X-Git-Url: https://git.dogcows.com/gitweb?a=blobdiff_plain;f=src%2Fcreate.c;h=5e2171b561917a74c2bde424a9eb7e0648c970a0;hb=8da503cad6e883b30c05749149084d24319063b4;hp=910b4a239d684180b0ebeb4c4132495331737935;hpb=91b2d65e9dccae981f308a659d6c7bdd32285598;p=chaz%2Ftar diff --git a/src/create.c b/src/create.c index 910b4a2..5e2171b 100644 --- a/src/create.c +++ b/src/create.c @@ -1,13 +1,13 @@ /* Create a tar archive. Copyright (C) 1985, 1992, 1993, 1994, 1996, 1997, 1999, 2000, 2001, - 2003 Free Software Foundation, Inc. + 2003, 2004, 2005, 2006, 2007, 2009, 2010 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 under the terms of the GNU General Public License as published by the - Free Software Foundation; either version 2, or (at your option) any later + Free Software Foundation; either version 3, or (at your option) any later version. This program is distributed in the hope that it will be useful, but @@ -17,42 +17,108 @@ You should have received a copy of the GNU General Public License along with this program; if not, write to the Free Software Foundation, Inc., - 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. */ + 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA. */ -#include "system.h" - -#if !MSDOS -# include -# include -#endif - -#if HAVE_UTIME_H -# include -#else -struct utimbuf - { - long actime; - long modtime; - }; -#endif +#include #include #include "common.h" #include -#ifndef MSDOS -extern dev_t ar_dev; -extern ino_t ar_ino; -#endif +/* Error number to use when an impostor is discovered. + Pretend the impostor isn't there. */ +enum { IMPOSTOR_ERRNO = ENOENT }; struct link { dev_t dev; ino_t ino; - size_t nlink; + nlink_t nlink; char name[1]; }; + +struct exclusion_tag +{ + const char *name; + size_t length; + enum exclusion_tag_type type; + bool (*predicate) (int fd); + struct exclusion_tag *next; +}; + +static struct exclusion_tag *exclusion_tags; + +void +add_exclusion_tag (const char *name, enum exclusion_tag_type type, + bool (*predicate) (int fd)) +{ + struct exclusion_tag *tag = xmalloc (sizeof tag[0]); + tag->next = exclusion_tags; + tag->name = name; + tag->type = type; + tag->predicate = predicate; + tag->length = strlen (name); + exclusion_tags = tag; +} + +void +exclusion_tag_warning (const char *dirname, const char *tagname, + const char *message) +{ + if (verbose_option) + WARNOPT (WARN_CACHEDIR, + (0, 0, + _("%s: contains a cache directory tag %s; %s"), + quotearg_colon (dirname), + quotearg_n (1, tagname), + message)); +} + +enum exclusion_tag_type +check_exclusion_tags (struct tar_stat_info const *st, char const **tag_file_name) +{ + struct exclusion_tag *tag; + + for (tag = exclusion_tags; tag; tag = tag->next) + { + int tagfd = subfile_open (st, tag->name, open_read_flags); + if (0 <= tagfd) + { + bool satisfied = !tag->predicate || tag->predicate (tagfd); + close (tagfd); + if (satisfied) + { + if (tag_file_name) + *tag_file_name = tag->name; + return tag->type; + } + } + } + + return exclusion_tag_none; +} + +/* Exclusion predicate to test if the named file (usually "CACHEDIR.TAG") + contains a valid header, as described at: + http://www.brynosaurus.com/cachedir + Applications can write this file into directories they create + for use as caches containing purely regenerable, non-precious data, + allowing us to avoid archiving them if --exclude-caches is specified. */ + +#define CACHEDIR_SIGNATURE "Signature: 8a477f597d28d172789f06886806bc55" +#define CACHEDIR_SIGNATURE_SIZE (sizeof CACHEDIR_SIGNATURE - 1) + +bool +cachedir_file_p (int fd) +{ + char tagbuf[CACHEDIR_SIGNATURE_SIZE]; + + return + (read (fd, tagbuf, CACHEDIR_SIGNATURE_SIZE) == CACHEDIR_SIGNATURE_SIZE + && memcmp (tagbuf, CACHEDIR_SIGNATURE, CACHEDIR_SIGNATURE_SIZE) == 0); +} + /* The maximum uintmax_t value that can be represented with DIGITS digits, assuming that each digit is BITS_PER_DIGIT wide. */ @@ -61,6 +127,10 @@ struct link ? ((uintmax_t) 1 << ((digits) * (bits_per_digit))) - 1 \ : (uintmax_t) -1) +/* The maximum uintmax_t value that can be represented with octal + digits and a trailing NUL in BUFFER. */ +#define MAX_OCTAL_VAL(buffer) MAX_VAL_WITH_DIGITS (sizeof (buffer) - 1, LG_8) + /* 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. */ @@ -79,6 +149,29 @@ to_octal (uintmax_t value, char *where, size_t size) while (i); } +/* Copy at most LEN bytes from the string SRC to DST. Terminate with + NUL unless SRC is LEN or more bytes long. */ + +static void +tar_copy_str (char *dst, const char *src, size_t len) +{ + size_t i; + for (i = 0; i < len; i++) + if (! (dst[i] = src[i])) + break; +} + +/* Same as tar_copy_str, but always terminate with NUL if using + is OLDGNU format */ + +static void +tar_name_copy_str (char *dst, const char *src, size_t len) +{ + tar_copy_str (dst, src, len); + if (archive_format == OLDGNU_FORMAT) + dst[len-1] = 0; +} + /* 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. @@ -101,6 +194,81 @@ to_base256 (int negative, uintmax_t value, char *where, size_t size) while (i); } +#define GID_TO_CHARS(val, where) gid_to_chars (val, where, sizeof (where)) +#define MAJOR_TO_CHARS(val, where) major_to_chars (val, where, sizeof (where)) +#define MINOR_TO_CHARS(val, where) minor_to_chars (val, where, sizeof (where)) +#define MODE_TO_CHARS(val, where) mode_to_chars (val, where, sizeof (where)) +#define UID_TO_CHARS(val, where) uid_to_chars (val, where, sizeof (where)) + +#define UNAME_TO_CHARS(name,buf) string_to_chars (name, buf, sizeof(buf)) +#define GNAME_TO_CHARS(name,buf) string_to_chars (name, buf, sizeof(buf)) + +static bool +to_chars (int negative, uintmax_t value, size_t valsize, + uintmax_t (*substitute) (int *), + char *where, size_t size, const char *type); + +static bool +to_chars_subst (int negative, int gnu_format, uintmax_t value, size_t valsize, + uintmax_t (*substitute) (int *), + char *where, size_t size, const char *type) +{ + uintmax_t maxval = (gnu_format + ? MAX_VAL_WITH_DIGITS (size - 1, LG_256) + : MAX_VAL_WITH_DIGITS (size - 1, LG_8)); + 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 (gnu_format) + { + 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) + { + 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; + /* NOTE: This is one of the few places where GNU_FORMAT differs from + OLDGNU_FORMAT. The actual differences are: + + 1. In OLDGNU_FORMAT all strings in a tar header end in \0 + 2. Incremental archives use oldgnu_header. + + Apart from this they are completely identical. */ + 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, _("value %s out of %s range %s..%s; substituting %s"), + value_string, type, minval_string, maxval_string, + sub_string)); + return to_chars (negsub, s, valsize, 0, where, size, type); + } + else + ERROR ((0, 0, _("value %s out of %s range %s..%s"), + value_string, type, minval_string, maxval_string)); + return false; +} + /* 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 @@ -115,101 +283,61 @@ to_base256 (int negative, uintmax_t value, char *where, size_t size) SUBSTITUTE the address of an 0-or-1 flag recording whether the substitute value is negative. */ -static void +static bool to_chars (int negative, uintmax_t value, size_t valsize, uintmax_t (*substitute) (int *), char *where, size_t size, const char *type) { - int base256_allowed = (archive_format == GNU_FORMAT - || archive_format == OLDGNU_FORMAT); + int gnu_format = (archive_format == GNU_FORMAT + || archive_format == OLDGNU_FORMAT); /* 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_octal (value, where, size - 1); + return true; } - - /* 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 ? -1 : 1 << (LG_256 - 1); - to_base256 (negative, value, where + 1, size - 1); - } - - /* 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_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 + else if (gnu_format) { - uintmax_t maxval = (base256_allowed - ? MAX_VAL_WITH_DIGITS (size - 1, LG_256) - : MAX_VAL_WITH_DIGITS (size - 1, LG_8)); - 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"; + /* Try to cope with the number by using traditional GNU format + methods */ - if (negative) + /* Generate the base-256 representation if the number fits. */ + if (((negative ? -1 - value : value) + <= MAX_VAL_WITH_DIGITS (size - 1, LG_256))) { - char *p = STRINGIFY_BIGINT (- value, valbuf + 1); - *--p = '-'; - value_string = p; + where[0] = negative ? -1 : 1 << (LG_256 - 1); + to_base256 (negative, value, where + 1, size - 1); + return true; } - else - value_string = STRINGIFY_BIGINT (value, valbuf); - if (substitute) + /* 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) { - int negsub; - uintmax_t sub = substitute (&negsub) & maxval; - 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, _("value %s out of %s range %s..%s; substituting %s"), - value_string, type, minval_string, maxval_string, - sub_string)); - to_chars (negsub, s, valsize, 0, where, size, type); + static int warned_once; + if (! warned_once) + { + warned_once = 1; + WARN ((0, 0, _("Generating negative octal headers"))); + } + where[size - 1] = '\0'; + to_octal (value & MAX_VAL_WITH_DIGITS (valsize * CHAR_BIT, 1), + where, size - 1); + return true; } - else - ERROR ((0, 0, _("value %s out of %s range %s..%s"), - value_string, type, minval_string, maxval_string)); + /* Otherwise fall back to substitution, if possible: */ } + else + substitute = NULL; /* No substitution for formats, other than GNU */ + + return to_chars_subst (negative, gnu_format, value, valsize, substitute, + where, size, type); } static uintmax_t @@ -228,25 +356,25 @@ gid_substitute (int *negative) return r; } -void +static bool gid_to_chars (gid_t v, char *p, size_t s) { - to_chars (v < 0, (uintmax_t) v, sizeof v, gid_substitute, p, s, "gid_t"); + return to_chars (v < 0, (uintmax_t) v, sizeof v, gid_substitute, p, s, "gid_t"); } -void +static bool major_to_chars (major_t v, char *p, size_t s) { - to_chars (v < 0, (uintmax_t) v, sizeof v, 0, p, s, "major_t"); + return to_chars (v < 0, (uintmax_t) v, sizeof v, 0, p, s, "major_t"); } -void +static bool minor_to_chars (minor_t v, char *p, size_t s) { - to_chars (v < 0, (uintmax_t) v, sizeof v, 0, p, s, "minor_t"); + return to_chars (v < 0, (uintmax_t) v, sizeof v, 0, p, s, "minor_t"); } -void +static bool 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, @@ -261,6 +389,7 @@ mode_to_chars (mode_t v, char *p, size_t s) && S_IRGRP == TGREAD && S_IWGRP == TGWRITE && S_IXGRP == TGEXEC && S_IROTH == TOREAD && S_IWOTH == TOWRITE && S_IXOTH == TOEXEC && archive_format != POSIX_FORMAT + && archive_format != USTAR_FORMAT && archive_format != GNU_FORMAT) { negative = v < 0; @@ -282,25 +411,19 @@ 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, 0, p, s, "mode_t"); + return to_chars (negative, u, sizeof v, 0, p, s, "mode_t"); } -void +bool off_to_chars (off_t v, char *p, size_t s) { - 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, 0, p, s, "size_t"); + return to_chars (v < 0, (uintmax_t) v, sizeof v, 0, p, s, "off_t"); } -void +bool time_to_chars (time_t v, char *p, size_t s) { - to_chars (v < 0, (uintmax_t) v, sizeof v, 0, p, s, "time_t"); + return to_chars (v < 0, (uintmax_t) v, sizeof v, 0, p, s, "time_t"); } static uintmax_t @@ -319,36 +442,50 @@ uid_substitute (int *negative) return r; } -void +static bool uid_to_chars (uid_t v, char *p, size_t s) { - to_chars (v < 0, (uintmax_t) v, sizeof v, uid_substitute, p, s, "uid_t"); + return to_chars (v < 0, (uintmax_t) v, sizeof v, uid_substitute, p, s, "uid_t"); } -void +static bool uintmax_to_chars (uintmax_t v, char *p, size_t s) { - to_chars (0, v, sizeof v, 0, p, s, "uintmax_t"); + return to_chars (0, v, sizeof v, 0, p, s, "uintmax_t"); } -void -string_to_chars (char *str, char *p, size_t s) +static void +string_to_chars (char const *str, char *p, size_t s) { - strncpy (p, str, s); - p[s-1] = 0; + tar_copy_str (p, str, s); + p[s - 1] = '\0'; } -/* Writing routines. */ +/* A directory is always considered dumpable. + Otherwise, only regular and contiguous files are considered dumpable. + Such a file is dumpable if it is sparse and both --sparse and --totals + are specified. + Otherwise, it is dumpable unless any of the following conditions occur: -/* Zero out the buffer so we don't confuse ourselves with leftover - data. */ -static void -clear_buffer (char *buffer) + a) it is empty *and* world-readable, or + b) current archive is /dev/null */ + +static bool +file_dumpable_p (struct stat const *st) { - memset (buffer, 0, BLOCKSIZE); + if (S_ISDIR (st->st_mode)) + return true; + if (! (S_ISREG (st->st_mode) || S_ISCTG (st->st_mode))) + return false; + if (dev_null_output) + return totals_option && sparse_option && ST_IS_SPARSE (*st); + return ! (st->st_size == 0 && (st->st_mode & MODE_R) == MODE_R); } + +/* Writing routines. */ + /* 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. */ @@ -363,26 +500,71 @@ write_eot (void) set_next_block_after (pointer); } -/* Write a GNUTYPE_LONGLINK or GNUTYPE_LONGNAME block. */ +/* Write a "private" header */ +union block * +start_private_header (const char *name, size_t size, time_t t) +{ + union block *header = find_next_block (); + + memset (header->buffer, 0, sizeof (union block)); + + tar_name_copy_str (header->header.name, name, NAME_FIELD_SIZE); + OFF_TO_CHARS (size, header->header.size); + + TIME_TO_CHARS (t, header->header.mtime); + MODE_TO_CHARS (S_IFREG|S_IRUSR|S_IWUSR|S_IRGRP|S_IROTH, header->header.mode); + UID_TO_CHARS (getuid (), header->header.uid); + GID_TO_CHARS (getgid (), header->header.gid); + MAJOR_TO_CHARS (0, header->header.devmajor); + MINOR_TO_CHARS (0, header->header.devminor); + strncpy (header->header.magic, TMAGIC, TMAGLEN); + strncpy (header->header.version, TVERSION, TVERSLEN); + return header; +} -/* FIXME: Cross recursion between start_header and write_long! */ +/* Create a new header and store there at most NAME_FIELD_SIZE bytes of + the file name */ -static union block *start_header (const char *, struct tar_stat_info *); +static union block * +write_short_name (struct tar_stat_info *st) +{ + union block *header = find_next_block (); + memset (header->buffer, 0, sizeof (union block)); + tar_name_copy_str (header->header.name, st->file_name, NAME_FIELD_SIZE); + return header; +} + +#define FILL(field,byte) do { \ + memset(field, byte, sizeof(field)-1); \ + (field)[sizeof(field)-1] = 0; \ +} while (0) +/* Write a GNUTYPE_LONGLINK or GNUTYPE_LONGNAME block. */ static void -write_long (const char *p, char type) +write_gnu_long_link (struct tar_stat_info *st, const char *p, char type) { size_t size = strlen (p) + 1; size_t bufsize; union block *header; - struct tar_stat_info foo; - - memset (&foo, 0, sizeof foo); - foo.stat.st_size = size; - - header = start_header ("././@LongLink", &foo); + char *tmpname; + + header = start_private_header ("././@LongLink", size, time (NULL)); + FILL (header->header.mtime, '0'); + FILL (header->header.mode, '0'); + FILL (header->header.uid, '0'); + FILL (header->header.gid, '0'); + FILL (header->header.devmajor, 0); + FILL (header->header.devminor, 0); + uid_to_uname (0, &tmpname); + UNAME_TO_CHARS (tmpname, header->header.uname); + free (tmpname); + gid_to_gname (0, &tmpname); + GNAME_TO_CHARS (tmpname, header->header.gname); + free (tmpname); + + strcpy (header->header.magic, OLDGNU_MAGIC); header->header.typeflag = type; - finish_header (header, -1); + finish_header (st, header, -1); header = find_next_block (); @@ -402,100 +584,177 @@ write_long (const char *p, char type) set_next_block_after (header + (size - 1) / BLOCKSIZE); } +static size_t +split_long_name (const char *name, size_t length) +{ + size_t i; + + if (length > PREFIX_FIELD_SIZE + 1) + length = PREFIX_FIELD_SIZE + 1; + else if (ISSLASH (name[length - 1])) + length--; + for (i = length - 1; i > 0; i--) + if (ISSLASH (name[i])) + break; + return i; +} + +static union block * +write_ustar_long_name (const char *name) +{ + size_t length = strlen (name); + size_t i, nlen; + union block *header; + + if (length > PREFIX_FIELD_SIZE + NAME_FIELD_SIZE + 1) + { + ERROR ((0, 0, _("%s: file name is too long (max %d); not dumped"), + quotearg_colon (name), + PREFIX_FIELD_SIZE + NAME_FIELD_SIZE + 1)); + return NULL; + } + + i = split_long_name (name, length); + if (i == 0 || (nlen = length - i - 1) > NAME_FIELD_SIZE || nlen == 0) + { + ERROR ((0, 0, + _("%s: file name is too long (cannot be split); not dumped"), + quotearg_colon (name))); + return NULL; + } + + header = find_next_block (); + memset (header->buffer, 0, sizeof (header->buffer)); + memcpy (header->header.prefix, name, i); + memcpy (header->header.name, name + i + 1, length - i - 1); + + return header; +} + /* Write a long link name, depending on the current archive format */ static void write_long_link (struct tar_stat_info *st) { - if (archive_format == POSIX_FORMAT) - xheader_store ("linkpath", st); - else - write_long (st->link_name, GNUTYPE_LONGNAME); -} + switch (archive_format) + { + case POSIX_FORMAT: + xheader_store ("linkpath", st, NULL); + break; + + case V7_FORMAT: /* old V7 tar format */ + case USTAR_FORMAT: + case STAR_FORMAT: + ERROR ((0, 0, + _("%s: link name is too long; not dumped"), + quotearg_colon (st->link_name))); + break; + + case OLDGNU_FORMAT: + case GNU_FORMAT: + write_gnu_long_link (st, st->link_name, GNUTYPE_LONGLINK); + break; -/* NOTE: Cross recursion between start_header and write_extended */ + default: + abort(); /*FIXME*/ + } +} static union block * -write_extended (union block *old_header, char type) +write_long_name (struct tar_stat_info *st) { - union block *header, hp; - struct tar_stat_info foo; - size_t size; - char *p; + switch (archive_format) + { + case POSIX_FORMAT: + xheader_store ("path", st, NULL); + break; + + case V7_FORMAT: + if (strlen (st->file_name) > NAME_FIELD_SIZE-1) + { + ERROR ((0, 0, _("%s: file name is too long (max %d); not dumped"), + quotearg_colon (st->file_name), + NAME_FIELD_SIZE - 1)); + return NULL; + } + break; - if (extended_header.buffer || extended_header.stk == NULL) - return old_header; /* Prevent recursion */ - - xheader_finish (&extended_header); - size = extended_header.size; - memset (&foo, 0, sizeof foo); - foo.stat.st_mode = S_IFREG|S_IRUSR|S_IWUSR|S_IRGRP|S_IROTH; - time (&foo.stat.st_ctime); - foo.stat.st_atime = foo.stat.st_ctime; - foo.stat.st_mtime = foo.stat.st_ctime; - foo.stat.st_size = size; + case USTAR_FORMAT: + case STAR_FORMAT: + return write_ustar_long_name (st->file_name); - memcpy (hp.buffer, old_header, sizeof (hp)); - - header = start_header ("././@PaxHeader", &foo); - header->header.typeflag = type; + case OLDGNU_FORMAT: + case GNU_FORMAT: + write_gnu_long_link (st, st->file_name, GNUTYPE_LONGNAME); + break; + + default: + abort(); /*FIXME*/ + } + return write_short_name (st); +} - finish_header (header, -1); +union block * +write_extended (bool global, struct tar_stat_info *st, union block *old_header) +{ + union block *header, hp; + char *p; + int type; + time_t t; - p = extended_header.buffer; + if (st->xhdr.buffer || st->xhdr.stk == NULL) + return old_header; - do + xheader_finish (&st->xhdr); + memcpy (hp.buffer, old_header, sizeof (hp)); + if (global) { - size_t len; - - header = find_next_block (); - len = BLOCKSIZE; - if (len > size) - len = size; - memcpy (header->buffer, p, len); - if (len < BLOCKSIZE) - memset (header->buffer + len, 0, BLOCKSIZE - len); - p += len; - size -= len; - set_next_block_after (header); + type = XGLTYPE; + p = xheader_ghdr_name (); + time (&t); } - while (size > 0); - - xheader_destroy (&extended_header); + else + { + type = XHDTYPE; + p = xheader_xhdr_name (st); + t = st->stat.st_mtime; + } + xheader_write (type, p, t, &st->xhdr); + free (p); header = find_next_block (); memcpy (header, &hp.buffer, sizeof (hp.buffer)); return header; } +static union block * +write_header_name (struct tar_stat_info *st) +{ + if (archive_format == POSIX_FORMAT && !string_ascii_p (st->file_name)) + { + xheader_store ("path", st, NULL); + return write_short_name (st); + } + else if (NAME_FIELD_SIZE - (archive_format == OLDGNU_FORMAT) + < strlen (st->file_name)) + return write_long_name (st); + else + return write_short_name (st); +} + /* Header handling. */ /* 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 tar_stat_info *st) +union block * +start_header (struct tar_stat_info *st) { union block *header; - name = safer_name_suffix (name, 0); - assign_string (&st->file_name, name); - - if (sizeof header->header.name <= strlen (name)) - { - if (archive_format == POSIX_FORMAT) - xheader_store ("path", st); - else - write_long (name, GNUTYPE_LONGNAME); - } - - header = find_next_block (); - memset (header->buffer, 0, sizeof (union block)); - - assign_string (¤t_stat_info.file_name, name); + header = write_header_name (st); + if (!header) + return NULL; - strncpy (header->header.name, name, NAME_FIELD_SIZE); - header->header.name[NAME_FIELD_SIZE - 1] = '\0'; - /* Override some stat fields, if requested to do so. */ if (owner_option != (uid_t) -1) @@ -503,8 +762,10 @@ start_header (const char *name, struct tar_stat_info *st) if (group_option != (gid_t) -1) st->stat.st_gid = group_option; if (mode_option) - st->stat.st_mode = ((st->stat.st_mode & ~MODE_ALL) - | mode_adjust (st->stat.st_mode, mode_option)); + st->stat.st_mode = + ((st->stat.st_mode & ~MODE_ALL) + | mode_adjust (st->stat.st_mode, S_ISDIR (st->stat.st_mode) != 0, + initial_umask, mode_option, NULL)); /* Paul Eggert tried the trivial test ($WRITER cf a b; $READER tvf a) for a few tars and came up with the following interoperability @@ -529,61 +790,103 @@ start_header (const char *name, struct tar_stat_info *st) above, thus making GNU tar both a universal donor and a universal acceptor for Paul's test. */ - if (archive_format == V7_FORMAT) + if (archive_format == V7_FORMAT || archive_format == USTAR_FORMAT) MODE_TO_CHARS (st->stat.st_mode & MODE_ALL, header->header.mode); else MODE_TO_CHARS (st->stat.st_mode, header->header.mode); - if (st->stat.st_uid > MAXOCTAL7 && archive_format == POSIX_FORMAT) - xheader_store ("uid", st); - else - UID_TO_CHARS (st->stat.st_uid, header->header.uid); - - if (st->stat.st_gid > MAXOCTAL7 && archive_format == POSIX_FORMAT) - xheader_store ("gid", st); - else - GID_TO_CHARS (st->stat.st_gid, header->header.gid); + { + uid_t uid = st->stat.st_uid; + if (archive_format == POSIX_FORMAT + && MAX_OCTAL_VAL (header->header.uid) < uid) + { + xheader_store ("uid", st, NULL); + uid = 0; + } + if (!UID_TO_CHARS (uid, header->header.uid)) + return NULL; + } - if (st->stat.st_size > MAXOCTAL11 && archive_format == POSIX_FORMAT) - xheader_store ("size", st); - else - OFF_TO_CHARS (st->stat.st_size, header->header.size); + { + gid_t gid = st->stat.st_gid; + if (archive_format == POSIX_FORMAT + && MAX_OCTAL_VAL (header->header.gid) < gid) + { + xheader_store ("gid", st, NULL); + gid = 0; + } + if (!GID_TO_CHARS (gid, header->header.gid)) + return NULL; + } + + { + off_t size = st->stat.st_size; + if (archive_format == POSIX_FORMAT + && MAX_OCTAL_VAL (header->header.size) < size) + { + xheader_store ("size", st, NULL); + size = 0; + } + if (!OFF_TO_CHARS (size, header->header.size)) + return NULL; + } - TIME_TO_CHARS (st->stat.st_mtime, header->header.mtime); + { + struct timespec mtime = set_mtime_option ? mtime_option : st->mtime; + if (archive_format == POSIX_FORMAT) + { + if (MAX_OCTAL_VAL (header->header.mtime) < mtime.tv_sec + || mtime.tv_nsec != 0) + xheader_store ("mtime", st, &mtime); + if (MAX_OCTAL_VAL (header->header.mtime) < mtime.tv_sec) + mtime.tv_sec = 0; + } + if (!TIME_TO_CHARS (mtime.tv_sec, header->header.mtime)) + return NULL; + } /* FIXME */ if (S_ISCHR (st->stat.st_mode) || S_ISBLK (st->stat.st_mode)) { - st->devmajor = major (st->stat.st_rdev); - st->devminor = minor (st->stat.st_rdev); + major_t devmajor = major (st->stat.st_rdev); + minor_t devminor = minor (st->stat.st_rdev); - if (st->devmajor > MAXOCTAL7 && archive_format == POSIX_FORMAT) - xheader_store ("devmajor", st); - else - MAJOR_TO_CHARS (st->devmajor, header->header.devmajor); + if (archive_format == POSIX_FORMAT + && MAX_OCTAL_VAL (header->header.devmajor) < devmajor) + { + xheader_store ("devmajor", st, NULL); + devmajor = 0; + } + if (!MAJOR_TO_CHARS (devmajor, header->header.devmajor)) + return NULL; - if (st->devminor > MAXOCTAL7 && archive_format == POSIX_FORMAT) - xheader_store ("devminor", st); - else - MAJOR_TO_CHARS (st->devminor, header->header.devminor); + if (archive_format == POSIX_FORMAT + && MAX_OCTAL_VAL (header->header.devminor) < devminor) + { + xheader_store ("devminor", st, NULL); + devminor = 0; + } + if (!MINOR_TO_CHARS (devminor, header->header.devminor)) + return NULL; } - else + else if (archive_format != GNU_FORMAT && archive_format != OLDGNU_FORMAT) { - MAJOR_TO_CHARS (0, header->header.devmajor); - MINOR_TO_CHARS (0, header->header.devminor); + if (!(MAJOR_TO_CHARS (0, header->header.devmajor) + && MINOR_TO_CHARS (0, header->header.devminor))) + return NULL; } - + if (archive_format == POSIX_FORMAT) { - xheader_store ("atime", st); - xheader_store ("ctime", st); + xheader_store ("atime", st, NULL); + xheader_store ("ctime", st, NULL); } else if (incremental_option) - if (archive_format == OLDGNU_FORMAT) + if (archive_format == OLDGNU_FORMAT || archive_format == GNU_FORMAT) { - TIME_TO_CHARS (st->stat.st_atime, header->oldgnu_header.atime); - TIME_TO_CHARS (st->stat.st_ctime, header->oldgnu_header.ctime); + TIME_TO_CHARS (st->atime.tv_sec, header->oldgnu_header.atime); + TIME_TO_CHARS (st->ctime.tv_sec, header->oldgnu_header.ctime); } header->header.typeflag = archive_format == V7_FORMAT ? AREGTYPE : REGTYPE; @@ -594,12 +897,13 @@ start_header (const char *name, struct tar_stat_info *st) break; case OLDGNU_FORMAT: + case GNU_FORMAT: /*FIXME?*/ /* Overwrite header->header.magic and header.version in one blow. */ strcpy (header->header.magic, OLDGNU_MAGIC); break; case POSIX_FORMAT: - case GNU_FORMAT: + case USTAR_FORMAT: strncpy (header->header.magic, TMAGIC, TMAGLEN); strncpy (header->header.version, TVERSION, TVERSLEN); break; @@ -616,52 +920,30 @@ start_header (const char *name, struct tar_stat_info *st) { uid_to_uname (st->stat.st_uid, &st->uname); gid_to_gname (st->stat.st_gid, &st->gname); - + if (archive_format == POSIX_FORMAT - && strlen (st->uname) > UNAME_FIELD_SIZE) - xheader_store ("uname", st); - else - UNAME_TO_CHARS (st->uname, header->header.uname); + && (strlen (st->uname) > UNAME_FIELD_SIZE + || !string_ascii_p (st->uname))) + xheader_store ("uname", st, NULL); + UNAME_TO_CHARS (st->uname, header->header.uname); if (archive_format == POSIX_FORMAT - && strlen (st->gname) > GNAME_FIELD_SIZE) - xheader_store ("gname", st); - else - GNAME_TO_CHARS (st->gname, header->header.gname); + && (strlen (st->gname) > GNAME_FIELD_SIZE + || !string_ascii_p (st->gname))) + xheader_store ("gname", st, NULL); + GNAME_TO_CHARS (st->gname, header->header.gname); } 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. If BLOCK_ORDINAL - is not negative, is the block ordinal of the first record for this - file, which may be a preceding long name or long link record. */ void -finish_header (union block *header, off_t block_ordinal) +simple_finish_header (union block *header) { size_t i; int sum; char *p; - /* Note: It is important to do this before the call to write_extended(), - so that the actual ustar header is printed */ - if (verbose_option - && header->header.typeflag != GNUTYPE_LONGLINK - && header->header.typeflag != GNUTYPE_LONGNAME - && header->header.typeflag != XHDTYPE - && header->header.typeflag != XGLTYPE) - { - /* These globals are parameters to print_header, sigh. */ - - current_header = header; - /* current_stat_info is already set up. */ - current_format = archive_format; - print_header (block_ordinal); - } - - header = write_extended (header, XHDTYPE); - memcpy (header->header.chksum, CHKBLANKS, sizeof header->header.chksum); sum = 0; @@ -684,226 +966,386 @@ finish_header (union block *header, off_t block_ordinal) 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. */ -static int -zero_block_p (char *buffer) +/* 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. If BLOCK_ORDINAL + is not negative, is the block ordinal of the first record for this + file, which may be a preceding long name or long link record. */ +void +finish_header (struct tar_stat_info *st, + union block *header, off_t block_ordinal) { - int counter; + /* Note: It is important to do this before the call to write_extended(), + so that the actual ustar header is printed */ + if (verbose_option + && header->header.typeflag != GNUTYPE_LONGLINK + && header->header.typeflag != GNUTYPE_LONGNAME + && header->header.typeflag != XHDTYPE + && header->header.typeflag != XGLTYPE) + { + /* FIXME: This global is used in print_header, sigh. */ + current_format = archive_format; + print_header (st, header, block_ordinal); + } - for (counter = 0; counter < BLOCKSIZE; counter++) - if (buffer[counter] != '\0') - return 0; - return 1; + header = write_extended (false, st, header); + simple_finish_header (header); } + void -init_sparsearray (void) +pad_archive (off_t size_left) { - if (! sp_array_size) - sp_array_size = SPARSES_IN_OLDGNU_HEADER; - sparsearray = xmalloc (sp_array_size * sizeof *sparsearray); + union block *blk; + while (size_left > 0) + { + blk = find_next_block (); + memset (blk->buffer, 0, BLOCKSIZE); + set_next_block_after (blk); + size_left -= BLOCKSIZE; + } } -static off_t -find_new_file_size (int sparses) +static enum dump_status +dump_regular_file (int fd, struct tar_stat_info *st) { - int i; - off_t s = 0; - for (i = 0; i < sparses; i++) - s += sparsearray[i].numbytes; - return s; -} + off_t size_left = st->stat.st_size; + off_t block_ordinal; + union block *blk; + + block_ordinal = current_block_ordinal (); + blk = start_header (st); + if (!blk) + return dump_status_fail; + + /* Mark contiguous files, if we support them. */ + if (archive_format != V7_FORMAT && S_ISCTG (st->stat.st_mode)) + blk->header.typeflag = CONTTYPE; + + finish_header (st, blk, block_ordinal); + + mv_begin_write (st->file_name, st->stat.st_size, st->stat.st_size); + while (size_left > 0) + { + size_t bufsize, count; -/* 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. */ + blk = find_next_block (); -/* There is little point in trimming small amounts of null data at the head - and tail of blocks, only avoid dumping full null blocks. */ + bufsize = available_space_after (blk); -/* FIXME: this routine might accept bits of algorithmic cleanup, it is - too kludgey for my taste... */ + if (size_left < bufsize) + { + /* Last read -- zero out area beyond. */ + bufsize = size_left; + count = bufsize % BLOCKSIZE; + if (count) + memset (blk->buffer + size_left, 0, BLOCKSIZE - count); + } -static int -deal_with_sparse (char *name, union block *header) + count = (fd <= 0) ? bufsize : safe_read (fd, blk->buffer, bufsize); + if (count == SAFE_READ_ERROR) + { + read_diag_details (st->orig_file_name, + st->stat.st_size - size_left, bufsize); + pad_archive (size_left); + return dump_status_short; + } + size_left -= count; + set_next_block_after (blk + (bufsize - 1) / BLOCKSIZE); + + if (count != bufsize) + { + char buf[UINTMAX_STRSIZE_BOUND]; + memset (blk->buffer + count, 0, bufsize - count); + WARNOPT (WARN_FILE_SHRANK, + (0, 0, + ngettext ("%s: File shrank by %s byte; padding with zeros", + "%s: File shrank by %s bytes; padding with zeros", + size_left), + quotearg_colon (st->orig_file_name), + STRINGIFY_BIGINT (size_left, buf))); + if (! ignore_failed_read_option) + set_exit_status (TAREXIT_DIFFERS); + pad_archive (size_left - (bufsize - count)); + return dump_status_short; + } + } + return dump_status_ok; +} + + +/* Copy info from the directory identified by ST into the archive. + DIRECTORY contains the directory's entries. */ + +static void +dump_dir0 (struct tar_stat_info *st, char const *directory) { - size_t numbytes = 0; - off_t offset = 0; - int file; - int sparses = 0; - ssize_t count; - char buffer[BLOCKSIZE]; + bool top_level = ! st->parent; + const char *tag_file_name; + union block *blk = NULL; + off_t block_ordinal = current_block_ordinal (); - if (archive_format == OLDGNU_FORMAT) - header->oldgnu_header.isextended = 0; + st->stat.st_size = 0; /* force 0 size on dir */ - if (file = open (name, O_RDONLY), file < 0) - /* This problem will be caught later on, so just return. */ - return 0; + blk = start_header (st); + if (!blk) + return; - init_sparsearray (); - clear_buffer (buffer); + if (incremental_option && archive_format != POSIX_FORMAT) + blk->header.typeflag = GNUTYPE_DUMPDIR; + else /* if (standard_option) */ + blk->header.typeflag = DIRTYPE; - for (;;) - { - /* Realloc the scratch area as necessary. FIXME: should reallocate - only at beginning of a new instance of non-zero data. */ + /* If we're gnudumping, we aren't done yet so don't close it. */ - if (sp_array_size <= sparses) + if (!incremental_option) + finish_header (st, blk, block_ordinal); + else if (gnu_list_name->directory) + { + if (archive_format == POSIX_FORMAT) { - sparsearray = - xrealloc (sparsearray, - 2 * sp_array_size * sizeof (struct sp_array)); - sp_array_size *= 2; + xheader_store ("GNU.dumpdir", st, + safe_directory_contents (gnu_list_name->directory)); + finish_header (st, blk, block_ordinal); } - - count = safe_read (file, buffer, sizeof buffer); - if (count <= 0) - break; + else + { + off_t size_left; + off_t totsize; + size_t bufsize; + ssize_t count; + const char *buffer, *p_buffer; - /* Process one block. */ + block_ordinal = current_block_ordinal (); + buffer = safe_directory_contents (gnu_list_name->directory); + totsize = dumpdir_size (buffer); + OFF_TO_CHARS (totsize, blk->header.size); + finish_header (st, blk, block_ordinal); + p_buffer = buffer; + size_left = totsize; - if (count == sizeof buffer) + mv_begin_write (st->file_name, totsize, totsize); + while (size_left > 0) + { + blk = find_next_block (); + bufsize = available_space_after (blk); + if (size_left < bufsize) + { + bufsize = size_left; + count = bufsize % BLOCKSIZE; + if (count) + memset (blk->buffer + size_left, 0, BLOCKSIZE - count); + } + memcpy (blk->buffer, p_buffer, bufsize); + size_left -= bufsize; + p_buffer += bufsize; + set_next_block_after (blk + (bufsize - 1) / BLOCKSIZE); + } + } + return; + } + + if (!recursion_option) + return; + + if (one_file_system_option + && !top_level + && st->parent->stat.st_dev != st->stat.st_dev) + { + if (verbose_option) + WARNOPT (WARN_XDEV, + (0, 0, + _("%s: file is on a different filesystem; not dumped"), + quotearg_colon (st->orig_file_name))); + } + else + { + char *name_buf; + size_t name_size; + + switch (check_exclusion_tags (st, &tag_file_name)) + { + case exclusion_tag_all: + /* Handled in dump_file0 */ + break; - if (zero_block_p (buffer)) + case exclusion_tag_none: { - if (numbytes) + char const *entry; + size_t entry_len; + size_t name_len; + + name_buf = xstrdup (st->orig_file_name); + name_size = name_len = strlen (name_buf); + + /* Now output all the files in the directory. */ + for (entry = directory; (entry_len = strlen (entry)) != 0; + entry += entry_len + 1) { - sparsearray[sparses++].numbytes = numbytes; - numbytes = 0; + if (name_size < name_len + entry_len) + { + name_size = name_len + entry_len; + name_buf = xrealloc (name_buf, name_size + 1); + } + strcpy (name_buf + name_len, entry); + if (!excluded_name (name_buf)) + dump_file (st, entry, name_buf); } - } - else - { - if (!numbytes) - sparsearray[sparses].offset = offset; - numbytes += count; - } - else + free (name_buf); + } + break; - /* Since count < sizeof buffer, we have the last bit of the file. */ + case exclusion_tag_contents: + exclusion_tag_warning (st->orig_file_name, tag_file_name, + _("contents not dumped")); + name_size = strlen (st->orig_file_name) + strlen (tag_file_name) + 1; + name_buf = xmalloc (name_size); + strcpy (name_buf, st->orig_file_name); + strcat (name_buf, tag_file_name); + dump_file (st, tag_file_name, name_buf); + free (name_buf); + break; - if (!zero_block_p (buffer)) - { - if (!numbytes) - sparsearray[sparses].offset = offset; - numbytes += count; - } - else - /* The next two lines are suggested by Andreas Degert, who says - they are required for trailing full blocks to be written to the - archive, when all zeroed. Yet, it seems to me that the case - does not apply. Further, at restore time, the file is not as - sparse as it should. So, some serious cleanup is *also* needed - in this area. Just one more... :-(. FIXME. */ - if (numbytes) - numbytes += count; - - /* Prepare for next block. */ - - offset += count; - /* FIXME: do not clear unless necessary. */ - clear_buffer (buffer); + case exclusion_tag_under: + exclusion_tag_warning (st->orig_file_name, tag_file_name, + _("contents not dumped")); + break; + } } +} - if (numbytes) - sparsearray[sparses++].numbytes = numbytes; - else +/* Ensure exactly one trailing slash. */ +static void +ensure_slash (char **pstr) +{ + size_t len = strlen (*pstr); + while (len >= 1 && ISSLASH ((*pstr)[len - 1])) + len--; + if (!ISSLASH ((*pstr)[len])) + *pstr = xrealloc (*pstr, len + 2); + (*pstr)[len++] = '/'; + (*pstr)[len] = '\0'; +} + +/* If we just ran out of file descriptors, release a file descriptor + in the directory chain somewhere leading from DIR->parent->parent + up through the root. Return true if successful, false (preserving + errno == EMFILE) otherwise. + + Do not release DIR's file descriptor, or DIR's parent, as other + code assumes that they work. On some operating systems, another + process can claim file descriptor resources as we release them, and + some calls or their emulations require multiple file descriptors, + so callers should not give up if a single release doesn't work. */ + +static bool +open_failure_recover (struct tar_stat_info const *dir) +{ + if (errno == EMFILE && dir && dir->parent) { - sparsearray[sparses].offset = offset - 1; - sparsearray[sparses++].numbytes = 1; + struct tar_stat_info *p; + for (p = dir->parent->parent; p; p = p->parent) + if (0 < p->fd && (! p->parent || p->parent->fd <= 0)) + { + tar_stat_close (p); + return true; + } + errno = EMFILE; } - return close (file) == 0 && 0 <= count ? sparses : 0; + return false; } -static int -finish_sparse_file (int file, off_t *sizeleft, off_t fullsize, char *name) +/* Return the directory entries of ST, in a dynamically allocated buffer, + each entry followed by '\0' and the last followed by an extra '\0'. + Return null on failure, setting errno. */ +char * +get_directory_entries (struct tar_stat_info *st) { - union block *start; - size_t bufsize; - int sparses = 0; - ssize_t count; + DIR *dirstream; + while (! (dirstream = fdopendir (st->fd)) && open_failure_recover (st)) + continue; - while (*sizeleft > 0) + if (! dirstream) + return 0; + else { - start = find_next_block (); - memset (start->buffer, 0, BLOCKSIZE); - bufsize = sparsearray[sparses].numbytes; - if (! bufsize) - abort (); - - if (lseek (file, sparsearray[sparses++].offset, SEEK_SET) < 0) - { - (ignore_failed_read_option ? seek_warn_details : seek_error_details) - (name, sparsearray[sparses - 1].offset); - break; - } + char *entries = streamsavedir (dirstream); + int streamsavedir_errno = errno; - /* If the number of bytes to be written here exceeds the size of - the temporary buffer, do it in steps. */ - - while (bufsize > BLOCKSIZE) + int fd = dirfd (dirstream); + if (fd < 0) { - count = safe_read (file, start->buffer, BLOCKSIZE); - if (count < 0) + /* The dirent.h implementation doesn't use file descriptors + for directory streams, so open the directory again. */ + char const *name = st->orig_file_name; + if (closedir (dirstream) != 0) + close_diag (name); + dirstream = 0; + fd = subfile_open (st->parent, + st->parent ? last_component (name) : name, + open_searchdir_flags); + if (fd < 0) + fd = - errno; + else { - (ignore_failed_read_option - ? read_warn_details - : read_error_details) - (name, fullsize - *sizeleft, bufsize); - return 1; + struct stat dirst; + if (! (fstat (fd, &dirst) == 0 + && st->stat.st_ino == dirst.st_ino + && st->stat.st_dev == dirst.st_dev)) + { + close (fd); + fd = - IMPOSTOR_ERRNO; + } } - bufsize -= count; - *sizeleft -= count; - set_next_block_after (start); - start = find_next_block (); - memset (start->buffer, 0, BLOCKSIZE); } - { - char buffer[BLOCKSIZE]; + st->fd = fd; + st->dirstream = dirstream; + errno = streamsavedir_errno; + return entries; + } +} - clear_buffer (buffer); - count = safe_read (file, buffer, bufsize); - memcpy (start->buffer, buffer, BLOCKSIZE); - } +/* Dump the directory ST. Return true if successful, false (emitting + diagnostics) otherwise. Get ST's entries, recurse through its + subdirectories, and clean up file descriptors afterwards. */ +static bool +dump_dir (struct tar_stat_info *st) +{ + char *directory = get_directory_entries (st); + if (! directory) + { + savedir_diag (st->orig_file_name); + return false; + } - if (count < 0) - { - (ignore_failed_read_option - ? read_warn_details - : read_error_details) - (name, fullsize - *sizeleft, bufsize); - return 1; - } + dump_dir0 (st, directory); - *sizeleft -= count; - set_next_block_after (start); - } - free (sparsearray); -#if 0 - set_next_block_after (start + (count - 1) / BLOCKSIZE); -#endif - return 0; + restore_parent_fd (st); + free (directory); + return true; } + + +/* Number of links a file can have without having to be entered into + the link table. Typically this is 1, but in trickier circumstances + it is 0. */ +static nlink_t trivial_link_count; + /* Main functions of this module. */ void create_archive (void) { - char *p; + struct name const *p; + + trivial_link_count = name_count <= 1 && ! dereference_option; open_archive (ACCESS_WRITE); + buffer_write_global_xheader (); if (incremental_option) { @@ -913,66 +1355,87 @@ create_archive (void) collect_and_sort_names (); - while (p = name_from_list (), p) - if (!excluded_name (p)) - dump_file (p, -1, (dev_t) 0); + while ((p = name_from_list ()) != NULL) + if (!excluded_name (p->name)) + dump_file (0, p->name, p->name); blank_name_list (); - while (p = name_from_list (), p) - if (!excluded_name (p)) + while ((p = name_from_list ()) != NULL) + if (!excluded_name (p->name)) { - size_t plen = strlen (p); + struct tar_stat_info st; + size_t plen = strlen (p->name); if (buffer_size <= plen) { while ((buffer_size *= 2) <= plen) continue; buffer = xrealloc (buffer, buffer_size); } - memcpy (buffer, p, plen); + memcpy (buffer, p->name, plen); if (! ISSLASH (buffer[plen - 1])) - buffer[plen++] = '/'; - q = gnu_list_name->dir_contents; + buffer[plen++] = DIRECTORY_SEPARATOR; + tar_stat_init (&st); + q = directory_contents (gnu_list_name->directory); if (q) while (*q) { size_t qlen = strlen (q); if (*q == 'Y') { + if (! st.orig_file_name) + { + int fd = open (p->name, open_searchdir_flags); + if (fd < 0) + { + open_diag (p->name); + break; + } + st.fd = fd; + if (fstat (fd, &st.stat) != 0) + { + stat_diag (p->name); + break; + } + st.orig_file_name = xstrdup (p->name); + } 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); + dump_file (&st, q + 1, buffer); } q += qlen + 1; } + tar_stat_destroy (&st); } free (buffer); } else { - while (p = name_next (1), p) - if (!excluded_name (p)) - dump_file (p, 1, (dev_t) 0); + const char *name; + while ((name = name_next (1)) != NULL) + if (!excluded_name (name)) + dump_file (0, name, name); } write_eot (); close_archive (); - + finish_deferred_unlinks (); if (listed_incremental_option) write_directory_file (); } /* Calculate the hash of a link. */ -static unsigned -hash_link (void const *entry, unsigned n_buckets) +static size_t +hash_link (void const *entry, size_t n_buckets) { - struct link const *link = entry; - return (uintmax_t) (link->dev ^ link->ino) % n_buckets; + struct link const *l = entry; + uintmax_t num = l->dev ^ l->ino; + return num % n_buckets; } /* Compare two links for equality. */ @@ -984,716 +1447,504 @@ compare_links (void const *entry1, void const *entry2) return ((link1->dev ^ link2->dev) | (link1->ino ^ link2->ino)) == 0; } +static void +unknown_file_error (char const *p) +{ + WARNOPT (WARN_FILE_IGNORED, + (0, 0, _("%s: Unknown file type; file ignored"), + quotearg_colon (p))); + if (!ignore_failed_read_option) + set_exit_status (TAREXIT_FAILURE); +} + + +/* Handling of hard links */ + /* Table of all non-directories 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. */ static Hash_table *link_table; -/* 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 the top level - of 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_INFO 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, int top_level, dev_t parent_device) +/* Try to dump stat as a hard link to another file in the archive. + Return true if successful. */ +static bool +dump_hard_link (struct tar_stat_info *st) { - union block *header; - char type; - union block *exhdr; - char save_typeflag; - time_t original_ctime; - struct utimbuf restore_times; - off_t block_ordinal = -1; - - /* FIXME: `header' might be used uninitialized in this - function. Reported by Bruno Haible. */ - - if (interactive_option && !confirm ("add", p)) - return; - - if (deref_stat (dereference_option, p, ¤t_stat_info.stat) != 0) + if (link_table + && (trivial_link_count < st->stat.st_nlink || remove_files_option)) { - if (ignore_failed_read_option) - stat_warn (p); - else - stat_error (p); - return; - } + struct link lp; + struct link *duplicate; + off_t block_ordinal; + union block *blk; - original_ctime = current_stat_info.stat.st_ctime; - restore_times.actime = current_stat_info.stat.st_atime; - restore_times.modtime = current_stat_info.stat.st_mtime; + lp.ino = st->stat.st_ino; + lp.dev = st->stat.st_dev; -#ifdef S_ISHIDDEN - if (S_ISHIDDEN (current_stat_info.stat.st_mode)) - { - char *new = (char *) alloca (strlen (p) + 2); - if (new) + if ((duplicate = hash_lookup (link_table, &lp))) { - strcpy (new, p); - strcat (new, "@"); - p = new; - } - } -#endif + /* We found a link. */ + char const *link_name = safer_name_suffix (duplicate->name, true, + absolute_names_option); - /* See if we want only new files, and check if this one is too old to - put in the archive. */ + duplicate->nlink--; - if ((0 < top_level || !incremental_option) - && !S_ISDIR (current_stat_info.stat.st_mode) - && current_stat_info.stat.st_mtime < newer_mtime_option - && (!after_date_option || current_stat_info.stat.st_ctime < newer_ctime_option)) - { - if (0 < top_level) - WARN ((0, 0, _("%s: file is unchanged; not dumped"), - quotearg_colon (p))); - /* FIXME: recheck this return. */ - return; - } - -#if !MSDOS - /* See if we are trying to dump the archive. */ - - if (ar_dev && current_stat_info.stat.st_dev == ar_dev && current_stat_info.stat.st_ino == ar_ino) - { - WARN ((0, 0, _("%s: file is the archive; not dumped"), - quotearg_colon (p))); - return; - } -#endif - - if (S_ISDIR (current_stat_info.stat.st_mode)) - { - char *directory; - char const *entry; - size_t entrylen; - char *namebuf; - size_t buflen; - size_t len; - dev_t our_device = current_stat_info.stat.st_dev; - - errno = 0; - - directory = savedir (p); - if (! directory) - { - if (ignore_failed_read_option) - savedir_warn (p); - else - savedir_error (p); - return; - } + block_ordinal = current_block_ordinal (); + assign_string (&st->link_name, link_name); + if (NAME_FIELD_SIZE - (archive_format == OLDGNU_FORMAT) + < strlen (link_name)) + write_long_link (st); - /* Build new prototype name. Ensure exactly one trailing slash. */ + st->stat.st_size = 0; + blk = start_header (st); + if (!blk) + return false; + tar_copy_str (blk->header.linkname, link_name, NAME_FIELD_SIZE); - len = strlen (p); - buflen = len + NAME_FIELD_SIZE; - namebuf = xmalloc (buflen + 1); - memcpy (namebuf, p, len); - while (len >= 1 && ISSLASH (namebuf[len - 1])) - len--; - namebuf[len++] = '/'; - namebuf[len] = '\0'; + blk->header.typeflag = LNKTYPE; + finish_header (st, blk, block_ordinal); - if (! is_avoided_name (namebuf)) - { - /* 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 - some subtle reason for the exclusion that I don't know, but the - current behavior is broken.'' I do not know those subtle - reasons either, so until these are reported (anew?), just allow - directory blocks to be written even with old archives. */ + if (remove_files_option) + queue_deferred_unlink (st->orig_file_name, false); - block_ordinal = current_block_ordinal (); - current_stat_info.stat.st_size = 0; /* force 0 size on dir */ + return true; + } + } + return false; +} - /* FIXME: If people could really read standard archives, this - should be: +static void +file_count_links (struct tar_stat_info *st) +{ + if (hard_dereference_option) + return; + if (trivial_link_count < st->stat.st_nlink) + { + struct link *duplicate; + char *linkname = NULL; + struct link *lp; - header - = start_header (standard_option ? p : namebuf, ¤t_stat_info); + assign_string (&linkname, st->orig_file_name); + transform_name (&linkname, XFORM_LINK); - but since they'd interpret DIRTYPE blocks as regular - files, we'd better put the / on the name. */ + lp = xmalloc (offsetof (struct link, name) + + strlen (linkname) + 1); + lp->ino = st->stat.st_ino; + lp->dev = st->stat.st_dev; + lp->nlink = st->stat.st_nlink; + strcpy (lp->name, linkname); + free (linkname); - header = start_header (namebuf, ¤t_stat_info); + if (! ((link_table + || (link_table = hash_initialize (0, 0, hash_link, + compare_links, 0))) + && (duplicate = hash_insert (link_table, lp)))) + xalloc_die (); - if (incremental_option) - header->header.typeflag = GNUTYPE_DUMPDIR; - else /* if (standard_option) */ - header->header.typeflag = DIRTYPE; + if (duplicate != lp) + abort (); + lp->nlink--; + } +} - /* If we're gnudumping, we aren't done yet so don't close it. */ +/* For each dumped file, check if all its links were dumped. Emit + warnings if it is not so. */ +void +check_links (void) +{ + struct link *lp; - if (!incremental_option) - finish_header (header, block_ordinal); - } + if (!link_table) + return; - if (incremental_option && gnu_list_name->dir_contents) + for (lp = hash_get_first (link_table); lp; + lp = hash_get_next (link_table, lp)) + { + if (lp->nlink) { - off_t sizeleft; - off_t totsize; - size_t bufsize; - union block *start; - ssize_t count; - const char *buffer, *p_buffer; - - buffer = gnu_list_name->dir_contents; /* FOO */ - totsize = 0; - if (buffer) - for (p_buffer = buffer; *p_buffer; ) - { - size_t size = strlen (p_buffer) + 1; - totsize += size; - p_buffer += size; - } - totsize++; - OFF_TO_CHARS (totsize, header->header.size); - finish_header (header, block_ordinal); - p_buffer = buffer; - sizeleft = totsize; - while (sizeleft > 0) - { - if (multi_volume_option) - { - assign_string (&save_name, p); - save_sizeleft = sizeleft; - save_totsize = totsize; - } - start = find_next_block (); - bufsize = available_space_after (start); - if (sizeleft < bufsize) - { - bufsize = sizeleft; - count = bufsize % BLOCKSIZE; - if (count) - memset (start->buffer + sizeleft, 0, BLOCKSIZE - count); - } - memcpy (start->buffer, p_buffer, bufsize); - sizeleft -= bufsize; - p_buffer += bufsize; - set_next_block_after (start + (bufsize - 1) / BLOCKSIZE); - } - if (multi_volume_option) - assign_string (&save_name, 0); - goto finish_dir; + WARN ((0, 0, _("Missing links to %s."), quote (lp->name))); } + } +} - /* 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. */ +/* Assuming DIR is the working directory, open FILE, using FLAGS to + control the open. A null DIR means to use ".". If we are low on + file descriptors, try to release one or more from DIR's parents to + reuse it. */ +int +subfile_open (struct tar_stat_info const *dir, char const *file, int flags) +{ + int fd; - if (! recursion_option) - goto finish_dir; + static bool initialized; + if (! initialized) + { + /* Initialize any tables that might be needed when file + descriptors are exhausted, and whose initialization might + require a file descriptor. This includes the system message + catalog and tar's message catalog. */ + initialized = true; + strerror (ENOENT); + gettext (""); + } - /* 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. */ + while ((fd = openat (dir ? dir->fd : AT_FDCWD, file, flags)) < 0 + && open_failure_recover (dir)) + continue; + return fd; +} - if (one_file_system_option && !top_level - && parent_device != current_stat_info.stat.st_dev) +/* Restore the file descriptor for ST->parent, if it was temporarily + closed to conserve file descriptors. On failure, set the file + descriptor to the negative of the corresponding errno value. Call + this every time a subdirectory is ascended from. */ +void +restore_parent_fd (struct tar_stat_info const *st) +{ + struct tar_stat_info *parent = st->parent; + if (parent && ! parent->fd) + { + int parentfd = openat (st->fd, "..", open_searchdir_flags); + struct stat parentstat; + + if (parentfd < 0) + parentfd = - errno; + else if (! (fstat (parentfd, &parentstat) == 0 + && parent->stat.st_ino == parentstat.st_ino + && parent->stat.st_dev == parentstat.st_dev)) { - if (verbose_option) - WARN ((0, 0, - _("%s: file is on a different filesystem; not dumped"), - quotearg_colon (p))); - goto finish_dir; + close (parentfd); + parentfd = IMPOSTOR_ERRNO; } - /* Now output all the files in the directory. */ - - /* FIXME: Should speed this up by cd-ing into the dir. */ - - for (entry = directory; - (entrylen = strlen (entry)) != 0; - entry += entrylen + 1) + if (parentfd < 0) { - if (buflen < len + entrylen) + int origfd = open (parent->orig_file_name, open_searchdir_flags); + if (0 <= origfd) { - buflen = len + entrylen; - namebuf = xrealloc (namebuf, buflen + 1); + if (fstat (parentfd, &parentstat) == 0 + && parent->stat.st_ino == parentstat.st_ino + && parent->stat.st_dev == parentstat.st_dev) + parentfd = origfd; + else + close (origfd); } - strcpy (namebuf + len, entry); - if (!excluded_name (namebuf)) - dump_file (namebuf, 0, our_device); } - finish_dir: - - free (directory); - free (namebuf); - if (atime_preserve_option) - utime (p, &restore_times); - return; + parent->fd = parentfd; } - else if (is_avoided_name (p)) - return; - else - { - /* Check for multiple links. */ - - if (1 < current_stat_info.stat.st_nlink && link_table) - { - struct link lp; - struct link *dup; - lp.ino = current_stat_info.stat.st_ino; - lp.dev = current_stat_info.stat.st_dev; - - if ((dup = hash_lookup (link_table, &lp))) - { - /* We found a link. */ - char const *link_name = safer_name_suffix (dup->name, 1); +} - dup->nlink--; - - block_ordinal = current_block_ordinal (); - assign_string (¤t_stat_info.link_name, link_name); - if (NAME_FIELD_SIZE <= strlen (link_name)) - write_long_link (¤t_stat_info); +/* Dump a single file, recursing on directories. ST is the file's + status info, NAME its name relative to the parent directory, and P + its full name (which may be relative to the working directory). */ - current_stat_info.stat.st_size = 0; - header = start_header (p, ¤t_stat_info); - strncpy (header->header.linkname, link_name, NAME_FIELD_SIZE); +/* FIXME: One should make sure that for *every* path leading to setting + exit_status to failure, a clear diagnostic has been issued. */ - /* Force null termination. */ - header->header.linkname[NAME_FIELD_SIZE - 1] = 0; +static void +dump_file0 (struct tar_stat_info *st, char const *name, char const *p) +{ + union block *header; + char type; + off_t original_size; + struct timespec original_ctime; + struct timespec restore_times[2]; + off_t block_ordinal = -1; + int fd = 0; + bool is_dir; + struct tar_stat_info const *parent = st->parent; + bool top_level = ! parent; + int parentfd = top_level ? AT_FDCWD : parent->fd; + void (*diag) (char const *) = 0; - header->header.typeflag = LNKTYPE; - finish_header (header, block_ordinal); + if (interactive_option && !confirm ("add", p)) + return; - /* FIXME: Maybe remove from table after all links found? */ + assign_string (&st->orig_file_name, p); + assign_string (&st->file_name, + safer_name_suffix (p, false, absolute_names_option)); - if (remove_files_option && unlink (p) != 0) - unlink_error (p); + transform_name (&st->file_name, XFORM_REGFILE); - /* We dumped it, and we don't need to put it in the - table again. */ - return; - } + if (parentfd < 0 && ! top_level) + { + errno = - parentfd; + diag = open_diag; + } + else if (fstatat (parentfd, name, &st->stat, fstatat_flags) != 0) + diag = stat_diag; + else if (file_dumpable_p (&st->stat)) + { + fd = subfile_open (parent, name, open_read_flags); + if (fd < 0) + diag = open_diag; + else + { + st->fd = fd; + if (fstat (fd, &st->stat) != 0) + diag = stat_diag; } + } + if (diag) + { + file_removed_diag (p, top_level, diag); + return; + } - /* This is not a link to a previously dumped file, so dump it. */ + st->archive_file_size = original_size = st->stat.st_size; + st->atime = restore_times[0] = get_stat_atime (&st->stat); + st->mtime = restore_times[1] = get_stat_mtime (&st->stat); + st->ctime = original_ctime = get_stat_ctime (&st->stat); - if (S_ISREG (current_stat_info.stat.st_mode) - || S_ISCTG (current_stat_info.stat.st_mode)) +#ifdef S_ISHIDDEN + if (S_ISHIDDEN (st->stat.st_mode)) + { + char *new = (char *) alloca (strlen (p) + 2); + if (new) { - 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 than 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_info.stat) - < (current_stat_info.stat.st_size / ST_NBLOCKSIZE - + (current_stat_info.stat.st_size % ST_NBLOCKSIZE != 0))) - { - int counter; - - block_ordinal = current_block_ordinal (); - header = start_header (p, ¤t_stat_info); - header->header.typeflag = GNUTYPE_SPARSE; - header_moved = 1; + strcpy (new, p); + strcat (new, "@"); + p = new; + } + } +#endif - /* 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. */ + /* See if we want only new files, and check if this one is too old to + put in the archive. - sparses = deal_with_sparse (p, header); + This check is omitted if incremental_option is set *and* the + requested file is not explicitely listed in the command line. */ - /* See if we'll need an extended header later. */ + if (!(incremental_option && !is_individual_file (p)) + && !S_ISDIR (st->stat.st_mode) + && OLDER_TAR_STAT_TIME (*st, m) + && (!after_date_option || OLDER_TAR_STAT_TIME (*st, c))) + { + if (!incremental_option && verbose_option) + WARNOPT (WARN_FILE_UNCHANGED, + (0, 0, _("%s: file is unchanged; not dumped"), + quotearg_colon (p))); + return; + } - if (SPARSES_IN_OLDGNU_HEADER < sparses) - header->oldgnu_header.isextended = 1; + /* See if we are trying to dump the archive. */ + if (sys_file_is_archive (st)) + { + WARNOPT (WARN_IGNORE_ARCHIVE, + (0, 0, _("%s: file is the archive; not dumped"), + quotearg_colon (p))); + return; + } - /* 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. */ + is_dir = S_ISDIR (st->stat.st_mode) != 0; - OFF_TO_CHARS (current_stat_info.stat.st_size, - header->oldgnu_header.realsize); + if (!is_dir && dump_hard_link (st)) + return; - /* 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. */ + if (is_dir || S_ISREG (st->stat.st_mode) || S_ISCTG (st->stat.st_mode)) + { + bool ok; + struct stat final_stat; - current_stat_info.stat.st_size = find_new_file_size (sparses); - OFF_TO_CHARS (current_stat_info.stat.st_size, header->header.size); + if (is_dir) + { + const char *tag_file_name; + ensure_slash (&st->orig_file_name); + ensure_slash (&st->file_name); - 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); - } - } + if (check_exclusion_tags (st, &tag_file_name) == exclusion_tag_all) + { + exclusion_tag_warning (st->orig_file_name, tag_file_name, + _("directory not dumped")); + return; } - sizeleft = current_stat_info.stat.st_size; + ok = dump_dir (st); - /* Don't bother opening empty, world readable files. Also do not open - files when archive is meant for /dev/null. */ + fd = st->fd; + parentfd = top_level ? AT_FDCWD : parent->fd; + } + else + { + enum dump_status status; - if (dev_null_output - || (sizeleft == 0 - && MODE_R == (MODE_R & current_stat_info.stat.st_mode))) - f = -1; - else + if (fd && sparse_option && ST_IS_SPARSE (st->stat)) { - 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; - } + status = sparse_dump_file (fd, st); + if (status == dump_status_not_implemented) + status = dump_regular_file (fd, st); } + else + status = dump_regular_file (fd, st); - /* If the file is sparse, we've already taken care of this. */ - - if (!header_moved) + switch (status) { - block_ordinal = current_block_ordinal (); - header = start_header (p, ¤t_stat_info); - } + case dump_status_ok: + case dump_status_short: + file_count_links (st); + break; - /* Mark contiguous files, if we support them. */ + case dump_status_fail: + break; - if (archive_format != V7_FORMAT && S_ISCTG (current_stat_info.stat.st_mode)) - header->header.typeflag = CONTTYPE; + case dump_status_not_implemented: + abort (); + } - isextended = header->oldgnu_header.isextended; - save_typeflag = header->header.typeflag; - finish_header (header, block_ordinal); - if (isextended) - { - int sparses_emitted = SPARSES_IN_OLDGNU_HEADER; + ok = status == dump_status_ok; + } - 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 (ok) + { + if (fd < 0) { - if (f < 0 - || finish_sparse_file (f, &sizeleft, - current_stat_info.stat.st_size, p)) - goto padit; + errno = - fd; + ok = false; } - else - while (sizeleft > 0) - { - if (multi_volume_option) - { - assign_string (&save_name, p); - save_sizeleft = sizeleft; - save_totsize = current_stat_info.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_info.stat.st_size - sizeleft, bufsize); - goto padit; - } - sizeleft -= count; - - /* 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, - ngettext ("%s: File shrank by %s byte; padding with zeros", - "%s: File shrank by %s bytes; padding with zeros", - sizeleft), - 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) + else if (fd == 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) + if (parentfd < 0 && ! top_level) { - char const *qp = quotearg_colon (p); - WARN ((0, 0, _("%s: file changed as we read it"), qp)); + errno = - parentfd; + ok = false; } - 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); + else + ok = fstatat (parentfd, name, &final_stat, fstatat_flags) == 0; } - goto file_was_dumped; - - /* File shrunk or gave error, pad out tape to match the size we - specified in the header. */ + else + ok = fstat (fd, &final_stat) == 0; - 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); - } - goto file_was_dumped; + if (! ok) + file_removed_diag (p, top_level, stat_diag); } -#ifdef HAVE_READLINK - else if (S_ISLNK (current_stat_info.stat.st_mode)) + + if (ok) { - char *buffer; - int size; - size_t linklen = current_stat_info.stat.st_size; - if (linklen != current_stat_info.stat.st_size || linklen + 1 == 0) - xalloc_die (); - buffer = (char *) alloca (linklen + 1); - size = readlink (p, buffer, linklen + 1); - if (size < 0) + if ((timespec_cmp (get_stat_ctime (&final_stat), original_ctime) != 0 + /* Original ctime will change if the file is a directory and + --remove-files is given */ + && !(remove_files_option && is_dir)) + || original_size < final_stat.st_size) { - if (ignore_failed_read_option) - readlink_warn (p); - else - readlink_error (p); - return; + WARNOPT (WARN_FILE_CHANGED, + (0, 0, _("%s: file changed as we read it"), + quotearg_colon (p))); + set_exit_status (TAREXIT_DIFFERS); } - buffer[size] = '\0'; - assign_string (¤t_stat_info.link_name, buffer); - if (size >= NAME_FIELD_SIZE) - write_long_link (¤t_stat_info); + else if (atime_preserve_option == replace_atime_preserve + && set_file_atime (fd, p, restore_times) != 0) + utime_error (p); + } - block_ordinal = current_block_ordinal (); - current_stat_info.stat.st_size = 0; /* force 0 size on symlink */ - header = start_header (p, ¤t_stat_info); - strncpy (header->header.linkname, buffer, NAME_FIELD_SIZE); - header->header.linkname[NAME_FIELD_SIZE - 1] = '\0'; - header->header.typeflag = SYMTYPE; - finish_header (header, block_ordinal); - /* nothing more to do to it */ + ok &= tar_stat_close (st); + if (ok && remove_files_option) + queue_deferred_unlink (p, is_dir); - if (remove_files_option) - { - if (unlink (p) == -1) - unlink_error (p); - } - goto file_was_dumped; - } -#endif - else if (S_ISCHR (current_stat_info.stat.st_mode)) - type = CHRTYPE; - else if (S_ISBLK (current_stat_info.stat.st_mode)) - type = BLKTYPE; - else if (S_ISFIFO (current_stat_info.stat.st_mode)) - type = FIFOTYPE; - else if (S_ISSOCK (current_stat_info.stat.st_mode)) - { - WARN ((0, 0, _("%s: socket ignored"), quotearg_colon (p))); - return; - } - else if (S_ISDOOR (current_stat_info.stat.st_mode)) + return; + } +#ifdef HAVE_READLINK + else if (S_ISLNK (st->stat.st_mode)) + { + char *buffer; + int size; + size_t linklen = st->stat.st_size; + if (linklen != st->stat.st_size || linklen + 1 == 0) + xalloc_die (); + buffer = (char *) alloca (linklen + 1); + size = readlinkat (parentfd, name, buffer, linklen + 1); + if (size < 0) { - WARN ((0, 0, _("%s: door ignored"), quotearg_colon (p))); + file_removed_diag (p, top_level, readlink_diag); return; } - else - goto unknown; + buffer[size] = '\0'; + assign_string (&st->link_name, buffer); + transform_name (&st->link_name, XFORM_SYMLINK); + if (NAME_FIELD_SIZE - (archive_format == OLDGNU_FORMAT) < size) + write_long_link (st); + + block_ordinal = current_block_ordinal (); + st->stat.st_size = 0; /* force 0 size on symlink */ + header = start_header (st); + if (!header) + return; + tar_copy_str (header->header.linkname, st->link_name, NAME_FIELD_SIZE); + header->header.typeflag = SYMTYPE; + finish_header (st, header, block_ordinal); + /* nothing more to do to it */ + + if (remove_files_option) + queue_deferred_unlink (p, false); + + file_count_links (st); + return; + } +#endif + else if (S_ISCHR (st->stat.st_mode)) + type = CHRTYPE; + else if (S_ISBLK (st->stat.st_mode)) + type = BLKTYPE; + else if (S_ISFIFO (st->stat.st_mode)) + type = FIFOTYPE; + else if (S_ISSOCK (st->stat.st_mode)) + { + WARNOPT (WARN_FILE_IGNORED, + (0, 0, _("%s: socket ignored"), quotearg_colon (p))); + return; + } + else if (S_ISDOOR (st->stat.st_mode)) + { + WARNOPT (WARN_FILE_IGNORED, + (0, 0, _("%s: door ignored"), quotearg_colon (p))); + return; + } + else + { + unknown_file_error (p); + return; } if (archive_format == V7_FORMAT) - goto unknown; + { + unknown_file_error (p); + return; + } block_ordinal = current_block_ordinal (); - current_stat_info.stat.st_size = 0; /* force 0 size */ - header = start_header (p, ¤t_stat_info); + st->stat.st_size = 0; /* force 0 size */ + header = start_header (st); + if (!header) + return; header->header.typeflag = type; if (type != FIFOTYPE) { - MAJOR_TO_CHARS (major (current_stat_info.stat.st_rdev), header->header.devmajor); - MINOR_TO_CHARS (minor (current_stat_info.stat.st_rdev), header->header.devminor); + MAJOR_TO_CHARS (major (st->stat.st_rdev), + header->header.devmajor); + MINOR_TO_CHARS (minor (st->stat.st_rdev), + header->header.devminor); } - finish_header (header, block_ordinal); + finish_header (st, header, block_ordinal); if (remove_files_option) - { - if (unlink (p) == -1) - unlink_error (p); - } - goto file_was_dumped; - -unknown: - WARN ((0, 0, _("%s: Unknown file type; file ignored"), - quotearg_colon (p))); - if (! ignore_failed_read_option) - exit_status = TAREXIT_FAILURE; - return; - -file_was_dumped: - if (1 < current_stat_info.stat.st_nlink) - { - struct link *dup; - struct link *lp = xmalloc (offsetof (struct link, name) - + strlen (p) + 1); - lp->ino = current_stat_info.stat.st_ino; - lp->dev = current_stat_info.stat.st_dev; - lp->nlink = current_stat_info.stat.st_nlink; - 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) - abort (); - lp->nlink--; - } - + queue_deferred_unlink (p, false); } -/* For each dumped file, check if all its links were dumped. Emit - warnings if it is not so. */ +/* Dump a file, recursively. PARENT describes the file's parent + directory, NAME is the file's name relative to PARENT, and FULLNAME + its full name, possibly relative to the working directory. NAME + may contain slashes at the top level of invocation. */ + void -check_links () +dump_file (struct tar_stat_info *parent, char const *name, + char const *fullname) { - struct link *lp; - - if (!link_table) - return; - - for (lp = hash_get_first (link_table); lp; - lp = hash_get_next (link_table, lp)) - { - if (lp->nlink) - { - WARN ((0, 0, _("Missing links to '%s'.\n"), lp->name)); - } - } + struct tar_stat_info st; + tar_stat_init (&st); + st.parent = parent; + dump_file0 (&st, name, fullname); + if (parent && listed_incremental_option) + update_parent_directory (parent); + tar_stat_destroy (&st); }