,13 @@ finish_sparse_file (int file, off_t *sizeleft, off_t fullsize, char *name) while (bufsize > BLOCKSIZE) { -#if 0 - if (amount_read) - { - count = safe_read (file, start->buffer + amount_read, - BLOCKSIZE - amount_read); - bufsize -= BLOCKSIZE - amount_read; - amount_read = 0; - set_next_block_after (start); - start = find_next_block (); - memset (start->buffer, 0, BLOCKSIZE); - } -#endif - /* Store the data. */ - count = safe_read (file, start->buffer, BLOCKSIZE); if (count < 0) { - char buf[UINTMAX_STRSIZE_BOUND]; - ERROR ((0, errno, - _("Read error at byte %s, reading %lu bytes, in file %s"), - STRINGIFY_BIGINT (fullsize - *sizeleft, buf), - (unsigned long) bufsize, name)); + (ignore_failed_read_option + ? read_warn_details + : read_error_details) + (name, fullsize - *sizeleft, bufsize); return 1; } bufsize -= count; @@ -786,34 +775,15 @@ finish_sparse_file (int file, off_t *sizeleft, off_t fullsize, char *name) if (count < 0) { - char buf[UINTMAX_STRSIZE_BOUND]; - - ERROR ((0, errno, - _("Read error at byte %s, reading %lu bytes, in file %s"), - STRINGIFY_BIGINT (fullsize - *sizeleft, buf), - (unsigned long) bufsize, name)); + (ignore_failed_read_option + ? read_warn_details + : read_error_details) + (name, fullsize - *sizeleft, bufsize); return 1; } -#if 0 - if (amount_read >= BLOCKSIZE) - { - amount_read = 0; - set_next_block_after (start + (count - 1) / BLOCKSIZE); - if (count != bufsize) - { - ERROR ((0, 0, - _("File %s shrunk, padding with zeros"), - name)); - return 1; - } - start = find_next_block (); - } - else - amount_read += bufsize; -#endif + *sizeleft -= count; set_next_block_after (start); - } free (sparsearray); #if 0 @@ -824,10 +794,6 @@ finish_sparse_file (int file, off_t *sizeleft, off_t fullsize, char *name) /* Main functions of this module. */ -/*---. -| ? | -`---*/ - void create_archive (void) { @@ -837,32 +803,48 @@ create_archive (void) if (incremental_option) { - char *buffer = xmalloc (PATH_MAX); + size_t buffer_size = 1000; + char *buffer = xmalloc (buffer_size); const char *q; - char *bufp; collect_and_sort_names (); while (p = name_from_list (), p) if (!excluded_name (p)) - dump_file (p, (dev_t) -1, 1); + dump_file (p, -1, (dev_t) 0); blank_name_list (); while (p = name_from_list (), p) if (!excluded_name (p)) { - strcpy (buffer, p); - if (p[strlen (p) - 1] != '/') - strcat (buffer, "/"); - bufp = buffer + strlen (buffer); + size_t plen = strlen (p); + if (buffer_size <= plen) + { + while ((buffer_size *= 2) <= plen) + continue; + buffer = xrealloc (buffer, buffer_size); + } + memcpy (buffer, p, plen); + if (buffer[plen - 1] != '/') + buffer[plen++] = '/'; q = gnu_list_name->dir_contents; if (q) - for (; *q; q += strlen (q) + 1) - if (*q == 'Y') - { - strcpy (bufp, q + 1); - dump_file (buffer, (dev_t) -1, 1); - } + while (*q) + { + size_t qlen = strlen (q); + if (*q == 'Y') + { + if (buffer_size < plen + qlen) + { + while ((buffer_size *=2 ) < plen + qlen) + continue; + buffer = xrealloc (buffer, buffer_size); + } + strcpy (buffer + plen, q + 1); + dump_file (buffer, -1, (dev_t) 0); + } + q += qlen + 1; + } } free (buffer); } @@ -870,64 +852,73 @@ create_archive (void) { while (p = name_next (1), p) if (!excluded_name (p)) - dump_file (p, (dev_t) -1, 1); + dump_file (p, 1, (dev_t) 0); } write_eot (); close_archive (); if (listed_incremental_option) - write_dir_file (); + write_directory_file (); +} + + +/* Calculate the hash of a link. */ +static unsigned +hash_link (void const *entry, unsigned n_buckets) +{ + struct link const *link = entry; + return (uintmax_t) (link->dev ^ link->ino) % n_buckets; +} + +/* Compare two links for equality. */ +static bool +compare_links (void const *entry1, void const *entry2) +{ + struct link const *link1 = entry1; + struct link const *link2 = entry2; + return ((link1->dev ^ link2->dev) | (link1->ino ^ link2->ino)) == 0; } -/*----------------------------------------------------------------------. -| Dump a single file. Recurse on directories. Result is nonzero for | -| success. P is file name to dump. PARENT_DEVICE is device our parent | -| directory was on. TOP_LEVEL tells wether we are a toplevel call. | -| | -| Sets global CURRENT_STAT to stat output for this file. | -`----------------------------------------------------------------------*/ +/* Dump a single file, recursing on directories. P is the file name + to dump. TOP_LEVEL tells whether this is a top-level call; zero + means no, positive means yes, and negative means an incremental + dump. PARENT_DEVICE is the device of P's + parent directory; it is examined only if TOP_LEVEL is zero. + + Set global CURRENT_STAT to stat output for this file. */ /* FIXME: One should make sure that for *every* path leading to setting exit_status to failure, a clear diagnostic has been issued. */ void -dump_file (char *p, dev_t parent_device, int top_level) +dump_file (char *p, int top_level, dev_t parent_device) { union block *header; char type; union block *exhdr; char save_typeflag; + time_t original_ctime; struct utimbuf restore_times; - off_t restore_size; - /* FIXME: `header' and `upperbound' might be used uninitialized in this + /* FIXME: `header' might be used uninitialized in this function. Reported by Bruno Haible. */ if (interactive_option && !confirm ("add", p)) return; - /* Use stat if following (rather than dumping) 4.2BSD's symbolic links. - Otherwise, use lstat (which falls back to stat if no symbolic links). */ - - if (dereference_option != 0 -#if STX_HIDDEN && !_LARGE_FILES /* AIX */ - ? statx (p, ¤t_stat, STATSIZE, STX_HIDDEN) - : statx (p, ¤t_stat, STATSIZE, STX_HIDDEN | STX_LINK) -#else - ? stat (p, ¤t_stat) : lstat (p, ¤t_stat) -#endif - ) + if (deref_stat (dereference_option, p, ¤t_stat) != 0) { - WARN ((0, errno, _("Cannot add file %s"), p)); - if (!ignore_failed_read_option) - exit_status = TAREXIT_FAILURE; + if (ignore_failed_read_option) + stat_warn (p); + else + stat_error (p); return; } + original_ctime = current_stat.st_ctime; restore_times.actime = current_stat.st_atime; restore_times.modtime = current_stat.st_mtime; - restore_size = current_stat.st_size; #ifdef S_ISHIDDEN if (S_ISHIDDEN (current_stat.st_mode)) @@ -945,13 +936,14 @@ dump_file (char *p, dev_t parent_device, int top_level) /* See if we want only new files, and check if this one is too old to put in the archive. */ - if ((!incremental_option || listed_incremental_option) + if ((0 < top_level || !incremental_option) && !S_ISDIR (current_stat.st_mode) && current_stat.st_mtime < newer_mtime_option && (!after_date_option || current_stat.st_ctime < newer_ctime_option)) { - if (!listed_incremental_option && parent_device == (dev_t) -1) - WARN ((0, 0, _("%s: is unchanged; not dumped"), p)); + if (0 < top_level) + WARN ((0, 0, _("%s: file is unchanged; not dumped"), + quotearg_colon (p))); /* FIXME: recheck this return. */ return; } @@ -961,454 +953,31 @@ dump_file (char *p, dev_t parent_device, int top_level) if (ar_dev && current_stat.st_dev == ar_dev && current_stat.st_ino == ar_ino) { - WARN ((0, 0, _("%s is the archive; not dumped"), p)); - return; - } -#endif - - /* Check for multiple links. - - We maintain a list of all such files that we've written so far. Any - time we see another, we check the list and avoid dumping the data - again if we've done it once already. */ - - if (current_stat.st_nlink > 1 - && (S_ISREG (current_stat.st_mode) - || S_ISCTG (current_stat.st_mode) - || S_ISCHR (current_stat.st_mode) - || S_ISBLK (current_stat.st_mode) - || S_ISFIFO (current_stat.st_mode))) - { - struct link *lp; - - /* FIXME: First quick and dirty. Hashing, etc later. */ - - for (lp = linklist; lp; lp = lp->next) - if (lp->ino == current_stat.st_ino && lp->dev == current_stat.st_dev) - { - char *link_name = lp->name; - - /* We found a link. */ - - while (!absolute_names_option && *link_name == '/') - { - static int warned_once; - if (!warned_once) - { - warned_once = 1; - WARN ((0, 0, _("Removing leading `/' from link names"))); - } - link_name++; - } - if (strlen (link_name) >= NAME_FIELD_SIZE) - write_long (link_name, GNUTYPE_LONGLINK); - assign_string (¤t_link_name, link_name); - - current_stat.st_size = 0; - header = start_header (p, ¤t_stat); - if (header == NULL) - { - exit_status = TAREXIT_FAILURE; - return; - } - strncpy (header->header.linkname, - link_name, NAME_FIELD_SIZE); - - /* Force null truncated. */ - - header->header.linkname[NAME_FIELD_SIZE - 1] = 0; - - header->header.typeflag = LNKTYPE; - finish_header (header); - - /* FIXME: Maybe remove from list after all links found? */ - - if (remove_files_option) - if (unlink (p) == -1) - ERROR ((0, errno, _("Cannot remove %s"), p)); - - /* We dumped it. */ - return; - } - - /* Not found. Add it to the list of possible links. */ - - lp = (struct link *) - xmalloc ((size_t) (sizeof (struct link) + strlen (p))); - lp->ino = current_stat.st_ino; - lp->dev = current_stat.st_dev; - strcpy (lp->name, p); - lp->next = linklist; - linklist = lp; - } - - /* This is not a link to a previously dumped file, so dump it. */ - - if (S_ISREG (current_stat.st_mode) - || S_ISCTG (current_stat.st_mode)) - { - int f; /* file descriptor */ - size_t bufsize; - ssize_t count; - off_t sizeleft; - union block *start; - int header_moved; - char isextended = 0; - int upperbound; -#if 0 - static int cried_once = 0; -#endif - - header_moved = 0; - - if (sparse_option) - { - /* Check the size of the file against the number of blocks - allocated for it, counting both data and indirect blocks. - If there is a smaller number of blocks that would be - necessary to accommodate a file of this size, this is safe - to say that we have a sparse file: at least one of those - blocks in the file is just a useless hole. For sparse - files not having more hole blocks than indirect blocks, the - sparseness will go undetected. */ - - /* Bruno Haible sent me these statistics for Linux. It seems - that some filesystems count indirect blocks in st_blocks, - while others do not seem to: - - minix-fs tar: size=7205, st_blocks=18 and ST_NBLOCKS=18 - extfs tar: size=7205, st_blocks=18 and ST_NBLOCKS=18 - ext2fs tar: size=7205, st_blocks=16 and ST_NBLOCKS=16 - msdos-fs tar: size=7205, st_blocks=16 and ST_NBLOCKS=16 - - Dick Streefland reports the previous numbers as misleading, - because ext2fs use 12 direct blocks, while minix-fs uses only - 6 direct blocks. Dick gets: - - ext2 size=20480 ls listed blocks=21 - minix size=20480 ls listed blocks=21 - msdos size=20480 ls listed blocks=20 - - It seems that indirect blocks *are* included in st_blocks. - The minix filesystem does not account for phantom blocks in - st_blocks, so `du' and `ls -s' give wrong results. So, the - --sparse option would not work on a minix filesystem. */ - - if (ST_NBLOCKS (current_stat) - < (current_stat.st_size / ST_NBLOCKSIZE - + (current_stat.st_size % ST_NBLOCKSIZE != 0))) - { - off_t filesize = current_stat.st_size; - int counter; - - header = start_header (p, ¤t_stat); - if (header == NULL) - { - exit_status = TAREXIT_FAILURE; - return; - } - header->header.typeflag = GNUTYPE_SPARSE; - header_moved = 1; - - /* Call the routine that figures out the layout of the - sparse file in question. UPPERBOUND is the index of the - last element of the "sparsearray," i.e., the number of - elements it needed to describe the file. */ - - upperbound = deal_with_sparse (p, header); - - /* See if we'll need an extended header later. */ - - if (upperbound > SPARSES_IN_OLDGNU_HEADER - 1) - header->oldgnu_header.isextended = 1; - - /* We store the "real" file size so we can show that in - case someone wants to list the archive, i.e., tar tvf - . It might be kind of disconcerting if the - shrunken file size was the one that showed up. */ - - OFF_TO_CHARS (current_stat.st_size, - header->oldgnu_header.realsize); - - /* This will be the new "size" of the file, i.e., the size - of the file minus the blocks of holes that we're - skipping over. */ - - find_new_file_size (&filesize, upperbound); - current_stat.st_size = filesize; - OFF_TO_CHARS (filesize, header->header.size); - - for (counter = 0; counter < SPARSES_IN_OLDGNU_HEADER; counter++) - { - if (!sparsearray[counter].numbytes) - break; - - OFF_TO_CHARS (sparsearray[counter].offset, - header->oldgnu_header.sp[counter].offset); - SIZE_TO_CHARS (sparsearray[counter].numbytes, - header->oldgnu_header.sp[counter].numbytes); - } - - } - } - else - upperbound = SPARSES_IN_OLDGNU_HEADER - 1; - - sizeleft = current_stat.st_size; - - /* Don't bother opening empty, world readable files. Also do not open - files when archive is meant for /dev/null. */ - - if (dev_null_output - || (sizeleft == 0 - && MODE_R == (MODE_R & current_stat.st_mode))) - f = -1; - else - { - f = open (p, O_RDONLY | O_BINARY); - if (f < 0) - { - WARN ((0, errno, _("Cannot add file %s"), p)); - if (!ignore_failed_read_option) - exit_status = TAREXIT_FAILURE; - return; - } - } - - /* If the file is sparse, we've already taken care of this. */ - - if (!header_moved) - { - header = start_header (p, ¤t_stat); - if (header == NULL) - { - if (f >= 0) - close (f); - exit_status = TAREXIT_FAILURE; - return; - } - } - - /* Mark contiguous files, if we support them. */ - - if (archive_format != V7_FORMAT && S_ISCTG (current_stat.st_mode)) - header->header.typeflag = CONTTYPE; - - isextended = header->oldgnu_header.isextended; - save_typeflag = header->header.typeflag; - finish_header (header); - if (isextended) - { -#if 0 - int sum = 0; -#endif - int counter; -#if 0 - union block *exhdr; - int arraybound = SPARSES_IN_SPARSE_HEADER; -#endif - /* static */ int index_offset = SPARSES_IN_OLDGNU_HEADER; - - extend: - exhdr = find_next_block (); - - if (exhdr == NULL) - { - exit_status = TAREXIT_FAILURE; - return; - } - memset (exhdr->buffer, 0, BLOCKSIZE); - for (counter = 0; counter < SPARSES_IN_SPARSE_HEADER; counter++) - { - if (counter + index_offset > upperbound) - break; - - SIZE_TO_CHARS (sparsearray[counter + index_offset].numbytes, - exhdr->sparse_header.sp[counter].numbytes); - OFF_TO_CHARS (sparsearray[counter + index_offset].offset, - exhdr->sparse_header.sp[counter].offset); - } - set_next_block_after (exhdr); -#if 0 - sum += counter; - if (sum < upperbound) - goto extend; -#endif - if (index_offset + counter <= upperbound) - { - index_offset += counter; - exhdr->sparse_header.isextended = 1; - goto extend; - } - - } - if (save_typeflag == GNUTYPE_SPARSE) - { - if (f < 0 - || finish_sparse_file (f, &sizeleft, current_stat.st_size, p)) - goto padit; - } - else - while (sizeleft > 0) - { - if (multi_volume_option) - { - assign_string (&save_name, p); - save_sizeleft = sizeleft; - save_totsize = current_stat.st_size; - } - start = find_next_block (); - - bufsize = available_space_after (start); - - if (sizeleft < bufsize) - { - /* Last read -- zero out area beyond. */ - - bufsize = sizeleft; - count = bufsize % BLOCKSIZE; - if (count) - memset (start->buffer + sizeleft, 0, - (size_t) (BLOCKSIZE - count)); - } - if (f < 0) - count = bufsize; - else - count = safe_read (f, start->buffer, bufsize); - if (count < 0) - { - char buf[UINTMAX_STRSIZE_BOUND]; - ERROR ((0, errno, - _("Read error at byte %s, reading %lu bytes, in file %s"), - STRINGIFY_BIGINT (current_stat.st_size - sizeleft, - buf), - (unsigned long) bufsize, p)); - goto padit; - } - sizeleft -= count; - - /* This is nonportable (the type of set_next_block_after's arg). */ - - set_next_block_after (start + (count - 1) / BLOCKSIZE); - - if (count == bufsize) - continue; - else - { - char buf[UINTMAX_STRSIZE_BOUND]; - ERROR ((0, 0, - _("File %s shrunk by %s bytes, padding with zeros"), - p, STRINGIFY_BIGINT (sizeleft, buf))); - goto padit; /* short read */ - } - } - - if (multi_volume_option) - assign_string (&save_name, NULL); - - if (f >= 0) - { - struct stat final_stat; - if (fstat (f, &final_stat) != 0) - ERROR ((0, errno, "%s: fstat", p)); - else if (final_stat.st_mtime != restore_times.modtime - || final_stat.st_size != restore_size) - ERROR ((0, errno, _("%s: file changed as we read it"), p)); - if (close (f) != 0) - ERROR ((0, errno, _("%s: close"), p)); - if (atime_preserve_option) - utime (p, &restore_times); - } - if (remove_files_option) - { - if (unlink (p) == -1) - ERROR ((0, errno, _("Cannot remove %s"), p)); - } - return; - - /* File shrunk or gave error, pad out tape to match the size we - specified in the header. */ - - padit: - while (sizeleft > 0) - { - save_sizeleft = sizeleft; - start = find_next_block (); - memset (start->buffer, 0, BLOCKSIZE); - set_next_block_after (start); - sizeleft -= BLOCKSIZE; - } - if (multi_volume_option) - assign_string (&save_name, NULL); - if (f >= 0) - { - close (f); - if (atime_preserve_option) - utime (p, &restore_times); - } - return; - } - -#ifdef HAVE_READLINK - else if (S_ISLNK (current_stat.st_mode)) - { - int size; - char *buffer = (char *) alloca (PATH_MAX + 1); - - size = readlink (p, buffer, PATH_MAX + 1); - if (size < 0) - { - WARN ((0, errno, _("Cannot add file %s"), p)); - if (!ignore_failed_read_option) - exit_status = TAREXIT_FAILURE; - return; - } - buffer[size] = '\0'; - if (size >= NAME_FIELD_SIZE) - write_long (buffer, GNUTYPE_LONGLINK); - assign_string (¤t_link_name, buffer); - - current_stat.st_size = 0; /* force 0 size on symlink */ - header = start_header (p, ¤t_stat); - if (header == NULL) - { - exit_status = TAREXIT_FAILURE; - return; - } - strncpy (header->header.linkname, buffer, NAME_FIELD_SIZE); - header->header.linkname[NAME_FIELD_SIZE - 1] = '\0'; - header->header.typeflag = SYMTYPE; - finish_header (header); /* nothing more to do to it */ - if (remove_files_option) - { - if (unlink (p) == -1) - ERROR ((0, errno, _("Cannot remove %s"), p)); - } + WARN ((0, 0, _("%s: file is the archive; not dumped"), + quotearg_colon (p))); return; } #endif - else if (S_ISDIR (current_stat.st_mode)) + if (S_ISDIR (current_stat.st_mode)) { - DIR *directory; - struct dirent *entry; + char *directory; + char const *entry; + size_t entrylen; char *namebuf; size_t buflen; size_t len; dev_t our_device = current_stat.st_dev; - /* If this tar program is installed suid root, like for Amanda, the - access might look like denied, while it is not really. - - FIXME: I have the feeling this test is done too early. Couldn't it - just be bundled in later actions? I guess that the proper support - of --ignore-failed-read is the key of the current writing. */ + errno = 0; - if (access (p, R_OK) == -1 && geteuid () != 0) + directory = savedir (p, current_stat.st_size); + if (! directory) { - WARN ((0, errno, _("Cannot add directory %s"), p)); - if (!ignore_failed_read_option) - exit_status = TAREXIT_FAILURE; + if (ignore_failed_read_option) + savedir_warn (p); + else + savedir_error (p); return; } @@ -1417,16 +986,16 @@ dump_file (char *p, dev_t parent_device, int top_level) len = strlen (p); buflen = len + NAME_FIELD_SIZE; namebuf = xmalloc (buflen + 1); - strncpy (namebuf, p, buflen); + memcpy (namebuf, p, len); while (len >= 1 && namebuf[len - 1] == '/') len--; namebuf[len++] = '/'; namebuf[len] = '\0'; - if (1) + if (! is_avoided_name (namebuf)) { - /* The "1" above used to be "archive_format != V7_FORMAT", GNU tar - was just not writing directory blocks at all. Daniel Trinkle + /* The condition above used to be "archive_format != V7_FORMAT". + GNU tar was not writing directory blocks at all. Daniel Trinkle writes: ``All old versions of tar I have ever seen have correctly archived an empty directory. The really old ones I checked included HP-UX 7 and Mt. Xinu More/BSD. There may be @@ -1447,11 +1016,6 @@ dump_file (char *p, dev_t parent_device, int top_level) files, we'd better put the / on the name. */ header = start_header (namebuf, ¤t_stat); - if (header == NULL) - { - exit_status = TAREXIT_FAILURE; - return; /* eg name too long */ - } if (incremental_option) header->header.typeflag = GNUTYPE_DUMPDIR; @@ -1503,8 +1067,7 @@ dump_file (char *p, dev_t parent_device, int top_level) bufsize = sizeleft; count = bufsize % BLOCKSIZE; if (count) - memset (start->buffer + sizeleft, 0, - (size_t) (BLOCKSIZE - count)); + memset (start->buffer + sizeleft, 0, BLOCKSIZE - count); } memcpy (start->buffer, p_buffer, bufsize); sizeleft -= bufsize; @@ -1512,17 +1075,15 @@ dump_file (char *p, dev_t parent_device, int top_level) set_next_block_after (start + (bufsize - 1) / BLOCKSIZE); } if (multi_volume_option) - assign_string (&save_name, NULL); - if (atime_preserve_option) - utime (p, &restore_times); - return; + assign_string (&save_name, 0); + goto finish_dir; } /* See if we are about to recurse into a directory, and avoid doing so if the user wants that we do not descend into directories. */ if (no_recurse_option) - return; + goto finish_dir; /* See if we are crossing from one file system to another, and avoid doing so if the user only wants to dump one file system. */ @@ -1531,74 +1092,449 @@ dump_file (char *p, dev_t parent_device, int top_level) && parent_device != current_stat.st_dev) { if (verbose_option) - WARN ((0, 0, _("%s: On a different filesystem; not dumped"), p)); - return; + WARN ((0, 0, + _("%s: file is on a different filesystem; not dumped"), + quotearg_colon (p))); + goto finish_dir; } /* Now output all the files in the directory. */ - errno = 0; /* FIXME: errno should be read-only */ - - directory = opendir (p); - if (!directory) - { - ERROR ((0, errno, _("Cannot open directory %s"), p)); - return; - } - /* FIXME: Should speed this up by cd-ing into the dir. */ - while (entry = readdir (directory), entry) + for (entry = directory; + (entrylen = strlen (entry)) != 0; + entry += entrylen + 1) { - /* Skip `.', `..', and excluded file names. */ - - if (is_dot_or_dotdot (entry->d_name)) - continue; - - if ((int) NAMLEN (entry) + len >= buflen) + if (buflen <= len + entrylen) { - buflen = len + NAMLEN (entry); - namebuf = (char *) xrealloc (namebuf, buflen + 1); -#if 0 - namebuf[len] = '\0'; - ERROR ((0, 0, _("File name %s%s too long"), - namebuf, entry->d_name)); - continue; -#endif + buflen = len + entrylen; + namebuf = xrealloc (namebuf, buflen + 1); } - strcpy (namebuf + len, entry->d_name); + strcpy (namebuf + len, entry); if (!excluded_name (namebuf)) - dump_file (namebuf, our_device, 0); + dump_file (namebuf, 0, our_device); } - closedir (directory); + finish_dir: + + free (directory); free (namebuf); if (atime_preserve_option) utime (p, &restore_times); return; } - - else if (S_ISCHR (current_stat.st_mode)) - type = CHRTYPE; - else if (S_ISBLK (current_stat.st_mode)) - type = BLKTYPE; - else if (S_ISFIFO (current_stat.st_mode) - || S_ISSOCK (current_stat.st_mode)) - type = FIFOTYPE; + else if (is_avoided_name (p)) + return; else - goto unknown; + { + /* Check for multiple links. + + We maintain a table of all such files that we've written so + far. Any time we see another, we check the table and avoid + dumping the data again if we've done it once already. */ + + if (1 < current_stat.st_nlink) + { + static Hash_table *link_table; + struct link *lp = xmalloc (sizeof *lp + strlen (p)); + struct link *dup; + lp->ino = current_stat.st_ino; + lp->dev = current_stat.st_dev; + strcpy (lp->name, p); + + if (! ((link_table + || (link_table = hash_initialize (0, 0, hash_link, + compare_links, 0))) + && (dup = hash_insert (link_table, lp)))) + xalloc_die (); + + if (dup != lp) + { + /* We found a link. */ + char const *link_name = dup->name; + + free (lp); + + if (! absolute_names_option) + for (; *link_name == '/'; link_name++) + { + static int warned_once; + if (!warned_once) + { + warned_once = 1; + WARN ((0, 0, + _("Removing leading `/' from link names"))); + } + } + + if (NAME_FIELD_SIZE <= strlen (link_name)) + write_long (link_name, GNUTYPE_LONGLINK); + assign_string (¤t_link_name, link_name); + + current_stat.st_size = 0; + header = start_header (p, ¤t_stat); + strncpy (header->header.linkname, link_name, NAME_FIELD_SIZE); + + /* Force null termination. */ + header->header.linkname[NAME_FIELD_SIZE - 1] = 0; + + header->header.typeflag = LNKTYPE; + finish_header (header); + + /* FIXME: Maybe remove from table after all links found? */ + + if (remove_files_option && unlink (p) != 0) + unlink_error (p); + + /* We dumped it. */ + return; + } + } + + /* This is not a link to a previously dumped file, so dump it. */ + + if (S_ISREG (current_stat.st_mode) + || S_ISCTG (current_stat.st_mode)) + { + int f; /* file descriptor */ + size_t bufsize; + ssize_t count; + off_t sizeleft; + union block *start; + int header_moved; + char isextended = 0; + int sparses = 0; + + header_moved = 0; + + if (sparse_option) + { + /* Check the size of the file against the number of blocks + allocated for it, counting both data and indirect blocks. + If there is a smaller number of blocks that would be + necessary to accommodate a file of this size, this is safe + to say that we have a sparse file: at least one of those + blocks in the file is just a useless hole. For sparse + files not having more hole blocks than indirect blocks, the + sparseness will go undetected. */ + + /* Bruno Haible sent me these statistics for Linux. It seems + that some filesystems count indirect blocks in st_blocks, + while others do not seem to: + + minix-fs tar: size=7205, st_blocks=18 and ST_NBLOCKS=18 + extfs tar: size=7205, st_blocks=18 and ST_NBLOCKS=18 + ext2fs tar: size=7205, st_blocks=16 and ST_NBLOCKS=16 + msdos-fs tar: size=7205, st_blocks=16 and ST_NBLOCKS=16 + + Dick Streefland reports the previous numbers as misleading, + because ext2fs use 12 direct blocks, while minix-fs uses only + 6 direct blocks. Dick gets: + + ext2 size=20480 ls listed blocks=21 + minix size=20480 ls listed blocks=21 + msdos size=20480 ls listed blocks=20 + + It seems that indirect blocks *are* included in st_blocks. + The minix filesystem does not account for phantom blocks in + st_blocks, so `du' and `ls -s' give wrong results. So, the + --sparse option would not work on a minix filesystem. */ + + if (ST_NBLOCKS (current_stat) + < (current_stat.st_size / ST_NBLOCKSIZE + + (current_stat.st_size % ST_NBLOCKSIZE != 0))) + { + int counter; + + header = start_header (p, ¤t_stat); + header->header.typeflag = GNUTYPE_SPARSE; + header_moved = 1; + + /* Call the routine that figures out the layout of the + sparse file in question. SPARSES is the index of the + first unused element of the "sparsearray," i.e., + the number of elements it needed to describe the file. */ + + sparses = deal_with_sparse (p, header); + + /* See if we'll need an extended header later. */ + + if (SPARSES_IN_OLDGNU_HEADER < sparses) + header->oldgnu_header.isextended = 1; + + /* We store the "real" file size so we can show that in + case someone wants to list the archive, i.e., tar tvf + . It might be kind of disconcerting if the + shrunken file size was the one that showed up. */ + + OFF_TO_CHARS (current_stat.st_size, + header->oldgnu_header.realsize); + + /* This will be the new "size" of the file, i.e., the size + of the file minus the blocks of holes that we're + skipping over. */ + + current_stat.st_size = find_new_file_size (sparses); + OFF_TO_CHARS (current_stat.st_size, header->header.size); + + for (counter = 0; + counter < sparses && counter < SPARSES_IN_OLDGNU_HEADER; + counter++) + { + OFF_TO_CHARS (sparsearray[counter].offset, + header->oldgnu_header.sp[counter].offset); + SIZE_TO_CHARS (sparsearray[counter].numbytes, + header->oldgnu_header.sp[counter].numbytes); + } + } + } + + sizeleft = current_stat.st_size; + + /* Don't bother opening empty, world readable files. Also do not open + files when archive is meant for /dev/null. */ + + if (dev_null_output + || (sizeleft == 0 + && MODE_R == (MODE_R & current_stat.st_mode))) + f = -1; + else + { + f = open (p, O_RDONLY | O_BINARY); + if (f < 0) + { + if (! top_level && errno == ENOENT) + WARN ((0, 0, _("%s: File removed before we read it"), + quotearg_colon (p))); + else + (ignore_failed_read_option ? open_warn : open_error) (p); + return; + } + } + + /* If the file is sparse, we've already taken care of this. */ + + if (!header_moved) + header = start_header (p, ¤t_stat); + + /* Mark contiguous files, if we support them. */ + + if (archive_format != V7_FORMAT && S_ISCTG (current_stat.st_mode)) + header->header.typeflag = CONTTYPE; + + isextended = header->oldgnu_header.isextended; + save_typeflag = header->header.typeflag; + finish_header (header); + if (isextended) + { + int sparses_emitted = SPARSES_IN_OLDGNU_HEADER; + + for (;;) + { + int i; + exhdr = find_next_block (); + memset (exhdr->buffer, 0, BLOCKSIZE); + for (i = 0; + (i < SPARSES_IN_SPARSE_HEADER + && sparses_emitted + i < sparses); + i++) + { + SIZE_TO_CHARS (sparsearray[sparses_emitted + i].numbytes, + exhdr->sparse_header.sp[i].numbytes); + OFF_TO_CHARS (sparsearray[sparses_emitted + i].offset, + exhdr->sparse_header.sp[i].offset); + } + set_next_block_after (exhdr); + sparses_emitted += i; + if (sparses == sparses_emitted) + break; + exhdr->sparse_header.isextended = 1; + } + } + if (save_typeflag == GNUTYPE_SPARSE) + { + if (f < 0 + || finish_sparse_file (f, &sizeleft, + current_stat.st_size, p)) + goto padit; + } + else + while (sizeleft > 0) + { + if (multi_volume_option) + { + assign_string (&save_name, p); + save_sizeleft = sizeleft; + save_totsize = current_stat.st_size; + } + start = find_next_block (); + + bufsize = available_space_after (start); + + if (sizeleft < bufsize) + { + /* Last read -- zero out area beyond. */ + + bufsize = sizeleft; + count = bufsize % BLOCKSIZE; + if (count) + memset (start->buffer + sizeleft, 0, BLOCKSIZE - count); + } + if (f < 0) + count = bufsize; + else + count = safe_read (f, start->buffer, bufsize); + if (count < 0) + { + (ignore_failed_read_option + ? read_warn_details + : read_error_details) + (p, current_stat.st_size - sizeleft, bufsize); + goto padit; + } + sizeleft -= bufsize; + + /* This is nonportable (the type of set_next_block_after's arg). */ + + set_next_block_after (start + (bufsize - 1) / BLOCKSIZE); + + + if (count != bufsize) + { + char buf[UINTMAX_STRSIZE_BOUND]; + memset (start->buffer + count, 0, bufsize - count); + WARN ((0, 0, + _("%s: File shrank by %s bytes; padding with zeros"), + quotearg_colon (p), + STRINGIFY_BIGINT (sizeleft, buf))); + if (! ignore_failed_read_option) + exit_status = TAREXIT_FAILURE; + goto padit; /* short read */ + } + } + + if (multi_volume_option) + assign_string (&save_name, 0); + + if (f >= 0) + { + struct stat final_stat; + if (fstat (f, &final_stat) != 0) + { + if (ignore_failed_read_option) + stat_warn (p); + else + stat_error (p); + } + else if (final_stat.st_ctime != original_ctime) + { + char const *qp = quotearg_colon (p); + WARN ((0, 0, _("%s: file changed as we read it"), qp)); + if (! ignore_failed_read_option) + exit_status = TAREXIT_FAILURE; + } + if (close (f) != 0) + { + if (ignore_failed_read_option) + close_warn (p); + else + close_error (p); + } + if (atime_preserve_option) + utime (p, &restore_times); + } + if (remove_files_option) + { + if (unlink (p) == -1) + unlink_error (p); + } + return; + + /* File shrunk or gave error, pad out tape to match the size we + specified in the header. */ + + padit: + while (sizeleft > 0) + { + save_sizeleft = sizeleft; + start = find_next_block (); + memset (start->buffer, 0, BLOCKSIZE); + set_next_block_after (start); + sizeleft -= BLOCKSIZE; + } + if (multi_volume_option) + assign_string (&save_name, 0); + if (f >= 0) + { + close (f); + if (atime_preserve_option) + utime (p, &restore_times); + } + return; + } +#ifdef HAVE_READLINK + else if (S_ISLNK (current_stat.st_mode)) + { + char *buffer; + int size; + size_t linklen = current_stat.st_size; + if (linklen != current_stat.st_size || linklen + 1 == 0) + xalloc_die (); + buffer = (char *) alloca (linklen + 1); + size = readlink (p, buffer, linklen + 1); + if (size < 0) + { + if (ignore_failed_read_option) + readlink_warn (p); + else + readlink_error (p); + return; + } + buffer[size] = '\0'; + if (size >= NAME_FIELD_SIZE) + write_long (buffer, GNUTYPE_LONGLINK); + assign_string (¤t_link_name, buffer); + + current_stat.st_size = 0; /* force 0 size on symlink */ + header = start_header (p, ¤t_stat); + strncpy (header->header.linkname, buffer, NAME_FIELD_SIZE); + header->header.linkname[NAME_FIELD_SIZE - 1] = '\0'; + header->header.typeflag = SYMTYPE; + finish_header (header); /* nothing more to do to it */ + if (remove_files_option) + { + if (unlink (p) == -1) + unlink_error (p); + } + return; + } +#endif + else if (S_ISCHR (current_stat.st_mode)) + type = CHRTYPE; + else if (S_ISBLK (current_stat.st_mode)) + type = BLKTYPE; + else if (S_ISFIFO (current_stat.st_mode)) + type = FIFOTYPE; + else if (S_ISSOCK (current_stat.st_mode)) + { + WARN ((0, 0, _("%s: socket ignored"), quotearg_colon (p))); + return; + } + else if (S_ISDOOR (current_stat.st_mode)) + { + WARN ((0, 0, _("%s: door ignored"), quotearg_colon (p))); + return; + } + else + goto unknown; + } if (archive_format == V7_FORMAT) goto unknown; current_stat.st_size = 0; /* force 0 size */ header = start_header (p, ¤t_stat); - if (header == NULL) - { - exit_status = TAREXIT_FAILURE; - return; /* eg name too long */ - } - header->header.typeflag = type; if (type != FIFOTYPE) @@ -1611,10 +1547,13 @@ dump_file (char *p, dev_t parent_device, int top_level) if (remove_files_option) { if (unlink (p) == -1) - ERROR ((0, errno, _("Cannot remove %s"), p)); + unlink_error (p); } return; unknown: - ERROR ((0, 0, _("%s: Unknown file type; file ignored"), p)); + WARN ((0, 0, _("%s: Unknown file type; file ignored"), + quotearg_colon (p))); + if (! ignore_failed_read_option) + exit_status = TAREXIT_FAILURE; }